Data were collected through a cluster-randomized stepped-wedge trial to evaluate the impact of an intervention that introduced PPIUD services in six tertiary health facilities in Tanzania. The trial was registered with clinicaltrials.gov (NCT02718222), and the full study protocol has been published elsewhere [13]. The study also received ethical approval from the National Institute of Medical Research (NIMR) in Tanzania (protocol number: NIMR/HQ/R.8a/Vol.IX/2006). The study received a human subjects exemption from the institutional review board at Harvard University (protocol number IRB15–1605) as only de-identified data were received by the Harvard evaluation team.
Study design and deviations from the stepped-wedge protocol
For this cluster-randomized stepped-wedge trial, six large, tertiary care facilities were selected by AGOTA, the implementing agency for the intervention, to provide coverage of PPIUD services for different regions of Tanzania. The stepped-wedge design was selected to measure intervention effectiveness because it is characterized by staggered intervention implementation in all study facilities, which ensured that all women delivering in study facilities could potentially benefit from the intervention. The evaluation team matched facilities in pairs based on annual delivery caseload, and within each pair, one facility was randomly assigned to Group 1 (early intervention implementation) and the other to Group 2 (late intervention implementation). The matched pair group assignments were as follows: Dodoma General Hospital in Dodoma (Group 1) and Mt. Meru Hospital in Arusha (Group 2), Muhimbili National Hospital in Dar es Salaam (Group 1) and Sekou-Toure Regional Referral Hospital in Mwanza (Group 2), and Mbeya Zonal Referral Hospital in Mbeya (Group 1) and Tumbi-Piwani Regional Referral Hospital in Kibaha (Group 2).
After randomization, there were two key deviations from the stepped-wedge protocol. First, before data collection started, the evaluation team decided to drop Sekou-Toure Regional Referral Hospital from the evaluation because the hospital served as a family planning model facility for the country and had an ongoing PPIUD intervention, which would make it difficult to isolate the effect of the newly implemented FIGO/AGOTA intervention. Data for the evaluation were only collected in the remaining five hospitals. Second, significant delays in intervention implementation in the Group 2 hospitals (Tumbi-Pwani Regional Referral Hospital and Mt. Meru Hospital) led to insufficient data collected after intervention implementation began. Group 2 hospitals were scheduled to start intervention implementation on 15th September 2016, but implementation was delayed until 17th November 2016, 1 month before the end of data collection, providing data for only 1 month rather than the planned 3 months. As a result, we have dropped the intervention period for the Group 2 hospitals and will consider the Group 2 hospitals as control facilities that are only observed in a state where they do not receive the intervention even as the Group 1 hospitals receive the intervention. This set-up of the data allows us to conduct the analysis as a treatment/control study of a cluster-randomized trial using a difference-in-difference approach. The difference-in-difference approach compares the change in an outcome that is observed in the Group 1 (treatment) hospitals between the pre- and post-intervention periods relative to the change in the outcome that is observed in the Group 2 (control) hospitals over the same period of time. The key identifying assumption of this analytic approach, referred to as the “parallel trends” assumption, is that the change in the outcome in the treatment hospitals between the pre- and post- periods would have been the same as the observed change in the control hospitals over the same period had the treatment hospitals not received the intervention. More specifically, the average outcome in the two groups would have evolved in parallel over time in the absence of the intervention, even if the average outcome between the Group 1 hospitals and Group 2 hospitals in the period had differed in the pre-period, before the Group 1 hospitals received the intervention. In this manner, any deviation from the relative parallel trend of the outcome into the post-intervention period between the Group 1 and Group 2 hospitals can be attributed to the effect of intervention on the outcome. On 15th January 2016, data collection commenced in both Group 1 and Group 2 hospitals, and the analysis will consider only the initial 8 months of data collection (15th January 2016 – 15th September 2016), before intervention implementation was to take place in the Group 2 hospitals. Group 1 hospitals began implementing the intervention in mid-May 2016, providing 4 months of data during the pre-intervention period and 4 months of data during the post-intervention period.
