Oncology Survival
Analysis

Cancer survivors are dying of other things

As cancer treatment improves, more patients survive their primary diagnosis. But cancer survivors carry a high burden of cardiometabolic conditions: diabetes, hypertension, and obesity. These comorbidities may independently drive mortality in ways that aren't addressed by oncology-focused care. Understanding this is critical for survivorship programs, integrated care models, and risk stratification.

10 cycles, 8 notebooks, publication-grade analysis

I pooled 10 NHANES survey cycles (1999-2018) with NCHS linked mortality files, constructing a nationally representative cohort of 51,168 adults including 4,715 cancer survivors. The analysis follows a full epidemiologic pipeline from cohort construction through sensitivity analyses.

Cancer survivors with diabetes die faster

Kaplan-Meier curves show clear separation in survival between cancer survivors with and without diabetes. The gap widens over follow-up time, consistent with the cumulative cardiovascular toll of uncontrolled glycemia.

Kaplan-Meier survival curves: cancer survivors by comorbidity status

Diabetes: HR 1.31 after full adjustment

Progressive Cox regression models show that diabetes remains a significant predictor of mortality even after adjusting for age, sex, race/ethnicity, and smoking. Hypertension is borderline significant. Obesity shows a paradoxical protective effect, consistent with published survivorship literature.

Forest plot: fully adjusted Cox PH model

Dose-response: more conditions, higher mortality

Each additional cardiometabolic condition stacks the risk. Cancer survivors with all three conditions (diabetes + hypertension + obesity) have 30% higher mortality than those with none (HR 1.30, 95% CI 1.08-1.57). This dose-response relationship strengthens the causal argument.

Survival by number of cardiometabolic conditions among cancer survivors

Diabetes hits younger survivors hardest

Subgroup analysis reveals that the diabetes-mortality association is strongest among younger cancer survivors (age 20-59: HR 1.66) and attenuates with age (60-79: HR 1.36, 80+: HR 1.33). This has direct implications for survivorship screening guidelines.

Subgroup analysis: diabetes effect by age, sex, and cancer type

PH assumption: mostly met, mild violations flagged

Schoenfeld residual tests support the proportional hazards assumption for the primary exposures of interest (diabetes, hypertension, smoking). Mild violations are present for age and obesity, which is unsurprising given that both effects evolve non-proportionally over a 20-year follow-up window. The diabetes effect estimate is stable across the sensitivity analyses below, so the flagged violations are noted rather than rebuilt with time-varying coefficients in this version.

Schoenfeld residual plots

Log-log survival plots

Robust across sensitivity analyses

The diabetes-mortality association persists across every sensitivity test: alternative diabetes definitions (HbA1c-only), exclusion of early deaths (first 2 years), sex-stratified models, and time-period stratification. This consistency strengthens confidence in the primary finding.

Sensitivity analysis: diabetes HR across alternative specifications

Clinical and strategic implications

How I built this

Pooled 10 NHANES cycles (1999-2018) by merging 7 survey components per cycle (demographics, cancer history, diabetes, body measures, blood pressure, smoking, HbA1c) with fixed-width mortality linkage files. Custom parser handles concatenated follow-up fields in the mortality data. Cancer defined as self-reported physician diagnosis (MCQ220). Diabetes defined as composite: physician diagnosis OR HbA1c >= 6.5%. Cox models with progressive adjustment (unadjusted, age-sex, fully adjusted). Schoenfeld residuals for PH diagnostics. 8 notebooks, 16 publication-quality figures.

Built with

Cox proportional hazards in Python using lifelines. Raw NHANES XPT files, NCHS mortality linkage, PH diagnostics, sensitivity analyses. Reproducible from the raw data.