Association between unrealistic comparative optimism and self‐management in individuals with type 2 diabetes: Results from a cross‐sectional, population‐based study

Abstract Background and aims Unrealistic comparative optimism (UO), as the erroneous judgement of personal risks to be lower than the risks of others, could help explain differences in diabetes self‐management. The present study tested the hypothesis that individuals with type 2 diabetes who underestimate their comparative heart attack risk, have a lower adherence regarding recommended self‐management. Methods We used data from individuals with type 2 diabetes participating in the German KORA (Cooperative Health Research in the Region of Augsburg) GEFU 4 (self‐administered health questionnaire 2016) study. UO was estimated by comparing participants' subjective comparative risk for having a heart attack within the next 5‐years (ie, “higher than others,” “average,” “lower than others”), with their objective comparative 10‐year cardiovascular disease risk based on the Framingham equations. We estimated binary logistic and linear regression models to analyze which characteristics were associated with UO and to test the association between UO and participants' self‐management behaviors (ie, regular self‐monitoring of body weight, blood sugar, and blood pressure, regular foot care, keeping a diabetes diary, and having a diet plan), and their sum score, respectively. All models were adjusted for socio‐demographic and disease‐related variables. Results The studied sample included n = 633 individuals with type 2 diabetes (mean age 70.7 years, 45% women). Smokers and males were more likely to show UO than nonsmokers and females. Furthermore, a higher blood pressure and a higher body mass index were associated with a higher likelihood of UO regarding heart attack risk. However, UO was not significantly associated with patient self‐management. Conclusions Unfavorable health behavior and risk factors are associated with UO. However, our results suggest that UO with regard to perceived heart attack risk may not be a relevant factor for patient self‐management in those with type 2 diabetes.


| BACKGROUND
Type 2 diabetes is a major health concern worldwide and causes enormous societal costs. 1,2 Previous studies have shown that good selfmanagement can help slow down progression of the disease, prevent the occurrence of comorbidities, [3][4][5]

reduce mortality, and increase
health-related quality of life. 6,7 Unrealistic comparative optimism (UO) has been frequently suggested as a promising construct to explain health behavior and adherence in healthy and unhealthy individuals, and to ultimately tailor and improve interventions. 8,9 UO describes the tendency for people to make the erroneous assumption that they are less likely than others to experience a negative (health) event, for example, a heart attack. [9][10][11] The personal risk perception, relative and absolute, has been identified as a relevant factor for explaining preventive behavior. 12 Furthermore, other authors have reported that UO plays a role in all factors included in the Health Belief Model. 8 Therefore, UO might help explain differences in preventive behaviors, for example, self-management in patients with type 2 diabetes. 8,9 As Shepperd et al described, it is expected that individuals who show UO would show less preventive behaviors, that is, self-management. 13 In individuals with type 2 diabetes, the risk for a wide range of cardiovascular disease (CVD) is about 2-fold compared to individuals without diabetes. 14 Indeed, myocardial infarction (MI) accounts for more than 50% of all death in individuals with type 2 diabetes. 15 Therefore, an accurate risk perception with regard to MI is especially important for individuals with type 2 diabetes. Studies analyzing UO regarding MI on an individual level are uncommon and mainly concentrate on predictors of UO. 13 For example, Avis et al found that higher education was associated with a lower probability for UO. 16 Furthermore, Radcliffe and Klein reported that disease-specific education was associated with a lower probability for UO. 17 Moreover, Ayanian and Cleary found that smokers older than 64 years were more likely to show UO regarding their MI risk than smokers younger than 64 years. 18 In contrast, Strecher et al reported that young smokers (18-29 years), individuals with lower education levels, and females were more likely to show UO, compared to smokers older than 29 years, individuals with higher education levels, and males, respectively. 19 There have been few studies that have investigated the association between UO and health behavior where UO was determined by comparing a subjective and an objective risk, on an individual level. 13 In a study that is unrelated to diabetes and heart attack risk, Dillard et al reported higher rates of unpleasant alcohol-related events, for example, hangover or memory loss, among unrealistically optimistic individuals. 20 We found no studies on the association between UO and self-management in individuals with type 2 diabetes.
In this study, we measured individual-level UO with regard to the risk of suffering a MI with a method that is very similar to the way it has been proposed by Avis et al. 16 We compared participants' comparative risk judgments for having a heart attack (ie, "higher than that of other patients with type 2 diabetes of the same age," "about the same as that of other patients with type 2 diabetes of the same age," "lower than that of other patients with type 2 diabetes of the same age") with their objectively calculated individual comparative risk of having a CVD based on the Framingham risk equations. Subsequently, we examined the characteristics associated with UO, and tested the hypothesis that individuals who show UO have a lower adherence rate with regard to recommended self-management, in a sample of individuals with type 2 diabetes.

