Participants and procedures
This study utilized data from the latest Health and Retirement Study (HRS), conducted every two years from 1992 to 2020. The HRS is a nationally representative survey that captures the sociodemographic data of middle-aged and retired older adults from approximately 23,000 US households and is supported by the National Institute on Aging (NIA U01AG009740) [34]. Similar to studies evaluating the retired population [35], we e only consider the individuals who are “completely retired” between 2004 and 2020. Following this screening, data missingness issues were significantly mitigated as data from the remaining 1238 individuals (aged 50 to 104) were mostly nonmissing. Note that the availability of physical activity data prompted the choice of 2004 as the base period for this study (See Table 1 for the complete descriptive statistics breakdown).
Measures
Healthy aging
Based on the augmented Rowe and Kahn’s model (i.e., Mclaughlin et al.‘s level 3 construct), we assess the disability, physical, and cognitive functioning domains of healthy aging. The disability domain estimates retired older adults’ ability to independently perform ten activities of daily living (ADL) and the instrumental activities of daily (IADL): “bathing, eating, dressing, walking across a room, and getting in or out of bed, using a telephone, taking medication, handling money, shopping, and preparing meals.” We reverse-coded and summed the composite scores of these ten items such that they range from 0 = ‘inability to perform all tasks’ to 10 = ‘ability in all tasks.’ Deviating slightly from the parent model’s prescription (suggesting the maximum score for this domain), we believe it is reasonable to classify individuals with a disability domain score ≥ 9 as having ‘sufficient ability’ since approximately 0% of the sample scored 10 [22, 27, 36].
The physical functioning domain assesses retired older adults’ capacity to engage in nine activities requiring ‘mobility’ and the use of ‘large muscles’: “walking several blocks, walking one block, walking across the room, climbing several flights of stairs, climbing one flight of stairs, sitting for two hours, getting up from a chair, stooping or kneeling or crouching, and pushing or pulling a large object.” We reverse-coded and summed the composite scores of these items such that they range from 0 = ‘poor functioning’ to 9 = excellent functioning;’ with a physical functioning domain intercept ≥ 8 indicating high physical functioning [22, 27].
The cognitive functioning domain captures retired older adults’ overall cognitive health using the total cognition score. The total cognition score is an ordinal outcome (ranging from 0 to 35) and is the sum of the weighted composite scores of the total recall and mental status indices — word recall (range 0—20); serial 7 subtraction (range 0—5); backward counting (range 0—2); objects (range 0—2), date (range 0—4), and president/vice president recognition (range 0—2). A cognition score tending to 35 entails excellent cognition in all cognition indices. A cognitive domain intercept ≥ 22 indicates high cognitive functioning [22, 37, 38].
Multimorbidity
study participants reported doctor’s diagnoses of the following eight chronic conditions: “1) high blood pressure or hypertension; 2) diabetes or high blood sugar; 3) cancer or a malignant tumor of any kind except skin cancer; 4) chronic lung disease except asthma such as chronic bronchitis or emphysema; 5) heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems; 6) stroke or transient ischemic attack (TIA); 7) emotional, nervous, or psychiatric problems; and 8) arthritis or rheumatism.” Based on relevant research, we grouped these conditions into five different multimorbidity classes as follows: cardiometabolic (diabetes, heart problems), neurological (hypertension, stroke, and psychiatric problems), musculoskeletal (arthritis), respiratory (chronic lung disease), and cancer. Each class is coded 0 (no chronic condition) and 1 (at least one chronic condition) and entered as distinct variables to ascertain the effects of each class on healthy aging. This approach is appropriate because although all classes of multimorbidity predict healthy aging, there is a low probability of correlation amongst them since having one class of chronic condition does not necessarily influence the occurrence of another in one individual [39, 40].
Physical activity and intensity
retired older adults reported their ‘frequency’ of engaging in light, moderate, and vigorous physical activity intensities. Light physical activities include “vacuuming, home repairs, sports, or other mildly energetic acts.” Moderate physical activities include “gardening, walking, dancing, stretching, and cleaning the car.” Vigorous physical activities included “running or jogging, cycling, playing tennis, digging, gym workouts, and swimming.” The frequency of each physical activity intensity was reverse-coded and ranged from 1 = none, 2 = one to three days per month, 3 = one day per week, 4 = more than one day per week, and 5 = daily physical activities.
Furthermore, we included established healthy aging covariates such as age, sex, race, marital status, educational attainment, socioeconomic status, body mass index (BMI), smoking and drinking status, and wave to control for retired older adults’ sociodemographic characteristics [15, 22]. Age is entered as a continuous variable ranging from 50 to 104 years old. We coded sex as 1 = female and 2 = male. Race ranged from 1 = white/Caucasian, 2 = black/African American, 3 = other. We coded marital status as 1 = “married, married (spouse absent), partnered, and 0 = separated, divorced, widowed, and never married, separated, and divorced.” Educational attainment is entered as 1 = less than high school, 2 = GED, 3 = high school graduate, 4 = some college, 5 = college and above. Based on Amadeo et al., socioeconomic status is calculated as the sum of total assets (less debt) and is coded as (-141,999 to 6030 = 0) poverty class, (6030.1 to 42,000 = 1) lower-middle class, (42,001 to 104,000 = 2) middle class, (104,001 to 201,000 = 3) upper middle class, (201,001 to 608,000 = 4) wealthy, and (608,001 and above = 5) super wealthy [41]. BMI is a continuous variable ranging from 12.6 to 58.6. Smoking status = 0 for non-smokers and 1 for smokers. Drinking frequency ranged from 0 = none to 7 = daily drinking. We included wave and wave2 to account for the influence of repeated measurements taken on individuals over different waves and for retired older adults’ healthy aging status depending on time [42].
