Objective: To transform data from an acute hospital in Singapore to the common data model (CDM) developed by the Observational Medical Outcomes Partnership (OMOP) and apply analytical tools on the data to uncover the pharmacological treatment patterns of patients newly diagnosed with diabetes, hypertension or depression.

Method: We converted source data files to the OMOP CDM. We then executed a federated query on the transformed data, which returned the sequences of drug exposure in patients newly diagnosed with diabetes, depression or hypertension. Treatment pathway comparisons were made using sunburst diagrams.  

Results: We converted the data of 250,000 patients who had visited Khoo Teck Puat Hospital, a tertiary care hospital in Singapore from January 2013 and December 2016, into the OMOP CDM. The data contained approximately 1.1 million rows of diagnoses, 5.2 million rows of ordered medications and 15.5 million lab records.

Three cohorts of patients with depression (n=251), hypertension (n=3,175) and diabetes (n=1,006) mellitus with their index prescriptions between 2015 and 2016 were identified. For diabetes, metformin was most often prescribed as the first medication (53.5%), while for hypertension and depression there was a more even spread of medication use, with mirtazapine (37.1%) and fluvoxamine (19.1%) being the more commonly used first medication for depression, and amlodipine (22.1%) being the more commonly used first line medication for hypertension. Sunburst charts of the treatment pathways also reveal the relative frequencies at which subsequent agents are used as second, third or fourth line therapies. 

Sulfonylureas were these most common second line agent used in diabetes. Interestingly, among the agents used, glipizide was the most common (used in 43.3% of all patients), followed by tolbutamide (7.1%) whereas arguably safer options such as gliclazide and glimepiride were used less frequently, at 5.5 and 0.6%, respectively.

Conclusion: The pathways reveal that there is considerable heterogeneity in treatment patterns for hypertension and depression, whereas for diabetes, metformin is the most common first line agent, (used in 53.5% of patients). We show that the same query previously used in a similar study on external data sources can be applied consistently to our dataset, demonstrating the feasibility of use of the CDM. These easily applied queries provide regulators valuable insights on real world drug utilization patterns, adherence to recommended treatment guidelines.

Use of electronic health records to evaluate treatment pathways - a Common Data Model approach