Understanding the context of healthcare utilisation for children under-five with diarrhoea in the DRC: based on Andersen behavioural model

This study used data from the Multiple Indicator Cluster Surveys (MICS) carried out in the DRC in 2018. The surveys were conducted by the National Statistical Office (NSO) and United Nations International Children’s Emergency Fund (UNICEF), in collaboration with relevant DRC Ministries. During data collection, following UNICEF guidelines, written informed consent was obtained from all participants (mothers) following an explanation of study objectives, assurance of the confidentiality of their identity, and a guarantee that there was no disadvantage to not participating in the study. All data were recruited anonymously via study identification numbers. A two-stage sampling method was used. At the first stage, three strata within each province, except Kinshasa, were created. Within each stratum, primary sample units (PSU) were selected with probability proportional to population size. In the second stage, 30 households were drawn from each of the 721 clusters, with 21,630 households selected in total. Details of these MICS have been previously described elsewhere [36]. For this study, MICS data were accessed and analyzed with the authorization of UNICEF.

The MICS were designed to collect updated information on the situation of women and children nationally. Due to the intrinsic structure of the questionnaire, only women who had children under 5 years old were included, a total of 20,245 participants. In our study, we focus on 2918 children who had diarrhoea in the last 2 weeks. Of that group, 292 participants were excluded due to missing values in maternal age, child age, sex of the child, marital status, maternal educational attainment, maternal smoking status, maternal alcohol use, wealth index, children’s health insurance, residence region, dyspnoea in the last 2 weeks, fever or cough in the last 2 weeks, seeking advice or treatment for diarrhoea in the last 2 weeks, Oral Rehydration Salts (ORS) or oral zine use, and type of services accessed for diarrhoea. The final analyses included a total of 2626 individuals.

Predisposing characteristics (P) are personal characteristics that exist prior to the onset of specific episodes of illness. In this study, P included age, sex, marital status, and maternal educational attainment as they have been previously used in studies examining risk factors for healthcare utilisation. Our study also included other factors such as children’s number in a household, smoking status, and alcohol use as they are often considered in studies using the BM. Age was a continuous variable, while the six others were categorical variables: sex (girl/boy), marital status (married/cohabiting/not married), maternal educational attainment (no formal school, primary school, junior high school, senior middle school, senior high school and above), children’s number in a household (1-13), maternal smoking status (yes/no), and maternal alcohol use (yes/no).

Enabling resources (E) are factors that make health services resources available to the family. In prior studies, the most commonly used variables in the category were income and financial situation. The current study attempted to capture E by examining five variables: wealth index (0-10), health insurance (yes/no), residence region (rural/urban), improved water sources (yes/no) and improved sanitation facilities (yes/no). The use of a wealth index is generally considered an accurate mechanism to determine socioeconomic status (SES) within a population [37, 38]. The wealth index [39], which captures underlying long-term wealth through household asset information, was constructed by principal components analysis using the information on ownership of consumer goods (e.g. refrigerators, televisions, cars, trucks, bicycles, and motorcycles), materials used in household construction (e.g. wood, bricks, rocks, and cement), household electricity, access to drinking water and water for general use, and improved sanitation facilities [36, 40]. Therefore, individuals’ SES was determined through the wealth index. Drinking water sources are defined as improved drinking water sources (piped water, boreholes or tubewells, protected dug wells, protected springs, rainwater, and packaged or delivered water), and unimproved sources (unprotected springs and wells, surface water, and other sources) [41]. Improved sanitation facilities are those designed to hygienically separate excreta from human contact (Flush/pour flush toilets connected to piped sewer systems, septic tanks or pit latrines; ventilated improved pit latrines, composting toilets or dry pit latrines with slabs), while unimproved sanitation facilities (Pit latrines without a slab or platform, hanging latrines or bucket latrines and open defecation). Improved and unimproved water sources or sanitation facilities were represented as dichotomous variables, with “1″ representing “unimproved” and “2″ representing “improved”, respectively for both water sources and sanitation (Additional file 1: Table 1) [32].

We measured children’s health needs (N) by using the following common factors that impacted the possibility of healthcare-seeking behaviour, including fever (yes/no), cough (yes/no), dyspnoea (yes/no), and eating less (No, eating more/Same as before/Yes, eating less) in the last 2 weeks.

Four variables were used in this study to quantify children’s healthcare usage for diarrhoea: seeking advice or treatments for diarrhoea in the last 2 weeks (yes/no), the type of services accessed for diarrhoeal health issues (including 16 items, the options given for “where did you seek advice or treatment?” were classified as the public health sector, private health sector, other sources, and none (Additional file 1: Table 1) [42], ORS use in the last 2 weeks (yes/no), and oral zinc use in the last 2 weeks (yes/no).

Structural equation modelling (SEM) was conducted using SPSS AMOS, version 24 (IBM-SPSS, Chicago), to validate the hypothesized direct and indirect associations among individual characteristics, health services quality contextual factors, and healthcare-seeking behaviours. To determine whether the relationships among the latent variables constructed were as suggested by BM, a confirmatory factor analysis (CFA) was conducted. Guided by the BM, four latent variables were examined: predisposing characteristics, enabling characteristics, health needs, and health services use for children with diarrhoea.

In addition to the chi-squared test (χ²), the goodness of fit index (GFI), the adjusted goodness of fit index (AGFI), and the comparative fit index (CFI) were examined. In all cases, the values ranged from 0 to 1 and reflected the improvement in the fit of a hypothesized model over a model of independence among the measured variables, with values over 0.95 indicative of a good fit. Eventually, the root means squared error of approximation (RMSEA) was used as a measure of model fitness per degrees of freedom, with values less than 0.06 considered desirable.

