OnCo
ideasIdea

Open, benchmarked algorithms for lines of therapy and progression from routine data

Publish the exact rules used to work out from messy hospital records which treatment a patient was on and when it stopped working, and test them all on the same data.

Real-world studies depend on deriving line of therapy, progression and response from records, but every vendor and academic group uses proprietary rules, so results are not comparable. The proposal is an open library of versioned derivation algorithms with a benchmark dataset of chart-reviewed cases, run as a public leaderboard, so studies can cite algorithm version and known error rates.

Hypothesis
Open benchmarked algorithms will reduce between-study variation in real-world PFS estimates for the same cohort by half and become the reference method in regulatory RWE submissions.
Rationale
In genomics, open benchmarked pipelines (GATK, Genome in a Bottle) made variant calling comparable across labs; RWE has no equivalent for its core derived variables.
What would test it
Assemble 2,000 chart-reviewed patients across three cancers as a benchmark; invite groups to submit algorithms; publish accuracy and inter-algorithm agreement.
Maturity
early clinical
Who has to act
research
Cost to try
Small (under $1M)
Years to first evidence
2
Bottlenecks it attacks
  • Data silos · Records, scans, genomes and outcomes sit in separate systems that cannot talk. Every patient's experience is lost to the next.
  • Weak real-world evidence and registries · We do not reliably know what happens to patients after approval, so we cannot tell which drugs deliver in practice.

Connected

5top