Impact of hypoglycemic episodes on health-related quality of life of type-2 diabetes mellitus patients: development and validation of a specific QoLHYPO^© questionnaire

First, a review of the literature was conducted to identify the domains and descriptors of HRQoL in T2DM patients who experienced hypoglycemia. Subsequently, a focus group with four clinical experts evaluated the domains and descriptors found in the review and provided any additional information that may not had been identified in the literature [20].

Based on the conclusions of the focus group, a script was designed for conducting semi-structured interviews with patients. The aim of these interviews was to find out the patients’ perceptions regarding the HRQoL domains and to draw up an initial list of the items for inclusion in the preliminary version of the questionnaire [21]. Patients 30 years of age or older who had been diagnosed with T2DM for at least 5 years, and who had had an HbA1c test in the previous 3 months and a hypoglycemic episode in the 6 months prior to the study were interviewed.

Finally, the preliminary version of the questionnaire was reviewed and validated by the same experts who participated in the first focus group. The first version of the questionnaire resulted from this discussion.

In order to determine the ease with which patients answered the first version of QoLHYPO^©, face validity and feasibility were assessed. A pilot test with T2DM patients was performed to evaluate the clarity of the wording of the items, their importance and the perceived burden (difficulty and time spent on the questionnaire), using a 5-point Likert scale (0 = not at all satisfactory to 5 = very satisfactory). The face validity and feasibility test population included adult patients ≥30 years old, diagnosed with T2DM for at least 5 years, with an HbA1c test in the previous 3 months and who had a hypoglycemic episode in the 6 months prior to the study.

Following this assessment, items that had scored equal to or greater than 4 by at least 75% of patients were selected for inclusion in the second version of QoLHYPO^© questionnaire.

Statistical analysis was performed using SPSS Statistics v.20 software.

To reduce the number of items, floor and ceiling effects and Rasch analysis were applied. Floor and ceiling effects were determined if at least 35% of patients had used the lowest (floor) or highest (ceiling) response categories. Items that presented with floor and/or ceiling effects were removed from the questionnaire. It does not exist a cut-off value to define floor and ceiling effects, it usually is an arbitrary value which depends on the study population [22]. Since the study sample is composed by a high proportion of patients (58%) with mild hypoglycaemia episodes, in order to be more sensitive to the severity of hypoglycaemias a cut-off value of 35% was used. Rasch analysis is a mathematical modeling technique that is widely used in the development of questionnaires [23‐27]. The main characteristic of the model is that item and person parameters can be estimated independently of the characteristics of the sample and the difficulties of the items, respectively [28]. Rasch analysis estimates the thresholds or degrees of difficulty of the response categories for each item. These thresholds are displayed graphically by means of characteristic curves, which must appear in the same ascending order as the category responses (the higher the score, the better the HRQoL). These graphics allow identifying redundant response categories where the curves of different categories overlap [29]. The presence of redundant categories indicates an excess of response categories, so they were recoded and merged. Items that still displayed overlapping or unordered curves after recoding were removed from the questionnaire.

After assessment of the characteristic curves, the fit of the items to the Rasch model was analyzed through the infit and outfit statistics calculation. Both statistics are derived from the squared standardized residuals, with a goodness of fit range between 0.5 and 1.5 in polytomous responses [30, 31]. Items whose infit and outfit values exceeded the established range were eliminated from the questionnaire so that the final version of the QoLHYPO^© was constituted by the remaining ones.

A total of 150 T2DM patients comprise the sample. Sample size estimation was based on the minimum number required for Rasch parameter estimation with a stability of 0.5 logits and 99% confidence [32]. An additional 7% of patients were recruited for possible losses. Patients had to meet the following criteria: be at least 30 years old, have been diagnosed with T2DM for a minimum of 5 years, have had at least one HbA1c test in the 3 months prior to the study, have presented at least one hypoglycemic episode in the 6 months prior to the study, and have given their consent to participate in the study.

The R package eRm was used to perform Rasch analysis [33, 34].

