The effect of a postpartum intrauterine device programme on choice of contraceptive method in Tanzania: a secondary analysis of a cluster-randomized trial

Vertical global health programmes often evaluate success with a narrow focus on programmatic outcomes. However, evaluation of broader patient-centred and unintended outcomes is critical to assess impacts on patient choice and autonomy. Here, we evaluate the effects of a postpartum intrauterine device (PPIUD) intervention on outcomes related to contraceptive method choice. The stepped-wedge cluster randomized contolled trial (RCT) took place in five Tanzanian hospitals. Hospitals were randomized to receive immediate (Group 1; n = 11 483 participants) or delayed (Group 2; n = 8148 participants) intervention. The intervention trained providers on PPIUD insertion and counselling. The evaluation surveyed eligible women (18+, resided in Tanzania, gave birth at a study hospital) on provider postpartum contraceptive counselling during pregnancy or immediately postpartum. In our completed study, participants were considered exposed (n = 9786) or unexposed (n = 10 145) to the intervention based on the location and timing of their birth (no blinding). Our secondary analysis examined differences by intervention exposure on the likelihood of being counselled on IUD only, multiple methods, multiple method durations, a broad method mix; and on the number of methods women were counselled across two samples: all eligible women, and only women who reported receiving any contraceptive counselling. Among all eligible women, counselling on the IUD alone was 7% points higher among the exposed (95% confidence interal (CI): 0.02, 0.12). Among women who received any counselling, those exposed to the intervention were counselled on 1.12 fewer contraceptive methods (95% CI: 0.10, 2.34). The likelihood of receiving counselling on any non-IUD method decreased among those exposed, while the likelihood of being counselled on an IUD alone was 14% points higher among the exposed (95% CI: 0.06, 0.22), suggesting this intervention increased IUD-specific counselling but reduced informed contraceptive choice. These findings underscore the importance of broad metrics that capture autonomy and rights (in addition to programmatic goals) at all stages of health programme planning and implementation. We performed a secondary analysis of data from a large cluster-randomized stepped-wedge PPIUD trial in Tanzania to evaluate the impact of the PPIUD intervention on outcomes related to freedom of informed contraceptive method choice. Since the intervention was primarily focused on training providers in hospitals, it was cluster randomized to avoid contamination, while the stepped-wedge design allowed the intervention to eventually reach all study hospitals. The research described here is part of a broader evaluation of the PPIUD Project implemented by the International Federation of Gynacology and Obstetrics (FIGO) (de Caestecker et al., 2018). This initiative began in 2013, with the goal ‘to address the gap in the continuum of maternal health care and to provide for the postpartum contraceptive needs of women by increasing the capacity of healthcare professionals to offer PPIUDs by training community midwives, health workers, doctors, and delivery unit staff, as appropriate, in counselling and insertion of PPIUD’ (de Caestecker et al., 2018). In 2016, FIGO brought this programme to Tanzania via their national affiliate, the Association of Gynaecologists and Obstetricians of Tanzania (AGOTA), with the goal of introducing and institutionalizing PPIUD for those seeking antenatal care and facility-based births. AGOTA implemented the FIGO PPIUD Project at six public referral hospitals throughout Tanzania (Dodoma General Hospital in Dodoma, Muhimbili National Hospital in Dar es Salaam, Mt. Meru Hospital in Arusha, Tumbi-Pwani Regional Referral Hospital in Kibaha, Mbeya Zonal Referral Hospital in Mbeya and Sekou-Toure Regional Referral Hospital in Mwanza). Doctors and nurses were trained in counselling and insertion. Hospitals were selected to receive the intervention by AGOTA, the implementer of the intervention, in order to provide coverage of PPIUD services for different geographic regions of Tanzania among tertiary care facilities. The size of facilities varied, with smaller referral hospitals such as Mbeya having fewer providers (58 combined junior doctors and trained Ob/Gyns) than the larger Muhimbili National Hospitals (240 combined junior doctors and trained Ob/Gyns). The FIGO/AGOTA programme trained providers on cadre-appropriate skills and knowledge to support the implementation of PPIUD. AGOTA organized a series of trainings in the six referral hospitals and surrounding satellite antenatal care clinics whose patients deliver at the referral hospitals. Hospital doctors were trained on PPIUD insertion and removal, while nurses and midwives in the satellite clinics were trained to integrate PPIUD counselling into routine antenatal family planning counselling. The