PSM在预防医疗中的完整应用链路The Complete PSM Application Chain in Preventive Medicine
PSM工程化落地到真实预防医疗场景，需要解决三个实际问题：数据从哪来、匹配质量如何保证、结果如何让支付方认可。Engineering PSM for real-world preventive medicine requires solving three practical problems: data sourcing, matching quality assurance, and making results acceptable to payers.
第一步：数据接入与准备Step 1: Data Integration and Preparation
PSM需要三类数据：①基线特征变量（年龄、BMI、血压、血脂、吸烟史、基线风险评分等）；②干预分配记录（谁参与了干预，何时开始）；③结局指标（随访期内的健康结果变化）。所有基线变量必须在干预开始之前测量。PSM requires three data types: ①Baseline characteristics (age, BMI, blood pressure, lipids, smoking history, baseline risk score); ②Treatment assignment records (who received the intervention and when); ③Outcome metrics (health outcome changes during follow-up). All baseline variables must be measured before intervention begins.
第二步：计算倾向评分与执行匹配Step 2: Calculate Propensity Scores and Execute Matching
使用逻辑回归模型以所有基线协变量为输入，预测每位患者"接受干预的概率"——即倾向评分。采用1:1最近邻匹配加卡钳（通常为PS标准差的0.2倍）。匹配完成后，必须检验所有协变量的标准化均值差（SMD）是否均小于0.1。Use logistic regression with all baseline covariates to predict each patient's probability of receiving the intervention — the propensity score. Apply 1:1 nearest-neighbor matching with caliper (typically 0.2 × SD of propensity score). After matching, verify all covariate SMDs are below 0.1.
| 指标Metric | 匹配前Pre-match | 匹配后Post-match | 标准Standard |
|---|---|---|---|
| 年龄 SMDAge SMD | 0.31 | 0.04 | <0.1 |
| BMI SMD | 0.24 | 0.06 | <0.1 |
| 基线风险 SMDBaseline Risk SMD | 0.35 | 0.03 | <0.1 |
第三步：估计ATT与置信区间Step 3: Estimate ATT and Confidence Intervals
在配对样本上计算干预组和匹配对照组的结局差异，得到ATT（处理组平均处理效应）。使用Bootstrap重采样（建议≥1000次）计算95%置信区间。Calculate the outcome difference between matched treatment and control groups to obtain ATT (Average Treatment Effect on the Treated). Use Bootstrap resampling (≥1000 iterations recommended) for 95% confidence intervals.
PSM不是一行代码，而是一条需要严格质量控制的数据流水线。能通过完整质量检查清单的PSM报告，才是可以拿去找支付方谈结算的证据。PSM is not a single line of code — it is a data pipeline requiring rigorous quality control. Only PSM reports that pass a complete quality checklist constitute evidence acceptable for settlement negotiations.