The present study provides nationally representative data on the prevalence and comorbidity of known diabetes among older non-institutionalized adults in Germany. The overall 12 month-prevalence of known diabetes was high at 13.6% with no difference between men and women. Although we did not collect information about the type of diabetes, it can be generally assumed that the vast majority of diabetes identified from older adults aged over 50 years were type-2 diabetes. Persons with known diabetes were significantly more likely to suffer from additional chronic health conditions than persons without diabetes, irrespective of potential confounders. As expected, these associations were most pronounced for diabetes-concordant comorbidities, i. e. conditions in the pathophysiological pathway of diabetes. Severity of diabetes classified as grade 1–4 according to the type and number of concordant comorbidities positively and independently correlated with age and the number of diabetes-discordant comorbidities as well as with specific diabetes-discordant conditions including depression, chronic lower respiratory disease, musculoskeletal disease, and severe hearing impairment.
Prevalence of known diabetes
The prevalence of diabetes varies considerably between countries  and even between regions within a given country . This may be partly explained by differences in ethnic or socioeconomic background as well as differences in health care systems . However, comparisons between studies are often compromised by methodological differences regarding data collection mode, the age range of the study population, and the diagnostic criteria used to define diabetes. In the present study, we used 12-month-prevalence estimates. Previous sex and age stratified analysis of GEDA 2009 data including study participants of all age groups showed that lifetime prevalence estimates were consistently higher than 12-month-prevalence estimates with absolute overall differences of 1.8% among women and 1.0% among men . The largest differences were found among women in the age group 30–44 (2.5%) and 65+ years (2.7%) , probably reflecting gestational diabetes.
In Germany, several nationwide health surveys have been conducted since the reunification using standardized computer-assisted interview technique, either via telephone or face-to-face interview [11, 32], Heidemann C, Du Y, Schubert I, Rathmann W, Scheidt-Nave C: Prevalence and temporal trend of known diabetes mellitus. Results of the German Health Interview and Examination Survey for Adults (DEGS1). Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 2013. Forthcoming]. In these previous surveys, age-specific prevalence estimates in older age groups are similar to those observed in the present study . In contrast, prevalence estimates obtained from health insurance claim data [14, 15] have been consistently higher than results of national health surveys. This has partly been explained by selection bias, as most analyses are derived from AOK (Allgemeine Ortskrankenkasse) data. The AOK is a large German sickness fund insuring a particularly high proportion of older persons, persons with multiple concurrent health problems and persons on social welfare. Information on health insurance provider was obtained in GEDA, which permitted the calculation of diabetes prevalence estimates stratified by insurance company. Persons insured by the AOK were significantly more likely to have known diabetes compared to those insured by other providers (data not shown). This was also demonstrated in previous national health interview surveys conducted by telephone or postal questionnaires as well as in recent national health examination surveys based on two-stage stratified random sampling from local population registries [16, 32], Heidemann C, Du Y, Schubert I, Rathmann W, Scheidt-Nave C: Prevalence and temporal trend of known diabetes mellitus. Results of the German Health Interview and Examination Survey for Adults (DEGS1). Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 2013. Forthcoming]. The prevalence estimates of known diabetes among persons with AOK insurance observed in our study were similar to published prevalence estimates based on data from the AOK Hesse . A recent pooled analysis of several regional population-based studies demonstrated considerable regional differences in the prevalence of known type-2 diabetes; in agreement with our results, prevalence estimates were highest in the eastern parts of Germany . The underlying reasons are subject to ongoing investigation (http://www.kompetenznetz-diabetes-mellitus.net). Extending comparisons to results of national health surveys in other western countries, our recalculated age and sex-specific prevalence estimates were comparable to those reported in the US National Health and Nutrition Examination Survey 2003–2006  and in the French Nutrition and Health Survey 2006–2007  among persons 50–74 years of age. Among persons 75 years of age and above, our prevalence estimates exceed those obtained in the US study  and in the annual Health Survey for England 1994–2006 . However, differences in the definition of known diabetes need to be considered, e. g. lifetime vs. 12-month prevalence estimates and exclusions of women with gestational diabetes.
