Cost-effectiveness of therapeutics for COVID-19 patients: a rapid review and economic analysis

Population

The population considered within the EAG report has been divided into two broad groups. The first group consists of people who have been hospitalised due to COVID-19 and the second group consists of people who are at high risk of requiring hospital care due to COVID-19. Patients who were hospitalised for reasons other than COVID-19 and contracted COVID-19 in hospital and were at high risk of requiring hospital care for COVID-19 were categorised within the second group. For brevity, all patients not hospitalised due to COVID-19 who are at high risk of hospitalisation will be termed ‘non-hospitalised patients’ noting the aforementioned caveat regarding patients who contract COVID-19 in hospital, whereas patients who have been hospitalised directly because of COVID-19 are referred to as ‘hospitalised patients’.

Following discussions with NICE, the definition for patients at high risk was aligned to that considered within the Platform Adaptive trial of NOvel antiviRals for eArly treatMent of COVID-19 In the Community (PANORAMIC) clinical study,8 with the exception that being aged 50 years or over was not considered to be a high-risk factor.

The aim of treatment differs between both groups. For patients hospitalised due to severe or critical COVID-19, the aim of treatment is to reduce the immunoinflammatory response of the body and prevent clinical deterioration. For non-hospitalised patients, the aim of treatment is to prevent viral replication and damp inflammation, thus reduce the probability of the development of severe symptoms that could lead to hospitalisation or death.

Interventions

The interventions listed within the NICE scope7 are shown in Tables 1–3 based on marketing authorisation in the UK at the time of writing. Table 1 contains the interventions with marketing authorisation in the UK, Table 2 contains the interventions with conditional marketing authorisation in the UK and Table 3 contains the interventions with no marketing authorisation in the UK. Each table contains the generic name of the intervention, its branded name and the company manufacturing it, the class of intervention, the mode of administration and recommended dose. Table 1 provides the indication for the drug, while Tables 2 and 3 provide the population in key studies for the intervention.

Multiple interventions are indicated for the prevention of severe COVID-19. Severe disease in adults is defined as having clinical signs of pneumonia plus at least one of the following: respiratory rate > 30 breaths/minute, severe respiratory distress, or saturation of peripheral oxygen < 90% on room air and would require hospitalisation. 9

Comparators

The comparators within the decision problem include all of the interventions contained in Tables 1–3, when used in the same position as a particular intervention and additionally standard of care (SoC) which would be dependent on the severity of the patient’s illness. SoC is defined as any treatment widely accepted by the NHS, which is routinely funded by the NHS with no strong rationale to appraise it, for example supplemental oxygen and dexamethasone. SoC has evolved throughout the COVID-19 pandemic, which means that randomised controlled trials (RCTs) conducted comparing interventions against SoC may not be directly comparable as SoC has improved over time.

Outcome measures

The NICE scope7 lists nine possible outcomes to explore: mortality; requirement for respiratory support; time to recovery; hospitalisation (requirement and duration); time to return to normal activities; virological outcomes (viral shedding and viral load); post-COVID-19 symptoms; adverse effects of treatments; and health-related quality of life (HRQoL). All model outcomes, except virological outcomes were assessed; these were excluded as these would be of more relevance to decision problems that included transmission and due to the prioritisation of other endpoints given the limited time available.

The cost-effectiveness of the eight treatments was expressed in terms of incremental cost-effectiveness ratios (ICERs) which were reported in terms of cost per QALY gained. A patient lifetime horizon was used to take differential mortality between treatments into account.

Subgroups

Due to time constraints, the only subgrouping considered was related to whether oxygen was required upon admission to hospital entry. This was considered important as the licensed indication and the clinical outcomes for some of the appraised interventions depend on the level of oxygen support required. The EAG is aware that other possible criteria for selecting subgroups include but are not limited to: age; immune system competence; comorbidities; seroprevalence; vaccination status; and the predominant SARS-CoV-2 variant but did not have the time to explore the impact of these characteristics.

Rationale for using living systematic reviews

COVID-19 clinical research has accelerated dramatically worldwide, with over 5000 registered trials investigating therapeutic interventions for COVID-19. 17 The need for rapid information on COVID-19 has resulted in a paradigm shift, especially in the communication of scientific results. Traditional systematic reviews can date quickly but ‘living’ systematic reviews search for evidence much more regularly than standard reviews and incorporate relevant new evidence as it becomes available. This is important in the context of COVID-19, in which the evidence-base is rapidly changing as new data emerge. The ability of a ‘living’ systematic review and NMA to regularly update and incorporate relevant new evidence as it becomes available makes it the best type of evidence synthesis, in the opinion of the EAG, to inform this pragmatic rapid evaluation. This approach has been recommended by best practice recommendations3 which stated that ‘HTA agencies should consider the use of existing “living” clinical evidence reviews and meta-analyses to inform their clinical effectiveness decisions’ as ‘Using these sources will reduce duplication of work and may allow for quicker assessments’.

The EAG did not have the time to attempt to untangle the impact of differences between studies in terms of aspects such as the dominant SARS-CoV-2 variant, SoC, vaccination status, outcome definition and age of participants and caution that the results may not be directly comparable between interventions. The EAG also did not have time to: validate the data within the living systematic reviews; to quality assess the component studies; or to remove studies that were not using the appropriate doses. To recognise this uncertainty the EAG has run ‘mean’, ‘high’ and ‘low’ efficacy scenarios in the cost-effectiveness analyses (see Analyses undertaken) to allow decision-makers an indication of how cost-effectiveness changes with different efficacy assumptions.

Selection criteria for the living systematic reviews

Several living systematic reviews that incorporate emerging trial data and allow for analysis of comparative effectiveness of multiple COVID-19 treatments have been robustly developed and published. 12,17–19 Two sources were selected as they provided detailed relevant outcome data from individual studies and up-to-date evidence synthesis to inform the model.

The first source is the COVID-NMA initiative,11,12 supported by the World Health Organization (WHO) and Cochrane which is a living systematic review of registered randomised trials, in which all available evidence related to COVID-19 is regularly collected, critically appraised and synthesised using pairwise comparisons and NMA methods. These analyses are updated every 2 weeks and results can be accessed via a web interface (https://covid-nma.com/).

The second source is the metaEvidence initiative,18 supported by the University Hospital of Lyon and the University of Lyon which is also a living meta-analysis and evidence synthesis of therapies for COVID-19 and is an emerging online resource that provides direct access to the efficacy and safety results reported in the studies for potential drugs for the treatment of COVID-19. The risk of bias, synthesised by meta-analysis, is also reported. The analyses are updated within a target time of ˂ 24 hours with results accessed through a web interface (www.metaevidence.org/COVID19.aspx).