FIGO/AGOTA initiative identified and trained a cascade of master trainers to carry out a ‘training the trainer’ approach for counselling and insertion. Counselling training sessions included ‘information on the advantages of PPIUD’, presentation of visual aids and role-playing of potential counselling scenarios (de Caestecker et al., 2018). The training was also designed to include content on method removal. Trained nurses and midwives were then expected to provide family planning counselling to women seeking antenatal care that included information about the PPIUD as part of a wide range of contraceptive methods. The central FIGO team in London maintained a data ‘dashboard’ to provide real-time feedback to clinicians and project leaders from the hospitals included in the PPIUD Project in Tanzania and the other intervention countries. The dashboard reported numbers of: (1) deliveries; (2) PPIUD insertions; (3) women counselled on PPIUD; (4) PPIUD removals; (5) women followed-up; (6) PPIUD expulsions; (7) providers trained to insert PPIUD; and (8) providers trained to counsel on PPIUD. FIGO coordinators in London regularly communicated dashboard statistics with each other, Tanzanian affiliate staff and providers. Providers were given real-time feedback about their progress towards meeting project goals, all tied to PPIUD counselling and insertion, with the primary marker of success being a calculation from dashboard numbers of the percent of all deliveries leading to a PPIUD insertion. A detailed description of the intervention can be found in de Caestecker et al. (2018). This analysis is part of a mixed-methods sequence of learning motivated by the qualitative portion nested within this larger RCT. As results from in-depth interviews with women who received antenatal care under PPIUD intervention conditions began to reveal a tendency for counselling to focus on the IUD to the exclusion of other methods (Senderowicz et al., 2021), we turned to the large quantitative dataset to explore how widespread this phenomenon was among the broader study population. Our objective was to estimate the impact of the PPIUD intervention on individual-level overall access to information about contraceptive methods and a broad contraceptive method mix during antenatal, peripartum and immediate postpartum contraceptive counselling. The study received human subjects research approval from the National Institute of Medical Research (NIMR) in Tanzania (protocol number: NIMR/HQ/R.8a/Vol.IX/2006), and ethical approval as exempt by the institutional review board at Harvard University (protocol number: IRB15–1605). Respondents provided written informed consent to be interviewed, or thumbprints and a witness’s signature if they could not sign their names. A cluster-randomized stepped-wedge trial approach with a 1:1 allocation ratio for clusters was designed to evaluate FIGO intervention sites in Nepal, Sri Lanka and Tanzania. A detailed description of that overarching trial design can be found in Canning et al., 2016. The trial is registered with clinicaltrials.gov {“type”:”clinical-trial”,”attrs”:{“text”:”NCT02718222″,”term_id”:”NCT02718222″}}NCT02718222. Because of substantial differences between the implementation of both intervention and study procedures between countries, results are presented here for Tanzania alone. In Tanzania, AGOTA selected six large public referral hospitals across different regions to receive the PPIUD intervention. Each hospital served as a cluster. There was no blinding. Given the range of hospital sizes, we employed a strategy of block randomization, in which clusters were matched on annual obstetric caseload into blocks of n = 2 clusters. Using Stata v14, EP generated a random number for each cluster and assigned the lower number within each pair to Group 1 (Dodoma General Hospital in Dodoma, Muhimbili National Hospital in Dar es Salaam, and Mbeya Zonal Referral Hospital in Mbeya) and the higher number to Group 2 (Mt. Meru Hospital in Arusha, Sekou-Toure Referral Hospital in Mwanza and Tumbi-Pwani Regional Referral Hospital in Kibaha). According to the stepped-wedge design, Group 1 hospitals received the early intervention, and Group 2 hospitals received the late intervention (Figure 1). After randomization, evaluators learned of a pre-existing PPIUD intervention at Sekou-Toure hospital in Mwanza (Group 2), and this hospital was dropped from the study. Stepped-wedge design Data collection began on 15 January 2016. Group 1 hospitals were scheduled to begin PPIUD trainings after 3 months of baseline data collection, while Group 2 hospitals were scheduled to receive the intervention after 9 months. Due to delays in training implementation and other logistical challenges, the actual timing of the rollout to each hospital varied slightly, and these delays are accounted for in the data analysis. The intervention began in Group 1 on 15 April 2016. Group 2 was due to begin on 15 September 2016, but project implementation actually began on 17 November 2016, approximately 