HFS-II scores in this study and for this patient group reflected the expected patterns in fear of hypoglycaemia. Patients who reported hypoglycaemia showed more fear than those who did not, and patients who reported more severe episodes of hypoglycaemia showed more fear that those who reported less severe episodes. Based on the MID results of this study, the difference in the Worry Scale of the HFS-II between not having and having reported hypoglycaemia without the need for assistance appears to be clinically meaningful to patients. This is particularly relevant because the vast majority of episodes of hypoglycaemia reported in this study were those which patients managed themselves without the need for assistance from others.
To determine MID, two classes of methods have been discussed in the literature, distribution-based methods and anchor-based methods [34
]. Distribution-based methods are based on mathematical calculations that involve standard deviation, score range or Cronbach's alpha. Anchor-based methods are based on judgment of treatment success [35
]. Among the group of anchor-based methods, either physician- or patient-based anchors can be applied. While distribution-based methods that involve calculations with the standard deviation, also depend on the heterogeneity of the study population [23
], anchor-based methods critically depend on the validity of patients' rating [23
] and on choosing response categories that reflect the importance of a change. No gold standard for determining MID currently exists. Therefore we followed the suggestion of Guyatt et al. 2002 that is more accurately to report a range of MID retrieved by different approaches than to recommend a single MID.
Wells et al., had patients compare their own health status with that of their peers. MID was calculated from the difference in score between patients who felt 'somewhat better than other patients' and patients who felt 'about the same' [32
]. Another approach is to have patients rate their own improvement due to treatment. Depending on the disease, MID is then calculated as the difference in mean scores between patients who report a 'small improvement' and those who felt 'a little worse', or between patients who report a 'small improvement' and those who reported 'no change' [22
]. Walters and Brazier surveyed change in condition over time to establish a MID for quality of life measures, calculated as the changes in score means between patients reporting 'an improvement' and those reporting 'no change' in condition [31
]. In our study, one of the anchors chosen to determine MID was based on treatment satisfaction ('less than satisfied' vs. 'satisfied') which we considered closely related to the construct of self-reported treatment success ('no change' vs. 'small improvement') frequently recommended for studies that examine treatment effects [27
The results for the MID using TSQM question 14 on treatment satisfaction (3.6) and TSQM questions 6 and 8 on side effects (3.6 and 3.9) were relatively consistent. Anchor-based MID estimates of the HFS-II were well within the range obtained from distribution-based methods. However, compared to the MID based on treatment satisfaction (n = 154), the MID's based on side effects were derived from the smaller number of patients (n = 29) who reported side effects. Also, hypoglycaemia does not appear to have been the only source for variation in treatment satisfaction. Hypoglycaemia may not have been the only side effect experienced by patients. For these reasons, these results should be considered exploratory. The unexpected low HFS scores for patients who were 'very dissatisfied' (5 patients, HFS-II 12.8) and 'extremely dissatisfied' (1 patient, HFS-II 2.0) compared to patients who were 'dissatisfied' (4 patients, HFS-II 16.5) or 'somewhat satisfied' (44 patients, HFS-II 13.5) may be due in part to the low number of patients in the categories. Another explanation is that perhaps these patients did not adhere to their prescribed medication regimen due to their dissatisfaction and, therefore, had less risk of hypoglycaemia.
While the response categories 'extremely dissatisfied', 'very dissatisfied', 'dissatisfied', and 'somewhat satisfied' were aggregated to 'less than satisfied', we did not aggregate the other response categories of the TSQM question on treatment satisfaction. In our opinion, aggregating 'satisfied', 'very satisfied', and 'extremely satisfied' and comparing these patients to patients 'less than satisfied' would have no longer constituted a minimum
clinically important difference [27
It is important to note, however, that our MID estimates may not apply to other types of patients with diabetes [37
]. Patients in this study had type 2 diabetes managed by combined oral anti-hyperglycemic medications, and the majority of them (71%) reported no hypoglycaemic episodes in the previous 6 months. Less than 4% reported severe or very severe episodes. Different MID estimates would likely be generated in the subgroup of patients with type 1 diabetes who experience frequent, recurrent episodes of severe hypoglycaemia. Nonetheless, the results of this study suggest that fear of hypoglycaemia, as measured by the HFS-II, can be a useful outcome variable in diabetes health services research, and that even relatively small differences in scores can be clinically meaningful to patients with type 2 diabetes mellitus using oral anti-hyperglycemic medications.
Observational studies can provide valuable information on effectiveness due to real-world settings and larger study populations [38
]. However, self-reported outcomes from a large number of sites also introduce bias and limitations. The participating physicians may not have always had complete knowledge about parallel prescriptions to their patients and patient's visits to other physicians or hospitals. Eventually, this might have led to incomplete data on patient's medical history, or inclusion of patients who would have otherwise been excluded. Episodes of hypoglycaemia were most likely underreported in our study population, since many patients with diabetes may not always recall or recognize symptoms of hypoglycaemia [6
], or may have limited knowledge about hypoglycaemia itself [40