Background: Older individuals with a cancer history have a lower risk of developing Alzheimer’s disease (AD). However, there is limited information on the relationship between cancer and subsequent risk of AD or Alzheimer’s disease and related dementias (ADRD) in racially and ethnically diverse populations. These seemingly counterintuitive findings may be influenced by various sources of bias, which can obscure the true nature of this association.
Objectives and Specific Aims: The objective of this dissertation is to further elucidate the association between cancer and AD, addressing study biases and investigating heterogeneity across population subgroups and cancer characteristics. The specific aims were: (1) to investigate differences in the association between cancer and the risk of developing ADRD across sociodemographic factors; (2) to examine the heterogeneity in ADRD risk across different cancer sites, tumor stages, and treatment modalities; and (3) to compare trajectories of functional limitations among individuals with and without cancer, and to investigate changes in functional limitations before and after cancer diagnosis among those with cancer.
Methods: For specific aims 1 and 2, we analyzed a prospective cohort from the Multiethnic Cohort Study (MEC) using a causal framework to estimate the overall burden of cancer on ADRD and the controlled direct effect of cancer on ADRD. For specific aim 1, we conducted stratified analyses by sociodemographic factors (i.e., race/ethnicity, sex, education level, and age at enrollment). We used risk ratios (RRs) to estimate the total and controlled direct effects of incident cancer on ADRD at 26-year follow-up, stratified by each sociodemographic factor. We used inverse probability weighting to control for confounding by sociodemographic and health characteristics. The total effect of cancer on ADRD risk, including the path mediated by death, was estimated using a weighted Aalen-Johansen estimator. The controlled direct effect of cancer on ADRD risk (intuitively, the effect of cancer on ADRD risk not mediated by death) was estimated using a weighted Kaplan-Meier estimator. For specific aim 2, we stratified analyses by cancer type, tumor stage, and treatment modality to explore cancer heterogeneity using the same approach as in the specific aim 1. For specific aim 3, we followed individuals aged 50 and older without history of cancer using data from the Health and Retirement Study from 1998 through 2020. Incident cancer was self- or proxy- reported. Functional limitation was assessed using self-reported independence in instrumental activities of daily living (IADL) and activities of daily living (ADL). We employed linear probability models with repeated measures to assess IADL and ADL independence probabilities.
Results: The study sample for specific aims 1 and 2 included 104,258 Medicare-linked participants in the MEC. Among 104,258 participants (55% women, 74% non-White), 24,083 were diagnosed with incident cancer and 23,147 were diagnosed with ADRD during 26-year follow-up. For specific aim 1, the RR for the total effect of cancer on ADRD, including the path mediated by death, was 0.77 (95% CI, 0.75-0.80) at 26-year follow-up. The RR for the estimated direct effect of cancer on ADRD not mediated by death was 1.36 (95% CI, 1.32-1.37) at 26-year follow-up. In stratified analyses, the inverse cancer-ADRD association in total effect analyses was more pronounced in women, Black participants, participants with lower education levels, and those aged ≥65 at baseline. In contrast, the positive cancer-ADRD association in controlled direct effect analyses was stronger among men, Native Hawaiian participants, participants with higher education levels, and those aged <65 at baseline For specific aim 2, the five most frequently reported cancer types were prostate cancer (24%), breast cancer (17%), colorectal cancer (11%), lung cancer (10%), and non-Hodgkin lymphoma (4.9%). Of the total cancer cases, 56% were diagnosed at an early stage and 58% of cancer patients underwent surgery as the first course of treatment. Overall, the estimated total effect of cancer on ADRD risk was protective across cancer types, stages, and treatments. However, the estimated total effect of breast (RR 1.02; 95% CI, 0.89 to 1.06) and prostate cancer (RR 1.00; 95% CI, 0.92 to 1.08) on ADRD risk was null at 26-year follow-up. Across all cancer characteristics, the estimated direct effect of cancer was associated with a higher risk of ADRD. An exception was observed with thyroid cancer, where the estimated direct effect was protective at 26-year follow-up (RR 0.82; 95% CI, 0.67 to 0.94).
For specific aim 3, among 15,972 participants (mean [SD] age, 66.1 [9.9] years), 23.1% reported having cancer during follow-up. The probability of IADL independence before cancer diagnosis was slightly lower compared to cancer-free individuals, with a risk difference of 1.3% (95% CI, -1.9 to -0.7). After diagnosis, individuals with cancer experienced a sharp 6% drop in probability of IADL independence, resulting in a 3.4% lower probability of IADL independence compared to their cancer-free counterparts (95% CI, -2.2 to -4.6). The annual rate of IADL decline after diagnosis was slower than for cancer-free individuals (-0.9% [95% CI, -1.2 to 0.7] and -1.3%, [95% CI, -1.4 to -1.3] respectively).
Conclusions: Our study provides additional evidence of the relationship between cancer and ADRD. The findings support the inverse association between cancer and ADRD observed in previous prospective cohorts when accounting for death as a competing event (i.e., total effect). However, cancer was associated with higher ADRD risk when death was treated as a censoring event (i.e., controlled direct effect). Additionally, we observed heterogeneity in the association across sociodemographic factors and cancer characteristics. Lastly, the slower rate of functional decline in cancer survivors compared to cancer-free individuals may partially explain the observed lower Alzheimer’s disease risk.
