Cost-effectiveness of birth-dose hepatitis B vaccination among refugee populations in the African region: a series of case studies

We examined African countries with existing national HepB BD vaccination policy, and then selected countries hosting large and protracted refugee populations in established camps [13]. We calculated the estimated annual birth cohort for each identified refugee population by multiplying the camp size by the crude birth rate from the country of origin. For each population, we compared life-years lost based on our decision tree model and the cost for each of three strategies for implementation of HepB BD: RI, RI plus universal HepB BD, and RI plus HepB BD provided only for newborns of HBsAg-positive mothers identified through RDT for HBsAg (RDT-directed HepB BD plus RI).

We adapted the disease-specific decision-tree models of Lu, et al. and Tu, et al. to estimate the impact and cost-effectiveness of implementing HepB BD by comparing the cost of adding that intervention with life-years lost due to complications of chronic hepatitis B in the refugee camp setting [14, 15]. We constructed the model using TreeAge Pro 2017 (Williamstown, Massachusetts, USA). Variables considered in the analysis are listed in Table 1. The model inputs include mortality due to either HBV-associated cirrhosis or hepatocellular carcinoma (HCC); we did not include fulminant HBV infection at time of acquisition. Additional variables included combined RI and HepB BD vaccine coverage and efficacy, and cost (administrative vaccine purchase and HBsAg RDT). We incorporated population-specific values for vaccine coverage in host country, HBsAg prevalence and fertility rate (but substituting Morocco for Western Sahara due to lack of data), and camp size for refugee camps in Djibouti, Mauritania and Algeria using published figures as shown in Table 1 [10, 12, 16‐20]. We did not consider direct medical costs for other complications of hepatitis B because access to medical care is limited in these settings and the likelihood of obtaining prolonged treatment for complications is low.

Our decision tree simulated the vaccination strategies, using maternal HBsAg rate (but substituting Morocco for Western Sahara due to lack of data) as a proxy for HBV prevalence, and calculating probabilities of infant infection based on vaccination coverage and efficacy (Fig. 1). We used a Markov cohort model, cycling yearly, to model hepatitis B progression (Fig. 2). The analytic horizon was from birth to death. We used a global discounting rate of 3% [21]. We modeled four states, with HBV inactive carrier status and chronic active hepatitis B moving between each other at fixed probabilities, chronic hepatitis capable of progressing to compensated cirrhosis or HCC, and compensated cirrhosis able to progress to decompensated cirrhosis or HCC. Chronic hepatitis is defined as persistent hepatic inflammation, whereas cirrhosis is defined as liver fibrosis affecting the liver’s synthetic functioning [22]. Compensated cirrhosis is defined as fibrosis with liver dysfunction that does not substantially impair the individual’s daily activity, whereas decompensated cirrhosis encompasses gastrointestinal bleeding, encephalopathy and massive ascites that are clinically apparent and usually fatal in the short term [23]. Transition probabilities of movement from one state to another during hepatitis B infection were based on published literature [24‐27].

For birth rate, life expectancy and HBSAg prevalence, we used the country of origin of the refugees, except for life expectancy from Western Sahara because the measure was not available; the estimate for Morocco was substituted. We obtained life expectancies from the World Bank and age-specific background mortality rates and vaccine coverage data from WHO tables [16, 28, 29].

To approximate the vaccine efficacy of RI plus universal HepB BD versus RI alone, we applied figures from a retrospective study of vaccine efficacy for preventing vertical transmission of HBV infection in Canada, comparing HBsAg prevalence rates among children who were vaccinated within 3 days of birth to those who were vaccinated more than 8 days after birth [30].

We used UNICEF prices for multi-dose monovalent hepatitis B vaccine (HepB BD), and program costs (vaccine purchase, injection supplies, vaccine distribution and service delivery) from a similar study in the Gambia, updating for inflation [26, 31]. We chose cost of the highly sensitive and specific Alere Determine™ HBsAg test (Waltham, MA), widely available in Africa, to represent HBsAg RDT [32].

We assumed RI costs in all countries were from government perspective. Costs of implementing HepB BD delivery were therefore calculated from the public payer perspective (i.e., the cost to the government to procure and deliver the vaccine).

Our model has several key assumptions.

On the public health level, we assumed that infants would receive HepB BD vaccination at the same rate as RI and all receiving HepB BD would go on to receive RI, as they receive health services within the camp setting. We varied this assumption by a margin of - 25% of estimate to 100% in the sensitivity analysis [33]. We assumed that the operational costs for performing RDT were similar to the operational costs for delivery of HepB BD vaccine as they both require trained fieldworkers, supplies and transport, and we varied this assumption in the sensitivity analysis as exact operational costs for RDT are not available. We assumed that external stressors are approximately equal between camps as all refugee camps in this analysis are supervised by the United Nations High Commissioner for Refugees. We used a residual susceptibility rate of 5% for likelihood of acquiring hepatitis B horizontally in the event of immunologic failure [34].

