The comorbidity burden of type 2 diabetes mellitus: patterns, clusters and predictions from a large English primary care cohort

The Clinical Practice Research Datalink (CPRD) is a database of anonymised electronic, primary health records. In January 2017, the CPRD held data on nearly 17 million active and historical patients registered with 714 general practices across the UK. It contains information on diagnoses, referrals, tests and therapy records, which are mainly recorded using Read clinical codes. Additional data is available for a subset of English practices (nearly 75% of English practices; 58% of all UK CPRD practices) which consented to participate in the CPRD linkage scheme and provided patient-level information. To obtain information on social deprivation at the level of the patient’s postcode, we used the linked information on the quintiles from the 2015 Index of Multiple Deprivation (IMD) measure, which aggregates data on income, employment, health and disability, education and training, barriers to housing and services, crime and living environment.

People registered with a general practice in England meeting CPRD data quality standards and with the first T2DM Read code recorded at any point between 1 April 2007 and 31 March 2017 were included. The inclusion criteria for this study were as follows: patient registered with a CPRD practice for at least 365 days before T2DM diagnosis, aged 35 years and older and no recorded diagnostic code for type-1 diabetes mellitus. In the UK, T2DM has been incentivised since 2004 through a national pay-for-performance scheme, the Quality and Outcomes Framework (QOF), along with another 20 clinical domains approximately, resulting in uniformity in Read code usage and recording. The index date was defined as the date of first recorded code for T2DM and the follow-up as the time between the index date and the earliest of date of death, transfer out of practice date and last date of data collection from the practice or the end of study period (31 March 2017). The lists of codes used to establish the presence of each comorbidity were downloaded from clinicalcodes.​org and CPRD@Cambridge websites.

We selected the following 18 conditions: coronary heart disease (CHD), chronic kidney disease (CKD), atrial fibrillation, stroke, hypertension, heart failure, peripheral vascular disease (PVD), rheumatoid arthritis, cancer, osteoporosis, depression, asthma, chronic obstructive pulmonary disease (COPD), dementia, severe mental illness (SMI), epilepsy, hypothyroidism and learning disability. The reporting of these conditions is financially incentivised under the QOF, and consequently, they are well-recorded in the CPRD. The presence of asthma, epilepsy and depression was determined using Read codes and prescription data, since these can be acute or resolvable. Each condition was considered to be present at the index date if it satisfied the definition criteria at the time of the T2DM diagnosis (Additional file 1: Table S1). Each condition was considered to be present during the follow-up period if it satisfied the definition criteria at the index date or at any time during the follow-up.

We calculated the age-standardised prevalence of each condition, stratified by gender, for patients from the least and most deprived areas. We also calculated the crude and age-standardised co-prevalence of each pair of comorbidities for the whole sample and stratified by gender, deprivation (least and most deprived areas) and age (using 35–54-, 55–74- and 75+-year-old age bands).

We longitudinally calculated the prevalence of each comorbidity present at the time of the T2DM diagnosis in the incidental cohort of patients with T2DM, for financial years (April to March) 2007/2008 to 2016/2017. To forecast the proportion of people diagnosed with T2DM in the next 10 years that will also have a particular comorbidity present at the time of diagnosis, we used linear regression on log-transformed, age-standardised prevalence. For clarity of results, we present the patterns for the six most prevalent conditions as the prevalence of remaining conditions remained relatively low and stable over the study period.

Lastly, we selected patients with two or more comorbidities present at the index date and used agglomerative hierarchical clustering to identify groups of similar conditions. Similarity was assessed using the tetrachoric correlation coefficient. Tetrachoric correlation estimates what the correlation for two binary variables would be if they were measured on a continuous scale. We used Ward’s linkage method to group conditions. At each linkage step, Ward’s method finds a pair of clusters that leads to a minimum increase in total within-cluster variance after merging. To avoid chaining (low prevalence comorbidities being sequentially linked to existing clusters), we excluded conditions with prevalence in a given group below 3%. Cluster analysis was stratified by gender, age bands (35 to 54 years, 55 to 74 years and ≥ 75 years old) and deprivation using the least and most deprived quintiles. We present the results for the whole sample. Stratified results are available in Additional file 1: Figure S12–S18. To assess the progression in clustering patterns, we performed the cluster analysis for conditions present at the time of T2DM diagnosis and those present at 2, 5 and 9 years after. We plotted the results in dendrograms and identified clusters using visual analysis. Dendrograms visually represent the clustering. The heights at which conditions fuse together correspond to their similarity. The earlier the branches merge, the more similar the groups of conditions are. The clustering structure was measured using the agglomerative coefficient, with values closer to zero suggesting tight clustering of objects and values closer to one suggesting less well-formed clusters. Due to differences in sample sizes, agglomerative coefficients should not be compared across groups. We used R version 3.4.2 for the analysis and data preparation.

We showed important changes in the comorbidity patterns in a large real-world cohort of people living with T2DM, using data from the UK primary care. Our findings are relevant to patients, clinicians and policy-makers and can inform on the healthcare needs and how best to prioritise and deliver care for people with T2DM. We identified alarming levels and trends of depression prevalence, which we estimated will continue to grow over the next decade. This could have major consequences for how to offer these patients integrated care. Health systems will have to respond to a growing need for diagnosis and management of mental health problems among people with T2DM, underpinned with established links between depression and poor glycaemic control [18], treatment adherence [19], diabetes complications [9] and mortality [20]. The differences in comorbidity patterns observed in groups stratified by gender and social deprivation highlight the need to address the present and increasing health inequalities, particularly with higher prevalence of comorbidities in patients from more deprived areas.