2 months later than planned. The present analysis uses quantitative data from the first contact with respondents in the immediate period following delivery of their index pregnancy. Women were eligible to participate if they had given birth at a study hospital during data collection, resided in Tanzania, and were over age 18. All eligible women were invited to participate. Trained data collectors collected survey data in the postnatal wards of study hospitals, administering preprogrammed tablet-based questionnaires to all women who provided informed consent. Consent to participate in the evaluation was sought at the individual level after cluster-level randomization. The survey included questions on fertility desires; experiences with family planning counselling during the antenatal period, peripartum and immediate postpartum periods; perceptions of PPIUD and contraceptive intentions. The predefined primary outcome of interest to this evaluation was the percent uptake of PPIUD, defined as the proportion of all women who received a PPIUD divided by the number of women who delivered in one of the study hospitals over the course of the study period. No subgroups were excluded from this end-point. Key predefined secondary outcomes as defined at the outset of the study were also focused on PPIUD-related outcomes (e.g. the percentage of women who receive PPIUD counselling and the percentage of PPIUD acceptors who have PPIUD expulsions). Predefined primary and secondary outcomes pertained to the cluster level. Analyses of these outcomes have been performed and reported on elsewhere (Huber-Krum et al., 2019; Hackett et al., 2020; Pearson et al., 2020). The current secondary data analysis builds on the results of our concurrent nested qualitative study to expand the scope of this inquiry for this study. Results from semi-structured in-depth interviews with women at antenatal clinics exposed to the PPIUD intervention suggested that antenatal family planning counselling was directive and biased to focus on the IUD to the exclusion of other methods. Here, we have developed a set of quantitative measures to test whether these qualitative results can be expanded to the study population more generally. As such, the purpose of this analysis is to examine the effect of the intervention on a set of person-centred family planning outcomes related to method mix and availability of choice (World Health Organization, 2021). The five outcomes of interest are measured on the individual level among women who reported having received any perinatal family planning counselling and include: (1) likelihood of being counselled on the IUD alone (and no other contraceptive methods); (2) likelihood of being counselled on multiple (more than one) methods; (3) likelihood of being counselled on multiple method durations of use; (4) likelihood of being counselled on a broad contraceptive method mix; and (5) number of methods counselled on. Additionally, to assess whether losses to the number of methods counselled were compensated for by gains in counselling under the PPIUD intervention, we explored the total number of contraceptive methods on which women received counselling among all women in the study, including those who reported not receiving any family planning counselling. For all outcomes, women reported on their contraceptive counselling throughout the antenatal, perinatal and immediate postpartum continuum. Women reported whether they had received counselling on each of the following contraceptive methods at any point throughout the perinatal period: female or male sterilization, injectables, implants, oral contraceptives, condoms, emergency contraceptives, diaphragm, cervical mucus observation, calendar-based methods, lactational amenorrhoea, withdrawal, or another method. We calculated Outcome 1 as a binary variable (received counselling at any time on any method other than or in addition to the IUD = 0, received counselling on only the IUD and no other method = 1). Outcome 2 was a binary variable (received counselling on only one method = 0, received counselling on more than one method = 1). Outcome 3 was calculated as a binary variable [received counselling on methods from one duration group (long-acting, only short-acting, only medium-acting or only-permanent) = 0, received counselling on methods from two or more durations groups = 1]. Outcome 4 was calculated as a binary variable (not receiving counselling on at least one method from each contraceptive attribute group = 0, receiving counselling on a method from each attribute group = 1) (Senderowicz, 2020). A more detailed explanation of the derivation of Outcomes 3 and 4 is presented in Appendix A. Finally, Outcome 5 was calculated as an ordinal count variable, which was the sum of the number of methods on which each woman received counselling. The outcomes of interest for this analysis are focused on the content of the contraceptive counselling received. In addition to affecting the content of counselling, however, we also expected