A mixed-effects multilevel Poisson regression analysis of diarrhea prevalence among under-five children in Indonesia

Abstract

Diarrheal disease is the second leading cause of death in under-five children. This study assessed the prevalence and risk factors of diarrhea among under-five Indonesian children using data from the 2017 Indonesia Demography and Health Survey.The analysis included 17,263 children living with their mothers and was based on a multistage stratified sampling design. Diarrhea was defined as maternal report of an episode within the two weeks preceding the survey. Descriptive statistics and chi-square tests were used for preliminary analysis. Determinants of diarrhea were examined using mixed-effects multilevel Poisson regression to account for the hierarchical data structure, with region specified as a random effect. All analyses were conducted using Stata. The model included region as a random effect, allowing baseline diarrhea prevalence to vary across regions. Female children had a lower prevalence of diarrhea than males (APR = 0.883; 95%CI:0.815-0.956), while children aged 1–2 years had a higher prevalence compared with infants. Higher household wealth was strongly protective, with children from the richest households showing a substantially lower prevalence (APR = 0.720; 95%CI:0.611-0.812). Environmental factors remained significant in the multilevel framework, as households with standard sanitation facilities had lower diarrhea prevalence (APR = 0.871; 95%CI:0.776-0.976), and rural residence was associated with higher prevalence. The small random-effect variance indicated limited residual regional heterogeneity after adjustment. Childhood diarrhea in Indonesia is shaped by interacting individual, environmental, and community-level factors. Socioeconomic disadvantage and inadequate sanitation remain major contributors to diarrhea prevalence among under-five children. The limited residual regional variation suggests that improvements in household living conditions and sanitation, particularly in rural areas, could substantially reduce disease burden. Multilevel analytical approaches are crucial for informing targeted and effective public health interventions.

Introduction

Diarrheal disease stands as the second most prevalent cause of death in children under the age of five, leading to the loss of approximately 525,000 young lives annually on a global scale [1]. This substantial impact highlights the pressing need for effective interventions to address the prevalence and impact of diarrhea in young children, particularly in low-resource settings [2]. In developing countries, an estimated 1.8 million individuals die as a result of diarrheal diseases annually and more than 80% of them occur in children under the age of five [3]. Geographically, South Asia and sub-Saharan Africa (SSA) combined account for 88% of the fatalities in this age group [1].Diarrhea is responsible for more child deaths than AIDS, malaria, measles, or the combined impact of all three diseases [1]. Children suffering from diarrhea face various difficulties, such as reduced appetite, electrolyte imbalance, malnutrition, increased vulnerability in comparison to other infectious diseases, there is an observed delay in both physical and cognitive development [4, 5]. As a result of its adverse effects on physical and cognitive development, diarrhea is associated with a broad spectrum of complications, leading to 72.8 million disability-adjusted life years. Additionally, it exacerbates the economic strain on families and healthcare systems [1].

In Indonesia, diarrhea is still considered an endemic disease and a potential outbreak that can lead to fatalities. In 2015, there were 18 reported diarrhea outbreaks in 11 provinces and 18 municipalities/cities, affecting 1,213 patients and resulting in 30 deaths. The Case Fatality Rate (CFR) for diarrhea in Indonesia was 2.47% during that year [6]. The Basic Health Survey (BHS) conducted in 2013 reveals that the occurrences of diarrhea across various age groups in Indonesia stood at 3.5%, with the prevalence during a given period registering at 7%. High rates of both incidence and prevalence were recorded in multiple provinces, notably Papua, South Sulawesi, Aceh, West Sulawesi, and Central Sulawesi. The rate at which toddlers experience diarrhea stands at 10.2% [7]. Additionally, diarrhea affects a quarter of children under the age of five, making it the primary cause of mortality in this age group [8, 9]. In this study, we applied the IDHS 2017 dataset due to its international comparability, methodological rigor, and standardization, which align with the study’s objectives. The IDHS 2017 allows for consistent analysis across countries and time periods, supported by well-established quality control measures such as pre-survey training, field monitoring, and post-survey validation (USAID, 2017). Additionally, while the BHS 2018 and 2023 datasets provide more recent data, they focus primarily on Indonesia’s specific health concerns, with flexibility in methodology that might introduce variations across years (Indonesia Ministry of Health, 2018, 2023). For the purposes of our study, which involved broader demographic and health trends rather than localized or time-sensitive data, the IDHS 2017 offered the most reliable and comparable foundation.

Numerous investigations have focused on identifying factors influencing the incidence of diarrhea. Factors such as the residential environment, employment status, family income, the age of the mother or caregiver, household size, availability of information, toilet facilities, access to potable water, child vaccination coverage, nutritional health, and the quantity of bedrooms in a home have all been identified as contributing elements to the prevalence of diarrhea [8, 11]. Research from regions such as Mbour, Senegal [12], Bangladesh [13], and Eastern Ethiopia [10] has shed light on various social and environmental condition determinants impacting the incidence of diarrhea. In Senegal, factors like mothers’ lack of employment, shared sanitation facilities, and failure to treat stored water have been highlighted [12]. In Bangladesh, the age and developmental benchmarks of children, mothers’ ages and occupation, residential zones, and types of sanitation services have been pinpointed as pivotal [13]. Meanwhile, in Ethiopia, ineffective waste management, absence of handwashing facilities, living in rural areas, and having two or more under-five children in the household were all significant contributors to the presence of diarrhea [10].

The current research investigates the variation in the prevalence of childhood diarrhea across Indonesia by employing a hierarchical analytical model that considers individual, environmental, and community-level factors. The analysis incorporates the seven big islands area (Java and Bali, Kalimantan, Sumatra, Maluku, Sulawesi, Papua, and Nusa Tenggara), the locale of residence (urban versus rural), and key demographic characteristics of children.

Unlike Siramaneerat & Agushybana [14], which primarily focused on family wealth, housing conditions, and residence using PLS-SEM, the present study provides a more comprehensive, multi-level analysis. In addition to confirming the influence of wealth and housing, this study examines critical individual-level factors (child’s age and gender, parental education, vaccination status, child health card, maternal mass media exposure, and dietary diversity) alongside community-level variables. By integrating these multiple layers of determinants, the study highlights the strong and independent contributions of individual, environmental, and community characteristics to diarrhea risk, identifying factors previously unexplored. The findings provide a deeper understanding of how demographic, environmental, and regional factors interact to influence diarrhea prevalence, offering valuable insights for the development of targeted intervention strategies and region-specific guidelines. Such evidence can guide effective public health measures aimed at reducing diarrhea among under-five children, ultimately improving health outcomes across diverse Indonesian communities.