Other sources of evidence, which primarily informed living guidelines,17,19 did not report the extracted outcome data from individual studies. As such, they precluded further synthesis and evaluation.

Assumption of transportability of relative treatment effects

A consequence of the need to use data from the living systematic reviews was that there was reduced scope for the EAG to undertake nuanced analyses with a key limitation being that the EAG had to assume that all relative treatment effects were generalisable to different settings. This meant that for each intervention, the same treatment effects, either hazard ratios (HRs) or relative risks (RRs), were assumed to be applicable regardless of study characteristics which include: the age, perceived severity, vaccination status and history of SARS-CoV-2 infection of patients; the SoC at that time; the geographical location; and the dosage of the intervention used. The EAG acknowledges that this assumption may be incorrect, which adds additional uncertainty to the clinical- and cost-effectiveness results.

Inclusion criteria and data extraction

Selected data were extracted for the interventions contained in Tables 1–3. Key model outcomes such as time to death, clinical improvement at day 28 or day 60 (defined as a hospital discharge or improvement on the scale used by trialists to evaluate clinical progression and recovery) and incidence of serious adverse events (SAEs) were initially extracted from the COVID-NMA living systematic review. 12 Where relevant outcome data were not available, these data were extracted from the metaEvidence living systematic review. 18 All data extractions (undertaken between 16 March and 18 May, updated between 25 and 31 May 2022 and updated again on 6 September 2022 and 19 December 2022) were undertaken by one reviewer (AS) and checked by a second reviewer (AP), with any discrepancies resolved by a third reviewer (SR). All data and evidence synthesis analyses were extracted from forest plots, tables and text generated by the COVID-NMA and metaEvidence web interface; checking of the extracted data by the EAG against the original RCT publications for accuracy could not be undertaken within the timescales of the project.

Adjustments made for changing SoC, SARS-CoV-2 variant, vaccination status and prior infection

The conditions under which each study was conducted were heterogeneous. Across time SoC has changed markedly, most particularly with reference to the widespread use of corticosteroids such as dexamethasone and change in SARS-CoV-2 variants. The vaccine roll-out in England has provided protection that was not available to patients recruited to early studies, similarly, there is likely to be an increased level of protection associated with prior infection. Ideally there would be attempts to establish the impact of different circumstances on the observed clinical effectiveness of interventions in studies, although this was not possible within the timescales of the project. As such, the EAG had to make a simplistic assumption that none of the changes were treatment effect modifiers, and that given this, the relative benefits observed in the studies were generalisable and could be applied to the estimated outcomes for patients with COVID-19 in England in Summer 2022. The EAG notes that this assumption is very unlikely to hold but believed that this approach was preferable to provide no estimates of effectiveness. The EAG further warns that comparing treatments, either in terms of clinical- or cost-effectiveness could be very misleading.

One notable comment is that there is a belief raised by stakeholders and confirmed by the clinical authors of this report that casirivimab/imdevimab does not work for the Omicron variant of SARS-CoV-2. Further guidance from the US Food and Drug Administration that ‘sotrovimab is not authorized in any US state or territory at this time’ (5 April 2022) as it is unlikely to be effective against the Omicron BA.2 subvariant. 20 Additionally, less than a fortnight before the report was completed, the WHO offered strong recommendations against the use of casirivimab/imdevimab in patients with COVID-19 and against the use of sotrovimab in patients with non-severe COVID-19. 21 As such, it is likely that the ‘low’ efficacy scenario described in Analyses undertaken may be the most appropriate scenario for casirivimab/imdevimab and for sotrovimab in patients with non-severe COVID-19, with the results from the ‘mean’ and ‘high’ efficacy scenarios reserved for consideration if there is a change in the SARS-CoV-2 variant. The robustness of any estimate of efficacy is uncertain.

Another stakeholder comment was the belief that monoclonal antibodies may have better effectiveness than antivirals in reducing supplemental oxygen use in patients treated in the community that are subsequently hospitalised. Clinical opinion provided to the EAG suggests that the effect of antivirals is uncertain. The EAG believes that the sensitivity analyses undertaken in conjunction with an incremental net monetary benefit approach, shortened to net monetary benefit (NMB) approach (see Analyses undertaken for further details) would allow the committee to consider alternative assumptions.

General model structure for hospitalised patients

The economic model was developed in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) and uses a partitioned survival approach (often referred to as area under the curve approach) with three mutually exclusive health states; (1) discharged from hospital and alive, (2) hospitalised with or without COVID-19 and (3) death from any cause (COVID-19 or due to other causes).

Movements between health states were not explicitly modelled. Instead, the partitioned model estimates health state occupancy at each time interval. Figure 1 shows a simplified schematic of the model structure. A daily cycle length is used until the end of parametric extrapolation, at day 70, after which a weekly cycle length is used. An initial daily cycle length was chosen to allow changes in treatment and/or hospitalisation and oxygen requirements that happen early in a patient’s stay to be modelled at a granular level. A cohort partitioned survival approach was used due to the limited time, the absence of individual patient-level data (IPD) that may allow a more complex model structure and the need to not explicitly model transitions between health states as would be required by a state transition model. A limitation of the partitioned survival approach is that it is not possible to track individual patients in the model which may have allowed a better representation of the patient experience.

FIGURE 1.

Simplified schematic of model structure (values are for illustration only).

While in hospital, the 8-point ordinal scale of clinical status (an inverted version of the scale originally developed for severe influenza requiring hospitalisation as recommended by the WHO) used in the Adaptive COVID-19 Treatment Trial (ACTT-1) RCT,25 and in the Remdesivir Effectiveness Evaluation Study26 is used. This ordinal scale is described in Table 6 and is used in the model to (1) define the population at baseline in terms of oxygen requirements at the start of treatment, and (2) estimate changes in hospital/oxygen requirements during the hospital stay.

When evaluating the interventions, patients enter the hospital model based on the marketing authorisation, where this has been granted, or in relation to the population in the key studies. Table 7 provides information of the ordinal scales within which the interventions can be used, in line with their marketing authorisation (or anticipated marketing authorisation) of each intervention as illustrated in Tables 1–3. Scale scales 1 and 2 describe patients with COVID-19 in the community while ordinal scales 3 or higher describe patients in hospital, although ordinal scale 3 does not require ongoing medical care. Only ordinal scales 3 or higher are relevant for the hospital model.

Movements (improvement or worsening) between the different hospitalisation/oxygen requirements over time are modelled with each scale being associated with cost and HRQoL implications. During their hospital stay, patients are distributed according to their hospital/oxygen requirement derived from the placebo arm of the ACTT-1 study and additional assumptions where necessary. Figure 2 provides an illustration of movement between ordinal scales for patients who needed supplemental oxygen on hospital entry and when treated with SoC. The area above Ordinal Scale 7 denotes patients who have died; the area below Ordinal Scale 3 signifies patients discharged from hospital.

