How precise cancer care is

Every part of cancer care is limited by how precise it is: how little tumour a test can find, how close to it a surgeon can cut, how tightly a beam is aimed, how well a dose fits the person. This page sets them side by side: 50 technologies across 4 of 4 areas of care. 49 of the 50 carry a sourced figure for how precise they are today, quoted in the source's own words; the rest say no figure was found rather than guess one. 12 next steps have been shown to help a patient, and 17 are named as not established. 12 technologies are graded as sold beyond the evidence. Use this to know what to ask, not to decide a case.

How to read a row

Each row says what precision is measured in for that technology, and gives today's figure only where a source states one, with the setting it came from and a fragment of the source's own words. The last line says whether the next step has been shown to help a patient, which is a different question from whether it is more precise.

Shown to help

A trial or guideline shows the extra precision changes an outcome a patient feels: survival, side effects, a second operation avoided.

Plausible, not shown

The reasoning is sound and the precision is real, and no trial has yet shown it changes an outcome a patient feels.

Not established

No source shows that more precision here helps a patient. Named rather than left blank, because readers are told it does.

Molecular detection and matching

How little tumour a test can find in blood or tissue, how often it finds something a drug can act on, and how often acting on it helps.

ctDNA residual disease tests after surgery: how many relapses one blood sample catches

MRD / molecular residual disease testing ctDNA MRD positivity (molecular residual disease after curative treatment)Tumour-informed versus tumour-naive ctDNA assays

Precision means

Share of patients who later relapse whose circulating tumour DNA was detected in a blood sample taken in the weeks after surgery, per cent

Today

49 per cent of the patients who relapsed had detectable ctDNA in plasma taken within 120 days of surgery; 25 per cent of all patients tested positive (197 patients with resected non-small-cell lung cancer in TRACERx, tracking a median of 200 mutations chosen from each patient's own tumour (Abbosh 2023))“Landmark analyses of plasma samples collected within 120 days after surgery revealed ctDNA detection in 25% of patients, including 49% of all patients who experienced clinical relapse” Abbosh et al., Nature 2023: tracking early lung cancer metastatic dissemination in TRACERx using ctDNA

What limits it

BiologySome tumours shed too little DNA to be found even before surgery. In TRACERx, the lung adenocarcinomas with no ctDNA before the operation were the slow-growing ones with good outcomes, which is reassuring for them and means the test has nothing to follow afterwards.“A lack of preoperative ctDNA detection distinguished biologically indolent lung adenocarcinoma with good clinical outcome” Abbosh et al., Nature 2023 (TRACERx)
SamplingA single tube of blood holds a limited number of DNA fragments, so at very low tumour fraction there may be no tumour fragment in the sample to find, however deep the sequencing.“the combination of low tumor fraction (TF) and limited number of DNA fragments restricts low-disease-burden monitoring through the prevailing deep targeted sequencing paradigm” Zviran et al., Nat Med 2020: genome-wide cell-free DNA mutational integration
SamplingNegative results are not all-clear results. In c-TRAK TN, seven women with triple-negative breast cancer relapsed without ctDNA ever being detected on three-monthly digital PCR testing.“Seven patients relapsed without prior ctDNA detection.” Turner et al., Ann Oncol 2023: results of the c-TRAK TN trial

Next step

Available, not routineTesting repeatedly rather than once. In TRACERx, sampling every three to six months after a negative landmark test found impending relapse in a further fifth of patients; in colorectal cancer, serial testing found 14 of 16 relapses.“3 to 6 monthly ctDNA surveillance identified impending disease relapse in an additional 20% of landmark-negative patients” Abbosh et al., Nature 2023 (TRACERx)

Continuous and near-continuous ctDNA monitoring Analysis of plasma cell-free DNA by ultradeep sequencing in patients with stages I to III colorectal cancer

What it would buy

Not establishedEarlier warning of relapse for more people. Whether that warning changes the outcome depends on having a treatment that works at that point, which has been shown in one setting (see the row on ctDNA-triggered treatment) and not in general. The ESMO working group's 2022 position was that residual disease detection had high clinical validity and no proven clinical utility.“Molecular residual disease/molecular relapse detection cannot be recommended in routine clinical practice, as currently there is no evidence for clinical utility in directing treatment.” Pascual et al., Ann Oncol 2022: ESMO recommendations on the use of circulating tumour DNA assays

How little tumour DNA a blood test can see: from tracking a few mutations to reading the whole genome

MRD / molecular residual disease testing Circulating tumour DNA fraction (and what a negative plasma result means)Tumour-informed versus tumour-naive ctDNA assays

Precision means

Lowest tumour fraction detectable in plasma, as a share of all cell-free DNA (10⁻⁵ is one tumour fragment in 100,000)

Today

No source found that states a figure. A gap, not a zero.

What limits it

SamplingDeep sequencing of a small panel runs out of molecules before it runs out of depth: the number of fragments in the tube, not the sequencer, sets the floor. OnCo did not find a peer-reviewed abstract stating the limit of detection of each commercial tumour-informed assay in comparable units, so no figure is printed for today's tests.“the combination of low tumor fraction (TF) and limited number of DNA fragments restricts low-disease-burden monitoring through the prevailing deep targeted sequencing paradigm” Zviran et al., Nat Med 2020
WorkflowTracking more mutations finds ctDNA sooner. In c-TRAK TN, a whole-exome-informed multivariant assay detected ctDNA before digital PCR or the tissue-free assay in 12 of 41 patients found by both.“Among patients detected by both assays (41 of 133), 12 of 41 (29.3%) had ctDNA detected earlier by multivariant tumor-informed assay with 1 (2.4%) earlier by the tissue-free assay.” c-TRAK TN tissue-free analysis, JAMA Oncol 2026

Next step

ResearchWhole-genome sequencing of cell-free DNA, which adds up the signal from thousands of a tumour's mutations instead of reading a few very deeply, reached a tumour fraction of 10⁻⁵ in the 2020 proof of principle.“Whole-genome sequencing (WGS) of cfDNA allowed ultra-sensitive detection, capitalizing on the cumulative signal of thousands of somatic mutations observed in solid malignancies, with TF detection sensitivity as low as 10-5.” Zviran et al., Nat Med 2020

Next-generation sequencing (NGS)cfDNA fragmentomics

What it would buy

Plausible, not shownFewer false reassurances after surgery and earlier detection of relapse. In DYNAMIC, a newer assay tracking an average of 29 mutations a patient separated those whose ctDNA cleared after chemotherapy (97 per cent five-year recurrence-free) from those whose did not (none), in a post hoc analysis. Whether the extra sensitivity changes outcomes has not been tested.“post hoc analysis of ctDNA clearance at EOT assessed by a new assay that evaluated an average of 29 tumor-derived mutations per patient predicted for a favorable 5-year recurrence-free probability of 97% versus 0% for ctDNA persistence” Tie et al., Nat Med 2025: DYNAMIC five-year outcomes

How far ahead of scans a ctDNA test warns of relapse

MRD / molecular residual disease testing ctDNA MRD positivity (molecular residual disease after curative treatment)Continuous and near-continuous ctDNA monitoring

Precision means

Time between first ctDNA detection and relapse seen clinically or on imaging, months

Today

Mean 8.7 months ahead of standard imaging (range 0.8 to 16.5) in colorectal cancer; median 7.9 months ahead of clinical relapse in early breast cancer (125 patients with stage I to III colorectal cancer in Denmark tested every three months (Reinert 2019); 55 women with early breast cancer after neoadjuvant chemotherapy (Garcia-Murillas 2015))“Serial ctDNA analyses revealed disease recurrence up to 16.5 months ahead of standard-of-care radiologic imaging (mean, 8.7 months; range, 0.8-16.5 months).” Reinert et al., JAMA Oncol 2019: ultradeep sequencing of plasma cell-free DNA in stage I to III colorectal cancer

What limits it

SamplingLead time over symptoms is not lead time over a scan. When c-TRAK TN acted on a positive result by scanning, 23 of 32 women already had metastases that imaging could see.“Of patients allocated to intervention, 72% (23/32) had metastases on staging at the time of ctDNA+” Turner et al., Ann Oncol 2023 (c-TRAK TN)
SamplingTesting every three months leaves gaps in which a relapse can grow. The c-TRAK TN investigators concluded that testing has to start earlier and be more sensitive or more frequent.“emphasising the importance of commencing ctDNA testing early, with more sensitive and/or frequent ctDNA testing regimes” Turner et al., Ann Oncol 2023 (c-TRAK TN)

Next step

ResearchMonthly, tumour-informed testing. A mathematical model fitted to 176 patients with stage I to III lung cancer predicted it would detect relapse at a median tumour size of 0.83 cm, about 140 days before imaging; this is a model prediction, not a trial result.“For informed monthly cancer relapse testing, the model predicts a median detection size of 0.83 cm and suggests that treatment failure can be detected 140 days earlier than with imaging-based approaches.” Avanzini et al., Sci Adv 2020: a mathematical model of ctDNA shedding predicts tumour detection size

Residual disease kinetics (BCR-ABL halving and ctDNA slopes)

What it would buy

Not establishedMonths of warning, which help only if treatment given in those months works better than treatment given at visible relapse. In c-TRAK TN, none of the five women who started pembrolizumab on a positive result cleared their ctDNA for good.“Of the five patients who commenced pembrolizumab, none achieved sustained ctDNA clearance.” Turner et al., Ann Oncol 2023 (c-TRAK TN)

Tumour-informed against tissue-free residual disease tests

Tumour-informed versus tumour-naive ctDNA assays MRD / molecular residual disease testing

Precision means

Median lead time from ctDNA detection to recurrence, months, by type of assay in the same patients

Today

7.6 months with a methylation-based tissue-free assay against 7.1 months with a whole-exome-informed multivariant assay, a difference that was not statistically significant (159 women with moderate- to high-risk triple-negative breast cancer sampled every three months in c-TRAK TN, 1,026 plasma samples (JAMA Oncol 2026))“Median lead times to recurrence were 7.6 (95% CI, 4.6-10.5) months with tissue-free and 7.1 (95% CI, 5.7-10.0) months with multivariant tumor-informed assay (HR, 1.46; 95% CI, 0.87-2.44; P = .15).” Tissue-free versus tumour-informed ctDNA assays in early triple-negative breast cancer, JAMA Oncol 2026

What limits it

WorkflowTumour-informed tests need a tumour sample to build the panel. In c-TRAK TN, 208 women registered, 185 had tumour sequenced and 171 had mutations that could be tracked.“Two hundred and eight patients registered between 30 January 2018 and 06 December 2019, 185 had tumour sequenced, 171 (92.4%) had trackable mutations” Turner et al., Ann Oncol 2023 (c-TRAK TN)
BiologyTissue-free tests cannot check a plasma signal against the tumour, so they must tell cancer DNA from blood-cell DNA some other way; most plasma mutations in people without cancer come from clonal haematopoiesis (see that row).“The vast majority of cfDNA mutations (81.6% in controls and 53.2% in patients with cancer) had features consistent with clonal hematopoiesis.” Razavi et al., Nat Med 2019: high-intensity sequencing reveals the sources of plasma cell-free DNA variants

Next step

In trialsTissue-free testing in trials for patients with no usable tumour sample, which the c-TRAK TN authors say their data support.“Comparable lead times between the tissue-free and multivariant tumor-informed assays support tissue-free MRD detection in a clinical trial setting, permitting ctDNA testing when tissue is not available.” c-TRAK TN tissue-free analysis, JAMA Oncol 2026

What it would buy

Plausible, not shownAccess to residual disease testing for people whose tumour tissue is used up, too small or unavailable, and a first result without waiting for a bespoke panel. Shown in one cancer and one trial cohort so far.

Using a negative ctDNA result to spare chemotherapy after bowel cancer surgery

MRD / molecular residual disease testing ctDNA MRD positivity (molecular residual disease after curative treatment)

Precision means

Share of patients given adjuvant chemotherapy, per cent, and the difference in recurrence-free survival against standard care, percentage points

Today

Chemotherapy given to 15 per cent with ctDNA-guided care against 28 per cent with standard care; two-year recurrence-free survival 93.5 against 92.4 per cent, meeting non-inferiority; five-year overall survival 93.8 against 93.3 per cent (455 patients with stage II colon cancer randomised 2:1 in DYNAMIC, followed for a median of 37 months and then 59.7 months (Tie 2022, 2025))“A lower percentage of patients in the ctDNA-guided group than in the standard-management group received adjuvant chemotherapy (15% vs. 28%; relative risk, 1.82; 95% confidence interval [CI], 1.25 to 2.65).” Tie et al., NEJM 2022: DYNAMIC

What limits it

BiologyA negative test lowers the risk of recurrence without removing it: in DYNAMIC, three-year recurrence-free survival among untreated ctDNA-negative patients was 92.5 per cent.“Three-year recurrence-free survival was 86.4% among ctDNA-positive patients who received adjuvant chemotherapy and 92.5% among ctDNA-negative patients who did not.” Tie et al., NEJM 2022: DYNAMIC
BiologyThe result did not carry over to stage III. In DYNAMIC-III, de-escalating ctDNA-negative patients cut oxaliplatin use and hospital admissions but recurrence-free survival was slightly lower and the trial did not meet its non-inferiority margin.“In ctDNA-negative patients, de-escalation reduced oxaliplatin use (34.8% versus 88.6%) and hospitalizations (8.5% versus 13.2%) but yielded slightly lower RFS than standard management (85.3% versus 88.1%), not meeting the non-inferiority margin.” Tie et al., Nat Med 2025: DYNAMIC-III

Next step

ResearchUsing the amount of ctDNA, not just its presence, to grade risk among positive patients: in DYNAMIC, a higher than median postoperative level carried a recurrence hazard ratio of 10.62.“A higher than median postoperative tumor-derived mutant molecules per milliliter plasma was associated with worse 5-year RFS (HR 10.62; P = 0.005).” Tie et al., Nat Med 2025: DYNAMIC five-year outcomes

DYNAMIC-III CIRCULATE-Japan (GALAXY / VEGA / ALTAIR)

What it would buy

Shown to helpFor stage II colon cancer, about half as many people having chemotherapy, with no loss of recurrence-free or overall survival at five years in one randomised trial. Not shown for stage III.“At a median follow-up of 59.7 months, 5-year RFS was 88% and 87% with ctDNA-guided and standard management, respectively (difference 1.1%, 95% confidence interval -5.8% to 8.0%), and 5-year overall survival is similar (93.8% versus 93.3%, hazard ratio (HR) 1.05; P = 0.887).” Tie et al., Nat Med 2025: DYNAMIC five-year outcomes

Starting treatment because a ctDNA test turned positive

Some evidence

ctDNA MRD positivity (molecular residual disease after curative treatment)MRD / molecular residual disease testing Signatera

Sold as more precise than shown: A positive ctDNA result is sometimes presented as a reason to treat in any cancer; the randomised survival benefit so far is in bladder cancer with atezolizumab, and treating on a positive result did not help in stage III colon cancer.