Methods

Source and study population

The dataset utilized in the study was sourced from the 2017 Indonesia Demographic and Health Survey (IDHS), conducted in partnership with the National Population and Family Planning Board (BKKBN), Statistics Indonesia (BPS), and the Ministry of Health (Kemenkes). Technical guidance for the survey was provided by the U.S. Agency for International Development (USAID) through the Demographic and Health Surveys (DHS) Program, which is dedicated to aiding developing nations in executing DHS surveys [15]. The survey provides comprehensive data on crucial health and demographic indicators, including maternal and child health, family planning, fertility, and nutrition.

When comparing the DHS 2017 dataset with other national datasets, such as the Basic Health Survey (BHS) 2018 and 2023, the BHS datasets are more context-specific, tailored to Indonesia’s health priorities and policy needs. The BHS may allow for more flexibility in data collection to adapt to evolving national health concerns, providing timely and detailed insights on local health issues [16]. However, this flexibility can introduce slight variability in methodology across years, which may impact on the dataset’s consistency compared to the IDHS.

While DHS 2017 excels in providing broad, internationally comparable health trends, the BHS is more adept at capturing Indonesia’s unique health dynamics. Both the DHS 2017 and BHS 2018 & 2023 datasets [16] offer valuable insights into Indonesia’s health landscape, but they serve slightly different purposes. IDHS 2017 is more reliable and valid for broad, internationally comparable health indicators [17], while BHS 2018 and 2023 may provide more accurate and timely information on Indonesia’s specific health concerns [16].

The IDHS 2017 dataset is recognized for its high reliability and validity, primarily due to its international standardization and rigorous quality control (QC) procedures. This dataset is designed to ensure consistency across multiple countries and time periods, making it a trusted source for demographic and health data globally. The IDHS employs strict pre-survey training, continuous field supervision, and post-collection consistency checks to verify data accuracy and minimize reporting and measurement errors, thereby strengthening both internal and external validity. In addition, the IDHS utilizes a multi-stage stratified sampling design that explicitly accounts for the hierarchical structure of the population, including children nested within households and households nested within communities and regions. This sampling framework ensures adequate representation of both urban and rural populations and supports valid inter-regional comparisons [17]. Through multi-stage random sampling, the 2017 Indonesia Demographic and Health Survey (IDHS) interviewed 49,250 eligible women aged 15–49 years across all provinces of Indonesia. In the first stage, 1,970 census blocks (CBs) from the 2010 Population Census of Indonesia were selected. Subsequently, 34 CBs were chosen from each province, and in the final stage, eight households were randomly selected from each CB. For the purposes of this study, data on under-five children residing with their mothers were extracted, yielding an analytical sample of 17,263 children. This hierarchical data structure further justified the use of mixed-effects multilevel regression analysis in examining factors associated with diarrhea prevalence.

Study variables and definitions

Diarrhea prevalence was first estimated to describe the overall burden of disease and was subsequently used as the outcome variable in the mixed-effects multilevel Poisson regression analysis to identify factors associated with diarrhea. The outcome variable in this study was the prevalence of diarrhea among under-five children, as reported by their mothers. Diarrhea prevalence was defined as the proportion of children under five years of age who experienced at least one episode of diarrhea during the two weeks preceding the survey interview, representing a two-week period prevalence. The responses to the inquiries were classified as either Yes (coded 1) or No (coded 0). Notably, no clinical examination data was obtained during the study. We assessed the risk of diarrhea by analyzing data across individual, environmental, and community levels. The selection of variables for analysis was based on previous research and current information from the specific region [8, 9, 18–20].

The study examined individual-level characteristics, including the child’s gender, child’s age, mother’s age, mother’s education, father’s education, child health card, mother’s access to mass media, received vaccination, dietary diversity score, and family wealth. Child gender was categorized as male (coded 1) or female (coded 2). The age of the child was documented in years, while the mother’s age was stratified as follows: 15–19 years (coded 1), 20–24 years (coded 2), 25–29 years (coded 3), 30–34 years (coded 4), 35–39 years (coded 5), 40–44 years (coded 6), and 45–49 years (coded 7). Parental educational attainment was classified into “no education” (coded 0), “primary level” (coded 1), “secondary level” (coded 2), or “higher level” (coded 3). Child health card was coded using a dichotomous response: yes (coded 1) or no (coded 0). The mother’s engagement with mass media, encompassing the use of newspapers or magazines, radio, television, internet, and mobile phone variables were dichotomized. Each form of mass media was coded using a dichotomous response: user (coded 1) or non-user (coded 0). Each form of received vaccination was coded using a dichotomous response as followed ever got vaccination (coded 1) or never (coded 0).

Dietary intakes were assessed using a semi quantitative food frequency questionnaire. To determine the dietary diversity score (DDS), we divided foods into 14 main groups: These groups were: (1) yogurt (2) solid food or semi-solid food (3) cheese, yogurt, other milk products (4) food made from beans, etc. (5) fish or shellfish (6) liver, heart, other organs (7) any fruits (8) mango, papayas, other vitamins and fruits (9) any dark green leafy vegetables (10) child pumpkin, carrots, squash, ets. (11) beef, pork, lamb, etc.(12) eggs (13) potatoes, cassava, or other tubes (14) bread, noodles, others food from grain. Each food group had 1 score. Total DDS was 10 and then it was divided to 3 quartiles: less than 4.0 (low diversity), 5.0–9.0 (moderate diversity), and 10.0–14.0 (high diversity).

Family wealth status was categorized using a five-quintile system derived from the household wealth index. This index incorporated various household assets and amenities, such as radios, televisions, refrigerators, bicycles, motorcycles, or cars, as well as housing conditions including access to electricity, type of flooring, wall/roofing materials, water source, latrine ownership, and bedroom availability. Quintiles were assigned based on ascending levels of wealth, ranging from the lowest to the highest. Further details on the computation of the wealth score can be referenced in the IDHS 2017 documentation [17]. Household evaluations were conducted by considering both the number and variety of consumer goods owned, which may range from televisions and bicycles to automobiles, alongside dwelling features like water sources, bathroom amenities, and flooring materials [20]. These evaluations employ principal component analysis. National wealth quintiles are established by assigning a wealth score to every standard household member (de jure), assessing everyone’s score within the household population, and subsequently dividing the distribution into five equal segments, each comprising 20% of the population [20].

