Advancing Survival in Nigeria: A Pre-post Evaluation of an Integrated Maternal and Neonatal Health Program

Introduction Nigeria contributes more obstetric, postpartum and neonatal deaths and stillbirths globally than any other country. The Clinton Health Access Initiative in partnership with the Nigerian Federal Ministry of Health and the state Governments of Kano, Katsina, and Kaduna implemented an integrated Maternal and Neonatal Health program from July 2014. Up to 90% women deliver at home in Northern Nigeria, where maternal mortality ratio and neonatal mortality rates (MMR and NMR) are high and severe challenges to improving survival exist. Methods Community-based leaders (“key informants”) reported monthly vital events. Pre-post comparisons of later (months 16–18) with conservative baseline (months 7–9) rates were used to assess change in MMR, NMR, perinatal mortality (PMR) and stillbirth. Two-tailed cross-tabulations and unadjusted and adjusted logistic regression analyses were conducted. Results Data on 147,455 births (144,641 livebirths and 4275 stillbirths) were analyzed. At endline (months 16–18), MMR declined 37% (OR 0.629, 95% CI 0.490–0.806, p ≤ 0.0003) vs. baseline 440/100,000 births (months 7–9). NMR declined 43% (OR 0.574, 95% CI 0.503–0.655, p < 0.0001 vs. baseline 15.2/1000 livebirths. Stillbirth rates declined 15% (OR 0.850, 95% CI 0.768–0.941, p = 0.0018) vs. baseline 21.1/1000 births. PMR declined 27% (OR 0.733, 95% CI 0.676–0.795, p < 0.0001) vs. baseline 36.0/1000 births. Adjusted results were similar. Discussion The findings are similar to the Cochrane Review effects of community-based interventions and indicate large survival improvements compared to much slower global and flat national trends. Key informant data have limitations, however, their limitations would have little effect on the results magnitude or significance. A pre-post design was used to evaluate change in mortality by comparing later with baseline incidence rates or ratios. In November, 2014, the MNH program revitalized, upgraded and expanded a dormant Community Based Health Management Information System (CBHMIS) in which key informants (KIs) reported monthly births, women’s and neonatal deaths and stillbirths. The KIs were mostly respected traditional community leaders with basic primary or higher education. In the case of death or disability, the new community leader became the KI. All traditional leaders had a scribe with basic primary or higher education who could read and write English or Hausa. As with the interventions, the CBHMIS was established to promote local ownership. A monitoring and evaluation (M&E) officer for each state, and various Local Engagement Consultants (LECs), who were communities/LGA members, mostly with previous similar data collection experience and CBHMIS familiarity, were engaged and trained by CHAI staff to oversee data collection. In November, 2014, the M&E officers led their state’s initial 2 days LGA level training sessions of the district heads, KIs and their scribes on the use of the program tools, forms, data templates and on the reporting structure. The training included information on MMR, NMR, SBR and their major causes and means of prevention. As a result of a CHAI site data audit in October 2015 that identified data irregularities and under-reporting, LGA level review meetings were held monthly through July 2015 and quarterly thereafter. Each M&E officer created a report of trends and gaps shared by the LECs at the meetings to resolve data quality issues. LECs also reviewed survival trends to identify and resolve program implementation limitations and share successes achieved to continue motivated and accurate data collection. The meetings produced a dramatic rise in reported events and stable reporting from May 2015 onwards. The M&E officers provided ongoing supervision to the LECs on a daily to weekly basis, depending on the geographic span of the LEC coverage area. From November 2014, over 1,500 KIs reported vital events for settlements (groups of 5–10 households) in the program LGAs. Data were collected with pen on paper, recording a ‘1’ as a stillbirth, livebirth or neonatal death and the women’s age (years) at delivery or at neonatal death (days) in Kano and Katsina. Kaduna reported gender only for deaths. Because gender influences survival (Sawyer 2012), the team retroactively classified missing gender using neonates’ (livebirths who survived or died ≤ 28 days) first names. Facility-based births and deaths were reported by families to the KIs. To maintain data collection simplicity, multiple gestation (which increases mortality