Identifying inequities in maternal and child health through risk stratification to inform health systems strengthening in Northern Togo

Objective In Togo, substantial progress in maternal and child health is needed to reach global development goals. To better inform clinic and community-based health services, this study identifies factors associated with maternal and child health care utilization in the Kara region of Northern Togo. Methods We conducted a population-representative household survey of four health clinic catchment areas of 1,075 women of reproductive age in 2015. Multivariable logistic regression was used to model individual and structural factors associated with utilization of four maternal and child health services. Key outcomes were: facility-based delivery, maternal postnatal health check by a health professional within the first six weeks of birth, childhood vaccination, and receipt of malaria medication for febrile children under age five within 72 hours of symptom onset. Results 83 percent of women who gave birth in the last 2 years delivered at a health facility. In adjusted models, the strongest predictor of facility delivery in the rural catchment areas was proximity to a health center, with women living under three kilometers having 3.7 (95% CI 1.7, 7.9) times the odds of a facility birth. Only 11 percent of women received a health check by a health provider at any time in the postnatal period. Postnatal health checks were less likely for women in the poorest households and for women who resided in rural areas.

This study was reviewed and approved by the Albert Einstein College of Medicine Ethics Review Committee (#005127) and the Comité de Bioéthique pour la Recherche en Santé (#031/2014/CBRS) in Togo. Secondary analysis of the de-identified dataset was deemed exempt from City University of New York IRB review. Prior to study participation, all participants provided written informed consent. All respondents were ages 18 and older and able to consent to the study as an adult. The study took place in the Kozah district of the Kara region in Togo, which covers an area of about 1,075 km2 with an estimated population of 225,259 [18]. Four catchment areas of public health clinics operated by the Ministry of Health were selected by the District Health Director based on the limited availability of maternal and child health services and where it was of interest to identify factors contributing to low utilization rates in the area. The four catchment areas were the urban area of Adabawere and three rural areas: Kpindi, Sarakawa and Djamdé. In 2014, the coverage rates of facility access were 54 percent in Adabawere, 23 percent in Kpindi, 15 percent in Sarakawa and 36 percent in Djamdé. The total population served by the four clinics is approximately 21,457 [18]. The present analysis draws upon baseline cross-sectional data collected as part of a larger study led by HTH and the Ministry of Health to reduce child mortality over a three-year period. For the main study, a sample size of 1,500 respondents at baseline and, again, post-intervention is needed in order to detect a 75 percent reduction in the under-five childhood mortality rate between the current baseline survey and a future post-intervention survey [19]. This sample size was calculated based on the following assumptions: baseline under-five child mortality rate of 88 deaths per 1,000 live births, design effect of 1.5, total fertility rate of 4.8 [2], a 20% non-response rate, with a 95% confidence level and a 5% margin of error in a population of 21,457 [18]. For the present secondary analysis, we anticipated adequate power to examine individual and structural-level differences in the four maternal and child health care outcomes of interest with the following assumptions: 50% prevalence for all indicators (given the potential for some indicators to be high prevalence, while others may be low prevalence), 80% power, and an alpha of 0.05. We considered a difference of at least twenty-percentage points between exposure groups to be programmatically meaningful. Under these specifications, a sample size of 206 is required for adequate power. We conducted a cross-sectional household survey among women between the ages of 18 to 49 between January 19 and February 24, 2015. A probability-based cluster sampling strategy was applied in two phases. First, the Kara region was stratified by four main catchment areas around the selected health centers (Adabewere, Kpindi, Sarakawa and Djamdé). Next, using geographic information system (GIS) technology, each catchment area was divided into 15 clusters of approximately equivalent size according to population density reports from Togo’s 2010 Census to comprise the primary sampling unit [18]. Within each cluster, data collectors conducted door-to-door recruitment. Starting points were chosen at random by dropping a pen on the map to determine a starting location, and spinning the pen to determine the starting direction to head in to avoid systematic bias in the types of households visited by data collectors. Households were considered eligible if at least one adult female resided within and was present at the time of recruitment. If more than one eligible female resided in the household, the respondent was randomly selected using a KISH table [20]. Based on 2010 census data and the understanding that not all households approached would have eligible females at home during fielding, we estimated we would need to approach 50% of households in each catchment area to achieve a sample size of 1,500 respondents. During data collection, however, field staff noted that the population exceeded census estimates in the urban area of Adabawere and thus 25% of households were approached in this area. Population-based weights were calculated to adjust for non-equal probability of selection (i.e., 50% coverage in each rural site and 25% coverage in Adabewere) and to adjust for participant non-response [21]. Data collectors were local individuals with at least some post-secondary education who were also fluent in French and Kabiyé. Many of the interviewers had previously worked on the 2013 Demographic and Health Survey and/or the 2010 Togo Census. All data collectors received a detailed three-day training led by HTH researchers on the ethical procedures for conducting research with human subjects, the procedures for data collection and obtaining informed consent, as well as the sampling strategy and administration of the interview questionnaire. Data collectors recorded participant responses using pen and paper questionnaires, with close supervision by the research team and HTH staff to ensure data quality and accuracy. Survey questions related to household demographics, maternal and newborn health, family planning, and child disease and treatment were drawn from modules from the French-language versions of the 2013 Demographic and Health Surveys (DHS) and 2010 Multiple Indicator Cluster Survey (MICS) in Togo, with a few minor adaptations. Following the maternal and newborn health module, women responded to questions related to receipt of antenatal, labor and delivery and postnatal care for live births within the two years preceding the survey. Women also responded to questions on birth history from the last 10 years, children currently living with them, and child disease and treatment status (e.g., fever, cough and diarrhea). To reflect a more recent child cohort, questions related to child disease and treatment refer to a woman’s most recent birth among all children born in the past 10 