Draft guidance

Impact of COVID-19

3.1 COVID-19 is the acute respiratory illness caused by the SARS-CoV-2 virus. It can range from mild to severe. In severe disease, excessive immune response to the virus may cause severe complications associated with hospitalisation and death. The need for organ system support, particularly respiratory support, is also a key feature of severe disease and can lead to substantial longer-term morbidity. COVID-19 may cause long-term symptoms that continue or develop after acute infection called 'long COVID'. These are health problems that fluctuate and can last several months or years, and which severely impact a person's physical and mental health and potentially affect their ability to work, attend school or do their usual activities. During draft guidance consultation, consultees highlighted the treatment gap for children. At the second evaluation committee meeting (referred to as 'second meeting' from now on) one clinical expert explained that COVID-19 rarely makes children unwell. But there is a small proportion of children with underlying conditions who have an increased risk of severe COVID-19 comparable with adults who have underlying conditions. Many people are at increased risk of hospitalisation or death from COVID-19, including people who are immunosuppressed (for example, people with primary immunodeficiency, people having chemotherapy, or people who have had a transplant) or who have comorbidities (such as heart disease, respiratory disease, diabetes, neurological conditions). Some immunocompromised people are at risk of persistent viral infection if their immune system cannot control the virus. Patient experts explained that the increased risk of hospitalisation and death has led to some people changing their treatments, lifestyle and behaviour during the COVID-19 pandemic because of the need to shield. Patient organisations emphasised the need for treatments to prevent progression to severe COVID-19. They considered that routine availability of these treatments would support a return to normality for many people who already have disease burden from other comorbidities. The committee agreed that the risk of hospitalisation and death, and other longer-term impacts of COVID-19, can result in severe physical and mental burden and that there is an unmet need in this population.

The rapidly evolving SARS-CoV-2 virus

3.2 The global COVID-19 pandemic has caused unprecedented challenges to the healthcare system and this is reflected in the evidence collected on COVID-19 and treatments for it. The SARS-CoV-2 virus has evolved throughout the pandemic, as has the healthcare system's ability to respond to the virus. New variants of the virus and subvariants, referred to as variants of concern, have emerged throughout the pandemic. The properties of each variant can differ, such as the levels of transmissibility and disease severity. The clinical experts explained that understanding of the disease has changed throughout the pandemic, with increasingly effective supportive care, vaccination, and greater natural immunity. The committee understood that overall hospitalisation and mortality from COVID-19 has reduced, and the incidence of COVID-19 pneumonitis in hospital has lowered, as has the need for supplemental oxygen or mechanical ventilation.

3.3 At the time of first evaluation committee meeting (referred to as 'first meeting' from here on), the dominant variant of concern in the UK was the Omicron sublineage BA.5. The Omicron variant (B.1.1.529) has multiple subvariants based on mutations in specific spike proteins. The clinical experts explained that changes in the epidemiology and context of COVID-19 have led to people with COVID-19 having different characteristics than seen earlier in the pandemic. At the second meeting, the committee saw the updated Omicron variant data published in the UK Health Security Agency's (UKHSA's) technical briefing 49. Based on all the UK sequenced samples between 26 December 2022 to 1 January 2023, BQ.1 was the dominant 'designated variant'. BQ.1 was not expected to increase the risk of severe COVID-19 compared with BA.5. The committee understood from this data that the BQ.1 subvariants account for a large proportion of the currently circulating variants in the UK. The committee noted the XBB.1.5 and CH.1.1 subvariants are some of the fastest growing variants in the UK. The clinical experts explained that people presenting at hospital with COVID-19 are mainly either unvaccinated or immunocompromised, or did not have an immune response to vaccines. They reported that 'viral persistence' from chronic infection is a concern in immunocompromised people because new variants or subvariants can develop if the viral infection persists. They also noted that offering a clinically ineffective treatment unable to clear the infection may increase the risk of future variants developing. The committee noted the changing nature of SARS-CoV-2 and context of the pandemic affect the generalisability of the evidence for any treatment being evaluated. It agreed that the most appropriate approach would be to consider how relevant the clinical data are to the current endemic context of the disease at the time of this evaluation, but noted that the context and relevant variants are still changing at a fast pace.

