Enfortumab vedotin with pembrolizumab for untreated unresectable or metastatic urothelial cancer when platinum-based chemotherapy is suitable [ID6332]

The company estimated long-term overall survival in its base case by applying independently fitted models to survival data from EV‑302 for both treatment arms. It did so because it thought that the proportional hazards assumption may not hold and that this would be clearer if the trial data was more mature. It also noted that there was a difference in the mechanisms of action between the treatment arms and that the proportional hazards assumption did not hold for progression-free survival (see section 3.8). The company selected the log-logistic model for both treatment arms, based on statistical fit and how well the survival estimates aligned with the clinical expert opinion it had received. The EAG agreed with the company and applied independently fitted log-logistic models for both treatment arms in its base case. The committee heard from the company that applying the generalised gamma model for the platinum-based chemotherapy with gemcitabine arm was also plausible. This is because the generalised gamma model provided overall-survival estimates within the range suggested by the clinical expert opinion received by the company. The clinical experts present during the committee meeting thought that the 10-year overall-survival estimate of 5% generated using the log-logistic model for the platinum-based chemotherapy with gemcitabine arm was likely to be optimistic. They noted that the results of the generalised gamma model (3%) was more plausible. The EAG explained that the generalised gamma model was plausible but it is more methodologically appropriate to apply the same type of model for both treatment arms. It noted that using the generalised gamma model for both arms would increase the company's cost-effectiveness estimates. The committee was aware that the choice of overall-survival model impacts the severity weighting calculation (see section 3.14). It concluded that both the log-logistic and generalised gamma models were plausible and it would consider both in its decision making (see section 3.14).

The company considered that the proportional hazards assumption was not met for progression-free survival. So, to extrapolate progression-free survival it fitted independent models for both treatment arms. It noted that hazards data from the clinical trial showed a pattern of initially increasing hazards that then decreased. It considered that standard parametric models did not appropriately capture the changing hazards and could overestimate observed hazards (people whose disease progress). So it preferred to use spline models, which are more flexible, for its base case. The company modelled transformed versions of its survival data, that is, log cumulative hazard and log cumulative odds. Specifically, it applied a spline model to hazard with 2 knots for the enfortumab vedotin with pembrolizumab arm and a spline model to odds with 3 knots for the platinum-based chemotherapy with gemcitabine arm. The EAG noted that with the company's approach, progression-free survival becomes higher than overall survival in the enfortumab vedotin with pembrolizumab arm at about 8 years. So, the company had to apply constraints in the model to adjust this. The EAG preferred parametric models that did not require similar constraints. The EAG noted that the log-logistic model also followed a similar pattern of initially increasing hazards that then decreased. So, it preferred to apply log-logistic models for both treatment arms in its base case. The committee highlighted that the choice of progression-free survival models had a minor impact on the cost-effectiveness estimates. But it was concerned that the spline model resulted in progression-free survival being greater than overall survival in the enfortumab vedotin with pembrolizumab arm and that this was not plausible. So it concluded that applying the log-logistic models for both treatment arms was reasonable.

In EV‑302 people could have enfortumab until disease progression or unacceptable toxicity, whereas pembrolizumab could only be used for a maximum of 35 3-week treatment cycles. In its original base case the company noted that it had applied a stopping rule of 24 months (which it said approximated 35 treatment cycles) for pembrolizumab. In its revised base case, the company noted that because the time-on-treatment Kaplan–Meier curve for pembrolizumab was complete, it used this directly in its model. It explained that some people may have missed doses, which meant that their permitted maximum number of cycles would occur after month 24. The committee understood that the Kaplan–Meier curve for pembrolizumab was not properly applied in the company's revised model. This was because the 24-month stopping rule from the company's original base case was still functional. The committee was concerned that the Kaplan–Meier curve showed that some people were still having pembrolizumab treatment beyond 24 months. The NHS England Cancer Drugs Fund lead also noted that pembrolizumab had a stopping rule of 35 3-week treatment cycles and that this could be beyond 24 months. The committee concluded that the full Kaplan–Meier curve for pembrolizumab should be used and properly implemented in the model to reflect the stopping rule for pembrolizumab of 35 3-week treatment cycles.

