Pharmacoeconomic Report Audit and Validation

The problem it addresses

Pharmacoeconomic reports often depend on subtle parameter assumptions. Common failure modes include whether drug cost is weighted by the on-treatment population, whether monthly discount rates are converted correctly, whether QALYs are adjusted from monthly cycles to years, whether Markov models include half-cycle corrections, and whether HRs are used to derive treatment-arm survival curves. Manual review requires rereading methods, recalculating costs, QALYs, and ICERs, and explaining differences. This skill targets PSM, Markov, and decision-tree reports by turning validation into a repeatable workflow: extract methods, rebuild the model, compare baselines, and explain differences.

How it works

• Method extraction: reads .docx, .pdf, or plain text, focusing on the methods section to extract survival parameters, utilities, cost logic, discount rates, baseline results, and scenarios.
• Model rebuild: uses Python and NumPy to rebuild the model from the report description, saving validation_model_rebuild.py with comments for parameter sources, formulas, and steps.
• Baseline comparison: produces cost, QALY, and ICER tables with reported values, rebuilt values, absolute and relative differences, and verdicts; bias ≤5% passes, 5%–15% requires explanation, and >15% suggests possible algorithm misunderstanding.
• Difference explanation: for ⚠️ and ❌ items, lists likely causes, evidence, and impact on conclusions rather than only reporting numeric bias.
• Report output: generates a Word validation report with an executive summary, numbered sections, three-line tables, terminology notes, and a code appendix.

Boundaries

It works best when the report has a reasonably complete methods section. Missing parameters should be marked as not provided or assumed, with impact noted. Scenario reverse-engineering is only a supplementary check and does not replace correct understanding of the methods. The output is not a substitute for formal pharmacoeconomic expert review, and conclusions still depend on the original report, guidelines, and data availability.