Key definitions

3.4 The committee noted that the marketing authorisations for the treatments which lower the risk of progression to severe COVID-19, including nirmatrelvir plus ritonavir, were based on evidence from populations with slightly different definitions of high risk. For example, some trials included people with at least 1 risk factor for severe COVID-19 whereas some had specific age requirements. Understanding of the prognostic effects of risk factors has developed throughout the pandemic, and therefore the available evidence may represent a heterogeneous population. The committee acknowledged the potential limitations of the available evidence but considered it was important to clearly define high risk and therefore treatment eligibility. PANORAMIC was a large UK platform trial that included people with many different potential risk factors, including chronic conditions and immunosuppression, and allowed enrolment of people aged over 50 years. It also allowed for clinical judgement of clinical vulnerability. The independent advisory group report commissioned by the Department of Health and Social Care ('the McInnes report' from here on) defined groups of people at highest risk for adverse COVID-19 outcomes, including hospitalisation and death. The 'UK interim commissioning policy on treatments for non-hospitalised patients with COVID-19' used the McInnes report to define high risk. The clinical experts noted that some treatments were available through the interim commissioning policy at the time of PANORAMIC enrolment. The interim commissioning policy's and McInnes report's high-risk definition would have influenced the risk level of people who enrolled in PANORAMIC. At the first meeting, the committee considered this in its evaluation of the clinical evidence. The committee considered the different definitions of risk and concluded that PANORAMIC included people who had a lower risk of severe COVID-19 compared with the McInnes high-risk definition.

Other key risk groups

3.5 The clinical experts gave examples of additional considerations around how high-risk groups are affected differently:

Age as an independent risk factor

3.6 PANORAMIC allowed enrolment of people aged over 50 years who did not have any comorbidities. The committee questioned the inclusion of age over 50 years as an independent risk factor for progression to severe COVID-19. The clinical experts considered that age was an important risk factor. They cited the International Severe Acute Respiratory and emerging Infections Consortium (ISARIC) study of mortality in the earlier stages of the pandemic that defined age over 50 years as a risk factor (Knight et al. 2020). They noted that age over 70 years may be an important determinant of mortality but also considered that the relationship between age and comorbidities is complex, particularly for immunocompromised people. One of the companies considered that age was an important risk factor but noted an ongoing debate about what age is appropriate for inclusion in the high-risk group. The clinical experts agreed it was challenging to define an exact age that defines high risk. The committee was concerned that making a recommendation based on age might cause inequality, given that age is a protected characteristic. For this reason, NICE technology appraisal guidance on medicines for cardiovascular disease do not include criteria based on age, despite it being a well-recognised risk factor. The committee noted that age is a protected characteristic and any recommendation including age would need to be assessed for impact on equity of treatment. At the first meeting, the committee concluded that more evidence was needed on the impact of age to justify including it as an independent factor that increases risk at similar levels to other risk factors defined in the McInnes report. This should include evidence, adjusted for these risk factors, from a vaccinated population who are infected with the Omicron variant. At the second meeting, the committee noted the additional evidence provided by consultees which showed a statistical relationship between age and comorbidities. The committee acknowledged that age is a risk factor for progression to severe COVID‑19. The committee considered that the relationship between age and comorbidities can be important in explaining risk of severe disease. The committee concluded that age over 70 years is likely to be confounded by underlying conditions which could also contribute to increased risk of severe disease. The committee also noted that additional evidence is needed to model age over 70 years as an independent subgroup for the mild COVID-19 setting. It said the evidence should include age-adjusted hospitalisation and mortality rates for the untreated population and relative treatment effects. The committee concluded that the McInnes report's definition of high risk included the most robust evidence of people who have a high risk for progressing to severe COVID-19, and this did not include age as an independent risk factor.

High-risk definition conclusion

3.7 The assessment group (AG) explained the approach used to model high-risk groups in its economic model (see section 3.19). At the first meeting, it assumed that people had general population survival, with a starting age of 56.6 years and the same hospitalisation rate as PANORAMIC. Therefore, no individual high-risk subgroups were modelled based on specific baseline characteristics, and these characteristics were explored in sensitivity analyses that represented the entire group eligible for treatment. The clinical experts acknowledged the difficulties of defining high risk by separate subgroups. The committee recognised that the decision problem for this evaluation required a definition of who has a high risk for progressing to severe COVID-19. It recognised the limitations of the model in characterising a group at high risk but considered the hospitalisation rate to be the most important variable for sensitivity to the clinical inputs (see sections 3.18 and 3.19). At the second meeting, the committee repeated these limitations of modelling separate high-risk groups and concluded that a single definition of high risk should be used. The committee noted that evidence at a subgroup level is limited and too uncertain to parameterise the model. For example, additional functionality, clinical or cost inputs and treatment-effectiveness assumptions would be required to make differential subgroup recommendations and this would not be practical or aligned with the decision problem. The committee did not see additional evidence to justify splitting the high-risk group. The committee considered that the McInnes report's definition of high risk was based on the most robust evidence of people who have a high risk for progression to severe COVID-19, and this did not include age as an independent risk factor. Another benefit of using this definition is that outcomes data has been collected on this well-defined cohort over the course of the pandemic, providing some evidence from vaccinated people who were infected with Omicron variants. The committee considered the use of the Q‑COVID risk calculator in clinical practice but concluded it had limited applicability because of the limitations of the economic model. The committee noted a wider definition of risk, from PANORAMIC, was included in the marketing authorisations for each of the treatments (see section 3.4). However, it concluded that the definition of risk in the McInnes report is the most robust definition. The committee acknowledged that the McInnes definition of high risk may be revised over time. Depending on the nature of the revisions, this guidance may need to be reviewed if a difference in clinical or cost effectiveness is expected.

