Cost-effectiveness of nirmatrelvir/ritonavir in COVID-19 patient groups at high risk for progression to severe COVID-19 in the Netherlands

Abstract

Background

Nirmatrelvir/ritonavir is indicated for the treatment of COVID-19 in symptomatic adults with increased risk for severe illness, not requiring supplemental oxygen yet. From a Dutch societal perspective, a cost-utility assessment of nirmatrelvir/ritonavir versus best supportive care (BSC) was conducted in three patient groups: (a) immunocompromised patients, (b) patients aged at least 60 years with one comorbidity, (c) patients aged at least 70 years. Groups were selected considering their relevance as high-risk groups, as described in Dutch and international guidelines and recommendations.

Methods

A one-year decision-tree, estimating costs and outcomes associated with a COVID-19 infection was coupled to a lifetime two-state Markov component simulating subsequent life-time survival and quality of life. Effectiveness estimates, informing the intervention preventing hospital admission or death, were based on real-world evidence by Lewnard and colleagues (2023) in a vaccinated population during a timeframe with predominance of the Omicron variant. Epidemiology relies on publicly available data, primarily sourced during the Omicron variant’s era. In the decision tree, clinically relevant event-related disutilities per disease course were applied to adjusted age-dependent Dutch-specific utility levels. In the Markov component, a disutility was considered for post-ICU patients. Costs rely on Dutch pharmacoeconomic guidelines and public data sources. The incremental cost-effectiveness ratio (ICER) was analysed as the main outcome, with a positive ICER indicating the cost associated with each additional quality-adjusted life year (QALY) gained by adopting the intervention.

Results

Nirmatrelvir/ritonavir was associated with an ICER of € 395 in the immunocompromised group (per patient: + 0.125 QALYs gained; + 0.130 life-years [LYs] gained; € 49 incremental cost), with an ICER of € 8700 in 60-plus patients with comorbidity (+ 0.054 QALYs; + 0.055 LYs; € 474 incremental cost), and with an ICER of € 13,021 among 70-plus patients (+ 0.053 QALYs; + 0.045 LYs; € 689 incremental cost). Results were most sensitive to the baseline hospitalization rates among high-risk individuals. Probabilistic sensitivity analyses indicate a high probability of being cost-effective (100, 94, 85% respectively), considering a willingness-to-pay threshold of € 20,000 per QALY.

Conclusions

From a Dutch societal perspective, over a lifetime horizon, nirmatrelvir/ritonavir is cost-effective versus BSC in the three groups analysed.

Background

The coronavirus disease 2019 (COVID-19) pandemic has had a substantial impact on the survival and quality of life of patients in the Netherlands [1]. The COVID-19 pandemic disproportionally affected the more vulnerable population groups, particularly those with increased age, chronic health conditions and lower incomes [1]. Associated with a wide variety of complications and clinical manifestations persisting for many weeks and even months, COVID-19 led to higher numbers of hospital and intensive care unit (ICU) admissions, adding pressure on national health systems in the Netherlands and around the world [2, 3].

COVID-19 related deaths in the general Dutch population decreased in 2023 compared with 2022, representing less than 2% of deaths as opposed to 5% in 2022 and to 13% in 2021. The roll-out of effective vaccines offering protection against COVID-19, together with the increased number of individuals that have had an infection with COVID-19, are considered to contribute to this favourable evolution. Among older adults (60-plus), COVID-19 mortality remained stable in 2023, whereas it decreased among other age groups [4]. This risk reduction, described in function of age groups within the general population, are expected to vary when focusing on specific populations defined by parameters other than age category (e.g. patients featuring comorbidity, or immunocompromised patients).

In the Netherlands, physicians use the SWAB-FMS guideline for the treatment of COVID-19 [5]. The SWAB-FMS guideline recommends treatment with nirmatrelvir/ritonavir only in the outpatient setting within 5 days of symptom onset, if (a) having a seriously reduced immune system regardless of vaccination status (e.g. patients with a haematological malignancy, after an organ transplant), or (b) being expected to be SARS-CoV-2 naive (unvaccinated and/or not having experienced COVID-19) while having an increased risk of serious course of disease as a result of comorbidities or conditions (for instance, vulnerable elderly patients or patients with severe cardiac or pulmonary conditions). Nirmatrelvir/ritonavir is currently the only treatment recommended within the SWAB-FMS guideline for ambulatory patients with COVID-19. Note that, the SWAB-FMS guideline explicitly states that treatment of hospitalized patients with nirmatrelvir/ritonavir is not recommended [5].