Associations between diabetes and other chronic conditions
In the present study, diabetes was significantly associated with a wide range of chronic conditions. The strength of these associations was most pronounced for diabetes-concordant comorbidities, including cardiometabolic risk factors, cardiovascular disease and chronic renal disease. These conditions are well known to be in the pathophysiological pathway of diabetes . Self-reported severe visual impairment was not significantly related to diabetes in our study after adjusting for age and other covariables. There are several possible explanations for this. First, visual impairment is strongly related to older age, causes other than diabetic retinopathy prevail in older compared to younger persons with diabetes [37–40]. Thus, the difference between persons with and without diabetes may be less pronounced in older age. Secondly, self-reported visual impairment does not permit to differentiate between uncorrectable and correctable visual impairment. Uncorrectable but not correctable visual impairment as assessed by measurements of visual acuity and automated refraction was significantly more prevalent among adults with than without diabetes in the National Health Interview and Examination Survey (NHANES) 1999–2004 after adjustment for confounders . Finally, the risk of severe visual impairment due to microvascular complications increases with the duration of diabetes  and long-standing diabetes may be underrepresented in our survey.
We also observed significant associations between diabetes and diabetes-discordant comorbidities, i. e. comorbidities less evidently related to diabetes. Previous studies reported associations of diabetes with chronic liver disease  and various types of cancer . Study results regarding the relation between diabetes and musculoskeletal conditions are less consistent [43, 44]. A relation between diabetes and certain site-specific cancers has been attributed to hyperinsulinemia, but the causal link is still subject to debate . We observed no independent association between diabetes and asthma or chronic bronchitis. Results of previous investigations of these associations are conflicting [46, 47]. The association of diabetes with upper gastrointestinal tract disease may imply helicobacter pylori infection as a possible common pathway of the two diseases . Unlike previous studies, we did not observe a significant association between a history of diabetes and self-reported severe hearing impairment . Also in contrast with earlier reports , an association between diabetes and depression was restricted to women in the present study. We cannot exclude that associations of diabetes with both conditions were biased towards the null in the present study due to non-participation.
Comorbidity patterns and severity of disease among diabetics
One of the key goals of diabetes management programs is to prevent diabetes-concordant complications involving target organ damage. In our study, 43% of persons with known diabetes reported at least one such comorbidity. Few previous studies have systematically analyzed comorbidity patterns in population-based samples of persons with diabetes. In the German DETECT study of diabetic patients recruited from a nationwide sample of general practices, half of patients with type-2 diabetes (50.2%) had at least one diabetes-related micro- or macrovascular complication . Similar results were also found in other studies conducted in Germany [18, 51]. In a population of patients with diabetes identified from South Glamorgan in the UK, Morgan et al. found that 25.2%, 9.6%, 18.1%, 16.5% and 2.0% of diabetic patients had CHD, cerebrovascular disease, diabetic foot, retinopathy and nephropathy, respectively, while 52% of diabetic patients had none of these studied micro- and macro-vascular complications . These results have been confirmed by additional regional investigations in the UK . In summary, results from previous studies regarding the proportion of persons with diabetes who also have severe diabetes-related end-organ disease are roughly in line with our observations. Direct comparisons between studies are precluded by differences in study design, setting, mode of data collection as well as the type of diabetes-related comorbidities considered.
Within the given limits of the available database, we classified diabetes severity based on the type and number of diabetes-concordant comorbidities. The grading system presented here is self-developed based on evidence derived from published studies of comorbidity. It is well known that treatment and prognosis of persons with diabetes mellitus or any other index disease is likely to be influenced by co-existing health conditions, whether they are in the pathophysiologic pathway of the index disease or not . This is particularly true among older persons with diabetes who tend to have multiple health conditions. In order to describe diabetes comorbidity in a comprehensive and systematic way, we applied the idea of the Cumulative Illness Rating Scale (CIRS), aggregating individual health problems according to organ systems [54, 55]. From a public health perspective, it seemed crucial to develop a grading system of diabetes comorbidity that would not only permit to differentiate between diabetes-concordant and diabetes-discordant comorbidities, but also between target organ complications of diabetes and systemic cardiometabolic conditions likely to coexist or even to precede the onset of diabetes, such as hyperlipidemia. The proportion of persons with diabetes mellitus who already have macro- or microvascular complications may serve as an indicator to monitor time trends and spatial distributions of diabetes management and quality of care.
