TA1027 · STA

Tebentafusp for treating advanced uveal melanoma

Not recommendedTechnology appraisal committee AJuly 2023

Source documents

Intervention

tebentafusp (Kimmtrak)
· molecular bridge linking cancer cells to T cells, binding to CD3 on T cells and gp100 peptide-HLA complex · intravenous

Conditions

hla-a*02:01-positive unresectable or metastatic uveal melanomaoncology · metastatic
advanced uveal melanomaoncology · metastatic

Comparators

NameType Established Committee preferred
pembrolizumabactive drugYesYes
ipilimumabactive drugYes
nivolumabactive drugYes
dacarbazineactive drugYes
investigator's choicestandard of careYes
ipilimumab plus nivolumabactive drug
best supportive carebest supportive care

Clinical trials

TrialDesignPhasePivotal
IMCgp100-202RCTPhase 3Yes
IMCgp100-102single_arm2

Economic model

partitioned survival (company)
Time horizon: 38 years (lifetime)
Cycle length: not specified
markov
Time horizon: lifetime

ICER

Above £100,000 (tebentafusp vs pembrolizumab) · very_high uncertainty
Confidential (PAS-dependent) (tebentafusp vs investigator's choice (pembrolizumab subgroup)) · high uncertainty
Confidential (PAS-dependent) · very_high uncertainty

Methodological decisions (13)

comparator selection

Which comparator to use from investigator's choice arm (pembrolizumab 82%, ipilimumab 13%, dacarbazine 6%)

Company: investigator's choice reflected lack of standard care

ERG: all scope comparators should be included in model

Committee: pembrolizumab is most relevant comparator, use subgroup of 82% who had pembrolizumab

ICER impact: decreases

cost assumption

Subsequent treatment costs: company applied one-off cost at progression plus 4 months best supportive care (based on McKendrick et al. 2016); ERG preferred per-cycle costs while in progressive disease state

Company: One-off costs at progression appropriate. 4-month BSC duration based on McKendrick study

ERG: Per-cycle costs more appropriate. One-off costs inappropriate and 4-month duration not related to time in progressive disease state

Committee: Some uncertainty in estimates but limited impact on results. Clinical experts noted treatment usually stops at progression so costs would not accrue

ICER impact: negligible

cost assumption

Treatment adherence consistency: company assumed 92% adherence in tebentafusp arm but 100% in pembrolizumab arm

Company: 92% adherence applied to tebentafusp only

ERG: Adherence unlikely 100% in either arm. Should either apply same adherence to both arms or apply no adherence correction to either

Committee: Adherence would not be 100% for pembrolizumab; some adjustment should be applied to maintain consistency

ICER impact: negligible

cost assumption

Best supportive care costs modelled as one-off costs versus per-cycle costs

Company: One-off costs for best supportive care

ERG: Per-cycle costs for best supportive care

ICER impact: negligible

crossover adjustment

Whether to adjust for crossover from investigator's choice to tebentafusp in overall survival analysis

Company: crossover adjustment not feasible using standard methods; trial protocol did not mandate crossover; statistical analysis inappropriate due to too few crossovers

ERG: crossover adjustment preferred; company's approach was methodologically most robust as it mirrored clinical practice

Committee: accept company's unadjusted approach as methodologically most robust

ICER impact: uncertain_direction

model structure

Progression-free survival and time on treatment extrapolation: company used piecewise approach (Kaplan-Meier data then generalised gamma tail at 15% at risk); ERG preferred fully parametric generalised gamma for both arms

Company: Piecewise approach appropriate: rapid decrease phase followed by flattening. Data shows 15% cut-point appropriate for extrapolation tail

ERG: Fully parametric generalised gamma more appropriate. Kaplan-Meier may overfit and cut-points are arbitrary

Committee: Both approaches reasonable. Given mature PFS data, differences have little impact on cost-effectiveness results

ICER impact: negligible

model structure

Piecewise versus fully parametric progression-free survival and time on treatment extrapolation

Company: Piecewise extrapolation

ERG: Fully parametric extrapolation

ICER impact: negligible

survival extrapolation

Choice of parametric model for extrapolating overall survival beyond trial data

Company: 3-knot spline model for tebentafusp arm (captures change in survival profile) and Weibull for investigator's choice arm

ERG: standard parametric models (log-logistic or generalised gamma) applied to both arms preferred