FIGURE 2.

Illustration of ordinal scale occupancy during hospital stay of a cohort admitted to hospital requiring supplemental oxygen and receiving SoC treatment.

Following Rafia et al. 22 the model assumes that all patients are discharged at 70 days. This may underestimate the costs and QALY losses associated with hospital care for the most efficacious drugs, although this is not expected to be a large limitation as the proportions of patients estimated to be in hospital at day 70 is very small. For example, in the mean efficacy scenario this proportion was zero for all interventions.

Pivotal clinical trials/studies for treatments for COVID-19 used in this economic evaluation tend to follow patients and typically collect key clinical outcomes after 28 days of follow-up. It is, therefore, necessary to extrapolate beyond the duration of studies to capture the life expectancy and HRQoL following hospital discharge from COVID-19. Following discharge, patients who were hospitalised with COVID-19 are at an elevated risk of death;27 emerging evidence suggests that some patients discharged continue to experience symptoms and have a reduced quality of life,28–37 may require readmission due to COVID-19,25,38–42 and are at an elevated risk to experience multiorgan dysfunctions27 (such as respiratory diseases, diabetes, cardiovascular, liver and kidney diseases) and may require long-term management/monitoring. 43 Within the model, HRQoL reductions and additional costs associated with COVID-19 have been included; for brevity this has been termed ‘long COVID’. In addition, the possibility of patients having an increased risk of death following COVID-19 has been modelled using a standardised mortality rate (SMR) applied to the mortality rates for an age- and sex-matched population.

Consequently, a seven-step approach is employed:

• Step 1: Use of a parametric function (hazard spline model with three knots) fitted to the relevant outcomes (time to death and time to discharge) for all patients on the SoC arm in Randomised Evaluation of COVid-19 thERapY (RECOVERY) study44 for the first 28 days, as used in Rafia et al. 22

• Step 2: This parametric function is adjusted to reflect the outcomes at day 28 as reported in the literature to reflect the benefit of using corticosteroids, which represent the current SoC for patients in need of supplemental oxygen. 43 The model was calibrated as detailed in Time to hospital death in patients initiating SoC (with or without corticosteroids).

• Step 3: Treatment effect in the form of HRs or RRs for the interventions were applied to the SoC curves. Data were missing for some interventions with respect to the HR for discharge and the HR for clinical improvement (see Results of the rapid evidence review). The EAG noted that these HRs were not large drivers of the cost-effectiveness results, and that there was no clear relationship between the two HRs and other results, such that an estimation could be made. As no values for interventions with data were markedly different from unity when compared with SOC, the EAG decided to use the values for SoC where data were missing, with a sensitivity analysis undertaken using a HR of 1.0 for all interventions which is likely to be favourable to casirivimab/imdevimab in relation to time of discharge and baricitinib/remdesivir in relation to clinical improvement.

• Step 4: As shown in Figure 2, ordinal scale occupancy in hospital is assumed to last until the distribution for overall survival (OS) and the distribution for time to discharge intersect. It was assumed in the model that none of the hospitalised cohorts would remain in hospital after 70 days.

• Step 5: Parametric extrapolation is employed to estimate the rates of death from day 28 until day 70 in the base case.

• Step 6: Use of mortality rates from the general population, adjusted by an SMR for the assumed mean duration of long COVID to reflect the elevated risk of death in patients with COVID-19 discharged from hospital.

• Step 7: Use of unadjusted mortality rate from the general population after the assumed mean duration of long COVID.

General model structure for non-hospitalised patients

The model structure used for assessing interventions for patients with COVID-19 and at high risk of hospitalisation is depicted in Figure 3. This is comprised of a decision tree which simulates whether hospitalisation is required, and for those patients who are hospitalised, whether supplemental oxygen is required on admission. Patients who are hospitalised were assumed to enter a partitioned survival model as described in General model structure for hospitalised patients.

FIGURE 3.

Structure of the decision tree used for the non-hospitalised cohort.

The EAG’s central estimate of the probability of patients at high risk of being hospitalised was taken from Patel et al. 13 which was a retrospective cohort study of non-hospitalised patients who received early treatment for, or were diagnosed with COVID-19 or had a positive polymerase chain reaction test between 1 December 2021 and 31 May 2022 using the Discover dataset in North-West London. Four thousand forty-four untreated patients did not receive treatment with a monoclonal antibody or an antiviral. The untreated cohort had a median age of 52 years and 86% had received two or more vaccinations. One hundred fourteen patients were hospitalised with a primary diagnosis of COVID-19 within 28 days of COVID-19 diagnosis equating to 2.82%. This value has been used in the base case. The EAG notes that this value aligns with the definition of high-risk detailed in a report45 produced for the Department of Health and Social Care, but believes that this, along with the sensitivity analyses conducted will be informative for decision-making.

Three additional sources for hospitalisation for high-risk patients in the community were also considered. These were: Hippisley-Cox et al.,46 the PANORAMIC study47 and Shields et al. 48

Data from Hippisley-Cox et al. 46 indicated an average risk of hospitalisation following a SARS-CoV-2 positive test was 1.45% based on approximately 1.3 million people in England, although the EAG notes that it would expect the value based on a positive SARS-CoV-2 test to be lower than based on having COVID-19 (which is people with SARS-CoV-2 who are symptomatic). The risk of hospitalisation was markedly increased in patients with Down syndrome, patients with kidney transplant, chemotherapy grade B or C and rare neurological conditions with midpoint HRs > 4. Many conditions were associated with increased risks of hospital admission, although vaccination and prior SARS-CoV-2 infection were associated with lower risks.

Data from the PANORAMIC study47 indicated a lower proportion of high-risk people were hospitalised with 96 of 12,525 (0.77%) untreated patients requiring hospitalisation. Stakeholders, however, stated that high-risk patients may not have been recruited to PANORAMIC and were instead treated by COVID Medicines Delivery Units (CMDU) and that the proportion of hospitalised patients would be underestimated.

Data reported by Shields et al. 48 indicated that the risk of hospitalisation for immunodeficient individuals who did not receive treatment was 15.9%.

In sensitivity analyses (see Analyses undertaken) the hospitalisation rate is changed from 2.82% to 1%, 5%, 10% and 20% to provide a comprehensive range for Committee discussion. These ranges would allow decision-makers to explore the cost-effectiveness of treatments in subgroups that are of greater, or lower, risk of hospitalisation.

