Identification of diabetes records and implementation of algorithm to determine type of diabetes in AIM-CISC

Description of workflow This document describes the diabetes-specific workflow used in the AIM-CISC project at the University of Edinburgh, led by Prof Bruce Guthrie. We describe how we identified diabetes records in Clinical Practice Research Datalink (CPRD) Aurum and how we implemented the modified CALIBER algorithm to determine type of diabetes. Identification of diabetes records in CPRD Aurum First, we used one ICD-10 and two Read v2 code-lists developed by Kuan and colleagues [1] to identify all records of type I diabetes, type II diabetes and other or unspecified diabetes in hospital and primary care records. The code-lists are available here: Hospital records of diabetes: https://github.com/spiros/chronological-map-phenotypes/blob/master/secondary_care/ICD_diabetes.csv Diabetes diagnoses in primary care records: https://github.com/spiros/chronological-map-phenotypes/blob/master/primary_care/CPRD_caliber_diabdiag_cprd.csv Other diabetes codes in primary care records: https://github.com/spiros/chronological-map-phenotypes/blob/master/primary_care/CPRD_caliber_dm_cprd.csv Implementation of algorithm to determine type of diabetes Since participants may have records of different types of diabetes in health records but types of diabetes are mutually exclusive from a clinical perspective, we then used these records to define type of diabetes by implementing the modified CALIBER 'Diabetes' phenotyping algorithm, which is outlined on the HDR UK phenotype library webpage (see https://phenotypes.healthdatagateway.org/phenotypes/PH152/version/304/detail/). The algorithm ensures that different types of diabetes are mutually exclusive. Thus, after implementing the algorithm, each participant can only ever have one type of diabetes. Date of diabetes We defined the minimum and maximum date of diabetes as earliest and latest date across all data sources, irrespective of type of diabetes. Number of records of diabetes For each participant, we defined the number of diabetes records as the number of any diabetes records across all data sources, irrespective of type of diabetes.

Prigge, R., Guthrie, B., Palmer, J., & De Ferrari, L. (2024). Identification of diabetes records and implementation of algorithm to determine type of diabetes in AIM-CISC (version 1.0). Zenodo. https://doi.org/10.5281/zenodo.12698044