Economic burden of female genital mutilation in 27 high-prevalence countries

Countries

The analysis included 27 countries, selected on account of having a high burden of FGM and having information on prevalence by type of FGM available. The included countries were: Benin, Burkina Faso, Central African Republic, Côte d'Ivoire, Cameroon, Chad, Djibouti, Egypt, Eritrea, Ethiopia, Ghana, Guinea, Gambia, Guinea-Bissau, Iraq, Kenya, Mali, Mauritania, Niger, Nigeria, Sudan, Senegal, Sierra Leone, Somalia, Togo, United Republic of Tanzania and Yemen. The adult prevalence of FGM and sources of this data are reported in online supplemental appendix 1.

Model structure

A dynamic Markov state-transition model with a time horizon of 30 years was constructed. The model states took into account the different life stages in which health complications from FGM may arise (childhood, reproductive age, periods of pregnancy and childbirth, and later life), and were broadly partitioned into individuals unaffected by FGM, individuals affected by type I/II FGM and individuals affected by type III FGM. This stratification of the population by type of FGM allows estimation of costs specific to FGM typology. The model was run separately for each country of the analysis, using national population age structures. States of the model are shown in online supplemental appendix A. The model was implemented in R v 3.6.19 using the HEEMOD package.10 The model code and underlying functions were extensively unit tested using RUnit.11

Model parameters and population structure

Probabilities of transitioning between states were derived from a systematic review and meta-analysis of the clinical and epidemiological literature (see table 1) and public databases. Briefly, model population size and age structure were determined by country from the World Population Prospects (WPP) medium variant revision,12 using the breakdown by 5-year age groups. Ageing within the model cohort over the 30-year time horizon was represented by dynamically changing transition probabilities between the 0–14, 15–49 and 50+ age groups according to the current population age structure. The model population was initiated according to the baseline national age structure and prevalence of FGM in each country. Specifically, the number currently at risk of FGM was the baseline 0–14-year age group minus those already exposed to FGM, while the adult and later life population with FGM was the FGM prevalence multiplied by the 15–49 and 50+ year population size.

For each country, the General Fertility Rate (GFR, total number of live births per 1000 women aged 15–49) was taken from the medium variant revision 2017 of the WPP for the years 2018–2047. GFR was used to determine the annual probability of a childbirth occurring in the cohort of women of reproductive age (15–49 years). The inflow of population to the model was determined by births of the WPP medium variant revision 2017 extrapolation for each year, using a secondary sex ratio (rate at birth) of 0.489 (105 males to 100 females) to determine annual female births.13 Further details of model parameters are provided in online supplemental appendix A, while details of model transitions are provided in online supplemental appendix B.

Annual risk of FGM

The annual risk of being subjected to FGM was estimated from the prevalence of FGM in the adult population by country, recorded in Demographic and Health Survey (DHS) or Multiple Indicator Cluster Survey (MICS) data. The data from these surveys is categorised as ‘cut, flesh removed’ or ‘sewn closed’, corresponding to type I/II and type III FGM, respectively. It is a limitation of the data that we cannot fully align the categories from population-based studies with the categorisations measuring health complications, but every effort was made to ensure that these categories aligned. In some studies measuring associations with complications, a category of ‘any FGM’ was used given the lack of specificity in the authors’ descriptions of FGM typology. Considering the prevalence of FGM at 15 years of age to be a cumulative incidence and thus representing an exponential decay in the unaffected population over 15 years, and assuming all cases of FGM occur in the 0–14-year age group, the annual incidence rate i was derived from the formula for the cumulative distribution function, where CI(t) is the cumulative probability of being exposed to FGM by year t, i is the annual incidence rate and t is the time after birth in years (here CI(t=15)=prevalence at age 15):

Finally, the annual incidence rate was converted to an annual probability using the exponential formula with t=1 denoting a 1-year probability:

Combining the two formulae with the derivation of i for t=15 years yields the following annual risk of FGM in girls aged 0-14

The annual probability was verified by calculating the evolution of a hypothetical cohort of individuals over 15 years (online supplemental appendix C)

Background mortality

Background mortality is incorporated to account for the lifespan during which FGM complications are incurred. Background mortality by age group was derived from the WHO Global Health Observatory indicator LIFE_0000000030, representing the probability of dying between ages x and x+n. The most recent year available was used. Five-year mortality rates for the 5-year age buckets reported by the Global Health Observatory were rescaled to 1-year mortality rates in the 0–14, 15–49 and 50–84 age groups of the model. Additional details are provided in online supplemental appendix D.