The efficacy and safety of nirmatrelvir/ritonavir was evaluated in the EPIC-HR trial in non-hospitalized, symptomatic adults with COVID-19, who were at high risk for progression to severe COVID-19 [6]. Patients treated with nirmatrelvir/ritonavir showed a significant relative risk reduction of hospitalization for COVID-19 or death from any cause through day 28 of 87.8%, compared with placebo (0.77% vs 6.31%; P < 0.001), together with a reduction in symptom days of 20% [25]. Given that the EPIC-HR trial was conducted during a period where the Delta variant was predominant in an unvaccinated population, it is useful to consider the current body of real-world evidence showing the safety and effectiveness of nirmatrelvir/ritonavir in the Omicron period with high vaccination rates [7,8,9,10,11,12]. Among those studies, only one large observational study, published by Lewnard and colleagues [7], was identified that reported the effectiveness of nirmatrelvir/ritonavir during the Omicron era in a vaccinated population while considering timing of symptom onset to treatment initiation (i.e. within 5 days of COVID-19 symptom onset). These eligibility criteria are in line with the label for nirmatrelvir/ritonavir from the European Medicines Agency (EMA) [62]. They reported, after adjustment for differences among treated and untreated people, that treatment with nirmatrelvir/ritonavir within 5 days of symptom onset was associated with a 79.6% (95% confidence interval [CI] 33.9–93.8%; p = 0.008) effectiveness in preventing hospital admission or death within 30 days [7]. Comparing with the efficacy reported in the EPIC-HR trial (i.e. 87.8%) [25], the estimates reported by Lewnard and colleagues are expected to lead to a more conservative analysis. Other estimates, not considering the provision of nirmatrelvir/ritonavir within 5 days of symptom onset, were available from singular observational studies and from studies synthesizing existing evidence. For instance, the study by Lewnard et al. [7] reported that provision of nirmatrelvir/ritonavir at any time (irrespective of the presence or timing of symptoms) was associated with a 53.6% (95% CI 6.6–77.0%) effectiveness in preventing hospital admission or death within 30 days [7]. A network meta-analysis conducted by Souza and colleagues based on 16 observational studies (where the duration of symptom onset or time of positive COVID-19 test to treatment initiation varied widely or was unavailable) reported a 59% reduction on risk of death (odds ratio [OR] = 0.41; 95% CI 0.35–0.52; moderate certainty of evidence) and a 53% reduction on risk of hospitalization (OR = 0.47; 95% CI 0.36–0.60, with very low certainty of evidence) [63].

There is an absence of published studies assessing the cost-effectiveness of nirmatrelvir/ritonavir in the Dutch setting. In the United States, an assessment covering several COVID-19 treatments, completed by the Institute for Clinical and Economic Review reported a cost-effective price benchmark range of $563–$868 per course [13]. More recently, a cost-utility analysis relying on a short-term decision tree structure coupled with a Markov trace, considering a cost per nirmatrelvir/ritonavir treatment course of $1,390, estimated an incremental cost-effectiveness ratio (ICER) of $8,931/QALY over lifetime with a 99% probability of being cost-effective when considering a $100,000 willingness-to-pay threshold [25].

The objective of this study was to assess the cost-effectiveness of nirmatrelvir/ritonavir in The Netherlands in patients at increased risk of severe disease.

Methods

This study consists of a cost-utility analysis of nirmatrelvir/ritonavir versus best supportive care (BSC) conducted in three patient groups that are deemed relevant for The Netherlands during the Omicron period and fit within the authorized indication. Such groups were selected considering their relevance as vulnerable and high-risk groups described in Dutch and international recommendations, such as those from the SWAB-FMS, the Joint Committee on Vaccination and Immunisation and the World Health Organization [15, 16].

The first group under analysis covers the immunocompromised patient population, as defined in appendix 1 of the SWAB-FMS guideline specific to patients with a very high risk of a serious course of COVID-19 [17]. This guideline targets patients with serious immune disorders associated or derived from: organ transplantation procedures, bone marrow or stem cell transplantation, malignant haematological diseases, haematological malignancies known to be associated with severe immunodeficiency, all cancer patients (solid tumours) who received chemotherapy and/or immune checkpoint inhibitors less than 3 months before their COVID-19 vaccinations, presence of primary immunodeficiency, and, immunosuppressant treatment with B-cell depleting medications or strongly lymphopenia-inducing medications [17]. Notably, the immunocompromised patient group is also listed by the Dutch National Institute for Public Health and the Environment (RIVM), as having a greater risk of serious progression of COVID-19 [18].

