3A.5 Health outcomes

Page last updated: September 2016

3A.5.1 Health outcomes

Nominate and justify the final health outcome that is considered to best reflect the comparative clinical performance of the interventions and will be presented as the denominator unit in the base-case incremental cost-effectiveness ratio (ICER), consistent with the approach justified in Subsection 3A.1.2.

Detail the health outcome(s) (intermediate and/or final) that inform the final outcome in the economic evaluation and whether these were reported directly in the clinical evaluation (Section 2), and, if not, summarise the transformations involved to obtain the final outcome.

If available, use quality-of-life or utility data reported in Section 2 to estimate QALYs in the model, or, justify the use of alternative indirect methods to estimate QALYs when direct data are available. Present both sets of methods and results, and compare the interpretation.

Present the results of any utility study as the point estimate of the mean elicited utility weight for each health state, and include its standard deviation and 95% confidence interval, where available.

If a claim is made for a change in a nonhealth outcome, or the submission identifies health-related outcomes in people other than the patient receiving treatment (eg quality-of-life benefits for family, decreased carer burdens), do not include these in the base-case evaluation; rather, present them as supplementary analyses (see Appendix 6).

Use of quality-of-life data from the clinical trials to estimate QALYs

Estimates of quality of life or utility from the within-trial evidence (from Section 2) may inform direct estimates of QALY gains in the intervention and comparator populations, or inform utility values applied to health states in a cost-effectiveness model.

If a MAUI has been used in an included study to estimate utility weights (as described in Subsection 2.4.3), state where and when the scoring algorithm was derived, and consider how applicable it is to the general Australian population. It is preferred that Australian-based preference weights are used in the scoring algorithm used to calculate utility weights.

If the initial patient-reported outcome measure is not a MAUI, provide detail of the measure and justification of its use in Subsection 2.4.3. In this subsection, describe a validated method of mapping the results into preference weights (see below). State whether Australian-based value sets are incorporated. If there is no reliable method of transforming the patient-reported outcome data into utility weights for the model, describe why this is not possible and detail whether the patient-reported outcome data from the trial can still be used to inform or validate the economic model.

Consider the duration over which the patient-reported outcome measure informing utilities was administered compared with the duration of the condition of interest. If a generic MAUI or patient-reported outcome measure is used, consider whether it captures all important disease- or condition-specific factors that might be relevant.

Address the following questions when incorporating trial-based patient-reported outcome data into the economic model:

• Are the participants representative of the population for whom listing is requested? (Refer to Subsection 3A.3, as needed.)
• If quality of life is not the primary outcome, is the trial adequately powered to detect a difference in the survey results? As with all secondary outcomes, assess the results with reference to the conclusion from the primary analysis of the trial.
• Is there a ‘healthy cohort effect’? (ie where the sickest patients are least likely to complete patient-reported outcome data forms, and therefore the data obtained has a bias towards healthier patients.) Consider the responder numbers and drop-outs. While generally associated with an overestimate of utility weights, the direction of any associated bias may depend on whether the treatment and comparator are associated with different utilities, the relative extent of the effect across different arms and health states, and the time spent in different health states. Identify any impact on the overall ICER.
• Is there potential for systematic bias where progressed health states are defined by nonsymptomatic events (ie identified by investigations that may or may not reflect clinical practice)? Provide details.
• Is it appropriate to pool patient-reported outcome data across arms of a trial? This may be appropriate where patient numbers are small and for posttreatment states, but not in other circumstances where treatment (rather than disease or condition) directly affects quality of life (eg because of serious adverse events and any associated long-term implications, or imposed limitations). Justify the approach, and, where possible, present results with and without pooling.
• Is there a risk of bias from a regression to the mean effect?⁴⁷ This may be more likely in instances where quality of life for the control arm is drawn from a trial other than a randomised controlled trial (eg instance from a pre-intervention population).

Use of other sources of data to estimate utility weights

Where utility weights or QALY changes cannot be directly estimated from data collected in the clinical studies from Section 2, or there are significant concerns about the reliability and relevance of trial-based utility, transform the Section 2 health outcomes to estimate QALY gains (eg by applying utility weights to the time spent in different health states that represent the experience of clinical outcomes).

