Network variables
To derive the TSNs we used the structural approach [5]. The structure and interactional variables included the total number of family members, having a life partner, children, friends, belonging to some community group, and the frequency of contact with family and friends. The total number of family members was obtained by adding all of members who lived under the same roof and outside the older adult's home (children, parents and close relatives) with each other.
Other variables included having a life partner, having children, having friends and belonging to a community group (groups for the elderly, religious, etc.). These were characterized as 0 = no and 1 = yes.
The frequency of personal contacts with friends was measured on a scale ranging from 0 to 5: 0 = no contact, 1 = less than one contact per month, 2 = monthly contact, 3 = contact every two weeks, 4 = weekly contact and 5 = daily contact. The frequency of personal contacts with the family members was obtained by adding the contacts with relatives that live inside (0-5 scale) and outside (0-5 scale) the house of the older adult, therefore, the scale was 0 to 10.
Other network variables included frequency of support received to perform BADL and IADL, which was measured using a scale of 0-5: 0 = absent, 1 = less than monthly frequency, 2 = monthly, 3 = every two weeks, 4 = weekly and 5 = daily. Economic support was interpreted as the monetary and in-kind (groceries, meals, groceries coupons) support that the older adult was receiving from the network members. The categories for economic support were 0 = did not receive economic support, 1 = received economic support. Satisfaction with family and satisfaction with friend's contacts was measured on a scale of 1 to 3, where: 1 = unsatisfied; 2 = more or less satisfied; 3 = totally satisfied.
Functional dependency and variables to characterize the sample
Functional dependency in basic and/or instrumental activities of daily living (BADL and/or IADL). The BADL were: lying down and getting of the bed, walking from the bed to a chair and vice versa, walking within the house, dressing/undressing, bathing, eating, using the toilet and maintaining sphincter continence. The IADL were using the telephone, shopping, preparing food housekeeping, doing laundry, mode of transportation, responsibility of own medication and ability to handle finances.
To ascertain functional dependency we identified whether the older adult had difficulty to perform BADL and IADL activities by applying the tests of Barthel [26], Katz [27] and Lawton [28]. For the purpose of the analysis, we considered as an independent older adult, a person who did not have limitations in performing BADL activities and had no limitations in up to one IADL activity. A dependent older adult was a person who had limitations in performing one or more BADL and two or more IADL activities. Functional dependency was classified as: 1 = dependent and 0 = independent.
Variables to characterize the sample: a) General characteristics of the older adults such as sex: men and women; age divided into two strata: 60-74 years and ≥ 75 years; schooling; geographic area of residence; 1 = Central (Mexico City), 2 = South (South Veracruz), 3 = West (Nayarit) and 4 = North (Durango).
b) Lifestyles: physical activity classified as 0 = regular (practicing exercise 3 or more times a week, ≥ 30 minutes every time), 1 = irregular or physical inactivity. Diet classified as 0 = healthy diet (daily consumption of fruits, vegetables and dairy products), 1 = unhealthy diet (no consumption or irregular consumption of any of these three groups of food). Smoking: 0 = no (non-smoker during lifetime), 1 = yes (smoker and/or smoked during lifetime).
c) Self-rated health that was measured with the question: In general, would you say your health is: 1) excellent; 2) very good; 3) good; 4) regular 5) poor? The answers allowed classifying the older adults in two types: positive self-rated health = 0, which comprised the categories excellent, very good and good, and negative self-rated health = 1 that included the categories regular and poor.
d) Medical history: number of chronic diseases; depression, which was measured using the Yesavage Geriatric Depression Scale [29] and was classified as 0 = absent, 1 = probable depression, 2 = established depression.
Statistical analysis
The relative and absolute frequency of categorical variables was determined, as well as the average and standard deviation of continuous variables.
The TSNs were derived from the structure and interactional variables (total number of family members, having a life partner, having children, having friends, belonging to a community group, and frequency of contact with family and friends) by using cluster analysis. Two clustering techniques were applied (hierarchical and non-hierarchical) [6, 7]. To carry out the cluster analysis first, the variables were standardized to eliminate the scale difference effects. Then, we performed a hierarchical clustering procedure using the Ward's minimum-variance method and multiple criteria available, such as pseudo-F statistics, pseudo-t2 statistic and Sarle's cubic clustering criterion to confirm the ideal number of clusters. Next, a non-hierarchical k-means cluster analysis was performed. These techniques allowed identifying the characteristics of homogeneous subgroups to establish the TSNs.
After deriving the network types, we examined the main characteristics of the older adults in each TSN, including the instrumental and economic support they received and their satisfaction with the network. The differences in means of continuous variables among social networks types were compared using one way analysis of variance (ANOVA) followed by a Bonferroni multiple-comparison test, and the differences in proportions for categorical variables among social networks types were compared with the chi-square test.
To evaluate the crude association between functional dependency and the TSN, and the association among dependency and the other covariates, we performed a bivariate analysis that allowed estimating the crude prevalence ratios (PR) and 95% confidence intervals (95% CI).
Due to the controversial results of previous studies regarding the existing differences in the relationship between social networks and functional disability of older adults depending on sex and age, we also determined whether age and sex were confounders or effect-modifiers by applying the stratified Mantel-Haenszel statistics. It was found that these variables were confounders but not effect-modifiers so only analysis for the whole sample was presented.
To obtain the adjusted association between functional dependency (dependent variable) and the TSNs (independent variables), a multiple regression (Poisson regression with robust variance) analysis was performed using the forward selection method. This statistical model estimates directly the PR and 95% CI as better alternative for logistic regression in cross-sectional studies with binary outcomes, because using odds ratio in a cross-sectional study would overestimate the risk ratio (or PR) when the outcome of interest is common (>10%) [30]. A stage entry of variable blocks was used. First, the TSN variables were entered; then, the block of socio-demographic variables and medical history; next the block of lifestyle variables. The variables entered in the model were relevant to the outcome variable, according to the literature [19, 20, 22, 23], and in the bivariate analysis the p value had to be ≤ 0.20. After each entry, the variables within the model were tested for removal based on a p value >0.05.
All analyses were performed with the statistical program Stata (Version 8.1) (Stata; Stata Corp., College Station, TX).