Predictors of health care use by adults 50 years and over in a rural South African setting

Background: South Africa’s epidemiological transition is characterised by an increasing burden of chronic communicable and non-communicable diseases. However, little is known about predictors of health care use (HCU) for the prevention and control of chronic diseases among older adults. Objective: To describe reported health problems and determine predictors of HCU by adults aged 50+ living in a rural sub-district of South Africa. Design: A cross-sectional study to measure HCU was conducted in 2010 in the Agincourt sub-district of Mpumalanga Province, an area underpinned by a robust health and demographic surveillance system. HCU, socio-demographic variables, reception of social grants, and type of medical aid were measured, and compared between responders who used health care services with those who did not. Predictors of HCU were determined by binary logistic regression adjusted for socio-demographic variables. Results: Seventy-five percent of the eligible adults aged 50+ responded to the survey. Average age of the targeted 7,870 older adults was 66 years (95% CI: 65.3, 65.8), and there were more women than men (70% vs. 30%, p<0.001). All 5,795 responders reported health problems, of which 96% used health care, predominantly at public health facilities (82%). Reported health problems were: chronic non-communicable diseases (41% - e.g. hypertension), acute conditions (27% - e.g. flu and fever), other conditions (26% - e.g. musculoskeletal pain), chronic communicable diseases (3% - e.g. HIV and TB), and injuries (3%). In multivariate logistic regression, responders with chronic communicable disease (OR=5.91, 95% CI: 1.44, 24.32) and noncommunicable disease (OR=2.85, 95% CI: 1.96, 4.14) had significantly higher odds of using health care compared with those with acute conditions. Responders with six or more years of education had a two-fold increased odds of using health care (OR=2.49, 95% CI: 1.27, 4.86) compared with those with no formal education. Conclusion: Chronic communicable and non-communicable diseases were the most prevalent and main predictors of HCU in this population, suggesting prioritisation of public health care services for chronic diseases among older people in this rural setting. The study used data from the MRC/Wits Agincourt Research Unit situated in Ehlanzeni Health District, Mpumalanga Province, South Africa. Trained local field workers collect and update vital events (births, deaths, and migration) on a yearly basis since 1992. This is complemented by additional information at different time intervals such as on HCU, education level, labour, migration, and household assets (22). The current population under follow-up in the MRC/Wits Agincourt Research study site as on 1st July 2011 was approximately 90,000 people in 16,000 households living in 27 villages (23). The study area covers about 420 km2. Tsonga is the most widely spoken language. One-third of the population are of Mozambican origin, having immigrated into South Africa mainly as war refugees in the early- and mid-1980s. Despite the current government’s development initiatives, which have led to improved housing, access to potable water, electricity and social security grants, infrastructure in the area is still limited. Unemployment rates remains high, with 60% of labour migration being accounted for by men aged 35–54 years and an increasing proportion of labour migrants seen among young men and women (22). The pattern of labour migration has resulted in a disproportionately higher proportion of older women permanently resident in the area. There are eight health facilities in the study area: one public health centre, six government satellite clinics, and one private community health centre in a public–private partnership. There are three referral hospitals situated 25 and 45 km from the study setting (23). The government of South Africa decentralised the provision of health services by dividing the country into 53 health districts to ensure that citizens have access to a comprehensive package of PHC and district hospital services. In the South African PHC model, the nurse is the provider of services in the clinics and comprehensive health centres – which are the first point of entry to the health system. Located within the reach of rural, semi-rural, and urban communities, these facilities are the cornerstone of the public health system through provision of comprehensive and integrated ‘preventive, promotional, curative and rehabilitation services’ (24). The range of services includes maternal and childcare, immunisation, family planning, treatment of sexually transmitted infections, minor trauma, and care for chronic diseases (e.g. diabetes and hypertension). Additional services provided by the health centres include 24-hour maternity services, accident and emergency services, up to 30 beds for observation for a maximum of 48 hours, a procedure room (not an operating theatre). With the exception of emergency cases which are referred to the hospitals (secondary level of care), the clinics and health centres offer services to ambulatory patients for 8 hours/day and 24 hours, respectively (24). This was a cross-sectional study of all eligible older adults aged 50 years and older in the study site. Of the total 10,249 older adults registered in the 2009 census database, 7,870 persons with permanent residency status were eligible and targeted for the 2010 HCU survey. Eligibility criteria for the interviews were 1) residency status of 21 days or more before the survey for those prospective participants who moved out of the study site after the 2009 census and relocated to the study site before the 2010 survey and 2) availability of the prospective participants at home after two revisits by field workers (Fig. 1). Sampling of eligible study participants. Field workers were trained for two days in the administration of the HCU questionnaire, as part of preparation for the general census. Field work was closely supervised for a week, after which a new training session was run to review and tackle challenges. Quality control followed a four-step system where field workers, supervisors, quality checkers, and data entry clerks assured good quality of the data: 1) The field workers double-checked all questionnaires before leaving the household of the interviewee, and again at the office before submission to the supervisors. 