Potential cost-effectiveness of prenatal distribution of misoprostol for prevention of postpartum hemorrhage in Uganda

Background: In settings where home birth rates are high, prenatal distribution of misoprostol has been advocated as a strategy to increase access to uterotonics during the third stage of labor to prevent postpartum hemorrhage (PPH). Our objective was to project the potential costeffectiveness of this strategy in Uganda from both governmental (the relevant payer) and modified societal perspectives. Methods: and Findings To compare prenatal misoprostol distribution to status quo (no misoprostol distribution), we developed a decision analytic model that tracked the delivery pathways of a cohort of pregnant women from the prenatal period, labor to delivery without complications or delivery with PPH, and successful treatment or death. Delivery pathway parameters were derived from the Uganda Demographic and Health Survey. Incidence of PPH, treatment efficacy, adverse event and case fatality rates, access to misoprostol, and health resource use and cost data were obtained from published literature and supplemented with expert opinion where necessary. We computed the expected incidence of PPH, mortality, disability adjusted life years (DALYs), costs and incremental cost effectiveness ratios (ICERs). We conducted univariate and probabilistic sensitivity analyses to examine robustness of our results. In the base-case analysis, misoprostol distribution lowered the expected incidence of PPH by 1.0% (95% credibility interval (CrI): 0.55%, 1.95%), mortality by 0.08% (95% CrI: 0.04%, 0.13%) and DALYs by 0.02 (95% CrI: 0.01, 0.03). Mean costs were higher with prenatal misoprostol distribution from governmental by US$3.3 (95% CrI: 2.1,4.2) and modified societal (by US$1.3; 95% CrI: -1.6,2.8) perspectives. ICERs were US$191 (95% CrI: 82, 443) per DALY averted from a governmental perspective, and US$73 (95% CI: -86, 256) per DALY averted from a modified societal perspective. Conclusions: Prenatal distribution of misoprostol is potentially cost-effective in Uganda and should be considered for national-level scale up for prevention of PPH. A decision model was developed in Microsoft Excel (2011) to compare two strategies: 1) misoprostol (600μg taken orally immediately following delivery of the baby) distributed to all mothers either at a prenatal visit in the third trimester or as an add-on to a safe delivery kit (SDK) and 2) only oxytocin (10 international units administered intramuscularly immediately following delivery of the baby), whose use is limited to births in health facilities, is available (Fig 1). Prenatal visits are ideal for distribution because over 90% of mothers will visit a skilled provider at least once [5]; and addition to a safe delivery kit is potentially efficient because it would leverage an already existent service, and could ensure that mothers are aware that the drug should be used as part of the delivery procedure. The delivery pathway trajectory i.e., place of delivery (health facility versus non-health facility birth) and assistance at delivery (skilled assistance, traditional birth attendant, relative or friend and unassisted delivery) is defined by wealth quintile from the Uganda Demographic and Health Survey. We follow women through to the immediate postpartum period in which they may receive prophylactic uterotonics (or not), may experience postpartum hemorrhage (or not), may access emergency obstetric care or not and may recover or die due to postpartum hemorrhage. We allow access to emergency obstetric care to vary by delivery pathway. Further, we account for the potential of misoprostol misuse through stillbirth or uterine rupture outcomes. This model structure is used to project the costs and outcomes (incident postpartum hemorrhage, mortality due to postpartum hemorrhage and disability adjusted life years) of a cohort of pregnant women in Uganda. We tracked the potential delivery pathways and outcomes of a mother in Uganda (Fig 1) from the prenatal period, through labor and delivery, to the immediate postpartum period in which she could experience PPH and either survive or die due to PPH. Data from the Uganda Demographic and Health Survey (UDHS) 2011 [5] show that 47.6% of births take place outside health facilities, without skilled attendance. Some commonly cited reasons for delivery outside health facilities include the inaccessibility of level III health facilities (the lowest level that should offer maternity services) either due to distance required to travel or lack of transportation and money to deliver in facilities [5,20]. Other reasons for delivery outside health facilities include: the sudden onset of labor and short labor which may preclude making it to the health facility; facility-based factors such as poor staff attitudes and lack of privacy; and sociocultural factors like the lack of power in decision-making in relation to delivery [21]. Maternal age, education, child birth order, region, rural or urban residence and economic status are important determinants of maternal delivery pathway trajectories—i.e., both place of delivery (health facility versus non health facility) and the type of assistance at delivery (skilled versus unskilled assistance). Younger (less than 20 years), more educated (with at least a secondary education) women, first order births, women from Kampala, those who resided in urban areas and women in the highest economic quintile were more likely to deliver in health facilities with skilled attendance [5,20]. The wealth index in the UDHS is a useful summary proxy of key variables (maternal education, region, urban versus rural residence and economic status), that drive maternal delivery trajectories [5,20]. Therefore depending on the UDHS wealth index [5], a mother could deliver either at a hospital or health center (with skilled assistance) or out of the health facility (either assisted by TBAs, friend or relative, or without assistance). Additionally, structuring the model according to wealth index provides a useful way to demonstrate the distribution (heterogeneity) of outcomes across wealth strata in a population [22]. The model allowed for differential access to uterotonics and emergency obstetric care (EmOC) by delivery pathway trajectory [23,24]. A mother could receive the first line drug, oxytocin if she delivered at a health facility (hospital or health center). If oxytocin was unavailable at the health facility or if she delivered outside a health facility (assisted by a TBA, relative/friend or by herself), she could receive either misoprostol or no uterotonic. If she experienced PPH, a mother may access EmOC at a hospital or not, depending on where she delivered. Finally she may die or survive. There are several published case reports of stillbirths and vaginal rupture following the misuse of misoprostol prior to birth of the baby [25–30]. We therefore accounted for this possibility in our model. The outcomes in the model were PPH, death due to PPH, DALYs and costs associated with delivery, PPH prevention and treatment. We estimated costs from the governmental (the relevant payer in Uganda) and modified societal perspectives [31]. The governmental perspective considered direct medical costs (including health worker time, drugs, sundries, and laboratory tests) and some direct non-medical costs (health facility overhead and capital costs). The modified societal perspective considered additional direct non-medical costs (out-of-pocket travel and upkeep costs), and indirect costs (time costs for the mother and her care giver). Future outcomes were discounted at 3% [32]. Data from UDHS, 2011 were used to estimate the probabilities for each delivery pathway trajectory of a mother in Uganda by wealth quintile (Table 1) [5]. 1 conditioned on wealth quintile 2 conditioned on delivery in health facility 3 Joint probabilities of non-health facility delivery and either unassisted or assisted by TBA †TBA = Traditional Birth Attendant Access to uterotonics was defined as the probability that a mother in the third stage of labor would receive a uterotonic. Data from two health facility based studies were used to estimate probabilities of receipt of oxytocin [23,24] and data from a demonstration project in Tanzania, an East African country similar to Uganda culturally and in maternal health seeking behaviors, were used to estimate probabilities of receipt of misoprostol (Table 2) [33]. The third author, a consultant obstetrician and gynecologist in Uganda provided expert professional opinion on access to EmOC for women who develop PPH. We define access to EmOC as access to a facility that provides the basic signal functions: parenteral antibiotics, and oxytocics, manual removal of placenta, removal of retained products and assisted vaginal delivery [23,24]. To obtain an estimate of the probability of PPH in unskilled births, we pooled estimates (using a random effects model, S1 Fig) of the probabilities of PPH in non-health facility births from the non-interventional arm of studies in which a uterotonic was compared to no uterotonic [10,34–36] in unskilled births (PPH probability: 0.122; 95% CI: 0.0678, 0.1763). The ancillary care, pre-labor risk assessment and practice of other components of AMTSL (other than oxytocin) may independently lower the underlying risk of PPH for women who give birth under skilled attendance in health facilities. To approximate this reduction in risk, we adjusted the above estimate and its lower and upper bounds by 63%, a ratio of the relative risk (RR) of PPH comparing active to expectant management of the third stage of labor (0.34) [37], and the RR of PPH comparing oxytocin to placebo (0.53) [38]. Oxytocin and misoprostol lower the risk of PPH; however misoprostol appears to have heterogeneous effects depending on whether it is used with or without skilled attendance at delivery (Table 3). Based on a Cochrane meta-analysis, we estimated the RR of PPH with oxytocin (versus placebo) in the setting of skilled attendance at delivery as 0.53 (95% CI: 0.38, 0.74) [38,39]. We pooled estimates from two trials (S2 Fig) [11,40], that compared the risk of PPH with misoprostol versus placebo in a setting of skilled assistance at delivery (RR 0.84; 95% CI: 0.73, 0.97); and the RR of PPH with misoprostol (versus placebo) in unskilled births was estimated from the trial of Derman et al. [10] as 0.53 (95% CI: 0.39, 0.74). These differences could be explained by differential skill level and practice of AMTSL in skilled deliveries—because the underlying risk of PPH is lower, this could make the independent