Partial review considering broader high-risk population

3.8 NICE were made aware of the Therapeutics Clinical Review Panel (TCRP) modelling group findings: risk of severe COVID-19 outcomes ('the Edmunds report' from here on) and a company request to submit additional evidence supporting the cost effectiveness of nirmatrelvir plus ritonavir in a broader population than the McInnes-defined high-risk group. The Edmunds report considered if any other groups of people have an equivalent risk to people with any condition in the McInnes-defined high-risk group. The Department of Health and Social Care considered that based on the report, age 70 and over, diabetes, and obesity were important risk factors that should be taken into account in a cost-effectiveness analysis. The committee therefore considered the findings of the report in detail and the additional evidence submitted by the company (see section 3.21).

3.9 The Edmunds report included analysis of 3 sources of evidence on outcomes of people with COVID-19, comprising large datasets from UK observational studies: OpenSAFELY (18.7 million people), Agrawal (30 million people) and Hippisley-Cox (1.3 million people). The OpenSAFELY data was collected during the Delta wave, while the Agrawal and Hippisley-Cox datasets were mostly collected during the Omicron wave. The OpenSAFELY data included unadjusted mortality rates whereas the other datasets included adjusted mortality and hospitalisation rates. The Edmunds report concluded that there was some evidence suggesting additional groups have an equivalent or greater risk of severe COVID-19 than people with certain autoimmune or inflammatory conditions such as rheumatoid arthritis or inflammatory bowel disease (the lowest risk level for people included in the McInnes high-risk. definition. These additional groups included people aged 70 and over, people with diabetes or living with obesity, and people with other conditions. The introduction to the Edmunds report noted 'the methods used are crude and any groups identified through this process would require closer scrutiny to better understand their risk and to what extent this might be modified by improved access to antivirals and therapeutics'. The Edmunds report also noted the limitations of the analysis in that the 3 sources of evidence used different definitions of risk groups, different definitions of outcomes, adjusted for different variables, and collected evidence during different waves of the pandemic.

3.10 The committee considered the evidence underpinning the Edmunds report's findings. It noted that the OpenSAFELY data had been collected during the Delta wave. The committee was aware that the Omicron variant and its sublineages, which have been dominant since December 2021, are less virulent than the Delta variant. However, while this would impact absolute risks, the information on relative risks between population groups was still relevant to the decision problem of the partial review. The committee noted that the aims of the independent advisory group (McInnes) and the TCRP modelling group (Edmunds) were slightly different. The McInnes group considered which groups are at highest risk from COVID-19 and would therefore be most likely to benefit from treatment. The Edmunds group addressed whether there are additional groups with a risk level at least as high as those who are already eligible for treatment. The committee noted the authors' comment that the extent to which risk may be modified by improved access to treatments for these additional groups would need close scrutiny. The committee was also aware that different methodology underpinned both reports. The McInnes group included more granular information on patient groups by specifying whether certain autoimmune or inflammatory conditions are active or uncontrolled, and when people are taking specific medications likely to affect their immune response to vaccination. The analysis underpinning the Edmunds report had used diagnosis codes to identify people with certain conditions. The committee considered that this would result in the groups having more heterogeneity in terms of risk by, for example, including people not currently taking immunosuppressants because of disease remission. The committee had previously discussed that COVID-19 risk may be more related to medication than diagnosis (see section 3.5). It considered that this heterogeneity would cause the overall risk estimate to mask a significant proportion of people who have much lower risk and are therefore less likely to benefit from treatment. The committee concluded that there was not robust evidence to suggest that the broader population identified by the Edmunds report were equivalent to the group defined in the McInnes report, in terms of having equivalent likelihood to benefit from treatment.

