Variation in the quality and out-of-pocket cost of treatment for childhood malaria, diarrhoea, and pneumonia: Community and facility based care in rural Uganda

Background A key barrier to appropriate treatment for malaria, diarrhoea, and pneumonia (MDP) in children under 5 years of age in low income rural settings is the lack of access to quality health care. The WHO and UNICEF have therefore called for the scale-up of integrated community case management (iCCM) using community health workers (CHWs). The current study assessed access to treatment, out-of-pocket expenditure and the quality of treatment provided in the public and private sectors compared to national guidelines, using data collected in a large representative survey of caregivers of children in 205 villages with iCCM-trained CHWs in mid-Western Uganda. Results The prevalence of suspected malaria, diarrhoea and suspected pneumonia in the preceding two weeks in 6501 children in the study sample were 45%, 11% and 24% respectively. Twenty percent of children were first taken to a CHW, 56% to a health facility, 14% to other providers and no care was sought for 11%. The CHW was more likely to provide appropriate treatment compared to any other provider or to those not seeking care for children with MDP (RR 1.51, 95% CI 1.42–1.61, p<0.001). Seeking care from a CHW had the lowest cost outlay (median $0.00, IQR $0.00-$1.80), whilst seeking care to a private doctor or clinic the highest (median $2.80, IQR $1.20-$6.00). We modelled the expected increase in overall treatment coverage if children currently treated in the private sector or not seeking care were taken to the CHW instead. In this scenario, coverage of appropriate treatment for MDP could increase in total from the current rate of 47% up to 64%. Conclusion Scale-up of iCCM-trained CHW programmes is key to the provision of affordable, high quality treatment for sick children, and can thus significantly contribute to closing the gap in coverage of appropriate treatment. A description of the study context and site is provided elsewhere [20]. In brief, iCCM implementation has been supported in 8 districts in mid-Western Uganda by the Malaria Consortium (a leading not for profit international organization specialising in maternal and child health) through a grant from the Canadian International Development Agency (CIDA) since August 2010. Following this and prior to the roll-out of the inSCALE interventions, a cross sectional survey was undertaken across the 8 participating districts. To be eligible for the inSCALE study, a sub-county had to have VHTs trained in iCCM by 31st January 2011. We excluded sub-counties that contained less than 10 villages and those where other Malaria Consortium projects had operated to avoid respondent fatigue. This yielded an overall total of 41 eligible sub-counties active in iCCM for surveillance. Five villages, with an additional five back-up villages were randomly selected per sub-county for surveillance. A list of all households in each community was supplied by the local parish council and verified by the field supervisor. From this list thirty-two households per community were selected and surveyed at random between May and August 2011 in accordance with the sample size requirements for the evaluation of the inSCALE trial [20]. The analyses presented in this paper are based on data collected in this 2011 inSCALE cross-sectional survey. Data on socioeconomic and demographic characteristics of households, symptoms of the most recent illness episode in children under 5 years of age in the two weeks preceding the survey, care seeking behaviour, treatments received and details of all the self-reported out-of-pocket costs associated with care seeking for the episode of illness, were collected from the primary carer. The survey questionnaires were based on Demographic Health Survey (DHS)/Multiple Indicator Cluster Survey (MICS) child health and heath economics survey instruments used extensively in low and middle income countries [21, 22], and are provided in S1 Questionnaire File. Draft versions of the questionnaire were piloted on site and updated in an iterative process to ensure intended meanings were accurately conveyed. Pictures of locally available drugs for common childhood illnesses were used to increase the accuracy of recall of treatments received for sick children (S1 Fig). The questionnaires were delivered in the local languages of the region (Luganda, Luo and Runyakitara). To ensure consistency, 10% of household interviews were repeated by field supervisors within a week of the original and discrepancies resolved. All questionnaires were double entered into a dedicated database and differences between copies verified and corrected. Range and consistency checks were run on data to ensure that missing/incorrect fields were identified and flagged for resolution. Case definitions of episodes of suspected malaria, diarrhoea and suspected pneumonia, appropriate treatment for each of these diagnoses, care seeking and cost of care seeking were developed in accordance with standard WHO/UNICEF guidelines for treatment of malaria, diarrhoea and pneumonia in the community and at health facilities [10, 23–26]. Detailed definitions are provided in S1A–S1D Table. In brief, cases of suspected malaria episodes