Committee: standard parametric approaches most appropriate; acknowledged different approach in each arm reasonable given novel mechanism, but concerned about plateau suggesting cure

ICER impact: increases

survival extrapolation

Overall survival extrapolation approach: company used 3-knot spline for tebentafusp (allowing plateau) and Weibull for investigator's choice; updated to piecewise model for tebentafusp with different parametric models for each arm; ERG preferred standard parametric models (log-logistic or generalised gamma) applied to both arms without plateau

Company: Spline model with plateau justified by novel mechanism of action and atypical hazard profile in tebentafusp arm (hazard increases then decreases). Piecewise model with Kaplan-Meier data for early phase and parametric tail for late phase

ERG: Standard parametric curves should be applied to both arms. Plateau assumption is inappropriate as uveal melanoma is not curative disease. Log-logistic or generalised gamma curves preferred

Committee: Standard parametric approach should be starting point. Company's approach appears to overestimate long-term survival. Standard parametric extrapolation preferred for both arms

ICER impact: increases

survival extrapolation

Choice of parametric curve for extrapolating overall survival: company preferred piecewise approach, ERG preferred standard parametric curves (generalised gamma or log-logistic)

Company: Piecewise extrapolation for overall survival

ERG: Standard parametric curves (generalised gamma or log-logistic) for overall survival extrapolation

Committee: Standard parametric curves for overall survival extrapolation

ICER impact: increases

treatment effect duration

2-year stopping rule: company included rule because <5% remained on treatment at 2 years; ERG and committee opposed

Company: 2-year stopping rule appropriate because model predicts <5% of people still on treatment beyond 2 years

ERG: No stopping rule should be applied; no clinical justification for stopping effective treatment

Committee: No stopping rule appropriate; clinical rationale not justified. Patient and clinical experts noted treatment is well-tolerated and no evidence treatment effect would wane after stopping

ICER impact: decreases

treatment effect waning

Whether and how to apply adherence correction for pembrolizumab relative to tebentafusp

Company: 92% adherence in tebentafusp arm with different assumption for pembrolizumab

ERG: Either include adherence correction for pembrolizumab to maintain consistency with tebentafusp arm, or apply no adherence correction to either arm

Committee: Treatment adherence would not be at 100% for pembrolizumab and some adjustment should be applied

ICER impact: uncertain_direction

utility source

Approach to derive utility values: company used time-to-death approach based on Hatswell et al. 2014, applied utilities from NICE's pembrolizumab guidance for advanced melanoma. ERG preferred health state utilities differentiated by progression-free vs progressed states using EQ-5D data from IMCgp100-202

Company: Time-to-death approach more appropriate because disease progression not good marker of quality of life in tebentafusp patients. EQ-5D data from trial had insufficient observations by time-to-death category and high missing data requiring imputation

ERG: Time-to-death approach inconsistent with partitioned survival model structure. Should use progression-state-specific utilities from trial EQ-5D data despite imputation needed

Committee: Time-to-death approach not consistent with model structure but clinical/patient experts agreed disease progression not good QoL marker. Choice of approach unlikely to be important driver of cost-effectiveness results

ICER impact: negligible

Evidence gaps

immature overall survivalOverall survival data immature with most gains accumulated beyond observed trial data; model driven by extrapolation with high uncertainty
short follow upTrial ongoing with multiple data cuts, clinical experts uncertain about post-progression survival mechanisms after tebentafusp
immature overall survivalMost gains in overall survival accumulated beyond observed trial data. Model driven by extrapolation with high uncertainty about long-term survival
no uk dataLimited information on NHS treatment practice for subsequent treatments; some treatments in model may not reflect NHS practice
short follow upTreatment effect after 2 years unclear; no clinical data on whether benefit would continue after stopping at 2 years
immature overall survivalOverall survival data from IMCgp100-202 is highly uncertain; trial is still ongoing and direct trial data could help reduce uncertainties around overall survival
no uk dataSystemic anti-cancer therapy dataset could provide additional survival data

Commercial arrangement

simple discount pas · confidential

Special considerations

End of life criteria met Severity modifier applied Innovation acknowledged Equality issues raised

Cross-references

methodology precedent — NICE's guide to the methods of technology appraisal 2013 regarding end of life criteria
regulatory framework — NICE's Cancer Drugs Fund methods guide (addendum) and recently updated health technology evaluations process and methods manual