The second group covers patients aged 60 plus with one comorbidity (e.g. obesity, diabetes, chronic obstructive pulmonary disease, kidney or liver disease, cancer, disabilities). This group was of interest for this analysis, given that the presence of comorbidities further increases the risk of developing severe COVID-19, in addition to being of advanced age alone. A threshold age of 60 years old was considered for this group of analysis as it is listed by the RIVM as a feature defining populations at high risk to progress to severe COVID-19, and since analyses of epidemiologic data show increased rates of hospital admission and mortality rates in the Netherlands [18, 19]. Challenges in terms of data availability, made further stratification (within this second group) by number of comorbidities unfeasible.

The third group consist of patients aged 70 plus, irrespective of comorbidities or immunocompromised status. This group was of interest for this analysis, given that several Dutch guidelines feature this age threshold alone as an important risk factor to develop severe COVID-19, with further increases in age further increasing the risk [5, 20]. Note that the patient groups defined for the analysis are not expected to be mutually exclusive between them due to their definitions, thus, an overlap between the patient groups defined in group 1 (i.e. immunocompromised), group 2 (i.e. 60-plus patients with comorbidity) and group 3 (i.e. 70-plus patients) exists. Consider that, whereas the immunocompromised patient group is meant to reflect current Dutch eligibility criteria for nirmatrelvir/ritonavir (i.e. as per SWAB-FMS guideline), the other two groups in the analysis consider broader population groups.

A model, capturing an initial short-term period (1 year), as well as long-term evaluation, was developed in Microsoft Excel considering a lifetime horizon, and both Dutch societal and health care perspectives were applied. This analysis aligns with the reference case described in the Dutch guidelines on economic evaluations from 2016, which was applicable at the time this analysis was conducted [21]. A discount rate of 1.5% for health benefits was applied in accordance with the Dutch guideline for pharmacoeconomic analyses [21]. A discount rate for costs is not applicable, as costs are only included during the first year of the analysis.

Model outcomes are reported in terms of total and incremental costs, quality-adjusted life years (QALYs) and life-years (LYs), incremental cost-effectiveness ratio (ICER), and net monetary Benefit (NMB) calculated considering a willingness-to-pay threshold of € 20,000. This threshold was calculated, in compliance with the Dutch guidelines on economic evaluations, based on the proportional shortfall and fair innings of the condition evaluated [21].

Sensitivity analysis, covering one-way sensitivity analyses to identify model parameter drivers, and probabilistic sensitivity analyses assessing the robustness of results, were conducted.

Model structure

The decision tree model component simulated the initial-year’s disease paths with a daily cycle length, resulting in three mutually exclusive outcomes of death, cured and cured but experiencing post-acute COVID syndrome (PASC). The decision tree component was coupled to a Markov model component, simulating patient survival and quality of life during the years after the initial year of the infection using a one-year model cycle. The model structure is represented in Figs. 1 and 2. Transmission-dynamic effects of nirmatrelvir/ritonavir treatment on infections and spread of COVID-19 were not included in the analysis. Furthermore, the analysis did not consider risk of reinfection; only a single infection event is modelled. Note that PASC events were included in the decision three component of the analysis, assuming that such event would accrue and complete within the same year of occurrence of the COVID-19 infection.

Fig. 1

Model schematics—initial decision tree component

Fig. 2

Model schematics—subsequent Markov component

Model implementation followed the guidelines from the International Society for Pharmacoeconomics and Outcomes Research Task Force on Good Modelling Practices [22]. Hybrid decision-tree Markov models have been used previously in economic evaluations for respiratory illnesses, including COVID-19 [13, 23].

Clinical inputs

Table 1 presents the clinical input parameters for the analysis. Informing relative treatment efficacy parameters associated with nirmatrelvir/ritonavir treatment within the 5 days of symptom onset, we relied on the reduction of hospitalizations and deaths (i.e. 79.6%) reported by Lewnard and colleagues [7] from its large observational study in a vaccinated population during the Omicron era, and on the reduction in symptom days reported from the EPIC-HR clinical trial (i.e. 20%) [6, 7, 24,25,26]. The risks of events and transitions, applicable to individuals in each of the three groups analysed, were preferably based on Dutch specific publicly available sources [19, 27,28,29]. When no local data was available, published items from outside the Netherlands were relied upon. We focused on the most recent six months of data availability, to reflect the Omicron era as close as possible in the analyses. If considered appropriate, adjustments were made to enhance assumptions being reflective of the Omicron era.

