Validity and reliability of EQ-5D-3L for breast cancer patients in Korea

Breast cancer is the most frequently diagnosed cancer in females and the leading cause of cancer-related deaths in both economically developed and developing countries [1]. In Korea, the crude breast cancer incidence rate was 63.7 per 100,000 persons in 2011 and has been increasing since 1999 [2]. The 5-year survival rate for breast cancer patients has improved, with 91.3 % reported between 2007 and 2011 in Korea [2]. Recent years have led to a range of treatments for breast cancer, and development of additional therapeutics is continually in progress.

Health-related quality of life (HRQoL) and traditional clinical outcomes (i.e., survival rate or tumor responses) are considered as significant outcomes of cancer care [3]. Tools to assess HRQoL are classified into disease-specific instruments and generic instruments. Preference-based instruments (the Health Utility Index [4, 5], the EuroQol 5D (EQ-5D-3L) [6], or the Short Form 6D [7]) are generic instrument generating utilities that yield a measure that combines both the length and quality of life [8]. Disease-specific instruments can be more sensitive for detecting clinically important differences or changes in specific disease groups in general, although general instrument is occasionally not inferior to disease-specific instrument [9]. However, they do not compare HRQoL of patients with other diseases [8]. Resource allocation for health care has been a critical issue, not only for its effectiveness, but also for efficiency of care; hence, it is important for guiding healthcare policy.

The EQ-5D-3L, a generic preference-based instrument that is widely used to measure HRQoL, can be used to assign preference values to various health states [10]. Additionally, EQ-5D-3L is a unique instrument with quality weight tariffs for use in Korean populations [11]. The psychometric properties of the Korean EQ-5D-3L have been studied for rheumatic disease [12] and colon cancer [13] patients. However, studies on validity and reliability of the EQ-5D-3L in breast cancer patients in Korea are scarce [14]. Hence, we evaluated the validity and reliability of the EQ-5D-3L to measure the HRQoL in post-operative breast cancer patients from Korea.

Among 1,002 recruited patients with breast cancer, 175 patients were excluded from this analysis: 13 duplicated, 2 with missing EQ-5D-3L data, and 160 patients who did not merge into the cancer registry data. Thus, a total of 827 patients were included in the study cohort. Among 150 participants who completed a second survey 1 week later, 67 participants replied and 54 subjects were used for test-retest reliability after excluding 13 participants for whom clinical information was not available. Demographic characteristics of the survey participants were shown in Table 1. All study participants were female with a mean age of 52.1 years. Most patients were married. The number of participants who had undergone breast-conserving surgery was 524 (63.4 %).

The proportion of patients reporting any problems of the EQ-5D-3L and the FACT-B subscale score distribution in the study participants was shown in Table 2. The proportion of patients reporting any problems was relatively higher in the pain/discomfort and anxiety/depression dimensions at 43.5 % and 41.5 %, respectively. The mean FACT-G score was 78.7 (SD 15.8) and mean TOI was 66.7 (12.3). The EQ-5D-3L index and VAS score according to demographic and clinical features were shown in Table 3. In the highly educated and unmarried group, and not in current radiotherapy group, the EQ-5D-3L index scores were higher than corresponding scores of the other groups. A worse stage at the time of diagnosis was associated with lower HRQoL scores; however, this trend did not reach the threshold for statistical significance. By post-hoc analysis, individuals with < 1 year since the time of diagnosis had a significantly lower EQ-5D-3L index as compared with those who had > 4 years since their diagnosis. The EQ-VAS score were not significantly different according to demographic and clinical characteristics.