The environmental condition-level characteristics comprised the source of drinking water, toilet condition, and the construction materials of the wall and floor of the house. Drinking water sources were categorized as unimproved (0) or improved (1). Unimproved drinking-water sources encompass a range of options such as unprotected (dug) wells, unprotected springs, carts with small tanks or drums, water delivered by tanker trucks, and surface water from rivers, dams, lakes, ponds, streams, canals, or irrigation channels, as well as bottled water. Improved sources of drinking water include piped water, public taps, standpipes, tube wells, boreholes, and protected dug wells and springs, as well as rainwater. Households utilizing bottled water for drinking are classified as using an improved source only if their water source for cooking and hand washing also originates from an improved source [16, 17]. The toilet’s condition was classified as unimproved (0) or improved (1). Toilets used by only one home, with or without a septic tank, are examples of improved toilet facilities. Unimproved toilet facilities are those that do not satisfy the standards for improved sanitation. These facilities might include flush or pour-flush to another location, pit latrines without a slab, or open pits [17, 18]. Wall materials were classified as either permanent (such as cement, brick, stone with lime, or ceramic tiles) and coded as 1, or non-permanent (including bamboo, cardboard, or wood plank) and coded as 0. Flooring types were distinguished as ground sand (coded 0), semi-permanent (coded 1), or permanent/concrete materials (coded 2). Additionally, residential areas were categorized as urban (coded 1) or rural (coded 0). Indonesia’s administrative divisions consist of seven regions: Java and Bali, Kalimantan, Sumatra, Maluku, Sulawesi, Papua, and Nusa Tenggara, coded from 1 to 7, respectively.

Statistical analysis

Statistical methods were employed to identify risk factors associated with the prevalence of diarrhea among under-five children. Understanding these relationships is essential for designing effective public health interventions. Descriptive statistics were used to summarize the characteristics of the study participants and their individual-, environmental-, and community-level attributes. Pearson’s chi-square tests were applied to examine bivariate associations between diarrheal illness among children under five years of age and parental socioeconomic and demographic characteristics, as well as community-level variables.

Subsequently, a mixed-effects multilevel poisson regression analysis was conducted to examine the relationships between explanatory variables and diarrheal disease, accounting for the hierarchical structure of the Indonesia Demographic and Health Survey (IDHS) dataset. Three multilevel models were specified. In Model 1, division was treated as an individual-level characteristic. In Model 2, division was considered an environmental-level characteristic. In Model 3, division was treated as a community-level characteristic.

In all models, region was specified as a random effect, while individual-, environmental-, and community-level variables were included as fixed effects. The inclusion of region as a random effect accounts for unobserved heterogeneity across regions in Indonesia, recognizing that regions may differ in economic, social, and cultural contexts that influence the prevalence of childhood diarrhea. This modeling approach allows the baseline risk of diarrhea to vary across regions while estimating the fixed effects of the explanatory variables. All statistical analyses were conducted using Stata statistical software.

Multilevel Poisson regression model specification

General mixed-effects multilevel Poisson regression model

Let denote the occurrence of diarrhea for child residing in region . The outcome variable was assumed to follow a Poisson distribution:

with a log-link function:

where:

•  is the expected prevalence of diarrhea.,

•  is the overall intercept.

•  represents fixed-effect predictors at the individual, environmental, and community levels.,

•  is fixed-effect coefficients.

•  is the random intercept for region , assumed to follow a normal distribution .

This mixed-effects specification allows the baseline risk of diarrhea to vary across regions while estimating the effects of observed covariates.

Model operationalization

Model 1: Individual-level model (Division as an individual characteristic)

Model 1 included individual-level characteristics only, with division treated as an individual-level fixed effect.

Model 2: Environmental-level model (Division as an environmental characteristic)

Model 2 extended Model 1 by including environmental-level characteristics, with division conceptualized at the environmental level.

Model 3: Community-level model (Division as a community characteristic)

Model 3 further incorporated community-level characteristics, treating division as a community-level contextual variable.

In all three models, region was specified as a random intercept to capture unobserved heterogeneity across Indonesian regions (Sumatra, Java and Bali, Nusa Tenggara, Kalimantan, Sulawesi, Maluku, and Papua). This approach accounts for regional differences in social, economic, cultural, and health-system contexts that may influence childhood diarrhea prevalence.

Results

Individual, environmental condition, and community-level characteristics

The study involved 17,263 mothers with under-five children. Approximately, 51.4% of the children were male. Most of the mothers (26.4%) were between the ages of 30 to 34, followed by 25–29 years (25.8%). Over 70.0% of both mothers and fathers had received education at the secondary level or higher. Furthermore, 95.6% of mothers had access to mass media. The analysis showed that the majority of the population (92.3%) possessed a child health card (n = 15,931), while a smaller proportion (7.7%) did not possess a child health card. A total of 4,182 respondents (24.23%) have never received a vaccination, meaning about a quarter of the population remains unvaccinated, which could raise public health concerns. Regarding dietary diversity score (DDS), 79.96% of respondents fall into the low dietary diversity category, while 17.29% have a moderate level of dietary diversity. Only 2.75% of respondents have a high dietary diversity score, indicating that a very small portion of the population consumes a broad and balanced variety of food groups. Around 27.5% of the mothers in the study were from families classified as having the lowest socioeconomic status based on family wealth. Moreover, 26.0% of the children belonged to households without access to an unimproved drinking water source, and 17.2% had a non-standard toilet condition. Most of the respondents had houses with permanent flooring (78.6%). The study encompassed a predominant portion (50.8%) of participants residing in rural areas, with the remaining (49.2%) living from urban areas across the country. Participants were chosen from diverse regions/divisions of the nation., including 26.6% from Sumatra, 30.3% from Java and Bali, 8.6% from Nusa Tenggara, 9.2% from Kalimantan, 15.2% from Sulawesi, 7.0% from Maluku, and 3.1% from Papua.

