This sequential explanatory mixed-methods study is guided by a conceptual framework (Fig. 1) highlighting factors to be explored including both invariable and potentially mutable factors to develop the evidence base that will allow women undergoing fistula repair to maintain their reproductive and overall health over time. The ClinicalTrials.gov identifier is NCT05437939.

Fig. 1

Summarized conceptual framework between predictors and adverse fistula outcomes

Study setting and sites

Uganda’s National Fistula Technical Working Group (est. 2002) has focused on increasing fistula surgery availability [25]. Fistula surgery is available at 20 centers of excellence in Uganda, with 25 trained surgeons within national and regional referral hospitals. Regional literature suggests fistula repair is successful among ~ 80% of affected women. [26, 27]

Research sites are approximately 12 fistula repair facilities across Uganda (Fig. 2). Some sites conduct routine surgeries within ongoing urogynecological services only, others conduct fistula repair camps only, and others combine both routine care and camp models. Patients under the care of study providers and research assistants at alternative care locations will be eligible for study participation if they meet study eligibility criteria.

Fig. 2

Partnering fistula repair study sites across Uganda

Longitudinal cohort study

Our study will recruit a longitudinal cohort of 1,000 women with successful (closed) vesicovaginal fistula repair (~ 48 h after surgery). Participants will be followed for 3 years in total, with data collected via questionnaire at baseline (surgery), 2 weeks (hospital discharge), 6 weeks, 3 months and quarterly thereafter (Fig. 3).

Fig. 3

Overview of study data collection and timeline

Inclusion and exclusion criteria

Inclusion criteria are vesicovaginal fistula, completed fistula surgery with confirmed closure, age 15 or above (where individuals 15–17 meet Ugandan legal criteria for emancipation), and capable and willing to provide informed consent. Exclusion criteria are do not live within a feasible location for follow-up, operationalized by return travel back to fistula repair facility and cellular telephone network. Potential participants with prior fistula repair will not be excluded.

In the case of fresh fistula limited in size (< 2cm) and time since occurrence (< 3 months), catheterization alone may successfully heal ~ 10% [28]. If a potential participant’s fistula is eligible for and undergoes catheterization instead of surgery and the fistula is confirmed to be closed, they will be eligible for study enrollment if they meet all other study eligibility criteria.

Study procedures

Study researchers will recruit participants into the cohort following confirmation of successful fistula repair. Potentially eligible women will be identified through review of urogynecology department surgical logbooks, patient medical records, and communication with fistula providers. After fistula repair, women stay at the repair facility for 14 days minimum for post-repair catheterization. The study researcher will approach women who meet inclusion criteria in-person ~ 48 h after surgery to screen for eligibility; explain the study procedures, risks and benefits, and participant rights; invite those eligible to participate; and formally enroll those who agree. The study researcher will then collect participant contact information and administer the baseline questionnaire.

Two-week data collection will occur prior to hospital discharge. Other follow-up data collection (6 weeks and quarterly, starting at 3 months) will occur over mobile phone. Participants with no phone will be provided with a study phone and airtime. Clinical exams will be conducted at baseline and 2 weeks post-surgery for fistula and repair characterization, and data from routine post-surgical follow-up appointments at 6 weeks and 3 months post-repair will inform outcome assessment. Women reporting symptoms on interviewer-administered questionnaire or through other study or clinical communication will return to the fistula repair facility for outcome assessment and clinical care following standard clinical procedures. Transportation costs will be reimbursed for all follow-up data collection required.

Participants will be followed through 36 months post-repair, regardless of outcome and subsequent care. Some participants may undergo multiple fistula surgeries during their study participation. If this occurs, participant follow-up will be adjusted to incorporate data collection at 6 weeks and 3 months following the subsequent surgery, after which regular quarterly data collection will continue through 36 months after the participant’s enrollment.

Measures

Primary predictors to be investigated include patient-related factors, fistula-related factors, fistula repair-related factors, and post-repair behaviors and exposures, collected via structured questionnaire at all follow-up data collection points and from medical records and clinical forms when examinations are performed (Table 1). Supplementary data to characterize incontinence among women with persistent incontinence without fistula breakdown/recurrence and to assess pregnancy-related exposures and outcomes among women reporting pregnancy.

Table 1 Study measurements and timing of data capture for longitudinal cohortData analysis

To identify predictors of post-repair fistula breakdown and recurrence (Objective 1), we will first calculate the incidence of post-repair fistula breakdown and recurrence and its 95% confidence interval (CIs) overall by dividing the number of events identified by the total person-time observed. The probability of event-free survival at defined time points will be calculated using the Kaplan–Meier estimate. We will then estimate the individual and joint-effects of the patient, fistula, fistula repair, and post-repair characteristics on time to post-repair fistula breakdown and recurrence in order to identify significant factors in time to post-repair fistula breakdown and recurrence. We will fit proportional hazards frailty survival models to jointly analyze times to post-repair fistula breakdown and recurrence [29]. These models will include a shared frailty at the subject level to accommodate within-subject correlation of times to breakdown and recurrence events and interactions of predictors with event type to accommodate potential differences in the association of predictors with times to the events. These models will also include a shared frailty at the provider level since patients will be clustered within providers within facilities. We will fit the frailty survival models using routines in Stata statistical analysis software (StataCorp, College Station, TX, USA). [30]