Table 1 Input parameters

The World Health Organization’s Guideline on Therapeutics and COVID-19 from November of 2023 was the source used for informing baseline hospitalization rates for the three patient groups included in this analysis [30]. The guideline defines three risk categories for which the recommendations apply, reporting for each category the estimated hospitalization rates. The immunocompromised patient group in our analysis, was covered by the group with the highest risk of hospitalization of 6% (i.e. patients with diagnosed immunodeficiency syndromes, or receiving immunosuppressants having undergone solid organ transplant or being concerned by autoimmune illness) [30]. Concerning the other two groups of interest in the analysis (i.e. 60-plus patients with comorbidity, and 70-plus patients), the 3% baseline hospitalization rate reported in the guideline, for patients at moderate risk of hospitalization, was used [30]. The proportion of hospitalized patients that were admitted into the ICU was estimated to 6.5%, by comparing the total admissions reported in the NICE stichting reports on general wards and on ICUs [28, 29]. Finally, the NICE stichting report on ICUs provided the proportion receiving ventilation of any type (88.9% over the whole reporting period) following admission.

The mortality risk associated with COVID-19 infections in the outpatient setting was assumed to be 0%, as severe cases are assumed to follow the in-hospital branches of the decision tree. Mortality risk associated with COVID-19 infections in the hospitalized general ward setting was estimated to be 8.85% for the immunocompromised group, 11.25% for the group of 60-plus patients with comorbidity, and 10.78% for the group of 70-plus patients. Such estimates resulted from crude mortality rates specific to 60-plus and 70-plus patients, which were calculated from admissions and death figures reported in the NICE stichting report on general wards [28]. For the group of 60-plus patients, the calculated crude mortality rate was adjusted, by applying the weighted average of the OR (i.e. 1.3) reported for the presence of comorbidity categories [32]. Such an adjustment was not applied for the group of 70-plus patients, as advanced age was already accounted for in the estimate. Further, no adjustment was applied for the group of immunodeficient patients, as the 95% CI associated with the OR reported for immunosuppression suggested non-significance [32]. The mortality risk associated with COVID-19 infections in the hospitalized ICU setting, i.e. 34.44% for the immunocompromised group, 40.70% for the group of 60-plus patients with comorbidity, and 42.68% for the group of 70-plus patients, was estimated using the same approach, while relying on data from the NICE stichting report on ICUs [29]. As these estimates were calculated based on the latest available data from the NICE stichting, relying on the last 6 months of data already reflective of high vaccination uptake in The Netherlands together with predominance of the Omicron variant, they are expected to be the best approximation available to inform these parameters based on local data.

Once survivors of COVID-19 make it into the Markov component, mortality is primarily informed by (age- and gender-specific) general population mortality rates (due to all causes) published by the Dutch Central Bureau of Statistics. Long term survival is therefore associated with the outcome that patients obtain at the end of the decision tree part of the model. However, mortality in immunocompromised individuals or in those featuring underlying disease is expected to be higher than the mortality of the Dutch general population. Thus, the general population mortality rates in the Markov model component were adjusted by applying standardized mortality ratios (SMR). An SMR of 1.85 was used for the group of 60-plus patients with comorbidity, based on the weighted average of the mortality OR per single comorbidity categories reported for a Danish cohort of 65-plussers [34]. For the immunocompromised group an SMR of 1.7 was sourced from a large European study on human immunodeficiency virus patients under combination antiretroviral therapy but featuring immune deficiency (median sample CD4 T-lymphocytes count of 186 among 60-plus patients, being less than 500 considered as immuno-depressed). For the group of 70-plus patients, no additional adjustment was implemented, as advanced age is already accounted for in life tables.

The proportion of non-hospitalized high-risk patients with subsequent PASC, set to 12.7%, was sourced from a specific Dutch cohort study [33]. The proportion of PASC in the survival state after hospitalization was extracted from a recent study and set to 27.5% for patients discharged from the general ward, and to 43.1% for patients discharged from the ICU [35].