Precision means

Difference in disease-free and overall survival when treatment is given on a positive ctDNA result, hazard ratio against placebo or standard care

Today

Disease-free survival hazard ratio 0.64 (median 9.9 against 4.8 months) and overall survival hazard ratio 0.59 (median 32.8 against 21.1 months) with atezolizumab against placebo (250 people with muscle-invasive bladder cancer who turned ctDNA-positive during a year of surveillance after cystectomy, randomised 2:1 in the double-blind IMvigor011 trial (Powles 2025))“The median overall survival was 32.8 months with atezolizumab, as compared with 21.1 months with placebo (hazard ratio for death, 0.59; 95% CI, 0.39 to 0.90; P = 0.01).” Powles et al., NEJM 2025: ctDNA-guided adjuvant atezolizumab in muscle-invasive bladder cancer (IMvigor011)

What limits it

BiologyA positive result identifies people who need something that works; it does not make a treatment work. In DYNAMIC-III, intensified chemotherapy for ctDNA-positive stage III colon cancer gave no benefit over standard treatment.“Escalated therapy did not improve outcomes over standard management (2-year RFS 51% versus 61%).” Tie et al., Nat Med 2025: DYNAMIC-III
SamplingBy the time three-monthly testing turned positive in triple-negative breast cancer, most women already had metastases visible on scans, so treatment would have started against established disease rather than microscopic residue.“Patients had a high rate of metastatic disease on ctDNA detection.” Turner et al., Ann Oncol 2023 (c-TRAK TN)

Next step

In trialsRandomised trials of new treatments, not more of the old ones, for ctDNA-positive disease, which is where the DYNAMIC-III authors point.“exploratory chemotherapy intensification conferred no RFS benefit, suggesting a need for novel strategies in ctDNA-positive disease” Tie et al., Nat Med 2025: DYNAMIC-III

A national platform trial that every ctDNA-positive patient can join ctDNA-triggered escalation in early TNBC

What it would buy

Shown to helpLonger survival in one setting, muscle-invasive bladder cancer treated with atezolizumab, and treatment spared for the 357 people whose tests stayed negative there (88 per cent disease-free at two years). Not shown in stage III colon cancer or triple-negative breast cancer.“Among 357 patients with persistent ctDNA-negative status, disease-free survival was 95% at the end of the 1-year monitoring period and 88% at 2 years.” Powles et al., NEJM 2025 (IMvigor011)

Which cancers a blood test can see: shedding by cancer type and stage

Liquid biopsy (ctDNA)Circulating tumour DNA fraction (and what a negative plasma result means)Cell-free DNA (cfDNA)

Precision means

Share of patients with detectable circulating tumour DNA, per cent, by cancer type and stage

Today

More than 75 per cent in advanced pancreatic, ovarian, colorectal, bladder, gastro-oesophageal, breast, melanoma, liver and head and neck cancers; under 50 per cent in primary brain, kidney, prostate and thyroid cancers; 48 to 73 per cent in localised colorectal, gastro-oesophageal, pancreatic and breast cancer (640 patients with a range of cancers tested by digital PCR (Bettegowda 2014))“We found that ctDNA was detectable in >75% of patients with advanced pancreatic, ovarian, colorectal, bladder, gastroesophageal, breast, melanoma, hepatocellular, and head and neck cancers, but in less than 50% of primary brain, renal, prostate, or thyroid cancers.” Bettegowda et al., Sci Transl Med 2014: detection of circulating tumour DNA in early- and late-stage human malignancies

What limits it

BiologyVery little DNA reaches the blood from each tumour cell that dies. A model fitted to 176 patients with stage I to III lung cancer estimated 0.014 per cent of a dying cell's DNA on average.“we inferred that, on average, 0.014% of a tumor cell's DNA is shed into the bloodstream per cell death” Avanzini et al., Sci Adv 2020
BiologyHow much a plasma result can be trusted depends on how much tumour DNA is in it. Even in metastatic prostate cancer, about a quarter of samples had 2 per cent tumour DNA or less.“Seventy-five point six percent of cfDNA samples had a ctDNA proportion greater than 2% of total cfDNA.” Wyatt et al., JNCI 2017: concordance of ctDNA and matched metastatic tissue biopsy in prostate cancer

Next step

ResearchReading other features of cell-free DNA, such as how it fragments across the genome or its methylation, so detection does not rest on finding mutations. In a case-control study of seven cancers, genome-wide fragmentation patterns detected 57 per cent to more than 99 per cent depending on cancer type, at 98 per cent specificity.“A machine learning model that incorporated genome-wide fragmentation features had sensitivities of detection ranging from 57% to more than 99% among the seven cancer types at 98% specificity” Cristiano et al., Nature 2019: genome-wide cell-free DNA fragmentation in patients with cancer

cfDNA fragmentomics Multi-cancer early detection (MCED)

What it would buy

Plausible, not shownBlood testing that works for people whose cancers shed little DNA, who are now the ones a negative plasma result is most likely to mislead.

Blood-cell mutations mistaken for cancer mutations in plasma tests

Liquid biopsy (ctDNA)Cell-free DNA (cfDNA)Clonal haematopoiesis (CHIP)

Precision means

Share of mutations found in plasma that come from blood-forming cells rather than the tumour, per cent

Today

53.2 per cent of cell-free DNA mutations in people with cancer, and 81.6 per cent in people without, had features of clonal haematopoiesis (124 patients with metastatic cancer and 47 controls, plasma and matched white blood cells sequenced across 508 genes (Razavi 2019))“The vast majority of cfDNA mutations (81.6% in controls and 53.2% in patients with cancer) had features consistent with clonal hematopoiesis.” Razavi et al., Nat Med 2019

What limits it

BiologyThe blood-cell mutations fall in genes that decide treatment. In 69 men with advanced prostate cancer, 7 (10 per cent) had clonal haematopoiesis variants in DNA repair genes used to decide PARP inhibitor eligibility, most often ATM, and these made up almost half of the repair-gene variants detected.“Overall, CHIP variants accounted for almost half of the somatic DNA repair gene variants detected.” Jensen et al., JAMA Oncol 2021: clonal haematopoiesis in DNA repair genes and prostate cancer plasma testing
BiologyClonal haematopoiesis becomes commoner with age, which is also when cancer is commonest.“Participant CHIP variants were exponentially correlated with older age (R2 = 0.82).” Jensen et al., JAMA Oncol 2021

Next step

Available, not routineSequencing a matched white blood cell sample alongside the plasma, so blood-cell mutations can be subtracted. The method exists; the papers that measured the problem recommend it for every clinical plasma test.“Clinical cfDNA testing should include a paired whole-blood control to exclude CHIP variants and avoid misdiagnosis.” Jensen et al., JAMA Oncol 2021

What it would buy

Plausible, not shownFewer people offered a PARP inhibitor, or told they carry a mutation, on the strength of a variant from their blood cells rather than their cancer.“CHIP interference variants could be distinguished from prostate cancer variants using a paired whole-blood control.” Jensen et al., JAMA Oncol 2021

Plasma against tissue for finding treatable mutations at diagnosis

Liquid biopsy (ctDNA)Comprehensive genomic profiling Guardant360 CDx

Precision means

Agreement between plasma and tissue for guideline biomarkers, per cent, and days to a result

Today

Concordance above 98.2 per cent for EGFR, ALK, ROS1 and BRAF, with every plasma-positive result confirmed in tissue; 80 per cent sensitivity of plasma for any guideline biomarker; median 9 days to a plasma result against 15 for tissue (282 people with newly diagnosed metastatic non-small-cell lung cancer, plasma tested with Guardant360 alongside usual tissue genotyping (NILE, Leighl 2019))“For FDA-approved targets (EGFR, ALK, ROS1, BRAF) concordance was >98.2% with 100% positive predictive value for cfDNA versus tissue (34/34 EGFR-, ALK-, or BRAF-positive patients).” Leighl et al., Clin Cancer Res 2019: NILE

What limits it

BiologyPlasma misses fusions and copy number changes more often than point mutations, and a negative plasma result can be a tumour that did not shed. ESMO recommends going back to tissue after a non-informative plasma result.“Reflex tumour testing should be considered following a non-informative ctDNA result, due to false-negative results with ctDNA testing.” Pascual et al., Ann Oncol 2022 (ESMO)
BiologyWhen tumour fraction is high enough, plasma finds what tissue finds: in metastatic prostate cancer samples above 2 per cent tumour DNA, every mutation in the matched biopsy was also in plasma.“In these patients, all somatic mutations identified in matched metastatic tissue biopsies were concurrently present in ctDNA.” Wyatt et al., JNCI 2017

Next step

Available, not routineSending plasma and tissue at the same time rather than in sequence. In NILE, adding plasma to tissue raised the number of patients with a guideline biomarker found from 60 to 89.“Utilizing cfDNA, in addition to tissue, increased detection by 48%, from 60 to 89 patients, including those with negative, not assessed, or insufficient tissue results.” Leighl et al., Clin Cancer Res 2019: NILE

FoundationOne CDx / Liquid CDx

What it would buy

Plausible, not shownA treatable mutation found in more people, and about a week sooner, before first treatment is chosen. That this changes survival has not been tested in a randomised trial OnCo found.

Comprehensive genomic profiling: from an actionable result to a matched treatment

Comprehensive genomic profiling Genomic profiling Cancer variant knowledgebases and molecular tumour boards

Precision means

Share of profiled patients who receive a treatment matched to their tumour's alteration, per cent

Today

Sequencing succeeded in 93.0 per cent of biopsies, 37.6 per cent had an actionable alteration and 17.8 per cent were assigned a matched treatment (5,954 people with treatment-refractory cancers screened at 1,117 sites in NCI-MATCH (Flaherty 2020))“Molecular profiling was successful in 93.0% of specimens. An actionable alteration was found in 37.6%. After applying clinical and molecular exclusion criteria, 17.8% were assigned (26.4% could have been assigned if all subprotocols were available simultaneously).” Flaherty et al., JCO 2020: molecular landscape and actionable alterations in NCI-MATCH

What limits it

Cost and accessA match needs a drug or a trial to be open at that moment. In the MSK-IMPACT programme of more than 10,000 patients, 11 per cent were enrolled in genomically matched trials; in the TARGET plasma programme, 41 of 100 patients had an actionable mutation and 11 received a matched therapy.“Patients were enrolled on genomically matched clinical trials at a rate of 11%.” Zehir et al., Nat Med 2017: MSK-IMPACT, prospective clinical sequencing of 10,000 patients
BiologyHow often an alteration is actionable depends heavily on the cancer: more than 35 per cent for urothelial cancers and under 6 per cent for pancreatic and small-cell lung cancer in NCI-MATCH.“Actionability rates differed among histologies (eg, > 35% for urothelial cancers and < 6% for pancreatic and small-cell lung cancer).” Flaherty et al., JCO 2020 (NCI-MATCH)
BiologyMost tumours carry more than the one alteration that is matched: 71.3 per cent of NCI-MATCH specimens had resistance-conferring mutations.“Multiple actionable or resistance-conferring tumor mutations were seen in 11.9% and 71.3% of specimens, respectively.” Flaherty et al., JCO 2020 (NCI-MATCH)

Next step

Available, not routineProfiling from plasma to match people to early-phase trials quickly, as the TARGET programme did with a 641-gene panel.“When a 2.5% variant allele frequency (VAF) threshold was applied, actionable mutations were identified in 41 of 100 patients, and 11 of these patients received a matched therapy.” Rothwell et al., Nat Med 2019: the TARGET study

Liquid biopsy (ctDNA)NCI-MATCH (EAY131)

What it would buy

Plausible, not shownFaster profiling widens the window in which a person can join a matched trial. The ceiling is set by how many people have a drug that works for their alteration: of people with advanced cancer in the United States, an estimated 8.33 per cent were eligible for a genome-targeted drug approved by 2018 and 4.90 per cent were estimated to benefit. Profiling matters a great deal to that minority, and it is a minority.“The percentage of patients with cancer estimated to benefit from genome-targeted therapy in 2006 was 0.70% (95% CI, 0.68%-0.72%), and in 2018, it had increased to 4.90% (95% CI, 4.85%-4.95%).” Marquart, Chen and Prasad, JAMA Oncol 2018: percentage of US patients with cancer who benefit from genome-driven oncology

What matching a drug to a tumour's mutation has been shown to buy

Insufficient evidence

Comprehensive genomic profiling Tumour-agnostic (tissue-agnostic) approval

Sold as more precise than shown: A profiling report that lists every actionable alteration can suggest more benefit than exists: the one randomised test of matching drugs off-label, SHIVA, found no gain in progression-free survival.

