Design and study population

In the present study, we analysed data from the baseline and second follow-up measurement of the population-based prospective ESTHER cohort study - a population-based cohort study of older adults conducted in Saarland, a state located in South West Germany. Between July 2000 and December 2002, in the federal state of Saarland, 9953 participants aged 50 to 74 years were recruited by their general practitioners (GPs) in the course of a health check-up that focused on early detection of chronic diseases. In Germany this health check-up is offered biennially to older adults. Baseline data shows that the ESTHER study sample is representative with respect to both demographic variables and chronic diseases of the general German population [14]. The study was approved by the Ethics Board of the University of Heidelberg and by the Ethics Board of the medical association of the State of Saarland. Written informed consent was obtained from all participants.

At baseline, participants completed a comprehensive self-administered questionnaire including the SF-12 to assess HRQOL; blood and urine samples were taken. The medical information was reported by the GPs. Between July 2005 and December 2007, 8270 elderly people aged 53 to 80 years participated in a second follow up ( = 90.0% response rate of those still alive). In the second follow-up, participants again completed a self-administered questionnaire that included the SF-12.

Measurements

In the predictor analysis of HRQOL in diabetes patients, the following baseline variables were included: HbA1c, treatment with insulin, number of complications, number of comorbid diseases, BMI, self-reported history of current or previous depression, HRQOL at baseline, gender, age, nationality, marital status, educational level, and smoking.

HRQOL at baseline and follow-up was measured with the SF-12- Questionnaire. The SF-12 is a widely used generic questionnaire that does not focus on specific disease groups. Several domains of HRQOL – such as physical, social and role functioning – are assessed. Items are weighted and totalled to provide both physical (PCS) and mental component scores (MCS) ranging from 0 to 100. A higher score in the respective summary scales indicates a higher quality of life. The questionnaire shows good psychometric criteria [5].

Diabetes was defined as follows: diabetes mellitus documented by the GP on the standardized health check-up form, or the use of anti-diabetic medication.

HbA1c is an established biochemical marker of diabetes control over the preceding two months. HbA1c analyses at baseline were done using high performance liquid chromatography in a central laboratory. Urinary albumin concentration was measured by immunonephelometry assay in a spot urine sample. The following diabetes-related complications were counted: cardiac complications (self-reported angina pectoris or physician-recorded coronary heart disease or self-reported myocardial infarction), self-reported history of stroke, self-reported circulatory disorder of the legs and nephropathy; the variable thus ranged from 0 (if no complications were recorded/reported) to 4 (if all four complications were recorded/reported). Nephropathy as a complication of diabetes was defined by albuminuria (urinary albumin≥20 mg/l). The following comorbid diseases (other than diabetes-related complications) were counted separately: physician-recorded hypertension, physician-recorded dyslipidemia, self- reported history of asthma, and self-reported history of cancer (composite variable ranging from 0 to 4). Self-reported depression was classified into “no” (no depression has ever been diagnosed) and “yes” (current or previous depression episode has been diagnosed). The participants' self-reported weight (in kilograms) and height (in metres) were used to calculate BMI (body mass index = weight in kilograms divided by height in metres squared). BMI and age were treated as continuous variables. Marital status was classified into “married” (referent), “divorced/widowed” and “unmarried”. Duration of school education was categorized into the categories ≤8, 9–10 (referent), 11–12, and >12 years. Cigarette smoking was classified into “never” (referent), “former” or “current” (variable ranging from 0 to 2).

Statistical analyses

In general, the mental and physical component summary scores (MCS and PCS) of the SF-12 are computed by using a linear weighting of the 12 items. However, if one or more SF-12 items are missing, a sophisticated algorithm has to be applied to replace missing values. In our study, 64.0% of the participants completed all of the SF-12 items both at baseline and after five years; 36.0% of the participants had one or more missing items in the SF-12 questionnaires. To deal with missing items, the robust modified regression estimation (MRE) methods were used [15], [16]. In 2004, Spiro and colleagues published a detailed technical report regarding the imputing of PCS and MCS for the Veterans SF-12 Health Survey (VR-12). The authors presented a sophisticated algorithm (including SAS macros) for estimating PCS and MCS in the context of missing values. However, Spiro et al. used the VR-12 form to measure HRQOL that is slightly different from the SF-12 version regarding the answer- format of four specific items. All 12 items of SF-12 and VR-12 are identically formulated.

