The Influence of Selection Bias on Racial Differences in Reproductive Aging and Accelerated Health Declines in the Study of Women's Health Across the Nation

Abstract

Aging cohort studies recruit participants at an age before most of the population will experience the study outcome to document its natural history and related risk factors. Usually assumptions regarding “normative” aging, including average age at onset, are based on data on White populations. However, evidence suggests that Black and Hispanic populations experience “weathering” or accelerated health declines compared to Whites due to cumulative experience of social, economic, and political marginalization. When population subgroups have a differential probability of inclusion, selection bias, i.e., a distortion caused by differential selection into or out of a study, will likely misinform understanding of aging in these populations.

By accounting for left truncation and left and right censoring in the Study of Women’s Health Across the Nation (SWAN), this dissertation addresses the main forms of selection bias in cohort studies and calculates their effects on estimates of racial/ethnic differences in reproductive aging and onset of cardio-metabolic risk. SWAN, a 25-year longitudinal cohort of midlife women, utilized a cross-sectional screener to recruit a cohort of Black, Chinese, Hispanic, Japanese, and White women.

Aim 1 created a framework for examining selection bias at study commencement and quantified how selection mechanisms from the cross-sectional screening study into the cohort biased estimates of racial/ethnic differences. Two main mechanisms were identified, eligibility and participation. Black and Hispanic women had the lowest eligibility rates stemming from high rates of surgical menopause. Their eligibility rates decreased with increasing age at a higher rate than in White women. Lower education was associated with lower eligibility for White women only. Participation was associated with demographic characteristics with no evidence of a “healthy volunteer” bias. Failure to account for selection at study commencement may underestimate racial/ethnic disparities in health, especially for Black and Hispanic women.

Aim 2 assessed the effects of selection bias on estimates of racial/ethnic differences in age at the final menstrual period (FMP). Results suggest that previous analyses that did not fully account for selection bias via left truncation and right censoring overestimated average age of FMP in Black and Hispanic women. In Black women, age of natural FMP was overestimated by an average of 1.1 years. Adjusting for selection bias, Black and Hispanic women had earlier natural and surgical FMPs compared to White women independent of socioeconomic factors and health behaviors.

Aim 3 assessed the effects of selection bias on estimates of racial/ethnic differences in the timing of cardio-metabolic outcomes including hypertension, isolated systolic hypertension, insulin resistance and diabetes. Selection at study commencement had greater effects on outcomes with earlier age at onset (hypertension), whereas selection via loss to follow up had greater effects on outcomes with later onsets (insulin resistance, diabetes). Addressing selection bias via left truncation, left and right censoring led to an average 20-year decrease in predicted median age of onset in cardio-metabolic outcomes but decreased the predicted difference in age at onset for Black and Hispanic women compared to White women. However, Black and Hispanic women still had significantly earlier onset of each outcome.

By providing a framework to account for and methods to mitigate selection bias, this dissertation documented accelerated health declines in Black and Hispanic women in SWAN across multiple reproductive and cardio-metabolic outcomes – highlighting the need for new and continued research/interventions aimed at the structural causes of racial disparities in health.