Introduction

Background

Rwanda’s community-based health insurance (CBHI) scheme covers more than 80% of the population, most of whom are in the informal sector [1]. The scheme has three main funding sources: member contributions, government subsidies, and donors, and operates mostly on a fee-for-service basis [2]. Members are entitled to a comprehensive benefits package covering drugs and medical services, and their contributions vary based on income level [3]. Co-payments are 200 Rwandan francs (RWF) at the health post level, and 10% of the total bill at higher levels of care [3]. As part of the scheme’s success, the government continues to face growing demand for a wide range of healthcare services, which it must balance with an estimated $58 per capita expenditure on health [4].

To strengthen the financial sustainability of the CBHI scheme, health technology assessment (HTA) is being introduced to support explicit, evidence-informed priority setting [5]. As a first step, a rapid cost-effectiveness analysis on dialysis for acute kidney injury (AKI) was undertaken [6].

Dialysis is a common topic of interest for low- and middle-income countries (LMICs) facing a growing burden of non-communicable diseases due to its high costs [7]. Across LMICs, only 2–5% of patients needing treatment receive it; for many, it is unaffordable [8]. At the time of analysis, six weeks of dialysis was covered by CBHI in Rwanda with a 10% co-pay, averaging 218,000 RWF out of pocket per patient [9]. For scale, this represents 25% of GDP per capita [10].

Some LMICs have conducted cost-effectiveness analyses on dialysis to inform their coverage decisions [11, 12]. However, these have been disproportionately focused on dialysis for patients with end-stage renal disease (ESRD) [13,14,15,16]. ESRD is the last stage of chronic kidney disease (CKD), which permanently impairs kidney function and renders patients ‘dialysis dependent’ to survive.

Dialysis is also used to treat AKI, which, unlike ESRD, temporarily impairs kidney function. AKI is reversible if diagnosed and treated early. Depending on the severity of a kidney injury, patients may only need dialysis for a limited period to allow for at least partial, and sometimes full recovery of kidney function.

LMICs bear a disproportionate amount of the globally estimated burden of AKI [17], and in these countries, it is commonly a disease of the young, often caused by a single, curable condition [17,18,19,20]. In Rwanda, the most common comorbidities include malaria; pneumonia; sepsis; pregnancy-related conditions such as eclampsia and hypertension; intoxication caused by treatment from traditional healers; and diabetes [21]. The median age for AKI patients in Rwanda is 38 years old, and the mortality rate is 34% [21]. However, barriers to optimal management of AKI in Rwanda remain. These include knowledge gaps among healthcare providers, sub-optimal diagnostic capacity, particularly in sub-tertiary hospitals, and limited treatment options [22].

The main treatment options for AKI in Rwanda are hemodialysis (HD) and peritoneal dialysis (PD). Evidence suggests little difference when comparing HD and PD in terms of their clinical outcomes or the risk of complications – though the evidence base remains moderate to poor [23].

Currently, all dialysis provision in Rwanda is exclusively delivered in the hospital setting. Most of this provision is HD, with a small proportion being PD. While in other settings, HD is often provided in hospitals and PD in smaller facilities or at home this type of PD was discontinued a few years ago in Rwanda. This was partially because of challenges in sourcing dialysate and difficulty in guaranteeing hygienic conditions for at-home PD.

Aim and objectives

At the time of writing, Rwanda’s CBHI benefits package officially covered up to six weeks of dialysis per patient with AKI. Dialysis for ESRD was not covered. However, the diagnosis of AKI versus CKD can sometimes be challenging, especially when there is a previously undiagnosed kidney dysfunction. Due to the considerable cost of providing dialysis, this study aimed to help the Rwandan CBHI scheme decide on the optimal delivery of dialysis services.

The objectives of the study were to rapidly assess (1) the cost-effectiveness and (2) the budget impact of providing PD versus HD for patients with AKI in the tertiary care setting in Rwanda. This may be the first study of its kind comparing dialysis modalities for AKI in LMICs where both HD and PD are provided exclusively at the tertiary care level.

Methods

Our rapid cost-effectiveness analysis used an ‘adaptive’ HTA (aHTA) approach, which adjusts HTA methods for time, data, and capacity constraints [24]. To respond to policy makers’ demand, the aim was to complete the assessment in six weeks. It was thus decided to use a rapid cost-effectiveness analysis which builds basic economic models using opportunistically or rapidly sourced local data [24]. The assessment used the international decision support initiative (iDSI) reference case for economic evaluation as a guide [6, 25]. Table 1 summarizes the application of the eleven iDSI reference case principles. Highlighted rows indicate principles that were adapted for time and data constraints.

Table 1 Methodological approach using the iDSI reference case
Full size table

Population and subgroups

The population for this analysis were dialysis-eligible patients with AKI in a tertiary care facility in Rwanda. Two age groups were considered: patients aged 15–49 and patients aged 50 and above.