Treatments for mild COVID-19

3.11 Current clinical management of mild COVID-19 (including hospital-onset COVID-19) in people who have a high risk for progression to severe COVID-19 includes treatments commissioned through a UK interim commissioning policy (see section 3.4). In November 2022, the policy was:

Assessment group's indirect comparison approach

3.12 In line with best practice guidance for assessing COVID-19 treatments (Elvidge et al. 2021), the AG used systematic reviews and network meta-analyses (NMAs) from publicly available sources. These reviews (COVID‑NMA and metaEvidence) are updated regularly as 'living' systematic reviews.

The mild COVID-19 setting included these clinical endpoints:

The AG highlighted some significant limitations of its approach, because of the changing nature of COVID-19 (see section 3.2). Each trial included in the analysis was done at a different time in the pandemic. The context of the disease has also changed with different circulating variants of concern, and changes in protection through vaccinations and greater natural immunity over time. Each of these limitations were compounded by significant differences in trial design, baseline characteristics and geographical locations. The AG explained that the analysis assumed any relative effect of treatment is transferable to current clinical management. The clinical experts commented that meta-analysing the trial results may not be appropriate. This is because the weighting of each trial in a meta-analysis may not consider the relevance of the context of each trial within the analysis, for example, with different variants. The committee recognised the high levels of uncertainty with each treatment effect and the context-specific nature of the evidence. To characterise the uncertainty, rather than use probabilistic sensitivity analysis, the AG ran scenarios using the mean and the upper and lower confidence limits of each efficacy estimate. This provided scenarios showing 'mean efficacy', 'lower efficacy' and 'higher efficacy' estimates. The AG cautioned the committee that the lower-efficacy and higher-efficacy scenarios had limitations because they represented a different uncertainty to that in the evidence base; they represented uncertainty on the estimates in the trial and were therefore sensitive to the number of events in each trial, rather than the context in which the trial happened. Therefore, they would not be sensitive to changes in efficacy against new circulating variants of concern. The committee understood the limitations of the scenario analysis. The committee considered it represented an attempt to address some aspects of uncertainty in the absence of alternative methods to model the uncertainty. At draft guidance consultation, consultees highlighted the lower efficacy scenarios were arbitrary and a probabilistic sensitivity analysis would be a better way to capture the uncertainty. The committee noted that the heterogeneity in the trial populations and the generalisability issues across the trials made the uncertainty challenging to parameterise. Therefore, the appropriate type of uncertainty would not have been captured in the probabilistic sensitivity analysis. Consultees also noted that the mortality assumptions meant that treatment in hospital had a higher mortality risk compared with standard care. In response, the AG updated this assumption and capped the mortality rate to equal 1 for the low-efficacy scenario.

Observational evidence

3.13 The committee considered the latest data from OpenSAFELY (non-randomised observational evidence from 40% of English GP practices). The OpenSAFELY database links with other national databases including the Office for National Statistics (ONS), inpatient hospital records, renal registries and Covid Medicines Delivery Units (CMDUs). The dataset is granular, updated regularly and reflective of the McInnes high-risk group during the Omicron wave in the UK. The committee acknowledged that this analysis of OpenSAFELY was done well and made efforts to account for confounding bias when possible. The analysis was done in a dynamic environment with changing treatment practices and linkages with various data sources which can increase risk of confounding bias. The committee noted the results of Hill and Mirchandani (2022) that compared the outcomes of a randomised controlled trial with non-randomised studies on COVID-19 treatments. The authors questioned the validity of non-randomised studies when their outcomes contradict the outcomes from a randomised controlled trial. The authors cautioned against using non-randomised evidence independent of randomised evidence for regulatory decisions. The committee was willing to accept the OpenSAFELY data on relative treatment effectiveness as supplementary evidence to the trial evidence and for modelling estimates for hospitalisation rates. The committee cautioned against solely relying on non-randomised evidence when making conclusions on treatment effect.