Additional studies (either published or done for the submission) may be needed to estimate utility weights for health states in the economic model. These studies should be identified (and copies provided) in Subsection 3A.2.1.

Describe the source(s) and method(s) (as described below) used to derive externally derived health state utilities, and justify their inclusion in the model. Depending on the clinical context and available data, there may be more than one acceptable source of utility weights. Where this is the case, reflect the uncertainty in selecting an optimal source of weights by reporting the sensitivity of the result to switching between the various sources of weights.

Address the questions regarding quality-of-life data derived from the clinical trials (above) that are applicable to any utility estimates obtained from alternative sources and methods.

Mapping of generic and disease-specific scales

Nonpreference-based patient-reported outcome measures will require a mapping algorithm to be transformed into preference-based measures to estimate utilities. Where this occurs, detail the source of the mapping algorithm. Describe the estimation sample (population demographic and clinical characteristics, sample size etc) and whether there is an external validation sample. Provide details of the source and target measures (eg index, dimensional), and the statistical model and methods used to estimate the mapping algorithm. Detail the statistical association or operations that constitute the algorithm. Discuss methods used to measure the algorithm performance and validity. Present the resulting predicted utilities with associated uncertainty. Discuss the applicability to the submission data, particularly in relation to the sample in which the algorithm was developed.

Scenario-based methods to indirectly elicit utility weights

Scenario-based methods use vignettes to describe the symptoms of a health state to a sample population, usually a representative general population sample, from which utility weights are elicited using an accepted preference-based method. Methods to elicit preferences include the standard gamble, time trade-off and discrete choice experiments, and other stated preference methods.

If using a scenario-based utility valuation to value health outcomes beyond the time horizon of the trial, include one or more health states captured and valued within the trial in the scenario-based study to validate the commonality of the trial-based and scenario-based utility weights.

Present supporting evidence for any claim of increased sensitivity of a scenario-based approach to identify real differences in utility.

Describe all stages of a scenario-based study in detail and explain efforts to minimise potential bias. It is difficult to minimise the many sources of analyst bias that are intrinsic to the scenario-based utility approach, including the nonblinded nature of the construction and presentation of the scenarios (eg incomplete inclusion and differential focus on alternative aspects of quality of life), the design of the methods to elicit values, and the analysis and interpretation of the results.

Population matching study method to indirectly elicit utility weights

This form of utility study involves the recruitment of a separate sample of patients with characteristics similar to those enrolled in the clinical trials reported in Section 2. Matched patients complete a MAUI reflecting their current health state, which informs the estimation of utility weights for the health states in the cost-effectiveness model. See Subsection 2.4.3 for further detail on MAUIs.

Potential sources of bias for such studies include the possibility of systematic differences between the clinical study participants and the matched patients, and the inability to blind the sampled patients from the objectives of the study. If there are important symptomatic medicine toxicities, the sampled patients should possibly have been exposed to the medicine and its toxicities at the time the MAUI is completed.

Matched patients should complete other patient-reported outcome measures that were completed by the trial participants, and the results of this concurrent instrument should be used to more closely match utility study participants to the clinical study population.

Published sources of utility weights

Utility estimates may be available from the literature. The validity of the derived utility weights depends on the applied elicitation methods and the relevance of the study populations. Present details of search strategies, and inclusion and exclusion criteria used to identify relevant utility studies. Assess the validity of all identified studies, including:

• how representative the health state in each identified study is of the health state in the economic evaluation (including the type and severity of symptoms, and the duration of the health state)
• how the health state was captured (eg MAUI, scenario based)
• how the preference was elicited (eg standard gamble, time trade-off)
• what sample was chosen to respond to the MAUI questionnaire or scenario (eg the general public, patients, carers, health care professionals)
• what assessment was made of the nature and direction of bias that might arise, given the sample and methods
• how the sensitivity analyses examined variation in the identified utility options.

Using different published studies to inform utility weights for alternative health states is discouraged because of the potential for inconsistency in the methods and populations from which utilities were derived.

Presentation of outcomes and health utility value information

If presenting a CUA, a format for summarising the minimum information on all modelled health outcomes (eg intermediate, final outcomes and events) contributing to the final health outcome in the economic evaluation, and any associated utilities or disutilities is suggested in Table 3A.5.1.

CI = confidence interval; QALY = quality-adjusted life year; SD = standard deviation