Prevalence of diarrhea across different Individual, environmental condition and community-level characteristics

The present study found that the overall two-week prevalence of diarrhea among under-five children in Indonesia was 14.13%. The prevalence rates of diarrhea among male (14.57%) and female (13.60%) children observed in this study are comparable to the overall national prevalence of 14.13% identified in the present analysis. Male children exhibited a slightly higher prevalence than female children. Child age was another significant factor, with diarrhea prevalence peaking at 19.84% for one-year-olds, compared to 13.89% for infants (0 years), 15.69% for two-year-olds, 11.86% for three-year-olds, and 9.12% for four-year-olds, leading to a p-value < 0.001. This trend indicates that younger children may be more susceptible to diarrheal diseases, possibly due to immature immune systems and dietary transitions. Additionally, a heightened prevalence of child diarrhea was observed among adolescent mothers (aged 15–19 years) at 22.98%. Similarly, the prevalence of diarrhea was higher when both parents had attained primary education levels (16.05% for mothers and 15.74% for fathers). Moreover, the analysis revealed that children whose mothers had no access to mass media had an 18.52% prevalence of diarrhea, while those with access reported a prevalence of 13.94%, with a p-value of 0.025. This suggested that a lack of access to mass media may result in lower awareness of health practices that can prevent diarrhea. Among individuals without a child health card, the prevalence of diarrhea was 17.34% (95% CI: 14.90–20.09.90.09). In comparison, those with a child health card had a lower prevalence of diarrhea of 13.86% (95% CI: 13.09–14.66). This suggests that possession of a child health card is associated with a reduced prevalence of health issues.

In terms of vaccination status, 17.60% of never-vaccinated children experienced diarrhea, compared to 13.10% of those who were ever vaccinated, with a statistically significant p-value < 0.001. This indicated that unvaccinated children have a higher prevalence of diarrhea. Regarding Dietary Diversity Score (DDS), the prevalence rates were 12.97% for low dietary diversity (1–4), 18.19% for moderate diversity (5–9), and 19.69% for high diversity (10–14), also showing a significant difference with a p-value < 0.001. Interestingly, higher dietary diversity was associated with higher rates of diarrhea, suggesting that children with more varied diets may be more exposed to pathogens from different food sources, warranting further investigation. Family wealth was strongly associated with diarrhea prevalence, decreasing from 16.06% in the poorest households to 10.14% in the richest households, with a significant p-value of 0.001. This finding indicated that wealthier families are likely to have better access to healthcare, sanitation, and nutrition, ultimately reducing the incidence of diarrhea. Additionally, the study foundthat diarrhea rates associated with specific environmental conditions, particularly among households without mass media access (18.52%) and those with semi-permanent floors (18.28%). Unimproved water and non-standard sanitation were also significant factors, with rates of 15.98% and 16.51%, respectively. In terms of location, rural residence was linked to a higher prevalence (15.30%) than urban residence, while Kalimantan (16.60%) emerged as the region with the most cases. Based on the Chi-square tests, gender was the only non-significant factor; all other individual and environmental variables were significantly associated with the outcome (P < 0.05) (Table 1).

Table 1.

Descriptive analysis and bivariate outcomes (χ2 tests) of childhood diarrhea in Indonesia; based on data from the Indonesia demographic and health survey conducted in 2017

Variables	N	%	Prevalence	95% CI	P-Value
Overall	17,263		14.13	(13.52–15.25)
Individual-level characteristics
Child’s gender					0.119
Male	8,868	51.4	14.57	(13.61–15.59)
Female	8,395	48.6	13.60	(12.65–14.61)
Child’s age					< 0.001**
0 years	3,385	19.6	13.89	(12.45–15.47)
1 years	3,533	20.5	19.84	(18.22–21.56)
2 years	3,403	19.7	15.69	(14.23–17.28)
3 years	3,449	20.0	11.86	(10.52–13.33)
4 years	3,493	20.2	9.12	(8.00–10.37)
Mother’s Age					< 0.001**
15–19	422	2.4	22.98	(18.08–28.74)
20–24	2,727	15.8	18.25	(16.49–20.15)
25–29	4,448	25.8	14.42	(13.11–15.84)
30–34	4,562	26.4	12.38	(11.17–13.69)
35–39	3,391	19.6	1289	(11.48–14.45)
40–44	1,414	8.2	11.34	(9.53–13.45)
45–49	299	1.8	9.94	(6.41–15.09)
Mother’s education					< 0.001**
No education	245	1.4	13.10	(8.60–19.46)
Primary	4,281	24.8	16.05	(14.62–17.59)
Secondary	9,631	55.8	14.25	(13.28–15.27)
Higher	3,106	18.0	10.36	(9.03–11.85)
Father’s education					< 0.001**
No education	229	1.3	11.55	(7.22–17.96)
Primary	4,954	28.7	15.74	(14.46–17.11)
Secondary	9,513	55.1	14.31	(13.37–15.31)
Higher	2,567	14.9	9.71	(8.36–11.24)
Mother’ s Access to Mass Media					0.025*
No access	768	4.4	18.52	(14.54–23.30)
Access	16,495	95.6	13.94	(13.18–14.74)
Child Health Card					0.005**
No	1,332	7.7	17.34	(14.90–20.09)
Yes	15,931	92.3	13.86	(13.09–14.66)
Received Vaccination					< 0.001**
Never	4,182	24.3	17.60	(16.07–19.24)
Ever got vaccination	13,081	75.7	13.10	(12.32–13.93)
Dietary Diversity Score (DDS)					< 0.001**
Low (1–4)	13,803	79.9	12.97	(12.20–13.79)
Moderate (5–9)	2,985	17.3	18.19	(16.42–20.11)
Hight (10–14)	475	2.8	19.69	(15.52–24.65)
Family wealth					< 0.001**
Poorest	4,742	27.5	16.06	(14.61–17.62)
Poorer	3,377	19.6	15.86	(14.33–17.51)
Middle	3,178	18.4	14.01	(12.63–15.52)
Richer	3,037	17.5	14.16	(12.58–15.91)
Richest	2,929	17.0	10.14	(8.79–11.68)
Environmental-level characteristics
Drinking water source					0.005**
Unimproved	4,492	26.0	15.98	(14.42–17.67)
Improved	12,771	74.0	13.52	(12.71–14.38)
Toilet Condition					0.003**
Not standard	2,973	17.2	16.51	(14.76–18.42)
Standard	14,290	82.8	13.66	(12.85–14.50)
Floor of the houses					0.012*
Ground-sand	1,291	7.5	13.16	(10.59–16.25)
Semi-permanent	2,402	13.9	18.28	(15.22–19.21)
Permanent	13,570	78.6	13.77	(12.94–14.66)
Community-level characteristics
Residence					0.002**
Urban	8,493	49.2	12.82	(11.83–13.88)
Rural	8770	50.8	15.30	(14.20–16.47)
Region					< 0.001**
Sumatra	4,590	26.6	15.92	(14.52–17.41)
Java and Bali	5,236	30.3	12.92	(11.79–14.14)
Nusa Tenggara	1,491	8.6	14.49	(12.43–16.82)
Kalimantan	1,581	9.2	16.60	(14.28–19.22)
Sulawesi	2,623	15.2	16.14	(14.35-0.18.35)
Maluku	1,209	7.0	13.38	(11.25–15.85)
Papua	533	3.1	10.05	(7.42–13.48)

