Participants were asked if they needed the help of another person for six personal care activities (bathing or showering, dressing, eating, using or getting to the toilet, getting around inside the home, and getting in/out of bed or chairs) because of a physical, mental or emotional problem [16]. A person was considered to have disability in ADLs if they needed help to perform one or more essential activities for personal self-care.

Participants were also asked to indicate, in one question, difficulties in handling routine needs, such as everyday household chores, doing necessary business, shopping, or getting around for other purposes [16]. A person was considered to have disability in IADLs if they needed assistance to perform routine activities.

Light-moderate and vigorous leisure-time PA was recorded in terms of frequency and duration [16]. Subsequently, time per week (min/week) in light-moderate (MPA) and vigorous PA (VPA) was calculated, and PA was estimated as MPA + 2*VPA [17]. In order to correct over-reporting and minimize the influence of outliers on results, PA was truncated to 1,680 min/week [18]. PA was also classified according to the 2020 WHO PA guidelines [10] into: (i) not meeting WHO recommendations (PA < 150 min/week), and (ii) meeting WHO recommendations (PA ≥ 150 min/week).

The leading underlying cause of death was obtained by the linkage of the participant data with the National Death Index (NDI) mortality data, based on probabilistic matches [19]. NCHS classifies the underlying cause of death into 10 categories, using the 10th revision of the International Statistical Classification of Diseases, Injuries, and Causes of Death (ICD-10). All-cause mortality, as well as CVD (ICD-10 codes: I00-I99) and cancer mortality (ICD-10 codes: C00-C97) were considered. Follow-up period (in months) was calculated from the interview date to the date of death for deceased participants or to censoring date (31 December 2019) for the rest of participants.

Selected covariables, including sociodemographic, lifestyle, and health-related factors that have been relevant on the previous scientific literature and availability in the NHIS dataset [9, 20]. Sociodemographic covariates included sex (men; women), age (years), ethnicity (non-Hispanic White; non-Hispanic Black; Hispanic; and other race), educational attainment (less than high school; high school grade or equivalent; and more than high school), and relationship status (married or living with a partner; divorced, separated, or widowed; and never married). Lifestyle covariates were smoking status (never; former; current) and alcohol drinking (never; former; current). As health-related factors, body mass index (BMI) was calculated as self-reported weight (in kg) divided by squared self-reported height (in cm) and classified as underweight or normal-weight (< 25 kg/m2), overweight (25–29.9 kg/m2), and obesity (≥ 30 kg/m2). Participants also reported if they have even been diagnosed with the following conditions by a physician: hypertension, cancer, diabetes, respiratory disease and cardiovascular disease. Functional limitations was classified based on information from 12 items that indicate having any difficulty doing specific activities by oneself and without any special equipment: (i) push or pull large objects, (ii) go out to things like shopping, (iii) participate in social activities, (iv) do things to relax at home or for leisure, (v) walk a quarter of a mile, (vi) walk up ten steps without resting, (vii) stand or be on feet for about two hours, (viii) sit for about two hours, (ix) stoop, bend, or kneel, (x) reach up over the head, (xi) use fingers to grasp or handle small objects, and (xii) lift or carry something as heavy as 10 pounds. If a person acknowledged having difficulty doing one or more of these activities was coded as “limited in any way” and considered to have functional limitations [16]. To prevent the exclusion of participants due to missing data (ranged 0% to 3.22%) a dummy category was created denoting lack of information for each variable.

Analyses were conducted using STATA version 14.0 (Stata Corp, College Station, TX, USA) for Macintosh, with the level of statistical significance set at p < 0.05. Baseline characteristics of the study participants by disability in ADLs and IADLs were presented as mean (standard deviation) for continuous variables and as percentage for categorical variables.

Cox proportional regression models were used to estimate hazard ratios (HR) and their 95% confidence intervals (CI) for the association of each disability type (i.e., disability in ADLs and disability in IADLs) with mortality from all-cause, CVD, and cancer. Four models with sequential adjustment for potential confounders were fitted. Model 1 was adjusted for sociodemographic factors (sex, age, ethnicity, education and relationship status). Model 2 was additionally adjusted for lifestyle factors (smoking and alcohol consumption). Model 3 was additionally adjusted for health-related factors (BMI, hypertension, CVD, cancer, diabetes, and respiratory disease). And model 4 was additionally adjusted for functional limitations and for disability in ADLs or IADLs, as appropriate; that is, analyses for disability in ADLs were additionally adjusted for disability in IADLs (yes, no) and functional limitations (yes, no); and analyses for disability in IADLs were additionally adjusted for disability in ADLs (yes, no) and functional limitations (yes, no).

We also explored in detail the association between PA and with all-cause, CVD, and cancer mortality in people with and without disability in ADLs and IADLs. Firstly, we modelled restricted cubic spline Cox regressions, with knots at 10th, 50th, and 90th, to examine dose-response relationship between continuous values of PA and mortality for each subgroup (i.e., people with and without disability in ADLs and IADLs). Next, we examined the association between meeting PA recommendations and mortality, stratifying by disability; we run Cox regressions adjusting for the same models described above and considering as reference the participants who did not meet the recommended PA. Last, we examined whether PA attenuated the impact of disability on all-cause, CVD and cancer mortality risk. Thus, we tested the association of combined disability and PA recommendations with mortality risk, considering people without the specific disability and meeting PA recommendations as the reference group, as appropriate.

All analyses accounted for the complex survey design employed in NHIS by considering sample weights, and primary sampling units and stratum for variance estimation. Sample weights were first corrected by dividing by the number of pooled waves (i.e., twenty-two). To minimize reverse causation, sensitivity analyses were performed by excluding participants with a follow-up period < 2 years or removing from analyses people with CVD or cancer at baseline.