Community health workers impact on maternal and child health outcomes in rural South Africa – a non-randomized two-group comparison study

Background: Home visits by paraprofessional community health workers (CHWs) has been shown to improve maternal and child health outcomes in research studies in many countries. Yet, when these are scaled or replicated, efficacy disappears. An effective CHW home visiting program in peri-urban Cape Town found maternal and child health benefits over the 5 years point but this study examines if these benefits occur in deeply rural communities. Methods: A non-randomized, two-group comparison study evaluated the impact of CHW in the rural Eastern Cape from August 2014 to May 2017, with 1310 mother-infant pairs recruited in pregnancy and 89% were reassessed at 6 months post-birth. Results: Home visiting had limited, but important effects on child health, maternal wellbeing and health behaviors. Mothers reported fewer depressive symptoms, attended more antenatal visits and had better baby-feeding practices. Intervention mothers were significantly more likely to exclusively breastfeed for 6 months (OR: 1.8; 95% CI: 1.1, 2.9), had lower odds of mixing formula with baby porridge (regarded as detrimental) (OR: 0.4; 95% CI: 0.2, 0.8) and were less likely to consult traditional healers. Mothers living with HIV were more adherent with co-trimoxazole prophylaxis (p < 0.01). Intervention-group children were significantly less likely to be wasted (OR: 0.5; 95% CI 0.3-0.9) and had significantly fewer symptoms of common childhood illnesses in the preceding two weeks (OR: 0.8; 95% CI: 0.7,0.9). Conclusion: The impact of CHWs in a rural area was less pronounced than in peri-urban areas. CHWs are likely to need enhanced support and supervision in the challenging rural context. The study was initiated in the rural Eastern Cape, in the catchment area of four clinics referring to Zithulele District Hospital, situated in the King Sabata Dalidyebo Subdistrict of the OR Tambo district, one of the poorest municipalities in South Africa [25]. The catchment population is approximately 40,000, with a density of (124 people/km2) [26] and the topography is hilly with deep river valleys and gorges. This, combined with poor road infrastructure, makes access to healthcare services challenging for much of the population. This study was an early Phase II non-randomized two-group comparison study evaluating the Philani Home-Visiting Model in the rural Eastern Cape of South Africa, called the Zithulele Mothers-to-be Assessment (ZiMBA). Recruitment started in August 2014 and ended in May 2017. Mothers were assigned based on the location of clinic: (1) Mapuzi and Tshezi, an area with home visiting by CHWs, which we will refer to as the “intervention” group or (2) Wilo and Luthubeni, areas with standard clinical care and no CHWs, as the “comparison” group. Philani CHWs focusing on maternal and child health, had been assigned to conduct home visits within designated geographical areas (i.e. intervention areas a year previously). The comparison area was matched with the intervention area based on clinic coverage, size, and distance from the district hospital by tar road, and a similar number of expected pregnancies, which was estimated from data of the Zithulele Births Follow-up Study (ZiBFUS) [27]. We ensured that a buffer zone existed between the two areas to avoid “cross-contamination” of intervention effects. See map – Fig. 1. Control and intervention areas, Eastern Cape, South Africa Philani Maternal, Child Health and Nutrition Trust is a Non-Governmental Organization (NGO) that has been operating in peri-urban townships of Cape Town since 1979. Philani Mentor Mothers (Community Health Workers) are positive-deviant women whose children are thriving despite living in poverty [9]. These mothers are recruited to work in the areas where they live and are primarily trained to address maternal and child health issues in their communities at the level of the household. They perform regular house-to-house visits in their designated neighborhoods and identify both pregnant mothers and malnourished children who are then offered to entry into the program and followed up with regular home visits. The Community Health Workers (CHWs) build strong relationships with their clients and encourage mothers to attend antenatal care, immunize their children, weigh them regularly and breastfeed exclusively for 6 months; they advise mothers about health-care regimes, mental-health issues, antenatal care, optimal infant feeding, accessing grants and accessing prevention of mother-to-child HIV transmission (PMTCT) [22]. CHW are also trained to identify and refer household members with possible TB and to support people living with HIV. A critical aspect of the intervention is the peer support provided to mothers who are struggling by women from their own community, who have faced similar challenges. In 2010, the program was expanded to a deeply rural area around Zithulele District Hospital, and by 2013 to the Coffee Bay area - both in the OR Tambo