Precision means

Median progression-free survival with a drug chosen by molecular profile against the doctor's choice, months, and hazard ratio

Today

2.3 months against 2.0 months (hazard ratio 0.88, p 0.41): no difference (195 heavily pretreated patients with any metastatic solid tumour randomised in SHIVA to a targeted drug matched to their tumour but used outside its licence, or to the doctor's choice (Le Tourneau 2015))“Median progression-free survival was 2·3 months (95% CI 1·7-3·8) in the experimental group versus 2·0 months (1·8-2·1) in the control group (hazard ratio 0·88, 95% CI 0·65-1·19, p=0·41).” Le Tourneau et al., Lancet Oncol 2015: SHIVA

What limits it

BiologyA mutation is a reason to try a drug, not proof it will work in that tissue. Of the first 27 NCI-MATCH substudies, each a phase 2 trial of a drug matched to an alteration, 7 met their response target.“which met its signal-seeking objective with 7/27 positive substudies (25.9%)” O'Dwyer et al., Nat Med 2023: the NCI-MATCH trial, lessons for precision oncology
WorkflowThe strongest survival figures for matching come from registries, which are not randomised. In the Know Your Tumor pancreatic cancer registry, the 46 people who received a matched treatment lived longer than the 143 with an actionable alteration who did not (hazard ratio 0.42), and the authors called for prospective testing.“molecularly guided treatments targeting oncogenic drivers and the DNA damage response and repair pathway warrant further prospective evaluation” Pishvaian et al., Lancet Oncol 2020: Know Your Tumor registry

Next step

In trialsTesting matched treatment in randomised trials and platform studies rather than off-label, which is what the SHIVA investigators recommended.“Off-label use of molecularly targeted agents should be discouraged, but enrolment in clinical trials should be encouraged to assess predictive biomarkers of efficacy.” Le Tourneau et al., Lancet Oncol 2015: SHIVA

NCI-MATCH (EAY131)MyPathway

What it would buy

Not establishedFor a drug approved for that alteration in that cancer, the benefit is whatever its own trials showed. For matching across cancers, off-label, the one randomised test found no improvement in progression-free survival.“The use of molecularly targeted agents outside their indications does not improve progression-free survival compared with treatment at physician's choice in heavily pretreated patients with cancer.” Le Tourneau et al., Lancet Oncol 2015: SHIVA

Tumour mutational burden: one cut-off across every cancer

Some evidence

Tumour mutational burden testing TMB-high (tumour mutational burden >= 10 mutations per megabase)Tumour-agnostic (tissue-agnostic) approval

Sold as more precise than shown: A single cut-off of 10 mutations per megabase is used across solid tumours, while high-TMB breast, prostate and brain tumours responded 15.3 per cent of the time in a pooled analysis, and in MyPathway tumours between 10 and 16 responded in 1 of 48.

Precision means

Mutations per megabase of DNA sequenced (mut/Mb) at the cut-off, and how much the same tumour's value varies between test panels

Today

10 mutations per megabase on FoundationOne CDx; 13 per cent of evaluable patients were above it, and 29 per cent of them responded to pembrolizumab against 6 per cent below it (790 evaluable patients with ten previously treated advanced cancer types in the KEYNOTE-158 prospective biomarker analysis (Marabelle 2020), the basis of the tumour-agnostic approval)“Objective responses were observed in 30 (29%; 95% CI 21-39) of 102 patients in the tTMB-high group and 43 (6%; 5-8) of 688 in the non-tTMB-high group.” Marabelle et al., Lancet Oncol 2020: tumour mutational burden and pembrolizumab in KEYNOTE-158

What limits it

WorkflowPanels differ in size, genes and software, so the same tumour can score differently on different tests, and the spread is widest at the high values that decide treatment.“Study results demonstrated that variability within and between panel TMB values increases as the WES TMB values increase.” Merino et al., J Immunother Cancer 2020: Friends of Cancer Research TMB harmonisation, phase I
BiologyA high count does not mean the same thing in every cancer. Pooling over 10,000 tumours, high-TMB breast, prostate and glioma tumours responded to checkpoint inhibitors 15.3 per cent of the time.“In cancer types that showed no relationship between CD8 T-cell levels and neoantigen load, such as breast cancer, prostate cancer, and glioma, TMB-H tumors failed to achieve a 20% ORR (ORR = 15.3%, 95% CI 9.2-23.4, P = 0.95)” McGrail et al., Ann Oncol 2021: high tumour mutation burden fails to predict response across all cancer types
BiologyThe cut-off may be in the wrong place. In MyPathway, 16 of 42 tumours at 16 or more mutations per megabase responded to atezolizumab, against 1 of 48 between 10 and 16.“In 42 patients with F1(CDx) TMB ≥16 mut/Mb, confirmed ORR was 38.1% [16/42; 95% confidence interval (CI), 23.6-54.4], and disease control rate was 61.9% (26/42; 95% CI, 45.6-76.4) versus 2.1% (1/48; 95% CI, 0.1-11.1)” Friedman et al., Cancer Discov 2022: atezolizumab for high-TMB tumours in MyPathway

Next step

ResearchHarmonising how panels report TMB against a whole-exome reference, and setting cut-offs by cancer type rather than one for all.“These harmonization efforts should improve consistency and reliability of panel TMB estimates and aid in clinical decision-making.” Merino et al., J Immunother Cancer 2020

Establishing guidelines to harmonize tumor mutational burden (TMB): in silico assessment of variation in TMB quantification across diagnostic platforms: phase I of the Friends of Cancer Research TMB Harmonization Project Atezolizumab Treatment of Tumors with High Tumor Mutational Burden from MyPathway, a Multicenter, Open-Label, Phase IIa Multiple Basket Study

What it would buy

Plausible, not shownFewer people given immunotherapy on a number that does not predict response in their cancer, and fewer missed because their panel scored them just below a line.

Microsatellite instability and mismatch-repair results: how often a local positive is right

MSI and mismatch-repair testing MSI-high (microsatellite instability by PCR or sequencing)Tumour-agnostic (tissue-agnostic) approval

Precision means

Positive predictive value of a locally reported MSI-high or mismatch-repair-deficient result, per cent

Today

92.1 per cent in 38 patients entering immunotherapy trials and 90.3 per cent in 93 patients from six hospitals; 3 of the 5 tumours that did not respond turned out to be microsatellite stable on retesting (Metastatic colorectal cancer called MSI or dMMR by local laboratories, retested centrally by immunohistochemistry and PCR (Cohen 2019))“The positive predictive value of MSI or dMMR status assessed by local laboratories was therefore 92.1% (95% CI, 78.5%-98.0%).” Cohen et al., JAMA Oncol 2019: primary resistance to checkpoint inhibitors and misdiagnosis of MSI or dMMR status

What limits it

WorkflowImmunohistochemistry and PCR can disagree on the same tumour; in the French cohort a third assay settled two of four such cases.“Testing for MSI with the HT17 assay confirmed the MSI or dMMR status in 2 of 4 cases showing discrepant results between immunohistochemistry and pentaplex polymerase chain reaction (ie, dMMR but microsatellite stable).” Cohen et al., JAMA Oncol 2019

Next step

Available, not routineTesting by both immunohistochemistry and PCR before immunotherapy, as the authors recommend.“Microsatellite instability or dMMR status should be tested routinely using both immunohistochemistry and polymerase chain reaction methods prior to treatment with immune checkpoint inhibitors.” Cohen et al., JAMA Oncol 2019

What it would buy

Plausible, not shownImmunotherapy reserved for people whose tumours are truly mismatch-repair deficient, where it works: in KEYNOTE-158, 30.8 per cent of 321 people with non-colorectal MSI-high cancers responded, with a median response of 47.5 months. In the small Cohen cohort, most primary resistance was explained by a wrong result.“Pembrolizumab demonstrated clinically meaningful and durable benefit, with a high ORR of 30.8%, long median duration of response of 47.5 months” Maio et al., Ann Oncol 2022: KEYNOTE-158 updated analysis in MSI-high or dMMR cancers

Multi-cancer early detection blood tests: stage I sensitivity and positive predictive value

Insufficient evidence

Multi-cancer early detection (MCED)Galleri

Sold as more precise than shown: A headline specificity of 99.5 per cent is accurate and is not the number a person being screened needs: the same validation found 16.8 per cent of stage I cancers, in NHS-Galleri about half of positive results were cancer, and the trial's primary endpoint was not met.

Precision means

Share of stage I cancers detected (sensitivity, per cent) and share of positive results that turn out to be cancer (positive predictive value, per cent)

Today

Stage I sensitivity 16.8 per cent at 99.5 per cent specificity in the case-control validation; in screening, 45.8 to 58.0 per cent of positive results were cancer and the test found 26.7 to 37.2 per cent of all cancers diagnosed in a screening round (4,077 people in the third CCGA substudy (Klein 2021); 142,250 people aged 50 to 77 randomised in NHS-Galleri, three annual rounds (Neal 2026))“sensitivity increased with stage [stage I: 16.8% (14.5% to 19.5%), stage II: 40.4% (36.8% to 44.1%), stage III: 77.0% (73.4% to 80.3%), stage IV: 90.1% (87.5% to 92.2%)]” Klein et al., Ann Oncol 2021: clinical validation of a targeted methylation-based multi-cancer early detection test

What limits it

BiologySensitivity climbs with stage, from 16.8 per cent at stage I to 90.1 per cent at stage IV in the validation set, so the test is least likely to find the cancers where finding them early would matter most.“Overall sensitivity for cancer signal detection was 51.5% (49.6% to 53.3%); sensitivity increased with stage” Klein et al., Ann Oncol 2021
SamplingIn screening, a positive result is cancer about half the time. In the first prospective study, 35 of 92 positives were cancer and false positives took a median of 162 days to resolve.“35 (38%) participants were diagnosed with cancer (true positives) and 57 (62%) had no cancer diagnosis (false positives).” Schrag et al., Lancet 2023: PATHFINDER
BiologyIn NHS-Galleri the test performed as a screening test and the trial did not show the outcome that matters: its primary endpoint, fewer stage III and IV cancers, was not met.“We reported elsewhere that the primary endpoint of a reduction in the incidence of stage III/IV cancer diagnoses in the intervention arm versus control arm was not met.” Neal et al., Nat Med 2026: performance of a multi-cancer early detection test in the NHS-Galleri trial

Next step

ResearchCombining different signals in cell-free DNA to raise early-stage sensitivity. In a case-control study, adding mutation analysis to genome-wide fragmentation patterns detected 91 per cent of patients with cancer; that is a laboratory comparison, not a screening result.“Combining our approach with mutation-based cell-free DNA analyses detected 91% of patients with cancer.” Cristiano et al., Nature 2019

cfDNA fragmentomics NHS-Galleri PATHFINDER 2 CCGA: the Circulating Cell-free Genome Atlas behind the Galleri methylation test

What it would buy

Not establishedIf early-stage sensitivity rose and the trials showed fewer late-stage diagnoses, fewer cancers found too late to cure. Neither has been shown: NHS-Galleri reported positive predictive values of 58.0, 50.4 and 45.8 per cent by round and did not meet its primary endpoint.“positive predictive values were 58.0% (419/722), 50.4% (261/518) and 45.8% (257/561)” Neal et al., Nat Med 2026 (NHS-Galleri)

Marrow residual disease in myeloma and leukaemia: clonoSEQ sequencing

NGS-based MRD (clonoSEQ and molecular MRD)clonoSEQ Multiparameter flow cytometry MRD

Precision means

Lowest frequency of cancer cells detectable among the marrow cells tested (10⁻⁶ is one in a million)

Today

Limit of detection 2.26 × 10⁻⁵ with 500 ng of DNA and 6.77 × 10⁻⁷ with 20 µg; about 1.9 cancer cells in the sample in both cases (FDA De Novo decision summary for the clonoSEQ assay, dilution series from 66 patients with lymphoid cancers including myeloma and acute lymphoblastic leukaemia (DEN170080, 2018))“A probit approach was used to determine the LoD to be 1.903 malignant cells” FDA De Novo decision summary DEN170080: clonoSEQ assay

What limits it

SamplingThe test counts cells, so its depth depends on how much marrow DNA is sent: the same assay reaches about one in 44,000 cells on a small sample and about one in 1.5 million on a large one. A result has to be read with the amount of DNA tested.“The LoD/LoQ by MRD frequency will vary based on the DNA input and the total nucleated cells that are evaluated by the assay.” FDA De Novo decision summary DEN170080: clonoSEQ assay
SamplingA marrow sample comes from one site. The International Myeloma Working Group added imaging to its residual disease categories because disease can persist outside the marrow.“sensitive imaging techniques can be used to detect the presence of residual disease outside of the bone marrow” Kumar et al., Lancet Oncol 2016: International Myeloma Working Group consensus criteria for response and minimal residual disease

Next step

Available, not routinePairing marrow sequencing with imaging to define residual disease inside and outside the marrow, as the 2016 consensus criteria allow.“the International Myeloma Working Group has defined new response categories of minimal residual disease negativity, with or without imaging-based absence of extramedullary disease” Kumar et al., Lancet Oncol 2016 (IMWG)

MRD negativity (myeloma, 10⁻⁵ / 10⁻⁶)Undetectable MRD (uMRD / MRD-negative)

What it would buy

Plausible, not shownA more reliable reading of who is in deep remission. In myeloma, a negative result goes with longer progression-free survival (hazard ratio 0.41) and overall survival (0.57) across 21 studies; that is a prognostic association, and whether stopping or reducing treatment on a negative result is safe is being tested rather than shown.“An MRD-negative status was associated with significantly better PFS overall (hazard ratio [HR], 0.41; 95% CI, 0.36-0.48; P < .001)” Munshi et al., JAMA Oncol 2017: minimal residual disease and survival in multiple myeloma, a meta-analysis

Surgery, imaging and pathology

How small a deposit a scan can see, how close to the tumour a surgeon can cut and know it is clear, and how reliably a slide is read.