Intervention effectiveness

Evidence reviews on what works to reduce FGM14–18 reveal a set of promising programmatic activities that may be contributing to observed declines in FGM in some settings. These include community engagement activities, including community dialogues and pledges; working with religious leaders, community leaders, peer educators and other opinion leaders; awareness raising through various communications channels; alternative rites of passage in places where FGM is considered a rite of passage; legislative and policy changes and other accountability mechanisms to penalise the practice of FGM; and school-based and health sector-based education initiatives. However, most studies explored the effects of interventions on intermediate outcomes, not on actual reductions in the practice of FGM. Further, most of the interventions were evaluated using pre- and post-test designs or quasi-experimental study designs, and did not have sufficient statistical power to detect changes in outcomes.

Given the limitations in study designs and lack of replication in a range of settings, we did not model incidence reductions as a result of specific interventions in the 27 study countries. Rather, this analysis opted for a continuation of the most recently observed incidence and two hypothetical change scenarios with partial and complete reduction in the incidence of FGM, the latter being an admittedly optimistic scenario but one contemplated in the Sustainable Development Goals. The authors further recognise that FGM prevalence is declining in some countries, while in others it is stable, and in yet others it is increasing. The assumption of continuation of current incidence in the absence of intervention is therefore debatable as it masks country level variation; however, given the variability in these trends and the unpredictability of their trajectory, the current incidence scenario provides a reference scenario against which to compare the projected declines reflected in the hypothetical intervention scenarios.

These hypothetical interventions were assumed to have an effectiveness of 50% (partial abandonment) and 100% (full abandonment) in the reduction of the annual risk of FGM for girls aged 0–14. The effect of interventions was assumed to be immediate from the first model year. Due to the irreversible nature of FGM, abandonment of FGM does not eliminate the prevalent population. Consequently, in our model abandonment only affects the number of FGM-affected individuals in the 0–14-year age groups. Abandonment was assumed to reduce the risk of FGM equally among all girls aged 0–14.

Resource utilisation

Healthcare utilisation associated with FGM was based on a review and meta-analysis of the literature on clinical complications associated with FGM.19–109 All complications for which a significant association with FGM was found, and for which corresponding interventions could be identified, were included in the healthcare utilisation calculations. Random effects meta-analyses were used to estimate the pooled (inverse variance) absolute difference in risk between FGM and non-FGM groups using STATA 16. The economic burden associated with FGM was calculated by linking each clinical complication with a clinical management strategy, sourced from the WHO OneHealth tool list of interventions,110 the scientific literature and expert opinion (table 1). As outlined in table 1, costs associated with specific health complications were assigned to either prevalent type I/II, type III or any FGM (type I/II/III) cases, depending on the associations supported by the literature review.

Country-specific healthcare utilisation rates for specific health complications were not included in the model as these data were not available. Some existing research suggests that women’s FGM status can also impact her healthcare utilisation, making her less likely to seek treatment.111 However, since most of the data on health risk are based on facility-based studies of women seeking care for some health services (most often maternal or reproductive health services), these studies suggest that women with FGM do seek care for health services and do receive treatments for FGM-related complications even if the proportion of women receiving treatment for particular health complications is not available. The healthcare costs calculated in this study constitute the costs of treating health conditions that may arise from FGM. These costs will be an overestimate of the actual healthcare costs of women seeking care for these conditions since many women will not seek care and health systems in the study countries may not currently routinely diagnose or offer care for these services.

