Cost-effectiveness of the non-pneumatic anti-shock garment (NASG): Evidence from a cluster randomized controlled trial in Zambia and Zimbabwe

Background: Obstetric hemorrhage is the leading cause of maternal mortality, particularly in low resource settings where delays in obtaining definitive care contribute to high rates of death. The non-pneumatic anti-shock garment (NASG) first-aid device has been demonstrated to be highly cost-effective when applied at the referral hospital (RH) level. In this analysis we evaluate the incremental cost-effectiveness of early NASG application at the Primary Health Center (PHC) compared to later application at the RH in Zambia and Zimbabwe. Methods: We obtained data on health outcomes and costs from a cluster-randomized clinical trial (CRCT) and participating study hospitals. We translated health outcomes into disability-adjusted life years (DALYs) using standard methods. Econometric regressions estimated the contribution of earlier PHC NASG application to DALYs and costs, varying geographic covariates (country, referral hospital) to yield regression models best fit to the data. We calculated cost-effectiveness as the ratio of added costs to averted DALYs for earlier PHC NASG application compared to later RH NASG application. Results: Overall, the cost-effectiveness of early application of the NASG at the primary health care level compared to waiting until arrival at the referral hospital was $21.78 per DALY averted ($15.51 in added costs divided by 0.712 DALYs averted per woman, both statistically significant). By country, the results were very similar in Zambia, though not statistically significant in Zimbabwe. Sensitivity analysis suggests that results are robust to a per-protocol outcome analysis and are sensitive to the cost of blood transfusions. Conclusions: Early NASG application at the PHC for women in hypovolemic shock has the potential to be cost-effective across many clinical settings. The NASG is designed to reverse shock and decrease further bleeding for women with obstetric hemorrhage; therefore, women who have received the NASG earlier may be better able to survive delays in reaching definitive care at the RH and recover more quickly from shock, all at a cost that is highly acceptable. The clinical data for this study was approved by the institutional review boards (IRBs) affiliated with the following institutions: University of California, San Francisco; University of Zambia, Lusaka; University of Zimbabwe-UCSF Collaborative Programme on Health Research; and the Department of Reproductive Health and Research of the World Health Organization. The cost data collected for this study did not involve human subjects, and was thus exempt from IRB oversight. This cost-effectiveness analysis builds on a previously reported clinical trial, summarized here. The clinical trial data belongs to UCSF and is freely available with a UCSF data sharing agreement. The cluster-randomized controlled trial of 38 PHCs in Zimbabwe and Zambia enrolled patients from 2009 to 2012. Eligible PHCs were peri-urban with at least 500 annual deliveries that referred obstetric hemorrhage (OH) cases (≥500 mL blood loss) to one of five study regional hospitals. Participants were admitted at the PHC and were consenting women with any obstetric hemorrhage etiology and hypovolemic shock. Women with antepartum hemorrhage with a viable fetus were excluded. PHCs were randomized to either the early application or later application group using a covariate-constrained procedure to ensure balance across intervention arms on number of deliveries, number of deliveries per midwife, distance to referral hospital (RH), and proportion of OH cases expected [16]. Women who presented at the PHC at < 24 weeks gestation (ectopic pregnancy, complications of abortion, or trophoblastic/molar pregnancy) were enrolled in the outpatient department, while women at ≥ 24 weeks were enrolled in the maternity department. Midwives at the PHC were trained to treat PPH with uterotonics and IV fluids, repair first- and second-degree perineal lacerations, and refer any patient with estimated blood loss > 500 mL to the RH. PHCs were not equipped to provide blood transfusions, surgery, or manual vacuum aspiration (MVA). Each PHC had access to a shared ambulance system to transfer patients to the RH. All eligible women had a perineal pad applied at study entry in the PHC to measure blood loss. Women in the early application arm received the NASG (Zoex Corporation, Coloma, CA 95613, USA) at the PHC and women in the later application arm received it at the RH per treatment protocol. All women were referred to the RH and were transported by ambulance, private vehicle, or taxi. Oxygen, IV fluids, uterotonics or uterine massage for uterine atony, suturing of lacerations, manual removal of placenta or retained tissues, MVA, surgery, and blood transfusions were available as needed at the RH. More detailed information regarding the design of the CRCT is provided elsewhere [5]. We use a per-protocol analysis [17]. Characteristics of women were similar between early and later NASG application groups (see Table 1) except for hemorrhage etiology. The early application group was composed of a higher proportion of women with uterine atony (42.1% vs. 28.7%) and a lower proportion with complications of abortion (15.6% vs. 36.2%) compared to the later application group. Per-protocol study characteristics ***p < 0.001, **p < 0.01, *p < 0.05. Note: Wilcoxon Rank Sum test utilized to test all continuous variables due to non-normality. Chi-square test used for categorical values except where noted. Disability-adjusted life years (DALYs) without age-weighting were used to quantify the burden of disease as a discounted sum of the number of years of life lost (YL) from early death and years lost due to disability (YLD) [18]. The timeframe of this analysis was the four-year period of the intervention. Disabilities over the women’s life were considered. YL was calculated as the difference between the woman’s age and her age-adjusted life expectancy within her country of residence for those women who died during the study. YLD was constructed as a composite of the morbidities for each woman who survived. This includes acute renal failure, acute respiratory distress syndrome, heart failure, cerebral impairment (seizures, unconsciousness, motor/cognitive loss), and severe anemia. The rate of