The Republic of Malawi is a small sub-Saharan African country bordering Zambia and Tanzania to the north and partly embedded into Mozambique to the south. The total population in 2011 was 15.4 million (20). Malawi is among the poorest and the least developed countries in the world, ranking 171 out of 187 countries in the Human Development Index (20). Eighty percent of the population lives in rural areas, and approximately 74% lives below the poverty line and 40% in severe poverty (20). The life expectancy at birth is 54.2 years, and the infant and under-five mortality rates are 53 and 110 per 1,000 livebirths respectively (20).

The transmission of malaria is perennial in Malawi, with a peak in the rainy season from November to April (21). The transmission of acute respiratory infections peaks in the beginning of the cold dry season in April to June (22).

We conducted a national household survey from the end of February to mid-April 2005, using a cluster-sample design with compact segments (23,24). We aimed for a nationally-representative sample of 600 children, equaling 3,000 households. Sample-size calculation took into account the expected proportion of interest (i.e. the proportion of children with cough and fever seeking care from a trained healthcare provider), the level of precision desired, and the expected design effect (23). Malawi's 1998 Population and Housing Census provided a complete list of enumeration areas (EAs) (25). The projected number of households in each EA was calculated based on census data, then divided by 100 and rounded to the nearest integer, which was considered to be the size of the EA in segments. This predetermined segment-size was based on the prediction that the survey team could complete interviews in 100 households per day. Subsequently, 30 EAs were chosen by systematic sampling with probability of selection proportional to the size of each EA in segments (23). When the survey team arrived at an EA, it was divided into the predetermined number of segments, using sketch maps, such that each segment in the EA had approximately the same number of households. Among these segments, one segment was then selected at random (23). All households in the segment were visited by interviewers who interviewed all primary caretakers of children aged less than five years; if found absent, households were revisited later the same day. Data were entered into Personal Digital Assistants (PDAs), using Visual Basic^® software (version 6) (Microsoft Corporation, Redmond, WA, USA) and analyzed with the SAS-callable version of SUDAAN^® software (version 9.01) (Research Triangle Institute, Research Triangle Park, NC, USA) to take account of the cluster-sampling design. Data collection was performed by one survey team of 6 interviewers and 2 supervisors.

In total, 2,697 households were selected for the survey and visited by survey teams. Figure 1 shows a flowchart of inclusion and exclusion of study participants. Interviews were completed in 2,610 (96.8%) of the households; reasons for non-completion are listed in Figure 1. Among the surveyed households, 871 households had no primary caretaker. Thus, a total of 1,787 primary caretakers were identified; 1,669 (93.4%) of them were eligible for the survey, i.e. had at least one child below five years of age and were interviewed.

The first part of the questionnaire recorded the demographic and socioeconomic characteristics of the household, including access to safe water, and was completed by all caretakers identified for the survey (26). The second part focused on illness symptoms and healthcare utilization among children aged below five years. This part was only completed if the child had cough and fever during the two weeks preceding the survey, which was determined by a screening question in the beginning of Part II of the questionnaire. A third part of the questionnaire was completed if the child had diarrhoea during the two weeks preceding the survey

Healthcare utilization (dependent variable): We identified 12 different types of healthcare providers in Malawi and defined healthcare utilization as primary visits to any of those. Providers were categorized as being trained or untrained based on whether or not they held a formal medical degree or medical qualifications. Trained providers were further categorized as working in private or government healthcare facilities (Figure 2).

Age of children: The following 5 age categories were used: <12, 12-23, 24-35, 36-47, and 48-59 completed months. In the analyses, age categories were dichotomized into <12 months and 12-59 months of age due to statistical uncertainty around the small number of subjects in 5 categories.

Education of mother: Mothers were asked if they had ever attended school, with ‘yes/no’ answers.

Wealth index: We defined socioeconomic status in terms of assets. Data on household assets were combined into a wealth index by principal components analysis. Households were assigned a standardized score for each asset, such as drinking-water source, type of toilet facility, electricity, and vehicles, according to a Malawi-specific index (27). The scores were summed for each household. Households were ranked and divided into terciles. The distribution of the summed up asset scores was right-skewed, i.e. the bulk of households had asset scores below the mean value. Consequently, if divided into quintiles, the lowest wealth quintiles would only slightly differ from one another in terms of asset scores, and a significant number of households with identical asset scores would group into different wealth quintiles due to the heaping of scores. Therefore, we opted to divide the sample into terciles. In the analyses, wealth quintiles were dichotomized into the poorest (the two lower terciles) and the wealthiest (the upper tercile) due to statistical uncertainty around the small number of subjects in three categories.