To analyze the construct validity of the QoLHYPO^© questionnaire, the responses to the questionnaire obtained from both cohorts at the baseline visit were used. For the assessment of the number of domains that constitute the QoLHYPO^© questionnaire, several exploratory factor analysis techniques were used (optimal coordinate, acceleration factor, Velicer’s minimum average partial test - original and modified, Horn’s parallel analysis) [36]. The basis of all these techniques is that every factor (or domain) is defined by an eigenvalue, so that the greater the eigenvalue, the greater the variance explained by the factor. These analyses were performed with the R package paramap statistical program [33, 37]. In addition, to evaluate whether the QoLHYPO^© questionnaire was measuring the same characteristics as specific questionnaire ADDQoL-19 and generic questionnaire EQ-5D-3 L, the multitrait-multimethod matrix was calculated using Spearman’s rho correlations between the scores of the various questionnaires.

The reliability of a questionnaire refers to the degree of consistency and to intra-observer reproducibility, i.e., the equivalence between repeated measurements in the same individual. It was evaluated using test-retest reliability and internal consistency based on the responses given to the QoLHYPO^© questionnaire by patients included in the reliability cohort during the baseline and follow-up visits [38‐40]. Internal consistency was measured using Cronbach’s alpha on the QoLHYPO^© total score. Test-retest reliability was assessed using the intra-class correlation coefficient (ICC) for the total score and Cohen’s Kappa coefficient for the responses to each item. Cronbach’s alpha values above 0.70 were regarded as satisfactory internal consistency [41]. ICC and Kappa values above 0.75 and 0.70, respectively, were regarded as indicators of good reliability [40].

SPSS v.20 software was used to compute Cronbach’s alpha and ICCs, and Cohen’s Kappa coefficient was computed using VassarStats software [42].

Responsiveness was measured based on the responses to the QoLHYPO^© questionnaire given by the patients, who constituted the responsiveness cohort. In the follow-up visit (at 3 months), patients had to report how their health status had changed with respect to the baseline visit. They responded to an anchor question to assess the evolution in their health status from the previous visit in relation with the T2DM, and therefore, with the hypoglycemic episodes. Answers were scored on a 5-point Likert scale: 1 = much worse, 2 = worse, 3 = no change, 4 = better and 5 = much better.

Responsiveness can be internal, referring to the ability of a measure to change over a predefined time period, or external, referring to the degree to which change in a measure relates to a change in a reference measure of clinical or health status [43]. It is important to notice that the objective of the responsiveness assessment was the ability of the questionnaire to detect changes, no the direction of the changes. Internal responsiveness was measured with standardized response mean (SRM) [44]:assuming that the group that had experienced improvement had responded to the anchor question with “better” or “much better” while those who had remained stable had responded “no change”. Values less than 0.2, between 0.2 and 0.5, between 0.5 and 0.8, and greater than 0.8 indicated slight, low, moderate and high sensitivity, respectively [43, 45]. External responsiveness was based on Receiver Operating Characteristics curves (ROC) [43, 46]. To assess external responsiveness, patients were coded as experiencing improvement (responded with “better” or “much better”) or no improvement (responded with “much worse”, “worse” or “no change”).

Minimal clinically important difference (MCID), defined as the smallest change in score in the questionnaire that could be considered significant, was based on the above-mentioned anchor questions, which allowed selecting patients who had experienced an improvement (“better”) [47]. Similarly, the professionals who recruited the patients also responded to an equivalent anchor question to assess their patients’ clinical course. The difference in score in the anchor question of the QoLHYPO^© questionnaire between the baseline visit and follow-up visit of patients whose physician and they themselves reported an improvement was used to determine MCID.

After reviewing the literature and holding the two focus groups with 4 T2DM management experts and 10 interviews with T2DM patients, the first version of the QoLHYPO^© questionnaire was developed. Table 1 shows the items included in the first version of the QoLHYPO^© questionnaire.

A total of 18 T2DM patients participated in the face validity and feasibility of the questionnaire. Related to questionnaire format, approximately 81 and 94% of participants scored 4 and 5 (from 1 = not at all satisfactory, to 5 = very satisfactory), respectively, indicating that they were satisfied with it. The mean time required to complete the questionnaire was 13 min (SD: 5 min., range 7-25 min).