Health-related quality of life inputs

To calculate the QALYs associated with each branch of the decision-tree component, a baseline age-dependent pre-infection Dutch-specific utility level was combined with a series of disutilities, each, associated with subsequent clinically relevant events taking place throughout each possible disease course (represented by each individual branch in the decision-tree) [36]. For the immunocompromised patient population and the group of 60-plus patients with comorbidity, the age-dependant baseline utility level was adjusted to reflect the reduced utility expected in individuals with underlying disease/comorbidity or being immunocompromised [37]. For the analysis focusing on the group of 70-plus patients, no adjustment was applied because the baseline age-dependent already reflects their age profile. Note that in the Markov model component, an additional disutility accounting for the poorer post-discharge quality of life featured by post-ICU patients irrespective of PASC-status was implemented during the 4 years following the year of discharge [38].

Treatment-related adverse events were not included in the analysis, as in the EPIC-HR clinical trial, the incidence of grade 3 or higher adverse events related to nirmatrelvir/ritonavir or placebo was recorded in about 0.5% of the patients in both arms. In the Netherlands, it is typical to only assign disutilities to grade ≥ 3 adverse events in economic evaluations. Table 1 presents all utility and disutility inputs informing the model.

Costs inputs

Costs associated with each of the clinically relevant events included in each of the branches describing the 1 year of the analysis, were calculated on a per-day basis, subsequently being aggregated into lumpsums at the end of each branch.

Direct costs related to a COVID-19 infection included two categories. First, drug acquisition costs were set to € 1,047.70 (as per the Dutch list price, from the G-standaard of the Z-Index in December of 2024) for a treatment course of nirmatrelvir/ritonavir in accordance with the EMA label (i.e. 300 mg nirmatrelvir [two 150 mg tablets] with 100 mg ritonavir [one 100 mg tablet] taken every 12 h for five days) [62], and to zero for BSC. Secondly, health care resource utilization cost, associated to events encountered by the patient throughout the branches of the decision tree, were based on Dutch unitary costs from the Dutch costing manual from 2016 (costs from 2014 inflated to 2024 levels relying on the consumer price index published by the Dutch Central Bureau of Statistics) and plausible frequencies [43]. Outpatient healthcare costs were estimated to € 365 in the nirmatrelvir/ritonavir arm and to € 292 in the BSC arm (see Table S1 in the supplement) [39]. Daily hospitalization costs of € 627 in the general ward and of € 2656 in the ICU were based on the Dutch costing manual (costs from 2014 inflated to 2024 levels) and combined with an average duration of 9.1 and 15.8 days respectively [2, 39]. Healthcare costs for inpatient services associated with PASC were estimated to be € 784 per case, resulting from the product of the daily average cost of hospitalization in general ward with a length-of-stay assumed to be half the length-of-stay associated with the initial infection. This product was adjusted multiplying by the average readmission rate sourced from the literature (see Table S2 in the supplement) [40]. A one-off post-discharge cost of € 3055 associated with all hospitalizations ending on a patient discharge, was estimated by multiplying the hourly cost of revalidation treatment by the number of hours estimated in the ENRAF–NONIUS Dutch revalidation guideline for COVID-19 patients for the group of patients discharged from the general ward and for the ICU [39, 41]. As very few grade 3 (or higher) AEs were observed in the EPIC-HR clinical study, showing a similar incidence in both treatment arms, costs associated with AE management were not included in the analysis [6]. End of life costs of € 4690 were considered, based on disease activity codes reported by the Dutch heath authority NZa and estimated nursing costs (see Table S2 in the supplement).

Societal costs covered informal care, transportation, and productivity losses. Informal care costs resulted from aggregating the products between the average cost per hour for informal care, by plausible hourly needs specific to discharges from the general ward (28 h per infection, i.e. 2 h per day for 2 weeks), from the ICU (56 h per infection, i.e. 2 h per day for 4 weeks), and in cases of PASC (8 h per week) [39]. No needs in terms of informal care were assumed for infections in the outpatient setting. Transportation costs were associated with disease management (cost of attending a point of care to receive treatment, undergo examinations), and the corresponding parking cost. The cost tariffs used in the calculation were extracted from the Dutch costing manual (costs from 2014 inflated to 2024 levels), and the number of trips considered are linked with the frequency of utilization of outpatient health services [39]. In case of hospitalization, one return trip was assumed. Productivity losses were implemented following the friction cost methodology in line with the Dutch costing manual [21]. Productivity losses associated with hospital stays, the mean length-of-stay was considered. Productivity losses associated with outpatient infections were associated with the mean number of symptom days [6]. The maximum friction period of 19.9 weeks (139 calendar days) was considered for deaths. The maximum friction period was also considered for PASC cases, given that the mean time to recovery in such case is higher than the friction period [42]. When needed, costs were inflated to 2024 levels, using the consumer price index published by the Dutch Central Bureau of Statistics [43].