The proportion of hospitalised patients requiring supplemental oxygen was estimated from an International Severe Acute Respiratory and Emerging Infection Consortium report49 where the requiring oxygen of any level on admission was calculated at 81% (55% HFO 16% non-invasive ventilation and 10% invasive ventilation). These proportions were assumed to be independent of treatment (intervention or SoC) due to lack of data, and it is plausible that if a person requires hospitalisation, then the intervention has not worked. The EAG ran a sensitivity analysis assuming that the proportion requiring supplemental oxygen was reduced to 75% [with 60% on HFO 10% NIV non-invasive ventilation and 5% invasive ventilation] for patients who have received an intervention.

The model applies an RR to account for other medical attended visits (MAVs) (i.e. visits other than hospital admission) compared to admissions. This RR was estimated from data in Nyberg et al. 50 and was equal to 1.37 (1.23% MAV rate divided by 0.9% hospitalisation rate). Only costs were considered for MAVs and incorporated a visit to an accident and emergency department.

Two key clinical outcomes were extracted from the living systematic reviews: RRs for hospitalisation or death, and RRs for day 28 all-cause mortality, which are shown in Figures 26 and 27, respectively. The RR for hospitalisation or death was assumed to apply for hospitalisations only due to the relatively low mortality rate compared to the admission rate. A separate RR was calculated for each intervention for deaths within hospital such that the overall RR for death at 28 days was consistent with the published estimate reported in Tables 4 and 5. This methodology assumes that there were no deaths among non-hospitalised patients in the first 28 days of the model. The EAG believes that this limitation would have a negligible impact on the ICER.

The EAG assumed that there would be no further active treatment in hospital for patients treated in the community, and thus patients receive SoC only. This decision was based on the following factors: the RRs for mortality for some of the interventions used in the community were substantially lower than the HRs for those treatments used in hospital. For example, the RR for death for nirmatrelvir/ritonavir was 0.04 while the median HR for death for baricitinib was 0.61, indicating that the residual effect of nirmatrelvir/ritonavir was larger than the impact of baricitinib, which had the most efficacious median value. Furthermore, there is no evidence for the synergistic effects (or not) of using multiple interventions.

In line with NICE’s final scope the model does not consider the impact of treatment on the transmission of SARS-CoV-2. 7 The community model may also slightly overestimate the costs associated with people in hospital at the time of catching COVID-19 as hospitalisation costs may be double-counted, although the EAG believes this will be of limited importance.

The modelling did not assess the logistical aspects of treatment in the community, but the EAG notes that this could be a large factor in deciding which treatments could be preferred, as oral treatments could be more acceptable to patients and healthcare systems than treatments that are given intravenously or subcutaneously.

Baseline characteristics after discharge

The economic model uses age and gender distributions to estimate both the rate of mortality beyond the duration of clinical evidence and to estimate HRQoL values for patients discharged from hospital and patients at high risk remaining in the community. The baseline mean age for the modelled hospitalised cohort was calculated from weekly Office for National Statistics (ONS) data51 reported in the middle of May 2022. For patients with COVID-19, these data included rates of hospital admissions per 100,000 people and number of deaths, by age bands. These values were multiplied by population data obtained from the ONS52 to estimate the absolute number of admissions and deaths by age band. The estimated number of discharged patients was calculated by subtracting the number of deaths from the number of admissions. Table 8 presents the estimated numbers and percentages calculated for admission, death and discharge conditional on age band.

If the midpoint of each age band represented the entire band, mean ages for admission, death and discharge are estimated at 70.6, 82.8 and 68.7 years, respectively. Without data to accurately estimate the age for people with COVID-19 at high risk of hospitalisation who do not get hospitalised, the EAG assumed that this equalled the age of patients who had not been hospitalised in order to maintain the average starting age for all comparators, which was 55 years in the base case.

The distribution between sexes was taken from an Intensive Care National Audit and Research Centre report53 which reported that 38.3% of patients admitted to hospital from May 2021, in a critically ill state due to confirmed COVID-19, were female.

Time to hospital death in patients initiating SoC (with or without corticosteroids)

The EAG used the following steps to estimate the survival of patients admitted to hospital due to COVID-19 and receiving SoC.

The Kaplan–Meier (KM) estimate for OS was taken from the control arm of the RECOVERY study,44 and was digitised which allowed pseudo-IPD to be reconstructed based on the algorithm developed by Guyot et al. 54 A spline model (hazard scale) with 3 knots was subsequently fitted to the pseudo-IPD using the R package flexsurv and employing a natural cubic spline function. This model was selected over standard parametric functions (such as the exponential, Weibull, Gompertz, Log-Normal, Log-Logistic, Gamma and Generalised Gamma) to increase the accuracy in the estimate and because parametric extrapolation beyond the observed period of the trial was limited to a maximum of 70 days. This distribution was then calibrated to the current data such that 73.5% of patients were alive for the population in need of oxygen and 86.0% of patients were alive for the population admitted with no need of supplemental oxygen at 28 days. These values were taken from a NICE rapid guideline19 assuming that the outcomes for patients without corticosteroid use were generalisable to patients requiring supplemental oxygen and the outcomes for those patients with corticosteroids were generalisable to patients not requiring supplemental oxygen. This decision was made as corticosteroids were only seen to be efficacious in patients not requiring supplemental oxygen. For illustration, Figure 4 shows the OS curves used in the model for SoC and remdesivir by oxygen requirement at hospital admission; the remdesivir data was calculated by applying the HR shown in Table 4.

FIGURE 4.

Illustration of OS curves used for the hospitalised cohort for SoC and remdesivir by oxygen requirement at entry.

Time to discharge for patients initiating SoC

The methodology for calculating time to discharge for patients receiving SoC was similar to that for time to death [see Time to hospital death in patients initiating SoC (with or without corticosteroids)] with the following changes. The KM estimate for time to discharge was taken from the control arm of the RECOVERY study,44 and a spline model (hazard scale) with three knots selected. This distribution was then calibrated to the current data such that 64.0% of patients for the population in need of supplemental oxygen and 80.4% of patients with no need of supplemental oxygen were discharged at 28 days. These values were taken from a NICE rapid guideline19 assuming that the outcomes for patients without corticosteroid use were generalisable to patients requiring supplemental oxygen and the outcomes for patients using corticosteroids were generalisable to patients not requiring supplemental oxygen. For illustration, Figure 5 shows the time to discharge curves used in the model for SoC and casirivimab/imdevimab by oxygen requirement at hospital admission; the data was calculated applying the HR shown in Table 4.

FIGURE 5.

Illustration of time to discharge curves used for the hospitalised cohort for SoC and casirivimab/imdevimab by oxygen requirement at entry.

