预防医疗的决策基础设施
The Decision Infrastructure for Preventive Medicine
将健康行为转化为可结算的医疗证据。预测 → 干预 → 归因 → 结算，跑通预防医疗完整闭环。Turning health behaviors into billable medical evidence. Predict → Intervene → Attribute → Settle.
- 提前1-3年预测心脑血管风险，识别高危患者Predict cardiovascular risk 1-3 years ahead
- 辅助医生在疾病发生前主动介入Enable physicians to intervene before disease onset
- HIS无缝对接，风险信号实时推送至医生工作站Seamless HIS integration, real-time risk signals
- 产出因果归因证据，向支付方证明干预真实价值Generate causal evidence to prove intervention value
- 预测高赔付人群，在赔付窗口关闭前锁定干预机会Predict high-claim populations before payout window closes
- PSM因果证据量化干预降低赔付的真实效果PSM causal evidence quantifies real impact on claims
- 将预防结果转化为标准化可结算证据Convert prevention outcomes into billable evidence
- 支持险企从赔付方转型为主动健康管理者Help insurers shift to proactive health managers
- 员工群体心脑血管风险预测与分层Cardiovascular risk prediction across workforce
- 量化干预ROI，将健康投入与医疗费用降低强绑定Quantify intervention ROI linked to cost reduction
- 因果证据报告支持企业向保险方申请费率优惠Causal evidence supports premium negotiation
- API订阅按需接入，无需自建健康数据团队API subscription, no in-house health data team needed
- 开放真实世界患者数据授权，支持靶点发现与患者分层License real-world data for target discovery
- 带干预轨迹的动态数据集，价值远超静态历史数据Dynamic datasets with intervention trajectories
- PSM因果分析可直接复用于临床试验设计优化PSM analysis applicable to clinical trial design
- 产出符合NMPA要求的真实世界证据NMPA-compliant RWE for drug registration
基于患者风险因子，自动生成针对性干预建议。请先完成风险预测，或直接点击生成示例方案。Auto-generates targeted interventions based on patient risk factors. Run prediction first or click to generate example plan.
基于PSM因果分析结果，生成可提交给保险公司和医疗机构的标准化结算证据报告。Generate standardized settlement evidence reports for insurers and healthcare institutions based on PSM causal analysis.
total_cholesterol, ldl, hdl, triglycerides, exercise_days,
smoking, drinking, family_history, diabetes, hypertension
可结算证据Billable Prevention Evidence
指通过科学方法量化的、可被医疗支付体系（医保、商业保险、企业雇主）正式认可并用于费用结算的预防干预效果证明。不同于传统医疗记录，可结算证据须包含因果归因链路——即证明干预与健康结果改善之间存在统计意义上的因果关系，而非单纯相关性。ReHealth Core 通过 PSM 倾向评分匹配生成符合真实世界证据（RWE）标准的因果报告，作为向支付方提交的结算依据。 Quantified proof of preventive intervention outcomes, formally recognized by healthcare payers (public insurance, commercial insurers, employers) for reimbursement. Unlike traditional medical records, billable evidence requires a causal attribution chain — proving the intervention caused health improvement, not mere correlation. ReHealth Core generates RWE-standard causal reports via PSM for submission to payers.
PSM 因果归因Propensity Score Matching
倾向评分匹配（PSM）是准实验因果推断的核心方法，通过对干预组和对照组在年龄、性别、基线风险等混杂变量上的统计匹配，消除选择偏倚，从而将干预效果从噪声中剥离出来。这一方法论已在 Lancet、NEJM 等顶级期刊被广泛采用，是真实世界证据（RWE）生成的标准路径之一。 PSM is the core quasi-experimental method for causal inference. By statistically matching treatment and control groups on confounders (age, sex, baseline risk), it eliminates selection bias and isolates the true intervention effect. Widely adopted in Lancet and NEJM, it is a standard pathway for Real-World Evidence generation.
真实世界证据Real-World Evidence (RWE)
区别于随机对照试验（RCT），真实世界证据来源于常规诊疗、健康管理、可穿戴设备等真实医疗场景中积累的数据。国家药监局（NMPA）、美国 FDA 均已发布 RWE 指南，ReHealth Core 的因果证据报告可直接用于医保谈判、险企费率优化和药企注册申报。 Unlike RCTs, Real-World Evidence derives from routine clinical care, health management programs, and wearables. Both NMPA and FDA have issued RWE guidelines. ReHealth Core's causal reports are directly applicable to insurance negotiations, premium optimization, and drug regulatory submissions.
医疗级联邦学习Medical-Grade Federated Learning
联邦学习允许多方机构在不共享原始数据的前提下协同训练模型——患者数据始终留存于医院本地，仅上传经过差分隐私处理的模型梯度。从根本上解决医疗数据"能用不能看、能算不能拿"的合规困境，满足《个人信息保护法》与《数据安全法》的严格要求。 Federated learning allows multi-party institutions to collaboratively train models without sharing raw data — patient data stays on-premise, only differentially-private model gradients are uploaded. This solves the core healthcare data compliance dilemma, meeting PIPL and Data Security Law requirements.