Analysis
This study employed the mixed effects model with an unstructured covariance matrix and random intercept to estimate the longitudinal association between multimorbidity, physical activities, and the disability, physical, and cognitive functioning domains of healthy aging [43]. The mixed effects model is a multilevel statistical technique utilizing fixed and random effects to analyze nested data [43,44,45]This method is beneficial for analyzing variations within (fixed effects) and between (random effects) individuals over time and has been widely applied in the literature [46, 47]. Hence, we used the individuals’ ID and wave to construct nine mixed effects models. Each healthy aging domain was constructed three times, with each model assessing the interaction between the frequency of PA intensities and the five multimorbidity classes. Specifically, we introduced the three different PA intensities separately for each healthy aging domain model and interacted with them the multimorbidity classes of cancer, cardiometabolic, neurological, musculoskeletal, and respiratory conditions. We specify the base models for the three PA intensities below:
$$\begin{aligned} H{A_{dit}}\, = \,{\pi _{0i}}\, + \,{\beta _{1 – 5}}M{M_{it}}\, + & \\ & {\beta _6}PA{L_{fit}}\, + \,{\beta _{7 – 31}}{\text{(}}M{M_{it}} \times PA{L_{fit}}{\text{)}} \\ & {\beta _{32 – 37}}TV{C_{it}}\, + \,{\beta _{38 – 40}}TI{C_i} \\ & {\beta _{41 – 49}}(WAV{E_t} \times WAV{E_t})\, + \,I{D_i} \\ & WAV{E_t}\, + \,WAV{E^2}_t\, + \,{ \in _{it}} \\ \end{aligned}$$
(1)
$$\begin{aligned} H{A_{dit}}\, = \,{\pi _{0i}}\, + \,{\beta _{1 – 5}}M{M_{it}}\, + & \\ & {\beta _6}PA{M_{fit}}\, + \,{\beta _{7 – 31}}{\text{(}}M{M_{it}} \times PA{M_{fit}}{\text{)}} \\ & {\beta _{32 – 37}}TV{C_{it}}\, + \,{\beta _{38 – 40}}TI{C_i} \\ & {\beta _{41 – 49}}(WAV{E_t} \times WAV{E_t})\, + \,I{D_i} \\ & WAV{E_t}\, + \,WAV{E^2}_t\, + \,{ \in _{it}} \\ \end{aligned}$$
(2)
$$\begin{aligned} H{A_{dit}}\, = \,{\pi _{0i}}\, + \,{\beta _{1 – 5}}M{M_{it}}\, + & \\ & {\beta _6}PA{V_{fit}}\, + \,{\beta _{7 – 31}}{\text{(}}M{M_{it}} \times PA{V_{fit}}{\text{)}} \\ & {\beta _{32 – 37}}TV{C_{it}}\, + \,{\beta _{38 – 40}}TI{C_i} \\ & {\beta _{41 – 49}}(WAV{E_t} \times WAV{E_t})\, + \,I{D_i} \\ & WAV{E_t}\, + \,WAV{E^2}_t\, + \,{ \in _{it}} \\ \end{aligned}$$
(3)
Where: i represents the retired older adults in this study from individual 1 to 1,238. And t represents the sixteen waves of the HRS utilized for this study from years 2004 to 2020.
The dependent variable (HAdit) represents the disability, cognitive and physical functioining healthy aging domains (d) for individual i in wave t. MMit represents the five multimorbidity classes of cancer, cardiometabolic, musculosckeletal, neurological, and respiratory conditions for individual i in wave t. PALfit represents the frequency (f) of engaging in light PA intensities for individual i in wave t. PAMfit represents the frequency (f) of engaging in moderate PA intensities for individual i in wave t. PAVfit represents the frequency (f) of engaging in vigorous PA intensities for individual i in wave t. TVCit represents the time-varying covariates (age, SES, BMI, drinking and smoking status) for individual i in wave t. TICit captures the time-invariant covariates (race, sex, and marital status) for individual i. Where: π0i = β0 + µ0i represents the individual-level healthy aging intercept and random effects. ID is the random effect term for individual i, representing the deviation of individual i from the population mean. WAVEt and WAVE2t is the random effect term for time and point t, representing the deviation of time point t from the overall time trend considering that changes across time could be nonlinear. \({ \in _{it}}\) is the residual error term, representing the difference between the observed and predicted values of healthy aging.
Data management
After reshaping the data of the 1,238 individuals to its long format, the observations for the various models fell short of the supposed 11,142 total. Assuming that data are ‘missing at random,’ we invoked the Multivariate Imputation by Chained Equations (MICE) technique to address data missingness issues for the nine mixed effects models [48, 49]. Afterward, we verified that the sample statistics of the imputed data were quantitatively similar to that of the original data (See Tables S1 to S3 in the Additional File). Furthermore, we conducted sensitivity analysis by performing a complete case analysis for the nine models in this study. Therefore, obtaining similar results would bolster the robustness of our findings. All analysis were performed using STATA version 18 [50].
Add Comment