Opportunities and Challenges in Developing a Cohort of Patients with Type 2 Diabetes Mellitus Using Electronic Primary Care Data

Abstract

PurposeTo develop a cohort of patients with T2DM treated with insulin using CPRD to obtain an accurate diagnosis date. This was used to analyse time from T2DM diagnosis to first ever insulin prescription between 01/01/2000 and 30/06/2012, for patients in England and Wales.MethodsPatients aged 18 years and over at diagnosis, were included if prescribed an anti-diabetic drug and were excluded if first diagnosis-specific code was inconsistent with a T2DM diagnosis. Diagnosis codes were split into 8 categories based on whether they related to specific T2DM or non-specific diabetes codes. Patients were excluded if they had non-specific diagnosis codes and were prescribed insulin as their first-ever treatment for diabetes. Descriptive statistics for time from T2DM diagnosis to insulin initiation were calculated.ResultsTwo hundred and fifty-six codes were identified which were consistent with a first-ever diagnosis of T2DM. 7 codes were considered to clearly define a diagnosis of T2DM, which were reported for 64% of patients. The final cohort comprised 11,917 patients and the median time to first insulin prescription from the date of diagnosis was 4.4 years.ConclusionsA clear definition of cohort development is required to compare and interpret results from studies. Use of diagnosis and product codes is essential when examining use of drugs such as insulin, where competing diagnoses need to be considered separately.

UK phenomics platform for developing and validating electronic health record phenotypes: CALIBER

“…The CPRD uses Read 42 terms (112 806 terms; subset of the International Health Terminology Standards Development Organization SNOMED-CT [Systematized Nomenclature of Medicine Clinical Terms]) 42 to record information. Prescriptions are recorded using Gemscript (a commercial derivative of the NHS Dictionary of Medicines and Devices [dm+d]) 43 (72 664 entries). The CPRD contains >10 billion rows of data from >15 million patients (from all the contributing primary care practices, irrespective of consent to linkage) shown to be representative in terms of age, sex, mortality, and ethnicity 44–46 and of high validity 47 …”

Transforming and evaluating electronic health record disease phenotyping algorithms using the OMOP common data model: a case study in heart failure

“… 29 Prescriptions are recorded using Gemscript (a commercial derivative of the NHS Dictionary of Medicines and Devices [DM+D]) (approximately 67 000 terms). 30 CPRD data have been shown to be representative of age, sex, mortality, and ethnicity and of high diagnostic validity. 22 , 31–33 …”

Transparent Reporting on Research Using Unstructured Electronic Health Record Data to Generate ‘Real World’ Evidence of Comparative Effectiveness and Safety

“…For example, a researcher could use an algorithm that identifies 'diabetic' patients from free-text EHR notes for a comparative evaluation of anti-diabetic drugs. However, without detail on how the phenotype was defined and possible limitations of this definition, it will not be apparent to reviewers or decision makers whether the algorithm is expected to be sensitive or specific, whether it identifies new-onset versus established diabetes mellitus, how the algorithm considers timing (if at all), or how it distinguishes between type 1, type 2, and gestational diabetes [45,46]. Publishing the criteria used by chart reviewers would be an important step toward increasing transparency.…”