include all children with reported fever, excluding those with a negative malaria blood-test, suspected pneumonia episodes include all children with reported fast breathing and cough, and those with an episode of diarrhoea will have passed at least three watery stools in a 24hr period during the illness episode. The survey allowed for interviewees to record all the care seeking locations visited for the illness episode in question, and follow-on questions on transport and costs incurred whilst seeking care were asked for the first two locations visited. Definitions of out-of-pocket cost are provided in S1D Table. Analyses were based on caregiver-reported episodes of suspected malaria, diarrhoea or suspected pneumonia (this was the most recent episode in the two weeks prior to the interview). The difference in proportions of appropriate treatment between care seeking locations (by combining all episodes of MDP, and again separated by type of illness episode) was evaluated using logistic regression, with the VHT as the treatment group (or ‘exposure of interest’), and other care providers as the null group. Regression models were calculated using general estimating equations with an exchangeable correlation structure to account for sub-county clusters. Relative risks were derived from the regression models by use of the marginal standardisation technique, and the 95% CIs estimated with the delta method [27]. There are a range of demand-side characteristics that could influence the chances of appropriate treatment for sick children, including household income, parents’ education, rural versus urban location, age of parents and children, gender and others. We made the a priori decision not to force those of the above indicators for which we had data (mothers age, household income, education, religion, occupation, age and gender of child) into the final regression model as there is good evidence to show that these primarily operate via their influence on care seeking behaviour [28, 29]. The objective of this paper is not to determine the root causes of appropriate treatment, rather to ascertain how ultimate care seeking location influences the probability of appropriate treatment, and the implications for future scale-up of community case management programmes for childhood illnesses. We nonetheless provide a comparison of the demographic profiles of households by the care seeking location first visited for reference. In accordance with the age-specific guidelines for classification and treatment of children (S1A and S1B Table) the samples for analysis was restricted to children between the ages of 2 months and 59 months, or 4 months and 59 months in the case of those with suspected or confirmed malaria. These represent the age ranges for community-based treatment of children with MDP. After dropping data outliers, the household financial costs associated with seeking care were converted from Ugandan shillings to US dollars (using 2011 currency exchange rates [30]). Both median and mean out-of-pocket costs of care seeking were calculated to provide an indication of the spread of the data. We assessed the theoretical overall improvement in appropriate treatment rates that could be gained from diverting those sick children for whom no care was sought or who were taken for care to the poorest-performing providers, to instead seeking care to the VHT as the first port of call. For this exercise, the overall appropriate treatment rate observed was described mathematically as the sum of the products of the proportions of MDP episodes taken to each care provider type (first port of call), and the proportions of episodes appropriately treated at the provider type in question (Eq 1). In each scenario, the children not seeking care or taken to poorly performing providers (i.e. βi for each of such providers as below) were instead switched in turn to care seeking to the VHT, and the overall rate of appropriate treatment re-estimated. Pc- Predicted coverage of appropriate treatment β- Proportion of sick children (i.e. with fever, diarrhoea, or pneumonia) taken to provider i ϒ- Proportion of children appropriately treated at provider i n- Total number of provider types (including ‘no care seeking’) in study area As a reference for future evaluations of iCCM programmes we calculated intra-class correlation coefficients (ICCs) to describe the distribution of the variation amongst children within and between sub-counties with respect to their probability of receiving appropriate treatment. These were calculated using the unweighted analysis of variance estimator (‘loneway’ command) in Stata 13.1: loneway calculates an ICC as a function of the F-statistic from a one way analysis of variance of appropriate treatment rate with cluster ID as the only predictor [31, 32]. Data were summarised as tables or Microsoft Excel (Microsoft Corp 2010) graphs, and statistical analysis carried out using Stata version 13.1 (StataCorp Texas USA). Ethical approval was provided by the Higher Degrees, Research and Ethics Committee Uganda (IRB0011353: 2011), the Uganda National Council of Science and Technology (HS958: 2011), and the Ethics Committee of the London School of Hygiene and Tropical Medicine (ref 5762: 2010, 2011). Informed written consent was obtained from all participants.