| Overview on the assessment of UO in the literature
The general approach to measure UO starts with measuring comparative risk perception. The comparative risk perception is assessed by asking individuals to rate their personal risk of experiencing an event of interest relative to an appropriate peer. These ratings can be assessed with either direct or indirect methods. 9,10 For the direct approach, participants are asked whether they consider themselves more likely, equally likely, or less likely to experience a certain event in comparison with their peers, for example, age group. 10 On a group level, the assumption is that if the mean comparative risk judgement of a group is below average, then this group shows UO at a group level. 9 For example, Weinstein used the direct approach and identified a lack of experience regarding the outcome of interest as a main predictor of UO at a group level. 23 However, this approach allows no conclusion about UO at an individual level. 9 The indirect approach combines two items. First, the participants are asked to rate their personal likelihood of experiencing the event of interest, and second, to rate the likelihood of experiencing the event of interest for the average person within their peer group. The difference score between both responses is considered the amount of comparative optimism or pessimism, respectively. 10 For example, Kim and Niederdeppe used an indirect approach and reported that comparative optimism had a moderating role in predicting intention to self-protect against H1N1. 24 Both the direct and the indirect approach, however, do not account for the actual individual-level risk of people. Hence, they do not determine whether the individuals' comparative judgments are actually unrealistic. 10 This can only be examined with the use of an objective comparator. 9,10 In other words, participants' estimates of whether they are equally likely, less likely, or more likely than others to experience a specific event, need to be compared with an objective comparator in order to test UO on an individual level. In health research, epidemiological risk equations are a practical option to measure people's objective risk to experience a specific event. 9,10,16,17,25

| Assessment of UO
We assessed UO using procedures modeled after the approach of Avis et al. 16 F I G U R E 1 Participant flow diagram First, we assessed the individuals' self-rated risk in comparison with other patients of their age with type 2 diabetes with the following question: "Do you believe that your personal risk of suffering a heart attack within the next 5 years is higher than that of other patients with type 2 diabetes of your age?" The response categories were: (a) "yes, I believe my personal risk is higher," (b) "I believe my risk is about the same," and (c) "no, I believe my risk is lower".
Second, in order to be able to compare the individuals' self-rated comparative risk with their actual comparative risk, we calculated the "office-based" Framingham risk (%), as described by D'agostino et al. 26 The score uses age, sex, body mass index (BMI), systolic blood pressure distinguished by treatment status, smoking status, and diabetes status to estimate the individual 10-year risk of suffering a CVD. 27 Third, we calculated the ratio (FR i ) of each individual's calculated Framingham risk (F i ) and the mean calculated risk of people of the respective age (FP i ). The FP i was estimated using a pseudo-binomial approach, calculating a binomial regression with logit link based on the distribution of calculated Framingham risks in our sample (FP i = exp(β 0 + age i × β age ). FP i was only regressed on age because participants were instructed to state their risk relative to other people of their age with diabetes. As described by Avis et al., we used the natural log transformation of the calculated ratio (ln(FR i )) and the cut-offs ln(0.75) and ln (1.33) in order to create a symmetric distribution and equal "risk distances". 16 (1)) and the dotted lines represent the cut-offs for ln(FR i ), that is, below average (ln(0.75)) and above average (ln(1.33)) to have a risk above average. Finally, we compared the self-rated risk with the calculated risk category. 16 When individuals self-rated their comparative risk as below average but their calculated comparative risk was average or above average, they were grouped with UO. Moreover, when individuals self-rated their comparative risk as average but their calculated comparative risk was above average, they were also grouped with UO. For Unrealistic comparative pessimism (UP), the grouping was done accordingly. See Table 1 for an overview.
Based on this approach, individuals with a low calculated risk (ln (FR i ) < ln(0.75)) could not be grouped with UO, and individuals with a high calculated risk (ln(FR i ) > ln(1.33)) could not be grouped with UP. To approach this conceptual limitation, we excluded individuals with a low calculated risk (ln(FR i ) < ln(0.75)) and individuals with a high calculated risk (ln(FR i ) > ln(1.33)) from all further analyses on UO (underestimation of comparative risk) and UP (overestimation of comparative risk), respectively. See Table 1 for an overview.