Margins and second operations after breast-conserving surgery

Margins after breast-conserving surgery, and second operations Lumpectomy (breast-conserving surgery)Resection margins (R0 / R1 / R2)

Precision means

share of women who have a second operation after breast-conserving surgery, per cent

Today

20.0 per cent had at least one reoperation within three months; adjusted rates across hospital trusts ran from 12.2 to 30.2 per cent (10th to 90th centile) (55,297 women having breast-conserving surgery in 156 English NHS trusts, April 2005 to March 2008, from hospital episode statistics)“One in five women who had breast conserving surgery in England had a reoperation.” Reoperation rates after breast conserving surgery for breast cancer among women in England: retrospective study of hospital episode statistics (BMJ 2012)

What limits it

SamplingThe margin is read by the pathologist days after the operation, so a positive margin is discovered only when a second operation is the remedy.“are identified days after surgery, necessitating a second surgery” Intraoperative Fluorescence Guidance for Breast Cancer Lumpectomy Surgery (NEJM Evidence 2023)
BiologyCancer with an in situ component is harder to clear: 29.5 per cent of English women whose tumour had carcinoma in situ coded had a reoperation, against 18.0 per cent with invasive disease alone.“2803 (29.5%) of the 9504 women with carcinoma in situ had at least one reoperation” Reoperation rates after breast conserving surgery in England (BMJ 2012)
WorkflowMuch of the variation is in the decision rather than the margin: among women whose margins were already clear, re-excision rates ranged from 0 to 70 per cent between surgeons in four US health systems.“For patients with negative margins, reexcision rates varied widely among surgeons (range, 0%-70%; P = .003)” Variability in reexcision following breast conservation surgery (JAMA 2012)
BiologyA clear margin lowers the odds of the cancer returning in the breast, by about half against a positive or close one, but a wider clear margin adds nothing measurable, so precision beyond a clear edge does not buy local control.“increasing the distance for defining negative margins is not significantly associated with reduced odds of LR” The association of surgical margins and local recurrence in women with early-stage invasive breast cancer treated with breast-conserving therapy: a meta-analysis (Ann Surg Oncol 2014)

Next step

Available, not routineFinding the positive margin during the operation, while the cavity is open, rather than days afterwards. Frozen section, cytology, specimen radiography, spectroscopy probes and fluorescent dyes all exist; none is in routine use everywhere, for the reasons the meta-analysis names.“IMA techniques have failed to penetrate routine practice due to limitations, including slow reporting times, technical demands, and logistics.” Diagnostic Accuracy of Intraoperative Techniques for Margin Assessment in Breast Cancer Surgery: A Meta-analysis (Ann Surg 2017)

Intraoperative margin assessment in breast-conserving surgery Pegulicianine

What it would buy

Shown to helpFewer second operations. The clearest gain so far came from agreeing what counts as clear rather than from a device: after the 2014 consensus endorsing "no ink on tumor", surgery after a first lumpectomy fell by 16 per cent in two US cancer registries, with no change in the rate of positive margins.“Surgery after initial lumpectomy declined by 16%” Trends in Reoperation After Initial Lumpectomy for Breast Cancer: Addressing Overtreatment in Surgical Management (JAMA Oncol 2017)

Checking breast margins during the operation: frozen section, specimen X-ray, spectroscopy and cavity shaving

Some evidence

Intraoperative margin assessment in breast-conserving surgery

Precision means

sensitivity and specificity for a positive margin during the operation, against final histology, per cent

Today

Pooled sensitivity and specificity: frozen section 86 and 96 per cent, imprint cytology 91 and 95, intraoperative ultrasound 59 and 81, specimen radiography 53 and 84, optical spectroscopy 85 and 87 (Meta-analysis of 35 studies with raw accuracy data against final permanent-section histology, searched to January 2016)“frozen section = 86%, 96%, 0.96 (n = 9); cytology = 91%, 95%, 0.98 (n = 11); intraoperative ultrasound = 59%, 81%, 0.78 (n = 4); specimen radiography = 53%, 84%, 0.73 (n = 9); optical spectroscopy = 85%, 87%, 0.88 (n = 3)” Diagnostic Accuracy of Intraoperative Techniques for Margin Assessment in Breast Cancer Surgery: A Meta-analysis (Ann Surg 2017)

What limits it

WorkflowThe two most accurate methods, frozen section and cytology, are resource intensive and slow to report, which the meta-analysis gives as the reason they are not used everywhere.“these methods are resource intensive and turnaround times for results have prevented widespread international adoption” Diagnostic Accuracy of Intraoperative Techniques for Margin Assessment (Ann Surg 2017)
PhysicsSpecimen radiography, an X-ray of the removed tissue, found about half of positive margins in the pooled studies: sensitivity 53 per cent.
PhysicsA radiofrequency spectroscopy probe (MarginProbe) found more positive margins than the control arm and also flagged many clear ones: its false-negative rate was 24.8 per cent and its false-positive rate 53.6 per cent, so extra tissue is taken from margins that were clear.“False-negative rates were 24.8 and 66.1 % and false-positive rates were 53.6 and 16.6 % in the device and control arms, respectively.” A randomized prospective study of lumpectomy margin assessment with use of MarginProbe in patients with nonpalpable breast malignancies (Ann Surg Oncol 2014)

Next step

Available, not routineAdding a spectroscopy probe to standard intraoperative imaging, tested in a 596-patient randomised trial in which no intraoperative pathology was allowed.“A total of 19.8 % (59 of 298) of patients in the device arm underwent a reexcision procedure compared with 25.8 % (77 of 298) in the control arm (6 % absolute, 23 % relative reduction).” A randomized prospective study of lumpectomy margin assessment with use of MarginProbe (Ann Surg Oncol 2014)

Pegulicianine A pragmatic trial network for intraoperative margin tools, paid on margin reduction

What it would buy

Shown to helpFewer second operations, shown in randomised trials: the spectroscopy probe lowered re-excision from 25.8 to 19.8 per cent, and routinely shaving an extra layer from the cavity walls, which needs no device, halved it from 21 to 10 per cent in a 235-patient trial. None of these trials measured local recurrence or survival.“as well as a lower rate of second surgery for margin clearance (10% vs. 21%, P=0.02)” A Randomized, Controlled Trial of Cavity Shave Margins in Breast Cancer (NEJM 2015)

Fluorescent dye imaging of the lumpectomy cavity (pegulicianine)

Insufficient evidence

Pegulicianine Fluorescence-guided surgery

Sold as more precise than shown: Pegulicianine is described as reducing second operations, and the pivotal trial showed one avoided in 9 of 62 patients with positive margins in a design that was not built to compare it with usual care.

Precision means

per-margin sensitivity and specificity for cancer left in the lumpectomy cavity, per cent

Today

Sensitivity 49.3 per cent and specificity 85.2 per cent per margin; dye-guided shaves removed cancer left behind after standard lumpectomy in 27 of 357 patients (INSITE, 406 patients with stage 0 to 3 breast cancer given pegulicianine before lumpectomy, randomised 10:1 to dye-guided surgery or control)“On per-margin analysis, pFGS specificity was 85.2%, and sensitivity was 49.3%.” Intraoperative Fluorescence Guidance for Breast Cancer Lumpectomy Surgery (NEJM Evidence 2023)

What limits it

BiologyThe dye lights where tumour enzymes cleave it, and it missed about half the margins that held cancer; the trial did not reach its own sensitivity threshold.“did not meet the prespecified threshold for sensitivity” INSITE (NEJM Evidence 2023)
SamplingThe 10:1 randomisation was there to stop surgeons taking smaller lumpectomies in anticipation of the dye, not to compare it with usual care, so the trial cannot say how much it lowers the re-excision rate.“thus randomization was not designed to provide a control group for evaluating device performance” INSITE (NEJM Evidence 2023)

Next step

Not yet written.

What it would buy

Plausible, not shownSome second operations avoided: among patients whose margins were positive, the dye-guided shaves spared a second operation in 9 of 62. Pegulicianine administration was stopped for adverse events in six patients. No comparison of re-excision, recurrence or survival against usual care has been reported.“Second surgeries were avoided by pFGS in 9 of 62 patients with positive margins.” INSITE (NEJM Evidence 2023)

5-aminolevulinic acid fluorescence for high-grade glioma surgery

Some evidence

5-Aminolevulinic acid (oral, for glioma surgery)Fluorescence-guided surgery

Precision means

share of patients with no contrast-enhancing tumour left on MRI within 72 hours of surgery, per cent

Today

65 per cent with 5-aminolevulinic acid against 36 per cent under white light (Randomised phase 3 trial, 270 patients with suspected malignant glioma judged completely resectable, full-analysis population at interim (Stummer 2006))“Contrast-enhancing tumour was resected completely in 90 (65%) of 139 patients assigned 5-aminolevulinic acid compared with 47 (36%) of 131 assigned white light” Fluorescence-guided surgery with 5-aminolevulinic acid for resection of malignant glioma: a randomised controlled multicentre phase III trial (Lancet Oncol 2006)

What limits it

BiologyThe trial enrolled only tumours a surgeon judged completely removable, so its result does not reach tumours that sit in or against parts of the brain that control speech, movement or vision.“with suspected malignant glioma amenable to complete resection of contrast-enhancing tumour” Stummer et al. (Lancet Oncol 2006)
PhysicsThe target is the contrast-enhancing tumour seen on MRI; glioma cells beyond the enhancing edge are not what the primary endpoint measured.

Next step

Not yet written.

What it would buy

Shown to helpLonger before the tumour grows back: six-month progression-free survival was 41.0 per cent against 21.1 per cent, with no difference in serious adverse events in the week after surgery. The abstract reports no overall survival result, so a survival gain from the dye is not shown by this trial.“Patients allocated 5-aminolevulinic acid had higher 6-month progression free survival than did those allocated white light” Stummer et al. (Lancet Oncol 2006)

Sentinel node biopsy in clinically node-negative breast cancer

Sentinel node biopsy Sentinel lymph node biopsy

Precision means

false-negative rate: share of women with cancer in the armpit whose sentinel nodes were clear, per cent

Today

9.8 per cent (75 of 766); overall accuracy 97.1 per cent (NSABP B-32, women with clinically node-negative invasive breast cancer who had sentinel node biopsy followed by full axillary dissection)“with a false-negative rate of 9.8% (75 of 766; 95% CI 7.8-12.2)” Technical outcomes of sentinel-lymph-node resection and conventional axillary-lymph-node dissection in patients with clinically node-negative breast cancer: results from the NSABP B-32 randomised phase III trial (Lancet Oncol 2007)

What limits it

SamplingThe false-negative rate rises when only one node is taken and after an excisional biopsy, and depends on where the tumour sits.“Removal of more than one SLN and avoidance of excisional biopsy are important variables in reducing the false-negative rate.” NSABP B-32 technical outcomes (Lancet Oncol 2007)

Next step

Available, not routineFor small tumours the trials moved the other way, from staging the armpit more precisely to not operating on it at all: in INSEMA, 4,858 women with T1 or T2 tumours, omitting axillary surgery was non-inferior to sentinel node biopsy for invasive disease-free survival, with less lymphoedema and arm pain, and axillary recurrence of 1.0 against 0.3 per cent.“omission of surgical axillary staging was noninferior to sentinel-lymph-node biopsy after a median follow-up of 6 years” Axillary Surgery in Breast Cancer: Primary Results of the INSEMA Trial (NEJM 2025)

INSEMA (Intergroup Sentinel Mamma)Doing less to the armpit

What it would buy

Not establishedMissing about one positive armpit in ten did not show up in survival: with sentinel node biopsy alone, overall survival, disease-free survival and regional recurrence were statistically equivalent to full dissection at about eight years. A lower false-negative rate in this setting has not been shown to change outcomes.“Overall survival, disease-free survival, and regional control were statistically equivalent between groups.” Sentinel-lymph-node resection compared with conventional axillary-lymph-node dissection in clinically node-negative patients with breast cancer: overall survival findings from the NSABP B-32 randomised phase 3 trial (Lancet Oncol 2010)

Staging the armpit after chemotherapy given before surgery: sentinel node and targeted axillary dissection

Targeted axillary dissection Sentinel node biopsy

Precision means

false-negative rate of axillary staging after neoadjuvant chemotherapy in women who started node-positive, per cent

Today

12.6 per cent with two or more sentinel nodes examined (ACOSOG Z1071); 14.2 per cent in women whose nodes became clinically negative (SENTINA) (ACOSOG Z1071, 525 women with biopsy-proven node-positive breast cancer who had sentinel node surgery and full dissection after chemotherapy, 136 institutions)“resulting in an FNR of 12.6% (90% Bayesian credible interval, 9.85%-16.05%)” Sentinel lymph node surgery after neoadjuvant chemotherapy in patients with node-positive breast cancer: the ACOSOG Z1071 (Alliance) clinical trial (JAMA 2013)

What limits it

SamplingAfter chemotherapy the sentinel node is found less often and misses more when few nodes are taken: in SENTINA the detection rate was 80.1 per cent and the false-negative rate 24.3 per cent when one node was removed.“The false-negative rate was 24.3% (17 of 70) for women who had one node removed” Sentinel-lymph-node biopsy in patients with breast cancer before and after neoadjuvant chemotherapy (SENTINA): a prospective, multicentre cohort study (Lancet Oncol 2013)
SamplingThe node known to hold cancer before treatment was not retrieved as a sentinel node in 23 per cent of patients, including six whose sentinel nodes were clear but whose clipped node still held cancer.“The clipped node was not retrieved as an SLN in 23% (31 of 134) of patients” Improved Axillary Evaluation Following Neoadjuvant Therapy for Patients With Node-Positive Breast Cancer Using Selective Evaluation of Clipped Nodes (JCO 2016)

Next step

Available, not routineClipping the biopsied node before chemotherapy and removing that exact node afterwards with the sentinel nodes (targeted axillary dissection).“TAD followed by ALND was performed in 85 patients, with an FNR of 2.0% (1 of 50; 95% CI, 0.05 to 10.7).” Implementation of Targeted Axillary Dissection (JCO 2016)

Targeted axillary dissection (MD Anderson prospective study)ACOSOG Z1071 (Alliance)

What it would buy

Plausible, not shownA more accurate answer to whether full axillary dissection, and the lymphoedema risk that comes with it, can be avoided after a good response to chemotherapy. The studies here measured false-negative rates, not axillary recurrence or survival, so the outcome gain is not yet shown.