People in EV‑302 could have 6 cycles of platinum-based chemotherapy with gemcitabine. Then, after a washout period, people whose disease did not progress could have avelumab maintenance treatment. In its base case model the company applied a 4.14-month (approximately 6 cycles) stopping rule for platinum-based chemotherapy with gemcitabine. It also applied a 60-month stopping rule for avelumab in line with TA788. To extrapolate long-term time on treatment for avelumab, it used a Weibull model. The EAG noted that the mean time on treatment estimated using the company's model, and the proportion of people on avelumab treatment at 1 year and 2 years, were higher than estimates from one of its clinical experts. The expert suggested that avelumab is normally used for less than 1 year (about 9 months) in the UK. The EAG preferred to apply the exponential model, which estimated the lowest mean time on treatment, but it acknowledged this was still higher than 12 months. The exact mean time-on-treatment data is considered confidential by the company and cannot be reported here. The clinical experts at the committee meeting explained that they could not confirm which of the extrapolation models (Weibull or exponential) provided the most plausible time-on-treatment estimates. They explained this was because avelumab maintenance had been recommended by NICE, and so available in routine practice, for less than 5 years (the maximum recommended treatment duration). The committee considered that it had not been presented with enough evidence to determine if either the company's or the EAG's models for time on avelumab reflected clinical practice. To reduce the uncertainty, it requested:

• further information on time on treatment for avelumab maintenance in the NHS, including mean time on treatment

• more justification for the choice of time-on-treatment model for avelumab maintenance.

The company assumed 30% of people who have platinum-based chemotherapy with gemcitabine would have avelumab maintenance treatment. It based this estimate on its clinical trial evidence. The EAG also used the same estimate for its base case. The clinical experts at the committee meeting estimated that around 60% to 70 % of people would be eligible to have avelumab maintenance treatment but that a proportion of these (about one-third) would choose not to have treatment. But they noted that the proportion of people having avelumab maintenance treatment varies across the NHS. The NHS England Cancer Drugs Fund lead noted that around 400 people in England who are eligible for immunotherapy currently have avelumab maintenance treatment. They agreed with the clinical experts that use of avelumab maintenance treatment was currently variable across the NHS. The clinical experts noted that the proportion of people who had avelumab maintenance in the clinical trial likely reflected its availability at the trial sites, not people's willingness to have it. They also explained that the overall-survival estimates would be lower if fewer people were to have avelumab than in the trial. The committee noted that the proportion of people having avelumab maintenance impacts both the clinical outcomes and the total costs in the comparator arm, and therefore the cost-effectiveness estimates. It concluded that the proportion of people on avelumab from the clinical trial (30%) was plausible and suitable for decision making but it would value further evidence relevant to the NHS, where available.

The company's base case model did not include a treatment-effect-waning assumption. The company explained that enfortumab vedotin was given until disease progression or unacceptable toxicity, and that some people are expected to have long-term treatment. It also noted that because independently fitted overall-survival hazard models had been applied for both treatment arms, any treatment effect waning would already be incorporated. The EAG also did not apply a treatment-effect-waning assumption in its base case. It cited Taylor et al. (2024), which suggests that if independently fitted hazard models gradually converge, treatment effect waning might already be accounted for in the model without an explicit treatment-effect-waning assumption being included. The EAG explained that the overall-survival hazard models for enfortumab vedotin with pembrolizumab and platinum-based chemotherapy with gemcitabine gradually converge over the 30-year time horizon. The EAG also noted that although pembrolizumab is stopped after 2 years, it is not clear if this would lead to treatment effect waning. It presented a range of scenario analyses that apply a treatment-effect-waning assumption for pembrolizumab but noted that these scenarios may overestimate the impact of treatment effect waning. That is, the scenarios may be biased against enfortumab vedotin with pembrolizumab. This is because people have treatment as a combination. So the effect of pembrolizumab may wane but people would still be having enfortumab vedotin, which may still provide benefit. The clinical experts explained to the committee that both treatments work synergistically to produce long-term immunological change in the body. They explained that there is some evidence showing long-term effect with pembrolizumab in people who had stopped treatment because of toxicity after about 10 months of initial treatment. The committee recalled that a proportion of people in the trial who had enfortumab vedotin with pembrolizumab remained progression-free beyond 24 months. It acknowledged the difficulty in plausibly separating a waning assumption for each treatment in a combination. It also took into account the EAG's and company's perspectives regarding inherent treatment waning in the model. It concluded that there is uncertainty in applying a treatment waning assumption for pembrolizumab in this case.