| Assessment of self-management
Our measures of diabetes self-management behavior were based on a compliance score introduced by Arnold-Wörner et al. 28 Within our study, we assessed the following self-management behaviors: monitoring of body weight (at least once per week), conducting regular foot care (checking for wounds at least once per week), measuring blood sugar (at least once a day for patients treated with insulin and at least once a week for all others), measuring blood pressure (at least once per week), keeping a diabetes diary, and having a diet plan. We asked participants to consider the last 6 months for their answers ((a) "daily," (b) "at least once per week," (c) "once or twice per month," (d) "less than once per month"). The respective cut-off points were based on recommendations by the European NIDDM (noninsulin-dependent diabetes mellitus) Policy Group 29 and the American Diabetes Association. 30 Furthermore, we combined the six self-managing behaviors into a self-management score. In this score, one point was attributed per criterion in every individual, as proposed by Arnold-Wörner et al. 28 A similar score has been shown to be highly predictive for all-cause mortality in patients with type 2 diabetes. 7

| Covariates
To calculate the Framingham risk (%), we derived BMI from body height measured at the respective baseline study and self-reported body weight at the time of GEFU 4. Age, sex, systolic blood pressure, blood pressure treatment status (medication), and smoking status were also based on self-report at GEFU 4. Other than that, we assessed whether participants' treatment regimen included the injection of insulin, as we assumed treatment with insulin as an indicator for disease severity. Furthermore, we assessed education (primary education, ≤10 years of school; secondary/tertiary education, >10 years of school) and whether participants had ever participated in a diabetes education program that was not part of routine care or during a hospital stay. Finally, we asked participants whether they had ever had a heart attack that was diagnosed by a physician.

| Statistical analysis
In a first step, we calculated frequencies and means with regard to measured characteristics and self-management behaviors-overall and stratified by the three categories of self-rated comparative risk, that is, "higher than others," "average," "lower than others". Second, we regressed the self-rated comparative risk on the Framingham variables (ie, age, sex, systolic blood pressure, blood pressure treatment status, BMI, and smoking status) and the variables education, participation in a diabetes education program, treatment with insulin, and history of MI. Likewise, UO and UP were regressed on the same set of variables in two separate binary logistic regression models.
Finally, we estimated binary logistic regression models and ordinary least square regression models to test the association between individual-level UO, UP, and the six measured self-management behaviors and their sum-score, respectively. We adjusted all models on the association with self-management for age, sex, BMI, blood pressure treatment status, systolic blood pressure, smoking status, education, participation in a diabetes education program, treatment with insulin, and history of MI. Additionally, we adjusted all models for self-rated risk. Thereby, we tried to disentangle the association between UO and self-management behavior from confounding by positive or negative self-view, that is, self-rated risk "lower than others" or "higher than others". As described by Humberg et al, the mere positivity of self-view needs to be differentiated from the Comparison between self-rated and calculated comparative risk Note: The cells with colored background were excluded from some parts of the analysis. Specifically, individuals with an objective relative risk below average (lighter gray) were excluded from analyses regarding UO because per definition they could not be grouped with UO. Likewise, individuals with an objective relative risk above average (darker gray), were excluded from analyses regarding UP because per definition they could not be grouped with UP. Abbreviations: UO, unrealistic comparative optimism; UP, unrealistic comparative pessimism.
erroneous positive self-view, that is, UO. 31 A P-value <.05 was considered to be statistically significant. Missing information in the items of the Framingham risk score was imputed using a predictive mean matching approach (see Table 2 for details). 32