**P < 0.01, * P < 0.05

A Mixed-Effects multilevel Poisson regression analysis of Individual, Environmental, and Community-Level factors on the prevalence of diarrhea among under-five children

Table 2presents the results of the mixed-effects multilevel Poisson regression analysis, reporting adjusted prevalence ratios (APR) with 95% confidence intervals (CI). Three hierarchical models were estimated to reflect the conceptual structure of the data. Model 1 included individual-level characteristics, Model 2 additionally incorporated environmental-level characteristics, and Model 3 represented the full model by further including community-level characteristics. In all models, region was specified as a random intercept to account for clustering and unobserved regional heterogeneity.Model comparison was conducted to evaluate relative model fit, whereas interpretation of determinants was based on the fully adjusted Model 3. The variance of the random intercept was small across all models, decreasing from 0.010 (95% CI: 0.002–0.057) in Model 1 to 0.004 (95% CI: 0.0005–0.040) in Model 2, and remaining low at 0.008 (95% CI: 0.001–0.050) in Model 3. Correspondingly, intraclass correlation coefficients (ICC) ranged from 0.003 to 0.006, indicating that only a small proportion of the total variability in childhood diarrhea prevalence was attributable to regional-level differences after accounting for observed covariates.The Wald chi-square tests confirmed that the fixed effects in all models jointly contributed to explaining variability in childhood diarrhea prevalence (p < 0.001). Model fit statistics showed that Model 1 had the lowest Akaike Information Criterion (AIC = 14164.5) and the highest log-likelihood value (LL = −7055.252), while Model 2 (AIC = 14406.8; LL = −7197.401) and Model 3 (AIC = 14409.88; LL = −7201.942) exhibited comparatively poorer fit. This pattern suggests that although the inclusion of environmental- and community-level variables increased model complexity, these variables explained limited additional variance beyond individual-level characteristics, likely due to shared variance with individual socioeconomic factors and the absorption of contextual heterogeneity by the random effects.Accordingly, Model 1 was identified as the most parsimonious model in terms of statistical fit, whereas Model 3 was retained as the primary model for interpretation, as it provides a comprehensive assessment of individual, environmental, and community-level determinants within a multilevel framework. All reported associations and conclusions regarding significant determinants of childhood diarrhea are therefore based on the results of Model 3.

Table 2.

A Mixed-Effects multilevel Poisson regression analysis for childhood diarrhea in Indonesia; Indonesia demographic and health Survey, 2017

	Model 1		Model 2		Model 3
	APR	95% CI	APR	95% CI	APR	95% CI
Individual-level characteristics
Child’s gender
Male	1.000
Female	0.883**	(0.815;0.956)
Child’s age
0 year	1.000
1 year	1.452**	(1.271;1.658)
2 years	1.288**	(1.130;1.467)
3 years	0.978	(0.851,1.123)
4 years	0.742**	(0.639;0.861)
Mother’s Age
15–19 years	1.000
20–24 years	0.885	(0.708;1.106)
25–29 years	0.750*	(0.601;0.936)
30–34 years	0.673**	(0.538;0.841)
35–39 years	0.671**	(0.533;0.843)
40–44 years	0.648**	(0.500,0.839)
45–49 years	0.681	(0.461,1.007)
Mother’s education
No education	1.000
Primary	0.907	(0.647;1.271)
Secondary	0.841	(0.597;1.184)
Higher	0.755	(0.524;1.089)
Father’s education
No education	1.000
Primary	1.305	(0.886;1.924)
Secondary	1.220	(0.826;1.804)
Higher	1.094	(0.722;1.657)
Mother’ s Access to Mass Media
No access	1.000
Access	1.043	(0.853;1.276)
Received Vaccination
Never	1.000
Ever got	1.050	(0.951;1.160)
Dietary Diversity Score (DDS)
Low (1–4)	1.000
Moderate (5–9)	1.154*	(1.023;1.301)
Hight (10–14)	1.187	(0.942;1.496)
Family wealth
Poorest	1.000
Poorer	0.968	(0.860;1.089)
Middle	0.915	(0.805;1.040)
Richer	0.912	(0.795;1.046)
Richest	0.720**	(0.611;0.812)
Constant	0.176**	(0.106;0.291)
Environmental-level characteristics
Drinking water source
Unimproved			1.000
Improved			0.915	(0.830;1.009)
Toilet Condition
Not standard			1.000
Standard			0.871*	(0.776;0.976)
Floor of the houses
Ground-sand			1.000
Semi-permanent			1.445**	(1.192;1.752)
Permanent			1.272**	(1.063;1.522)
Constant			0.131**	(0.111;0.155)
Community-level characteristics
Residence
Urban					1.000
Rural					1.171**	(1.079;1.272)
Constant					0.128**	(0.117;0.141)
Random effect: Region
var(constant)	0.010 (0.002;0.057)		0.004 (0.0005;0.040)		0.008 (0.001,0.050)
Model comparison
Wald Chi2	300.97**		24.19**		14.22**
Log likelihood	−7055.252**		−7197.401*		−7201.942*
ICC	0.006		0.003		0.005
AIC	14164.5		14406.8		14409.88

Wald Chi2, Wald Chi-square statistic; APR, adjusted prevalence ratio; CI, confidence interval; ICC, intraclass correlation coefficient;; AIC, Akaike’s information criteria;