effects of misoprostol less marked [41]. a Computed by multiplying the baseline probability of PPH in unskilled delivery by the ratio of the relative risk of PPH comparing active to expectant management of the third stage of labor b A random effects meta-analysis of incidence of PPH in the non-interventional arms of clinical studies comparing a uterotonic to no uterotonic c A random effects meta-analysis of trials that compared the risk of PPH with misoprostol versus placebo in a setting of skilled assistance at delivery Approximately 6.2% of women who develop PPH in hospitals in Uganda die [42]. This was used as the case fatality rate (CFR) for PPH for women who access EmOC after delivering in a health facility. In general, those who do not deliver in health facilities eventually make it into care, albeit belatedly, when they experience complications. Because of the late presentation, we assumed that these had a twice-higher CFR. A similar assumption has been used in a previous study of the economic costs of induced abortions in Uganda [43]. Misoprostol implementation projects and clinical trials have demonstrated very low rates of misuse of misoprostol and increases in likelihood of delivery with appropriate education. A study in Ghana found that 99% of mothers used misoprostol appropriately and the likelihood of health facility delivery increased from 30 to 69% [15]. In Uganda, the MamaMiso study found that 97% of women used misoprostol appropriately, and the 3% that took it after delivery show no adverse events [18]. However, there have been case reports of both still birth [25] and uterine rupture [25–30] following use of misoprostol for labor induction in women without previous cesarean deliveries. To account for this, we used the estimate of appropriate use from the MamaMiso study [18], and assumed conservatively that the 3% who took it before birth of the child experienced an adverse outcome—as there are more published case reports, we assume a 70% uterine rupture probability and 30% still birth probability. We then estimated the probability of death from uterine rupture in Uganda from Kaye et al. [42] as 0.118 and assumed that still birth was not fatal for the mother. We used a microcosting approach to estimate health resources used and costs for health facility and non-health facility delivery, and for management of PPH, still births and uterine rupture at a hospital (detailed inputs in S1 Table and summary of cost model in S2 Table). Total costs in each category were obtained by multiplying health resources used by unit costs. Estimates of health worker time, quantities of drugs and sundries and average length of hospital stay for PPH weighted by proportions of use at health centers and hospitals were obtained from the Uganda safe motherhood costing study [44]. We applied updated unit costs for drugs and sundries from local supplier price catalogue [45,46]. The unit cost of misoprostol was obtained from the Management Sciences for Health International Drug Price Indicator Guide (2012) [47]. To estimate the cost of a program of distribution of misoprostol, we further assumed that 200 health workers (to serve a cohort of 10,000 mothers) would undergo training for 5 days at US$10 per day, mother training by health workers would cost 20 nurse full time equivalents (1 FTE is the annual salary of 1 nurse for 1 year) plus an additional packaging cost (equivalent to 50% of the cost of a dose of misoprostol) (S3 Table). Annual health worker pay was estimated from published pay schedules for health workers in Uganda [48]. The unit cost for blood transfusion was obtained from a study in Malawi, which is similar to Uganda [49]. Laboratory test costs were obtained from a hospital-based analysis in Uganda [50]. Facility and overhead unit costs were estimated from the WHO-CHOICE [51], model for estimating unit costs for hospitalizations [52], and either weighted by the proportion of patients who experience overnight stays for normal deliveries or by the average length of stay for PPH in a hospital. Average travel and upkeep costs at health centers and hospitals and the average cost for a normal vaginal delivery by a TBA (assuming no travel costs to TBA) were obtained from a costing study of maternal health services in Uganda [53]. The cost of management of still birth was assumed to be the same as that for a normal vaginal delivery because the management protocols are the same. The management protocols for uterine rupture involve either a total or subtotal hysterectomy, or repair with or without tubal ligation. As we had no data on the costs of (or proportions of women receiving) each of these surgical interventions, we assumed women underwent a major surgical operation equivalent to the cost of a cesarean delivery estimated from the Uganda safe motherhood costing study [44]. We computed patient and care giver time costs under each delivery pathway, for treatment of PPH and complications resulting from misoprostol misuse. For health facility delivery and management of complications, the time spent was estimated as a sum of the average travel time to the health facility [53], and the average length of stay in the health facility [44]. For TBA births, we assumed the travel time and length of stay was the same as that in a health center. For assisted