Thoughts about health and patient-reported outcomes among people with diabetes mellitus: results from the DiaDec-study

The present cross-sectional study included data from participants who took part in the ‘Quality of life, disability, health care utilisation and costs in patients with diabetes: The role of depression’ study (DiaDec-study) [25]. The DiaDec-study sought to estimate the prevalence of diagnosed and undiagnosed depression in people with DM, to compare costs and health care utilisation, and to analyse HRQoL in people with DM with and without co-morbid depression. The participants were a random sample of the ‘pronova Betriebskrankenkasse’ (pronova BKK), a German statutory health insurance, aged between 18 years and 80 years and with a diabetes diagnosis (response 51%). Diabetes diagnosis was based on the ‘International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10)’, antihyperglycaemic medication and HbA_1c values, as described as diagnostic criteria in other published studies [26, 27]. Inclusion criteria were described in detail in the study protocol [25]. Data assessment comprises questionnaire data and individually linked longitudinal statutory health insurance data, 1 year before and 1 year after the survey.

The present study applied a similar approach to Dolan (as described below) to analyse data elicited by questions on thoughts about health. The questions were incorporated into the questionnaire at a later stage of the study, which was sent out once to the participants. Thus, of the 1860 participants who sent back their questionnaire 783 participants sent back a questionnaire including questions on ‘thoughts about health’. Fifty-seven participants were excluded due to missing data in the ‘thoughts about health’ instrument, resulting in a study population of 726 participants.

The questions on thoughts about health (Additional file 1, Appendix 1) are based on the DRM and were extracted from the study by Dolan [21] (2011) and translated into German.

The first question enquires about how often participants thought about their health the day before (reference to a single day). Participants can answer by specifying the frequency (‘not at all’, ‘a few times’, ‘many times’ and ‘continually’). The second question refers to the intensity of certain feelings that can occur during the presence of thoughts: ‘If you thought about your health yesterday, how did you feel about those thoughts?’. On a scale of 0 (does not apply at all) to 6 (applies exactly), answers can be given to a total of three negative feelings (angry, depressed and worried) and one positive feeling (happy). Three frequency groups were defined. The first group included all participants who had not thought about their health the day before. Respondents who thought about health ‘a few times’ were placed in the low frequency group, and those who thought about health ‘many times’ or ‘continually’ were placed in the high frequency group. As in the Dolan study [21] the ‘U-index method’ reported by Kahneman et al. (2004) [23] was used to determine the valence of thoughts. If ‘happy’ was the highest or joint-highest rated feeling, thoughts were labelled as ‘positive’; otherwise thoughts were labelled as ‘negative’.

Five groups were defined by combining these two group allocations: ‘no thoughts about health’, ‘low frequency and positive thoughts’, ‘high frequency and positive thoughts’, ‘low frequency and negative thoughts’ and ‘high frequency and negative thoughts’.

The 12-Item Short-Form Survey (SF-12) measures HRQoL and is a short version of the SF-36 [28]. The instrument was tested in people with DM and is available and validated in German [29, 30]. The SF-12 contains 12 questions representing eight health dimensions, e.g. physical function and emotional role function. The dimensions reflect a physical and mental component summary score of the SF-12.

The Problem Areas in Diabetes (PAID) scale identifies diabetes-related distress in people with DM [31, 32]. The instrument was validated in German [33] and is used to measure diabetes-specific HRQoL and also as a screening instrument for depressive symptoms among people with DM [31]. This instrument consists of 20 items relating to diabetes-specific stress situations. Each item can be answered on a 5-Likert scale from ‘no problem’ [0] to ‘major problem’ [4]. Diabetes-related distress was coded as ‘high diabetes-related distress’ (cut-off score ≥ 40) as recommended [31].

