Why Ordinary Medical Records Can't Be Used for Settlement
Traditional medical records are descriptive: they document what happened (diagnoses, prescriptions, test results). Payers verify these records and pay accordingly.
The challenge with preventive interventions is that the core "service outcome" is disease not occurring. The payer's question isn't "did the intervention happen?" but "did it actually prevent disease?" — a causal inference problem, not a record verification problem.
Key Distinction:Ordinary health records can prove "participants' blood pressure decreased after the program." Billable evidence must prove "the decrease was caused by the intervention, not because healthier people self-selected into the program." This requires causal inference, not descriptive statistics.
Six Core Elements of Billable Evidence
Intervention Description
Intervention type (lifestyle, follow-up, digital-health tools, etc.), duration, population characteristics, and data provenance. Medication or treatment content is managed separately by medical institutions under applicable rules.
RequiredBaseline Characteristics Comparison
Pre-matching baseline differences between treatment and control groups, SMD, proving group comparability.
RequiredPSM Matching Quality
Propensity score model, matching method (nearest neighbor/kernel), post-matching SMD improvement, caliper width.
RequiredAverage Treatment Effect (ATT)
Average causal effect of the treatment group relative to matched controls, with 95% CI and p-value.
RequiredHealth Outcome Metrics
Primary outcome (e.g., cardiovascular risk score) and secondary outcomes (blood pressure, lipids, hospitalization rate) changes.
RequiredScope and Limitations
Applicable population characteristics, follow-up duration limits, unmeasured confounders declaration, AI-generated content disclaimer.
RequiredThree Levels of Billable Evidence
Group Trend Evidence
Pre/post statistical comparison without fully eliminating confounders. Suitable for annual reports and internal performance, not recommended for insurance settlement.
PSM Causal Attribution Evidence
PSM helps reduce bias from observed confounding and reports ATT with confidence intervals. It can inform program evaluation and discussion by commercial insurers, employers, and partners; it does not control unobserved confounding.
Multi-center Federated Real-world Study Evidence
Multi-institution, federated studies can expand the observable data base and should fully disclose governance, design, quality checks, and limitations. Drug or device registration submissions require separate work under applicable NMPA requirements.