The percentage, frequency, mean value, and variance for each variable in the P, E and N factors used in the model are presented in Table 1. In the final analysis, of the 2626 participants included, 51.4% were boys, and the mean age was 1.61 years (SD = 1.28). Only about one-third of children under-five with diarrhoea (34.1%) had access to improved water sources, and 44.8% of children had access to improved sanitation facilities. Over 2 weeks before the survey, the proportion of participants who had fever, cough, dyspnoea, or eating less were 54.8, 40.9, 17.4, and 68.20% respectively. Amongst children under-five with diarrhoea, more than half of them (52.0%) did not receive any healthcare. Nearly one in five of children under-five with diarrhoea (19.7%) sought care in the public health sector, and a quarter of them sought care in the private health sector or at home.

In the hypothesis model, this study had three latent variables: P, E, and N, all of which were exogenous variables. A separate measurement model was developed for each of the latent variables and then independently validated by CFA.

As previously mentioned, the predisposing characteristics variable (P) was composed of eight indicators: child age, sex of the child, maternal age, marital status, maternal educational attainment, children’s number in a household, maternal smoking status, and alcohol use. As depicted in Fig. 1, all indicators had a strong positive correlation with the latent variable P except for the sex of the child, smoking status and alcohol use, where the correlation was very weak. Also, all indicators for the construct were statistically significant at p < 0.05 except for the sex of the child and smoking status (Table 2, p = 0.435 and p = 0.599, respectively). Therefore, the sex of the child, smoking status and alcohol use were excluded from the modified/nested model. Refer to the modification suggests (Additional file 1: Table 2) and as well as considering the practical significance, maternal educational attainment was moved from P to E (Modification Indices = 145.489), urbanization was removed from the enabling construct and coughing was removed from the health need construct.

The final enabling construct was composed of 5 indicators: maternal educational attainment, wealth index, health insurance, improved water source and improved sanitation facilities. The health need construct was composed of three indicators: the presence of fever, dyspnoea or eating less. As shown, all indicators had a strong positive correlation with the latent variable and were statistically significant (Figs. 3, 4).

After revising the measurement models for P, E, and N, three empirically derived models for health services uses for childhood diarrhoea were constructed based on the different versions of BM. Statistically significant relationships are depicted in Fig. 2. As shown in Fig. 2, considering hypothesis model A and hypothesis model B, E had a slight, non-significant influence on N (− 0.01, p = 0.85 and − 0.01, p = 0.76, respectively). In hypothesis model C, P had a slight, non-significant influence on healthcare utilisation (− 0.01, p = 0.52). None of the three hypothesis models has a good CFI, which indicates the model fitting results are not so satisfying. Applying Andersen’s model to the context of the DRC, minimal model modification was conducted, and the final model was acceptable, shown in Fig. 3 and Table 3, with a GFI of 0.972, CFI of 0.953 and RMSEA of 0.043 (90% CI: 0.040, 0.047). All factor loadings were significant (p < 0.001), indicating that the model was plausible. Figure 4 provides the standardized path coefficients. The model revealed the direct effects of latent variables. Among all the variables, “Need” had the largest direct effect on healthcare utilisation (β, direct effect = 0.135, p < 0.001), followed by “enabling resources” (β, direct effect = 0.051, p = 0.015). Indirect effects were also examined based on Andersen’s model. “Need” emerged as a mediator of the effect of “Predisposing” on “Utilisation” (β, indirect effect = 0.014; p = 0.009). All of the proposed relationships were positively associated with health services use. For example, higher enabling resources (especially household wealth [β = 0.88, p < 0.001]) and improved water sources [β = 0.60, p < 0.001]) and more health needs (especially fever [β = 0.54, p < 0.001]) were both positively associated with a greater chance of health services use. In this sample, greater predisposing characteristics (especially the number of children [β = 0.87, p < 0.001]) were associated with a higher likelihood of healthcare utilisation (Fig. 4).

One of the strengths of screening a national sample is that the standardized methodology used by MICS allows for comparisons with similar data from other countries. This permitted horizontal comparisons to be made by exploring different influential factors on health services. Additionally, a conceptual framework was applied based on Andersen’s health behavioural model, a widely accepted method for assessing factors associated with healthcare utilisation. This study then developed modified models to predict healthcare utilisation for diarrhoea in children under five years in the DRC.

As most of the examination was based on Anderson’s behavioural model, which posited a particular directionality of influence in health services use, BM was also confirmed by using a secondary dataset [31]. However, as cross-sectional data were used, the nature of any causal relationships among these three latent variables (predisposing characteristics, enabling factors, and need factors) and healthcare utilisation for children under five with diarrhoea in the DRC was unable to be ascertained. Furthermore, only one dataset was used to test different versions of Andersen’s model (traditional and empirically modified versions), and then to explore the relationships among the various latent factors. This work is expected to soon be replicated with different datasets from other developing countries to examine health services use in a variety of contexts.

Although the BM was explicitly employed as the theoretical background for the reviewed studies, there was a lack of consistency in these findings. As previously mentioned, the majority of reviewed studies’ operationalizations of the model revealed that only a small set of variables were commonly used and that there were many variations in the way these variables were categorized, especially concerning predisposing and enabling factors. This may stem from the secondary datasets used in the majority of the studies, which limited the variables available for study. Therefore, more primary studies exploring healthcare utilisation in the DRC or similar contexts are urgently needed to enrich understanding of this subject.