Study design. A cross-sectional multilevel study was developed using secondary data from the Colombian population-based survey Health, Wellbeing and Aging (Salud, Bienestar y Envejecimiento (SABE)), carried out in 2015. The SABE study included 23 694 individuals aged over 60 years living in urban and rural areas of the 32 Colombian departments (i.e., states). Participants were selected by a probabilistic, multistage and stratified sampling design. Detailed information about the SABE study and the sampling method can be found elsewhere(25).
Variables. The dependent variable of interest was “Multimorbidity”, defined as the coexistence of two or more non-communicable diseases: diabetes, hypertension, cancer, lung disease (chronic obstructive pulmonary disease, asthma, bronchitis or emphysema), heart disease, stroke, joint diseases (arthritis, rheumatism or osteoarthritis), osteoporosis, depressive symptoms and other mental illnesses (mental or psychiatric problem). Except for depression, which was assessed with the Yesavage Geriatric Depression Scale (26), all morbidities were self-reported.
The independent variables of interest were both at individual and at group level. At individual level, the following dichotomous and continues sociodemographic characteristics were used: Sex (female / male(ref)), Age (continues), Marital status (single, widowed or divorced / others(ref)), Ethnicity (non-white / white(ref)), Income (mean income less than $7.83/day / income of $7.83/day / more than $7.83/day(ref)), Educational level (secondary or lower level / higher level(ref)), Victim of armed displacement, defined as ever been displaced by armed conflict or violence (yes / no(ref)), Area of residence (rural / urban(ref)). Health and lifestyle-related variables included were Functional limitation defined as having a Barthel score of <100 (27) (yes / no(ref)), Body Mass Index (BMI) (low (<22) / overweight/obesity (>27) / normal weight (22–27) (ref))(28), Physical inactivity or a low level score of physical activity in the short-form International Physical Activity Questionnaire - IPAQ-SF (29) (yes / no(ref)) and Smoking (current or former smoker / nonsmoker(ref)). Childhood-related factors were also included as Self-perceived childhood economic situation (regular / poor / good(ref)) and Self-perceived childhood health status (poor or regular / good(ref)).
At group level, deprivation was addressed with official reports of the multidimensional poverty index (MPI) (30) in 2018 (31) and infectious diseases mortality rate in 2016 (32) of each department. The MPI includes 5 dimensions with 15 indicators of education, childhood and youth conditions (school attendance, childcare services), employment (informality, long-term unemployment), health (access, insurance), access to public utilities (water source, sewer system) and housing conditions (floors and walls material, overcrowding) (30). Group deprivation characteristics were analyzed as continues variables.
Statistical analysis. Descriptive analysis of the individual characteristics was based on the absolute and relative frequencies with 95% confidence intervals (95%CI) for categorical variables, and measures of central tendency and dispersion (mean and standard deviation (SD)) for quantitative variables. To identify differences between baseline characteristics, independent X2 test and t-test analyses were developed for qualitative and quantitative variables respectively. Variables with p-values below 20% were considered for the adjusted models. A first one-level logistic model was used to evaluate the associations between the individual variables with MM, and significant variables (p<0.05) were included in a multilevel stepwise backward model. To evaluate the variability of MM prevalence across states, the median OR (MOR) (33) was calculated. Correlation was evaluated between MM and state deprivation variables using Pearson correlation coefficients. A two-level stepwise structural equation model (SEM) was used to simultaneously adjust the individual and contextual effects.