district of the Eastern Cape. The CHW program was integrated into primary health-care services in the King Sabata Dalindyebo sub-district authority, and provides several advantages over other CHW programs, including a strong emphasis on care in households and the community rather than at primary care clinics, a total 6-week standardized training both in the classroom and the field, meticulous record-keeping of patient follow-ups, daily in-the-field support and supervision visits by supervisors ensuring accountability and the recruitment of positive-deviant mothers [9]. The Philani Mentor Mother program is not active in this area. Mothers and children in the comparison group have access to free primary health care at clinics and free maternity and child healthcare (up to 6 years) at government hospitals, which includes HIV care. All pregnant women attending antenatal care during the study period and who agreed to participate were recruited by trained interviewers stationed at each of the four clinics listed above, during the recruitment period. Women living in the areas covered by the four clinics who presented to Zithulele Hospital around the time of delivery, but who were not previously recruited, were also recruited into the study. Participants who were deaf, mute or with significant psychiatric issues at initial contact were excluded. All women signed an informed consent form. In addition, those under the age of 18 had a parent or guardian sign consent. Figure 2 depicts the participant flow chart. Of the 1490 mothers approached at antenatal clinics and the hospital, 180 women were not eligible due: no pregnancy (n = 21), miscarriages and intrauterine deaths (n = 53), early birth deaths (n = 13), refusal at baseline interview or other reasons (n = 11), and not completing the baseline interview (n = 82). A total of 1310 women were recruited into the study, stratified by location of clinic: (1) Mapuzi and Thezi (Intervention group; n = 636); or (2) Wilo and Luthubeni (Comparison group; n = 674). Baseline assessments were performed soon after the birth of the baby (median, 2 days) and at 6-months post-birth (median, 180 days). In the analysis, babies who died by 6-months (n = 21), twins or triplets (n = 17), and HIV-seropositive children (n = 3) were excluded. There were two maternal deaths by 6 months. Although child deaths did affect the overall study sample, maternal deaths did not, as long as there was a proxy for the mother’s interview (i.e. caregiver like a grandmother) who could complete the assessment. Participant flow chart - by intervention (Thezi-Mapuzi) and comparison (Luthubeni-Wilo) groups Background characteristics collected included maternal age, highest education level achieved (years), current employment or learner status, and the presence of a live-in partner or husband. Economic resources were identified as monthly income (> 2000 South African Rand [ZAR]), receipt of the child support grant, access to electricity and safe water, and the number of adults and children that live in the household. Mothers were asked if they ever used alcohol before pregnancy, and if they continued to use alcohol after discovering they were pregnant. Maternal antenatal cards detailed the medical care the mother received before pregnancy and the number of antenatal visits she attended. We collected previous pregnancy history including poor outcomes and the number of live births. Depressive symptoms were measured using the Edinburgh Postnatal Depression Scale (EPDS). The EPDS is a 10-item scale with four Likert-type responses for each item, with mothers self-reports indicating possible depressed mood with scores > 13 and probable clinical depression ≥18 that has been extensively used in South Africa [24, 28–30] and has been found to “reliably and validly measure perinatal depression symptom severity or screen for probable postnatal depression in African countries” (29, p1). HIV testing during pregnancy was confirmed either by self-report or by the mother’s antenatal card at baseline. Mother’s HIV status, receipt of ARVs before and during pregnancy, and disclosure of HIV status were also recorded. At baseline, data collectors record if the child lives with the father, and the father’s HIV status and substance use. current breastfeeding at the birth interview was recorded and how soon the infant was put to the breast. At the 6-months assessment, mothers self-reported whether they were exclusively breastfeeding for three and 6 months. If a mother was not exclusively breastfeeding, data collectors asked whether and when the mother started using formula, and whether she was mixing formula with baby porridge. Birthplace of the child was either at the hospital, clinic, or on the way to the hospital/clinic versus at home. Low birth-weight was defined as any infant that weighed less than 2500 g at the time of birth. Finally, Apgar scores were recorded at birth at 5 min, ranging from 1 to 10 using information from the government Road to Health Card (RtHC), which is a health record kept by the mother. Trained and certified interviewers