Grant: Project Grant

A family history of a common chronic disease, such as heart disease, can help to identify people with a higher-than-usual risk of developing a chronic disease in the future. Collecting information about chronic disease histories for individuals and their family members can be difficult because self-reported information is often not accurate for events from the past. Diagnosis codes in hospital records and physician billing records can be used to build personal and family disease histories for entire populations and are not affected by recall bias. However changes in disease coding systems over time and differences in healthcare systems around the world can affect the accuracy of disease history measures, particularly for complex diseases. Our study will compare two methods to create disease histories using healthcare data from Manitoba and Denmark and assess their value for predicting disease risk. One method uses a set of rules to assign diagnosis codes to a small number of clinician-defined categories and then translates these categories over time. The second method uses machine-learning methods to search for related diagnosis codes collected over time and then combines the codes into groups based on the strength of these relationships. Healthcare data from Denmark will be used to assess whether these methods will produce comparable results in a different healthcare setting. In both Manitoba and Denmark, we will build models to predict the risk of common chronic diseases, including heart disease and mental disorders, and describe the predictive value of personal and family disease histories from machine-learning and rule-based clinical classification methods. Software tools and code will be developed to share our methods. Our research will improve studies about population disease burden and outcomes for at-risk populations and will benefit scientists from Canadian and international healthcare data centres.