the intervention to affect the proportion of women who receive counselling at all. As a result, except where otherwise noted, we ran all analyses among two samples: (1) the sample of all respondents; and (2) the subsample of respondents who reported receiving any counselling. We used difference-in-difference linear probability models to estimate the effect of the intervention on the outcomes of interest, controlling the time period as a fixed effect and the hospital as a random effect. Coefficients for binary outcomes (IUD only; counselled on >2 methods; counselled on >2 methods durations; and counselled on a broad mix of methods) can be interpreted as the percentage point increase or decrease in the probability of the outcome associated with the intervention. Coefficients for the models for which the outcome was number of methods counselled can be interpreted as the difference in the number of methods counselled on associated with the intervention. We use an intent-to-treat approach, classifying women who received any maternity services at hospitals where the intervention had taken place as exposed, and women who received services when the hospitals had not yet received the intervention as unexposed. We present results for unadjusted and adjusted models. Adjusted models controlled for sociodemographic characteristics including women’s age, educational attainment, parity, marital status, religion and ‘fast track’ hospital service (a premium service at some hospitals that offers patients better amenities and a lower provider-to-patient ratio for a higher cost) as fixed effects. Results are presented with P-values associated with standard errors adjusted using cluster wild bootstrapping with Rademacher weights. This method is designed to correct for the inflation of precision associated with replications based on a small number of clusters. We present intraclass correlation coefficients (ICC) with associated standard errors from logistic models. To assess sensitivity to model specification, we also conducted multilevel mixed-effects regression models on the same outcomes of interest. We present the detailed methods and results of these analyses in Appendix C. A total of 22 691 women who delivered during the study period (15 January 2016–15 January 2017) in five hospitals were screened for eligibility (Figure 2). Of these, 21 033 met the inclusion criteria. Of those eligible, 1031 women (4.9%) declined to participate. Among eligible women who agreed to participate, 371 (1.9%) were missing data on primary study variables and were dropped from analyses in the full analytic sample. A total of N = 19 631 women were included in the full analytic sample. In the full analytic sample, 9305 (47.4%) reported that they did not receive any contraceptive counselling at all during their antenatal and perinatal care and were dropped for analyses to create the counselled analytic sample. The counselled analytic sample included N = 10 078 women. For Group 1, n = 5771 (n = 1874 at Dodoma General Hospital in Dodoma; n = 1220 at Muhimbili National Hospital in Dar es Salaam and n = 2478 at Mbeya Zonal Referral Hospital in Mbeya), and for Group 2, n = 4555 (n = 3153 at Mt. Meru Hospital in Arusha and n = 1353 at Tumbi-Pwani Regional Referral Hospital in Kibaha). In the counselled analytic sample, 5198 women (51.6%) were exposed to the intervention, while the remaining 4880 constitute our control group. The trial was planned for 1 year and was stopped when the expected duration was complete. Sample size calculations were performed for primary study end-points but were not performed post hoc for the secondary outcomes employed in this analysis. Given that our analytic samples were quite large (full analytic sample = 19 631 and counselled analytic sample= 10 078 observations), we operated under the assumption that these sample sizes were pragmatic for the purposes of our secondary analysis. Participant inclusion Table 1 describes the characteristics of study participants in the full analytic sample. Examining only women who reported ever receiving any contraceptive counselling perinatally, the intervention and control groups remained similar in terms of their sociodemographic characteristics and were not perfectly balanced. Those in the control group were, on average, 0.30 years younger, more likely to have at least a primary education, less likely to be married/cohabitating, more likely to be Catholic, Muslim or Protestant and less likely to be Evangelical Christian compared to their counterparts in the intervention group. There were no statistically significant differences between the respondents in Groups 1 and 2 (Pearson et al., 2020). Characteristics of respondents by intervention status, all enrolled participantsa The funder played no role in the study design; the collection, analysis or interpretation of data; the writing of the report; or the decision to submit for publication. All authors had full access to all the data in the study and accept the responsibility to submit for publication.