| Sensitivity analysis
The Framingham risk is supposed to be calculated only for individuals <75 years of age and without a prior CVD. Therefore, we excluded individuals >74 years or with a history of MI in our first sensitivity analysis (n = 356).
In our second sensitivity analysis, we approached the issue that individuals might have compared themselves within their sex, even though the question did not imply this. Therefore, we estimated the mean risk (FP i ) in a binomial regression based on age and sex (FP i = exp . We then tested the association between UO and the assessed characteristics, as well as the association between UO and self-management similar to our main analysis. In further sensitivity analyses, we examined the association between UO and self-management using different cut-offs for the calculated risk ratio ln(FR i ). We tested very sensitive cut-offs, that is, ln(0.86) < ln(FR i ) > ln(1.16), and very specific cut-offs, that is, ln(0.60) < ln(FR i ) > ln(1.66)).
Finally, multiple previous studies did not exclude individuals with a low comparative risk or a high comparative risk from analysis including UO or UP, respectively. Therefore, in another sensitivity analysis, we repeated our main analysis without the exclusion of individuals based on their objective comparative risk.

T A B L E 2
Characteristics for the complete sample and self-rated risk groups Note: The variables used for calculating the Framingham risk were essential to our study. Within the 633 individuals who self-rated their comparative MI risk, we found 67 missing values for systolic blood pressure, 3 missing values for smoking status, and 11 missing values for BMI. In order to avoid loss of power for our analysis, we decided to apply a predictive mean matching approach, as introduced by Little 32 within the variables that were relevant to the calculation of the Framingham risk. The imputation was performed with the R package "Mice". 33 The self-management score was composed by adding the six self-managing behaviors into a single score, in which one point was attributed per criterion in every individual (See Methods). Abbreviations: D. education, diabetes education program (yes); MI, myocardial infarction.

| Ethical considerations
The study was approved by the Bavarian Medical Association  Details are shown in Figure 1 (see also Appendix Table A3). The mean self-management score was about the same in all groups of self-rated risk.
All details on the analyzed characteristics are shown in Table 2.
Characteristics of individuals with missing information regarding their self-management or their self-rated heart attack risk (n = 113) are reported in Appendix Table A4. Individuals with missing information were more likely to smoke and less likely to have higher education compared to individuals without missing information.

| Associations between the individuals' characteristics and self-rated risk, UO, and UP
Overall, 32% of the participants (200 of 633) rated their MI risk lower than that of others, while only 9% (54 of 633) rated their risk higher The association between UO, UP, and the individuals' characteristics in the main analysis (1) "Lower than others" (n = 200) (2) "Higher than others" (n = 54) Note: The association of patient characteristics with low comparative risk perception, high comparative risk perception, UO, and UP was examined in four binary logistic regressions (1 through 4). In (1), participants with average and high comparative risk perception were used as reference to the participants with a low comparative risk perception. In (2), participants with average and low comparative risk perception were used as reference to the participants with a high comparative risk perception. In (3), participants at average or high objective comparative risk and who were not grouped with UO were used as reference to participants with an average or high objective comparative risk but who were grouped with UO. In (4), participants at low or average objective comparative risk and who were not grouped with UP were used as reference to participants with a low or average objective comparative risk but who were grouped with UP. Abbreviations: UO, unrealistic comparative optimism; UP, unrealistic comparative pessimism.
than that of others. Males and individuals with a history of MI were more likely to self-rate themselves with a higher than average risk of suffering a MI in the future than females and individuals without a history of MI, respectively (Table 3). Individuals treated for high blood pressure were less likely than individuals without blood pressure treatment to self-rate their risk lower than that of other type 2 diabetes patients of their age (Table 3).
Within the studied sample, 32% of individuals (202 of 633) showed UO-that is, have a higher or equal calculated Framingham risk compared to other patients with type 2 diabetes of the same age but think their risk is average or lower than average, respectively. On the other hand, 23% (148 of 633) showed UP-that is, have a lower or equal calculated Framingham risk compared to other patients with type 2 diabetes of the same age but think their risk is average or higher than average, respectively (Table 1).
Males, smokers, individuals with a higher BMI, a higher blood pressure, and no history of MI were more likely than females, nonsmokers, individuals with a lower BMI, lower blood pressure, and no history of MI, to underestimate their comparative risk, that is, to show UO (Table 3). Accordingly, males, smokers, individuals with a higher blood pressure, and individuals with no history of MI were less likely than females, nonsmokers, individuals with a lower blood pressure, and individuals with a history of MI, to overestimate their comparative risk, that is, to show UP (Table 3). Furthermore, older individuals were less likely than younger individuals to show UP.

