Abstract
INTRODUCTION: Cancer is inversely associated with cognitive impairment. Whether this is due to statistical handling of attrition (death and censoring) is unknown. METHODS: We quantified associations between cancer history and incident cognitive impairment among Health, Aging, and Body Composition Study participants without baseline cognitive impairment or stroke (n = 2604) using multiple competing-risks models and their corresponding estimands: cause-specific, subdistribution, and marginal hazards, plus composite-outcome (cognitive impairment or all-cause mortality) hazards. All-cause mortality was also modeled. RESULTS: After covariate adjustment (demographics, apolipoprotein E ε4, lifestyle, health conditions), cause-specific and marginal hazard ratios (HRs) were similar to each other (≈ 0.84; P values < 0.05). The subdistribution HR was 0.764 (95% confidence interval [CI] = 0.645-0.906), and composite-outcome Cox model HR was 1.149 (95% CI = 1.016-1.299). Cancer history was positively associated with all-cause mortality (HR = 1.813; 95% CI = 1.525-2.156). DISCUSSION: Cause-specific, subdistribution, and marginal hazards models produced inverse associations between cancer and cognitive impairment. Competing risk models answer slightly different questions, and estimand choice influenced findings here. HIGHLIGHTS: Cancer history is inversely associated with incident cognitive impairment. Findings were robust to handling of competing risks of death. All models also addressed possible informative censoring bias. Cancer history was associated with 16% lower hazard of cognitive impairment. Cancer history was associated with 81% higher all-cause mortality hazard.