weighed children (kilograms) and measured their length (centimeters) using a measuring mat at birth and 6-months. Birth weight (g/kg) was also recorded from the RtHC. Infant anthropometric data was then converted to z-scores based on the World Health Organization’s (W.H.O) age-adjusted norms [31]. A z-score below − 2 standard deviation (SD) was considered a serious growth deficit: <− 2 for height-for-age z-scores (HAZ) was considered stunted, < − 2 for weight-for-age z scores (WAZ) was considered underweight, and < − 2 for weight-for-height z scores (WHZ) were considered wasted. Growth scores of > 5 SD or < − 5 SD may have represented measurement errors and were replaced by 5 or − 5 in order to reduce the effect of outliers in our analyses (i.e. HAZ [n = 21, 0.6%], WAZ [n = 20, 0.6%], and WHZ [n = 123, 3.5%]). Data collectors asked the mothers what immunizations the child had received up to and including 6-months (plus Vitamin A) and confirmed these answers with the child’s RtHC. Mothers were asked at the 6-months assessment if the child had experienced any child illness in the past 2 weeks including diarrhea, vomiting, cough, shortness of breath, fever, and/or runny nose. Mothers were asked if the mother has taken the child to the hospital, clinic, traditional healer, private doctor, or private pharmacy in the past 3 months. WHO gross motor developmental milestones were measured at 6-months. The presence or absence of the following motor skills were assessed: (1) sitting without support; (2) standing with support; (3) hands-and-knees crawling; (4) walking with assistance; and (5) standing alone [32]. Depending on the child’s age (age unit: months), a child was determined to be either on target or behind target based on the assessment of the trained interviewer. Among the mothers living with HIV (MLH), the PMTCT cascade focused on six tasks: (1) receipt of antiretroviral (ARVs) before the birth of the baby; (2) nevirapine (NVP) administered to the infant after delivery; (3) continuation of Anti-retroviral Therapy (ART) for mother post-delivery; (4) infant received HIV PCR testing; (5) maintenance of a single feeding method (i.e. exclusive breastfeeding for 3 months); and (6) the infant received co-trimoxazole by the 6-months assessment. The key baseline demographics and maternal characteristics were compared between intervention and comparison groups, using the Student’s t-test or Mann-Whitney U test for continuous variables, and the chi-square (χ2) test or Fisher’s exact test for categorical variables. We also examined participants who were lost to follow-up (11%; n = 147), by comparing their baseline characteristics to mothers who were retained at 6-months assessment (n = 1142/1310) stratified by intervention and comparison groups. Our primary analysis compared the distribution of maternal and child outcomes by intervention and comparison groups at 6-months using t-test and χ2 test for continuous and categorical outcomes, respectively. Among MLH, we compared additional tasks in the PMTCT cascade between intervention and comparison groups using χ2 test. Longitudinal mixed-effects models were used to assess the intervention effect for maternal and child outcomes that were measured over time. In particular, we applied linear mixed-effects regression models with restricted maximum likelihood estimation for continuous outcomes, and mixed-effect logistic regression models for binary outcomes. The mixed-effects models for maternal and child outcomes assessed over time, included fixed-effects for maternal HIV status, access to electricity and safe water, having total grant income above 2000 South African Rand (ZAR), intervention, time, and interaction between intervention and time. The time variable used in the longitudinal models for maternal outcomes corresponds to time-points (baseline and 6-months) where mothers were assessed in the study, and for child outcomes refer to the actual age of children (age unit: months). For child outcomes assessed only at 6-months, we omitted the child random-effects. Multivariate negative binomial regression models were fitted for count outcomes, where the likelihood ratio test was adapted to assess whether the negative binomial model was more appropriate than the Poisson model. For child outcomes that were assessed over time from baseline to 6-months, the effect modification was assessed by fitting interaction terms between maternal HIV status, time, and intervention in the longitudinal models. Similarly, for those child outcomes that were only measured at 6-months, this was assessed by including an interaction term between maternal HIV status and intervention in models. Further examination was carried out to assess whether maternal HIV status modifies the effect of intervention on outcomes. All mixed-effects models accounted for repeated measures for participants by including random intercepts. Random slope for time was also assessed, and if needed, included in the models in addition to fixed-effects and random intercepts. All analyses were conducted using Stata SE software Version 15.