** P < 0.01

*P < 0.05

In multilevel mixed-effect poisson regression analysis, child’s gender, child’s age, dietary diversity score (DDS), mother’s age, family wealth, toilet condition, floor of the houses and residence were significant factors for the prevalence of diarrhea among under-five children. Females had 11.7% lower prevalence of experiencing diarrhea compared to males (APR: 0.883, 95% CI 0.815–0.956). Moreover, 1-year-old children had a 45.2% higher prevalence of developing diarrhea compared to children aged 0 years (APR: 1.452, 95% CI 1.271–1.658). Similarly, 2-year-old children had 28.8% higher prevalence of experiencing diarrhea compared to 0-year-olds (APR: 1.288, 95% CI 1.130–1.467). In contrast, 4-year-old children had 25.8% lower prevalence of developing diarrhea compared to children aged 0 years (APR: 0.742, 95% CI 0.639–0.861).Moreover, maternal age was inversely associated with the prevalence of childhood diarrhea. Compared with children born to adolescent mothers aged 15–19 years, those whose mothers were aged 25–29 years had a significantly lower prevalence of diarrhea (APR = 0.750; 95% CI: 0.601–0.936). The protective effect became more pronounced among children of mothers aged 30–34 years (APR = 0.673; 95% CI: 0.538–0.841), 35–39 years (APR = 0.671; 95% CI: 0.533–0.843), and 40–44 years (APR = 0.648; 95% CI: 0.500–0.839). Although children of mothers aged 20–24 years and 45–49 years also showed lower prevalence ratios, these associations were not statistically significant, as their confidence intervals included unity.

Furthermore, children with a moderate dietary diversity score (DDS 5–9) had a significantly higher prevalence of diarrhea compared with those with a low dietary diversity score (DDS 1–4) (APR = 1.154; 95% CI: 1.023–1.301), whereas high dietary diversity (DDS 10–14) was not significantly associated with diarrhea prevalence. Children from the richest households had a significantly lower prevalence of diarrhea compared with those from the poorest households (APR = 0.720; 95% CI: 0.611–0.812), whereas no significant differences were observed among children from poorer, middle, or richer households.Regarding environmental-level characteristics, sanitation facilities were significantly associated with diarrhea prevalence. Children living in households with standard toilets had a lower prevalence of diarrhea compared with those in households with non-standard toilets (APR = 0.871; 95% CI: 0.776–0.976).Housing floor materials were also significant determinants. Children residing in households with semi-permanent floors (APR = 1.445; 95% CI: 1.192–1.752) and permanent floors (APR = 1.272; 95% CI: 1.063–1.522) had a higher prevalence of diarrhea compared with those living in houses with ground-sand floors. In contrast, drinking water source was not significantly associated with diarrhea prevalence after adjustment. Furthermore, children living in rural areas hada significantly higher prevalence of experiencing childhood diarrhea compared to those in urban areas (APR: 1.171, 95% CI: 1.079-1.272 ).

Discussion

This study applied a mixed-effects multilevel Poisson regression approach to appropriately account for the hierarchical structure of the data and regional clustering in childhood diarrhea prevalence. Although region was specified as a random intercept in all models, the estimated random-effect variances and intraclass correlation coefficients (ICCs) were consistently small across Models 1–3. These findings indicate that, after adjusting for observed covariates, regional-level differences contributed minimally to the overall variability in childhood diarrhea prevalence. This suggests that disparities in diarrhea risk are driven predominantly by individual- and household-level characteristics rather than by unobserved regional contextual factors.The progressive decline in random-intercept variance from Model 1 to Model 2, followed by a slight increase in Model 3, reflects partial absorption of contextual heterogeneity by environmental- and community-level variables. However, the persistently low ICC values imply that these higher-level characteristics explained only a limited proportion of between-region variation. This may be attributed to the relative homogeneity of contextual exposures across regions or the strong correlation between environmental conditions and individual socioeconomic status, leading to shared explanatory power across model levels.Model comparison results further support this interpretation. Model 1 demonstrated the best statistical fit, as indicated by the lowest AIC and highest log-likelihood, suggesting that individual-level factors alone accounted for a substantial proportion of the explainable variation in childhood diarrhea prevalence. The addition of environmental- and community-level variables in Models 2 and 3 increased model complexity without yielding meaningful improvements in overall model fit. This finding aligns with previous multilevel studies in low- and middle-income settings, where proximal individual and household determinants such as wealth, maternal characteristics, and sanitation practices often dominate distal contextual influences. Nevertheless, despite its comparatively poorer fit, Model 3 was retained for substantive interpretation because it provides a more comprehensive and theoretically grounded assessment of diarrhea determinants within a multilevel framework. Incorporating environmental and community factors allows for a clearer understanding of how individual risks operate within broader living contexts, even when their independent statistical contributions are modest. Importantly, retaining the full model reduces the risk of omitted-variable bias and strengthens the interpretability of adjusted prevalence ratios for key determinants.The prevalence of diarrhea among male (14.57%) and female (13.60%) children was generally comparable to the national prevalence observed in this study (14.13%), with neither sex exhibiting levels that were substantially higher or lower than the national average. At the regional level, variations in diarrhea prevalence were evident across Indonesian regions, suggesting that the observed sex-specific differences should be interpreted within broader regional contexts. These regional disparities likely reflect differences in socioeconomic conditions, sanitation infrastructure, and access to health services, as documented in previous studies on childhood diarrhea and health inequalities [21, 22]. Thus, the findings indicate that the burden of childhood diarrhea is more strongly influenced by regional and contextual factors than by sex alone, underscoring the importance of region-specific public health interventions, particularly in areas with persistently higher prevalence.

By analyzing individual, environmental, and community-level factors through the lens of the Social-Ecological Model (SEM), the research illustrated how various levels of influence shape health behaviors and outcomes. The SEM emphasized that health was not only determined by individual characteristics but also by interactions with social environments, communities, and broader societal structures [19]. The findings underscored the distinct contributions of each level in influencing diarrheal disease outcomes, providing insights into targeted public health interventions.