After questionnaire face validity and feasibility, the second version of the questionnaire was developed. It included four general items about the impact of hypoglycemia on patients’ HRQoL (frequency and severity of hypoglycemia, knowledge to control hypoglycemia and frequency of use of glucometer) and two sets of questions with 16 and 21 items, respectively. First set of questions was related to the patients’ feelings when their blood sugar drops. The second one, explores patients’ perceptions in relation to their drops in blood sugar. Each set of questions had five categories of response (0 = Never, 1 = Rarely, 2 = Sometimes, 3 = Often, 4 = Always).

A total of 160 patients with T2DM from 12 Spanish autonomous regions were included in the pilot test. Of these patients, 20 did not meet the inclusion criteria so the final sample consisted of 140 T2DM patients. Table 2 shows the main characteristics of the study population.

Table 3 shows the answers to the 4 general items about the hypoglycemia in each phase of the study. As in Phase 1 a high proportion of patients had mild to very mild hypoglycemia (57.67%), a high cut-off value (35%) was used to assess floor and ceiling effects in order to make the questionnaire more sensitive to the severity of the hypoglycemia. After floor and ceiling effects testing (Table 4), 11 items were discarded. The first Rasch analysis was therefore conducted on the remaining 26 items. To estimate the threshold of HRQoL of the response categories for each item, a graphical representation of characteristic curves for each item was used. It allowed assessing the scale scoring system through a graphic display of the probability of patients answering each response category for each item, according to the individual’s ability (ability = HRQoL). Each item provided five response options, where 0 was the least favorable case (worse HRQoL) and four was the most favorable (better HRQoL).

For most of the items, the probabilities of response overlapped, suggesting that the patients were not using the full range of options allowed by the scoring system, which implied that the categories were redundant (see Additional file 1). These results showed the need to recode the responses by merging the categories: “Never” with “Rarely” and “Often” with “Always”, so that only three response categories remained. Figure 2 shows the characteristic curves for items 5.14 (not overlapping) and 6.5. (overlapping).

After recoding the response categories, a second Rasch analysis was conducted. New characteristic curves continued showing overlapping in 11 items (see Additional file 2). They were therefore removed from the questionnaire. Rasch analysis was again performed on the remaining set of 15 items (26-11 = 15) with three ordered response categories to assess proper fit of the items to the Rasch model. Two items had infit and outfit values that exceeded the 0.5-1.5 range and were therefore discarded. Finally, another Rasch analysis was conducted on the resulting set of 13 items, which determined the goodness of fit of the items for measuring the construct of interest (Table 5). These items constituted the final version of the QoLHYPO^© questionnaire (see Additional file 3).

The QoLHYPO^© questionnaire responses reported by the participants included in Phase 2 (n = 227) were used for the construct validity analysis.

All the techniques used for the exploratory factor analysis showed that the QoLHYPO^© questionnaire consisted of a single factor and was therefore unidimensional. The multitrait-multimethod matrix indicated that there was a significant correlation (rho = 0.557; p < 0.001) between the QoLHYPO^© and the DM-specific questionnaire ADDQoL-19. A significant correlation was likewise observed between the dimensions of the generic HRQoL questionnaire, EQ-5D-3 L, and the scores obtained in the QoLHYPO^© questionnaire [mobility rho = − 0.216 (p = 0.001), self-care rho = − 0.259 (p < 0.01), usual activities rho = − 0.032 (p < 0.01), pain/discomfort rho = 0.265 (p < 0.01), anxiety/depression rho = − 0.346 (p < 0.01), VAS rho = − 0.446 (p < 0.01)].

The internal consistency of the QoLHYPO^©, which was measured using Cronbach’s alpha, was high at both visits (visit 1 = 0.912; visit 2 = 0.901). The values obtained as a result of test-retest showed that the QoLHYPO^© has good reliability [ICC = 0.920 (CI95%: 0.890-0.942); and kappa> 0.60 for all items].

Internal responsiveness was determined based on patients’ self-reported health improvement. The results showed a standardized effect size of 0.676, indicating moderate internal responsiveness (between 0.5 and 0.8). External responsiveness assessment using ROC curve analysis did not yield significant results.

Analysis of MCID showed that the QoLHYPO^© questionnaire was able to detect improvement in health when there was a 3.2 point increase between visits.