Intervention
The intervention sought to improve women’s access to PPFP through improved counseling during ANC and through the introduction of immediate PPIUD insertion services in health facilities. The intervention was implemented by FIGO in partnership with AGOTA. Specific intervention components included: 1) information education and communication (IEC) materials on PPFP, including leaflets and a video that played in the waiting room; 2) provider training on PPFP counseling and PPIUD insertion techniques; 3) provision of equipment, including Kelly’s forceps to insert the IUD; and 4) regular monitoring and support provided by FIGO and AGOTA. All four elements of the intervention were implemented in the three Group 1 hospitals, and counseling and IEC materials were also made available in satellite clinics surrounding the Group 1 hospitals where many women received ANC services before delivering in the study hospitals. The intervention was implemented in two stages: AGOTA first conducted a training of trainers (TOT) from each intervention hospital, and then trainers provided cascade training to Ob/Gyns, residents and midlevel providers in their hospital approximately 1 month later. The post-intervention period is considered to have started after the cascade training was completed in the Group 1 hospitals.
Data collection
Trained Research Assistants with previous experience conducting surveys were posted in post-natal wards of study hospitals where they conducted an interviewer-administered survey with women who consented to participate. Research Assistants were employed by AGOTA to collect data over the full project implementation period, but they were managed by the local research organization, Management and Development for Health (MDH), during the evaluation data collection period. The women’s survey collected socio-demographic data, information on PPFP counseling, including timing of counseling and information about the birth and PPFP decision-making. In addition, providers completed a survey about PPIUD insertion for women choosing a PPIUD as their PPFP method before discharge from the hospital. All data were collected using pre-programmed tablets using the CommCare application by Dimagi.
Outcomes of interest
The key outcomes of interest for this evaluation were counseling on PPIUD and choice of the PPIUD after delivery, as PPIUD insertion was a newly offered service after intervention implementation. Counseling on PPIUD was measured through women’s self-report, and a woman was considered to have been counseled if she reported PPIUD counseling during antenatal care or during her stay at the hospital for delivery. Choice of PPIUD was measured as a dichotomous variable based on both the woman’s report and the provider’s report of PPIUD insertion. Occasionally, a woman would choose to have a PPIUD inserted after she completed her survey, and the insertion would be reported only on the provider survey. If either the woman or the provider reported PPIUD insertion, the woman was considered to have chosen the PPIUD.
Analytic sample
A total of 16,930 women who delivered during the study period (15th January 2016 – 15th September 2016) in five hospitals were screened for study eligibility (age 18 or older, delivered in one of the five study hospitals, and resident of Tanzania), 15,912 (94%) were eligible (ineligibility primarily due to age under 18 years), and 15,264 (96%) of them consented to participate (Fig. 1). A total of 14,950 women with complete information on the outcomes of interest and key covariates were retained for the analysis (98% of those who consented to participate) – 8968 in Group 1 hospitals and 5982 in Group 2 hospitals. The CommCare data collection application required a response to each question, and missing data are due to participant refusal to give a response or ending the survey early.
Analysis
We conducted a difference-in-difference analysis to evaluate the effect of intervention exposure, defined as delivering in an intervention hospital during the implementation period, on the two outcomes of interest: PPIUD counseling and choice of PPIUD. Linear probability models were used to estimate the effect of the intervention in percentage points (pp). In all models, we controlled for hospital and month fixed effects. We present an unadjusted model showing the effect of intervention exposure on each outcome controlling only for the hospital and month fixed effects and an adjusted model which includes women’s socio-demographic characteristics. Characteristics include woman’s age, education, parity, marital status, religion, and whether the woman was being seen in the “fast track” or normal track service. Fast track services cost more than normal track services and typically have better amenities and a lower provider to patient ratio.
Next, we measure the intervention adherence-adjusted effect of the intervention on choice of PPIUD. Some women were not exposed to the intervention for a variety of reasons, including inconsistent implementation of the intervention counseling, because they received ANC in a facility that did not offer counseling on PPIUD, or because they did not attend ANC services. The adherence-adjusted approach assumes that all of the effect of the intervention is through counseling and allows us to measure the effect of the intervention on choice of PPIUD among women who were counseled on PPIUD. A linear probability model is used to estimate the adherence-adjusted effect, which is equivalent to a standard instrumental variables (IV) approach [14].
We also present an analysis of the determinants of women’s choice of PPIUD among women who were counseled, controlling for hospital and month fixed effects. This analysis focuses on measured aspects of quality in counseling, including timing of counseling, whether IEC materials were used (leaflet given and video seen), whether they were given an opportunity to ask questions during counseling and the types of information they recall from the PPIUD counseling they received, and women’s socio-demographic characteristics that are associated with choice of PPIUD.
Due to the small number of clusters (hospitals) included in our analysis, all of our models adjust standard errors using the cluster wild bootstrapping method with Webb weights, a six-point distribution that reduces spurious precision due to replications based on a small number of clusters [15]. This approach produces corrected standard errors for all point estimates presented herein.