Assessing Risk Factors for Short Birth Interval Hot Spots using Geographically Weighted Regression: Findings from a Nationally Representative Survey Data

Short Birth Interval (SBI) is a public health problem in most low- and lower-middle-income countries. Understanding geographic variations in SBI, particularly SBI hot spots and associated factors, may help intervene with tailored programs. This study identied the geographical hot spots of SBI in Bangladesh and the factors associated with them. ` Methods We analyzed women’s data extracted from the 2017/18 Bangladesh Demographic and Health Survey and the healthcare facility data extracted from the 2017 Service Provision Assessment. Moran’s I was used to examine the spatial variation of SBI in Bangladesh whereas the Getis-Ord G* i (d) was used to determine the hot spots of SBI. The Geographical Weighted Regression (GWR) was used to explore the spatial variation of SBI on explanatory variables. The explanatory variables included in the GWR were selected using the exploratory regression and ordinary least square regression model. in and No high maternal parity, and were signicantly associated with in the Sylhet region. 34 less at the rst birth a protective in the Rajshahi and Khulna


Introduction
Despite the tremendous strides in improving maternal and child health, particularly in the Millennium Development Goals period of 2000-2015, maternal and child morbidity and mortality continue to be a major problem in low-and lower-middle-income countries (LMICs). Globally, almost 300,000 women die annually from causes related to pregnancy and childbirth, and 94% of these occur in LMICs 1 . Also, over 80% of 5.2 million global under-ve mortality occurs in LMICs, although they only account for 52% of the global under-ve population 2 . Around half of these deaths occur within the rst 28 days of children's lives and an additional 1.5 million occur within 1-11 months of birth 2 . Anemia, placental abruption, placenta previa, uterine rupture, preterm birth, low birth weight and congenital malformations are some of the dominant causes for many of these deaths, and they are preventable [1][2][3][4] . Inadequate birth spacing is incontrovertibly linked to many of these adverse health outcomes 5,6 . The reason is the lack of su cient time to return to the normal pregnancy metabolic state before the next pregnancy that affects women's nutritional, physical and emotional health 7 . Moreover, pregnancy Page 3/18 in short intervals decreases maternal foetal concentrations, especially during the second and third trimesters, which can weaken connective tissue by preventing collagen cross-linking, thus increasing the risk of adverse pregnancy outcomes 8 . The World Health Organization recommends this interval to be at least 33 months. A shorter duration than this is identi ed as a short birth interval (SBI) 9 . Around one-fourth of pregnancies that end with live births in LMICs occur in SBIs [10][11][12] .
Previous research on SBI in LMICs primarily focused on sociodemographic risk factors of mothers, children, and other members of households [10][11][12][13][14] . Another key variable of interest is the area-level difference in SBI rates, measured across geographical locations (e.g., urban/rural and administrative divisions) [10][11][12][13][14] . An important aspect of this variable is its ability to modify other risk factors of SBI, such as partners' educational attainments and children's survival status 15 . The reason is that in LMICs, socio-economically advantaged and disadvantaged people usually live in clusters. As a result, risk factors of SBIs could be different between various clusters in an area. However, in literature, these differences are not receiving the necessary attention and most researchers are often producing overall estimates for broader jurisdictions. This overall estimate masks the local-level variation and, thereby, limits our ability to formulate policies and programs targeting speci c areas or segments of people where much-needed public health interventions are needed 15 . Consequently, in many LMICs, we see mismatches between service requirements and service availability and misuse of limited but valuable resources. This is particularly true for Bangladesh, where healthcare policies and programs are usually adopted nationally. This approach directly or indirectly considers a uniform situation across the country and does not account for locallevel needs. With this study, using two nationally representative samples of Bangladesh, we aimed to determine the area-speci c differences in SBI and explore their risk factors.