Redistribution of patients according to supplemental oxygen/hospitalisation requirements

In order to estimate costs and QALYs during an average hospital stay, it was necessary to model how patients move between the 8-point ordinal scale as each scale has different consequences in terms of the costs of treatment and the HRQoL of the patient. Hospitalised patients with COVID-19 may receive supplemental oxygen, defined as LFO, HFO and mechanical ventilation. However, during their hospital stay, patients may require more or less intensive management. Hospitalised patients are divided into five states, which correspond to ordinal scales 3 to 7.

Treatment effects for interventions compared with SoC

The treatment effects for interventions are summarised in Tables 4 and 5. Where data were not available for clinical improvement or time to discharge a value of 1.0 was used as the model results were not sensitive to these values within the observed range associated with other interventions. A value of 1.0 indicates that the level of clinical improvement and time to discharge are the same for an intervention and for SoC. The RRs for clinical improvement were only applied for improvements of two ordinal scales or more as per the outcome definition in the early trials. 56

Duration of treatment/number of doses

The dosage information data were taken from the NICE COVID-19 rapid guideline. 19 Where either the dosage or the duration of treatment was not available, this information was taken from alternative sources. Table 10 summarises the dosage information used in the model.

Mortality rate assumed posthospitalisation and for those people who did not require hospital admission

The unadjusted rate of mortality for the general population is taken from the England and Wales life table 2018–20. 58 After discharge, patients hospitalised with COVID-19 were assumed to be at an elevated risk of death while they have long COVID. An SMR of 7.7 (7.2–8.3) was applied based on the RR reported by Ayoubkhani et al. 27 which was estimated from 47,780 patients treated for COVID-19 in NHS hospitals and discharged alive, using matched-controls and which had a median follow-up of 140 days. This SMR was also applied to patients at high risk in the community for the period in which they were simulated to have long COVID.

Serious adverse events

While the living systematic reviews allowed the RRs related to SAEs to be extracted, on inspection the ERG identified that these were not events related to the unwanted impacts of the interventions but were conditions related to severe COVID-19. As such, many interventions were associated with less SAEs than SoC, which is generally atypical for efficacious pharmacological treatments. As the model was explicitly tracking the severity of patients using the 8-point ordinal scale the EAG decided to omit SAEs from the model. The degree to which this is favourable, or unfavourable to specific treatments is unknown.

Long COVID

To facilitate modelling, the authors have not strictly adhered to previously defined definitions of long COVID but have taken a simplistic approach with sensitivity analyses undertaken to assess the uncertainty of the impact of long COVID in the base case.

The prevalence of long COVID within the wider community has been taken from an ONS report dated 6 May 2022,59 which in supplementary tables reports adjusted model estimates for long COVID of any severity and at any point since the last vaccine of: 8.7% of double-vaccinated patients and 8.0% of triple-vaccinated patients, who had the Omicron BA 1 variant; and 15.9% of double-vaccinated patients and 8.6% of triple-vaccinated patients, who had the Delta variant. Having noted the relatively wide CIs for the ONS estimates, the difference depending on vaccination status (with no data reported for unvaccinated patients) and the method it proposes to use for estimating the duration of long COVID (described below), the EAG assumed that 10% of patients in the community who were at high risk of severe COVID-19 but did not need hospitalisation would experience long COVID. The EAG was not aware of any evidence on the impact of community treatment on the incidence of long COVID and thus assumed that long COVID was independent of treatment. The degree to which this is favourable, or unfavourable to specific treatments is unknown. More recent data has been released (with a date of July 2022,60 although this did not influence the decision made by the EAG).

The ONS released an updated report in December 2022 on the prevalence of ongoing symptoms following COVID-19 in the UK. 14 This stated that of people with self-reported long COVID, defined as ‘symptoms continuing for more than four weeks after the first suspected coronavirus (COVID-19) infection that were not explained by something else’ 87% of people had been first infected by COVID-19 (or suspected they had) at least 12 weeks earlier, 55% were infected at least 1 year previously, and 27% at least 2 years previously.

The EAG fitted simple parametric distributions to the three reported estimates of at least 12 weeks duration (87% with long COVID at 12 weeks, 55% at 1 year and 27% at 2 years). A Gamma distribution (shape = 74.373, scale 1.089), a Weibull distribution (shape = 1.065, scale 82.049) and a log-normal distribution [mean = 3.998, standard deviation 1.212 (both on the log scale)] were observed to fit the data well. The mean survival times from these distributions were 81.0 weeks (Gamma), 80.1 weeks (Weibull) and 113.6 weeks (log-normal).

The plot of the log-normal and the Weibull parametric fits, which have the highest and lowest mean times of the distributions that fitted the data well are shown in Figure 6, with the log-normal used in the base-case. For the log-normal distribution, approximately 30% of people still have symptoms at 2 years, 10% at 5 years and 3% at 10 years.

FIGURE 6.

Assumed duration of long COVID.

For its base case, the EAG assumed the log-normal distribution was most appropriate as data may be administratively censored in that patients who had COVID-19 only 6 months ago could not report having symptoms at 2 years, and the EAG noted that the proportions of people reporting long COVID symptoms for at least 2 years had increased from 19% in a June report61 to 27% in the December report. The EAG undertook sensitivity analyses halving and doubling the mean duration, the range of which includes the mean from the Gamma distribution. The EAG notes that its analyses are simplistic as formal survival analysis methods were not used, and that it does not assume that all patients must have long COVID for at least 4 weeks, as used in some definitions but believes that the analyses undertaken are informative for decision-making despite this limitation.

From Evans et al. 62 it is estimated that at approximately 6 months, 51.7% of patients with non-missing data (n = 830) reported that they had not recovered from COVID-19; this value increases to 71.2% when patients stating they were not sure if they had recovered were included. The patients included in the study were hospitalised early in the pandemic (between March and November 2020) and it is unclear how generalisable this result is to patients hospitalised in 2022. The best-fitting log-normal and Weibull distributions shown in Figure 6 estimate the proportions of patients not recovered from long COVID to be 72.9% and 74.5% at 26 weeks which is similar to the value reported in Evans et al. 62 when those not sure if they have recovered are included. Given the uncertainty in patients who stated they were not sure if they had recovered, a simplistic assumption was made that all patients hospitalised due to COVID-19 would suffer long COVID. The EAG was not aware of any evidence on the impact of hospital treatment on the incidence of long COVID and thus it was assumed that this is independent of treatment. The degree to which this is favourable, or unfavourable to specific treatments is unknown.

While the simplistic approach used does not capture any potential differential severities of long-term effects based on initial severity of COVID-19,63 the impact of vaccination status, or the consequences of any organ damage caused by long COVID the authors believe that this method is still informative for decision-making.