Generalisability of trial evidence to current endemic context

3.14 The committee acknowledged that most trials informing the clinical efficacy data pre-dated the Omicron variant, which was the dominant circulating variant of concern at the time of this evaluation. Clinical experts said extrapolating data from past trials was misleading because epidemiology and virus characteristics have changed (see section 3.2). The clinical experts and the committee considered it appropriate to consider how the clinical evidence would generalise to the endemic setting. It considered the main generalisability concerns to be:

In vitro evidence

3.15 In vitro (laboratory) evidence may provide additional information on whether there is a realistic clinical possibility that a treatment retains efficacy against currently circulating variants. In vitro neutralisation assays can be used to assess if treatments can neutralise new variants, which can then be used to infer whether they retain clinical effectiveness over time as the virus evolves. An advantage of in vitro evidence is that it can be generated much faster than clinical trial evidence. A large body of in vitro evidence suggests that specific COVID-19 treatments may no longer show neutralisation activity against some circulating Omicron variants. In the first meeting, the committee could not comment on the validity of in vitro data and welcomed comments in response to consultation. Because of this, NICE commissioned an 'in vitro expert advisory group' made up of experts in infectious disease, virology, vaccine epidemiology, immunology and pharmacology (see the in vitro expert advisory group report in the committee papers for TA878). The group developed a decision framework to link in vitro neutralisation data to clinical outcomes and helped the committee use the framework to interpret the in vitro evidence. The committee understood this framework and also noted the latest in vitro evidence.

3.16 The in vitro evidence considered by the committee was against newly circulating variants and was available shortly before the second meeting. Because the COVID-19 landscape is rapidly evolving, a systematic review of the in vitro data was not possible. One in vitro study (Imai et al. 2023) investigated the effectiveness of the antivirals against BQ.1.1 and XBB. This showed that some antiviral treatments retain the ability to neutralise a range of SARS-CoV-2 variants and subvariants, including those circulating at the time of this evaluation and that are rapidly increasing in prevalence (at the time of the second meeting). The committee also considered the in vitro evidence that was systematically collected and summarised by multiple organisations including the 'Stanford Coronavirus Resistance Database'. For further details on the in vitro evidence, see the in vitro slides in the committee papers. By using the framework and the evidence, the committee concluded that there was no in vitro evidence showing reduced clinical efficacy of nirmatrelvir plus ritonavir across the variants tested.

Relative treatment effects

3.17 For the mild COVID-19 setting, the clinical experts considered the relative treatment effects of each treatment to be uncertain without considering the wider context of the trials (see section 3.2). The committee noted the potential for bias in all the comparisons because the indirect comparison used pairwise analysis rather than a network to produce its comparisons. They also noted that multiple interventions could be required and cautioned against the side-by-side comparison of treatment effects (as a fully incremental analysis). The committee considered that the heterogeneity of trial outputs and generalisability contributed greater uncertainty to the decision problem.

Model structure and key drivers of cost effectiveness

3.18 The economic model for this appraisal was developed by the AG and informed by a publication (Rafia et al. 2022) that evaluated COVID-19 treatment in a pre-hospital setting. The AG used a decision tree model structure for treatments in the mild COVID-19 (non-hospital) setting that joined with a partitioned survival model in the severe COVID-19 (hospital) setting. The decision tree had either an active treatment or standard care arm offered to people with COVID-19. People were hospitalised at a baseline standard care rate, or not hospitalised. Those that were hospitalised entered the partitioned survival model. This section of the model had 3 mutually exclusive health states: discharged from hospital and alive, hospitalised with or without COVID-19, and death (from COVID-19 or any other cause). For people in hospital, the level of respiratory support was assumed based on COVID-19 severity, with associated costs and disutilities by health state. The clinical inputs for each of the clinical efficacy scenarios were from the indirect treatment comparison (see section 3.12). The AG fitted parametric distributions to long COVID data from the ONS. Consultees highlighted that the long COVID duration was underestimated and should be higher than the 108.6 weeks used by the AG. In response, the AG updated the model which estimates that 30% of people will still have symptoms at 2 years, 10% at 5 years and 3% at 10 years. The AG assumed that 100% of people in the severe COVID-19 setting and 10% in the mild COVID-19 setting would have long COVID. Consultees noted that the proportion should be reduced for the severe COVID-19 setting and increased for the mild COVID-19 setting. The AG considered its original assumption to be conservative and therefore appropriate because alternative evidence was not available at the time of the second meeting. The committee noted that the treatment efficacy was highly uncertain and the most important driver of cost effectiveness, but also noted the following other key drivers of model outputs:

Hospitalisation rates

3.19 The rate of hospitalisation is a key driver of model outputs (see section 3.18) with multiple potential evidence sources. Hospitalisation rate is one of the key model input variables that define the group at high risk. To closely align with the marketing authorisations, for the first meeting the AG used a hospitalisation rate of 0.77% from PANORAMIC in its base case to generate the decision-making incremental cost-effectiveness ratios (ICERs). PANORAMIC was reflective of the current COVID-19 landscape, including the Omicron variant. However draft guidance consultation comments further highlighted that PANORAMIC would have excluded people at higher risk who were eligible for treatment through UK interim clinical commissioning policies (see section 3.4). Consultees provided a range of hospitalisation rates identified through targeted reviews. The committee saw overall hospitalisation rates defined by the McInnes high-risk definition including: OpenSAFELY 2.41% (untreated but eligible using McInnes definition), 1.37% (untreated but eligible group without contraindications to nirmatrelvir plus ritonavir) and 2.82% (DISCOVER-NOW database, UK observational study of people covered in the McInnes report). Hospitalisation rates also varied across different conditions, including between 4.15% and 4.40% for advanced kidney diseases and 15.90% (study of people with primary and secondary immunodeficiency [Shields et al. 2022]). In the first meeting the clinical experts agreed, given the committee's preferred definition of high risk (see section 3.7), that 0.77% could be an underestimation because the highest risk group may have been underrepresented in PANORAMIC (see section 3.4). They acknowledged the difficulty of determining hospitalisation rate without analysing the baseline population and all appropriate groups at risk. The rate is likely to vary substantially based on types of underlying conditions in the high-risk group, with potentially higher rates for severely immunocompromised people, such as people who have had a transplant and people having chemotherapy. The committee acknowledged significant uncertainty in estimating the hospitalisation rate for the population who have high risk of progression to severe COVID-19. Based on the strength of the evidence it concluded that it was likely to fall between the underestimate of PANORAMIC at 0.77% and the estimate of 2.82% from the DISCOVER-NOW database. The committee concluded that the hospitalisation rate for the McInnes high-risk group is between 2.41% and 2.82% based on OpenSAFELY and DISCOVER-NOW. For people contraindicated to nirmatrelvir plus ritonavir the hospitalisation rate is assumed to be about 4% as an upper limit using advanced renal disease as proxy from OpenSAFELY.

Assessment group's analysis for partial review

3.20 In the partial review, the broader population of people aged 70 and over was represented in the AG's model by adjusting the mean age and baseline hospitalisation rates. These were both informed by data provided in confidence by the PANORAMIC trial team, so cannot be reported here. It was noted that changing the mean age for hospitalised and non-hospitalised patients had little impact on the cost-effectiveness estimates. The AG modelled 3 scenarios:

Company analysis for partial review

3.21 The company had provided a range of cost-effectiveness analyses for different populations. It did not include an analysis for people aged 70 and over. Instead, it presented an analysis of people aged 70 and over and 18 to 69-year-olds with at least 1 pre-existing condition, which the committee considered most closely resembled the decision problem of the partial review. The company used the baseline hospitalisation rate from PANORAMIC for people aged 70 and over and 18 to 69-year-olds with at least 1 pre-existing condition. It had also assumed that the relative risk of hospitalisation or death with nirmatrelvir plus ritonavir was 0, because there were no events in the age 70 and over subgroup of the nirmatrelvir plus ritonavir arm in the EPIC‑HR trial. The AG noted this was methodologically incorrect because it assumes there will never be any COVID‑19-related hospitalisations or deaths for people aged 70 and over having nirmatrelvir plus ritonavir. It noted this should have a continuity correction in the model to adjust for small numbers of events. In addition, because the company's analysis included people aged under 70, the mean age modelled was much lower than that modelled by the AG. The AG noted that this would reduce cost-effectiveness estimates because more quality-adjusted life years (QALYs) will be gained per death avoided. The committee concluded that the analysis provided by the company was not relevant to the decision problem of the partial review and so preferred to use the AG's cost-effectiveness estimates in its decision-making.

Time to discharge

3.22 The amount of time spent in hospital is a key driver of cost effectiveness because of hospitalisation costs. Evidence on each treatment showed a relative reduction in time spent in hospital. The AG had previously noted the time to discharge evidence was collected during the early stages of the pandemic, which could lead to substantial generalisability concerns because the context of care has changed in the endemic setting. The committee noted that in clinical practice, time to discharge can sometimes overestimate time in high-cost health states because it can depend on multiple factors (for example, waiting for a negative COVID-19 test). Time to discharge was also considered more important for people who are being discharged to a care home. The committee also noted that clinical experts in both meetings explained that people hospitalised with COVID-19 have very different symptoms at present (the time of this evaluation) compared with early stages of the pandemic. Also that the population is heterogeneous (see sections 3.2 and 3.3). The AG included scenarios that removed treatment effects on time to discharge and clinical improvement at 28 days to try and account for these potential uncertainties. At the first meeting the committee considered these scenarios to be plausible but conservative if treatments had effects outside of hospitalisation and mortality. The committee was not presented with additional evidence on time to discharge or clinical improvement and was uncertain about the treatment benefit in the endemic setting. The committee concluded it was reasonable to remove these treatment effects.