Comparators

The intervention was tertiary care delivered PD, compared with tertiary care delivered HD. In the base case, the model assumed that patients receive the maximum allotted care covered by CBHI. This included three sessions per week for six weeks of HD or six weeks of continuous ambulatory PD for hospitalized patients.

Model structure and assumptions

A de novo Markov model was used to reflect the costs and effects of the initial acute condition of AKI combined with the long-term health effects that can follow the condition. The cycle length was one year (Appendix 1). The model was informed by the published literature and validated through consultation with local and international nephrology specialists. Patients enter the model at the tertiary care facility, starting on hospital HD or hospital PD. Over time, patients may stay with the same modality or switch modalities. They may develop complications or not, and subsequently fully recover rendering them dialysis independent or partially recover with no further treatment. They may then die from AKI progressing to CKD, from co-morbidities, or from natural causes.

Modelling perspective and scenarios

A payer perspective was used. The payer perspective included all direct medical costs to Rwanda Social Security Board (RSSB) plus salaries, overhead, and depreciation of the HD machines paid by the Ministry of Health. A ‘decreased provision’ scenario was also explored, which assumed the actual number of sessions patients receive on average was five instead of the full eighteen sessions [9].

Evidence for model parameters

Given an initially short timeline, a pragmatic approach was taken to select data to inform the model. Available local data was supplemented by a rapid literature search and sources known to the authors (Appendix 2). Where needed, gaps were addressed based on personal communication involving Rwandan nephrologists and international experts.

Cost and resource use

Costs and resource use data were sourced primarily from a 2018 RSSB Utilization and Expenditure Review on Dialysis made available by RSSB [9]. These were supplemented by published data from other jurisdictions and assumptions made by co-authors on this study. The total cost for HD, PD, and palliative care are expressed as per patient unit costs and reflect the cost of providing HD and PD at the tertiary care level (Table 2). Direct medical costs (including catheters, drugs, lab tests, kits, dialysate, other consumables, and palliative care) were sourced from an average across four facilities for HD and one facility for PD. Direct non-medical costs (costs of healthcare professionals, overheads, and depreciation) were estimated by combining local reports, peer-reviewed literature, and personal communication and allocated per patient based on patient volumes. Only one facility provides a minimal amount of PD, and thus there remain uncertainties regarding the PD unit costs. All costs are incurred during the six weeks of treatment; no additional costs of complications were included. For further details of costs included, see Appendix 3.

Table 2 Input parameters
Full size table

Costs in the model are expressed in RWF and are inflated to 2022 prices using the Consumer Price Index [26]. They are converted to current US dollars using the latest available exchange rate of 1 United States dollar (US$):1030 RWF [27].

Effectiveness

Clinical effectiveness parameters were drawn from several sources. Population mortality rates were sourced from the World Health Organization’s Global Health Observatory Data repository[28]. Mortality rates for AKI patients undergoing dialysis were from a local observational study [21]. Transition probabilities were from an Indonesian study that compared HD and PD for patients with ESRD [14], and on assumptions made by co-authors of this study.

Quality-adjusted life years (QALYs) were the primary health outcome in this study. Utility values for dialysis patients with AKI were sourced from two studies from Argentina and Canada, which used the EuroQoL EQ-5D-3L [29, 30]. Utility values for dialysis independence was sourced from Garay et al. [30].

Discounting

Costs and outcomes were discounted at a standard 3% per annum after the first year following the iDSI Reference Case, though most costs are incurred in the first year[25]. The impact of varying the discount rate between 0 to 5% was explored in a sensitivity analysis [31].

Thresholds

The base case analysis uses a cost-effectiveness threshold of 0.5 × GDP per capita (RWF 444,074 or US$431). This is broadly in line with recently estimated values for Rwanda based on cross-country studies of US$325 to US$426 (2022), or 39%–51% of GDP per capita[32,33,34].

Analyses

The cost-effectiveness analysis was completed in TreeAge software (version 2023 R1.2), and the budget impact analysis in Microsoft Excel. Uncertainty was analyzed using one-way sensitivity analysis and probabilistic sensitivity analysis (PSA), with distributions set according to standard practice for different parameter types.

Results

Base-case results

Overall, the intervention (PD) was less expensive and less effective relative to the comparator (HD). Table 3 presents the incremental costs and QALYs for the intervention (PD) and status quo (HD), stratified by age. The total estimated per patient cost for PD was RWF 1,824,886 (US$1,771) compared with a total estimated cost of HD of RWF 2,059,354 (US$1,999). The expected net QALYs lost in delivering PD compared with HD were -0.62 for patients aged 15–49 and -0.27 for those over 50, the latter due to the older cohort’s increased mortality rate.

