2.1 Data source
The current study used the most recent Demographic and Health Survey of Bangladesh conducted during 2017-18 (hereby referred to as BDHS 2017-18). The National Institute for Population Research and Training (NIPORT) has conducted BDHS 2017-18 under the stewardship of the Ministry of Health and Family Welfare (MoHFW) of Bangladesh. The survey provided crucial information on information on childhood mortality levels, maternal and child health, fertility and fertility preferences, utilization of family planning methods, newborn care, women’s empowerment, selected non-communicable diseases (NCDS) and availability and accessibility of health and family planning services at the community level. The survey follows a two-stage stratified sample design. Further details regarding sample design, survey instruments, fieldwork and training of staff, data collection and processing, and response rates are available in the BDHS 2017-18 reports [24].
We used the data for 8759 children aged under-five years born to 7562 mothers aged 15–49 years in Bangladesh. However, we dropped the records of 361 children who were not alive during the survey period and had no information regarding their morbidity status. Therefore, the analytical sample for this study is 8398 children under five years of age.
2.4 Explanatory variables
Guided by extant research, we identified relevant factors that are associated with the occurrence of morbidity among children [5, 12, 13]. Accordingly, we included relevant explanatory variables, conditional upon their availability in BDHS 2017-18. The child-related characteristics are – age in years (less than one, one, two, three, four) and gender (male, female). The parent-related characteristics are – mother’s level of education (no formal education, upto primary, secondary and above), father’s level of education (no formal education, upto primary, secondary and above). The household-level factors are – household sanitation condition (poor, average, good), household members drink treated water (no, yes), type of handwashing place (Private space, Public place, No handwashing place), shares toilet with other households (Not shared, Shared by two households (HH), Shared by three HH, Shared by four and more HH), wealth quintile (poorest, poor, middle, rich, richest), the religion of household head (Islam, Hinduism, others). Further, the season during the interview (Summer, Monsoon, Winter), place of residence (City Corporation, urban areas other than City Corporation, Rural areas), and administrative division (Dhaka, Chittagong, Barisal, Khulna, Mymensingh, Rajshahi, Rangpur, Sylhet) were also included.
Taking a cue from extant research, the household sanitation condition variable was constructed from three variables – type of source of drinking water, type of sanitation facility and the number of members per room in the household [25]. Respondents were asked about the source of household drinking water and as per prevalent standards, we recoded the source of household drinking water into two categories – “unimproved” coded as 0 (consisting of “dug, open well”, “river”, “pond”, “truck” and “bottled” categories from the original variable) and “improved” coded as 1 (consisting of “piped”, “tube well”, “hand pump”, “covered well” and “rainwater” categories from the original variable) [26]. Similarly, we recoded the type of household toilet facility into – “unimproved” coded as 0 (consisting of “defecation in open fields” and “traditional pit latrine” categories from the original variable) and “improved” coded as 1 (consisting of “ventilated improved pit latrine” and “flush toilet” categories from the original variable) [26]. Similarly, households with less than 3 members per room were coded as “1” and those with 3 or more members were coded as “0”. After this, we added the three variables to obtain a household sanitation condition score. Households with a score of 3, score of 2 and score less than 2 were categorized as having “good”, “average” and “poor” sanitation condition respectively.
To avoid multicollinearity, we coded a new wealth quintile variable after excluding information on household water source and toilet facility. The wealth quintile variable was prepared using standard procedures that are documented elsewhere [27].
2.5 Statistical methods
We performed bivariate and multivariate analysis to realize the study objectives. Owing to the categorical nature of the outcome variable, the bivariate association was examined using the chi-square test for association. Equivalently, multivariable analysis was performed by estimating multinomial regression models. The multivariate association of morbidity status of children with the explanatory variables was shown using relative risk ratios. Relative risk ratio gives the risk (multiple times) of having comorbidity (or single morbidity) compared to having no morbidity among those children belonging to a particular category of an explanatory variable given the effect of all the other explanatory variables remain constant [28]. To show effectively the effect of different risk factors on child morbidity status, we have summarized the regression output into graphs of predicted probability [29].
We checked for multicollinearity in the regression model and found the mean value of the variance inflation factor (VIF) to be less than 1.25. Therefore, multicollinearity is negligible [30]. Further, the Hausman-McFadden test revealed that our estimated model did not violate the independence from irrelevant alternatives (IIA) assumption [31]. All statistical estimations were performed using the STATA software version 13.0 [32].