Sensitivity analyses

One-way sensitivity analyses and probabilistic sensitivity analysis were conducted for all model inputs. In the one-way sensitivity analyses, parameters varied relying on the upper and lower bounds of the 95% CI (which were recorded directly from the source when available or calculated based on the reported standard error or standard deviation). A 20% standard error was used by default when no uncertainty measures were available for a given input at the source (see Table S4 in the supplement). Further, NMB was used in the one-way sensitivity analyses to illustrate the sensitivity of cost-effectiveness results (as opposed to using ICERs, given that variations tested in some of the parameters resulted in dominant results) enhancing the interpretability of results in the tornado diagram potentially avoiding the complex interpretation of negative ICERs. Input distributions for the probabilistic sensitivity analysis were based on recommendations by Briggs et al. [44, 45] and are featured in Table S4 in the supplement.

Results

Discounted results from the Dutch societal perspective over the lifetime horizon for each of the groups under analysis are presented in Table 2. Compared with BSC, nirmatrelvir/ritonavir was associated with 0.130 additional LYs and 0.125 additional QALYs in the group of immunocompromised patients; with 0.055 additional LYs and 0.054 additional QALY’s in the group of 60-plus patients with comorbidity, and with 0.045 additional LYs and 0.053 additional QALYs in the group of 70-plus patients. Nirmatrelvir/ritonavir was associated with incremental costs of all groups analysed of € 49, € 474 and € 689 respectively, when compared to BSC.

Table 2 Base case results (discounted, lifetime horizon)—Dutch societal perspective

Therefore, nirmatrelvir/ritonavir was found to be cost-effective in all groups analysed, with an ICER well below the applicable willingness-to-pay threshold of € 20,000/QALY gained. When compared to BSC, the ICER (QALYs) was € 395 in the group of immunocompromised patients, € 8700 in the group of 60-plus patients with comorbidity, and € 13,021 in the group of 70-plus patients. The associated NMBs, at the said willingness-to-pay threshold, were € 2444, € 615 and € 369, respectively.

Discounted results from the Dutch health care payer perspective over the lifetime horizon for each of the groups under analysis are presented in Table S3. Under this perspective, nirmatrelvir/ritonavir remained cost-effective in all three groups considered, featuring an ICER of € 3588 when compared to BSC in the group of immunocompromised patients (i.e. 0.125 incremental QALYs and € 447 incremental cost), of € 13,009 in the group of 60-plus patients with comorbidity (i.e. 0.056 incremental QALYs and € 709 incremental cost), and of € 13,404 in the group of 70-plus patients (i.e. 0.053 incremental QALYs and € 709 incremental cost). The associated NMBs, at the said willingness-to-pay threshold of € 20,000/QALY gained, were € 2046, € 381 and € 349, respectively.

By the end of the infection period, a mortality risk reduction was estimated for patients in the nirmatrelvir/ritonavir treatment arm compared with patients in the BSC arm: 0.13% versus 0.63% in the group of immunocompromised patients, and of 0.08% versus 0.39% in the group of 60-plus patients with comorbidity and in the group of 70-plus patients. Furthermore, considering only patients surviving the infection period, the proportion of patients progressing to PASC was 1.1% lower in the group of immunocompromised patients treated with nirmatrelvir/ritonavir compared with BSC, and 0.6% lower for the group of 60-plus patients with comorbidity and the group of 70-plus patients.

Figures 3, 4 and 5 present the tornado diagrams of the one-way sensitivity analyses for each of the three groups analysed. The baseline proportion of patients hospitalized among high risk individuals, was the most impactful parameter on cost-effectiveness results across all groups of interest. The mortality risk associated with hospitalizations in the general ward, and the efficacy of nirmatrelvir/ritonavir reducing hospitalization rates, are followed as the second and third most impactful parameters in the group of immunocompromised patients. In the group of 60-plus patients with comorbidity and the group of 70-plus patients, the drug acquisition cost associated with nirmatrelvir/ritonavir and its efficacy reducing hospitalization rates are followed as the second and third most impactful parameters.