severe anemia was defined as hemoglobin value less than 7.0 g/dL at hospital discharge. There was no evidence of statistically significant differences between earlier and later NASG application across mortality and morbidity outcomes. The odds of death in the early application group were 64% lower (OR 0.36 (95% CI: 0.08 – 1.53) than the later application group (Table 2). There were no morbidities in the early application group and 0.2% in the later application group. There was no statistically significant difference in severe anemia at discharge between groups. As morbidities and mortalities were rare, there may have not been adequate statistical power to detect an effect [5]. There was no statistically significant difference between minutes from study entry to death or study exit between the two groups; however, women in the early application group recovered from shock at a significantly faster rate, 165 min for early application vs. 209 min for later application (OR 1.28 (95% CI: 1.05-1.57). On average, the later application group received the NASG 2.5 hours after the early application group. Study treatments and outcomes αMedian (IQR); βHazard Ratio. We estimated costs using micro-costing methods. Resource use was estimated from clinical trial records. Unit costs were collected from pharmacies, blood banks, and hospital administrators in local currencies and converted into international dollars [19]. Costs of clinical resources at the PHC and RH were summed for each individual (Table 3). As the NASG was applied to both groups, only the timing differed; costs of the NASG (material/cleaning/training) were estimated and described below for reference only. Unit costs by country, 2010 (IU)1 1International dollars. 2Cost is amortized over 72 uses and includes cleaning. 3Averaged across countries and includes provider opportunity cost. 4Mean of first 2 units shown; actual costs in analysis are $135 for first unit and $90 per each additional unit. The material cost of the NASG per use was estimated as $1.04, based on an approximate price of $75 and an estimated life of 72 uses per garment (personal communication Neil McConnochie, BlueFuzion to Suellen Miller). The cost of cleaning the NASG included bleach, bucket for immersion, personal protection equipment, and personnel, and was estimated at $0.50 per use. The total estimate for cost of NASG was $1.54 per use. Training costs included transportation, facilities, materials, and personnel costs. The model assumes training has a 10-year life, which is conservative given that training is not designed to require a refresher. Estimates of training cost per patient were based on actual costs collected from one facility where two hundred participants attended a stand-alone (NASG only) training. The base-case estimate for the cost of training per NASG use was $1.62. Costs of uterotonics, oxytocin per ampoule (10 IU) and misoprostol per dose, were collected from hospital pharmacies and hospital administrators in one facility in Zambia and one facility in Zimbabwe. Cost of blood transfusions was based on cost per unit of blood in each country during the study period. The cost of one unit of blood and uterotonics in Zimbabwe were significantly higher than in Zambia. In Zimbabwe blood cost $135 for the first unit and $90 per each additional unit, compared to $42 per unit in Zambia. Blood was not always available during the study period. Emergency hysterectomy (EH) costs for complications due to intractable uterine atony and complications of abortion were collected. Costs include personnel, equipment, anesthesia, and operating room costs. Emergency hysterectomies were conservatively estimated to require 6 personnel over 60 to 90 minutes. No other surgeries were included in this analysis, as etiologies differed, and some etiologies require surgery (ruptured uterus, ruptured ectopic pregnancy). We estimated a series of models using Stata 13.0 (StataCorp, College Station, TX, USA). We anticipated that variances would differ across clusters due to variations in adherence to treatment protocol; we include random effects to allow individual-level differences to vary across clusters [20]. The general models for the random effects specifications are as follows: where Eic is the probability of a disability-adjusted life year of individual i in cluster c and Cic is the costs of individual i in cluster c; Jc is an intervention indicator for cluster c where j = 1 for early application group and j = 0 for later application group; Zimc is an indicator for Zimbabwe (1 = Zimbabwe, 0 = Zambia) for each cluster c; and uic is the error term. In order to determine Eic, the probability of a disability-adjusted life year of individual i in cluster c, we specified 4 random intercept models. In Model 1, the model was specified as above but excluded the country indicator. In Model 2, we added the country indicator. In Model 3, we estimated an interaction term between country indicator Zimc and early application indicator Jc to understand country-specific effects of early application. In Model 4, we omitted the country indicator and instead used 4 referral hospital indicators, RHc, where the Lusaka hospital was the reference. We re-specified Model 4 for Zambia only (Model 4a) to improve statistical power as 13 of the 15 deaths occurred in Zambia. Models 5-8a were specified identical to Models 1-4a substituting cost of individual i in cluster c for outcome. We used a likelihood ratio test to compare model fit for Models 1–4 and 5–8. We compared the effectiveness and costs between the earlier and later NASG application groups to calculate the incremental cost-effectiveness ratio (ICER) [21]. The ICER is the difference between the costs and effectiveness of the groups, given by δ1/β1. We conducted a sensitivity analysis by simulating probabilistic clinical resource costs and mortalities to provide insight to their contribution to the ICER. We varied the unit cost of blood from $20 to $200 while keeping the blood transfusion rate (number of units per individual) and all other variables constant to reflect the probable range of costs of blood within sub-Saharan Africa. We also assessed how varying the relative odds ratio of death given timing of NASG application would impact the ICER, and whether the results from our primary models were consistent when stratifying by severe shock at study entry, defined by mean arterial pressure <60 mm Hg.