| Association between UO, UP, and the participants' self-management
Overall, we found no statistically significant association between the measured UO or UP and the six self-management behaviors (see Tables 4   and 5). However, the association of UO with self-management ( Table 4, model 2) was predominantly negative in its direction (OR < 1), while the association of a positive self-view, that is, rating the personal risk lower than that of others, with self-management was predominantly positive (OR > 1).

| Sensitivity analysis
In the subset of individuals under 75 years of age and without a prior CVD, we found very similar associations as reported for our main analysis. (Appendix Table A1 upper half).
When the objective comparator was based on a comparison between the calculated individual risk and the mean risk of individuals of the respective age and sex, smoking and a higher blood pressure were still significantly associated with UO and UP. However, the associations between sex, BMI, and UO and UP were not statistically significant anymore. Detailed results are provided in the lower half of Appendix Table A1.
The results of the sensitivity analyses, like those in the main analysis, showed no consistent and statistically significant associations between UO and patient self-management (Appendix Table A2). Finally, in our main analysis, we did not observe a statistically significant association between UO and self-management behavior.

| DISCUSSION
The relatively high frequency of unrealistically optimistic responses compared to unrealistically pessimistic responses on a group level, as well as on an individual level, was not surprising. Similar results have been reported in previous studies, 16,17 and with respect to other negative events on a group level, 8,11,35 and on an individual level. 25,36 One reason for the predominantly optimistic responses may be the person-positivity bias. 9,37 Person-positivity bias states that individuals dehumanize the "average person," which leads to a worse rating of the "average person," 37 and hence, to a better self-rating. 9 Most of the results regarding participant characteristics that were associated with UO are in line with findings from previous studies.
Individuals with a history of MI were less likely to show UO concerning heart attack risk in our study. Likewise, the very first studies by Weinstein 11,23 or Helweg-Larsen and Shepperd 10 found that personal experience was associated with less prevalent UO. 10,11,23 Homko et al. 38 reported that in a sample of individuals with type 2 diabetes, males had a lower comparative risk perception than females when they were asked to compare their CVD risk with others of their age and sex. 36 In our main analysis, we observed that males were also more likely than females to show UO. However, when the objective comparator was based on a comparison between the calculated individual risk and the mean risk of individuals of the respective age and sex in our sensitivity analysis, this association was no longer statistically significant. Therefore, it is likely that the association in our main analysis resulted from males and females comparing themselves to other individuals of their age and sex, even though the question did not imply this.
Smokers were more likely than nonsmokers to show UO in our study. Strecher et al also reported that smokers were more likely than nonsmokers to show UO. 19 Furthermore, Ayanian and Cleary reported that many smokers did not perceive themselves at increased MI risk when asked to compare themselves with nonsmokers. 18 The association between increased blood pressure and UO, which was very robust towards any alterations in our sensitivity analyses, has not been reported in previous studies that examined UO. Therefore, smokers and individuals with higher blood pressure seem to underestimate the increased heart attack risk that results from their respective behavior or characteristic.
The results of our main analysis show that UO and UP were not associated with the measured self-management behaviors. This was surprising, because theory suggests that UO is a relevant factor in explaining health behavior. 8,10,13 As Shepperd et al described, we would have expected that individuals who showed UO would show less preventive behaviors, that is, self-management. 13 However, our results suggest that UO is not a relevant target when aiming to improve the adherence to self-management recommendations in individuals with type 2 diabetes.
