Background
Economic evaluations are an increasingly important aspect of drug development. The cost-utility analysis of a novel therapy is typically assessed in terms of cost per quality-adjusted life-year (QALY) gained, where QALYs reflect both survival and health-related quality of life (HRQoL) [
1,
2]. Health-state utility values, which represent an individual’s preference for different health outcomes as captured by patient questionnaires, are commonly used by researchers to calculate QALYs. These values are generated by applying an appropriate utility algorithm to the patients’ responses to the questionnaires, and are depicted as an interval scale in which 1 represents perfect health and 0 reflects a health state equivalent to death [
1,
2]. The most common type of questionnaire used to calculate health-state utility is the EuroQol-5 Dimensions (EQ-5D), a simple questionnaire that takes only a few minutes to complete, and can be applied to patients with any disease type [
3]. By providing health-state utility values, preference-based instruments such as the EQ-5D allow health service providers with a means to compare QALYs across different patient groups and disease types, which can aid in decisions regarding broader healthcare resource allocation [
1,
2].
The measurement of HRQoL in the oncology setting is usually carried out using cancer-specific instruments rather than generic preference-based measures as they focus on relevant health problems, and tend to capture more clinically meaningful differences [
4,
5]. The use of core questionnaires, such as the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 (EORTC QLQ-C30) [
6] is standard practice, with these often supplemented with condition-associated modules. Preference-based measures of health status, however, are rarely used. Two recent trials evaluating novel therapies for patients with multiple myeloma, for example, assessed HRQoL using EORTC QLQ-C30 with or without its myeloma-specific module EORTC QLQ-MY20, yet neither trial collected EQ-5D data [
6,
7]. From an economic perspective, preference-based instruments are required as they can be readily applied at the population level for economic analysis. In the absence of data from preference-based instruments, researchers may utilize suboptimal, non-specific measures, such as clinical response levels, to derive utility values [
8].
Mapping algorithms is an alternative means of relating HRQoL scores from core and disease-specific modules to generic utility values. Several studies have demonstrated the feasibility of mapping EORTC HRQoL data to EQ-5D values in cancer patients [
4,
9‐
13], including those with breast cancer [
10,
12], prostate cancer [
13], esophageal cancer [
4], and gastric cancer [
11]. However data on the ability to map EORTC HRQoL data to EQ-5D in patients with multiple myeloma are limited [
14,
15]. The goal of the current study was to develop a mapping algorithm that uses HRQoL data from the EORTC QLQ-C30 (with or without QLQ-MY20 data) to estimate EQ-5D utility values in patients with multiple myeloma, to facilitate economic evaluation of novel therapies for multiple myeloma.
Discussion
This study reports the mapping of EORTC QLQ-C30 scores (with and without QLQ-MY20 scores) to EQ-5D utility values in patients with multiple myeloma. The ability to estimate utility values based on HRQoL data is beneficial in this setting since recent major trials of novel agents in multiple myeloma have not collected preference-based data [
6,
7]. Although the information on physical and mental health provided by EORTC QLQ-C30 and QLQ-MY20 questionnaires offer great insight to clinicians, the questionnaires are not pertinent to cost-utility analyses as they are not easily translated into global health-utility values.
Mapping algorithms is an alternative means of relating HRQoL scores from core and disease-specific modules to generic utility values, and similar mappings have been developed in other cancer populations. In a study of 48 patients with gastric cancer, Kontodimopoulos and colleagues [
11] demonstrated the ability of QLQ-C30 scores to predict 15D, Short Form-6D (SF-6D) and, to a lesser extent, EQ-5D utility values. McKenzie and Van der Pol [
4] mapped QLQ-C30 scores to EQ-5D in 199 patients with inoperable esophageal cancer in a similar type of study using regression analysis techniques. In both studies, however, the model fit was somewhat lower (adjusted R-square 0.61 for both) compared with current mapping (adjusted R-square of 0.69). In a study of 280 patients with hormone-refractory prostate cancer, Wu and colleagues [
13] developed models to predict EQ-5D utility values based on QLQ-C30, Functional Assessment of Cancer Therapy-Prostate (FACT-P) scores, and patient demographics using ordinary least square regression analysis. However, this mapping can only be used when both QLQ-C30 and FACT-P scores are collected within the same study.