Other demographic variables: Child's sex (male/female) and place of residence (urban/rural) were also included.

Severity of illness: In our screening question, we identified children who had cough and fever during the two weeks preceding the survey. Additional symptoms, reported by caretakers, were collected only for children who had both of these symptoms. Malawi's Ministry of Health has adopted the guidelines of the World Health Organization's Integrated Management of Childhood Illness (IMCI) for the case management of childhood illnesses at the first-line health facilities (28). We applied the IMCI classifications, with modifications, to categorize the symptoms of illness. We defined two illness categories each comprising two subcategories: moderately-severe illness comprising “malaria without pneumonia” and “malaria with pneumonia”; and severe illness comprising “severe pneumonia or very severe febrile disease”, and “possible serious bacterial infection.” Children aged 2 months up to 5 years with cough and fever without any danger signs were classified as having “malaria without pneumonia”, and children with cough and fever accompanied with fast breathing but without any danger signs were classified as having “malaria with pneumonia.” Children aged 2 months up to 5 years, with cough and fever and any of the following danger signs: being unable to eat or drink, convulsions, lethargy, unconsciousness, and/or stiff neck, were classified as having “severe pneumonia or very severe febrile disease.” Children aged 1 week up to 2 months, with cough and fever, were classified as having “possible serious bacterial infection.”

Direct and indirect costs: Costs associated with seeking care were measured in both monetary terms, i.e. direct costs, and in terms of time, i.e. indirect costs. The direct costs included travel costs, user fees for consultation and/or hospitalization, and the costs of drugs. Direct costs were measured as the sum of out-of-pocket payments in local currency—the Malawi Kwacha (MWK). The official exchange rate in 2012 was 1 US$=269.5 MWK (29). Indirect costs were measured as time spent travelling and waiting when seeking care. The opportunity cost, i.e. the costs of activities forgone due to time spent on seeking care, was assumed to be lost income. Hence, time spent on seeking care was transformed into lost income per minute as its monetary equivalent, using income data from the Malawi Integrated Household Survey (30). Waiting time was assumed to be 75 minutes for all caretakers, in accordance with other studies (31).

The study received ethical approval from the Ethics Review Committees of the College of Medicine, Malawi and the Centers for Disease Control and Prevention, Atlanta, GA, USA, prior to the enrollment of participants. Informed consent was obtained by the research assistants for each participant prior to his/her enrollment in the study. Informed consent for this study conformed to the 45 Code of Federal Regulations, Part 46.101c and 46.102d requirements.

We used univariate analyses to assess the effect of the six predictive variables (child's age, sex, education of mother, wealth index, place of residence, and illness severity) on seeking care from a trained healthcare provider compared to seeking care from an untrained provider or not seeking care at all. Only primary healthcare visits were included and, thus, successive visits were excluded from analysis. Odds ratios (ORs) and 95% confidence intervals (95% CI) were calculated for each variable, and the p value of a Wald F-test was evaluated to determine whether associations were significant (p<0.05).

All six variables and all two-factor interaction terms were entered into a multiple logistic regression model. This model was reduced to the final model by means of a hierarchical backward elimination procedure. Significance was determined by evaluating the p value of Wald F-test (p<0.05). If an independent variable was not significant but, when removed, it changed the OR of a significant variable by ≥10%, and it was retained in the model to adjust for confounding effect; t-tests were used for comparing means in travel time and direct costs incurred when seeking care with regard to wealth index, place of residence, and type of healthcare provider visited. Furthermore, we calculated the total cost each caretaker incurred when seeking care and investigated total cost relative to income as described elsewhere (31).

A trained healthcare provider was visited for 48.0% (n=208; 95% CI 41.0-55.1) of the illness episodes that had occurred among the surveyed children in the two weeks preceding the survey. An untrained provider was visited for 37.9% (n=164; 95% CI 31.2-45.1) of illness episodes. The remaining 14.1% (n=61; 95% CI 10.7-18.4) of illness episodes did not result in a healthcare visit. The distribution of healthcare providers visited is illustrated in Figure 2. The most common providers worked in private health centres (n=60; 13.9%; 95% CI 10.0-18.8), were street vendors (n=60; 13.9%; 95% CI 10.0-18.8), or worked in pharmacies (n=49; 11.3%; 95% CI 8.3-15.2).

In univariate analyses, child's age and wealth index of the household were associated with seeking care from a trained healthcare provider compared to seeking care from an untrained provider or not seeking care at all (Table 2). Children aged 12-59 months (OR 0.52; 95% CI 0.30-0.90) and children living in the poorest households (OR 0.51; 95% CI 0.30-0.87) were less likely to be taken to a trained healthcare provider when ill compared to older children and those living in wealthier households respectively.  Children with moderately-severe illness (OR 0.72; 95% CI 0.47-1.12) were less likely than children with severe illness to visit a trained healthcare provider; however, this association was not statistically significant.