Methods
This study draws on data from the 2017/18 Bangladesh Demographic and Health Survey (BDHS), which is a nationally representative data source and provides estimates of reproductive health, maternal and child health.
The survey is part of the Demographic and Health Survey Program conducted in 90 LMICs. In Bangladesh, the Ministry of Health and Family Welfare supervised the survey; its partner organizations, The National Institute of Population Research and Training along with Mitra and Associates (an independent research rm) implemented this survey at the eld level. Several development partners, including UNFPA and UNDP, provided nancial support for this survey.
Following a two-stage strati ed random sampling approach, the survey collected data from women of  years old living in the selected households. At the rst stage of sampling, the survey selected 675 Enumeration Areas (EAs, clusters) covering urban and rural areas as well as eight administrative divisions of Bangladesh. The EAs were selected randomly from a list of 293,579 EAs created by the Bangladesh Bureau of Statistics as part of conducting the 2011 National Population Census, the most recent population Census in Bangladesh. The household listing operation was conducted at the second stage of sampling and 30 households were selected from each EA through probability proportional to the sample size. A total of 20,160 households were selected, of which data collection was undertaken in 19,457 households with over 96% inclusion rate. There were 20,376 eligible women in the selected households. Of them, data was collected from 20,127 women with a response rate of 98.8%. Finally, data of 5,941 women were included in this study by applying the following inclusion criteria: (i) the woman had at least two pregnancies, of which the most recent one ended with live birth within ve years of the survey date, (ii) the second most recent pregnancy ended with live birth or termination, and (ii) the end dates of both pregnancies and the interval were recorded.
The survey also collected the geographical location of each EA using the Global Positioning System (GPS). The GPS reading was made at the center of each EA, while efforts were made to ensure adequate satellite signal strength. For this, the data collectors ensured that they were not near any tall building or under any big tree. The points recorded were then randomly displaced to 5 kilometers in the rural area and 2 kilometers in the urban area.
The DHS recorded those displaced EA points in a shape le (geographical data le) and released it along with the survey data.
We also used geographical data of the 2017 Bangladesh Health Facility Survey (BHFS), a nationally representative survey of healthcare facilities. This dataset includes 1524 healthcare facilities selected randomly throughout the country covering primary, secondary and tertiary level healthcare facilities. A detailed description of the sampling procedure of both surveys has been published in their survey reports 16,17 .

Outcome variable
The outcome variable is SBI, de ned as an interval of at least 33 months between the two most recent births. The BDHS recorded this data in months by subtracting the date of birth of the most recent child to the date of birth or termination of the second most recent child. These dates were collected from the birth registration reports or immunization cards. If these were not available, mothers were requested to recall their memories. These women were referred to memorable events like the national or local election, ood to help them recall their memories to estimate the accurate date of births.

Explanatory variable
The explanatory variables considered in this study were identi ed through a comprehensive literature search in the following ve databases: Medline, Embase, Web of Science, CINHAL, and Google Scholar. A pre-designed search strategy was used with relevant keywords, including birth interval, birth spacing, and short birth interval. To identify the key factors, special attention was paid to the ve studies conducted in Bangladesh 13,14,18−20 and the studies conducted in some other LMICs 10-12 . The factors were age at birth (≤ 19, 20-34, ≥ 35), age at rst childbirth (≤ 19, 20-34, ≥ 35), educational status of women and their husbands (no education, primary, secondary, higher), and women's employment status (employed, not employed), sex of households' head (male, female), women's exposure to mass media (little exposed, moderately exposed, highly exposed) and the number of children ever given birth (≤ 2, > 2). Survival (yes vs no) of the second most recent child was also considered. The average distances from respondents' houses to the nearest healthcare facilities that offer reproductive healthcare services were also considered an explanatory variable. The average distance was calculated at the divisional level using the administrative boundary link method based on the geographical variables of the 2017/18 BDHS and 2017 SPA datasets 21 .