Drug acquisition costs

While the EAG acknowledges that some stock may have already been acquired before this appraisal, recommendations are to inform future commissioning decisions and so the list price is used in this report. Drug acquisition costs, both list prices and prices with Patient Access Scheme (PAS) discounts applied were provided to the EAG by NICE. All analyses in this report have used list prices, with analyses using the PAS for baricitinib, sotrovimab and tocilizumab included in a confidential appendix. Table 11 summarises the list prices used in the model. Three drugs had list prices which were not publicly available: casirivimab/imdevimab, molnupiravir and tixagevimab/cilgavimab. No cost-effectiveness results are presented for these interventions in this report but will be contained in a confidential appendix; however, results related to QALYs gained through the use of these treatments will be presented. For corticosteroids, daily costs were assumed to be negligible compared to the in-hospital day cost and were not included for simplicity.

Administration costs

The EAG assumed that the costs associated with treatment administration while in hospital would be incorporated in the unit costs associated with hospitalisation (see Unit costs associated with hospitalisation). NICE provided the EAG with information from the CMDU relating to how indicative local tariffs were calculated. The costs included elements for: staffing, administrative support, dispensing, clinical consumables, couriering medicines, travel, stationery and hiring rooms, but excluded medical review to assess drug interactions and any changing in permanent staffing structures.

The costs associated with providing oral antivirals were estimated to be £410 per person, whereas the costs associated with intravenous (i.v.) infusions were estimated to be £820 per person. For simplicity, the costs associated with administering injections were assumed to be the same as oral antivirals. Within the analyses it has been assumed that there is likely to be a delay in patients receiving intravenous casirivimab/imdevimab and that a subcutaneous version would be used instead.

Stakeholders commented that due to drug interactions, additional time related to medication reviews would be required for some interventions and that these costs should be incorporated in the model. The need for additional time and the costs associated with this time are both uncertain so they have not been included in the model. However, the EAG comments that due to the NMB approach taken, any determined costs associated with additional medication review could be subtracted from the NMB values, allowing the NICE appraisal committee, and other stakeholders, to determine the relative cost-effectiveness of interventions.

Unit costs associated with hospitalisation

Following stakeholder comments some sources for costs have been updated and the latest version of the National Schedule of NHS costs (2020–21) has been used. 64 A stakeholder suggested alternative sources for the costs associated with ordinal scales 3, 4 and 5 which were as follows: for ordinal scale 3, a weighted average of currency codes DZ11R to DZ11V (Lobar, Atypical or Viral Pneumonia, without Interventions); for ordinal scale 4, a weighted average of currency codes DZ11N to DZ11Q (Lobar, Atypical or Viral Pneumonia, with Single Interventions); and for ordinal scale 5 a weighted average of DZ11K to DZ11M (Lobar, Atypical or Viral Pneumonia, with Multiple Interventions). These appeared plausible and the EAG estimated the cost per bed-day by dividing the average costs per currency code by the average length of stay per currency code taken from the 2017/2018 National Schedule of NHS cost65 as no later data on length of stay existed. However, the results lacked face validity as the estimated average cost per bed-day was less in ordinal scale 5 than in ordinal scales 3 and 4.

The EAG used an alternative approach which generated plausibly valid costs per bed-day for the ordinal scales. These costs were larger than those in the report sent out for stakeholder consultation, apart from ordinal scale 3 which is lower. In response to the consultation on the ACD a stakeholder indicated that there was a better way to calculate the costs of a bed day in hospital in ordinal scales 4 or 5 as the cost per finished cost episode (FCE) was used to approximate the cost per bed day and there can be multiple FCEs per admission. The EAG has reviewed these approaches and agrees that these are an improvement and preferred the approach where the costs of being in ordinal scale 5 were greater than in ordinal scale 4. The EAG calculated values using this approach (described below) and produced slightly higher cost values than the consultee; the higher costs were used in the model. The NHS currency codes used to estimate the costs for ordinal scales 4, 6 and 7 are detailed in Table 12.

The mean length of stay associated with COVID-19 was estimated from NHS Digital, Hospital Episode Statistics for England. Admitted Patient Care statistics, 2020–2168 using primary diagnosis codes U07.1 and U07.2 which suggested a weighted average of 10.6 days. From the same source, the weighted mean number of FCEs per admission for U07.1 and U07.2 was 2.29. The cost per FCE for ordinal scale 4 was calculated using the National Schedule of NHS costs66 as the weighted average of HRG codes DZ11R, DZ11S, DZ11T, DZ11U and DZ11V (lobar, atypical or viral pneumonia, without interventions) for non-elective long stay which was a value of £3524. Using the same source, the cost per FCE for ordinal scale 5 was calculated as the weighted average of HRG codes DZ11N, DZ11P and DZ11Q (lobar, atypical or viral pneumonia, with single intervention) for non-elective long stay which was £5411. The use of DZ11 has previously been used as a proxy for COVID-19 costs in a published paper. 70

The costs for ordinal scale 4 and 5 were calculated as the average cost per FCE (£3524 and £5411, respectively), multiplied by the mean number of FCEs per admission (2.29) and divided by the mean length of stay (10.6 days); this results in a cost per day of £759.28 in ordinal scale 4 and £1165.70 in ordinal scale 5.

Costs associated with COVID-19 for outpatients or following discharge

Health-related quality of life

NICE’s preferred measurement of HRQoL is the EQ-5D74 which asks participants to value five domains of health (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) on either a three-level scoring system [EuroQol-5 Dimensions, three-level version (EQ-5D-3L)] or a five-level scoring system [EuroQol-5 Dimensions, three-level version (EQ-5D-5L)]. A value of 1.00 generated by this instrument indicates perfect health whereas a value of 0.00 indicates a state equivalent to death.

Mean efficacy results for patients requiring supplemental oxygen on admission to hospital

The results of the mean efficacy analysis for patients requiring supplemental oxygen on admission to hospital are shown in Table 13. All interventions were estimated to have a cost per QALY gained compared to SoC below £12,000.

High efficacy results for patients requiring supplemental oxygen on admission to hospital

The results of the high efficacy analysis for patients requiring supplemental oxygen on admission to hospital are shown in Table 14. All interventions were estimated to have a cost per QALY gained compared to SoC below £12,000. The costs associated with tocilizumab are lower than for other drugs due to the assumed higher rate of discharge of patients as the remaining interventions do not have data and assumed to have the same discharge rate as SoC.

Low efficacy results for patients requiring supplemental oxygen on admission to hospital

The results of the low efficacy analysis for patients requiring supplemental oxygen on admission to hospital are shown in Table 15. Remdesivir and baricitinib/remdesivir were dominated by SoC due to providing no additional QALYs at an increased price. The ICER for baricitinib was below £9000, that for tocilizumab was below £29,000.