Depressive symptoms were assessed using the Depression Module of the Patient Health Questionnaire (PHQ-9) [34]. The PHQ-9 has been validated in studies with people with DM and in German [34‐37]. Participants answer how often they have felt affected by nine different conditions over the previous 2 weeks. The following four answers can be given: ‘not at all’ [0], ‘several days’ [1], ‘more than half the days’ [2] and ‘nearly every day’ [3] [34]. We used the categorical diagnostic algorithm to obtain an indication of clinically relevant depression [38]. This adds a clinical perspective to the results of the other patient-reported outcomes. ‘Depressive symptoms’ were coded if at least two questions were answered with ‘more than half the days’ or ‘nearly every day’ and if item one (‘little interest or pleasure in doing things’) or item two (‘feeling down, depressed or hopeless’) were included. Item nine (‘thoughts that you would be better off dead or of hurting yourself in some way’) was also counted if ‘several days’ were reported [34, 37, 39].

The following variables, which previous studies have found to be associated with repetitive thoughts (e.g. worries and rumination) and specific thoughts at the time of the diabetes diagnosis (e.g. ‘nothing I could do about complications’), were included in our study: age, sex, education, type of diabetes, diabetes duration and comorbidity [6, 13, 40‐42]. Most of these variables were collected in the DiaDec-study survey in 2013. Education was coded as ‘no graduation’, ‘other graduation’ and ‘university degree’, type of diabetes was coded as ‘type 1’, ‘type 2’, ‘other’ and ‘not known to participants’, and diabetes duration was coded as ‘less than 10 years’ and ‘10 or more years’. Comorbidity was measured using the health insurance data in 2012 based on the results of the 80 hierarchical morbidity groups according to the German morbidity-oriented risk structure compensation scheme (total number of comorbidities) [43].

Further variables which might be associated with thoughts about health were also included, i.e. characteristics related to socioeconomic status such as employment and marital status. In addition to comorbidities, people’s own perception of their state of health is also relevant. This variable was therefore added. Furthermore, the mode of diabetes treatment was included, as it may indicate the severity of the disease. Various requirements, for instance remembering to take medication, can also lead to treatment being perceived more intensely in everyday life.

The 2013 survey coded employment as ‘yes’ or ‘no’. Marital status was coded as ‘single’, which also included people who are widowed, divorced or live permanently separated, and by ‘married and living together’. Migration background was coded as ‘yes’ if the participant or at least one parent did not possess German nationality by birth, otherwise it was coded as ‘no’ [44]. The ‘In general, would you say your health is … ’ item of the SF-12 questionnaire, coded as ‘excellent’, ‘very good’, ‘good’, ‘fair’ and ‘poor’, was selected to assess perceived health status. Mode of diabetes treatment was identified using available drug information in the 2012 health insurance data and coded in relation to oral glucose-lowering drugs and insulin as ‘yes’ or ‘no’.

Descriptive summaries for patient characteristics were obtained (depending on the distribution of the variables by frequencies, percentages, means (M) ± standard deviations (SD)). We compared participant characteristics between the ‘thought groups’.

Multivariate linear and logistic regression models were fitted to assess associations between thought groups and outcomes. These models estimate regression coefficients (betas) and odds ratios (OR) with 95% confidence intervals (CI). Separate models were fitted for the outcomes, using the physical and mental summary score of the SF-12 as metric dependent variables and diabetes-related distress (low vs. high) and depressive symptoms (yes vs. no) as dependent binary variables. In addition, we checked for potential artefacts from dichotomising the variables ‘diabetes-related distress’ and ‘depressive symptoms’ and performed sensitivity analyses with the respective score. Two final models were considered for each outcome, one only including the four thought groups as independent variables vs. no thoughts, the other including all preselected sociodemographic and health-related variables: age, sex, education (other graduation or university degree vs. no graduation), employment (yes vs. no) and marital status (married and living together vs. others) and the following health-related variables: diabetes duration (≥10 years), mode of diabetes treatment (oral glucose-lowering drugs (yes vs. no), insulin (yes vs. no)) and comorbidity (> 3 comorbidities). Perceived health status (fair to poor vs. good to excellent) was added for the PAID and PHQ-9 outcomes. Type of diabetes was excluded because of a high number of unknown type or missings.

Data analysis was performed using SAS software, Version 9.4 (SAS Institute Inc., Cary, NC). The significance level was 5% and two-sided if not stated otherwise.