Unit costs

Unit costs were derived from the WHO OneHealth tool appendix on Intervention Assumptions,110 which describes human resources, medicines, consumables and other resources required for individual clinical interventions. Details of these interventions and unit costing are given in online supplemental appendix E. When no information was available in the OneHealth tool for specific complications (genital wounds and ulcers, clitoral neuroma, keloids of labia, post-traumatic stress syndrome), a targeted literature review was carried out to identify published evidence of unit costs (online supplemental appendix F); however, no appropriate unit costs were identified. In such cases, unit costs were based on expert opinion in combination with sources describing the clinical management of these complications. Global unit costs were used for commodities, while country-specific unit costs from WHO-CHOICE were used for costs of outpatient consultations and inpatient bed days.112 As no country costs were available for Somalia, values from the Central African Republic were used as a proxy.

Unit costs were inflated from base year US$ (2010 for inpatient and outpatient consultation costs; 2015 for commodity costs sourced from OneHealth Tool) to 2018 US$ using country-specific GDP deflators.113

All global unit costs are summarised together with relevant clinical complications in table 1. Country-specific costs of outpatient consultations and inpatient days are included in online supplemental appendix J.

For reporting, costs are stratified by acute complications (bleeding, abscesses, etc), uro-gynaecological complications (urinary tract infections, neuroma, etc), obstetric and neonatal complications (caesarean sections, dystocy, etc) and psychological/sexual complications (anxiety, depression, etc), as outlined in table 1.

Sensitivity analysis

We undertook one-way deterministic sensitivity analyses using plausible ranges for the parameters anticipated to have a substantial impact on the model results, including annual incidence of different types of FGM, background mortality rates, annual births, and sub-categories of costs: acute complications, infections, scarring and obstetrics (detailed in online supplemental appendix G). Due to lack of data on distribution or plausible ranges of these parameters, we varied point estimates from 0.5× to 2× the base case value, except for cost categories where we varied point estimates from 0.25× to 4× the base case value. Cost sub-categories were examined in sensitivity analysis in aggregate rather than individual risks and unit costs, due to the many constituent parts of these sub-categories. A total of 27 model parameters were examined, and sensitivity analysis was undertaken at the country level to account for differences in context and country-specific consultation costs. An overview of the parameters varied in the sensitivity analysis and the range is given in online supplemental appendix G.

Limitations

A few limitations of the analysis should be noted. Three countries for which FGM prevalence data were available were not included because data on the breakdown of types of FGM were not available. Two of these countries, Indonesia and Liberia, have a high prevalence of FGM (49% and 44%, respectively), but since different risks are assigned by type of FGM, and since these countries did not have data available by FGM type, they were excluded from the analysis.

The present model uses uniform unit costs of commodities across countries and country-specific unit costs for outpatient consultations and inpatient days. While this is a simplification, we believe the basic commodities required for most of the interventions included here are not subject to large variation in procurement cost and, additionally, the costs of commodities are modest compared with the costs of consultations and inpatient days.

The model makes assumptions around treatment-seeking behaviour. In our work we model the economic impact assuming all healthcare needs result in formal healthcare contact. While we acknowledge that a proportion of healthcare need would result in informal care seeking, or no care seeking at all, we present this analysis in the context of the move towards universal health coverage (UHC). Consequently, our results should be interpreted as the economic burden in a setting with or close to achieving UHC.

We do not model the economic impact of mortality or indirect costs of productivity loss. As such, our model presents only direct costs to the healthcare system associated with FGM and is likely to be an underestimate of the true negative economic and societal impact of FGM.

We recognise that the economic burden will differ between countries, and in particular that our estimates may overestimate costs in settings where rates of FGM are systematically and consistently declining.

Finally, although we include costs related to psychological and sexual health complications, our model does not capture impacts on quality of life or the intangible burden associated with pain, sexual dysfunction or other consequences of FGM. A complementary review of the qualitative literature on FGM and psychosocial well-being discusses these effects.128