Statistical analysis
We rst determined the variation in the prevalence of SBI across the places of residence and administrative divisions using the chi-square test. The proportions of SBI and the explanatory variables were estimated across the EAs (clusters). The survey weight was applied using STATA's svy command to calculate the proportions. The estimated proportions were then merged with the GPS cluster locations and examined whether any geographical difference persists in the distribution of SBI in Bangladesh. For this, the hot spot analysis was conducted with the Moran's I and Getis-Ord G*i (d)statistics to examine spatial variation and clustering of SBI across EAs, respectively. We considered a False Discovery Rate (FDR) correction while using the Getis-Ord G*i (d) statistics to account for multiple, dependent tests. The importance of considering the FDR correction method in DHS data has been described elsewhere 22 . We ran Ordinary Least Square Regression (OLS) to identify the predictors of observed spatial patterns of SBI in Bangladesh. We checked the model assumptions for OLS and multicollinearity 23,24 . For this, the variables included in the OLS were rst determined carefully by using explanatory regression, a data mining tool was used to select the variables as Stepwise Regressions do. The explanatory regression model identi es the variables to be included in the OLS that meet the model's assumptions.
The OLS ts a linear regression to all of the data in the study area. Therefore, it did not answer the questions, (i) why clustering (if any) of SBI occurs in Bangladesh? and (ii) what factors are associated with the observed clustering? It is also important to know whether the relationships between the outcome variables and explanatory variables vary across areas and which explanatory variables show substantial in uence. The GWR answers these questions. We ran GWR with the variables that met the assumptions of the OLS model, as recommended in the previous studies 25,26 . The advantage of this approach is that the model produces an estimate for each EA

Background characteristics of the respondents
This study includes data of 5,941 women who came from 672 clusters in the 2017/18 BDHS. The crude and agestandardized characteristics of the study sample are shown in Table 1. The average age of participants at their most recent births was 25.93 years (SD ± 5.13). On average, they received 6.12 years of education (SD ± 3.70) and gave birth to 2.85 (± 1.18) children. More than a quarter of the total live births occurred in SBI (26.26%).

Hot spots and cold spots of short birth interval in Bangladesh
We found evidence of statistically signi cant clustering of SBI in the study area (Moran's I = 0.330590, p < 0.01). The Getis-Ord G statistic revealed the high clustering across EAs (p < 0.01) (Fig. 1). A relatively high number of SBI hot spots were found in the Sylhet division, and SBI cold spots were found in parts of the Rajshahi and Khulna divisions.

Model comparisons: OLS and GWR
The results of the OLS model are presented in

Predictors of short birth interval: hot spots and cold spots
The cluster-wise coe cients of the GWR model are plotted in Fig. 2a-f. In the Sylhet division, where a majority of the SBI hotspots are located, the signi cant predictors of SBI are no formal education of husbands (Fig. 2c), women doing no formal jobs (Fig. 2d), having three or more children (Fig. 2e), and experiencing the death of a Page 8/18 child (Fig. 2f). On the contrary, in the Rajshahi and Khulna divisions where most of the SBI coldspots were located, maternal age of 34 years or less at the rst birth (Fig. 2b) was a signi cant protector of SBI.