Utility value assumptions

3.23 The AG used UK age- and sex-adjusted utility values (EQ‑5D‑3L) for the baseline utility estimates in the model. The AG did not apply additional utility decrements in the mild COVID-19 setting for people who did not have long COVID. The age- and sex-adjusted UK general population utility estimates were used for this population instead. During consultation on the AG's draft report, stakeholders critiqued this assumption. They said this may not capture the full benefit of the treatments compared with standard care and disadvantaged community-based treatments. The AG agreed this was a simplified assumption, but scenario analysis showed it had limited impact on the final ICERs. The committee agreed with the AG's assumption and acknowledged the minor impact on the ICERs. For the severe COVID-19 setting, the AG used utility decrements from a recently published cost-effectiveness analysis of remdesivir (Rafia et al. 2022). The utility decrements were originally from a population with recurrent Clostridioides difficile infection and influenza. The same in-hospital utility decrements were also applied across ordinal scales 3 to 5. The ordinal scale was an 8‑point scale (1 to 8) used to define progression of COVID-19 severity in the model. During consultation, stakeholders critiqued the use of utility decrements from a non-COVID-19 population. An alternative approach for a utility study was proposed. The approach was to use COVID-19 severity-specific vignettes with EQ-5D-3L questionnaires completed by the UK general population. Some stakeholders also highlighted recent COVID-19 utility-specific systematic reviews that could be used. The AG said a vignette study would not be possible because of the restricted timelines. Across both settings, the AG did not find alternative COVID-19 utility decrements from the stakeholder-suggested systematic reviews. The AG used post-discharge long COVID utility decrements from Evans et al. 2022. The same utility decrement was assumed regardless of ordinal scale status at hospital admission. At AG report consultation, stakeholders suggested an alternative source of post-discharge utility decrements split by history of ordinal scale status. The AG explained that the model structure was unable to allocate post-discharge utility based on historical ordinal scale admission status. It also said that these utility decrements are only applied for the duration of long COVID and are not a key driver of ICERs. The committee agreed with the AG's rationale and the long COVID utility decrement assumptions.

Long COVID costs

3.24 In the first meeting, the AG assumed the annual per person management costs of long COVID to be comparable with chronic fatigue syndrome (£1,013). The clinical experts explained there were differences between people with long COVID who were in hospital compared with those not in hospital. People in hospital would be more likely to have severe complications that incur greater costs from multisystem complications. The AG considered the costs had minimal impact on the cost-effectiveness estimates because they were only applied for the duration of long COVID. But, it also provided scenario analyses with increased average yearly costs (£2,500). The committee agreed these scenarios had minimal effect on the cost-effectiveness estimates but considered that any new UK-specific evidence on long COVID costs should be included if available. During draft guidance consultation a consultee said the AG's base-case long COVID cost underestimates the true burden of long COVID and provided an alternative higher cost from Vos-Vromans et al. 2017. The AG accepted this new evidence and inflated the cost to £2,267 per year (to reflect 2021/2022). The committee agreed with the updated base-case value.

Administration costs

3.25 The AG did not originally include administration costs for oral or subcutaneous treatments. For intravenous treatments a cost of £221 was assumed based on NHS reference code SB12Z. After consultation, the AG updated the assumptions in the model with costs provided by NHS England. NHS England provided CMDU deployment costs for the administration of oral antivirals (£410) and neutralising monoclonal antibodies (£820). Some companies disagreed with using CMDU deployment costs because these include costs based in secondary care. However, future delivery may be in primary care, which would likely reduce these costs. The NHS England representative explained that the delivery of service is subject to change. In future, integrated care boards will be responsible for treatment delivery currently done by the CMDUs. They also noted that these costs were calculated before implementation of nirmatrelvir plus ritonavir which may increase resource use because of expected requirements to assess contraindications. During draft guidance consultation, consultees did not agree with the administration costs used in the AG base case. Some consultees said additional pharmacist per hour costs (about £352.49) should be added for assessment of nirmatrelvir plus ritonavir interactions with other treatments. Other consultees argued that the prescribing cost for nirmatrelvir plus ritonavir should be lower and between £75 to £117 because e-consultations and telephone triage options factor in the assessment of contraindications by clinicians already familiar with doing them. The AG explained that changes in administration costs can be evaluated by looking at differences in net monetary benefit. The committee considered the differences in administration costs in relation to the net monetary benefit outcomes, noting the uncertainty about future delivery models. In the partial review, the committee was mindful that with the end of free testing and the closure of CMDUs, there may be challenges around patient identification and delivering treatment to a broader population. These could have cost and resource implications for the healthcare system that were not fully captured in the model. Although these could not be quantified, the committee considered that they were likely to increase the cost-effectiveness estimates.

