Using electronic health record data to evaluate heterogeneity in clinical outcomes of people with diabetes

Abstract

Diabetes is highly heterogeneous condition with a broad range of potential characteristics that can influence a person's outcomes. Overall, this thesis aimed to use electronic health record data to evaluate heterogeneity in clinical outcomes across different groups of people with diabetes.

Electronic health record data are a powerful resource for diabetes research and provide large quantities of longitudinal data that is collected as part of routine clinical care. This makes it ideal data for studying clinical outcomes in people with diabetes. This type of data is particularly valuable where randomised controlled trials, considered the gold standard in many areas of research, are limited. For example, infections are a leading cause of morbidity and mortality in people with diabetes, however few diabetes clinical trials evaluate infection related outcomes. Many clinical trials also have restrictive entry criteria and often exclude certain groups of people all together, for example type 3c diabetes, an understudied condition but that is estimated to make up 5-10% of diabetes cases.

In this thesis, electronic health record data was used to define robust diabetes research cohorts and in 3 research studies to evaluate important clinical outcomes relevant to people with diabetes.

Transforming raw electronic health record data into robustly coded datasets ready for analysis is challenging and time consuming. Therefore, we firstly developed a data-processing framework to reproducibly define standardised research cohorts of people with diabetes. This framework was developed and demonstrated using primary care data from the Clinical Practice Research Datalink and linked datasets but is generalisable to any coded electronic health record dataset.

We then compared risk factors for major respiratory infections (Covid-19, influenza, and pneumonia) in people with type 1 and type 2 diabetes. We demonstrated that clinical risk factors such as poor glycaemic control and obesity were consistently associated with severe respiratory infections, but that HbA1c and BMI associated risks varied by age and sex. In contrast, sociodemographic risk factors such as age, sex, ethnicity, and deprivation, showed clear differences by infection type.

We then used Mendelian randomisation to evaluate the potential causal role of higher BMI and higher HbA1c on common infections. We found strong evidence of a causal effect of higher BMI on bacterial and fungal skin infections in primary care and in hospital, identifying a potential target for intervention.

Finally, we evaluated treatment outcomes in people with type 3c diabetes on major oral glucose-lowering therapies. We demonstrated that oral therapies were effective in people with type 3c diabetes, and for most of the cohort the average HbA1c response was similar to type 2 diabetes. We also found that pancreatic exocrine insufficiency was associated with modestly reduced HbA1c response and tolerability.