Study setting
KG Halli is the field site of the Urban Health Action Research Project (UHARP) being implemented by the Institute of Public Health in Kadugondanahalli (KG Halli) since 2009. KG Halli was purposefully selected for the UHARP to study how access to quality healthcare could be improved in a poor urban community with a pluralistic healthcare system. A cross sectional survey was conducted to understand self reported illness and health seeking profile. The residents as well as healthcare providers in KG Halli have identified unaffordable healthcare expenses as one of the major issues in the area [
21]. The institutional ethics committee from Institute of Public Health, Bengaluru, India approved this study.
KG Halli is one of the 198 administrative units of Bangalore, a metropolitan capital of Karnataka. KG Halli has a population of over 44,500 individuals in an area of less than a square kilometer. KG Halli has two recognized slums and comprises of people from Karnataka as well as migrants from other Indian states. Majority of the population in the community are daily wageworkers. KG Halli has a mixed healthcare system with two government facilities run by municipal and state government and around 32 private healthcare facilities. Services offered by government facilities are heavily subsidized and, in principle, free for people living below the poverty line. Private facilities that include many single-doctor clinics and four hospitals work on fee-for-service basis.
Sample size
Considering the 8.6 % of overall prevalence of self-reported chronic condition as found in the earlier study in KG Halli [
7], 95 % confidence interval and 1 % of precision, we estimated the minimum sample size needed for our survey to be 3286 individuals. We added another 50 % of this number in order to cover for non-response. A few factors made us to account for high non-response rate.
The population of KG Halli comprises largely of migrants who often keep shifting their residence. In the course of our project activities and the earlier survey, we would find many households empty or closed for long time. Also the community in the area is weary of participating in surveys – as they often are approached by various agencies dealing with marketing of commercial products and/or as part of welfare projects/schemes. Majority of adults in the community are daily-wage workers who are often not at home during the day. There are many nuclear families where all the adults might be at work and so it’s likely that such households will not have an adult respondent at home when data collectors approached the houses. Considering the average household size of 4.7, we aimed to survey a minimum of 1047 households in KG Halli.
Data collection and measurements
As part of the UHARP project, a baseline census was conducted in 2009–2010 to understand the socio- demographics and health related aspects of the community. We conducted a follow-up survey in 2012–2013 in KG Halli to monitor changes in socio-demography, prevalence of self-reported illness, health-seeking and healthcare expenditure. The baseline survey revealed a high prevalence of self reported chronic conditions, especially that of diabetes and hypertension. These patients were incurring high out of pocket expenses from these conditions [
7]. As part of the action research project varied strategies were employed, three community health assistants were identified from the same community and were trained for over a period of one year. The community health assistants started conducting regular house-to-house visits creating awareness on chronic conditions in general and for diabetes and hypertension in particular. They directed them to appropriate healthcare services in the area. Periodic meetings with healthcare providers in the area were conducted to discuss health issues of the population identified form the baseline survey and also provide local solutions for the same. In this paper we selectively analyze these parameters in reference to chronic conditions. Community health workers collected data at household level using a structured questionnaire. They administered a questionnaire to available and willing family member aged 18 years or above. They selected every tenth family in a sequential order, starting from the Vinobhanagar area, a southern end of KG Halli. In case of refusal or unavailability of eligible respondent, the immediate next household replaced the household. The data collectors took informed verbal consent from the participants before administering the questionnaire. The completed questionnaires were verified and revisits to surveyed households were made on the following day for any corrections or missing data. The research team verified 10 % of the questionnaires from the survey. The methods for the survey including the tool for data collection were similar to those used for the baseline census. While we briefly outline methods used for this survey, kindly refer to the earlier publication [
7] for detailed data collection method.
Three binary outcome variables were defined for assessing prevalence of chronic conditions. These were the ‘absence’ (coded as ‘0’) or ‘presence’ (coded as ‘1’) of: i) any chronic condition, ii) diabetes and iii) hypertension. We considered a chronic condition to be present when a respondent reported having prescribed or taking medications on a daily basis for at least 30 days preceding the survey. A chronic condition is defined as an illness or impairment that lasts for a long duration. The minimum time period for an illness to be considered chronic varies depending on the source of definition, ranging from three months to one year [
7]. The names of chronic conditions were initially recorded using the lay terms reported by respondents and later categorized by the researchers, to the extent possible, into specific conditions. Based on the names of the reported chronic conditions, the presence or absence of diabetes and hypertension were also recorded.
Similarly, three binary outcome variables were defined to assess health seeking for chronic conditions. These were type of health services sought (‘private’ coded as ‘0’, ‘government’ coded as ‘1’) for: i) a chronic condition, ii) diabetes and iii) hypertension. For this study, we coded the outcome variable based on the nature of the health facility where the first consultation occurred. We compared values of these outcome variables with the findings from baseline census conducted three years ago.
Apart from comparison with the earlier study, we examined association of these outcome variables with a set of predictor variables. Predictor variables included sex (‘male’ or ‘female’), age (transformed into three age groups: (0- ≤ 40; > 40- ≤ 60; > 60), per capita income per month (as income quintiles), religion (‘Islam’, ‘Hindu’, and ‘Christian’) and the household poverty status (‘above’ or ‘below’ the poverty line) as established by the type of ration card (a proof of identity which establishes the economic status of a family) possessed by the household. While examining predictors for health seeking, we included an additional predictor in form of the tier of the healthcare services sought. Three tiers of healthcare services were defined based on where the person with a chronic condition was being managed at the time of the survey: i) ‘clinics/health centers’, ii) ‘referral hospitals’ with in-patient facilities and iii) ‘super-specialty hospitals’ attached to medical schools. Though there are overlaps in the provision of services across clinics/health centers, referral hospitals and super-specialty hospitals, they roughly correspond to primary, secondary and tertiary healthcare services, respectively.
Data analysis
The data were entered using EpiData Entry software 3.1 (The EpiData Association, Odense, Denmark). The data was checked for errors and missing values before being analyzed using STATA 11.2 (StataCorp, Texas, USA).
The prevalence of self-reported chronic conditions is reported as a percentage with 95 % confidence interval. To identify the predictors of self-reported chronic conditions, a multivariable logistic regression model was developed using all aforementioned predictors. The interaction between predictor variables was checked and two-way interaction terms that were significant at p < 0.05 were included in a multivariable logistic regression model. Similar to a backward elimination technique, the predictors that were not significant at p < 0.05 were then dropped sequentially while comparing models for goodness of fit (using a likelihood-ratio test) until no further improvement was possible. A similar process was used to develop the final multivariable models for all other outcome variables. Multi colinearity was assessed by using post-estimation commands. The final models are represented with the adjusted odds ratio (OR), 95 % confidence interval and p values.