Hospitalisation costs

3.26 The AG used unit costs per hospital bed-day from the NHS National Schedule of NHS costs. During AG report consultation, the AG updated the costs for ordinal scales 3, 4 and 5 based on stakeholder suggestions. During draft guidance consultation, consultees said the approach to costing ordinal scales 4 and 5 underestimated the true cost. The AG agreed with the changes suggested and updated the costs. The final codes were as follows:

Nirmatrelvir plus ritonavir for mild COVID-19

3.27 ICERs and net monetary benefits were calculated for nirmatrelvir plus ritonavir in the mild COVID-19 setting for the McInnes group. The committee looked at the pairwise ICERs compared with standard care presented by the AG. The committee reviewed results for the low-, mean- and high-efficacy scenarios (see section 3.12). The committee noted its preferred assumptions to include combinations of the following:

The ICERs for nirmatrelvir plus ritonavir compared with standard care using a) mean and b) low efficacy treatment effect and a 2.41% hospitalisation rate were:

Based on the committee's preferred assumptions, it considered that nirmatrelvir plus ritonavir was likely a cost-effective use of NHS resources compared with standard care, for people with high risk of severe COVID‑19, as defined by the McInnes report's criteria. This includes people in hospital with mild COVID-19 who are at high risk of progressing to severe COVID‑19.

The committee also considered the mean- and low-efficacy scenarios using a hospitalisation rate of 0.77% from PANORAMIC which more closely approximated the marketing authorisation population for nirmatrelvir plus ritonavir. The ICERs were above £20,000 per QALY gained and nirmatrelvir plus ritonavir was likely not a cost-effective use of NHS resources in this broader lower risk population.

In the partial review, the committee considered the mean- and low-efficacy scenarios using the baseline hospitalisation rate from the PANORAMIC subgroup of people aged 70 and over and relative treatment effect from EPIC-SR. The ICER in the mean-efficacy scenario was £61,454 per QALY gained, and nirmatrelvir plus ritonavir was dominated in the low-efficacy scenario. The committee noted that the efficacy estimate from EPIC-SR had assumed no mortality benefit because it was not clear from the information available whether there had been any deaths in the trial. The committee accepted that this assumption may overestimate the ICER but noted that there is substantial uncertainty associated with any estimates from very few events. It considered that assuming a small mortality benefit, for example, as used in the EPIC-HR estimate, was not expected to reduce the ICER below £20,000 per QALY gained. The committee also considered the scenarios using the PANORAMIC hospitalisation rate and relative treatment effect from the COVID-NMA including EPIC-HR. In both the mean- and low-efficacy scenarios the ICERs were above £20,000 per QALY gained. The committee noted that these were likely to be underestimates given concerns about the generalisability of the EPIC-HR data to this population (see section 3.17). The committee concluded that nirmatrelvir plus ritonavir was likely not a cost-effective use of NHS resources in the broader population identified in the Edmunds report.

Uncaptured benefits

3.28 Clinical experts said hospitalisation and mortality rates are becoming less relevant clinical efficacy measures for COVID-19 treatments. They explained this was because of the changing COVID-19 landscape (see section 3.2). In future COVID-19 evaluations, higher QALY gains or cost savings could be captured if the model includes the impact of treatments on the following outcomes:

Equality issues

3.29 The committee considered potential equality issues, including:

Addressing health inequalities

3.30 The committee noted the equalities issues outlined in section 3.29, and considered flexibility as part of the principles that guide the development of NICE guidance and standards. This emphasises the importance of considering the distribution of health resources fairly within society as a whole, and factors other than relative costs and benefits alone. It noted that the issues raised could affect some people with protected characteristics disproportionately which would contribute to health inequality. The committee said that in theory it would be willing to accept an ICER slightly more than what is usually acceptable if it addressed such health inequalities. However, it noted that departing from NICE's usual range needs to be done with caution, because it risks displacing funding from more cost-effective treatments elsewhere in the NHS, with an overall net loss of health gain. Even considering greater flexibility, the ICERs for nirmatrelvir plus ritonavir in a broader population than the McInnes group were substantially higher than what is considered a cost-effective use of resources.