Prior to fitting multivariable models, we will calculate the estimated correlation of all potential predictors to identify any highly correlated groups of predictors. We will not include such groups of predictors in any multivariable models. We will assess the adequacy of the proportional hazards assumption through inspection of Schoenfield residuals as a function of time. In the event our data violate the proportional hazards assumption, we will modify our modelling approach to accommodate interactions or stratification, as is most appropriate for the data. We will subsequently fit one multivariable proportional hazards regression model to document the comparative relationship between patient, fistula, fistula repair, and post-repair characteristics and the hazard rate of post-repair fistula breakdown and recurrence integrating all independent variables that were associated with the outcome in bivariable analyses at a conservative p threshold of p < 0.1. Final model selection will be determined via Akaike’s Information Criteria [52]. Secondary analyses will assess time to post-repair fistula breakdown (< 3 months post-repair) and time to fistula recurrence (≥ 3 months post-repair) separately, and by fistula etiology (obstetric versus iatrogenic), although our study is not powered for secondary outcomes.

Other methods will be used to better understand the contribution of risk factors of fistula repair breakdown and recurrence. To overcome the biases inherent to observational research in understanding causal effects [31], we will conduct secondary analyses employing propensity score methods to account for systematic differences between exposed and unexposed participants, allowing effects to be interpreted as causal, similar to a randomized experiment [32,33,34]. For these analyses, we will estimate a series of models for each key modifiable factor of interest, first predicting the probability of exposure using key baseline and other measures deemed to be relevant for developing the treatment weight, followed by analyses of the exposure and outcome incorporating the treatment weight, as described above. Finally, we will seek to construct a classification rule based on predictors using techniques such as recursive partitioning and random forests using routines in R (R Foundation for Statistical Computing, Vienna, Austria) to identify groups of women defined by the exposure characteristics with high probability of adverse outcome. [35]

To identify predictors and characteristics of post-repair incontinence (Objective 2), we will first estimate the proportion of women who experience post-repair incontinence and associated 95% confidence interval at multiple time points (e.g., 6m, 12m, 2y and 3y). Our primary analysis of predictors of persistent post-repair incontinence will focus on incontinence at 3 months, the time point by which incontinence resolvable through surgery will have resolved per expert opinion. We will first estimate bivariable relationships between each predictor and post-repair incontinence using multi-level mixed effects logistic regression modeling procedures in Stata to accommodate our clustered data [36]. Subsequently, we will estimate one multivariable model to understand the independent and joint effects of patient, fistula, fistula repair, and post-repair characteristics on post-repair incontinence at 3m, integrating all independent variables that were associated (p < 0.1) with the outcome in bivariable analyses, addressing correlation as described for Objective 1. To identify predictors of incident post-repair incontinence, we will assess incident post-repair incontinence and factors associated with time to incident post-repair incontinence using the survival analysis methods described for Objective 1. Finally, we will conduct analyses of binary predictors of interest employing propensity scores and seek to develop classification rules following the methods described for Objective 1.

Sample size

The longitudinal cohort study sample size for (up to n = 1000 women) was calculated to provide adequate power to detect a minimum difference in effect by exposures on risk of fistula repair breakdown or recurrence and incontinence of 20% (Objectives 1 and 2) using the log-rank test for two-sample comparison of survivor functions (Objective 1) and the Pearson’s chi-squared two-sample proportions test (Objective 2). These effect differences were determined to be clinically significant based on expert opinion. Power calculations were developed using Stata’s power procedure, with values α = 0.05 and 1−β = 0.80 [36]. Prior research on fistula recurrence risk suggests that factors of interest for our survival analyses (Objective 1) may have hazard ratios ranging from 1.0 to 3.4 [9]. Fig. 4 illustrates the minimum sample size required for estimation of effect estimates ranging from 1.1 to 2.0 with parameters α = 0.05 and 1−β = 0.80, demonstrating adequate power for two-sample comparison of survivor functions (i.e., time to fistula repair breakdown, time to fistula recurrence) with a sample size of 1000 for effect estimates (hazard ratios) of 1.2 or higher, illustrating a 20% or higher risk difference, accommodating some loss to follow-up. Other research has reported repair breakdown or recurrence rates of approximately 15%. With our target sample size of 1000, we anticipate being able to estimate this incidence with precision range 2.5% (i.e., between 12.8 and 17.4%).