Robot-assisted cancer surgery

Tested, no benefit

Robotic & minimally invasive surgery Surgical robots: da Vinci, Hugo, Versius and single-port systems

Sold as more precise than shown: Robotic surgery is described in terms of precision, and in randomised trials of prostate and rectal cancer surgery that precision did not give better function, margins or cancer control than the comparator, while in early cervical cancer the minimally invasive approach did worse.

Precision means

difference from open or laparoscopic surgery in function, margins and cancer control, in randomised trials

Today

Urinary and sexual function did not differ at 6 or 12 weeks; positive margins were 15 per cent robotic against 10 per cent open, not significantly different (Randomised phase 3 trial, 326 men having radical prostatectomy for localised prostate cancer in Brisbane, early outcomes (Yaxley 2016))“These two techniques yield similar functional outcomes at 12 weeks.” Robot-assisted laparoscopic prostatectomy versus open radical retropubic prostatectomy: early outcomes from a randomised controlled phase 3 study (Lancet 2016)

What limits it

WorkflowIn rectal cancer the robot did not lower conversion to open surgery or the rate of cancer at the circumferential margin (5.1 against 6.3 per cent), with surgeons of varying robotic experience.“robotic-assisted laparoscopic surgery, when performed by surgeons with varying experience with robotic surgery, does not confer an advantage in rectal cancer resection” Effect of Robotic-Assisted vs Conventional Laparoscopic Surgery on Risk of Conversion to Open Laparotomy Among Patients Undergoing Resection for Rectal Cancer: The ROLARR Randomized Clinical Trial (JAMA 2017)
WorkflowIn early cervical cancer the minimally invasive approach, 84.4 per cent laparoscopic and 15.6 per cent robotic, gave lower disease-free and overall survival than open radical hysterectomy.“minimally invasive radical hysterectomy was associated with lower rates of disease-free survival and overall survival than open abdominal radical hysterectomy among women with early-stage cervical cancer” Minimally Invasive versus Abdominal Radical Hysterectomy for Cervical Cancer (NEJM 2018)

Next step

Not yet written.

What it would buy

Shown to helpWhat randomised trials show the robot buying is recovery rather than cancer control: in iROC, 338 UK patients having bladder removal, robotic surgery gave a median of 82 against 80 days alive and out of hospital in 90 days, with fewer clots and wound complications and no difference in recurrence or mortality at a median of 18.4 months. The authors called the clinical importance uncertain.“However, the clinical importance of these findings remains uncertain.” Effect of Robot-Assisted Radical Cystectomy With Intracorporeal Urinary Diversion vs Open Radical Cystectomy on 90-Day Morbidity and Mortality Among Patients With Bladder Cancer: A Randomized Clinical Trial (JAMA 2022)

PSMA PET: detection at biochemical recurrence by PSA level, and lymph node sensitivity at staging

PSMA PET

Precision means

share of scans that find the cancer at biochemical recurrence, by PSA level in ng/ml, per cent

Today

33, 45, 59, 75 and 95 per cent of scans positive at PSA 0 to 0.19, 0.2 to 0.49, 0.5 to 0.99, 1 to 1.99, and 2 ng/ml or more (Meta-analysis of 37 articles and 4,790 patients having gallium-68 PSMA PET, biochemical recurrence after prostatectomy or radiotherapy (Perera 2020))“For PSA categories 0-0.19, 0.2-0.49, 0.5-0.99, 1-1.99, and ≥2ng/ml, the percentages of positive scans were 33%, 45%, 59%, 75%, and 95%, respectively.” Gallium-68 Prostate-specific Membrane Antigen Positron Emission Tomography in Advanced Prostate Cancer: Updated Diagnostic Utility, Sensitivity, Specificity, and Distribution of Prostate-specific Membrane Antigen-avid Lesions: A Systematic Review and Meta-analysis (Eur Urol 2020)

What limits it

PhysicsAt first staging PSMA PET misses most small nodal deposits: before prostatectomy its per-patient sensitivity for pelvic nodes was 0.40, with specificity 0.95, in a 277-man phase 3 imaging trial.“the sensitivity and specificity of 68Ga-PSMA-11 PET were 0.40 and 0.95, respectively” Diagnostic Accuracy of 68Ga-PSMA-11 PET for Pelvic Nodal Metastasis Detection Prior to Radical Prostatectomy and Pelvic Lymph Node Dissection: A Multicenter Prospective Phase 3 Imaging Trial (JAMA Oncol 2021)
PhysicsSensitivity depends on the size of the deposit: in 583 men it was 26.3 per cent overall and rose with deposit size, and adjusting for a 5.5 mm deposit size cut-off tripled it to 90.2 per cent.“PSMA PET/CT showed a better sensitivity as LNM tumor deposit size increased” Staging Accuracy and Prognostic Value of Prostate-Specific Membrane Antigen PET/CT Strongly Depends on Lymph Node Tumor Burden (J Clin Med 2024)

Next step

Available, not routineReplacing CT and bone scan with PSMA PET for staging high-risk disease before curative treatment, as proPSMA tested: accuracy 92 against 65 per cent, with less radiation.“PSMA PET-CT is a suitable replacement for conventional imaging, providing superior accuracy, to the combined findings of CT and bone scanning.” Prostate-specific membrane antigen PET-CT in patients with high-risk prostate cancer before curative-intent surgery or radiotherapy (proPSMA): a prospective, randomised, multicentre study (Lancet 2020)

proPSMA PSMA PET → PSMA radioligand therapy PSMA-PET-guided metastasis-directed therapy as a curative strategy in oligorecurrent prostate cancer

What it would buy

Not establishedA changed plan more often: management changed after 28 per cent of first-line PSMA scans against 15 per cent after conventional imaging. Whether finding disease earlier and treating it lengthens life has not been shown; the trials here measured accuracy and management change.“First-line conventional imaging conferred management change less frequently” proPSMA (Lancet 2020)

Low-dose CT lung screening: nodules, size thresholds and false positives

Low-dose CT lung screening Lung-RADS and nodule reporting Pulmonary nodule

Precision means

share of positive screening scans that turn out not to be cancer, per cent

Today

24.2 per cent of low-dose CT screens were positive over three rounds, and 96.4 per cent of those positives were false (NLST, 53,454 people at high risk of lung cancer at 33 US centres, three annual screens)“A total of 96.4% of the positive screening results in the low-dose CT group and 94.5% in the radiography group were false positive results.” Reduced lung-cancer mortality with low-dose computed tomographic screening (NEJM 2011)

What limits it

BiologyA size threshold trades false alarms for missed cancers: applied retrospectively to NLST, Lung-RADS lowered the false-positive rate after baseline from 21.8 to 5.3 per cent and sensitivity from 93.8 to 78.6 per cent.“Lung-RADS may substantially reduce the false-positive result rate; however, sensitivity is also decreased.” Performance of Lung-RADS in the National Lung Screening Trial: a retrospective assessment (Ann Intern Med 2015)

Next step

Available, not routineMeasuring nodule volume and growth between scans rather than diameter alone, as NELSON did: 9.2 per cent of screened men had an additional scan for an indeterminate result and 2.1 per cent were referred for a suspicious nodule.“The overall referral rate for suspicious nodules was 2.1%.” Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial (NEJM 2020)

NLST & NELSON (low-dose CT screening)Optellum Virtual Nodule Clinic AI malignancy scores to end repeat scans and biopsies for benign lung nodules

What it would buy

Plausible, not shownFewer people sent for further scans, biopsies and operations for nodules that are not cancer, while keeping the lung cancer deaths screening prevents: in NELSON the 10-year rate ratio for lung cancer death was 0.76 in screened men.“There were low rates of follow-up procedures for results suggestive of lung cancer.” NELSON (NEJM 2020)

PD-L1 scoring: how often pathologists agree

PD-L1 expression testing (22C3, SP142, SP263)PD-L1 TPS (tumour proportion score)PD-L1 IC score (immune-cell score, SP142)PD-L1 CPS (combined positive score)

Precision means

agreement between pathologists scoring the same slides, intraclass correlation coefficient from 0 to 1

Today

0.832 to 0.882 for scoring tumour cells; 0.172 to 0.229 for scoring immune cells (13 pathologists scoring 90 non-small cell lung cancers stained with four PD-L1 assays (Rimm 2017))“the concordance between pathologists' scoring for PD-L1 expression in tumor cells ranged from ICCs of 0.832 (95% CI, 0.820-0.844) to 0.882 (95% CI, 0.873-0.891) for each assay, while the ICCs from immune cells for each assay ranged from 0.172 (95% CI, 0.156-0.189) to 0.229 (95% CI, 0.211-0.248)” A Prospective, Multi-institutional, Pathologist-Based Assessment of 4 Immunohistochemistry Assays for PD-L1 Expression in Non-Small Cell Lung Cancer (JAMA Oncol 2017)

What limits it

WorkflowThe assays are not interchangeable: the SP142 assay stained less PD-L1 in both tumour and immune cells than the others on the same tumours.“The assay using the SP142 antibody is an outlier that detected significantly less PD-L1 expression in tumor cells and immune cells.” Rimm et al. (JAMA Oncol 2017)
BiologyPD-L1 varies across a tumour, so a biopsy is a sample of a patchy signal as well as a slide to be read.“limited by factors such as intraobserver/interobserver bias and intratumor heterogeneity” Clinical Validation of Artificial Intelligence-Powered PD-L1 Tumor Proportion Score Interpretation for Immune Checkpoint Inhibitor Response Prediction in Non-Small Cell Lung Cancer (JCO Precis Oncol 2024)

Next step

ResearchAI-assisted scoring of whole slides. In a ring study of 21 pathologists at four institutions scoring the combined positive score in triple-negative breast cancer, agreement rose from 0.618 to 0.931 with AI assistance.“the intraclass correlation coefficient (ICC) value was increased from 0.618” Artificial intelligence enhances whole-slide interpretation of PD-L1 CPS in triple-negative breast cancer: A multi-institutional ring study (Histopathology 2024)

AI scoring of biomarkers and grade on pathology slides One digital PD-L1 scale that maps across all the competing assays Harmonise PD-L1 testing for triple-negative breast cancer around one scored assay, with external quality assurance

What it would buy

Not establishedThe same answer about eligibility for checkpoint inhibitors whichever pathologist reads the slide. In an external lung cancer cohort of 430 slides an AI score predicted progression-free survival about as well as pathologists did, which shows agreement, not that patients do better.“PD-L1 TPS assessed by AI analyzer correlates with that of pathologists, with clinical performance also being comparable when referenced to PFS.” AI-powered PD-L1 TPS interpretation in NSCLC (JCO Precis Oncol 2024)

HER2-low scoring: telling 0 from 1+

HER2-low and HER2-ultralow HER2-low (IHC 1+ or IHC 2+/ISH-negative)

Precision means

agreement between pathologists on HER2 immunohistochemistry 0 versus 1+, per cent of cases scored the same

Today

26 per cent concordance between 0 and 1+, against 58 per cent between 2+ and 3+ (18 pathologists reading scanned slides of 170 breast cancer biopsies on the four-point scale (Fernandez 2022))“there was only 26% concordance between 0 and 1+ compared with 58% concordance between 2+ and 3+” Examination of Low ERBB2 Protein Expression in Breast Cancer Tissue (JAMA Oncol 2022)

What limits it

WorkflowLaboratories disagree too: in College of American Pathologists proficiency surveys of about 1,400 laboratories, 19 per cent of cases reached 70 per cent concordance or less on 0 against 1+.“19% of cases read by the laboratories generate results with less than or equal to 70% concordance for IHC ERBB2 score 0 vs 1+” Fernandez et al. (JAMA Oncol 2022)
BiologyDisagreement clusters in small biopsies, slides with few tumour cells and tumours that stain unevenly.“Discordance was also associated with lower tumor cell counts, biopsy samples, and heterogeneous staining patterns.” Artificial Intelligence (AI)-Assisted Human Epidermal Growth Factor Receptor 2 (HER2) Scoring in Breast Cancer: Diagnostic Agreement and Understanding Discordance (Lab Invest 2026)

Next step

Available, not routineAI-assisted HER2 scoring. In a reader study of 201 cases, agreement among three pathologists on 1+ tumours rose from 26.5 to 70.7 per cent with AI assistance; a commercially available system raised overall multirater agreement from 79.8 to 88.6 per cent across 853 slides in a second study.“where the concordance significantly increased from 46.2 to 68.4% and from 26.5 to 70.7%, respectively.” Augmented interpretation of HER2, ER, and PR in breast cancer by artificial intelligence analyzer: enhancing interobserver agreement through a reader study of 201 cases (Breast Cancer Res 2024)

AI scoring of biomarkers and grade on pathology slides AI quantification of HER2-low and HER2-ultralow Calibrated reference slides so every lab scores HER2-low the same way

What it would buy

Plausible, not shownFewer people wrongly scored 0 and so not offered trastuzumab deruxtecan, which lengthened both progression-free and overall survival against chemotherapy in HER2-low disease in DESTINY-Breast04. Whether AI scoring changes who is treated, or how they do, has not been tested.“trastuzumab deruxtecan resulted in significantly longer progression-free and overall survival than the physician's choice of chemotherapy” Trastuzumab Deruxtecan in Previously Treated HER2-Low Advanced Breast Cancer (NEJM 2022)

AI help to find prostate cancer on biopsy slides

Paige Prostate Detect Digital pathology & AI

Precision means

pathologists' sensitivity for cancer on prostate needle biopsy slides, per cent

Today

Average sensitivity 74 per cent unaided and 90 per cent with the AI system, with no significant change in specificity (97 per cent unaided) (Reader study: three general pathologists reading 304 biopsy whole-slide images, then the same slides about four weeks later with Paige Prostate Alpha, against a genitourinary pathology reference (Raciti 2020))“With Paige Prostate Alpha, the average sensitivity for pathologists significantly increased to 90% with no statistically significant change in specificity.” Novel artificial intelligence system increases the detection of prostate cancer in whole slide images of core needle biopsies (Mod Pathol 2020)

What limits it

WorkflowThe study was a reader exercise, not practice: three pathologists read 304 slides in eight hours, and the authors say it is not known how such systems change diagnosis in routine work.“However, the impact of these artificial intelligence systems on pathologic diagnosis is not known.” Raciti et al. (Mod Pathol 2020)
Cost and accessAny slide AI needs the glass slides scanned first, so it reaches only laboratories with a digital pathology service.

