Understanding the determinants of postnatal care uptake for babies: A mixed effects multilevel modelling of 2016–18 Papua New Guinea Demographic and Health Survey

Background: Papua New Guinea (PNG) recorded 22 neonatal deaths out of every 1,000 livebirths in 2019. Some of these deaths are related to complications that arise shortly after childbirth; hence, postnatal care (PNC) utilisation could serve as a surviving strategy for neonates as recommended by the World Health Organisation. National level study on determinants of PNC uptake in PNG is limited. Utilising the Bronfenbrenner’s Ecological Model of Human Development, the study aimed at assessing determinants of PNC utilisation for babies by their mothers aged 15–49 in PNG. Methods: The study used data from the women’s file of the 2016–18 PNG Demographic and Health Survey (2016–18 PNGDHS) and a sample of 4,908 women aged 15–49 who had complete information on the variables of interest to the study. Nineteen (19) explanatory variables were selected for the study whereas PNC for babies within first two months after being discharged after birth was the main outcome variable. At 95% confidence interval (95% CI), six multilevel logistic models were built. The Akaike Information Criterion (AIC) was used to assess models’ fit. All analyses were carried out using STATA version 14.0. Results: Generally, 31% of the women utilised PNC for their babies. Women with primary education [aOR = 1.42, CI = 1.13–1.78], those belonging to the middle wealth quintile [aOR = 1.42, CI = 1.08–1.87], working class [aOR = 1.28, CI = 1.10–1.49], women who had the four or more ANC visits [aOR = 1.23, CI = 1.05–1.43], those with twins [aOR = 1.83, CI = 1.01–3.29], women who belonged to community of medium literate class [aOR = 1.75, CI = 1.34–2.27] and those of moderate socioeconomic status [aOR = 1.60, CI = 1.16–2.21] had higher odds of seeking PNC for their babies. The odds to seek PNC services for babies reduced among the cohabiting women [aOR = 0.79, CI = 0.64–0.96], those at parity four or more [aOR = 0.77, CI = 0.63–0.93], women who gave birth to small babies [aOR = 0.80, CI = 0.67–0.98] and residents in the Highlands region [aOR = 0.47, CI = 0.36–0.62]. Conclusions: Maternal education, wealth quintile, occupation, partner’s education, ANC visits, marital status, parity, child size at birth, twin status, community literacy and socioeconomic status as well as region of residence were associated with PNC uptake for babies in PNG. Variation in PNC uptake for babies existed from one community/cluster to the other. There is the need to strengthen public health education to increase awareness about the benefits of seeking PNC services for babies among women in PNG. Such programs should consider maternal and community/cluster characteristics in their design. The study adopted a cross sectional survey design and made use of women’s file of the 2016–18 Papua New Guinea Demographic and Health Survey (2016–18 PNGDHS). The 2016–18 PNGDHS is the third in the series of the DHS surveys conducted in the country. The survey was implemented by the National Statistical Office (NSO) of PNG. The survey started data gathering from October 2016 to December 2018. All necessary technical and advisory support were provided by the NSO. The ICF International provided technical assistance through the DHS Program, which offers support and technical assistance for the implementation of population and health surveys. The survey covered areas such as fertility, family planning, breastfeeding practices, nutritional status of children, maternal and child health, childhood immunisation, adult and childhood mortality, women’s empowerment, domestic violence, malaria, awareness and behaviour concerning HIV/AIDS and other sexually transmitted infections (STIs). The survey applied a stratified sampling technique and in all, a total of 18,175 women aged 15–49 were identified for individual interviews. However, 15,198 women were completely interviewed which yielded a response rate of 84%. Details of the sampling procedures, pretesting of instrument, fieldwork, data processing and analysis can be obtained from the 2016–2018 PNGDHS report [10]. Meanwhile, the present study focused on 4,908 women aged 15–49 who had complete information about the variables of the study. During the 2016–18 PNGDHS, all women who had birth(s) in the 5 years preceding the survey were asked if they had postnatal check-up for their children after exiting the health facilities where they delivered. This was posed as “Did any health care provider or a traditional birth attendant check on (NAME)’s health in the two months after you left?” accompanied by “yes”, “no” and “don’t know” responses. Therefore, the outcome variable for this study was ‘‘postnatal care for babies within first two months after exiting the facility where baby was born’’, defined as having received a postnatal check-up for the baby within first two months after exiting place baby was delivered. Those who affirmed “Yes” were recoded as “1” and “No” recoded as “0”. For precision in responses, “don’t know” responses were excluded from the analysis. Also, the outcome variable excludes pre-discharge checks for babies within facility where births took place to aid assess babies that actually received PNC services after discharge. Nineteen (19) explanatory variables were selected for the study [21–23]. These are: age, education, wealth