Mean efficacy results for patients requiring no supplemental oxygen on admission to hospital

The results of the mean efficacy analysis for patients not requiring supplemental oxygen on admission to hospital are shown in Table 16. All interventions were estimated to have a cost per QALY gained compared to SoC below £12,000.

High efficacy results for patients requiring no supplemental oxygen on admission to hospital

The results of the high efficacy analysis for patients not requiring supplemental oxygen on admission to hospital are shown in Table 17. All interventions were estimated to have a cost per QALY gained compared to SoC below £9000.

Low efficacy results for patients requiring no supplemental oxygen on admission to hospital

The results of the low efficacy analysis for patients not requiring supplemental oxygen on admission to hospital are shown in Table 18. Remdesivir and baricitinib/remdesivir were estimated to be dominated by SoC due to producing no additional QALYs at an additional cost. Baricitinib had a cost per QALY below £6000.

Mean efficacy results for patients at high risk of hospitalisation

The results of the mean efficacy analysis for patients at high risk of hospitalisation are shown in Table 19. Nirmatrelvir/ritonavir was estimated to have a cost per QALY compared to SOC of below £7000 with the remaining interventions having an ICER above £30,000.

High efficacy results for patients at high risk of hospitalisation

The results of the high efficacy analysis for patients at high risk of hospitalisation are shown in Table 20. Nirmatrelvir/ritonavir was estimated to have a cost per QALY compared to SOC of below £6000, sotrovimab was estimated to have a cost per QALY compared to SOC of below £19,000 and remdesivir was estimated to have a cost per QALY compared to SOC of below £25,000.

Low efficacy results for patients at high risk of hospitalisation

The results of the low efficacy analysis for patients at high risk of hospitalisation are shown in Table 21. Nirmatrelvir/ritonavir was estimated to have a cost per QALY compared to SOC of below £12,000 while remdesivir and sotrovimab both had ICERs of over £100,000 compared with SoC.

Amending the duration of long COVID

The NMB results when the duration of long COVID is doubled (to 227.6 weeks) and halved (to 56.8 weeks) are shown in Figures 10–12 for people admitted to hospital requiring supplemental oxygen, those admitted to hospital with no need for supplemental oxygen and those treated in the community at high risk of hospitalisation, respectively, using a WTP of £20,000 per QALY. Corresponding data using a WTP of £30,000 per QALY are shown in Appendix 4 (see Figures 35–37).

FIGURE 10.

The NMB results for patients admitted to hospital who require supplemental oxygen when the duration of long COVID is halved and doubled. Assuming a WTP of £20,000.

FIGURE 11.

The NMB results for patients admitted to hospital who do not require supplemental oxygen when the duration of long COVID is halved and doubled. Assuming a WTP of £20,000.

FIGURE 12.

The NMB results for patients in the community with COVID-19 who are at high risk of hospitalisation when the duration of long COVID is halved and doubled. Assuming a WTP of £20,000.

For patients in hospital, longer durations of COVID reduced NMBs, as there were more survivors with long COVID when treatment was beneficial. In contrast, NMBs were increased in patients at high risk in the community as treatments stopped patients being hospitalised and therefore reduced the numbers assumed to have long COVID. There were one instances where the NMB had a different sign compared with the base case when the duration of COVID was changed which was for sotrovimab in the mean efficacy scenario when the duration of long COVID was doubled. There was a moderate impact on the ICERs generated for hospitalised patients in the mean scenario typically changing between +/– £2000 per QALY. The impact was greater for remdesivir when used in the community although in this instance the initial ICERs were large.

Amending the hospital admission percentage for people with COVID-19 in the community at high risk of hospitalisation treated with SoC

The NMB results when the hospitalisation admission percentage for people with COVID-19 in the community at high risk of hospitalisation treated with SoC was changed from 2.82% to 1%, 5%, 10% and 20% are shown in Figure 13 assuming a WTP of £20,000 per QALY and Appendix 4 (see Figure 38) assuming a WTP of £30,000 per QALY. The proportion of patients with COVID-19 at high risk of being hospitalised being admitted to hospital makes a large difference to the NMB with values increasing as the admission proportion increases. All interventions had a positive NMB when the proportion of patients hospitalised was increased to 10.00% and the mean efficacy scenario was used independent of the WTP assumed. Although remdesivir and sotrovimab had negative NMBs when the low efficacy scenario was used even when the admission percentage was increased to 20%. The ICERs versus SoC changed considerably based on the proportion of high-risk patients hospitalised. The ICERs when assuming 1%, 10% and 20% and the mean efficacy were: nirmatrelvir/ritonavir (£24,647, dominant and dominant), remdesivir (£280,819, £16,170 and £1512) and sotrovimab (£111,318, £4870 and dominant).

FIGURE 13.

The NMB results for patients in the community with COVID-19 who are at high risk of hospitalisation when the hospital admission percentage was changed. Assuming a WTP of £20,000.

Amending the age of people with COVID-19 in the community at high risk of hospitalisation treated with SoC

The NMB results when the age assumed for people with COVID-19 in the community at high risk of hospitalisation treated with SoC was changed from 55 years to 50 and 60 years are shown in Figure 14, assuming a WTP of £20,000 per QALY, Appendix 4 (see Figure 39) shows results using a WTP of £30,000 per QALY. Where the RR of day 28 mortality is lower than one the NMBs decrease as the age of the patients increases because less QALYs are gained when a death is prevented. However, the EAG notes that there is no explicit link between risks of poor outcomes and age, and it is likely that all other things being equal, older patients are at a higher risk and that the results could be misleading. However, there was only a moderate impact on the ICERs generated for hospitalised patients in the mean scenario typically changing between +/– £2000 per QALY.

FIGURE 14.

The NMB results for patients in the community with COVID-19 who are at high risk of hospitalisation when the age was changed from 55 years to 50 and 60 years. Assuming a WTP of £20,000 per QALY.

Using a HR of unity for all interventions in relation to time to hospital discharge and time to clinical improvement

The NMB results when all interventions and SoC had the same impact on time to hospital discharge and time to clinical improvement are shown in Figure 15 for patients requiring supplemental oxygen and in Figure 16 for patients not requiring supplemental oxygen, assuming a WTP of £20,000 per QALY. Figures 40 and 41 in Appendix 4 use a WTP of £30,000 per QALY. This sensitivity analysis did not change the patterns or the sign of the NMBs. These parameters were not large drivers of the ICER. The only noticeable change in the ICER was that for tocilizumab which increased by about £3000 compared with SoC.

FIGURE 15.

The NMB when a HR of unity was used for all interventions in relation to time to hospital discharge and time to clinical improvement for patients requiring supplemental oxygen. Assuming a WTP of £20,000 per QALY.