For disease-specific variables, we made several simplifying assumptions about disease progression to facilitate demonstration of the model. We assumed that all mother-to-child HBV transmission happened in the perinatal period, as prior studies have shown that later vertical transmission is less than or equal to 1% [35, 36]. We did not calculate disability-adjusted life years (DALYs) lost or incidental costs to families or communities as the progression from fulminant disease (decompensated cirrhosis or HCC) to death is likely rapid (within one year) in the absence of dedicated health services as shown in the literature [37, 38]. A recent study supported this assumption, finding that globally, years of life lost due to hepatitis make up 98% of the burden of the disease encapsulated by disability-adjusted life years (DALYs) [39].

We assumed that infants did not die from fulminant hepatitis B, as the rate of death from perinatal hepatitis is low (less than 1 in 1000) [40]. We used overall HBV transmission rates from an HBV-infected mother (both hepatitis B e antigen-positive and -negative) to her child from the literature [24]. We also did not include the effect of co-infection with HIV as reported HIV rates in the studied areas are low enough (0.1 to 1.6%) to not significantly alter the results of our model [41].

We compared number of life-years lost from HBV attributable HCC- and cirrhosis-related premature death among the RI, RI plus universal HepB BD, and RDT-directed HepB BD plus RI scenarios and calculated cost per life year saved and incremental cost-effectiveness ratio (ICER) under each scenario.

We conducted one-way sensitivity analyses, varying the parameters one by one with the range of values shown in Table 1. To estimate the prevalence of HBsAg prevalence in refugee populations, we used sources from the published literature [42] and examined sensitivity around these values.

To determine the sensitivity around estimates of stochastic variables, we performed a second-order Monte Carlo simulation with 1000 iterations. We varied vaccination coverage, probability of HBV carriers transitioning to chronic hepatitis, cirrhosis and HCC, and the cost of operations according to triangular distributions, with the lower and upper bounds of vaccination coverage representing the worst and best performing countries in the African region [43]. We varied the transition to complications probability bounds by 25%, and varied the operational cost, with the lower bound representing the lowest estimate for the percentage of operational cost relative to vaccine cost determined by the Expanded Program on Immunization Costing study [44], and the upper bound of operational cost varied upward to 2 USD to account for vaccine deployment during rainy season or extreme condition. We assumed a fixed cost for the vaccine and RDT.

The impact of HepB BD vaccination was similar across multiple camp situations; it would save 506 life-years per 10,000 refugees per year among refugees in Djibouti, 126 life-years per 10,000 refugees per year among refugees in Algeria, and 2204 life-years per 10,000 refugees per year among refugees in Mauritania (Table 2).

The Monte Carlo simulations showed a trend toward increased effectiveness as cost increased, but a relative ceiling of effectiveness was seen despite increasing cost when the proportion of children receiving RI approached 1.

In each case, the cost per year of life saved by addition of HepB BD is much less than the average gross domestic product (GDP) of the host country (0.83 USD versus 3342 USD in Djibouti, 1.59 USD versus 14,613 USD in Algeria and 0.18 USD versus 3835 USD in Mauritania) [45]. In all countries, RDT-directed HepB plus RI was less cost-effective than RI alone or RI plus universal HepB BD.

Using triangular probability distributions, we found that none of the confidence intervals around transition probabilities in the Markov model affected the final cost, effectiveness, or ICER by more than 2% in any analysis. In sensitivity analyses, the baseline vaccination rate had the greatest impact on net health benefit. For each country, net life years gained varied across a range of 2.1 to 2.4 years when considering a − 25% of estimate to 100% in baseline vaccination rate. Variance in protection of RI plus universal HepB BD against HBV infection was the second greatest contributor to uncertainty in net benefits, given the uncertainty of protection from prior studies [30]. The ICER was more favorable for RI plus universal HepB BD than for a test and treat strategy (Fig. 3). The incremental cost-effectiveness of RI plus universal HepB BD over RI alone was most affected by the probability that RI is protective against hepatitis B. The greatest variability in net benefits upon sensitivity analysis was seen in the Mauritanian refugee population (Fig. 4). Sensitivity analyses were similar for all countries.

In the Monte Carlo simulation, we observed overlap in both cost and effectiveness for all three approaches, although RI plus universal HepB BD was both more effective and more cost-effective in terms of dollars spent per year of life saved than either RI alone or RDT-directed HepB plus RI under most scenarios. All approaches remained very cost-effective under all simulated scenarios.

This analysis has several limitations. Firstly, our probability point estimates for disease modelling and cost were drawn from existing literature and may not necessarily reflect the populations under study. We used a static model that does not incorporate horizontal transmission of HBV, which limits our ability to calculate life-years saved through protection from post-perinatal transmission. However, the benefits of RI to protect against horizontal transmission of HBV are already well-known [52]. We assumed that all refugees receiving HepB BD vaccination would receive RI, which may be interrupted in unstable situations. It is also possible that refugees may not accept vaccination due to cultural reasons or the perception that complications of hepatitis are too far in the future to warrant preventive measures now. We attempted to account for this possibility through the sensitivity analysis around actual vaccination rates in our model. [33], which is why we used the lower of the two estimates of vaccine coverage in order to generate a more conservative model.

We made several simplifying assumptions about the natural history of HBV infection, including a 100% one-year mortality rate for patients diagnosed with HCC and presuming that HCC is non-treatable [53]. The survival rates of patients with decompensated cirrhosis and HCC are unknown among populations with limited access to medical care. We did not include disability adjustment because we assume that there is rapid progression to death in the refugee population.