Modular update to NICE manuals EQ-5D-5L value set: Evidence review document

EEPRU purposively selected 39 decisions from 37 technology appraisals, published between 2016 and August 2024, that reported using EQ-5D-3L utility values in the economic model. The selected case studies broadly reflect the range of disease areas considered by NICE, but the sample was not formally designed to precisely mirror our portfolio of guidance topics. They assessed how using the new 5L value set instead of the 3L value set would have affected estimates of QALYs and ICERs. HST decisions were excluded because of the relatively small number of published evaluations and because they are usually for childhood diseases, whereas the EQ-5D-5L measure and its value set were designed for use in adults only. Also, there is often uncertainty in utility estimates and other model assumptions and resulting cost-effectiveness estimates for HST topics because of the difficulties with evidence generation for rare diseases. HealthTech evaluations were excluded because utility estimates are usually sourced from the literature, where it is often unclear which version of the EQ-5D was used (3L or 5L).

EEPRU found that using the EQ-5D-5L value set increased the total QALY gains for both the intervention and the comparator in all appraisals reviewed. If the increase in QALY gains for the intervention was greater than the increase in QALY gains for the comparator, there was an increase in the incremental QALY gains and consequently a decrease in the ICER. When the increase in intervention QALY gains was less than the increase in comparator QALY gains, the ICER increased.

The impact on cost effectiveness differed between cancer medicines and medicines that treat non-cancer conditions. It also depended on whether the intervention's health benefits were driven by extending how long people live or improving health-related quality of life (table 1).

• Cancer medicines, which represent over half of NICE's technology appraisals portfolio (table 2), became more cost effective with the 5L value set: ICERs decreased by a median of 12%.

• Medicines that did not affect how long people with non-cancer conditions live, which represent approximately 20% of NICE's technology appraisals portfolio, became less cost effective: ICERs increased by a median of 59%. The technology appraisals in this category were treatments for migraine, chronic sialorrhoea, ulcerative colitis, obstructive sleep apnoea, prurigo nodularis, hidradenitis suppurativa, alopecia areata, atopic dermatitis and plaque psoriasis.

• For medicines that helped extend how long people with non-cancer conditions live, which represent just over 20% of NICE's technology appraisals portfolio, the results were mixed: some medicines became more cost effective and some became less cost effective. In most cases (7 of the 11 recommendations), ICERs decreased. Across all medicines in this category the median change in ICERs was a 9% decrease, based on 11 ICERs from 10 technology appraisals (1 topic included 2 separate populations, meaning 2 separate ICERs were reported).

In EEPRU's analysis, the largest impacts of the 5L value set were seen for treatments that did not affect how long people lived: the ICERs for the medicines analysed increased by an average of 59%. We recognise that this is a substantial impact on cost-effectiveness estimates, but caution that the estimate for this category of medicines is uncertain because the sample size was small (11 ICERs from 10 published technology appraisals) and cannot be generalised to all quality-of-life improving interventions. The impact of using the 5L value set is likely to be different depending on the size of the treatment benefit and which aspects of health-related quality of life are improved with treatment (as measured by the 5 dimensions in the EQ-5D). Medicines can improve health-related quality of life in many ways, so no sample would be able to fully represent all such medicines. It is also important to note that the incremental QALY gains were very small for most of the treatments in this category. This means that ICERs are very sensitive to changes in the costs included in the health economic model, such as the medicine's price. Even a small reduction in medicine price, or other costs, could substantially reduce the ICER. Confidential commercial data held within NICE suggests the relationship between ICERs and prices is such that we do not anticipate a significant negative impact on NICE recommendations and patient access to these types of treatments.

The DSU used 2 samples to examine the impact of adopting the EQ-5D-5L value set on the application of severity weightings. The first sample (referred to from here as the 'main sample') was the same 39 decisions used in EEPRU's analysis of the impact of the 5L value set on cost-effectiveness estimates. The distribution of AS and PS levels (the 2 criteria for qualifying for the severity modifier) across these 39 decisions was broadly representative of that seen in NICE's technology appraisals decisions. The AS and PS distributions were similar to those of the sample used in the September 2024 review of the implementation of the severity modifier, as well as recent monitoring data on the uptake of the severity modifier (measured by the cumulative mean QALY weight used in technology appraisals published since January 2022). Ensuring representativeness was important, because a skewed sample could bias the estimates of the impact of the 5L value set on AS, PS or overall severity weightings. The DSU's main sample was supplemented by a second ('sensitivity') sample of 18 appraisal decisions, totalling 57 decisions. The additional 18 decisions were selected because they either met the criteria for the severity weighting or were close to the cutoffs for qualifying for the weighting. This sample contained a higher proportion of decisions for severe conditions than the main sample, as shown by the mean severity weights for each: the mean severity weight in the sensitivity sample was higher (1.267 compared with 1.133 in the main sample).

The DSU analyses showed that moving from the 3L to the 5L value set is unlikely to change the severity weighting in most technology appraisal decisions. This is because using the 5L value set appears to have a greater impact on AS than PS, but the severity weighting in most technology appraisal decisions is determined by PS. Although large reductions in AS can occur when using the 5L value set (the DSU identified 2 decisions with large AS decreases of 3 QALYs or more), such changes would not always alter the severity weighting used in decision making. This is partly because the cutoffs that determine whether a treatment qualifies for the severity weighting span broad ranges of AS and PS. A treatment will qualify for the lower severity weighting (where QALYs gained are multiplied by 1.2) if AS is anywhere between 12 and 18 QALYs. The higher weighting (where QALYs gained are multiplied by 1.7) is applied when AS is at least 18 QALYs, and no weighting is applied when AS is less than 12 QALYs. In only 1 of the 2 cases identified by the DSU did the large reduction in AS lead to a change in the severity weighting.

For cancer medicines, AS consistently increased when using the 5L value set (albeit by a relatively small amount: mean 0.55 QALYs), whereas PS was largely unchanged. The pattern for medicines that treat non-cancer conditions was less clear: both AS and PS usually decreased, with variable magnitude.

Larger reductions in both AS and PS were seen in technology appraisals for people with younger average starting ages in the cost-effectiveness models. This seems to suggest that adopting the 5L value set may be less favourable for younger people than for older people. That is, younger people may be less likely to receive the severity weighting. But, the DSU analysis did not identify any evidence that the severity weighting is systematically worse for younger people when using the 5L value set. In the 2 technology appraisal decisions that had a lower severity weighting with the 5L value set, 1 was in younger people (average starting age of 4 years) and the other was in older people (average starting age of 60 years).