Comparing the EQ-5D-3 L and EQ-5D-5 L: studying measurement and scores in Indonesian type 2 diabetes mellitus patients

A cross-sectional study was conducted from July 2016 to April 2017. A secondary care setting in South Sulawesi and two primary care settings in Central Sulawesi were included. In particular, these were Jaury Academic Hospital in Makassar and the Puskesmas/primary healthcare centers (PHCs) in Simpong and Kampung Baru in Luwuk Banggai, respectively. This study was approved by the Medical Ethics Committee of Universitas Gadjah Mada Yogyakarta, Indonesia (document number KE/FK/1188/EC, 12 November 2014, amended 16 March 2015).

Participants were T2DM outpatients with a minimum age of 18 years. The participants were informed of the study objectives and study procedure. The researcher or research assistants obtained signed informed consent forms from the participants. For the participants with disabilities or difficulties in reading, consent was based on confirmation from their caregiver who accompanied them during treatment at a health facility. The caregiver played a role in providing support to the participants as they filled in the instruments. It is important to note that all decisions on the exact health states chosen originated from the participants. In this study, all participants were treated by a consulting resident internal medicine who gave his/her consent to the data collection during the participant’s T2DM consultation (in primary and secondary care).

EQ-5D 3 L and 5 L consist of two parts: the EQ-5D descriptive system classification and the EQ visual analogue scale (EQ-VAS). The EQ-5D descriptive system comprises five items on its HRQoL dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Each dimension in the 3 L version [10] is completed with three response options: no problem, some problems, and confined to bed/unable/extreme problems, yielding a possible 243 (3⁵) unique health states. A single digit expresses the level selected for that specific dimension. Therefore, the five-digit number for five dimensions describes a specific health state. For example, ‘11111’ indicates ‘no problems on any of the five dimensions’, while ‘23231’ indicates ‘some problems walking, unable to wash or dress, some problems with performing usual activities, extreme pain/discomfort, and no anxiety/depression’. The 5 L [8] has five scale options to choose from: no problem, slight problems, moderate problems, severe problems, and extreme problems/unable. The 5 L instrument yields 3125 (5⁵) unique health states. For example, ‘12345’ indicates ‘no problems walking, slight problems washing or dressing, moderate problems doing usual activities, severe pain/discomfort and extreme anxiety/depression’. The EQ-VAS presents the participants’ self-rated health on a scale of 0 (worst imaginable health) to 100 (best imaginable health). The time frame for the EQ-VAS is ‘today’, meaning that participants were asked to describe their health state during the day they were interviewed. We used the 3 L and 5 L Bahasa Indonesia versions of the EQ-5D, produced by the EuroQol Group using a standardized translation protocol [17] and having been proved as valid and reliable questionnaires in Indonesian patient groups [13‐16].

After introducing the researchers and explaining the purpose of the study, a brief description to the participants was provided on how to use the EQ-5D instruments. An explanation of the concept of HRQoL as an aid on how they should describe their health state was presented. The participants were given the opportunity to ask questions throughout the data collection process. For EQ-VAS, we asked the participants to describe their health state and provide the most appropriate score to define their health state. Three research assistants were hired to collect the data. As a sequence, participants first classified their health state on the 5 L items, then provided their data (sociodemographic and clinical conditions), followed by the 3 L.

According to socio-demographic data (gender, age, T2DM duration, occupation, level of education, and dependence on a caregiver) were obtained from self-reporting. In this study, participants were classified into two age categories based on the retirement age of Indonesian people (56 years): productive age (below 56 years) and retirement age (56 years and above). As for employment status, participants were defined as in active employment when they were still actively working, and unemployed if they reported not having a job. Those whose main responsibilities were for their family members and household chores were classified as housewives.

Data on the clinical conditions, such as the type of therapy, T2DM-related complications, and comorbidities were obtained from treating physicians. Self-reported data from participants was used in the cases data could not be collected through the treating physicians. In this study, participants were defined as having comorbidities if they suffered from other diseases, such as asthma, gastritis and gout problems. Participants were defined as having complication and comorbidities if they suffered from other diseases and T2DM complications; for example, a participant with comorbid cancer and hypertension as a complication of diabetes.

Test-retest reliability was analyzed using sequential measurements. Participants involved in this phase were those who visited the specific health facility twice. The time interval between the two measurement times was four weeks as the participants were scheduled to meet their consulting resident internal medicine each month. Notably, an additional question was asked before the participants completed the instruments the second round: ‘Has there been any major change in your health state between the first time you completed the instruments last month and today? For example, have you been hospitalised, had an accident, experienced a natural disaster or have been bereaved’? Participants who answered ‘yes’ were excluded from the final sample.

For self-reported health state profiles obtained from the two versions of EQ-5D, we calculated the percentage of participants who responded to each level of each dimension. To determine the consistency of the participants’ answers, we checked the redistribution pattern, i.e., the consistency of individual participants’ scores in both versions. A consistent response pair was defined as a 3 L response which is at most one level away from the 5 L response (e.g., a participant chose level 1 in 3 L and chose level 2 in 5 L). When the 5 L level was more than 1 level away from the 3 L level (e.g., a participant chose level 1 in 3 L and chose level 3 in 5), this was labelled inconsistent [11]. Next, we converted their scores on 3 L to 5 L as follows: 1 in 3 L equals 1 in 5 L, 2 in 3 L equals 3 in 5 L, and 3 in 3 L equals 5 in 5 L [12]. The ceiling effect was defined as the proportion of participants who reported not having problems in any of the five EQ-5D dimensions (health state ‘11111’) for both 3 L and 5 L. This statistic is often used to assess the discriminatory power of health-state classification systems [18, 19]. As Indonesia only has the EQ-5D-5 L value set, not the 3 L [20], to obtain consistent 3 L and 5 L utility index scores, the UK 3 L and 5 L value sets [21, 22] were used.