The committee considered the severity of the condition (the future health lost by people living with the condition and having standard care in the NHS). The committee may apply a greater weight to QALYs (a severity modifier) if technologies are indicated for conditions with a high degree of severity. For its base case, the company provided absolute (8.18) and proportional (0.83) QALY shortfall estimates in line with NICE's health technology evaluations manual. The EAG also estimated absolute (7.94) and proportional (0.84) QALY shortfalls for its base case. Both base cases were below the threshold for a QALY weighting for severity (absolute shortfall of 12.0 or proportional shortall of 0.85) to be applied to the QALYs. The committee understood that the shortfall estimates were based on results from EV‑302. The company explained that in EV‑302 people in the platinum-based chemotherapy with gemcitabine arm could have subsequent treatments (such enfortumab vedotin monotherapy, erdafitinib and sacituzumab) that are not recommended in the NHS. This could have improved their overall survival beyond what would be expected in the NHS. So, this makes the severity modifier calculation uncertain. The company explained that it had not collected sufficient data about when people switched treatment in the clinical trial to allow it to remove the treatment effect of these non-standard treatments from the overall-survival curves. It also considered that using real world evidence such as the Systemic Anti-Cancer Therapy data to inform overall survival for people having platinum-based chemotherapy with gemcitabine would be flawed. It noted that NICE has only recently (in 2022) recommended avelumab maintenance treatment for people whose disease has not progressed after platinum-based chemotherapy. The company said that because of this, the current overall-survival data is not mature enough to include the full impact of avelumab availability in the NHS. The EAG agreed with the company. It noted that overall survival for people having platinum-based chemotherapy with gemcitabine is likely overestimated in the model, but the magnitude of the difference on the QALYs is uncertain. The company reported that it consulted an additional clinical expert. They agreed with the EAG's expert (see section 3.10) that avelumab maintenance is normally given for less than a year in the NHS. The company noted that avelumab was given for longer in the trial. So, the company argued, the overall survival in the trial was likely further overestimated for people having platinum-based chemotherapy with gemcitabine. The clinical experts at the committee meeting noted that there are now treatment options such as atezolizumab and avelumab maintenance available in the NHS. These treatments make it plausible for people on standard care to have overall survival of around 16 months, as seen in EV‑302. The company provided additional data showing that overall survival was lower for people who had subsequent treatments with taxanes (such as paclitaxel) compared with enfortumab vedotin monotherapy. But the data did not include the impact of other subsequent treatments that are relevant to the NHS, such as atezolizumab, sacituzumab and erdafitinib. Finally, the company explained that 5 of the 7 overall-survival extrapolation models explored for the platinum-based chemotherapy with gemcitabine arm estimated proportional QALY shortfall values between 0.85 and 0.87. It argued that this suggests a QALY weighting for severity should be applied. This included the generalised gamma model, which both the company and EAG noted were plausible (see section 3.7). The committee commended the company's transparency in selecting its base case overall-survival model regardless of the resulting severity weighting. But the committee understood that the company's model included a round up function. This meant that its proportional QALY shortfall estimate using the generalised gamma curve was slightly below the 0.85 threshold. But, proportional QALY shortfall values of 0.85 to 0.87 were estimated in the EAG's base case model for 5 of the 7 overall-survival models for the platinum-based chemotherapy with gemcitabine arm, regardless of rounding up. The committee questioned the interchangeability of the overall-survival models. It heard from the EAG that there are different hazard functions underpinning each model, which would need to be considered individually. But the EAG reiterated that the generalised gamma model was plausible (see section 3.7). The clinical experts also reiterated that that the 10-year overall-survival estimate predicted using the generalised gamma curve (3%) was plausible (see section 3.7). The committee recalled that a substantial burden of disease on people with urothelial cancer and their carers was reported by the patient experts (see section 3.1). The committee noted that the point estimates of proportional QALY shortfall showed that a QALY weighting of either 1 or 1.2 may be plausible. This was because extrapolating overall survival with either log-logistic or generalised gamma models was reasonable (see section 3.7). To reduce the uncertainty, the committee requested additional evidence on the overall survival of people having standard care treatments that are generalisable to NHS practice. This should include the impact of avelumab maintenance treatment. But, where unavailable, additional evidence on overall survival for people having platinum-based chemotherapy without avelumab maintenance may also be considered to better understand the potential impact of modelling avelumab maintenance on the severity modifier.