The Question
Investigation 13 shows that mortality can be predicted from health data alone. But people aren’t just their diagnoses — they’re also their bank accounts, their education, their marital status. The social determinants of health (SDOH) literature is clear: income is associated with longevity. The question is whether adding income, education, and marital status to a model that already knows measured biomarkers — BP, lab cholesterol, HbA1c, CRP, hemoglobin, serum creatinine — improves prediction. NHANES provides a stronger incremental-prediction test than HRS because it includes laboratory measurements, not just self-reported conditions.
ADM Prediction (Made Before Running Models)
Predicted winner: Health+SDOH ML, but modest gain. Chetty et al. (2016) reported a large income-associated life-expectancy gap. Income, health behaviors, conditions, and biomarkers are also strongly related, so their predictive information may overlap. The interesting question isn’t whether wealth matters — it’s whether the tested SDOH variables add information beyond what measured health captures.
Prediction confirmed. Income, education, and marital status add only +0.003 AUC (0.908 → 0.910) with overlapping CIs. The SDOH-only model reaches AUC 0.650, while the health-only model reaches 0.908. Together, these results show a strong observed income gradient but little incremental predictive discrimination from SDOH after the measured health panel is included.
Results
ROC Curves (Three Models)
Feature Importance (Top 8, Health+Bio+SDOH)
Income Quintile → Mortality Rate
Predictive Overlap Across Feature Sets
The NHANES Advantage
The HRS version of this investigation included self-reported conditions. NHANES provides a stronger predictive comparison because it has measured biomarkers: up to four averaged blood pressure readings, lab-verified cholesterol, HbA1c, CRP, hemoglobin, serum creatinine, and albumin. Even with those measurements in the model, however, this observational comparison cannot identify the causal pathway connecting income, health, and mortality.
The gradient itself is stark — the poorest income quintile has 28.0% ten-year mortality, the highest quintile 12.3% (relative risk 2.28×, 15.7 percentage-point spread). This is not a null result about poverty; it is a narrow result about incremental prediction. The observational design cannot tell us which causal pathways produced the gradient.
The ADM Insight
For this specific 10-year mortality model, income, education, and marital status add +0.003 AUC after measured biomarkers are included, with overlapping confidence intervals. The decision-relevant result is narrow: these SDOH variables add little ranking information for this model and cohort. It does not mean wealth is unimportant for health, nor that a blood panel substitutes for social context in care or policy.
Data source: CDC NHANES 1999–2018 (ten two-year cycles pooled) linked to NCHS Linked Mortality Files (NDI match through December 2019). Mortality-eligible respondents only (ELIGSTAT=1).
Cohort: Adults aged 25–90 with at least 10 years of mortality follow-up or an observed death within 10 years. Final analytic sample: 28,636 adults, 6,626 deaths (23.1%).
Three models compared: (1) Domain: Charlson-style log-relative-risk score with Gompertz age component — conditions, smoking, BMI only. (2) Health + Biomarkers ML: GradientBoosting on chronic conditions, smoking, BMI plus measured BP, lipids, HbA1c, CRP, hemoglobin, serum creatinine, albumin. (3) Health + Bio + SDOH ML: same architecture plus poverty income ratio quintile (INDFMPIR), education (DMDEDUC2), and marital status (DMDMARTL).
Predictive-overlap test: Fit SDOH-only, health-only, and combined models under the same folds. Compare health-only with health+SDOH to estimate incremental discrimination. AUC differences are not treated as additive effects or formal mediation.
Evaluation: 5-fold stratified cross-validation. Bootstrap 95% CIs from 1,000 resamples. Survey weights applied for nationally representative gradient estimates (NHANES 2-year weights divided by number of pooled cycles), but equal weighting for ML training.
Cross-sectional SDOH measurement. Poverty income ratio is measured at one interview. Lifetime income trajectory (childhood poverty, income volatility, retirement drop) may matter more than a snapshot.
SDOH panel is narrow. Only PIR, education, and marital status. Adding neighborhood deprivation, occupation, or discrimination exposure could change the picture — those variables are known to have effects not fully captured by individual income.
Pooled cycles span two decades. Policy changes over 1999–2018 (ACA, SNAP expansions, etc.) may have shifted the SDOH-health relationship. The pooled estimate smooths across a period when the income-mortality gap widened.
No mediation claim. This study compares predictive discrimination. A formal mediation analysis would require a causal estimand, explicit temporal assumptions, confounder control, and an appropriate effect-scale estimator; differences in AUC do not supply those elements.
Reverse causation. Serious illness causes medical bankruptcy, job loss, and divorce. A cross-sectional design cannot separate "poverty made you sick" from "sickness made you poor."