In the present analysis, data from the observational ESPRIT (Effect Study on Patient-Reported outcomes in Insulin glargine Treatment) study were used. The study was conducted between 2005 and 2008 in the Netherlands. In this study, diabetes specialists invited patients with type 2 diabetes who were in sub-optimal control (HbA1c > 7%) to participate. Inclusion criteria were: type 2 diabetes mellitus, an age of 18 years or older, current treatment with basal insulin and a clinical need to initiate insulin glargine. The decision to initiate treatment with insulin glargine as basal therapy was made at the discretion of the treating diabetes specialist. The study did not interfere with clinical practice and only included filling out a questionnaire booklet that took about 15 min at three consecutive periodic consultations. The questionnaire booklet was provided to the patient by the treating physician. Clinical data were retrieved from the medical chart. In view of the observational and non-invasive nature, the present study was not subject to the Dutch Medical Research Involving Human Subjects Act. Standardised study information and informed consent were provided to all patients and diabetes specialists.
In total, 510 type 2 diabetes patients from 116 out-patient clinics were included. Refusal rates were unknown. Sixty-three patients were found to be in good glycaemic control upon inclusion (HbA1c ≤ 7%) and were therefore excluded from analysis, resulting in a study population of N = 447 in sub-optimal glycaemic control. Fifty-two per cent of the patients (N = 233) received NPH insulin as basal therapy prior to initiating insulin glargine, 29% (N = 130) of the patients received premixed insulin and 19% (N = 84) received insulin detemir as basal therapy. Of the total population, 240 patients (54%) also received rapid-acting (meal-time related) insulin as co-therapy prior to therapy optimisation.
Demographic and clinical data were obtained at baseline, 3 and 6 months by means of self report and included age, gender, weight, height, time since diagnosis, previous medication use, hypoglycaemic episodes, the presence of diabetes-related complications, and co-morbidities. HbA1c and Fasting Blood Glucose (FBG), measured according to DCCT standards were retrieved from the medical chart. Adverse events were recorded by the treating physician, treated accordingly and reported to the researchers.
HRQoL is a multi-dimensional concept , in this study operationalised as a low level of diabetes symptom distress and fear of hypoglycaemia and a high level of emotional well-being.
Diabetes-related symptom distress was measured using the revised version of the Diabetes Symptom Checklist (DSC-r), that has been shown to have good psychometric properties [18, 19]. The DSC-r consists of 34 items grouped into eight symptom sub-scales: Hyperglycaemia, Hypoglycaemia, Cognitive distress, Fatigue, Cardiovascular distress, Neuropathic pain, Neuropathic sensibility and Ophthalmologic function. Each item asks about the presence of complaints (yes/no) and if present, to score the level of distress on a 5-point Likert type scale (i.e. ‘Have you experienced numbness in the hands?’ (yes/no), ‘How troublesome was this symptom for you?’ (Likert scale)). Scores are then transformed to a 0–100 score to obtain the DSC-r total distress score, with higher scores indicating more diabetes symptom distress. This transformation was also applied to the DSC-r sub-scales.
The Dutch version of the 13 item Worry sub-scale of the Hypoglycaemia Fear Survey (HFS-w)  was used to measure worries about hypoglycaemia. To facilitate interpretation of data, HFS-w scores were also transformed to a 0–100 scale, with higher scores indicating higher worry about hypoglycaemia.
The WHO-5 well-being index , a validated 5-item instrument, was used to assess general emotional well-being. The WHO-5 covers positive mood (good spirits, relaxation), vitality (being active and waking-up fresh and rested), and general interests (being interested in things). Item scores are summated to provide a total well-being score, transformed to a 0–100 scale, with lower scores indicating poorer well-being.
In the present study, the Cronbach’s alpha’s for the HFS, DSC-r total score and WHO-5 were 0.88, 0.94 and 0.90, respectively, confirming high internal consistency.
Generalised estimating equations (GEE) analysis was used for all longitudinal analyses. To test whether HbA1c, fasting blood glucose, hypoglycaemia, BMI, insulin dosage (insulin glargine and rapid-acting insulin) and scores on HFS-w, DSC-r and WHO-5 changed during the study period for the total population, time was modelled as an independent dummy variable and the outcome measure as dependent variable. For these analyses, effect sizes (Cohen’s d) for 6 month follow-up scores compared to baseline scores were calculated, to gain insight in clinical relevance.
To study whether improvement in HbA1c, by means of intensifying insulin therapy was related to improvement in HRQoL, GEE analyses were carried out with HbA1c as independent variable and HFS-w score, DSC-r score and WHO-5 score, respectively as dependent variable.
Variables with a right-skewed distribution (time since diagnosis, the number of diabetes complications, the number of co-morbidities, the number of symptomatic, nocturnal and severe hypoglycaemic episodes during the past 3 months, HFS-w total score and DSC-r total and sub-scores) were transformed with a natural logarithm for the purpose of statistical testing. In case of a right-skewed distribution, the median, 25th and 75th percentiles are presented.
We have studied the following variables as potential confounders: age, sex, time since diagnosis, BMI, the number of diabetes-related complications and the number of co-morbidities by adding them to the GEE model and looking if the coefficient of the main independent variable changed by more than ten per cent. Time-varying covariates (BMI and the number of symptomatic, nocturnal and severe hypoglycaemic episodes) were treated as such in the analysis. In case confounding, existed analyses were corrected for that specific variable. Moreover, we tested effect modification for sex, as a stronger association between glycaemic control and depression has previously been observed in women . In case of effect modification, analyses were stratified for sex.
Observational studies run a high risk of missing data, due to the naturalistic setting and low level of monitoring. The present study was no exception; at the start of the study, 19% of patients had missing data on the HFS-w, 23% on the DSC-r and 10% on the WHO-5. At 3 month follow-up, these percentages increased to 35, 37 and 27%, respectively. At 6 month follow-up they further increased to 42, 40 and 33%. Missing data were imputed using multiple imputations [23, 24]. The imputation model consisted of age, sex, time since diagnosis, the number of complications, the number of co-morbidities, HbA1c, FBG, the number of symptomatic, nocturnal and severe hypoglycaemic episodes, type of insulin used, and scores on the remaining two HRQoL instruments. Five datasets were generated using this technique. Analyses on these datasets were combined using Rubin’s rules for multiple imputations .