However, there are four items - regarding “role physical” (RP) and “role emotional” (RE) - that differ in their response format between SF-12 and VR-12.

The two items regarding RP are:

• “During the past 4 weeks, have you had any of the following problems with your work or other regular daily activities as a result of your physical health?”
  1. “Accomplished less than you would like”
  2. “Were limited in the kind of work or other activities”.

The two items regarding RE are:

• “During the past 4 weeks, have you had any of the following problems with your work or other regular daily activities as a result of any emotional problems (such as feeling depressed or anxious)?”
  1. “Accomplished less than you would like”
  2. “Didn't do work or other activities as carefully as usual”

For each of these four items the SF-12-response format is dichotomized: yes/no. In contrast the VR-12-response format comprises a 5-point Likert scale: none of the time, a little of the time, some of the time, most of the time, all of the time.

The SF-12 does not tolerate any missing item whereas incomplete responses in VR-12 can be dealt with with the MRE. In order to apply the MRE-method, we converted the responses of the four SF-12-items diverging in response format to the VR-12 response format. The SF12-response “yes” was converted to the VR-12-response format “most of the time” while the SF-12-response “no” was converted to the VR-12 response “a little of the time”. As a validation check for the transformation algorithm the correlation between SF-12 scale scores and VR-12 scale scores was computed and showed a high correlation (r>0.99). After transforming the SF-12 scoring format into the VR-12 format, it was possible to use the already published imputation methods for the missing values. Depending on the missing-pattern at hand, the algorithm selects the corresponding imputation model from the database. In order to allow the imputing process to take place with only a minimum level of accuracy, the minimum R-square value required was kept at the default-value of 0.6.

Regression analyses were conducted separately for the dependent variables MCS and PCS. For the linear regression analysis to predict MCS and PCS respectively after five years a four-step approach was applied. Firstly, a reference model was defined comprised of a set of control variables: gender, age, nationality, marital status, educational level, and smoking. Secondly, from the pool of the ESTHER data, the following variables of interest for a prognostic model were selected (guided by theory and literature [9], [17]): HbA1c, count of diabetes-related complications (self-reported angina pectoris, physician-recorded coronary heart disease, self-reported myocardial infarction, self-reported history of stroke, self-reported circulatory disorder of the legs, nephropathy), count of comorbid diseases (physician-recorded hypertension, physician-recorded dyslipidemia, self- reported history of asthma, self-reported history of cancer), self-reported depression, BMI, and treatment with insulin. Thirdly, univariate analyses were conducted with the selected variables. The associations between PCS (or MCS) and the variables of interest were measured by calculating the Pearson Correlation coefficient. The t-test was used to calculate p-values. Variables yielding p-values<.20 in the correlation analysis were retained as candidate variables for the multiple regression analysis. We used the level of 0.20 as a screening criterion for selection because a more traditional level (such as 0.05) could fail to identify variables known to be important [18], [19]. Finally, candidate variables were entered into a multivariable regression model that already contained the control variables of the reference-model. All multivariate analyses were adjusted for baseline PCS and baseline MCS. Thus, in the final model, the effects of the prognostic variables of interest were adjusted for the effect of the reference-model variables. For all variables, standardized β-coefficients were calculated. All statistical analyses were performed using SAS (version 9.2) for Windows.

Predictors of mental health

MCS and PCS at baseline were both significantly associated with MCS at follow-up. This reflects the evidence gained from other studies that, in general, HRQOL measured at a specific point in time is a good predictor for HRQOL at a subsequent time measurement point [20]. In addition, we found that self-reported history of depression at baseline was significantly predictive for reduced mental health after five years. Most of the previous studies report cross-sectional findings with negative associations between depression and HRQOL [21]. Prospective data indicate that depression has a negative impact on therapeutic progression [22]. The prospective association between previous depression episodes and decreased MCS at five years later suggests that screening for depression and subsequent treatment may be important issues. Antidepressants are often recommended as the most appropriate treatment in elderly patients with depression. However, a recent meta-analysis showed that the treatment effect of antidepressants in the elderly is rather small [23]. Furthermore, in adults 50 years or older with chronic conditions such as diabetes antidepressants increase the risk of falling [24]. Therefore, beyond development of optimal medication, treatment strategies should take psychosocial aspects into account [25], [26].