In the multiple logistic regression model, child's age and wealth index continued being associated with seeking care from a trained healthcare provider (Table 2). The effect of potential confounding variables was assessed, and no confounders were found. Similarly, no significant interactions were observed.

Results of t-tests demonstrated that time spent travelling when seeking care was significantly associated with place of residence and type of healthcare provider visited (Table 3). Children living in rural areas spent more time travelling than those in urban areas (mean: 68.9 minutes vs 14.1 minutes) when visiting any type of healthcare provider. In addition, children who visited a trained healthcare provider spent more time travelling than those visiting an untrained provider (mean: 76.8 minutes vs 26.2 minutes).

Furthermore, living in the wealthiest households, living in an urban area, and obtaining care from a trained healthcare provider and/or from private healthcare facilities were significantly associated with higher direct costs (Table 3). However, when estimating total cost relative to income, caretakers living in rural areas spent a higher proportion (3.2%) of their monthly income when seeking care than caretakers living in urban areas (3.0%). This result was not significant (Table 4).

To determine the representativeness of children for whom cost data were available, we compared those with children for whom data were missing by using chi-square tests. Our results indicated that a higher proportion of children for whom data on travel time were available was living in the wealthiest households (52.5%; p=0.04) and was taken to a trained healthcare provider (65.4%; p<0.001) compared to children who had missing data on travel time (47.5% and 34.6% respectively). Furthermore, a higher proportion of children who had data on direct costs had an educated mother (56.6%; p<0.01) whereas a lower proportion was taken to a trained healthcare provider (48.1%; p<0.001) compared to children who had missing data on direct costs (43.4% and 51.9% respectively).

Socioeconomic inequities in healthcare are still evident in Malawi and, therefore, addressing barriers to healthcare is a key priority for improving child health and achieving the Millennium Development Goal to reduce child mortality (18). One of the first responses should be improved access of the poorest households to healthcare services. This could be accomplished by reducing logistical and financial barriers. Geographic availability of healthcare facilities could be improved through targeted interventions, increased availability of health facilities in rural areas, and improved infrastructure (2,6,18,31). This has been recognized by the Government of Malawi, which increased the number of rural health centres from 219 to 258 between 2003 and 2010 (11). Furthermore, basic training of untrained healthcare providers might ensure higher standards of care and referral from these sources. In addition, financial access to care from the private health sector could be improved by reducing costs. This could be achieved by implementation of a fair user-fee system where the poorest people are guaranteed access and treatment through a waiver system, health insurance programmes, or payment by installments (36,43). Since 2005, the Government of Malawi has sought to reduce financial barriers to healthcare by introducing service-level agreements with the CHAM's facilities to allow free access to maternal and child health services. By 2010, seventy-two of 150 private facilities had such free access (11). Other priorities could be community-based interventions to improve healthcare-seeking patterns. Health communication efforts could reinforce caretakers’ ability to perceive the severity of illness and initiate the appropriate response (2,19,38). Finally, it should be emphasized that multisectoral strategies to reducing poverty and socioeconomic inequities are crucial to the success of specific interventions as recognized by the Government of Malawi (2,36).

Future research on healthcare performance by different care providers in Malawi would improve our evaluation of the likely benefits or hazards of observed utilization patterns. Continued research on the trends in healthcare utilization will help assess the success of the initiatives made by the Government of Malawi and the development partners.

Our study had a number of limitations. The cross-sectional study design precluded the identification of a causal relationship between predictor and outcome variables. In addition, relying on retrospective self-reports increased the susceptibility to recall bias and social desirability bias. A two-week recall period was used in minimizing the problem of recall. Our illness classifications were presumptive and, as shown elsewhere, some symptoms are not recognized by caretakers, which may have led to misclassification (33,42). It should also be noted that we only studied a subset of pneumonia and malaria episodes according to the IMCI guidelines, namely those with cough and fever. Another limitation was that the study was conducted during the rainy season in Malawi and, thus, the prevailing illness and healthcare utilization patterns could be different from the dry season. Finally, missing of data in the cost analyses was a limitation of our study, and children for whom cost data were available were not representative of all children. Therefore, our cost results should be interpreted with caution.

We demonstrated that, in a poor African country, like Malawi, significant inequities exist in obtaining healthcare from a trained provider for childhood illnesses and in the costs associated with seeking care. Continued efforts to reduce these barriers are needed to narrow the gap in health and healthcare equity in Malawi.