While we think this is a step in the right direction, criteria of diabetes severity in population-based epidemiological studies need to be refined and consented to be clinically relevant and to permit comparisons between studies. In particular, there is need for studies including objective measures of concomitant cardiometabolic risk factors (e. g. blood pressure, serum lipids), glucose control (glycosylated hemoglobin), inflammation (high sensitivity C-reactive protein) and diabetes-related complications (neuropathy, diabetic foot, diabetic retinopathy, diabetic nephropathy). In addition, characteristics of patient complexity (living alone, depressive symptoms, cognitive impairment) need to be considered. In the present study, we observed a significant and independent association between diabetes severity and depression, chronic lower respiratory disease, musculoskeletal disease, and severe hearing impairment. These results may indicate an interactive effect of diabetes and other chronic conditions on cardiovascular target organ damage. In fact, results from epidemiological studies have suggested that CVD may be related to depression , chronic obstructive pulmonary disease , osteoarthritis , and rheumatoid arthritis . Further, all these conditions tend to co-exist with CVD in older individuals and may hence interfere with effective patient counseling and treatment . Hearing impairment may interfere with the severity of diabetes. In a retrospective analysis of laboratory and audiometric data of diabetic patients, progression of diabetes correlated with worsening of hearing ability .
Strengths and limitations
GEDA 2009 is a nationally representative recent health telephone survey with a large sample size and comprehensive information on self-reported physician-diagnosed chronic conditions. We assessed systematically concordant and discordant comorbidities of diabetes. However, there are several limitations.
First and most importantly, we have to consider selection bias. This national health survey is confined to the non-institutionalized population. Regarding the population in private households, selection bias due to the exclusion of persons residing in households without landline telephones is possible. There is evidence that the proportion of persons exclusively using a mobile phone is increasing, particularly among younger adults living on their own . As the present analysis was confined to persons aged 50 years and older, selective underrepresentation of this particular subgroup can be expected to be small. However, given a cooperation rate of 51.2% at the respondent level, we cannot exclude selection bias due to other reasons. Survey adjustment weights were applied as computed from deviations between study participants and census data for the non-institutionalized German population within strata of age, sex, educational attainment, and region (see Additional file 1 and Additional file 2). Nevertheless, non-responders may well differ from study participants with respect to other characteristics relevant to the study variables of major interest and these differences may have biased our results. Data regarding diabetes-related risk factors based on census data are scarce and limited to smoking and BMI computed from self-reported body weight and height. Comparing age specific prevalence estimates of smoking and BMI status obtained in the present study to estimates derived from the Microcensus 2009 did not demonstrate significant underestimation of these risk factors (see Additional file 3).
Secondly, GEDA 2009 was not specifically designed for the investigation of diabetes and its complications, hence information on treatment, type or duration of diabetes was not collected. Type 2 diabetes can be assumed to predominate in the adult population. However, information on duration of diabetes and metabolic control would have been useful to test the hypothesis that these factors are related to higher comorbidity. This should be addressed in future studies of diabetes comorbidity. Third, the definition of some chronic conditions was rather crude. For example, we asked survey participants about any cancer without differentiation between specific types of cancer. However, some cancers are more common in patients with diabetes, while prostate cancer occurs less often in men with diabetes compared to those without diabetes [45, 63]. Furthermore, some of the disease categories such as ‘musculoskeletal disease’ probably included pathogenetically heterogeneous conditions.
Finally, information on diabetes and all other 20 chronic health conditions was self-reported and verification by medical records and/or laboratory tests was not possible. Assessment of health conditions by self-report bears the risk of misclassification due to over- or underreporting. Persons with diabetes may be more likely to report asymptomatic diabetes-related health conditions such as hypertension and hyperlipidemia than persons without a diagnosis of diabetes, due to higher health care services utilization . On the other hand, survey participants, in particular older persons may be unable to name or to memorize medical diagnoses correctly. A validation of self-reported diagnoses against objective health data (e. g. biochemical measurements and current medication use) in the Utrecht Health Project demonstrated that assessment by self-report is likely to lead to underestimation of disease prevalence estimates. The magnitude of bias varies according to the type of disease as well as the population studied . Nevertheless, our prevalence estimates for diabetes and other highly prevalent health conditions are well in line with estimates from other populations-based studies [33, 34] as well as one recent German study conducted in the primary care setting . The validity of our indicator for diabetes is further supported by the fact that associations of known diabetes to sociodemographic variables and major established risk factors of diabetes were all highly significant and in the expected directions (Table 1).