Fig. 3

One way sensitivity analysis results for the base case (immunocompromised patients)

Fig. 4

One way sensitivity analysis results for the base case (60− plus patients with comorbidity)

Fig. 5

One way sensitivity analysis results for the base case (70− plus patients)

Figures 6, 7 and 8 present the cost-effectiveness planes of the probabilistic sensitivity analysis for each of the three groups analysed. Results are robust across groups with compact iteration clouds, in particular for the group of 60-plus patients with comorbidity and the group of 70-plus patients, featuring a probabilistic mean result very close from the deterministic result. In the group of immunocompromised patients, all iterations were located under the willingness to pay threshold, thus, being cost-effective throughout. Further, 43% of iterations were located in the south-east quadrant (indicating dominance). In the group of 60-plus patients with comorbidity and the group of 70-plus patients, most iterations were concentrated in the north-east quadrant (97.6% and 99.9%, respectively), with most iterations suggesting a cost-effective result by being located under the willingness to pay threshold (94 and 84.8%, respectively). In the group of immunocompromised patients, the probability of being cost-effective at a willingness-to-pay threshold of € 20,000 was 100%, this probability was 94% in the group of 60-plus patients with comorbidity, and 85% in the group of 70-plus patients.

Fig. 6

Probabilistic sensitivity analysis results for the base case (immunocompromised patients)

Fig. 7

Probabilistic sensitivity analysis results for the base case (60− plus patients with comorbidity)

Fig. 8

Probabilistic sensitivity analysis results for the base case (70− plus patients)

Discussion

The objective of this study was to determine the cost-effectiveness of nirmatrelvir/ritonavir versus BSC in three patient groups at increased risk of severe disease, considering the Dutch societal perspective and a lifetime horizon. The three groups analysed, reflecting three different approaches to define the higher risk of developing a severe course of COVID-19 infection, consisted of the immunocompromised patient population, 60-plus patients with comorbidity, and 70-plus patients.

The base case analysis results for all three groups suggest that the reduction in costs associated with healthcare utilization, productivity losses, informal care, transportation, and end of life costs, do not fully offset the additional cost of treatment in the intervention arm, resulting in low to moderate incremental costs. Increases in LYs and in QALYs across all three groups analysed reflect a decrease in mortality versus BSC. Given that nirmatrelvir/ritonavir reduces the risk of hospitalization (compared to BSC alone) and reduces the duration of the symptomatology period, the positive differential in LYs and in QALYs are expected to be robust. Nirmatrelvir/ritonavir is therefore cost-effective versus BSC alone in all groups analysed. For all three groups, probabilistic sensitivity analyses, providing for parameter uncertainty, support the robustness of favourable cost-effectiveness, as it estimates that nirmatrelvir/ritonavir would have 100% probability of being cost-effective at a willingness-to-pay threshold of € 20,000 in the group of immunocompromised patients, a 94% probability in the group of 60-plus patients with comorbidity, and an 85% probability in the group of 70-plus patients. Scenario analysis results, considering the limitation to a health care perspective only, also resulted in favourable cost-effective results for nirmatrelvir/ritonavir versus BSC alone in all groups analysed, albeit with higher ICER results for each group than in the base case analysis.

A key limitation associated with this study was the lack of data at a sufficiently high granularity level to inform our analysis. Appropriately defining and populating each specific risk-group profile for the Netherlands was challenging, and various assumptions had to be made. Continuous changes in policies concerning the reporting of testing and mortality as the pandemic in the Netherlands evolved, led to data being heterogeneous over the periods of reporting. Specifically, data representing the latest situation in the Omicron era and reflecting a high uptake on vaccination, were not always available, having to rely on data that did overlap with the Delta period in some instances. This applied not just to epidemiologic datasets, but also to specific datapoints used for adjusting for the specific characteristics defining each of the population groups considered in the analysis. The epidemiological data for the group of 70-plus patients was the most complete, since age stratification is commonly included in data sets used as source for analyses. For the group of immunocompromised patients and for the group of 60-plus patients with comorbidity, additional adjustment to published data had to be made as a second-best approach. This limitation may have introduced biases, associated with differences in population characteristics, diverse geographical settings and different timeframes considered.