There are characteristics of our study design that might help explain some of our null results. One explanation could be the domain specificity of UO. Weinstein showed that mean comparative risk judgments varied between different health threats. 23 Hence, the measure of UO and the outcome of interest need to be directly associated with each other. Five of our self-management measures, that is, regular self-monitoring of body weight, blood sugar, and blood pressure; We tried to disentangle the association between UO and selfmanagement behavior from confounding by a positive self-view. Therefore, we included positive self-view, that is, self-rated risk "lower than others," as an additional covariate in our regression model. The results suggest that UO and positive self-view have opposing effects on self-management. Therefore, future studies should consider similar adjustments when examining the association between UO and health behavior.
Our study has several limitations. It is a general concern in surveys that self-report data suffer from recall bias. However, it is of even greater concern in our study where we based the objective comparator, that is, Framingham risk, on self-reported data. Nonetheless, a study by Okura et al supports the use of self-reported information on at least MI and hypertension, as they reported a very high correlation between self-report and clinical records, that is, 98% and 88%, respectively. 39 Furthermore, we had no information on the year that the participants had a MI or participated in a disease-specific education program, so we could not adjust for the time past between these events and data collection. Moreover, person-positivity bias might have affected the participants' responses to our subjective risk question. 37 Future studies could consider not making participants compare themselves with an "average person" but with one specifically described comparator that represents an average person. For example, Chock found that comparative optimism with regard to the healthfulness of lifestyle decreased when college students were asked to compare themselves with their best friend. 40 Another concern is that we assessed MI risk perception while comparing it with the CVD risk. However, due to the similarity of risk factors for MI and CVD and the resulting linearity between the absolute risks for MI and CVD, asking for CVD risk is justifiable. 41 Finally, our comparative risk question instructed participants to compare their risk with the risk of other patients with type 2 diabetes of their age. Hence, the instruction did not include sex specificity as most of the previous studies did. 9,16 Accordingly, our main analyses compared the individual comparative risk perception with the ratio between the calculated individual risk and the mean risk of people of the respective age. However, as it is possible that participants compared themselves with peers of the same age and sex, we also estimated the objective comparative risk based on a comparison between the calculated individual risk and the mean risk of individuals of the respective age and sex. Although the overall pattern of associations was qualitatively quite similar, some of the associations of our main analysis were no longer statistically significant. Given this result, we cannot exclude the possibility that some of the participants might have compared themselves with other individuals of their age and sex, even though the comparative risk question did not imply this. Therefore, we would recommend using an age and sex specific question in the future. Another possible issue in our study is selection bias. Of 746 individuals with type 2 diabetes, 113 individuals had missing information that we could not impute. On average, these individuals had a lower education and were more likely to smoke than individuals without missing information. Finally, due to the observational cross-sectional design of our study, reverse causation and residual confounding cannot be excluded.
One strength of this study lies in the strong theoretic foundation of the methodological approach that takes into account several ideas from previous studies to overcome general issues in the field, for example, the distinction of the positivity of self-view, 31

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from KORA (https://www.helmholtz-muenchen.de/en/kora/for-scientists/ cooperation-with-kora/index.html) but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. However, data can be requested through an individual project agreement with KORA via the online portal KORA.passt (https://epi.helmholtz-muenchen.de/).

ORCID
Florian M. Karl https://orcid.org/0000-0002-3155-4941 data was substantial for diabetes, hypertension, myocardial infarction and stroke but not for heart failure.  (1) and (3), participants at average or high objective comparative risk and who were not grouped with UO were used as reference to participants with an average or high objective comparative risk but who were grouped with UO. In model (2) and (4), participants at low or average objective comparative risk and who were not grouped with UP were used as reference to participants with a low or average objective comparative risk but who were grouped with UP. Abbreviations: UO, unrealistic comparative optimism; UP, unrealistic comparative pessimism.