Table 3 Incremental cost-effectiveness analysis
Full size table

At a threshold of 0.5 × GDP per capita (RWF 444,074 or US$431), the analysis suggests that HD provision, as the standard of care, was cost-effective compared with PD provision for patients aged 15–49, with an ICER below the threshold at RWF 378,174 (US$367). Notably, the interpretation of the ICER (Table 3) is reversed, because incremental costs and effects are both negative[35]. In other words, the ICER falls below the threshold, and thus the comparator (HD) is considered cost-effective.

For patients above 50, the analysis suggests that PD was the preferred option compared with HD with an ICER of RWF 868,399 (US$843). The same reverse interpretation of the ICER also applies to this scenario; as the ICER is above the threshold, PD is the preferred option (Table 3).

One-way sensitivity analysis

One-way deterministic sensitivity analysis was applied to individual parameters that affected the ICER most. Varying the costs of HD kits, HD commodities, HD salaries, and HD overhead by ± 30% increased the ICER when the cost of each parameter increased, and decreased the ICER when the cost of each parameter decreased. When varying the cost of PD dialysate, PD commodities, PD salaries, and PD overhead by ± 30%, increasing the costs of the parameters decreased the ICER, and decreasing the costs increased the ICER. In other words, this suggests that there may be opportunities to reduce HD-related costs and enhance the favorability of the ICER; but the same is not true for PD. See Appendix 4 for a tornado diagram.

Scenario analysis

In the reduced provision scenario, HD appeared to be cost-effective compared with PD. Again, for both age groups, the ICER falls below the threshold (Table 4).

Table 4 Scenario analysis
Full size table

Probabilistic sensitivity analysis

PSA was used to estimate the joint impact of uncertainty in all input parameters. Gamma distributions were applied to costs and beta distributions to health utilities (Table 2). By randomly sampling from each parameter distribution, 10,000 Monte Carlo simulations of incremental costs and incremental effects were obtained. The results of the PSA are presented in two figures. An incremental cost-effectiveness scatterplot (Fig. 1) and a cost-effectiveness acceptability curve (CEAC) (Fig. 2), which both summarize the impact of uncertainty in relation to the threshold[36]. At the threshold of 0.5 × GDP (RWF 444,074 or US$431) and above, HD provision has a 56% probability of being cost-effective relative to PD.

Fig. 1
figure 1

Incremental cost-effectiveness scatterplot

Full size image
Fig. 2
figure 2

Cost-effectiveness acceptability curve

Full size image

Budget impact analysis

The budget impact analysis rapidly assessed the five-year (2020–2024) costs associated with providing dialysis services in four scenarios. In the baseline scenario, ‘HD preferred,’ a stable distribution of HD (91%) to PD (9%) was maintained (i.e., status quo). Alternative scenarios included scenario 1–5% annual shift to PD provision over 5 years; scenario 2–10% annual shift to PD provision over 5 years; and scenario 3—HD provision maintained at 91% but with reduced costs for HD kits. Compared to an ‘HD preferred’ baseline scenario, shifting to PD coverage would generate some savings. Maintaining an HD-preferred strategy and decreasing the cost of HD kits could achieve significantly more savings because HD kits represent more than half of the overall baseline cost of dialysis (Table 5).

Table 5 Budget impact results by scenario
Full size table

Discussion

At a threshold of 0.5 × GDP per capita, the analysis suggests that HD is cost-effective compared with PD for most AKI patients receiving dialysis in Rwanda, i.e., those aged 15–49 years. The budget impact analysis suggests that shifting to PD would cost less than maintaining the status quo over five years. However, it also suggests that maintaining the HD status quo and decreasing the cost of the HD kit, which is a major cost driver in providing HD, would save even more than shifting to PD. A reduction in the cost of the kit could also reduce the overall co-pay for the patient.

One-way sensitivity analysis suggests that decreasing the cost of HD commodities and kits decreases the ICER. This may be achievable, as one hospital, Rwanda Military Hospital (RMH), procures HD kits directly from a local supplier and pays about half the price of those procured for other hospitals providing dialysis [9]. If all hospitals were to get the RMH price for kits, the ICER would decrease to below the threshold for all ages, and HD would be more cost-effective relative to PD. Indeed, this information has led to ongoing price negotiations for the kits for all facilities.

This study contributes to a sparse literature on dialysis for AKI. A recent systematic review published after the time of analysis identified only seven other studies on dialysis for patients with AKI, with mixed results and a recent increase in industry-sponsored studies [37]. Our study is thus a valuable, independent contributor to this sparse literature.