Next step

Not yet written.

What it would buy

Not establishedFewer missed cancers on biopsy. The extra detections were mostly smaller, lower-grade tumours, where finding more can mean more overdiagnosis as well as fewer misses, so a benefit to patients is not established.“With Paige Prostate Alpha, pathologists more often correctly classified smaller, lower grade tumors” Raciti et al. (Mod Pathol 2020)

Pancreatic cancer margins: how the specimen is examined decides the answer

R0 and R1 margins in pancreatic cancer: the 1 mm rule and standardised specimen reporting Resection margins (R0 / R1 / R2)

Precision means

share of pancreatic cancer resections reported as R1 (cancer within 1 mm of a margin), per cent

Today

76 per cent with a standardised examination protocol, against 14 per cent in the same centre before it (111 consecutive pancreatic head resections for ductal adenocarcinoma, 2005 to 2006, against 188 from 2002 to 2004 examined without a standardised protocol, single centre (Esposito 2008))“The R1 rate for resections carried out with the standardized protocol was 76%.” Most pancreatic cancer resections are R1 resections (Ann Surg Oncol 2008)

What limits it

SamplingThe more of the circumferential margin is sampled, the more often cancer is found at it.“Sampling of the circumferential RM was more extensive in the SP series and correlated with RM status.” Redefining the R1 resection in pancreatic cancer (Br J Surg 2006)
WorkflowThere was no widely accepted standard for examining these specimens when the studies were done, so R1 rates from different centres were not comparable.“there is currently no widely accepted standardized protocol for pathological examination, especially with respect to resection margins” Esposito et al. (Ann Surg Oncol 2008)

Next step

Available, not routineA standardised protocol for every specimen: multicolour inking of each margin, axial slicing and extensive sampling, with R1 defined as cancer within 1 mm.“Standardized examination influences the reporting of RM status.” Verbeke et al. (Br J Surg 2006)

What it would buy

Plausible, not shownA margin result that means something for the patient: under the standardised protocol, survival tracked margin status, and under the old one it did not. Whether acting on the more accurate result changes outcomes has not been tested.“Survival correlated with RM status for the entire SP series (P < 0.001), but not for the NSP series.” Verbeke et al. (Br J Surg 2006)

Radiotherapy delivery

How many millimetres of healthy tissue are added around the target because the tumour moves or the set-up varies, and how much dose the organs nearby receive.

The safety margin around the target

IMRT / IGRT (modern external beam)GTV, CTV and PTV (target volumes)

Precision means

Margin added around the clinical target volume to make the planning target volume, in mm, computed from the spread of systematic (Σ) and random (σ) geometric errors

Today

Margin = 2.5Σ + 0.7σ, to give at least 95 per cent of the dose to the whole target in 90 per cent of patients (Analytical model of set-up and preparation errors across a population of patients; the recipe most departments build their margins from)“For a sigma(p) of 3.2 mm, this recipe can be simplified to 2.5 Sigma + 0.7 sigma'.” van Herk et al., Int J Radiat Oncol Biol Phys 2000

What limits it

PhysicsThe recipe leaves out rotations and changes in the shape of the target, so the margin it gives is a floor, not a guarantee.“Because this margin excludes rotational errors and shape deviations, it must be considered as a lower limit for safe radiotherapy.” van Herk et al. 2000
WorkflowErrors made once, in preparing the plan, are repeated at every session and cost far more margin than errors that vary from day to day.“Large execution (random) variations lead to CTV underdosage for a large number of patients, while the same level of preparation (systematic) errors leads to a much larger underdosage for some of the patients.” van Herk et al. 2000

Next step

In trialsDaily imaging on the treatment couch, followed by re-planning to that day's anatomy (online adaptive radiotherapy), so that the margin covers only what is left after correction.“Quantitative evidence is strongest for dosimetric recovery, organ-at-risk sparing, workflow feasibility, and selected toxicity reduction; evidence for survival improvement remains limited and heterogeneous.” Srinivasan and Njeh, Cancers 2026 (narrative review)

Adaptive radiotherapy (online replanning)MR-guided adaptive radiotherapy

What it would buy

Plausible, not shownA smaller margin means less healthy tissue in the high-dose region. Where a margin reduction has been tested on its own terms, in prostate stereotactic radiotherapy (MIRAGE) and brain radiosurgery, it reduced or did not increase side effects; neither trial measured survival.Kishan et al., MIRAGE, JAMA Oncology 2023

Clinicians drawing the target differently

Variation between clinicians in drawing the radiotherapy target GTV, CTV and PTV (target volumes)

Precision means

Agreement between the outlines different clinicians draw for the same patient (overlap indices and surface distances, in mm), and what the difference does to dose

Today

Measured in 119 studies across all major tumour sites with metrics too varied to pool; 21 per cent reported what the variation did to the dose (Systematic review of studies published 2000 to 2014 with more than two observers; median 7 observers)“Statistical analysis of results was reported in 68% (n=81) and dosimetric consequences in 21% (n=25) of studies.” Vinod et al., Radiotherapy and Oncology 2016

What limits it

SamplingStudies use different metrics and analyses, so there is no single figure for how much clinicians disagree.“There was considerable heterogeneity in the use of metrics and methods of analysis.” Vinod et al. 2016
WorkflowA margin added for set-up error does not correct a wrongly drawn outline: the same outline is used at every session.

Next step

Available, not routineAI auto-contouring, reviewed and edited by a clinician, which NICE allowed into NHS use in 2023 while evidence is collected.“AI technologies must be used with healthcare professional review of contours.” NICE HTG695 (2023), recommendation 1.1

AI auto-contouring and adaptive planning

What it would buy

Not establishedMore consistent outlines would make the dose a patient receives depend less on which clinician drew it. Whether that changes control or side effects has rarely been measured, because most studies stop at the outline.Vinod et al. 2016

AI auto-contouring for radiotherapy planning

Some evidence

AI auto-contouring and adaptive planning

Sold as more precise than shown: AI contouring is sold as making radiotherapy more accurate; the evidence NICE reviewed shows outlines of similar quality to manual ones, produced faster, with every outline still needing professional review.

Precision means

How closely AI-drawn outlines match expert ones, how many need editing, and the time to an approved outline

Today

Contours of similar quality to manual contouring, with most structures needing only minor edits, and quicker even including review time (NICE early value assessment of nine AI contouring technologies for NHS external beam radiotherapy, September 2023)“Clinical evidence suggests that AI technologies generally produce similar quality contours as manual contouring, with most structures needing only minor edits.” NICE HTG695 (2023)

What limits it

WorkflowEvery AI outline must be reviewed and edited by a trained professional before it is used.“AI technologies must be used with healthcare professional review of contours.” NICE HTG695 (2023)
SamplingNICE listed what is not yet known, including what AI contouring does to dose and the errors and adverse events it causes.“contouring errors and adverse events associated with AI autocontouring” NICE HTG695 (2023), recommendation 1.4

Next step

Available, not routineThree years of NHS evidence generation, after which NICE decides whether the technologies can be adopted routinely.NICE HTG695 (2023), recommendation 1.3

What it would buy

Not establishedFaster planning and more consistent outlines, which NICE judged could free clinicians' time. A benefit to patients in control or side effects has not been shown.“The evidence suggests that AI autocontouring is quicker than manual contouring even when including time for healthcare professional review and edits.” NICE HTG695 (2023)

Imaging the prostate before every session

IMRT / IGRT (modern external beam)Fiducial markers

Precision means

How often the target's position is checked by imaging and corrected before the beam is switched on

Today

Daily against weekly image guidance: late rectal toxicity lower with daily imaging (hazard ratio 0.71); recurrence-free survival, the primary endpoint, not different (hazard ratio 0.81) (Phase 3 trial, 470 men with localised prostate cancer at 21 French centres, 70 to 80 Gy, median follow-up 4.1 years)“Late rectal toxicity (grade ≥1) was significantly lower in the daily group” de Crevoisier et al., Int J Radiat Oncol Biol Phys 2018

What limits it

BiologyThe prostate moves between and during sessions with bladder and bowel filling; imaging before the beam does not see movement during it.
SamplingIn the same trial overall survival was worse with daily imaging, unexplained in the abstract, alongside a worse interval free of other cancers.“Overall survival was worse in the daily group than in the weekly group” de Crevoisier et al. 2018

Next step

In trialsTracking the prostate during the beam on a standard linear accelerator (kilovoltage intrafraction monitoring), with sub-millimetre accuracy reported, now in a 10-centre trial.“Submillimeter accuracy of KIM was observed in simulations, phantom experiments, and clinical trials.” Keall et al., Int J Radiat Oncol Biol Phys 2026

Respiratory gating and tumour tracking systems (RPM, ABC, Calypso, Synchrony)SBRT / SABR (stereotactic radiotherapy)

What it would buy

Plausible, not shownFewer rectal side effects with daily imaging in one randomised trial, and better control of PSA in a post hoc analysis; no gain in the primary endpoint.de Crevoisier et al. 2018

Surface-guided radiotherapy

Surface-guided radiotherapy (SGRT)Patient positioning and surface guidance systems (AlignRT, Catalyst, ExacTrac Dynamic)

Precision means

Agreement between the patient's tracked skin surface and the reference position, in mm, and how well the surface stands in for the target

Today

Sub-millimetre accuracy reported for frameless brain radiosurgery (Institutional evaluations of surface guidance in radiosurgery, summarised by the American Association of Physicists in Medicine Task Group 302 (full text))“They have shown submillimeter accuracy of the system with various treatment parameters” Al-Hallaq et al., AAPM Task Group 302, Medical Physics 2022

What limits it

PhysicsThe cameras see skin, not the tumour, so accuracy falls for deep targets and when the surface does not move with them.“the accuracy for inferring the location of the internal target is degraded when the surface is not a good surrogate for the target, or when the surface is not visible to the cameras.” AAPM Task Group 302 (full text)
PhysicsThe systems have a limited field of view and cannot image very dark skin, which makes the technique less usable for some patients.“SGRT systems have certain technical limitations such as a finite FOV and an inability to image very dark skin tones.” AAPM Task Group 302 (full text)

Next step

Available, not routineCombining surface tracking with internal X-ray or thermal imaging, as some systems already do, so the surface is checked against anatomy beneath it.“potentially combining thermal imaging with surface imaging as implemented in the Brainlab ExacTrac Dynamic system.” AAPM Task Group 302 (full text)

What it would buy

Shown to helpOpen face masks instead of closed ones in head and neck radiotherapy, with equivalent set-up accuracy and significantly better tolerability, in a 230-patient randomised trial. The benefit shown is experience of treatment, not cancer control.“At treatment completion patient experience metrics, including tolerability, overall experience and distress, were significantly more favourable for both open-mask cohorts compared to closed masks” Nicholson et al., OPEN trial, Radiotherapy and Oncology 2026

Breath-hold radiotherapy for left-sided breast cancer

Deep inspiration breath-hold (DIBH)Respiratory motion management (4D-CT, gating, breath-hold, tracking)

Precision means

Mean heart dose, in Gy

Today

Mean heart dose 1.1 Gy in voluntary breath-hold against 1.8 Gy free breathing; 95 per cent of women had a lower heart dose (UK HeartSpare Stage II: 101 women at 10 UK centres with part of the heart in the field on a free-breathing plan)“Mean cardiac doses (Gy) for free-breathing and VBH techniques, respectively, were: heart 1.8 and 1.1, LAD 12.1 and 5.4, maximum LAD 35.4 and 24.1 (all P<0.001).” Bartlett et al., Clinical Oncology 2017

What limits it

BiologyNo heart dose has been found below which there is no added risk: coronary events rose in proportion to mean heart dose, so the dose left after breath-hold still counts.“Rates of major coronary events increased linearly with the mean dose to the heart by 7.4% per gray (95% confidence interval, 2.9 to 14.5; P<0.001), with no apparent threshold.” Darby et al., New England Journal of Medicine 2013
PhysicsSurface tracking confirms the chest is in position, not where the heart is.“it cannot directly provide information regarding the position of the heart during DIBH.” AAPM Task Group 302 (full text)
BiologyBreath-hold depends on the patient being able to hold a deep breath repeatedly.

Next step

Not yet written.

What it would buy

Plausible, not shownA lower projected risk of heart disease decades later: a 3.4 Gy reduction in mean heart dose, the largest in a review of ten studies, corresponds to a 13.6 per cent smaller projected rise in risk. This is projected from dose, not counted in a trial of breath-hold.“reduction in mean heart dose is equivalent to a 13.6% reduction in the projected increase in risk of heart disease.” Smyth et al., Journal of Medical Radiation Sciences 2015

Tumours that move with breathing: gating, breath-hold and tracking

Respiratory motion management (4D-CT, gating, breath-hold, tracking)Respiratory gating and tumour tracking systems (RPM, ABC, Calypso, Synchrony)Fiducial markers

Precision means

Movement of the target with breathing, in mm, and the residual movement after it is managed

Today

Motion management is recommended when the target moves more than 5 mm and the patient can tolerate it (American Association of Physicists in Medicine Task Group 76 report on respiratory motion in thoracic, abdominal and pelvic tumours)“If target motion is greater than 5 mm, a method of respiratory motion management is available, and if the patient can tolerate the procedure, respiratory motion management technology is appropriate.” Keall et al., AAPM Task Group 76, Medical Physics 2006

What limits it

BiologyBreathing differs from session to session and between patients, so motion has to be measured for each patient.“Included in this guide is the recommendation that tumor motion should be measured (when possible) for each patient for whom respiratory motion is a concern.” AAPM Task Group 76
Cost and accessGating, breath-hold, forced shallow breathing and tracking all add equipment, time and quality assurance.“Respiratory motion management involves further resources, education and the development of and adherence to QA procedures.” AAPM Task Group 76

Next step

In trialsReal-time tracking of the target during the beam on a standard linear accelerator, without extra hardware; tested in prostate stereotactic radiotherapy and now in a 300-patient, 10-centre trial.“KIM is being made available through commercialization with an industry partner, and through a 300-patient, 10-center clinical trial.” Keall et al., Int J Radiat Oncol Biol Phys 2026

SBRT / SABR (stereotactic radiotherapy)MR-guided adaptive radiotherapy

What it would buy

Not establishedSmaller margins on moving targets, and the safe delivery of stereotactic doses to them. Tracking improved dose coverage of the target in trials; a difference in control or side effects has not been reported.“a significant dosimetric impact of KIM deployment for target volume coverage” Keall et al. 2026

MRI-guided adaptive stereotactic radiotherapy for prostate cancer

Some evidence

MR-guided adaptive radiotherapy Adaptive radiotherapy (online replanning)

Sold as more precise than shown: MR-guided adaptive radiotherapy is sometimes presented as more likely to cure; the randomised evidence is for fewer acute side effects after prostate stereotactic radiotherapy, and survival benefit across adaptive radiotherapy is limited.