Two previous studies have reported mapping algorithms involving multiple myeloma. Rowen and colleagues [
14] derived utility values from QLQ-C30 scores in patients with multiple myeloma, but used a custom-designed preference-based measure (EORTC-8D), rather than the more generic EQ-5D. Versteegh and colleagues [
15] mapped QLQ-C30 scores to EQ-5D using a dataset derived from a HOVON trial of patients with previously untreated multiple myeloma (HOVON 24); the predictive value of this model was validated in a population of patients with non-Hodgkin lymphoma (HOVON 25). In this model, including only significant predictors, all other domains of QLQ-C30 were the same as in our study, however the model fit was not as good (R-square 0.51). Neither of the two mapping studies in multiple myeloma addressed the contribution of EORTC QLQ-MY20 scores.
The HRQoL scales identified in this analysis as significant predictors of utility values, such as Global Health Status/QoL, Physical Functioning, and Pain, are similar to HRQoL scales that have been pre-selected as clinically relevant in previous assessments [
7,
20]. Our final model included Global Health Status/QoL, Physical Functioning, Pain, Insomnia, and Future Perspective when both QLQ-C30 and QLQ-MY20 were used. Using QLQ-C30 alone, the final model also included Emotional Functioning alongside Global Health Status/QoL, Physical Functioning, and Pain. The adjusted R-squared value for both models was around 0.70, signifying a strong association and predictability. The high R-squared values and correlation between observed and predicted utility values of 0.84, suggest that the algorithm performs very well. A limitation to the above findings is the reverse association found between Insomnia and EQ-5D (i.e. worse insomnia associated with higher utility) when testing HRQoL domains from both the QLQ-C30 and QLQ-MY20 instruments. Unlike all other statistical associations, this is a counterintuitive finding. Insomnia was no longer a significant predictor when only QLQ-C30 domains were used as predictors for EQ-5D. Beside the almost equal predictability of HRQoL domains from the QLQ-C30 alone as compared to testing of HRQoL domains from both the QLQ-C30 and the QLQ-MY20, this provides an additional argument for using the QLQ-C30 as a stand-alone instrument for mapping onto EQ-5D. Table
2 also indicates possible floor effects on disease-specific symptom scores Dyspnea, Constipation, Appetite Loss, Diarrhea, and Nausea/Vomiting, all with median scores of 0.0, suggesting that these symptoms were not perceived at all by more than half of all patients. The five above-mentioned disease-specific symptoms are known to be reported in the context of novel treatments, and it is therefore possible that these scores in particular have been underreported in the context of this study. However, since our model displays comparably strong predictive power, independent of the symptom group chosen (see Tables
4 and
6), this does not reduce the predictability of the mapping algorithm presented here. Finally, further validation of the predictive performance of the algorithms presented here using external containing all three instruments is recommended.
Conclusion
The derived mapping algorithm establishes a link between EORTC HRQoL data and the EQ-5D as a utility-based outcomes measure specifically in patients with multiple myeloma. The results of the study are very encouraging in terms of the predicting power of mappings from QLQ-C30 scores alone or in combination with QLQ-MY20. This mapping algorithm will be a beneficial tool for deriving utility values from data obtained using the QLQ-C30 with or without the myeloma-specific instrument, and will thus enable to conduct cost-utility analyses. From a clinical perspective, the results may provide good guidance with regards to HRQoL domain selection for the purpose of primary analysis. Given the quantity of HRQoL data typically generated, a detailed discussion on all available HRQoL domains is often difficult to accomplish in the context of peer-reviewed publications. The results presented here provide an indication of the HRQoL domains that may be of particular interest in best describing a myeloma patient’s general health state and quality of life.
Acknowledgements
The authors received editorial support from Kathy Boon, PhD and Adriana Stan, PhD, of Excerpta Medica. The authors were fully responsible for the content of this manuscript and editorial decisions concerning it.
Competing interests
IP and JI are employed by Evidera which provides consulting and other research services to pharmaceutical, device, government, and non-government organizations. In this position, they work with a variety of companies and organizations; no payment or honoraria directly from these organizations for services rendered is received. PL is employed by and holds stock options in Celgene. The remaining authors have declared no conflicts of interest relevant to this study.
Authors’ contributions
IP, PL and CDW conceived and designed the study. CDW, KJ and FED collected and assembled data. PL, JI, CK, CDW, KJ and FED analyzed and interpreted data. CK, CDW and KJ assisted with the provision of materials and patients. All authors wrote and approved the manuscript.