FACT-B subscale scores and summary measurements by dimension and level of the EQ-5D-3L were shown in Table 4. All FACT-B subscale scores and combined index of subjects with no problem were significantly higher than those of subjects with any problem in all EQ-5D-3L dimensions. Difference in PWB scores was larger than other FACT-B subscales in mobility dimension, whereas difference in PWB scores was larger in self-care and usual activities dimensions. Difference in EWB scores was the largest in the anxiety/depression dimension. Table 5 showed correlation coefficients among the EQ-5D-3L index, EQ-VAS score, and FACT-B scores. TOI scores showed the highest correlation with both the EQ-5D-3L index and EQ-VAS scores at 0.557 and 0.456, respectively, and the SWB scores showed the lowest correlation with the EQ-5D-3L index. Rates of agreements in the EQ-5D-3L dimensions ranged from 85.2 to 96.3 % across dimensions. Kappa statistics ranged from 0.32 in mobility to 0.70 in pain/discomfort and anxiety/depression dimensions. ICCs of the EQ-5D-3L index and EQ-VAS scores were 0.70 and 0.48, respectively (Table 6).

In this study, we aimed to assess the validity and reliability of EQ-5D-3L for measuring HRQoL in post-operative breast cancer patients. Our findings supported the validity of the EQ-5D-3L index. Values for this index were lower in the current radiation therapy, shorter duration of disease since diagnosis, and low education groups. Most of known-group validity features fulfilled our hypothesis. EQ-5D-3L indexes were lower for progressive stages at diagnosis and in the current chemotherapy group, although the differences in the EQ-5D-3L index between groups were not statistically significant. Our results corroborated the findings of Chae and Seo that patients who are younger, more highly educated, and with a longer time since diagnosis have a higher HRQoL [22]. HRQoL score differences by age group were not consistent with previous reports of HRQoL measurements in breast cancer patients [16, 22, 23]. Age has different impacts on HRQoL domains. Ahn et al’s study on breast cancer survivors showed better social and emotional function but poorer physical function with increasing age [24].

The FACT-B subscale scores for problem reporting in each EQ-5D-3L dimension were significantly different in the current analysis. The EQ-5D-3L index was moderately correlated with FACT-B subscale scores, ranging from 0.377 at FWB to 0.553 at PWB subscales, except for SWB. This finding was similar to those of Lee et al. who studied patients with breast cancer in Singapore [26]. In that study, a correlation between the EQ-5D-3L index and FACT-B subscales was the lowest at 0.11 on the SWB subscales and the highest at 0.66 on the PWB subscales [26].

Reliability of EQ-5D-3L in this study was inconclusive. Kappa values for EQ-5D-3L dimensions varied depending on domain and confidence intervals of kappa values were wide in mobility, self-care and usual activities. The ICCs of the EQ-5D-3L index was reasonable but EQ-VAS were somewhat low. Kappa values on mobility dimensions were the lowest at 0.32, but showed some agreement with the Landis and Koch criteria [25] at an agreement rate of 94.4 %. Low reliability statistics could be due to small sample size, change of health states between 2 time periods, and high proportion of no problem in mobility, self-care and usual activities dimensions. In the previous Singapore study, ICCs for the EQ-5D-3L index were reported to as 0.81 in the English version and 0.69 in the Chinese version [27]. Similarly, another Korean study of colorectal cancer patients reported kappa values for the EQ-5D-3L dimensions ranging from 0.32 to 0.56 and the ICC of the EQ-5D-3L index of 0.55 [13]. Compared to our present study, an earlier Taiwan study on cervical cancer patients reported somewhat higher kappa values for the EQ-5D-3L dimensions and the ICC of the EQ-5D-3L index [28].

The findings of the present study cannot be generalized since only postoperative breast cancer patients at 1 tertiary hospital were enrolled. Another limitation was that we did not investigate patient recurrence status, time since treatment completion, performance status at the visit day, or comorbidity. Because these parameters could affect the HRQoL, more comprehensive data are needed in the future to assess the psychometric properties of the HRQoL instruments. In addition, we did not assess the responsiveness of the EQ-5D-3L in breast cancer interventions; hence, further studies are warranted. Lastly, we did not reassess physical status at retest period, thus evidence on test-retest reliability was inconclusive.