The EPIC-HR trial, being conducted in the period where the Delta variant was predominant while being restricted to unvaccinated persons [6], was not optimal to inform the efficacy of nirmatrelvir/ritonavir for reducing the risk of hospitalization in this analysis. Instead, the results from the study by Lewnard et al. [7] on a vaccinated population and in a timeframe with predominance of the Omicron variant, reporting an estimated effectiveness of 79.6% in preventing hospital admission or death within 30 days of a positive test for SARS-CoV-2 when treatment was administered within 5 days of symptom onset, was chosen to inform this parameter [7]. However, as nirmatrelvir/ritonavir efficacy data specific for the three groups included in this analysis were not available in Lewnard et al., the same efficacy rate had to be applied to all three groups.

Note that a total of 12 studies were identified that demonstrated the effectiveness of nirmatrelvir/ritonavir in reducing hospitalisation and mortality in several geographies [7, 11, 46,47,48,49,50,51,52,53,54,55]. Furthermore, in various studies, nirmatrelvir/ritonavir has also been shown to prevent COVID-19 hospitalisation and death in older patients (60-plus) and in patients with comorbidities [7, 48, 49, 51, 52, 56].

For the group of immunocompromised patients, an additional limitation came from the impossibility to fully match the defining characteristics described by the SWAB-FMS guideline, with the defining characteristics used in the sources providing the estimates that were used to inform that analysis [17]. An approximate match to the concept of ‘immunocompromised’ needed thus to be accepted as a proxy. Note that our model did not explicitly include risk of reinfection after and side effects of nirmatrelvir/ritonavir treatment. Obviously, clinical trial data—as well as current data from other sources—is still too limited to fully assess these aspects. Further real-world data will help establishing these aspects and careful monitoring of treatment is therefore warranted, with potential future model adaptations reflecting possible new insights. A last limitation concerning data inputs relates to the quality of life estimates. These were implemented in the model using utility scores estimated from reports of EQ-5D data, by applying utility decrements or disutilities associated with clinical events to the adjusted mean baseline utility for the Dutch general population. Unfortunately, such utility decrements specific to the Dutch setting were not available, requiring us to use UK utilities as a second-best approach.

The model structure developed was considered to appropriately reflect the individual costs and benefits associated with nirmatrelvir/ritonavir treatment. A straightforward static modelling approach was chosen to align with the abovementioned scarcity of detailed data. A dynamic model can assess disease transmission effects, but it requires more detailed data—also on contact patterns—as well as a substantial number of additional estimates that were not available. The static nature of the model used in this analysis cannot assess the impact on disease transmission in the population and, therefore, does not account for the impact that a change in disease transmission between individuals could have on survival and quality of life results. In addition, given that this analysis followed the standard societal cost perspective and corresponding categories, as specified in the Dutch guidelines on economic evaluations from 2016 [21], further economic savings to society and to the broader economy were not evaluated. This analysis did consider the potential for featuring PASC after infection in both, the hospitalized and the non-hospitalized branches of the decision tree. Further research on the extent of the consequences of PASC in the long term for the patient and the health system at large, could justify future analyses focusing on this aspect.

Other economic evaluations on treatments for COVID-19, published in international journals, report similar conclusions regarding the cost-effectiveness of nirmatrelvir/ritonavir. The special assessment conducted by US Institute for Clinical and Economic Review of outpatient treatments for COVID-19 (including nirmatrelvir/ritonavir) concluded that nirmatrelvir/ritonavir would meet standard US healthcare cost-effectiveness levels at an estimated ICER of US$ 21,000 per additional QALY while highlighting the impact of treatment efficacy against hospitalisation on results [13, 14]. Another recent analysis by Carlson et al. [25] estimated an ICER of US$ 8931 over a lifetime period for nirmatrelvir/ritonavir versus best supportive care, while pointing to baseline risk of hospitalization and treatment effectiveness parameters as drivers of results [25]. Last, one poster recently presented at the European Society of Clinical Microbiology and Infectious Diseases conference in April 2024, reported the cost-effectiveness of nirmatrelvir/ritonavir in the Dutch setting [57]. Offering a limited description of the methodology and the precise inputs used, it is challenging to determine the precise items driving the difference in results. Nevertheless, this study focuses on a different population for analysis, while it applies different assumptions regarding the effectiveness of nirmatrelvir/ritonavir on the reduction in hospitalizations and mortality.

Conclusions

From a Dutch societal perspective, nirmatrelvir/ritonavir is cost-effective versus BSC alone in the three groups analysed. Even though these results are expected to be reflective of the Omicron period of the endemic in The Netherlands, they may be considered illustrative for potential future situations with resurges of the COVID-19 pandemic.