Discussion
This study provides evidence that along with the socio-demographic factors known to be associated with a high prevalence of SBI in Bangladesh 14,19,27 , area-level variations are also important. This is the rst study in the Bangladesh context that explored the factors determining such area level variations, including factors responsible for SBI hot spots and cold spots. Unemployed women, those who gave birth to three or more children, experienced the death of a child, or whose husbands received no formal education were signi cantly more likely than others to be located in SBI hot spots. Women who gave their rst birth at the age of 19 years or earlier and 20-34 years were signi cantly more likely to be living in SBI cold spots. These ndings are robust as we have selected these variables following a proper statistical model-building technique. Therefore, we believe these ndings are reliable and implications in designing policies and tailored programs. Previous studies in Bangladesh consistently reported high rates of early marriage, relatively low age at rst birth, and low rates of formal education in the Sylhet division 31,32 . These characteristics, both individually and together, can affect SBI. Our results also suggest that these factors are the signi cant predictors of SBI in the SBI hot spots area in the Sylhet division. A possible reason for such association is that couples with these characteristics are less likely to access maternal healthcare services, including intrapartum, birthing, and post-partum care [33][34][35] .
Moreover, in the current form of maternal healthcare services delivery in Bangladesh, post-partum care visit on the fourth weeks of the live birth is dedicated to providing counselling regarding family planning and contraception 36 . This approach is not helpful in increasing family planning and contraception services because post-partum care visits at the fourth week of live birth are still very low in Bangladesh 36 . Indeed, many women in Bangladesh have a misapprehension that once a live birth has occurred, the issue of pregnancy is over, and it is unnecessary to visit a healthcare center for post-partum care, particularly at the fourth week of live birth. This tendency is even higher among women of disadvantaged backgrounds. Consequently, many women end up with another pregnancy in a short interval. Additionally, women with these characteristics are less likely to receive family planning counselling which is offered at the household level by family planning workers 37 .
Although the underlying reasons for such low use of services in the Sylhet division have yet not been explored, we believe this is mainly due to inadequate knowledge of reproductive goals 37 . Moreover, there are studies in Bangladesh, including the Sylhet division, that found women of disadvantaged backgrounds are highly in uenced by religious misconceptions. For instance, many couples believe that the religion Islam (the religion of over 90% of the population in Bangladesh) supports taking children as many as they want, and contraception use is comparable to the killing of humans [37][38][39] . Consequently, the current approach to family planning services, including visits to women's homes by family planning workers every 14 days to provide reproductive counselling and contraception, may not work effectively in this division. Indeed, several recent studies reported a high prevalence of unmet need for contraception and particularly modern contraception in Sylhet compared to the other divisions 40,41 . Also, the prevalence of unintended pregnancy in this division is higher than in other parts of Bangladesh 37,42 , and most of them occur in shorter intervals of the previous births 27 . Also, a relatively high proportion of men in the Sylhet division is either migrated aboard or locally 43 . Women having migrated partners are less likely to receive maternal healthcare services, a nding reported in Nepal 44 and Bangladesh 45 .
Consequently, they have inadequate knowledge regarding birth spacing.
Literature suggests that the prevalence of adverse pregnancy outcomes, including child mortality, is relatively high in the Sylhet division and low in the Rajshahi and Khulna divisions 46,47 and are aligned with the SBI hot spots and cold spots, respectively. There seems to be a two-way relationship between adverse pregnancy outcomes and SBI; adverse outcomes occur due to a relatively high number of births in shorter intervals and vice versa. Findings from the studies in other settings of LIMCs 48-50 demonstrate relatively high birth intervals among couples with fewer children. Couples experiencing the death of a child or even witnessing such an event among the neighbours, are usually motivated to take another child considering the uncertainty, often in a shorter interval 49 . Similarly, women who are not engaged in formal jobs are likely to take babies in short intervals 15 .
The ndings of this study highlight the need for tailored programs in Bangladesh in general and the Sylhet division in particular to reduce the prevalence of SBI. Strengthening reproductive healthcare service delivery, including intrapartum, delivery, postpartum, and postpartum contraceptive services should be prioritized.
Providing integrated reproductive healthcare services may help improve the current service delivery. Also, tailoring service modality considering the divisional level barriers is needed 36 , as it is not possible in the current uniform top-down policy approach 35,36 .
As far we know, this is the rst study that explored the hot spots and cold spots of SBI and its associated factors in Bangladesh. The explanatory variables considered in this study were chosen based on a comprehensive review of the existing literature and nally by following the proper statistical model building techniques. The data were collected from two nationally representative surveys conducted in the same year using validated questionnaires.
However, the analysis of cross-sectional data means that the ndings are correlational only. To ensure the privacy of the respondents, the BDHS displaced cluster locations that we used in plotting our results in maps, up to ve km in rural and two km in urban areas. Thus, the areas plotted in the maps as SBI hot spots or cold spots are slightly different from the actual areas from where data were collected, although divisions of data collection were the same. However, the ndings are still valid as our results only highlight the potential areas of SBI hot spots or cold spots. Moreover, besides the socio-demographic factors included in this study, area level and environmental factors could also be important predictors of SBI hot spots and cold spots in Bangladesh, but we could not consider those variables in our analysis as they were not available. However, our adjusted variables explained around 65% of the total occurrences of SBI hot spots and cold spots.

Conclusion
We found evidence of substantial geographical variations in SBI in Bangladesh. SBI hot spots are mainly located in the Sylhet division, and SBI cold spots are mainly located in parts of the Rajshahi and Khulna divisions. Divisional variations in socio-demographic characteristics of women and their husbands were the main reasons for such geographical variation in SBI hot spots and cold spots. Targeted and divisional level policies and programs to provide integrated intrapartum, birthing, and postpartum care, including postpartum contraception, are needed to reduce the prevalence of SBI in Bangladesh in general and in the Sylhet division in particular.

Declarations
Declaration of interests Hot spots and cold spots of short birth interval in Bangladesh Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.