Effects of a community health worker delivered intervention on maternal depressive symptoms in rural Tanzania

Maternal depression affects one in four women in sub-Saharan Africa, yet evidence on effective and scalable interventions is limited. Our objective was to evaluate the effect of a community health worker (CHW) delivered home visit responsive stimulation, health and nutrition intervention, and conditional cash transfers (CCTs) for antenatal care and child growth monitoring attendance on maternal depressive symptoms. We conducted a cluster-randomized controlled trial in 12 villages in rural Ifakara, Tanzania (September 2017 to May 2019). Study villages were randomly assigned to one of three arms: (1) CHW, (2) CHW + CCT and (3) Control. Pregnant women and mothers with a child <12 months were enrolled. Maternal depressive symptoms were assessed using a Tanzanian-Adapted version of the Hopkins Symptoms Checklist-25 (HSCL-25) after 18 months of follow-up. We used linear mixed-effects models to estimate intervention effects on HSCL-25 scores. Results showed that the CHW intervention significantly reduced HSCL-25 scores as compared with control [unadjusted mean difference (MD)-0.31, 95% confidence interval (CI)-0.47,-0.15]. The CHW + CCT intervention also appeared to lower HSCL-25 scores (MD-0.17, 95% CI-0.33,-0.01), but results were not statistically significant. Our findings showed that a low-intensity CHW-delivered home visit responsive stimulation, health and nutrition intervention, which did not explicitly aim to improve mental health, reduced maternal depressive symptoms, though the precise mechanisms of action remain unknown. CCTs for antenatal care and child growth monitoring appeared to provide limited to no additional benefit. Community-based integrated interventions that broadly consider maternal and child health, development and well-being have the potential to promote maternal mental health in rural Tanzania and similar settings.