risk) and individual exposure to specific interventions were not recorded, thus their effects could not be directly assessed. State was systematically reported. Data were collected at the lowest(settlement)level and reported up to the traditional ward, then the village, the District LGA (where data entry in Excel occurred), then to the LEC, state M&E officer, and finally to the CHAI national M&E officer where the data files were aggregated. Hard copy forms for out-of-range or inconsistent data reported between January 2015 and June 2016 were reviewed to correct electronic records. The Excel spreadsheets were securely transferred and imported into SPSS for Windows version 23 for analysis. The primary outcomes were women’s and neonatal mortality and stillbirth; perinatal mortality was a secondary outcome (see definitions in Table 1). Evaluation outcomes The evaluation (report to be posted on http://www.clintonhealthaccess.org along with an independent qualitative evaluation conducted by KPMG) analyzed CBHMIS vital events data and reviewed information from CHAI staff and program documents. The KIs were becoming familiar with data reporting between January and May 2015. Half as many births were reported in quarter 1 (Q1) and two-thirds in Q2 compared with Q3–Q5 (Table 2). Further training and monitoring stabilized the number of births reported from June 2015 onwards. As under-estimation of births would spuriously inflate mortality rates and over-estimate program effect, the evaluation conservatively used quarter 3 data (July–September 2015), when all the interventions were fully rolled out, as the baseline comparison group. Quarter 6 was the endline. More births were reported in Q6, with an increase in reporting villages and their population, but the similarity of Q5 and Q6 results (described below) suggests little (e.g., parallel if any) bias. Numbers of Births (Live and Stillbirths) and LGAs by State and Quarter Cross-tabulations with Chi square statistics and unadjusted and adjusted logistic regression analyses were conducted for all outcomes. Where inconsistent or out-of-range data correction was not possible, missing values were assigned for analysis. The regression Odds Ratios (OR), the 95% confidence intervals (CI), and two-tailed significance values (p) are presented. Neonatal mortality regression models were adjusted for gender and to adjust geographic differences in socio-economic status and program implementation state. All remaining outcomes were adjusted for state only. For adjusted analyses, missing gender was re-coded to the mean value. Data were analyzed in quarterly periods to minimize variability in event reporting by settlement. All 30 program LGAs contributed to each quarter. Variability in the number of monthly reporting settlements might influence the effects, however data were not adjusted for clustering given the large number of settlements and the small number of quarterly births and deaths per settlement, which nearly represents individual-level reporting. In the absence of extreme reporting bias, with the very large numbers reported upon, and large and highly significant results, adjustment for such clustering would have little effect on the findings magnitude or significance. As the vital events system reported all deaths, MMR and NMR analyses assume survival through 42 days postpartum and 28 days of life, respectively. Assuming neonates and women who did not die survived through 28 and 42 days, respectively, and excluding the few infant deaths beyond 28 days or women’s deaths from antepartum hemorrhage that did not result in births may minimally under-represent the denominators but with very large numbers of births, should have little effect on the results magnitude or significance. The three women aged 14 and the five aged 45–50 years were included in analysis as plausible obstetric and postpartum deaths (Fig. 3). No deliveries or deaths were reported in girls less than 14 years old. Only one women’s death > 50 years was reported; she was 60 years and excluded from analysis. Women (n = 12) and neonates (n = 29) with missing age were excluded from analysis. The timing of women’s deaths was not reported. Flow diagram and process used to merge and analyze event registry files Women’s and infants’ events were almost all reported as separate records, without unique linking identifiers. To attain the most accurate MMR denominator, the best woman-infant links were accomplished using geographic, event date and name fields. With imperfect matching, all (still and live) births constitute the MMR denominator; all women’s deaths are the numerator.