FIGURE 16.

The NMB when a HR of unity was used for all interventions in relation to time to hospital discharge and time to clinical improvement for patients not requiring supplemental oxygen. Assuming a WTP of £20,000 per QALY.

Changing the baseline distribution of supplemental oxygen requirements for people with COVID-19 in the community upon hospitalisation

The NMB results when the interventions were assumed to have a less severe distribution following treatment in the community are shown in Figure 17 assuming a WTP of £20,000 per QALY. Figure 42 in Appendix 4 provides NMBs assuming a WTP of £30,000 per QALY. This sensitivity analysis had a minor impact on the ICERs and did not change whether any NMBs were positive and negative.

FIGURE 17.

The NMB results when treatment in the community for high-risk patients was associated with less supplemental oxygen on admission to hospital. Assuming a WTP of £20,000 per QALY.

Applying a utility decrement of 0.02 per day for people in the community receiving i.v. treatment

The NMB results when a disutility of 0.02 per day for those receiving i.v. treatment in the community are shown in Figure 18 assuming a WTP of £20,000 per QALY and in Appendix 4 (see Figure 43) assuming a WTP of £30,000. This sensitivity analysis made no discernible change to the NMBs or ICERs.

FIGURE 18.

The NMB results when a disutility of 0.02 per day is assumed for patients receiving intravenous (i.v.) treatment in the community. Assuming a WTP of £20,000 per QALY.

Changing the SMR for people with long COVID

The NMB results when the SMR associated with long COVID is changed from 7.7 to 5.0 and 10.0 are shown in Figures 19–21 assuming a WTP of £20,000 per QALY. Figures 44–46 in Appendix 4 provide these data assuming a WTP of £30,000 per QALY. The change in the SMR for people with long COVID had little impact on the ICERs for hospital treatments. There was a greater impact for treatments in the community, but this did not change whether a NMBs was positive or negative.

FIGURE 19.

The NMB results for patients admitted to hospital who require supplemental oxygen when the SMR associated with long COVID is changed. Assuming a WTP of £20,000 per QALY.

FIGURE 20.

The NMB results for patients admitted to hospital who do not require supplemental oxygen when the SMR associated with long COVID is changed. Assuming a WTP of £20,000 per QALY.

FIGURE 21.

The NMB results for high-risk patients in the community when the SMR associated with long COVID is changed. Assuming a WTP of £20,000 per QALY.

Strengths of the economic analysis include

• The use of effectiveness data from living systematic reviews.

• An attempt by the EAG to align the results of SoC produced by the model with data observed in mid-2022.

• Uncertainty in the model inputs and assumptions has been explored in sensitivity analyses.

• The modelling attempts to capture movement between the 8-point ordinal scale to consider the costs and consequences of patient improvement and patient decline.

• The modelling explicitly attempts to take the impact of the longer-term implications of COVID-19 into consideration.

• The development of the model allows for a relatively quick evaluation of the treatments should more contemporary data become available.

Limitations of the analysis include

• The characteristics of the decision problem may have changed considerably since the pivotal trials for each intervention was conducted. Such changes include the emergence of new SARS-CoV-2 variants, the introduction of a vaccination programme, proportion of people with a history of prior SARS-CoV-2 infection and the widespread use of corticosteroids in SoC. The EAG assumed that none of these were treatment effect modifiers and that the treatment effects were generalisable which is likely to be incorrect for a proportion of interventions.

• No recent studies were identified using the Omicron BA5 most prevalent in England in the Summer of 2022.

• No head-to-head studies of interventions were identified that could be used in the modelling and the uncertainty regarding the most efficacious treatment is large.

• List prices are used for all interventions; results including PASs could not be provided in a publicly available document.

• For some interventions list prices were not publicly available. As such, no ICERs have been presented for these drugs.

• Uncertainty remains in the underlying rates of hospitalisation in patients with COVID-19 at high risk of hospitalisation under SoC.

• Uncertainty remains in the underlying rates of death in patients hospitalised due to COVID-19 who receive SoC.

• All deaths associated with people at high risk in the community are assumed to occur in hospital.

• SoC only was assumed to be provided to patients in hospital if they had been treated with an intervention in the community as the residual effects of some treatments used in the hospital were larger than treatments used in hospital.

• Treatments used in hospital were not assumed to affect the proportion of discharged people with long COVID and treatments used in the community were not assumed to affect the proportion of people not admitted to hospital with long COVID.

• All patients were assumed to be discharged from hospital at day 70, which could favour the more efficacious treatments in reducing hospital costs.

• No prior beliefs were incorporated relating to the clinical efficacy of the interventions. It may be clinically implausible that treatments that have a statistically significant beneficial HR relating to hospitalisation or death would be associated with increased RR of death at 28 days.

• The model did not consider secondary infections, which is likely to be unfavourable to the interventions.

• The model did not consider reinfections. It is unknown if this is favourable or unfavourable to the interventions.

• The model did not consider enablement benefits such as maintaining the capacity for operations or in avoiding delays in patients’ treatment that could arise due to either a reduced number of patients in hospital with COVID-19, or reduced staff absence due to COVID-19.

• No value of information analysis was conducted. This would allow funders to estimate the relative benefits of investing in future research.

• No analysis was conducted on whether it is logistically possible to treat patients in the community with COVID-19 and a high risk of hospitalisation with i.v. drugs.

Areas of future research

There is considerable uncertainty related to many aspects of this evaluation which hinders forming an accurate estimate of the ICER. A key uncertainty is the clinical effectiveness of interventions in conditions that do not replicate those in the pivotal studies. Contemporary research assessing the relative clinical effectiveness of interventions (and SoC) within head-to-head studies at current levels of vaccination, against the current predominant SARS-CoV-2 variant would be beneficial if the results could be obtained in a timely manner. Further data related to the probability of hospital admission and death for patients at high risk in the community would also improve the precision of the estimated ICERs as would ascertaining the average age of this population. If possible, analysing efficacy data by previously specified risk groups, such as age band, and underlying risk category, may allow more granular results to be obtained. The impacts of long COVID in terms of morbidity and mortality are currently uncertain and further research is required in this area. Value of information analyses could be undertaken to efficiently direct future research although it is clear that the efficacy of the interventions will be a key driver of any cost-effectiveness results.

Given current knowledge the EAG is happy that the results produced using relatively simplistic techniques supported with sensitivity analyses are informative to decision-makers. If data become available that show that the sum of the consequences for a cohort of homogenous people is not equal to the sum from a same-sized cohort of heterogeneous people then more complex modelling techniques, such as individual patient models may be required. More complex modelling could explore the benefits associated with the possibility of secondary infection and reinfection, and with wider aspects such as enablement benefit.