From a cost-effectiveness perspective, this analysis should be seen only as a starting point for discussion rather than a policy recommendation. Indeed, results of the PSA illustrate that HD is slightly more cost-effective at the threshold compared with PD, but there is still uncertainty. Moreover, policymakers raise several issues that the analysis cannot address. These include the cost or requirements for changing or expanding services; the cost-effectiveness of service delivery at lower levels of care; and the impact of removing patient co-pays.

Additionally, since the time of analysis, dialysis policies have changed in Rwanda. More coverage of dialysis is now available, as is kidney transplantation. Our analysis reflects the coverage at the time of analysis, and further analyses could be conducted to reflect current available health services.

Limitations

Our rapid cost-effectiveness analysis has several important limitations.

The study team’s approach to data collection was largely pragmatic, given time constraints. No attempts were made to synthesize the evidence for input parameters quantitatively or to systematically quality assure the data using available checklists. Data for the model came from several sources focused on patients with AKI where possible and supplemented by studies on CKD, author assumptions, and personal communication. Utility values, survival data, and transition probabilities are from various international sources. Importantly, the age of patients in papers from which utility values were sourced ranged from 45-to 65, while the average Rwandan dialysis patient is 38, and thus, utilities were overestimated. Local costing data were valuable in contextualizing the study. However, they were more focused on HD due to limited provision of PD, and they excluded the cost of infrastructure, overhead, and staff time for both HD or PD [9]. Other local reports, co-author assumptions and personal communication were used to fill data gaps, including staff time and equipment costs. Overhead was estimated as a percentage of operating costs (overhead + staff). Uncertainty remains about the costs and resources needed for PD because the local secondary data used reflected sparse provision of PD in hospitals. If time had allowed, the study would have benefited from more detailed costing on PD.

While the aim was to complete the analysis within six weeks, ultimately the assessment took about three to four months to complete.

Generalizability

A few factors may limit this study’s generalizability. First, these results reflect an analysis of PD and HD delivered in a tertiary care setting due to current practice and data availability. If the intervention had been PD delivered at lower-level facilities or at home, as is often the case in other countries, the analysis may have found PD to be much more cost-effective[13]. Second, these results are based on a time and data-constrained analysis that pragmatically sourced local and international data. This increased the chance of uncertainty and bias. Our findings may have limited generalizability to other contexts and should not be interpreted without caution alongside other studies on dialysis in LMICs. Other studies often focus on ESRD patients who can either get HD in hospital or PD in lower-level facilities or at home, and often conclude that a PD-first policy is preferable[13,14,15,16]. If policymakers in Rwanda were considering coverage options for patients with ESRD or lower-level provision of PD, a separate cost-effectiveness analysis and budget impact analysis would need to be undertaken to understand the implications of the new policy choice.

Reflections on ‘adaptive’ HTA

This cost-effectiveness analysis undertook an aHTA approach by deviating from what may be regarded as the ‘gold standard’ of HTA. This was done to reflect the local policymaker context, the availability of data, and general practicality constraints. The iDSI reference case served as a crucial principles-based framework to explore the suitability of the present analysis. Strategic choices were made on how to deliver evidence given the constraints, in a way that was still fit for policy makers’ purposes. This was done in two ways.

First, policymakers sought to conduct a rapid HTA to have 'proof of concept' for using HTA in decision-making. To conduct the analysis quickly, a topic was selected for which there were locally available cost data and supplemented by a pragmatic approach to collecting additional data as described in the limitations above. The implication of these choices naturally have impacts on the generalizability and potential bias of the analysis.

Second, the choice to exclude a 'no comparator' arm was made to reflect the local context. Dialysis is a hotly debated topic everywhere due to its high cost, and for ESRD, its limited effectiveness. From a purely economic perspective, some have argued that dialysis is an inefficient use of resources better spent elsewhere [38]. However, dialysis is a good illustration that priority-setting choices are not limited to cost-effectiveness. Dialysis is provided in many countries, including LMICs, on the grounds of financial risk protection and it being a moral imperative for universal health coverage [39]. The reality is that many LMICs have a shifting burden of disease, with existing coverage for dialysis services in LMICs being described as inadequate [40]. There is a real need to provide evidence to inform open debates about the optimal solution to providing dialysis.

Thus, the 'adaptive' choices in methodology made by the co-authors for this paper reflect a conscious effort to address policymaker needs in a deliberate departure from ‘gold standard’ HTA approaches. The pilot successfully raised awareness about HTA among key stakeholders, provided evidence for price negotiations, and identified key data needs that should be considered part of a strategy to support HTA development in the country [41, 42].

Conclusion

Our de novo model suggests that HD may be cost-effective from the payer perspective compared with PD, and significant cost savings may be achieved by reducing the costs of HD commodities. While the relative robustness of this economic evaluation was constrained by adopting an aHTA approach, it was nonetheless a useful policy tool for Rwandan policymakers as it helped build the foundation for evidence-based priority setting in the future.