At the individual level, several factors were significantly associated with the prevalence of childhood diarrhea. Key determinants included the child’s age, gender, maternal age, dietary diversity score (DDS), and household wealth status. The analysis revealed that female children had lower prevalence of diarrhea compared to males. This finding might reflect gender-based biological differences in susceptibility to infections [23–25]. Furthermore, age was a critical factor, with the highest risk observed among 1-year-olds and a declining trend as children grew older. These results indicated that younger children, particularly those in their first year, were more vulnerable to diarrhea, possibly due to weaning, increased exposure to contaminated food, and immature immune systems. A study in Southeast Asian countries [23] highlighted the role of social determinants of health and disease, including sex, in the occurrence of diarrhea. It was plausible that variations in the prevalence of diarrhea may be attributed to environmental exposures that differ by sex and age. To gain a comprehensive understanding of how male and female exposures differ and their respective effects on the risk of developing diarrhea, further investigation was warranted [24–26].

While this study found that female children had lower prevalence of diarrhea compared to male children, it is essential to consider the environmental factors that may affect female children differently. Despite this lower prevalence, female children in vulnerable households, especially those in poor living conditions, were still exposed to environmental risks such as inadequate sanitation and unsafe water sources, which remain major contributors to diarrhea [24, 25]. In many low-income settings, gender norms may also influence the roles children play within their households, which could indirectly impact exposure to diarrheal pathogens. For example, in low-income households, young girls might be tasked with household chores such as fetching water or cleaning, increasing their likelihood of exposure to contaminated water sources and unclean environments [24, 25]. While female children may generally have lower prevalence of diarrhea, they were not immunized to the environmental hazards that contribute to the disease, particularly in poverty-stricken settings [26].

This study showed that the prevalence of diarrhea remained highest at 1 year and decreased at 2 and 4 years, possibly due to factors such as changes in immune system development, dietary habits, and exposure to environmental pathogens. The transition from breastfeeding to solid foods and increased mobility may also contribute to the age-related pattern of diarrhea prevalence. This result was consistent with findings from national and international studies, providing consistent evidence of the age-related pattern of diarrhea prevalence [27–29].The study found a higher prevalence of child diarrhea among young mothers compared to older mothers. This may be due to factors such as inadequate knowledge of childcare and hygiene practices, limited access to healthcare services, and socioeconomic challenges faced by young mothers [30–33].

Children with moderate dietary diversity (DDS 5–9 had a higher prevalence of diarrhea compared to those with low dietary diversity (DDS 1–4). A low dietary diversity score (DDS) can indicate inadequate nutrition, which may weaken a child’s immune system and increase susceptibility to infections, including diarrheal diseases [34]. This may be due to increased exposure to different foods and potentially unsafe or contaminated food sources in regions where hygiene and food safety are not guaranteed. While a diverse diet is crucial for overall nutrition, these findings highlighted the need for safe food handling practices in contexts where diarrhea prevalence is high [33–35]. Wealthier households were associated with a lower prevalence of diarrhea, reflecting the protective role of higher socioeconomic status. Wealthier families were more likely to have access to better healthcare, sanitation facilities, and clean water, all of which contributed to the reduced risk of diarrhea in children [36]. Children from the poorest households faced significant environmental risks that contributed to diarrhea [11]. The study highlighted that children from the poorest households had much higher odds of developing diarrhea compared to those from wealthier families, largely due to the poor living conditions that accompany poverty [13, 24]. Children from low-income families were often exposed to substandard housing conditions, such as homes with dirt floors and inadequate waste disposal systems [4, 34]. These environments can harbor pathogens, increasing the risk of diarrhea. Additionally, lower SES was associated with a higher prevalence of diarrhea, possibly due to limited access to healthcare and nutrition, further exacerbating the risk. Children from poorer households were less likely to receive vaccinations or healthcare interventions, that could mitigate the incidence of diarrhea [36–38].

At the environmental level, sanitation conditions and household infrastructure, particularly toilet types and flooring materials, were significant predictors of childhood diarrhea. Children living in households with standard toilet conditions had a significantly lower prevalence of diarrhea compared to those with non-standard toilets [21, 31]. This finding underscored the well-established role of sanitation in preventing diarrheal diseases. Improved toilet facilities reduced the likelihood of fecal contamination, thus minimizing exposure to pathogens that cause diarrhea. Poor households were less likely to have access to standard toilets and sanitation infrastructure, increasing the likelihood of open defecation or the use of makeshift toilets, both of which can lead to fecal contamination in the living environment. This contamination, whether through water, food, or surfaces, was a significant driver of diarrheal diseases [21, 31].

The interesting finding that children living in houses with semi-permanent and permanent floors had significantly higher prevalence of diarrhea compared to those living in houses with ground-sand floors may be due to several factors. One potential reason could be the difference in the level of sanitation and hygiene between the different types of floors [37, 38]. Dwellings equipped with semi-permanent and permanent flooring options might potentially offered improved access to clean drinking water sources, which could led to increased exposure to waterborne pathogens and contaminants, contributing to a higher risk of diarrhea [39]. Additionally, the construction materials and maintenance of semi-permanent and permanent floors might create environments that facilitated the survival and transmission of diarrheal pathogens, leading to an increased risk of diarrhea for children living in such houses [37, 38].

In rural areas, where ground-sand floors were more common, there may be less overcrowding and thus a reduced risk of pathogen transmission. Further research is needed to explore this relationship. Poor households often lived in substandard housing conditions, such as homes with dirt or ground-sand floors, which breed grounds for bacteria and parasites. Although the study found that children in homes with semi-permanent or permanent flooring had a higher prevalence of diarrhea, it is important to note that these conditions were more likely to be overcrowded or unsanitary in low-income settings, exacerbating exposure to diarrhea-causing agents [2, 31].

At the community level, the study examined the effect of geographic and regional variability on childhood diarrhea outcomes. Children living in rural areas were found to have higher rates of diarrhea compared to those in urban areas. This finding reflected the persistent disparities in access to clean water, sanitation, and healthcare between rural and urban regions in Indonesia. Rural areas often face infrastructural challenges that hinder the provision of essential services like sanitation and clean water, which were crucial for preventing diarrhea. This observation was supported by several references [9, 19, 21, 23, 34] that highlighted socioeconomic disparities and challenges in resource allocation for rural communities as contributing factors to the differences in diarrhea prevalence between rural and urban areas. Furthermore, variances in health education methodologies and healthcare provision between rural and urban populations could also contribute. Urban inhabitants may benefit from improved access to knowledge concerning waste disposal and childcare techniques, which can aid in diarrhea prevention [34].