Precision means

Planning margin around the prostate, in mm

Today

2 mm with MRI guidance against 4 mm with CT guidance, 40 Gy in 5 fractions (MIRAGE phase 3 randomised trial, 156 men, single centre, 2020 to 2021)“Planning margins of 4 mm (CT arm) and 2 mm (MRI arm) were used to deliver 40 Gy in 5 fractions.” Kishan et al., MIRAGE, JAMA Oncology 2023

What limits it

SamplingOne centre, stopped early after an interim analysis, with acute endpoints at 90 days; the arms differed in imaging and margin together, so the trial cannot say which mattered.“Thus, the trial was closed to accrual early.” MIRAGE 2023
Cost and accessMR-linacs are slower per session and cost more than standard machines.
SamplingAcross adaptive radiotherapy the evidence is strongest for dose and side effects, and limited for survival.“evidence for survival improvement remains limited and heterogeneous.” Srinivasan and Njeh, Cancers 2026 (narrative review)

Next step

In trialsLonger follow-up of MIRAGE, and trials testing whether adapting the plan every day changes cancer control rather than only acute side effects.“Longer-term follow-up will confirm whether these notable benefits persist.” MIRAGE 2023

What it would buy

Shown to helpFewer moderate acute urinary and bowel side effects in one randomised trial: grade 2 or worse urinary toxicity 24.4 per cent against 43.4 per cent, and bowel toxicity 0 against 10.5 per cent, within 90 days. Cancer control was not an endpoint.“The incidence of acute grade 2 or greater GU toxic effects was significantly lower with MRI vs CT guidance (24.4% [95% CI, 15.4%-35.4%] vs 43.4% [95% CI, 32.1%-55.3%]” MIRAGE 2023

Radiosurgery for brain metastases: how much margin

Stereotactic radiosurgery (Gamma Knife, CyberKnife, linac SRS)Stereotactic radiosurgery (SRS)

Precision means

Margin added around the visible metastasis on MRI, in mm

Today

1 mm against 3 mm: local recurrence 6.7 per cent at 12 months, with no difference between margins; 5 of 6 biopsy-proven radionecrosis cases in the 3 mm group (Randomised trial by lesion, 49 patients with 80 brain metastases, linear accelerator with image guidance)“However, given the higher potential risk of RN with a 3-mm margin, a 1-mm GTV expansion is more appropriate.” Kirkpatrick et al., Int J Radiat Oncol Biol Phys 2015

What limits it

SamplingA small single trial; the radionecrosis difference (P = .10) was not statistically significant.“Biopsy-proven RN alone was observed in 6 lesions (5 in the 3-mm group, P=.10).” Kirkpatrick et al. 2015
WorkflowThe margin cannot be smaller than the error in drawing the metastasis and placing the patient, so it depends on imaging and immobilisation.

Next step

Not yet written.

What it would buy

Plausible, not shownLess healthy brain in the high-dose region without losing control of the treated tumour. The reduction in radiation damage is suggested, not proven.“The PTV/GTV ratio, volume receiving 12 Gy or more, and minimum dose to PTV were significantly higher in the 3-mm group (all P<.01)” Kirkpatrick et al. 2015

Where a proton beam stops: range uncertainty

Some evidence

Proton therapy Pencil-beam scanning and intensity-modulated proton therapy Proton range uncertainty and the range margin

Sold as more precise than shown: The proton's sharp stopping point is often presented as sparing everything beyond the tumour; in practice planners add several millimetres for range uncertainty, and the randomised lung trial found no reduction in lung inflammation.

Precision means

Uncertainty in the depth at which the beam stops, as a percentage of its range plus a fixed distance in mm

Today

Margin recipes from 2.5 per cent + 1.5 mm to 3.5 per cent + 3 mm; at 3.5 per cent + 1 mm, an 8 mm overshoot for a 20 cm beam in soft tissue (Recipes in use at five US proton centres, reviewed in 2012 (full text))“The 3.5% + 1mm range margin at MGH, for instance, results in a deliberate overshoot of 8mm for a 20 cm range field in soft tissue, which is quite substantial.” Paganetti, Physics in Medicine and Biology 2012 (full text in PubMed Central)

What limits it

PhysicsThe range depends on imaging, set-up, beam delivery and dose calculation, each with its own error.“The range in tissue is associated with considerable uncertainties caused by imaging, patient setup, beam delivery and dose calculation.” Paganetti 2012
BiologyThe biological effect is highest where the beam stops, higher than the standard conversion factor of 1.1 assumes.“The proton RBE increases with increasing LETd and thus with depth in an SOBP from ~1.1 in the entrance region, to ~1.15 in the center, ~1.35 at the distal edge and ~1.7 in the distal fall-off (when averaged over all cell lines, which may not be clinically representative).” Paganetti, Physics in Medicine and Biology 2014

Next step

Available, not routineMonte Carlo dose calculation and measuring the range in the patient, to reduce the margin.“In these cases Monte Carlo techniques might reduce the range uncertainty by several mm.” Paganetti 2012

In vivo dosimetry and patient-specific quality assurance Use LET, RBE and alpha/beta to decide when protons beat IMRT

What it would buy

Not establishedA smaller margin would give back part of the advantage protons are chosen for. In the one randomised comparison with modern photons in lung cancer, protons spared the heart but did not reduce severe lung inflammation or local failure.“No benefit was noted in RP or LF after PSPT.” Liao et al., Journal of Clinical Oncology 2018

FLASH: radiotherapy in a fraction of a second

Insufficient evidence

FLASH radiotherapy FLASH research accelerators (Oriatron, Mobetron FLASH, ProBeam FLASH)

Sold as more precise than shown: FLASH is described as sparing healthy tissue; that has been shown in animals, and the published human proton trial tested feasibility in 10 patients without a comparison arm.

Precision means

Dose rate, in Gy per second: FLASH is at least 40 Gy/s, against about 0.03 Gy/s conventionally

Today

Feasible in 10 people with painful bone metastases, single 8 Gy dose of proton FLASH, side effects consistent with conventional radiotherapy (FAST-01, the first human proton FLASH trial: nonrandomised, 12 treated sites, no comparison arm)“Adverse events were mild and consistent with conventional radiotherapy.” Mascia et al., FAST-01, JAMA Oncology 2023

What limits it

BiologyThe normal-tissue sparing that motivates FLASH was shown in mice.“no complications developed after doses of FLASH below 20 Gy for more than 36 weeks after irradiation” Favaudon et al., Science Translational Medicine 2014
PhysicsThe first proton FLASH treatments used a beam that passes through the patient rather than stopping in the tumour, giving up the proton's stopping point to reach the dose rate.“Bone metastases were treated on a FLASH-enabled (≥40 Gy/sec) proton radiotherapy system using a single-transmission proton beam.” FAST-01 2023
SamplingTen patients, palliative doses to limbs, no randomised comparison.

Next step

In trialsFurther trials in patients to test whether the sparing seen in animals occurs in people.“These findings support the further exploration of FLASH radiotherapy in patients with cancer.” FAST-01 2023

What it would buy

Not establishedIf the effect seen in mice holds in people, fewer side effects at the same tumour dose. That has not been shown in humans.Favaudon et al. 2014

Dose painting: a boost to the tumour visible on MRI

Dose painting and biologically guided radiotherapy

Precision means

Dose to the lesion visible on MRI within the prostate, in Gy, while organ-at-risk limits are kept

Today

Boost up to 95 Gy to the visible lesion on top of 77 Gy to the whole prostate: five-year biochemical disease-free survival 92 against 85 per cent (FLAME phase 3 randomised trial, 571 men with intermediate- and high-risk prostate cancer, median follow-up 72 months)“At 5-year follow-up bDFS was 92% and 85%, respectively.” Kerkmeijer et al., FLAME, Journal of Clinical Oncology 2021

What limits it

WorkflowThe boost is limited by the bladder and rectum, and by how reliably the lesion is seen and drawn on MRI.“Organ at risk constraints were prioritized over the focal boost dose.” FLAME 2021
SamplingNo difference in prostate cancer-specific or overall survival at this follow-up.“We did not observe differences in prostate cancer-specific survival” FLAME 2021

Next step

Not yet written.

What it would buy

Shown to helpBetter PSA-defined control without a measurable rise in late side effects or loss of quality of life, in one large randomised trial. Survival benefit not shown.“Both for late toxicity as health-related quality of life, differences were small and not statistically significant.” FLAME 2021

Dosing

How far a dose chosen by body size or by the highest tolerated amount sits from the dose a particular person needs.

Dosing chemotherapy by body surface area

Tested, no benefit

Body-surface-area dosing Cytotoxic chemotherapy

Sold as more precise than shown: Dosing in milligrams per square metre reads as a personalised dose and is, for most drugs in adults, a weakly individualised one.

Precision means

How much of the difference in drug clearance between adults on the same dose per square metre body surface area accounts for

Today

In most drugs studied, surface area does not reduce the spread of blood levels between adults at all (Review of pharmacokinetic studies of commonly used cytotoxics, 2007)“In (by far) most cases, use of BSA does not reduce the interindividual variation in the pharmacokinetics of adults” Flat-fixed dosing versus body surface area based dosing of anticancer drugs in adults: does it make a difference? (Oncologist 2007)

What limits it

BiologySurface area was adopted in 1958 because clearance scaled with it across species; between adults of the same species it tracks the enzymes and kidneys that clear a drug only loosely.
WorkflowIt is calculated from height and weight at one visit, so it does not follow weight loss, muscle loss or fluid during treatment.
WorkflowChanging it means changing the unit every protocol, pharmacy system and nomogram is written in.

Next step

Available, not routineReplace or correct it where a better variable exists: kidney function for carboplatin, genotype before fluoropyrimidines and thiopurines, measured blood level for infusional fluorouracil and oral targeted drugs, lean body mass in research.

Carboplatin dosing by kidney function: the Calvert formula and how kidney function is measured Fluorouracil dosing by measured blood level DPYD genotyping and DPD phenotyping before fluoropyrimidines Blood-level monitoring of oral targeted cancer drugs

What it would buy

Not establishedOn its own, replacing surface area with a flat dose changes nothing for a patient: the 2007 review concluded flat dosing is no worse rather than better. The gain comes from the drug-specific replacements below.Flat-fixed dosing versus body surface area based dosing of anticancer drugs in adults: does it make a difference? (Oncologist 2007)

Fluorouracil dosing by measured blood level

Some evidence

Fluorouracil dosing by measured blood level Fluorouracil (5-FU)

Precision means

Share of patients whose fluorouracil exposure, in milligram-hours per litre, falls in the target range on a surface-area dose

Today

20 to 30 per cent in range; about 40 to 60 per cent underdosed and 10 to 20 per cent overdosed (Review of studies linking fluorouracil plasma concentration to toxicity and efficacy, 2009)“only 20%-30% of patients treated with a 5-FU-based regimen have 5-FU levels that are in the appropriate therapeutic range” Pharmacokinetically guided dose adjustment of 5-fluorouracil: a rational approach to improving therapeutic outcomes (J Natl Cancer Inst 2009)

What limits it

WorkflowMonitoring needs a sample taken during the infusion and an assay with a fast turnaround; a 2009 review attributed the lack of uptake to the absence of a simple, fast and inexpensive method.
WorkflowThe randomised evidence used a weekly eight-hour infusion schedule, not the FOLFOX and FOLFIRI regimens given today.
BiologyIt measures the parent drug in plasma, not the active nucleotides inside the tumour cell, and it does not apply to capecitabine taken as a tablet.

Next step

In trialsRandomised trials on current regimens with survival endpoints; the 2026 meta-analysis found severe toxicity halved and response raised but could not establish a survival benefit.Comparative efficacy and safety of pharmacokinetically guided and body surface area-based 5-fluorouracil dosing in colorectal cancer: a systematic review and meta-analysis (Eur J Clin Pharmacol 2026)

What it would buy

Plausible, not shownIn the one randomised trial, response rose from 18.3 to 33.7 per cent and severe toxicity fell; median survival was 16 against 22 months, a difference that did not reach significance.“the objective response rate was 18.3% in arm A and 33.7% in arm B (P = .004)” Individual fluorouracil dose adjustment based on pharmacokinetic follow-up compared with conventional dosage: results of a multicenter randomized trial of patients with metastatic colorectal cancer (J Clin Oncol 2008)

DPYD testing before fluoropyrimidines

Strong evidence

DPYD genotyping and DPD phenotyping before fluoropyrimidines Fluorouracil (5-FU)Capecitabine

Precision means

Risk of death from the treatment itself, in carriers of a DPYD variant given a standard dose

Today

2.3 per cent of 566 carriers died of the treatment, against 0.1 per cent of 13,363 non-carriers (Systematic review of 35 studies and 13,929 patients given standard-dose fluoropyrimidines with baseline DPYD testing)“25.6 times increased risk of treatment-related death” Pathogenic DPYD Variants and Treatment-Related Mortality in Patients Receiving Fluoropyrimidine Chemotherapy: A Systematic Review and Meta-Analysis (Oncologist 2021)

What limits it

BiologyThe four variants tested for are common in European ancestry and miss variants that matter in others, and genotype explains only part of enzyme activity; plasma uracil measures the deficiency whatever its cause.
BiologyA reduced starting dose does not abolish the excess risk: in the Dutch prospective study severe toxicity was still 39 per cent in carriers against 23 per cent in wild-type patients.
WorkflowThe result has to arrive before cycle one, which is where testing most often fails in practice.