quintile, marital status, occupation, parity, health decision making, partners’ education (maternal factors/microsystem), sex of child, twin status and size of child at birth (child factors/microsystem); community literacy level, community socioeconomic status, residence and region (mesosystem); access to mass media, place of delivery and antenatal care (ANC) visits (exosystem); and covered by health insurance (macrosystem). For clarity of presentation, educational status was recoded into “no education”, “primary”, “secondary/higher”. Occupation recoded into “not working” and “working”; and partner’s education was recoded into “no education”, “primary” and secondary/higher”. Considering the current fertility rate of PNG which is 4.2 children per woman [10], total children ever born was recoded into “one birth”, “two births”, “three births” and “four or more births”. Access to mass media was determined from three principal variables: frequency of reading newspaper/magazine; frequency of listening to the radio; and frequency of watching television which were asked during the 2016–18 PNGDHS. Each of these variables had three responses: ‘not at all’, ‘less than once a week’, and ‘at least once a week’. A composite variable was created whereby all ‘less than once a week’ and ‘at least once a week’ responses were categorised as having access to mass media whilst ‘not at all’ was considered as not having access to mass media. Also, ANC visits were recoded into “less than four visits” and “four or more visits”. For child-level factors, twin status was recoded as “single birth” and “twins” and finally, child’s size at birth recoded as “large”, “average” and “small”. Health decision making was recoded as “alone”, “respondent/partner”, and “others”; place of delivery recoded into “home”, “health facility” and “others”; community literacy level (proportion of women who can read and write at all) was recoded into “low”, “medium” and “high”; and community socioeconomic status was recoded into “low”, “moderate” and “high”. Community literacy level was computed from the women who could read and write at all [24]. Also, community socioeconomic status was measured as the percentage of households in the poorest quintile of Papua New Guinea’s household wealth index [25]. Missing variables were low (3.4%) and were omitted. The present study assessed determinants of PNC uptake for babies in PNG. Based on the focus of the study, the following steps were involved in analysing the data. The weighting factor inherent in the dataset (v005/100000) and the “svy command” were applied to cater for over and under sampling biases and to account for the complex survey design and generalizability of the findings respectively. Next, computation of women who received PNC for their babies after two months of exiting place baby was born was done (data not shown) and further analysed providers of PNC. Thereafter, a univariate computation of independent variables was done to describe the sample whereas a bivariate analysis was done for the independent variables across PNC utilisation with their chi square test of independence reported. The chi square test of independence helped to gauge independent variables which were not associated with the outcome variable, hence, excluding such variables from the inferential analyses. Also, the “VIF” command was applied to assess collinearity among the explanatory variables and the results indicated no evidence of multicollinearity existing between the explanatory variables (Maximum VIF = 2.55, Minimum VIF = 1.02, Mean VIF = 1.46) (Appendix 1). Subsequently, at 95% confidence intervals (95% CIs), six (6) multilevel logistic models were built. The first was a null model (Model 0) to account for variability in PNC which can be attributed to the clustering of the primary sampling units (PSUs) without the effect of micro, meso, exo and macro-system. Further, Model I and Model II considered micro and mesosystem-level factors respectively. Model III and IV were fitted to cater for exosystem-level factors and macro system-level factors. Finally, a complete model containing all the factors (Model 0, I, II, III and IV) were constructed (Model V). The results for the fixed effects were presented in adjusted odds ratio (aOR) and any odds less than one was interpreted as reduced likelihood of PNC whilst an odds higher than 1 meant otherwise. Since the models were nested, the Akaike Information Criterion (AIC) was used to measure the model fit [26]. The random effects which are measures of variation of PNC utilisation across communities or clusters, were expressed using Intra-Class Correlation (ICC) and PSUs variance [26, 27]. These were calculated to gauge the variation of PNC utilisation across clusters and the proportion of variance explained by successive models. The entire analyses were done with the aid of STATA version 14.0. The present study dwelt on an already existing data and since the authors were not involved in the data gathering, no ethical clearance was sought. However, the authors sought for access to use the data from Measure DHS and after obtaining permission, the data was downloaded. The dataset is freely available to the public at https://dhsprogram.com/data/dataset/Papua-New-Guinea_Standard-DHS_2017.cfm?flag=1. However, Measure DHS report has documented details of ethical issues considered in gathering the 2016–18 PNGDHS data [10].