Choice Modelling in the Medical Context
A new medicine is granted market approval if clinical trials can show an acceptable risk-benefit profile compared to a control. The therapy’s safety and its efficacy against the treated disease are assessed through clinical endpoints, which embody the study’s targeted outcomes. While survival remains the most significant outcome in oncology, health-related quality of life (HRQoL) has received increasing attention as a clinical endpoint.
HRQoL is addressed through patient-related outcomes (PROs), which are measures obtained directly from patients through questionnaires or interviews. This procedure consists in asking individuals to rate various items in order to define their perceived physical and mental health. However, the use of questionnaires alone does not reflect patients’ preferences. Therefore, it does not provide the possibility to perform cost utility analyses directly, which are useful in guiding the decisions of clinicians and payers.
Discrete choice modelling can be used to bridge that gap since it provides a translation of HRQoL data into a preference-based measure. When choice-based methods – the subject of McFadden’s Nobel Prize in 2000 – are applied to the field of healthcare, patients are faced with various potential health states and are forced to make trade-offs among them. This allows for a robust comparison of the HRQoL benefits of drugs within similar diseases and thus for an evaluation going beyond pure clinical effectiveness.
Conventional surveys vs. choice modelling
To derive cost-utility analysis from PROs, discrete choice modelling adopts a completely different strategy from that used in HRQoL questionnaires like the cancer-specific Quality of Life Questionnaire (QLQ-C30). For instance, this widely used survey lists 30 questions related to the quality of life of cancer patients, their symptoms and ability to function in everyday life. Patients respond to those items – three of them shown as examples below – based on scales ranging from very negative to very positive:
- “Do you have any trouble taking a short walk outside of the house?”
- “Have you lacked appetite?”
- “How would you rate your overall health during the past week?”
The answers to the QLQ-C30 reveal the patient’s perspective on the treatment benefits and are usually interpreted in combination with the clinical picture. QLQ-C30 describes well the effectiveness of a therapy across several dimensions. Nevertheless, the questionnaire does not render the importance of each dimension to patients since it is not based on their preferences. This makes it difficult to abstract the benefits of a medicine on the HRQoL of various patient cohorts and, ultimately, to perform economic evaluations.
In the discrete choice framework, patients are not asked to rate a series of individual items, but rather to make choices between different scenarios. To do so, several items are selected and given a rating in order to build possible health states. Every health state consists of specific trade-offs that are explored explicitly. Subjects are thus confronted with decisions and must indicate their preference among the compared health states (see figure below). This setting allows to infer the significance of the items listed in the HRQoL questionnaire. As a result, choice modelling lays the foundation of multi-attribute utility instruments (MAUIs), which capture health state utility values through patients’ preferences. Those values are mapped on a continuum ranging from full health (1), via death (0), to health states regarded as worse than death (negative values). On average, the health state scores of adults range from 0.7 to 0.9 and decrease with age.
Figure: Choice between two health states involving trade-offs (highlighted)
Informing decisions through quality of life
Combining treatment preferences with survival data yields the utility measure called quality-adjusted life-years (QALYs). This measure adjusts the years of life remaining to reflect life quality, 1 QALY corresponding to 1 year lived in a state of full health. QALYs have been the basis of cost-utility analyses for payers in health care for some time now. The most famous example is the recommendations made by the National Institute for Health and Care Excellence (NICE) to the NHS in the United Kingdom. QALYs are used to compare the cost of health services and medical technologies for various diseases in light of the benefits they confer. This method has the great advantage of putting an objective parameter at the centre of the controversial pricing discussion. In addition, the ability to evaluate the utility of health states also improves the capacity of clinicians to make well-informed decisions regarding treatment. For example, it may help them gain insight into the preferences of patient subgroups in specific settings. In the end, however, the choice of individual patients is naturally more important than global values.
Enabling the appraisal of health states, discrete choice facilitates the comparison of PROs among clinical trials. As such, choice modelling establishes PROs as a robust endpoint. This opens new perspectives for pharmaceutical companies, as it can help demonstrate the added value of a new drug compared to competitor drugs in terms of HRQoL more effectively. In fact, clinical studies involving PROs as primary endpoints improve a therapy’s chances of approval. In cases where a new drug would fail to show superior survival rates, it is still possible that HRQoL might be improved. That scenario could justify approval in certain clinical settings. Therefore, it is in the interest of pharmaceuticals companies to use discrete choice modelling to develop validated clinical PRO endpoints in collaboration with the regulatory agencies, and to incorporate them into their trials.
NICE requires HRQoL to be reported by patients and the measure of utility to be based on the preferences of the population. This method has been validated for the generic EQ-5D, but robust benchmarks of public preferences for many other questionnaires are lacking. Disease-specific instruments, which produce more sensitive HRQoL profile measures than universal questionnaires, still need to be validated and benchmarked. In that context, the MAUCa research consortium (Multi-Attribute Utility in Cancer) has been setup within the Quality of Life working group of the EORTC to determine a utility measure for the cancer-specific QLQ-C30. This classification system is known as the EORTC Quality of Life Utility Measure-Core 10 dimensions (QLU-C10D). The objective is to record the HRQoL preferences in the general population by asking participants to choose between scenarios, each consisting of a survival time and a health state. The QLU-C10D has been validated in Australia and data has been collected in other countries, such as Germany. Studies involving cancer patients and survivors will be run at a later point in time to add the utility values of that group to the picture.
In order to ensure compatibility among studies, research groups are working to establish a standardised methodology for assessing HRQoL. Ideally, the whole process of discrete choice experiments should be covered, which corresponds to the range of services SurveyEngine offers: from design and data gathering to modelling. This scientific groundwork is not going unnoticed, as the increasing number of publications on the topic indicates. Although general consensus is still a distant goal, ambitious programmes like the MAUCa project are paving the way for a reproducible methodology. This is an exciting time for the development of utility-based approaches for the calibration of HRQoL questionnaires. Because choice modelling can provide a reproducible measure of the treatment benefits from the patient’s perspective, it is gaining momentum in healthcare.