Full details on the study design have been previously published (Sudfeld et al., 2019) . Briefly, we conducted a longitudinal cluster-randomized controlled trial (cRCT) where 12 villages stratified by urban–rural location were randomly assigned to one of three intervention arms: (1) CHW, (2) CHW + CCT and (3) Control. The CHW and CHW + CCT arms received the same integrated responsive stimulation, health and nutrition intervention. The CHW + CCT arm also received CCTs. The control arm did not receive any intervention and had access to the existing clinic-based healthcare services. In each village, enrolment in the trial continued until all pregnant women and women with a child <12 months of age were enrolled or until 50 participants were enrolled, whichever was reached first. Children with severe physical or mental impairment were excluded. The baseline survey was conducted in September to October 2017 prior to the start of the intervention; the midline survey was conducted in June to August 2018 after 9 months of intervention implementation; and the endline survey was conducted in January to May 2019 after 18 months at the completion of the intervention. Trained fieldworkers explained the intervention and associated study and obtained written informed consent from participants in their primary language. Ethical approval was received by the institutional review boards of the authors' institutions. The intervention has been described in detail elsewhere (Sudfeld et al., 2019) . Briefly, a CHW-delivered intervention was examined alone and in combination with CCTs. Female CHWs living in the study villages’ catchment areas delivered the intervention. The intervention-specific responsive stimulation component was a Tanzanian and Swahili adapted version of the UNICEF and WHO Care for Child Development package (World Health Organization and UNICEF, 2012). This package promoted caregivers’ sensitivity and responsiveness using developmentally appropriate stimulation activities (e.g. play and communication). It consisted of essential early child development knowledge, age-appropriate play and communication activities, toy making, parenting and problem-solving counselling, and promotion of caregiver responsiveness and sensitivity. The CHWs received a 1-week classroom-based intervention-specific training prior to the start of the intervention. This training covered both theoretical and practical aspects of early child development, age-appropriate play and communication activities, counselling of caregivers, problem-solving, and making of toys and other play materials. A 3-day refresher training was conducted after 9 months of implementation, half-way through the intervention. The health and nutrition components covered optimal antenatal, postnatal and newborn health and nutrition practices, care and danger sign identification, community-case management of childhood illness, and emergency and routine health facility referral. Training was received through a 1-year comprehensive government-certified training, which included both health and non-health-related topics. Additional health-related topics included infection and disease prevention and control, community-based reproductive health services and health promotion, home-based care, and health facility and community disease management. Non-health-related topics covered fundamentals of social work, basic life support skills, communication and customer service, computer literacy, citizenship and gender, management information systems, and fundamentals of entrepreneurship and life skills. CHWs conducted individualized, one-on-one home visits every 4–6 weeks to participating women during which they demonstrated age-appropriate play and communication activities for the child, observed the women practicing the newly learned activities, and provided support applying the new activities. In addition, CHWs advised women on optimal health and nutrition practices for themselves and their young child, counselled them on parenting and problem-solving strategies, and made health facility referrals for emergency and routine conditions. Finally, CHWs provided counsel and advice with respect to other topics covered in the government-based training. The CCT component consisted of monthly cash transfers of 10 000 Tanzanian shillings ($4.30) for ANC visits, with reimbursement for up to four ANC visits, or 5000 Tanzanian shillings ($2.20) for child growth monitoring and health visits, with reimbursement for up to one visit per month. Cash transfers were small relative to the $1.90 average daily per person income for smallholder farmers in Tanzania (Rapsomanikis, 2015). During the home visits, CHWs reimbursed pregnant women and women with young children after inspecting their health card or the child’s health card, respectively, to ensure the conditions for receiving the cash transfer were met. The intervention did not include a psychological component such as cognitive behavioural or behavioural activation therapy and did not explicitly address women’s mental health. One field co-ordinator supervised the CHWs throughout the intervention. Supervision included one-on-one biweekly meetings with each CHW, a monthly meeting with all CHWs, as well as monthly home visit spot-checks where the field co-ordinator accompanied CHWs during home visits. Further details on supervision are available elsewhere (Sudfeld et al., 2019) . Quantitative questionnaires collected data on women’s socioeconomic and demographic characteristics, and mental health, among other data related to the primary outcomes of the trial (Sudfeld et al., 2019) . Mental health was assessed using the Hopkins Symptoms Checklist-25 (HSCL-25), which captured depression (15 questions) and anxiety (10 questions) symptoms. All questions were asked by a male fieldworker in private areas, out of sight or earshot from other household or community members. Respondents rated how much 25 anxiety- and depression-related experiences and feelings bothered or distressed them in the past 2 weeks (1 = Not at all, 2 = A little, 3 = Quite a bit, 4 = Extremely). ‘Do not know’ responses were recoded as missing. A continuous HSCL-25 score (range 1–4) was calculated as the mean of the 25 items. Higher HSCL-25 score indicated worse depressive symptoms. Separate depressive symptoms and anxiety sub-scores were also calculated as the mean of the 15 and 10 items, respectively. The HSCL-25 was previously adapted and validated for diagnosis of symptoms consistent with major depressive disorder in HIV-positive pregnant Tanzanian women (Kaaya et al., 2002). In our sample, HSCL-25 revealed high internal consistency (α = 0.88) both among pregnant women and women with young children (α = 0.84 and α = 0.90, respectively). Perceived social support was assessed using the Duke University-University of North Carolina Functional Social Support Questionnaire (Broadhead et al., 1988). A household wealth index was constructed using principal components analysis of 11 items assessing asset ownership and housing quality. Intent-to-treat intervention effects of the CHW and CHW + CCT interventions compared with control were estimated using a linear mixed-effects model. Each cluster (i.e. each village) was modelled with a different starting point and trajectory over time. Within clusters, individuals were allowed different starting points, but similar trajectories over time. Model comparison indicated that differences in individual trajectories were statistically indistinguishable. Therefore, we favoured the more parsimonious model. All models controlled for baseline values, and therefore minimized the risk of regression to the mean. Per protocol, we also estimated the effect of the pooled CHW and CHW + CCT intervention arms compared with control, as there was no indication of additional benefit provided by the CCTs. We calculated unadjusted mean differences (MDs) and 95% confidence intervals (CIs). As a measure of effect size, we calculated standardized mean differences as the unadjusted MD divided by the pooled standard deviation (SD). To assess the variability of intervention effects at the cluster level, individual HSCL-25 scores were predicted from the fitted model, and cluster level predicted HSCL-25 scores were calculated as the mean of the individual model-predicted HSCL-25 scores in each cluster. All statistical tests were two-sided. Critical values were drawn from a t-distribution to account for the small number of clusters (Cameron et al., 2008). Results were considered statistically significant at P < 0.05. Descriptive analyses were conducted in Stata Version 15 (StataCorp, 2017) and model fitting was conducted in R Version 3.6.1 (R Development Core Team, 2017). The trial’s sample size was based on power calculations to detect meaningful changes in child development and linear growth, the co-primary outcomes (Sudfeld et al., 2019) . We conducted post-hoc power calculations to estimate the minimum detectable effect for the outcomes presented here. Using the same type-I error α = 0.05 and 80% power, and the sample coefficient of variation (0.05) and intra-cluster correlation (0.19 for HSCL-25, 0.17 for the depressive symptomsn sub-score and 0.15 for the anxiety sub-score), the available sample allowed us to detect effects of 0.32 SD in HSCL-25 scores, 0.39 SD in depressive symptoms sub-scores and 0.22 SD in anxiety sub-scores. As a sensitivity analysis, we examined the potential of baseline imbalance between randomized clusters to bias estimates by including multivariate adjustment for baseline covariates. We adjusted for the following a priori selected baseline covariates: woman’s age, education (whether she had completed secondary education or not), parity (whether she was multiparous or nulliparous), marital status (whether she was married/cohabitating or not), pregnancy status at baseline (pregnant or with a child <12 months of age), woman’s perceived social support score, household wealth index and child age. In line with CONSORT guidelines, missing data on baseline covariates were imputed (Moher et al., 2010) using fully conditional specifications (Li and Stuart, 2019) with 20 imputations (Graham et al., 2007). In addition, we assessed possible bias due to missing HSCL-25 data due to incomplete midline and endline interviews. Baseline characteristics of women without missing HSCL-25 data at midline or endline were compared with those of women with missing HSCL-25 data using t-tests. First, we compared women without missing HSCL-25 data at baseline and midline to women with missing HSCL-25 data at midline. Second, we compared women without missing HSCL-25 data at baseline and endline to women with missing HSCL-25 data at endline. Differences were considered statistically significant at P < 0.05. Inverse probability weights were derived from a logistic model predicting the probability of missing HSCL-25 data (i.e. probability of attrition) (Wooldridge, 2010) and applied to the effect estimates. Furthermore, we examined potential modification of the effect of the CHW and CHW + CCT interventions on HSCL-25 scores, and depressive symptoms and anxiety sub-scores by a priori defined factors: pregnancy status at baseline, maternal education, marital status, parity, symptoms consistent with depression at baseline, and number of CHW visits received (defined as having received at least 10 home visits or 90% of planned home visits, based on CHW report). To define symptoms consistent with major depressive disorder (which we refer to as depression for brevity), we used the Tanzanian cut-off of a mean score ≥1.06 on eight HSCL questions (Kaaya et al., 2002) and the standard cut-off of a mean score ≥1.75 on HSCL-25 (Derogatis et al., 1974). Interactions were considered statistically significant at P < 0.05.

• bliznashka, lilia
• yousafzai, aisha k.
• asheri, geofrey
• masanja, honorati
• sudfeld, christopher r.