2) The supervisors randomly checked the questionnaires for inconsistencies and blank questions before submitting to the quality checkers. 3) The quality checkers identified inconsistencies and other errors in the questionnaires before submitting to the data entry clerks. 4) Data entry clerks identified forms with errors during data entry and returned them to the field for correction, after which the whole process of quality control was engaged prior to final data entry. The questionnaire for the 2010 adult HCU survey was based on a HCU questionnaire used previously in the site to gather information on the older adult population (18). The questionnaire was used to collect information on socio-demographic variables, reception of any type of social grant, access to medical aid, need for and access to health care, type of disease, disability and hospitalisation. Age in years was calculated on 1st August 2010 using the census date of birth for all potential participants. Responders were then categorised into 10-year age intervals: 50–59, 60–69, and 70+. Years of formal education were obtained from the 2007 MRC/Wits Agincourt Research Unit database, which was the latest updated information. Years of education were categorised according to the WHO levels of education: no formal education, <6 years, and ≥6 years. Medical aid was categorised to reflect responders with: 1) medical aid to visit the doctor, 2) health insurance for specific disease, 3) medical aid in employer’s clinic/hospital, 4) access to free public hospital care, and 5) no medical aid/do not know. The variable ‘last time health care was needed’ was categorised into: 1) <1 year, 2) 1–3 years, 3) >3 years, and 4) never. In order to minimise errors due to recall bias, analysis of the predictors of HCU was restricted to responders who reported needing health care less than one year preceding the survey. The justification for using less than one year as the cut-off was based on the assumption that it is easier for responders to recall experiences with HCU in <1 year than in 1–3 years or >3 years. Due to the influx of Mozambican refugees into Agincourt sub-district, nationality of origin was grouped into South Africans and Mozambicans. Socio-economic status (SES) was constructed from a household asset score in the 2009 census data. A principal component factor analysis technique was used to construct SES based on 30 variables on access to water and electricity, type and size of dwelling, appliances, ownership of livestock, and transport available. Subsequently, responders were categorised into quintiles in the ascending order of lowest, middle low, middle, middle high, and highest SES (25). In order to ascertain employment status, the variable ‘looking for a paid job’ was categorised as yes or no. Reception of social grant was recorded as none, old age and disability. Hospitalisation, HCU (defined as the need for and access to health care, at least once, less than one year before the HCU survey in 2010) and disability were all binary variables (yes vs. no). Apart from acute conditions (fever and flu) and chronic communicable disease (HIV and TB), other types of illness were generated by recoding the reasons for visiting health facilities into NCD (hypertension, diabetes, stroke, sleep disorder, chronic pain in joints, depression, anxiety, cancer, and heart problems), injuries and others type of illness (musculoskeletal pain and nutritional deficiency). Actions taken during an illness episode less than one year before the survey were: visiting public and private health facilities. Other actions included practicing self-medication, consulting faith/traditional healers, and taking no action. Validation checks were done during data entry in MicrosoftSQL server 2005 database. Data were extracted into Stata 12.0 (College Station, TX, USA) for statistical analysis. At p-value of 0.05, bivariate analysis compared responders who used health care services with those who did not. The cut-off point for univariate binary logistic regression analysis was set at p-value of ≤0.2 and variables that were significantly associated with HCU were used to model the multivariate binary logistic regression analysis (p-value ≤0.05). Multiple imputation by chained equations (MICE) approach for categorical variables was used to impute for ‘socio-economic status’ and ‘looking for a paid job’, which had 1.3% and 12.4% missing values, respectively. Multiple imputation is a simulation-based method for analysing incomplete variables. It predicts missing values as close as possible to the true ones by replacing missing data with probable values based on other available information (26). Imputation is considered to have less estimation bias and valid statistical inference than list-wise deletion because the latter leads to loss of statistical power (27). Ethical clearance for the MRC/Wits Rural Public Health and Health Transitions Research Unit (Agincourt) was granted by the Committee for Research on Human Subjects (Medical) of the University of the Witwatersrand, Johannesburg, South Africa (Ref No. M960720).