The present study population (n = 726) included more male participants (Table 1). The majority of participants (94.7%) had no university degree. Three of four were currently not working and were married. One in six people had a migration background. Most participants (85.6%) stated that they had type 2 diabetes. Half of the participants were diagnosed ten or more years ago. Two thirds were treated with oral glucose-lowering drugs and about one third with insulin. The number of comorbidities varied from zero to 17 and most participants considered their general state of health to be good.

Group classification was carried out for 726 participants. This included those 334 participants (46.0%) who had not thought about their health the day before. A total 124 participants (17.1%) were assigned to the ‘low frequency and negative thoughts’ group and 155 participants (21.4%) to the ‘low frequency and positive thoughts’ group. The ‘high frequency and negative thoughts’ group included 88 participants (12.1%) and the ‘high frequency and positive thoughts’ group included 25 participants (3.4%). The characteristics of the participants within the thought groups are shown in Table 2.

Table 1 shows the mean and standard deviation of both summary scores of the SF-12 in the study population. The physical summary score and the mental summary score (Fig. 1) of the SF-12 differed between the thought groups. Participants in the ‘no thoughts’ group showed the highest mean scores in both scales compared to the other four groups. In contrast, the ‘high frequency and negative thoughts’ and ‘high frequency and positive thoughts’ groups showed the lowest mean scores in both scales. High diabetes-related distress was identified in 13.6% and depressive symptoms were present in 11.8% of the study population. As illustrated by Fig. 2, depressive symptoms and high diabetes-related distress were most common in people with high frequency and negative thoughts. Depressive symptoms and high diabetes-related distress were more common in the ‘low frequency and negative thoughts’ and ‘high frequency and positive thoughts’ groups than in the ‘low frequency and positive thoughts’ and ‘no thoughts’ groups.

Univariate linear regression models indicate that participants who reported to have thoughts about health were significantly more likely to have a lower physical and mental component summary score of the SF-12 compared to the reference group of participants with no thoughts (Table 3). Adjustment for sociodemographic and health-related variables did not alter ORs substantially. The univariate logistic regression model suggests that negative thoughts as well as high frequency and positive thoughts were significantly associated with high diabetes-related distress (Table 4). Multiple adjusted negative thoughts with high or low frequency remained significantly associated with diabetes-related distress. Participants who reported highly frequent negative or positive thoughts as well as low frequency and negative thoughts were significantly more likely to present depressive symptoms. Similar results were found after adjustment. In the two sensitivity analyses, we saw that linear regression models for diabetes-related distress and depressive symptoms lead to associations comparable to those presented in Table 4. Within the sensitivity analyses we saw a trend to stronger associations. This indicates, that our results are stable.

The DRM-based questions on thoughts about health proposed by Dolan (2011) were used in the sample of people with DM to measure the frequency of thoughts and to capture the related affective experiences of respondents. Presence of thoughts about health was identified in a number of people with DM in everyday life and was associated with the physical and mental component summary score of the SF-12, diabetes-related distress and depressive symptoms. Surprisingly, people with positive thoughts and negative thoughts alike have a lower HRQoL than people who had not thought about their health the day before. In addition, people with high frequency and positive thoughts have a higher chance of experiencing depressive symptoms than people who had not thought about their health the day before.

Contrary to Dolan’s (2011) survey of the general population in the USA [21], our study found more people who had not thought about their health the day before (26% vs. 46%). Suppression of unacceptable and/or unwanted repetitive thoughts and impulses defined as intrusive thoughts [45] is reported as a coping strategy for numerous diseases and may be one reason for these differences [46, 47]. On the other hand, a lower percentage of people in the ‘low frequency and positive thoughts’ group (21% vs. 40%) was found. The proportions of people who reported ‘low frequency and negative thoughts’, ‘high frequency and negative thoughts’ and the ‘high frequency and positive thoughts’ were almost identical in both studies.