Next step

Available, not routineTesting everyone before the first dose, as the European Medicines Agency recommended in 2020 and the FDA's revised boxed warning advised in February 2026, and settling the right size of reduction for the c.1236G>A and c.2846A>T variants.“test patients for genetic variants of DPYD prior to initiating treatment with capecitabine or 5-FU unless immediate treatment is necessary” FDA: Safety labeling update for capecitabine and fluorouracil (5-FU) on risks associated with DPD deficiency (2026)

DPYD genotyping and DPD phenotyping before fluoropyrimidines

What it would buy

Shown to helpFewer deaths and fewer hospital admissions from the first cycles. A matched analysis of 156 carriers given reduced doses against 775 wild-type controls found progression-free and overall survival not shortened overall, with shorter progression-free survival in c.1236G>A carriers.Survival of Patients With Cancer With DPYD Variant Alleles and Dose-Individualized Fluoropyrimidine Therapy: A Matched-Pair Analysis (J Clin Oncol 2023)

UGT1A1 testing before irinotecan

Some evidence

UGT1A1 genotyping before irinotecan Irinotecan (and liposomal irinotecan)

Precision means

Rate of febrile neutropenia in the first two cycles in people who clear irinotecan's active metabolite slowly

Today

6.5 per cent of 31 poor metabolisers given a 30 per cent reduced starting dose, against 24 per cent in a historical full-dose group (Prospective multicentre study of 350 evaluable patients genotyped before irinotecan, Netherlands)“incidence of febrile neutropenia in this group was 6.5% compared to 24% in historical UGT1A1 PMs” UGT1A1 genotype-guided dosing of irinotecan: a prospective safety and cost analysis in poor metaboliser patients (Eur J Cancer 2022)

What limits it

BiologyThe risk depends on the dose given: a meta-analysis of 821 patients found the odds of severe blood toxicity in *28 homozygotes raised at medium and high irinotecan doses but not at 100 to 125 mg per square metre.
BiologyEven on a reduced dose, exposure to the active metabolite SN-38 in poor metabolisers remained about 32 per cent above a standard-dosed comparison group.
Cost and accessVariant frequencies differ by ancestry, so a panel built for one population misclassifies another; the label reports *28 homozygosity in about 20 per cent of Black or African American, 10 per cent of white and 2 per cent of East Asian individuals.

Next step

Available, not routineGenotyping before irinotecan as standard, which the Dutch study argued for and found cost-saving by 183 euros a patient.

UGT1A1 genotyping before irinotecan

What it would buy

Shown to helpFewer episodes of fever with a low white cell count, at no measured cost in survival: a Dutch cohort of 779 patients found no significant difference in progression-free or overall survival between dose-reduced poor metabolisers and fully dosed others.Survival of patients with colorectal or pancreatic cancer who received UGT1A1 genotype-guided dosing of irinotecan in the Netherlands (2017-2024): a retrospective, multicentre cohort study (Lancet Reg Health Eur 2026)

TPMT and NUDT15 testing before thiopurine maintenance

Strong evidence

TPMT and NUDT15 genotyping before thiopurines Mercaptopurine

Precision means

Share of the planned mercaptopurine dose a child can tolerate, by genotype

Today

50 to 90 per cent if heterozygous for one gene, 30 to 50 per cent for both, 5 to 10 per cent if homozygous for either (Study of 1,028 children with acute lymphoblastic leukaemia on maintenance therapy, as reported in the mercaptopurine label)“heterozygous for either TPMT or NUDT15, 50-90%; heterozygous for both TPMT and NUDT15, 30-50%; homozygous for either TPMT or NUDT15, 5-10%” Mercaptopurine oral suspension: US prescribing information, section 12.5 (DailyMed)

What limits it

BiologyGenotype explains part of the variation only, so blood counts still set the dose month by month through two years of maintenance.
BiologyNUDT15 variants were identified only in 2014 to 2015 and explain much of the severe toxicity in East Asian and Hispanic patients that TPMT testing missed; the label reports two loss-of-function alleles in about 2 per cent of East Asian patients.
WorkflowRed-cell enzyme activity, the alternative to genotype, is misleading after a transfusion, which many children have had by then.

Next step

Available, not routineTesting both genes at diagnosis so the result is ready before maintenance, following the consortium dose tables updated in 2025.Clinical Pharmacogenetics Implementation Consortium (CPIC) Guideline for Thiopurine Dosing Based on TPMT and NUDT15 Genotypes: 2025 Update (Clin Pharmacol Ther 2026)

TPMT and NUDT15 genotyping before thiopurines

What it would buy

Shown to helpAvoids the weeks of marrow failure that a full dose causes in a child with no enzyme, in a treatment given daily for about two years.Mercaptopurine oral suspension: US prescribing information (DailyMed)

Carboplatin dosing by kidney function

Some evidence

Carboplatin dosing by kidney function: the Calvert formula and how kidney function is measured Carboplatin

Precision means

Share of patients whose carboplatin dose is more than 20 per cent away from the dose a measured filtration rate would give

Today

18.6 per cent with the best published equation, 25.5 per cent with Cockcroft-Gault, 14.2 per cent with a model built on the same cohort (2,471 white adults with cancer at Cambridge University Hospitals, compared with chromium-51 EDTA measured filtration)“reduced the fraction of patients with a carboplatin dose absolute percentage error > 20% to 14.17% in contrast to 18.62% for the BSA-adjusted CKD-EPI and 25.51% for the Cockcroft-Gault formula” New Model for Estimating Glomerular Filtration Rate in Patients With Cancer (J Clin Oncol 2017)

What limits it

BiologyCreatinine comes from muscle, so it flatters kidney function in people with little of it; in 545 patients with cancer, CT-defined sarcopenia roughly doubled the odds that creatinine and cystatin C estimates disagreed by more than 30 per cent.
WorkflowMeasured clearance needs an injected tracer and timed samples, so most centres estimate instead.
Cost and accessThe 2017 model was developed in white patients only, so its accuracy elsewhere is unestablished.

Next step

Available, not routineCystatin C alongside creatinine, or measured clearance where the two disagree; a 357-patient study built a carboplatin clearance equation with 1 per cent bias and 15 per cent imprecision that held in underweight, normal-weight and obese patients.A universal formula based on cystatin C to perform individual dosing of carboplatin in normal weight, underweight, and obese patients (Clin Cancer Res 2009)

What it would buy

Not establishedA dose closer to the exposure the regimen was tested at. OnCo found no trial showing that a more accurate filtration estimate changes survival or toxicity, and says so rather than assuming it.

Blood-level monitoring of oral targeted drugs

Insufficient evidence

Blood-level monitoring of oral targeted cancer drugs Small-molecule kinase inhibitors

Precision means

Share of patients whose trough blood level sits below the target for their drug

Today

25.8 per cent of 426 patients still below target at the third measurement, after a 39.0 per cent reduction against historical data (600 patients starting one of 24 oral targeted therapies, prospective multicentre study in the Netherlands)“Pharmacokinetically guided dosing reduced the proportion of underexposed patients at the third PK measurement by 39.0%” Therapeutic drug monitoring-based precision dosing of oral targeted therapies in oncology: a prospective multicenter study (Ann Oncol 2022)

What limits it

BiologyAn exposure target exists for only some drugs, and a trough level is a weaker guide where the exposure-response curve is flat.
WorkflowMuch of the shortfall is not the dose at all: adherence, food, timing and interacting medicines accounted for several of the interventions used.
Cost and accessAssays and a pharmacologist to read them are concentrated in a few centres.

Next step

ResearchRandomised trials with progression or toxicity endpoints rather than exposure endpoints.

Measure blood levels of oral targeted drugs and adjust doses to a target range

What it would buy

Not establishedMore people receiving the exposure the drug was approved at. Whether that lengthens life or delays progression has not been tested against a concurrent control in the sources read here.

Choosing the dose a drug is approved at

Some evidence

Project Optimus Sotorasib

Sold as more precise than shown: An approved dose is often read as the dose that works best; for drugs approved before dose optimisation it is the highest dose tested that people could tolerate for a month.

Precision means

Whether the approved dose was compared against a lower one before approval, and what the comparison found

Today

Pooled response 33 per cent at 960 mg against 26 per cent at a reduced starting dose, with no significant difference; 960 mg remains the labelled dose (Systematic review of 10 studies of sotorasib in KRAS G12C-mutated non-small cell lung cancer, the first drug for which the FDA required formal dose optimisation)“The pooled ORR was 33% (95% CI, 29% to 36%) for patients starting at 960 mg (n = 1,347) versus 26% (95% CI, 19% to 34%) for those starting at reduced dose (n = 130)” Standard Versus Reduced Sotorasib Dosing in KRAS G12C-Mutated Non-Small Cell Lung Cancer: A Systematic Review and Meta-Analysis (JCO Oncol Pract 2026)

What limits it

WorkflowDose-finding in phase 1 measures toxicity over about four weeks in a few dozen people, while targeted drugs are taken for months, so the dose-limiting window never sees the toxicity that makes people stop.
WorkflowComparing two doses in the registration programme costs patients and time, and the lower dose is commercially unattractive to the sponsor.
BiologyOnce a drug is approved at the higher dose, the randomised comparison that would settle it is hard to run: the sotorasib meta-analysis had 1,347 patients at 960 mg and 130 at a reduced dose.

Next step

Available, not routineThe FDA's August 2024 guidance asks sponsors to identify an optimised dosage before applying for approval; ponatinib is the worked example of a dose revised after approval, its label changed to 45 mg reducing to 15 mg on response after the three-dose OPTIC trial.FDA Approval Summary: Revised Indication and Dosing Regimen for Ponatinib Based on the Results of the OPTIC Trial (Oncologist 2022)

Project Optimus Ponatinib

What it would buy

Plausible, not shownFewer dose reductions and fewer people stopping a drug that was working: at 960 mg of sotorasib, 17 per cent reduced the dose and 10 per cent stopped for toxicity.Standard Versus Reduced Sotorasib Dosing in KRAS G12C-Mutated Non-Small Cell Lung Cancer: A Systematic Review and Meta-Analysis (JCO Oncol Pract 2026)

Muscle and fat measured on the staging scan

Insufficient evidence

CT body composition and sarcopenia measurement

Precision means

Rate of severe toxicity in people with low muscle compared with others on the same dose per square metre

Today

Toxicity in 50 per cent of sarcopenic against 20 per cent of non-sarcopenic patients (55 women with metastatic breast cancer starting capecitabine, muscle measured at the third lumbar vertebra on CT)“Toxicity was present in 50% of sarcopenic patients, compared with only 20% of nonsarcopenic patients (P = 0.03)” Sarcopenia as a determinant of chemotherapy toxicity and time to tumor progression in metastatic breast cancer patients receiving capecitabine treatment (Clin Cancer Res 2009)

What limits it

SamplingThresholds for low muscle differ between populations and studies, so the same scan can be classified either way.
WorkflowThe slice has to be in the scan: a chest-only CT does not reach the third lumbar vertebra.
BiologyThe association is prognostic. A 2026 systematic review in pancreatic cancer called the body-composition evidence hypothesis-generating rather than practice-changing.

Next step

ResearchProspective trials of doses set by lean body mass rather than surface area; the same review found dose relative to lean mass predicted dose-limiting toxicity where surface-area dosing did not.Beyond body surface area: CT-derived body composition as a predictor of chemotherapy toxicity and survival in pancreatic ductal adenocarcinoma: a systematic review (Cancer Treat Rev 2026)

Dose chemotherapy by muscle mass, not body surface area

What it would buy

Not establishedA warning before the first cycle for the people most likely to be harmed by it, from a scan already taken. No trial has shown that changing the dose on this basis improves outcomes.

Capping the dose in larger patients

Some evidence

Full weight-based chemotherapy dosing for people with obesity Body-surface-area dosing

Precision means

Share of patients with obesity given less than the full dose their body size calls for

Today

Up to 40 per cent (Practice-pattern studies reviewed for the ASCO guideline on chemotherapy dosing in obese adults, literature to 2010)“up to 40% of obese patients receive limited chemotherapy doses that are not based on actual body weight” Appropriate chemotherapy dosing for obese adult patients with cancer: American Society of Clinical Oncology clinical practice guideline (J Clin Oncol 2012)

What limits it

WorkflowCapping is a habit rather than a rule, so it is invisible unless prescriptions are audited.
BiologyThe guideline panel found the concern it rests on unfounded: it reported no evidence of increased short or long-term toxicity, with myelosuppression the same or less pronounced on full weight-based doses.
BiologyWeight is not composition: the 2021 update named sarcopenia and other body composition measures as an important area for future research.

Next step

Available, not routineApply the guideline, which since 2021 extends full approved doses to immunotherapy and targeted therapy as well as chemotherapy.Appropriate Systemic Therapy Dosing for Obese Adult Patients With Cancer: ASCO Guideline Update (J Clin Oncol 2021)

What it would buy

Plausible, not shownThe dose the regimen was tested at, particularly where the aim is cure. The guideline abstracts do not put a figure on the survival an uncapped dose returns.Appropriate chemotherapy dosing for obese adult patients with cancer: American Society of Clinical Oncology clinical practice guideline (J Clin Oncol 2012)

What is missing, named

1 of 50 rows have no sourced figure for how precise the technology is today.
0 of 50 rows do not yet say what the next step would buy a patient, and 7 have no next step written.
A figure here comes from the cohort or trial named beside it. Machines, laboratories and surgeons differ, so the precision available where you are treated may be better or worse.

Related: the precision roadmap, which follows each area from where it stands to the next step · what comes back after treatment · complementary approaches, graded, the model for the grades used here. OnCo is orientation, not medical advice.