The test-retest reliability of the dimension scores was assessed using the weighted kappa. We applied Landis JR & Koch GG standards [23] to determine the strength of agreement of the kappa values as follows: < 0.00 = poor, 0.00–0.20 = slight, 0.21–0.40 = fair, 0.41–0.60 = moderate, 0.61–0.80 = substantial, and 0.81–1.00 = almost perfect [20]. The test-retest reliability of the EQ-VAS and index scores were calculated using intra-class correlation coefficients (ICCs), two-way random effects and absolute agreements. The following reliability guideline was used for the strength of the ICC values: < 0.5 = poor, 0.5–0.75 = moderate, 0.75–0.90 = good and > 0.90 = excellent [24]. The discriminative power was calculated using the Shannon index (H′) and Shannon’s Evenness index (J’) [18, 19]. The Shannon index combines the absolute information content as expressed by the number of categories with the extent to which the information is evenly spread over these categories. On the other hand, the J’ expresses the relative information of a system or the evenness of the information distribution regardless of the number of categories. In case of an even distribution, when all levels are filled with the same frequency, J’ is equal to 1. Larger H′ and J’ values indicate more discriminatory performance. All the data were analysed using IBM SPSS Statistics for Windows version 23 (SPSS Inc., Cambridge, MA, USA), and statistical significance was set a priori at p < .05.

A total of 198 participants were interviewed (Table 1). The average age of the participants was almost 60 years, with 58% being female, and 70% of female participants reported being housewives as their main activity. Regarding the clinical conditions, more than 70% of participants were being treated with oral antidiabetic therapy (OAD), both monotherapy and OAD combinations, and 52% of participants reported T2DM-related complications. Furthermore, participants had various comorbidities, such as asthma (n = 6), gastritis (n = 5), and gout (n = 3).

Participants usually reported no problems (level 1) on both 3 L and 5 L, except for the pain/discomfort dimension with only 25 and 20% of participants reporting no problems on 3 L and 5 L, respectively. Therefore, pain/discomfort was more often reported at other 3 L and 5 L levels compared to the other EQ-5D dimensions (Table 2).

The range of index scores was broader in the 3 L than in the 5 L version, especially for negative values (Fig. 1). The lowest index score reported for the 3 L was − 0.349 (state ‘23333’), whereas this was − 0.263 (state ‘45554’) for the 5 L. The most frequently reported health state was ‘11121’ (slight problems in pain/discomfort and no problems in the other dimensions), i.e. 17% in the 3 L and 13% in the 5 L. There were 46 and 90, 3 L and 5 L health states reported in the study, respectively.

Of the participants who reported no problem (level 1) for a dimension on the 3 L, most (73–94%) reported the same on the 5 L, while 6–26% switched to slight problems (level 2) on the 5 L as shown in Table 3. The majority of the participants who reported moderate problems (level 2) on the 3 L indicated slight problems (level 2) on the 5 L (44–67%), while 20–28% switched to moderate problems (level 3) and 12–31% shifted to severe problems (level 4) on the 5 L. Most of the participants who indicated confined to bed/unable/extreme problems (level 3) on the 3 L indicated extreme problems (level 5) on the 5 L for the usual activities dimension, whereas most participants who reported extreme problems on 3 L redistributed into severe problems (level 4) for pain/discomfort and anxiety/depression. As for the self-care dimension, these percentages were equal. Redistribution occurred least frequently in the mobility dimension since no participant reported ‘confined to bed’ on the 3 L in that area. The inconsistent responses were ranging from 4% on self-care to 7.6% on the pain/discomfort and anxiety/depression dimensions. An example of such inconsistency was a participant choosing ‘no problems walking’ in 3 L (mobility level 1) and ‘severe problems walking’ in 5 L (mobility level 4).

Compared to the 3 L version, the 5 L system had a substantial gain in classification efficiency for each dimension, indicated by higher H′ values of all the dimensions. The J’ values were more similar among the two versions of EQ-5D as shown in Table 4, indicating that the degree of the potential use of the classification system was comparable between the two versions.

Fifty-three participants (26.8%) completed the instruments twice. By inclusion criterion, all reported no major changes in their health between the first and second data completion point. The weighted kappa of the 5 L dimensions for the 3 L was judged as slightly in agreement for the self-care dimension at 0.14, while the other four dimensions fair agreement existed: mobility at 0.25, usual activities at 0.23, pain/discomfort at 0.25 and anxiety/depression at 0.40. For the 5 L, the pain/discomfort dimension was judged as slightly in agreement at 0.19, while the other four dimensions were in fair agreement: mobility at 0.35, self-care at 0.30, usual activities at 0.37 and anxiety/depression at 0.39. The EQ-VAS ICCs were 0.35 and 0.32 for the 3 L and 5 L respectively. Moreover, the ICCs of the 3 L and 5 L index scores were 0.64 and 0.74 respectively, reflecting a moderate level of reproducibility (Table 5).