years. This slightly differs from the DHS, which represents the most recent birth among all prior births. A critical intervention in reducing maternal mortality is increasing women’s timely access to emergency obstetric services in order to manage life-threatening complications that occur during labor and delivery or postnatal periods. Given challenges in measuring the receipt of quality intrapartum care from women by self-report in a survey [22–24], we assessed facility-based delivery by asking women who had a live birth in the two years preceding the survey to report where their most recent birth occurred. Facility-based delivery is a widely used indicator of health care access, routinely tracked in health initiatives such as the Millennium Development Goals (MDGs) and Sustainable Development Goals (SDGs) to measure one component of effective intervention coverage. Responses were dichotomized into facility (e.g., public or private sector health post, community health center, hospital or clinic) or home births. Women with a live birth in the past two years were also asked about whether they received a postnatal health care check by a health professional (i.e., a doctor, clinical assistant, nurse-midwife or auxiliary midwife) at any time in the first 42 days following delivery. The standard of care for postnatal care is a health check within the first two days of birth [25]. However, less than two percent of women (N = 7 cases) reported a health check within this period. Given that nearly 2 in 5 maternal deaths occur in the postnatal period [26], we defined this outcome to include postnatal health care checks by a health professional at any time in the postnatal period (i.e., within the first 42 days of delivery). Current WHO recommendations related to postnatal care do not stipulate that health checks take place within a facility setting [27]. We also present these results by the type of health professional seen. We calculated the proportion of children who: (1) ever received the third dose of the pentavalent vaccine (or DTCoq3) which protects against diphtheria, pertussis, tetanus, hepatitis B and haemophilis influenzae type B, and (2) the proportion of children in the household under age five reported to have been febrile in the two weeks preceding data collection, who received an effective antimalarial treatment (ACT, arthemeter, arthesunate or artmefloquite or quinine) within 72 hours of symptom onset. These indicators of health service coverage were chosen to reflect key initiatives in child health in Togo, such as no-cost malaria treatment for children [5,6] and child vaccination through the Expanded Immunization Program (EIP) [7]. For child vaccination, cases were restricted to the youngest child born to women in the last ten years. Cases where vaccination status was unable to be determined (14.6%) were excluded from multivariable analysis for this outcome. In instances where more than one febrile child among women was reported in the past two weeks, the first reported case was analyzed. Personal characteristics of women examined include age, highest educational attainment, marital status (i.e., currently married or living together with a man versus not currently in a union), polygamous versus monogamous marriage, prior parity and ethnic group. We considered factors that affected the logistics of accessing care to include household wealth quartile, health insurance status, facility-proximity (distance in km), and site of residence. Household wealth status was measured using household assets on a scale ranging from 0 to 11 with one point given for each of the following household items (electricity, running water inside compound, corrugated tin roof, radio, television, mobile phone, refrigerator, motorbike, bicycle, car, or gas appliance stove). We used wealth quartiles as an index of household socioeconomic status measure in subsequent analysis. Health insurance was measured as women’s self-reported coverage status. Facility proximity was assessed as the distance to the nearest health facility (either a health post, community health center, clinic or hospital) within each cluster (less than 3km, between 3-5km, or greater than 5km by distance or by road). Distance to the nearest health facility was measured using global positioning systems (GPS) technology available through Google Maps to measure a three and five-kilometer radius from each health facility. Each of the 15 clusters within the four catchment areas were categorized “as the crow flies” according to whether more than 50% of households fell within the three-kilometer radius, within the five-kilometer radius, or exceeded this distance. We also considered differences in maternal and child health utilization across the four sites of residence (catchment areas of Adabewere, Djamdé, Sarakawa or Kpindi). Urban versus rural status was assessed by household residence in Adabewere, the only urban area, as compared to the three rural sites of Djamdé, Sarakawa and Kpindi. We present bivariate and multivariable logistic regression models that assessed the above risk factors for the four outcome measures for maternal and child health service use: delivery in a health facility, a postnatal health check for the mother by a health professional, receipt of the pentavalent vaccine for children under age 5, and febrile children under age five who received antimalarial treatment within 3 days of fever onset. We accounted for clustering among observations within the same primary sampling unit and weighted results to population estimates by using the “svy” function for complex survey data in Stata 14 (StataCorp, College Station, TX, USA). Descriptive statistics (raw sample sizes and population-weighted percentages) are reported for each health outcome. Given that the sample of women who reported on maternal (women with births in the last two years) and child (women with births in the last ten years) health outcomes differed, we disaggregate participant characteristics for each subgroup. We compared women’s sociodemographic characteristics and maternal and child health care utilization using chi-squared tests for association reporting within group prevalence estimates and multivariable logistic regression reporting adjusted odds ratios. Due to limited sample sizes of distinct ethnic groups, bivariate data on ethnicity are not presented in the data tables and ethnicity was not included in the final multivariable analysis. Multivariable models were developed in an iterative process, including variables significantly associated with the outcome of interest at the bivariate level and for which there was sufficient sample size for robust analysis (>5% prevalence within the total and rural specific strata). Pairwise missing data were excluded. Final models were selected to maximize goodness of fit indices using a backwards elimination strategy to identify variables most predictive of the outcome of interest, while adjusting for complex sampling and applying population-based weights. To account for potential time trends in the likelihood of childhood vaccination among women’s youngest child born in the past ten years, we adjusted for child age in the multivariable model of childhood vaccination status. As area of residence corresponded with perfect prediction of facility-based delivery, we present multivariable findings stratified by urban and rural residence.

• mccarthy, katharine j.
• braganza, sandra
• fiori, kevin p.
• gbeleou, christophe
• kpakpo, vivien
• lopez, andrew
• schechter, jennifer
• goodwin, alicia singham
• jones, heidi e.