Community perspectives of a 3-delays model intervention: A qualitative evaluation of saving mothers, giving life in Zambia

Background: Saving Mothers, Giving Life (SMGL), a health systems strengthening approach based on the 3-delays model, aimed to reduce maternal and perinatal mortality in 6 districts in Zambia between 2012 and 2017. By 2016, the maternal mortality ratio in SMGL-supported districts declined by 41% compared to its level at the beginning of SMGL-from 480 to 284 deaths per 100,000 live births. The 10.5% annual reduction between the baseline and 2016 was about 4.5 times higher than the annual reduction rate for sub-Saharan Africa and about 2.6 times higher than the annual reduction estimated for Zambia as a whole. Objectives: While outcome measures demonstrate reductions in maternal and perinatal mortality, this qualitative endline evaluation assessed community perceptions of the SMGL intervention package, including (1) messaging about use of maternal health services, (2) access to maternal health services, and (3) quality improvement of maternal health services. Methods: We used purposive sampling to conduct semistructured in-depth interviews with women who delivered at home (n=20), women who delivered in health facilities (n=20), community leaders (n=8), clinicians (n=15), and public health stakeholders (n=15). We also conducted 12 focus group discussions with a total of 93 men and women from the community and Safe Motherhood Action Group members. Data were coded and analyzed using NVivo version 10. Results: Delay 1: Participants were receptive to SMGL’s messages related to early antenatal care, health facility-based deliveries, and involving male partners in pregnancy and childbirth. However, top-down pressure to increase health facility deliveries led to unintended consequences, such as community-imposed penalty fees for home deliveries. Delay 2: Community members perceived some improvements, such as refurbished maternity waiting homes and dedicated maternity ambulances, but many still had difficulty reaching the health facilities in time to deliver. Delay 3: SMGL’s clinician trainings were considered a strength, but the increased demand for health facility deliveries led to human resource challenges, which affected perceived quality of care. Conclusion and Lessons Learned: While SMGL’s health systems strengthening approach aimed to reduce challenges related to the 3 delays, participants still reported significant barriers accessing maternal and newborn health care. More research is needed to understand the necessary intervention package to affect system-wide change. To explore the views of the community on the SMGL initiative, the study team used qualitative methods to gather insights from the community and the public health stakeholders who interacted with the SMGL program during the implementation period. The qualitative study was conducted in July 2016 during the fourth year of implementation. We purposively sampled a total of 171 individuals from communities in Mansa, Chembe, Lundazi, Nyimba, Kalomo, and Zimba. Of those sampled, we conducted in-depth interviews (IDIs) with 78 individuals representing women who delivered at home (n=20), women who delivered at a health facility (n=20), clinicians (n=15), community leaders (n=8), and public health stakeholders (n=15). We also purposively sampled 93 participants to participate in 12 focus group discussions (FGDs), with an average of 7 people per focus group, representing men (n=29; 4 FGDs) and women (n=33; 4 FGDs) from the communities and Safe Motherhood Action Group (SMAG) members (n=31; 4 FGDs). The SMAG members are government-established community health workers. We used both IDIs and FGDs to explore individual perspectives as well as group dynamics within a community in relation to understanding and uptake of health promotion messages and decisions on how, where, and when to seek and access MNH services. Semi-structured interview guides were used for both IDIs and FGDs (for focus group discussion guides, see Supplements 1–3; for key informant interview guides, see Supplements 4–7). The sample size was established to reach thematic saturation, which occurred when no new themes emerged from interviews. IDIs and FGDs lasted between 1 to 2 hours. The qualitative study sampled 171 individuals from 6 districts implementing the SMGL program in Zambia. The IDIs and FGDs were conducted in local languages—Cibemba, Cinyanja, and Citonga—and administered in-person by a trained qualitative research assistant who spoke the language of the assigned region. While a field guide was used to focus the interviews on research aims, participants directed the course of the conversation. All interviews were digitally recorded, and field notes were taken to supplement the transcriptions during analysis. All interviews were transcribed into English by trained research assistants and loaded onto a secure drive for review and quality checks. Written informed consent was obtained for each interview and FGD. To verify data quality, data were reviewed by 2 analysts during data collection, transcription, and data entry. Three levels of review were carried out: the first review was done immediately after each interview to ensure completeness of the interview; the second review ensured all data on the audio recordings were captured; and the third review was completed after transcription to ensure that translations preserved the original meaning. Data validity was achieved through triangulation of different data sources to cross-check for completeness of information. Transcribed interviews were imported into NVivo version 10 qualitative software (QSR International, Burlington, MA, USA) to facilitate the coding process. Deductive coding was employed by coders. Since SMGL used the 3-delays model as its underlying program theory, an initial code book was developed in which parent codes for each of the 3 delays were created, and child codes representing SMGL’s key interventions were organized under their respective parent code. The primary coder used this code book to group data by SMGL intervention and delay. A second coder reviewed all transcripts and noted disagreements, which were resolved by group consensus. Memo-writing was also used throughout the data analysis process to explore emerging themes. Additional codes were added as new themes emerged. The study team met frequently to discuss emerging themes and to consolidate and update the code book. After the initial analysis was completed, a third researcher reviewed the data, ensuring intercoder reliability. Ethical approval was granted by the ERES Converge Institutional Review Board (Ref. No. 2011-Oct-007).