Can lifestyle factors explain racial and ethnic inequalities in all-cause mortality among US adults? | BMC Public Health

Participants

Data came from the National Health Interview Survey (NHIS) linked to the National Death Index (NDI) using probabilistic record matching [22]. NHIS is an annual, nationally representative, cross-sectional household survey of the civilian non-institutionalized US population (i.e., active duty members of the US military and individuals living in an institution such as residential care facilities or prisons were not sampled). NHIS utilized a complex, multistage sample design that involved stratification, clustering, and oversampling of specific population subgroups. Every year approximately 35,000 households are enrolled, from which one adult is randomly selected for a face-to-face interview. An annual assessment of all lifestyle factors in sufficient detail started in 1997, and NHIS data up to 2014 have been linked to the NDI. Therefore, this study included pooled NHIS data from 1997 to 2014. The NDI contains information on vital status, time of death, and time last presumed alive with follow-up to December 31, 2015. Our sample was comprised of the adults (ages ≥ 18 years) who were not missing data on the exposure, mediators, outcome, and covariates; those with complete and missing data were largely similar across a range of characteristics (Supplementary Table S1). Participants over 85 years of age at the time of NHIS administration were removed given that their exact age was not available through the public use data files.

Measures

The outcome was time to all-cause mortality, operationalized as the time from the NHIS survey to death or last presumed alive. Race and ethnicity, the independent variable of interest, was self-reported and categorized as non-Hispanic White (reference category; henceforth White), non-Hispanic Black/African American (henceforth Black), and Hispanic/Latinx. We further distinguished all other non-Hispanic racial and ethnic groups (hereafter, non-Hispanic Other) for descriptive analyses, though sample size was too small for inclusion in the main analyses.

Participants’ report of the frequency and quantity of alcoholic beverage consumed in the past 12 months was converted to grams of pure alcohol consumed per day, assuming 14 g of pure alcohol per drink. Alcohol use was categorized according to the standards of the World Health Organization [23] and included: (1) never drinkers (no drinks in the past year and less than 12 drinks in any one year or entire life), (2) former drinkers (no drinks in the past year but have had at least 12 drinks in any one year), (3) category I (men: (0, 40] grams per day; women: (0, 20] grams per day; reference category), (4) category II (men: (40, 60] grams per day; women: (20,40] grams per day), (5) category III (men: >60 g per day; women: >40 g per day). With respect to smoking, participants were asked to report whether they (1) have smoked at least 100 cigarettes over their entire life, and (2) whether they currently smoked cigarettes. Smoking cigarettes was categorized as never smokers (reference category), former smokers, current someday smokers, and current everyday smokers. Based on self-reported height and weight, BMI was calculated and categorized according to current WHO guidelines as underweight (< 18.5 kg/m²), normal weight (18.5-24.99 kg/m²; reference category), pre-obesity (25-29.99 kg/m²) or obese (≥ 30 kg/m²) [24]. With respect to physical activity, participants reported how often and for how long they performed vigorous and light-moderate leisure-time physical activities of at least 10 min. No timeframe (e.g., over the past year, or past month) was specified. The length of moderate physical activity per week was calculated, assuming that 1 min of vigorous physical activity is equivalent to 2 min of moderate physical activity [25]. Physical activity was categorized as sedentary (0 min/week), somewhat active (< 150 min) or active (≥ 150 min; reference category), given the WHO recommendations of 150–300 min of moderate-intensity physical activity per week [26].

The covariates used in all models were age (continuous), sex, educational attainment, marital status, and survey year (continuous). Educational attainment was categorized as low (high school diploma or less), medium (some college but no bachelor’s degree), or high (bachelor’s degree or more), and was treated as a proxy for socioeconomic status; given its ubiquity in the extant literature, stability over time, and completeness of data (e.g., relative to income) in the NHIS. Marital status was a binary variable indicating whether the individual was married or living with partner.

Statistical analyses

Causal mediation analysis using the marginal structural approach with Aalen’s additive hazard models was used, as described by Lange et al. [27,28,29]. Briefly, this flexible approach uses a counterfactual framework and allows for the direct parameterization of natural ‘direct’ and ‘indirect’ effects through multiple mediators and exposure-mediator interactions. The total effect of race and ethnicity on mortality was decomposed into three components (Fig. 1): (1) the average pure indirect effect through each mediator (indicating differential exposure), (2) the average indirect effect of the mediated interaction between race and ethnicity and each mediator (indicating differential vulnerability), and (3) the average ‘direct’ effect of race and ethnicity independent of mediators and covariates. The model simultaneously included all mediators (lifestyle factors: alcohol use, smoking, BMI, physical activity) and covariates (age, educational attainment, marital status, and survey year), and we fit separate models for men and women. Aalen’s additive hazard models have the advantage of directly estimating additive interactions (reflecting differential vulnerability), which are of greater importance (relative to multiplicative interactions) for public health [30].

All analyses were completed in R 4.1.3, using the timereg package (version 2.0.2) [31]; the statistical code is publicly available (see below). The timereg package does not allow for complex sampling designs and survey weights were not utilized given the analytical and computational complexity of the analyses.

In a sensitivity analysis, causal mediation models were repeated without education included as a covariate, recognizing that race and ethnicity are deeply tied to SES in the US [32], and prior research shows SES differences in effects of lifestyle factors on mortality [14, 15].