Study population and setting
This cross-sectional study was conducted in two communities of similar size (~ 15,000 population) and demographics in rural, western Kenya. These two communities were chosen based on the level of their engagement with AMPATH’s BIGPIC program at the time of the study. AMPATH is a partnership between Moi University, Moi Teaching and Referral Hospital, North American universities led by Indiana University, and the Kenyan Government with a mission of teaching, care and research. BIGPIC began implementation in Community 1 in 2016 but had not yet rolled out to Community 2 at the time of our data collection.
We drew our sample from three different study populations. First, we sampled 100 BIGPIC Family microfinance group members, by targeting the full rosters of 7 groups randomly selected from all groups with at least 6 months of active membership in Community 1. Second, we identified a sample of 100 people in Community 1. Third, we identified a sample of 100 people in Community 2. Both community samples were randomly sampled from an enumeration of all community residents developed for a recent hypertension study in the area . Only participants age 18 years and above were eligible. Participants were not eligible to be interviewed twice if they happened to be sampled for both the microfinance group and community sample in Community 1. An a priori power calculation was conducted for our desired sample size of n = 300. For common outcomes (at 50% prevalence), we maintained at least 80% power to detect all prevalence ratios greater than 1.3. With less common outcomes (at 10% prevalence), the effect sizes necessary to maintain 80% power increased up to 2.2. Thus, we were well-powered to detect even small effects with common outcomes, but underpowered to detect small effects with uncommon outcomes.
The microfinance groups are the platform through which much of the BIGPIC programming is delivered. Individual members meet regularly to take out and repay group-funded loans, and they may choose to receive the full complement of other available interventions, including: 1) screening and care for diabetes and hypertension and 2) NHIF education. Some intervention spillover to non-microfinance community members is expected if they engaged in health screening or NHIF training at community-wide events.
Between November 2018–February 2019, trained local fieldworkers collected survey data from 300 participants. All data were self-reported using a tablet-based, quantitative survey administered in the local language of Kibukusu with REDCap software. The survey covered a range of topics including socio-demographic information, microfinance group experience, and health screenings (see survey instrument in Supplemental Material). All interviews were conducted at home in a private area, after obtaining informed consent. Ethical approval was provided by the Indiana University Institutional Review Board (#1705661852) and the Moi University Institutional Research and Ethics Committee (#00030702).
BIGPIC Family microfinance group membership was the primary exposure. Participants who self-reported being a current member of a BIGPIC Family microfinance group with a start date of at least 6 months prior to the interview date were considered exposed. As a sensitivity analysis, we also considered duration of participation as a continuous variable, based on the difference between start date and interview date.
The unexposed group was composed of non-microfinance group members of both Community 1 (the community in which the program had already rolled out) and Community 2 (the community in which the program had not yet rolled out). To evaluate whether to combine these communities for analysis, we assessed whether there were differences between them with respect to our key health outcomes and socio-demographic characteristics. We found no statistical differences between the two communities for our key outcomes, and no differences in socio-demographic characteristics with the exception of slightly different wealth index distributions (Supplemental Table 1).
We operationalized NHIF health insurance coverage as a potential outcome of BIGPIC Family group membership. Although NHIF education was provided through BIGPIC microfinance groups, the groups have not historically been used to directly enroll participants. Participants were asked to self-report if they had active NHIF coverage, which was confirmed with SMS messages to an NHIF information number.
As our primary health outcomes, we queried if participants had ever been screened for each of the following health conditions: HIV, diabetes, hypertension, tuberculosis, and cervical cancer. BIGPIC Family routinely provides screening and care for some conditions (diabetes and hypertension), but not others (HIV, tuberculosis, cervical cancer). We also assessed two disease management outcomes among the small subset of participants who reported a diagnosis of hypertension, diabetes, or HIV (n = 32): 1) current medication for condition, and 2) healthcare visit to manage condition within the last 6 months.
We also collected sociodemographic data on: sex, age, marital status, educational attainment, employment outside the home, and socio-economic status (SES). SES was operationalized by querying household assets. We calculated a weighted index of ownership of 20 items, then categorized participants in quartiles, aligned with the methodology of the 2014 Kenyan DHS) .
To understand the relationship between microfinance group membership and each of the health insurance, health screening, and disease management outcomes, we used log-binomial models to estimate prevalence ratios. We compared unadjusted results to results adjusted for age, sex, and marital status. To explore potential differences by sex and by SES, we ran models stratified by sex and by household assets dichotomized at the median. In the case of ‘0’ cell counts in the stratified analyses, we added 0.5 to each cell. To assess whether there were statistical differences in effect sizes by sex and SES, we introduced an interaction term between each of sex and household assets with microfinance group membership. Interaction terms with Wald p-values < 0.05 indicated statistically significant differences.
Since people who select into microfinance groups are likely to have different personality profiles compared to those that do not and this may also correlate with engagement in health screenings and care, we conducted a sensitivity analysis examining the relationships of interest among a study population restricted to only include microfinance group members. We then used duration of group membership as a proxy for the exposure, with a cutpoint dividing short-term members (< 12 months) from long-term members (greater than or equal to 12 months). We compared the prevalence of our health outcomes between short and long-term members to assess whether the results were of similar magnitude to our main findings.