To the best of our knowledge, this is the largest study of comorbidities in patients with T2DM in England. The quality of the data is very high for our study period, primarily due to data recording in line with the QOF and the financial incentives offered to UK primary care for the management of chronic and other conditions such as T2DM.

However, the study has limitations. First, due to the low prevalence of some conditions in general and in specific groups, some comorbidities were excluded from the cluster analysis for all or some strata. However, all conditions were included in the frequency analysis which provides a starting point for the analysis of grouping patterns of specific conditions. Second, we selected only 18 conditions for which recording quality was high, but patients may have additional comorbidities impacting on their disease management and quality of life. Third, some of these comorbidities, like CKD and CHD, are closely linked to T2DM, to the extent of them being considered its complications. However, the majority of patients with these conditions do not have T2DM, while the characterisation of these conditions is immaterial to our analyses. Fourth, to identify patients with depression, we used an algorithm analysing prescriptions as well as diagnostic codes. We were unable to discriminate uses of antidepressants for other conditions such as obsessive-compulsive or bipolar disorders; therefore, patients with other mental health conditions might have been incorporated into the depression group. Fifth, the predictions of future prevalence rates were obtained from linear regression models, which are dependent on certain assumptions such as the linearity of the trend. Sixth, some of the conditions we modelled may be present but undiagnosed in our cohort. Seventh, for the hierarchical clustering, each comorbidity is necessarily considered into a single cluster, which may not be the case [21]. Last, some diagnostic criteria were also changed during the study period, for example, the diagnostic criteria for hypertension. Therefore, the average number of comorbidities calculated in our sample is likely to be underestimated both due to the finite set of conditions we used and to non-diagnosis in practice.

We found that almost 75% of patients had at least one additional comorbidity at the time of T2DM diagnosis and 44% had at least two comorbidities. Prevalence of multiple conditions in addition to T2DM was lower than that reported in some clinical trials (90%) [22] or studies using administrative data (91.4%) [23] (84.6%) [24] but higher than in others (44%) [25]. However, our population was younger than in some studies, and we analysed a large but not exhaustive list of conditions. As expected, the burden of comorbidity increased with age, however, contrary to previous research [4, 8], which found a higher age-standardised prevalence of coexisting comorbidities in males or no gender difference, we found that the burden was higher in females. This reflects the pattern in the general population which shows that females tend to have more comorbid conditions than males [26]. This difference may relate to the surveillance bias with females being more likely to visit a general practitioner and therefore have a recorded diagnosis of comorbidity. In addition, previous studies tend to focus on conditions regarded as diabetes-concordant such as cardiovascular diseases and CKD [4]. Females with T2DM were found to have a lower probability of these having conditions and a higher prevalence of depression, which we included in our study [23]. The presence of mental health problems may have a significant impact on the ability of the patient to manage their condition, progression of T2DM [8, 16, 18]. Our findings of the high and increasing prevalence of depression in patients with T2DM imply that the inclusion of mental health conditions is essential in studies of comorbidities in this population. We found that the prevalence of all conditions except asthma and depression increased after diagnosis of T2DM. The fall in the prevalence of treated asthma during the follow-up may be related to the correlation between metformin use and decrease in asthma exacerbation [27]. Knowing that T2DM is highly correlated with obesity, as is asthma [28] and depression [29], it may be that patients after being diagnosed with T2DM work towards lowering their BMI, and therefore, both conditions may be resolved.

We observed a higher burden of comorbidity among people from the most deprived than the most affluent areas. Differences were also observed in the prevalence of specific conditions, notably higher prevalence of depression, CHD, asthma and COPD among people from the most deprived areas. This is consistent with other studies and may be explained by the higher prevalence of risk factors such as smoking, obesity and alcohol consumption [30, 31].

We found a very large increase in the prevalence of T2DM-comorbid depression, which is expected to rise over the next 10 years. The rising prevalence of depression and the large gender gap has also been observed for the general population [32]. There is an ongoing discussion over whether antidepressants are overprescribed [33, 34] which could explain the rise in depression observed in our analysis. Furthermore, the data may represent rises in conditions other than depression such as chronic pain for which antidepressants can be prescribed [35]. Although this discussion is inconclusive, the rise in antidepressant use in patients with T2DM should be a concern, with some evidence proposing that some antidepressants may be an independent risk factor for T2DM [36], suggesting that both conditions share similar risk factors. More research is needed to provide further insight into the increase in depression and antidepressants use in patients with T2DM. Nevertheless, people with both T2DM and depression may require tailored approaches of treatment for both conditions as depression was found to impair patients’ ability to manage their diabetes [15].

The observed and predicted stable or decreasing prevalence of comorbidities other than depression at the time of T2DM diagnosis may reflect the increase in the proportion of people diagnosed at a relatively early age [37]. This could mean that people are diagnosed with T2DM before they develop other comorbidities.

Our hierarchical clustering analysis showed that conditions regarded as diabetes-concordant (stroke, atrial fibrillation, CKD, CHD, hypertension, PVD and heart failure) tend to group together in all analysed groups. Cancer has been linked with different condition groups, depending on the analysed stratum. This may be due to the fact that we grouped all types of cancer into one condition. However, specific types of cancer may be more prevalent in different groups and be linked with the conditions sharing common risk factors. At the time of the T2DM diagnosis, the clusters seem to follow an expected pattern with lung diseases (asthma and COPD), mental health conditions (depression and SMI) and vascular conditions (PVD, CHD, stroke, atrial fibrillation and heart failure) grouped together. However, the grouping becomes more complex after the diagnosis with conditions needing different treatment and management likely to occur together. These complexities highlight the need for patient-centred approach. Furthermore, greater emphasis is needed on preventative actions and constant monitoring for conditions not closely related to the ones already experienced by the patient.