home delivery, one relative or friend was assumed to stay with the mother for an average of 3 days. Lost time was valued at national GDP per capita for 2012 ($547) assuming 264 workdays in a year and 8 work-hours in a day. All costs were converted to 2012 US Dollars using the Uganda Consumer Price Indicator for health [54] and the Bank of Uganda exchange rate on December 31, 2012 [55]. The average durations of disability due to PPH, vaginal stillbirth and uterine rupture in Uganda were assumed to be 1, 1 and 3 months respectively. The life expectancy at birth was 62.5 years [56], and the age distribution of pregnant women in Uganda was obtained from the UDHS [5]: 16.7% were less than 20 years, 69.7% were 24–34 years and 13.5% were 35–49 years old. The principal source of disability in PPH is anemia; therefore we applied the disability weight for severe anemia (0.164, 95% CI: 0.112, 0.228) [57]. Disability weights for stillbirth and uterine rupture were equated to values for moderate (0.123, 95% CI: 0.083, 0.176) and severe abdominal pelvic problem (0.326, 95% CI: 0.219, 0.451) respectively from the 2010 Global Burden of Diseases study [57]. We estimated DALYs as a sum of years lived with disability (YLD) and years of life lost (YLL). YLD was computed as a product of the probability of experiencing a disability, disability weight and the duration of each disability. To obtain YLL, we summed the years of life lost (up to the life expectancy) and weighted this by the expected probability of death. The total DALYs were obtained by summing these weighted values. Future YLL were discounted at 3%. Using this set up, we simulated a cohort of pregnant women in Uganda and computed expected costs, incidence of PPH, mortality, and DALYs for each strategy. We computed incremental costs and outcomes (changes in incidence of PPH and mortality due to PPH and DALYs averted) by subtracting the results for the status quo strategy from that of prenatal misoprostol distribution. We performed scenario analyses to evaluate our assumptions. First, because delivery pathways depended on wealth quintile, we performed a stratified analysis to estimate incremental cost and outcomes in each wealth quintile. We conducted threshold analyses to determine the per-patient cost of misoprostol distribution, the probability of misuse and costs of collection and destruction of unused misoprostol, that would result in an ICER of US$1,641 (3 x GDP per capita) from both government and societal perspectives. Lastly, we evaluated a range of odds ratios (OR) for health facility delivery to examine the impact changes in odds of health facility delivery due to the program could have on outcomes (for simplicity and due to lack of supporting data, the OR did not depend on wealth quintile). Univariate sensitivity analysis (USA) was conducted to determine the impact of uncertainty around parameters on incremental costs and outcomes. Parameters were assigned plausible ranges based either on 95% confidence intervals, or ±50% where assumed or based on expert opinion to represent greater uncertainty. Probabilities were capped at 1 if the upper bound exceeded this value. Unit costs for drugs and sundries were varied by ±20% because not much variation was expected as only two companies supply public sector facilities. Probability distributions were assigned to all parameters used in the model: a dirichlet distribution for the probability of belonging to one of the wealth quintiles and the age distribution of pregnancies; beta distribution for probabilities and disability weight; lognormal distribution for relative risks and normal distribution for costs, and life expectancy at birth for women and exponential distributions for contact hours per-patient for the different health workers, length of hospital stay, duration of PPH. We performed 10,000 second order Monte Carlo simulations and calculated the 95% credible interval (95% CrI) for incremental costs and outcomes (as the values at the 2.5th and 97.5th percentiles). A net benefit framework was used to compute the probability that prenatal distribution of misoprostol was cost-effective. Threshold ICERs for interventions in developing countries are derived from multiples of GDP per capita [13,51,58]. An intervention is considered highly cost-effective if the ICER is less than 1 x GDP per capita and cost-effective if less than 3 x GDP per capita. Willingness to pay (WTP) per DALY was varied from 0 to $1,800. On each occasion, the net benefit for each simulation was calculated using the formula: The probability that prenatal misoprostol distribution was cost-effective was computed by determining the proportion of simulations for which the net benefit was greater than zero. This was done over the values of WTP per DALY and a Cost effectiveness acceptability curve (CEAC) was generated. The data used in this modelling analysis were drawn from publicly available sources, assumptions and the expert opinion of a co-author. It did not involve “human participants” as described by the University of Washington Institutional Review Board, and therefore was exempt from prior ethics approval. All authors had full access to the data and the spreadsheet model, and reviewed and approved the final version of the manuscript. The authors declare no competing interests.