To date, the discussion about association between glycemic control and HRQOL in diabetes patients has been controversial. In an RCT study of 275 diabetes patients, Weinberger et al. [12] reported no association between glycemic control and HRQOL at the one year follow-up. A recent cross-sectional study of more than 1000 diabetes patients found a clear association between higher HbA1c levels and decreased HRQOL [8]. In our study, no influence of higher HbA1c levels on HRQOL after five years was noticeable.

Predictors of perceived physical health

In our study, the presence of diabetes related complications is strongly predictive for decreased HRQOL in diabetes patients after five years. In previous cross-sectional research, this inverse association was consistently reported [9]. We found, in addition, that a higher BMI at baseline and former or current smoking is associated with a lower PCS at five years later - a result that is in line with previous study results. Higher BMI is known to have a negative contemporaneous association to HRQOL in diabetes patients. This relationship appears to be independent of disease severity [27]. In a large multicenter RCT, it was shown that even a weight-management program could improve HRQOL in diabetes patients [28]. In addition to HRQOL, additional complications, and BMI at baseline, higher age was significantly related to lower PCS at five years later. As the PCS sum score is strongly related to the patient's physical ability to perform activities of daily life, this result was to be expected. Interestingly however, age was not associated with MCS at five years later. These results correspond to the findings of previous studies regarding the natural progression in HRQOL which showed that, in older age groups, PCS tends to decline over time whereas MCS still may improve, dependent on gender and age group [29].

Depression and age-related effects

Additional subgroup analyses showed that in people with a reported history of depression advanced age was significantly associated with higher MCS scores at five years later. What could be the cause of this finding? It has been hypothesized that responsiveness to negative emotions decreases with age, and that older people may, in fact, achieve greater emotional control due to their learning of more effective coping strategies over their lifespan [30], [31]. The reason for this reciprocal interaction between age and depression regarding MCS could be that people have better coping strategies of emotional distress with increasing age.

Treatment with insulin

Research findings on the association between treatment regimen and HRQOL in people with diabetes are controversial. The United Kingdom Prospective Diabetes Study Group (UKPDS) showed no detectable differences in HRQOL between the different treatment regimens. In contrast, in a cross sectional study that included 2056 individuals from a national sample in the United States, patients on insulin reported lower QOL scores than did those on oral medication [32]. In our study, however, we found no prospective association between insulin treatment and HRQOL. One reason for the controversial findings could be that the relationship between insulin treatment and HRQOL in diabetes patients is heterogeneous across age. In elderly people, insulin treatment may be less stigmatized than in younger patients. The latest research findings show that the omission of insulin injections is negatively correlated with age [33]. It may be possible that the elderly get used to taking insulin and that initial negative appraisal is modifiable over time.

Strengths and limitations

The major strength of our investigation is a longitudinal design that includes an observation time of five years. In addition, we could maintain the very high response rate of the ESTHER follow-up by using a sophisticated published algorithm for replacing missing values in the SF-12 questionnaire. Furthermore, the analysis of the prospective relationship between baseline variables and HRQOL was controlled for both demographic and lifestyle variables.

However, the study also has several limitations. At baseline we did not assess some highly relevant diabetes-related complications such as retinopathy, autonomic neuropathy and diabetic foot - which could lead to the underestimation of the complication rate as a predicting variable. Nevertheless, the results of our analysis show a strong prospective association between complications and PCS, and respectively, MCS. Another limitation of the study is that we are not able to detect relevant differences that are due to the manifestation time of diabetes.

Clinical implications

A recently published work of Landman et al. [7] showed that MCS and PCS are inversely associated with mortality in diabetes after a follow-up time of almost ten years. The authors recommend the measurement of HRQOL in clinical practice. Results of our study indicate that interventions intended to improve HRQOL in elderly diabetes patients should address glycemic control, smoking cessation, and weight management, to prevent short term as well as long term complications. In addition, screening of depression appears to be advisable, and treatment of depression should be evaluated in regard to the improvement of MCS in diabetes-patients.