Fig. 4

Sample size calculations for robust estimation of fistula breakdown or recurrence and persistent incontinence outcomes

Sample size calculations for comparisons between risk factors of post-repair urinary incontinence at 6 and 12 months were estimated using the Pearson’s chi-squared two-sample proportions test with parameters α = 0.05 and 1-β = 0.80. We have estimated risk differences of approximately 10 percentage points, across a range of possibilities, given the lack of informative estimates from the literature. For a potential comparison of 9 percentage points (e.g., from 1 to 10%) we achieve power of 0.80 at approximately 200 study participants (Fig. 4). On the other end of the range (e.g., a comparison between 50 and 60%; Fig. 4), statistical power of 0.80 is achieved with a minimum sample size of 800 participants.

Qualitative component

Qualitative research with key stakeholders will be conducted to expand inform the development of feasible and acceptable intervention concepts targeting risk factors identified from our longitudinal cohort findings (Objectives 1–2).

Study participants

We will enroll approximately 80 individuals in total, including women with fistula, family members, community members, clinical and social service providers, and government. We will purposively sample ~ 40 longitudinal cohort participants to reflect study variability in region and adverse outcome experience. Other key stakeholders (~ 40) will be identified through discussion with study investigators, site leads and research assistants, and other clinical and social service providers for fistula in Uganda, to maximize variability in respondent region and clinical and patient support roles. Identified individuals will be invited to participate over the phone, email, or in-person and those who are interested and are willing to provide informed consent will be scheduled for an in-depth interview with a trained qualitative interviewer at a convenient time and private location. Informed consent for all respondents will be conducted in person, with written or thumbprint confirmation obtained, as appropriate. To respect the privacy and confidentiality of longitudinal cohort participants, permission will first be sought from the research participant before recruiting potential family member or peer qualitative participants.

Study procedures

Based on our quantitative findings (Objectives 1 and 2), literature, and expert clinical and contextual experience, the research team will develop a semi-structured and open-ended in-depth interview guide for key stakeholder interviews to obtain a nuanced understanding of their perspectives on feasible and acceptable potential intervention opportunities for addressing key risk and causal factors associated with adverse outcomes. Exploration of intervention possibilities with stakeholders may employ constructs from health behavior theories COM-B (‘capability’, ‘opportunity’, ‘motivation’, and ‘behavior’) model (Fig. 5) and the theoretical domains framework (TDF) for understanding individual and contextual issues, and the Consolidated Framework for Implementation Research (CFIR) for pre-implementation assessment of factors important to successful implementation (i.e., intervention characteristics, inner setting (characteristics of implementing organization), outer setting (features of the external context or environment), and implementation process (strategies or tactics for implementation setup or delivery) [37, 38]. Interviews will be conducted in a private setting by an experienced qualitative interviewer and are anticipated to take 1–2 h. Participants may be asked to respond iteratively as new data arises during the qualitative process. Interviews will be audio recorded and translated into English and transcribed.

Fig. 5

COM-B (‘capability’, ‘opportunity’, ‘motivation’, and ‘behavior’) model to inform intervention exploration for qualitative objective. [39]

Data analysis

During the iterative interview and analysis process, we will combine COM-B [39], CFIR [37], and TDF analyses to identify a series of behavioral and implementation targets for each risk factor identified within our quantitative analysis, and for each of these we will 1) classify using the COM-B [39], 2) detail potentially modifiable determinants of behavior (e.g., barriers or facilitators) across CFIR [37] domains, 3), list the theoretical domain and techniques for behavior change using the TDF, and 4) develop and assess potential implementation strategies across multiple actors to achieve the desired change.

Further qualitative data analysis will follow a 2-stage systematic process [40]. The first stage will involve data coding and classification by reviewing the transcripts for potential conceptual categories, using the in-depth interview guide. Two types of codes will be employed: deductive and inductive/emergent. First, deductive codes that represent expected influences will be applied to the data; these will be taken from the existing literature and the theoretical orientation of the interview guide (i.e., COM-B, TFR and CFIR construct list). Next, inductive codes that emerge organically from the data will represent themes that were not expected by the researchers. Emergent themes will be identified based on recurrence rate and on similarities and differences noted across the texts. A codebook will be developed from the themes that will include a detailed description of each code, code inclusion and exclusion criteria, and examples of the code in use. Coded data will be analyzed to describe the different dimensions and commonalities of each theme, their distribution across socio-demographic variables, and the patterns and linkages between themes. Comparisons will be made to detect divergent views among participants and to contrast observations by sample population characteristics and type of key stakeholder.

Sample size

The qualitative sample size was selected on the basis of our prior experience with thematic saturation from our research among women with fistula in the Ugandan context; however, final sample size will be determined through iterative assessment of thematic saturation as data are collected across different participant types. [41]

Ethical approval

The study protocol was approved by the Makerere University School of Medicine Research and Ethics Committee (Ref#: 2021–277), the Uganda National Council for Science and Technology (REF#: HS2033ES), and the University of California, San Francisco Human Research Protection Program, Committee on Human Research (IRB# 21–33559). All individuals eligible for participation will undergo an informed consent process prior to enrollment; individuals unable to provide signatures for informed consent will provide thumbprint confirmation. All study procedures will be performed in accordance with the relevant guidelines and regulations.