Contrary to our expectations, the number of people who had thought about their health the day before and had predominantly negative thoughts was similar in both studies, although the target groups were different (general population vs. people with a chronic disease). It is assumed that people with diabetes are more vulnerable to the occurrence of negative thoughts. For example, negative thinking increases if individual consider themselves to be responsible for the disease [42]. Schabert et al. (2013) described how people with type 2 diabetes are confronted with the stigmatisation that they are responsible for their own situation, for example for being overweight or leading an unhealthy lifestyle [48]. Further studies reported concerns about hypoglycaemia and complications in people with DM [49, 50].

We found an association between positive thoughts and a lower HRQoL, which is not consistent with the results of Dolan (2011) where more frequent and positive thoughts were associated with higher valuations of state of health and only more frequent and negative thoughts were associated with lower health state valuations [21, 51]. A possible explanation for our finding may be a negative effect of unrealistically positive thoughts implying excessive expectations, as suggested by MacLeod and Moore [52]. However, comparison is difficult due to different outcome measures (preference-based valuation of HRQoL using Time-Trade-Off method vs. HRQoL measured by the SF-12), the heterogeneity of the study designs (e.g. online and telephone interviews vs. questionnaire) and the different target groups in the studies (general population vs. people with a chronic disease).

In line with other research [2, 7] we found an association between negative thoughts and higher diabetes-related distress and depressive symptoms. One reason for the unexpected associations identified in our study between frequent and positive thoughts and depressive symptoms could be an increased confrontation of diabetes and one’s own health in the case of an existing diagnosis. For instance, there is no increased risk of developing depression in people with prediabetes or undiagnosed diabetes [53]. Nevertheless, the confidence intervals are wide-ranging, which has to be taken into account when interpreting this association.

This explorative study applied the DRM-based questions on thoughts about health for the first time in the population with DM in order to obtain data on the frequency of thoughts about health and to distinguish between positive and negative thoughts based on an assessment of intensity of feelings. The DRM-based approach focuses on respondents’ actual experiences and represents an alternative to retrospective self-report questionnaires asking respondents about the typical content and/or style of their thinking. It has the potential to overcome the limitations of self-report questionnaires with regard to format (lengths of items and scales) and content validity (constructed experience biased by beliefs, subjective interpretations etc.). However, more research is needed with regard to the validity of the proxy questions collecting DRM-type data.

It was not possible to identify a positive association between thoughts about health and health outcomes. Given the limitations of cross-sectional data, longitudinal studies including various health outcomes (e.g. health promoting behaviours) are needed to better understand long-term consequences of positive and negative thoughts about health.

Given that the present study finds thoughts about health to be associated with worse patient-reported outcomes, a discussion of training in how to deal with thoughts in diabetes education programmes would seem pertinent. Two randomised controlled trials showed that people with type 2 diabetes who were additionally trained to experience their thoughts and feelings instead of changing or stopping them used more coping strategies, reported better diabetes self-management, and had better HbA_1c values [5, 54]. Further research must be carried out to discuss the implications of such interventions which may themselves already lead to an increased frequency of thoughts.

The main limitation of our study is the unproven validity and reliability of the questions used to assess the frequency and valence of thoughts about health. However, the study was designed as an explorative study to test the DRM-based approach to assessing thoughts about health in the population with DM - as proposed by Dolan [21]. The questions therefore still allowed us to collect data on actual experiences (thoughts about health and related feelings) with a very short recall period and without subjective interpretations of response categories by respondents. The ‘U-index method’ [21, 23] which was used to determine the valence of thoughts also helped to overcome respondents’ tendency to avoid extremes such as the answer category ‘very satisfied’. The U-index seems to be a useful method of identifying negative thoughts. However, it has to be mentioned that 16 out of 25 people with high frequency and positive thoughts in our study rated negative and positive thoughts equally. This might have led to a misclassification bias in our results.

Where values were missing in individual items of the questions (n = 12), assignment was performed as described in the online resource (Additional file 1, Appendix 1). This may lead to an inaccuracy in a small number of values. Exclusion of participants because of missing values may have caused biases in model fitting and estimation. Furthermore, the multiple regression models were fitted and analysed using the same data, such that associations might be overestimated.

A further strength is the possibility of combining primary and secondary data, providing both patient-reported outcomes and valid clinical data.