预防结算Prevention Settlement
将预防医疗干预产生的健康改善效果，通过标准化因果证据转化为可被医疗支付体系正式结算的收益项。ReHealth Core 以"预防结算"为核心目标函数，打通从风险预测到干预归因再到支付方认可的完整商业闭环，推动预防医疗从成本中心转变为收益中心。 Converting health improvements from preventive interventions into formally reimbursable items via standardized causal evidence. ReHealth Core uses prevention settlement as its core objective function — closing the full loop from risk prediction to causal attribution to payer recognition, transforming prevention from a cost center to a revenue center.
AUC 0.839 的含义What AUC 0.839 Means
AUC（受试者操作特征曲线下面积）是分类模型区分能力的标准度量，范围 0.5（随机猜测）到 1.0（完美区分）。AUC 0.839 意味着：从任意一对高低风险患者中，模型有 83.9% 的概率正确识别出谁的风险更高。在真实世界临床场景中，AUC > 0.8 通常被认为具有良好的临床实用价值。 AUC (Area Under the ROC Curve) measures a model's discriminative ability, ranging from 0.5 (random) to 1.0 (perfect). AUC 0.839 means the model correctly identifies the higher-risk patient in 83.9% of random pairs. In real-world clinical settings, AUC > 0.8 is considered to have strong practical utility.
采用 CatBoost 梯度提升树，输入 16 个临床特征维度（含血压、血脂、血糖、BMI、生活行为习惯、家族史等），输出连续心血管风险评分与 SHAP 特征重要性分析，支持实时推理（<200ms 响应）。 CatBoost gradient boosting with 16 clinical features (BP, lipids, glucose, BMI, lifestyle, family history), outputting continuous cardiovascular risk scores and SHAP feature importance. Real-time inference (<200ms response).
基于倾向评分匹配（PSM）的因果推断引擎，支持综合干预、药物干预、生活方式干预和数字健康干预四类场景，输出标准化 ATT（平均处理效应）及置信区间，符合真实世界证据（RWE）方法论要求。 PSM-based causal inference engine supporting four intervention types: comprehensive, medication, lifestyle, and digital health. Outputs standardized ATT (Average Treatment Effect) with confidence intervals, compliant with RWE methodology.
医疗级联邦学习架构，患者原始数据不出域，差分隐私保护聚合梯度，端到端加密传输。满足《个人信息保护法》《数据安全法》《网络安全法》三法合规，医疗器械软件注册备案中。 Medical-grade federated learning: raw patient data stays on-premise, differentially-private gradient aggregation, end-to-end encrypted transmission. Compliant with PIPL, Data Security Law, and Cybersecurity Law. Medical device software registration in progress.
标准 RESTful API，提供 HIS 医院信息系统对接、保险核保系统接入、企业 HR 健康平台集成三条预置通路。支持 JSON/CSV 数据格式，平均接入周期 < 2 周，按调用量订阅计费。 Standard RESTful API with three pre-built integration paths: hospital HIS, insurance underwriting systems, and enterprise HR health platforms. JSON/CSV support, average integration time <2 weeks, usage-based subscription pricing.
什么是预防医疗的"可结算证据"？What is "Billable Prevention Evidence"?
可结算证据是指通过科学方法（主要是 PSM 因果推断）量化干预效果后，生成的可被医保、商业保险或企业雇主正式认可并用于费用报销的标准化报告。核心在于"因果性"而非"相关性"——必须证明是干预本身导致了健康结果改善。这是 ReHealth Core 的根本创新：不仅预测风险，更为预防行为创造可被支付体系认可的经济价值。Billable prevention evidence is a standardized report generated after quantifying intervention outcomes via PSM causal inference, formally recognized by public insurance, commercial insurers, or employers for reimbursement. The key is causality, not correlation — proving the intervention itself caused health improvement.
PSM 因果归因具体如何工作？How Does PSM Causal Attribution Work?
PSM 分三步：① 对每位患者计算"倾向评分"——在其年龄、性别、基线风险等特征下接受干预的概率；② 将干预组与对照组中倾向评分相近的患者配对，形成统计上可比的两组；③ 比较配对后两组的健康结果差异，得出平均处理效应（ATT）及置信区间。这一方法消除了"健康人更倾向参与健康管理"的选择偏倚，使归因结论具有因果解释力。PSM works in three steps: ① calculate a propensity score for each patient; ② match treated and control patients with similar scores; ③ compare outcomes to derive the Average Treatment Effect (ATT) with confidence intervals — eliminating selection bias.
联邦学习如何保障患者数据不出院？How Does Federated Learning Keep Patient Data On-Premise?
在联邦学习架构下，模型训练分布在各机构本地服务器，患者原始数据全程不离开机构本地环境。只有经过差分隐私处理的模型梯度会被加密上传至聚合节点。即使节点遭受攻击，攻击者也无法从梯度中还原出任何个人患者信息。架构符合《个人信息保护法》第 23 条数据最小化原则。Under federated learning, model training runs on each institution's local servers — raw patient data never leaves. Only differentially-private model gradients are encrypted and uploaded. Even if the aggregation node is compromised, no individual patient data can be reconstructed.