The multilevel analysis also included random effects to account for regional variability. The results showed minimal variation in childhood diarrhea rates across different regions once other factors were considered. The low intraclass correlation coefficient (ICC) suggests that regional differences have a limited impact on diarrhea prevalence, with most variability explained by individual and environmental factors. This indicates that while geographic disparities exist, they are largely overshadowed by more immediate household-level conditions and socioeconomic factors.

The combination of these factors being female, 1 year old and from the poorest household can create a heightened vulnerability to diarrhea due to a confluence of environmental and social risks. Female children in their first year of life, particularly those in impoverished settings, may not only face direct exposure to pathogens through poor sanitation but also through societal and household roles that place them in risky environments. For instance, poor living conditions may mean that female infants and toddlers were cared for in overcrowded or unsanitary environments, where their exposure to contaminated food, water, or surfaces was inevitable [39–41].

The hierarchical levels of predictors identified in this study had important implications for public health policies aimed at reducing childhood diarrhea in Indonesia. At the individual level, targeted educational programs for mothers, particularly younger mothers, can help improve hygiene practices and child nutrition. At the environmental level, efforts to improve access to safe sanitation facilities and clean water were essential for reducing the burden of diarrheal diseases. Lastly, community-level interventions, particularly in rural areas, should focus on addressing infrastructure gaps and ensuring equitable access to healthcare and sanitation services. Overall, the findings suggested that addressing childhood diarrhea in Indonesia requires a comprehensive, multi-tiered approach that integrates individual, environmental, and community-level interventions. The study’s findings provided valuable evidence to inform public health policies by identifying the factors that most significantly contribute to diarrhea at the individual, environmental, and community levels. This understanding enabled the design of targeted interventions, such as improving child nutrition and promoting breastfeeding practices for high-risk groups.

Conclusion

The study’s a Mixed-effects multilevel poisson regression analysis identified several individual, environmental, and community-level factors significantly associated with childhood diarrhea in Indonesia. At the individual level, the analysis revealed that female children had a lower prevalence of diarrhea compared to males, while age had a non-linear effect, with the highest odds observed at age one, decreasing by age four. Older mothers were found to have children with lower prevalence of diarrhea, which highlighted the importance of maternal experience and education in improving child health outcomes [42, 43]. Moreover, children with a moderate dietary diversity score (DDS 5–9) had a higher prevalence of diarrhea compared to those with a lower DDS, suggesting that dietary diversity, while important for nutrition, may pose a risk when hygiene and food safety were inadequate. The socioeconomic status of families played a substantial role, with children from the wealthiest families having a significantly lower prevalence of diarrhea compared to those from the poorest families.

Environmental factors also emerged as crucial determinants of childhood diarrhea. Access to standard toilets was associated with a lower prevalence of diarrhea, indicating the importance of improved sanitation in reducing disease prevalence [33, 36]. Surprisingly, children living in homes with semi-permanent and permanent floors had a higher prevalence of diarrhea than those residing in houses with ground-sand floors. This finding may reflect confounding factors related to overcrowding or urban environments. Additionally, children residing in rural areas were found to have a higher risk of diarrhea than their urban counterparts, underscoring the disparities in access to clean water, sanitation, and healthcare between rural and urban regions. In terms of regional variability, the random effects model suggested that regional differences had a minimal impact on the prevalence of diarrhea after adjusting for other variables. From a multilevel perspective, regional variation in childhood diarrhea prevalence was minimal, as reflected by consistently small random-intercept variances and ICC values across all models, indicating that unobserved regional factors contributed little to overall variability after adjusting for observed covariates. Although Model 1, which included only individual-level characteristics, demonstrated the best statistical fit based on AIC and log-likelihood, suggesting that proximal individual and household determinants accounted for most explainable variation, Model 3 was retained for substantive interpretation because it provides a more comprehensive and conceptually robust assessment of diarrhea determinants across multiple levels. The limited additional explanatory power of environmental and community-level variables likely reflects contextual homogeneity across regions or strong correlations with individual socioeconomic factors. Collectively, these findings imply that interventions to reduce childhood diarrhea in Indonesia should prioritize individual- and household-level improvements particularly addressing socioeconomic inequalities, strengthening maternal knowledge and caregiving practices, and promoting safe food handling while continuing to invest in sanitation infrastructure, especially in rural areas.

In conclusion, public health interventions targeting childhood diarrhea in Indonesia should focus on improving sanitation, particularly in rural areas, and addressing socioeconomic disparities. Efforts to educate mothers on proper hygiene, safe food practices, and child nutrition are essential, especially for younger mothers and those in poorer households. Additionally, policies that enhance access to clean water and improve sanitation infrastructure, particularly in rural and low-income areas, can play a vital role in reducing the burden of diarrhea among Indonesian children.

Limitations and strengths

This study has several limitations that should be considered when interpreting its findings. Firstly, the use of a cross-sectional dataset limits the ability to establish causal relationships, making it necessary for future research to employ longitudinal designs to explore the pathways leading to diarrhea in children [44]. Secondly, reliance on respondents’ recall of past events may introduce recall bias [42], potentially impacting the accuracy of the reported data. Additionally, the analysis utilized secondary data, which constrained the inclusion of all possible factors influencing childhood diarrhea, particularly behavioral aspects that were not available in the dataset [44].

Author contributions

IS was responsible for conceptualizing and designing the study, analyzing, and interpreting the data, as well as drafting the manuscript. FA designed, supervised, and coordinated the study and critically reviewed the manuscript. PS. involved in the analysis and interpretation of the data. IS supported to write the discussion part. All authors thoroughly reviewed and approved the manuscript for submission.

Funding

This work was supported by the Faculty of Public Health, Diponegoro University, Indonesia under Grant (053/UN75.92/KS/2021).

Data availability

The datasets used in this study are publicly available from the Demographic and Health Survey (DHS) program at [https://dhsprogram.com]. Access requires registration and approval from DHS.

Declarations

Ethics approval and consent to participate

Ethical approval for this study was granted by the Ethics Committee of the Faculty of Public Health, Diponegoro University (Decision No. 82/EA/KEPKFKM/20212, dated March 22, 2021). The study was conducted in accordance with the Declaration of Helsinki. This research used secondary data from the Indonesia Demographic and Health Survey (IDHS), which follows ethical standards including obtaining informed consent from all participants prior to data collection.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.