Results Summary
PCORI funded the development of PCORnet®, the National Patient-Centered Clinical Research Network, to make research faster, easier, and less costly to conduct. PCORnet is made up of Partner Networks of healthcare systems, patients and communities, and health plans that harness the power of large amounts of health data.
PCORI supports brief, descriptive projects to assess the feasibility of conducting research using data gathered and shared securely through PCORnet. This project is one of several designed to test the network while addressing priorities identified by PCORI and its stakeholders.
What was the project about?
Patients with type 2 diabetes need medicine to control their blood sugar levels. Many types, or classes, of medicines are used to treat type 2 diabetes. Doctors may prescribe one or more classes of medicine for patients.
PCORnet created a shared database system that includes information about patients and the medicine they take to treat type 2 diabetes. The database uses information from patients’ electronic health records, or EHRs. The project team looked at the database to answer four questions:
- Can researchers identify patients with type 2 diabetes using information from EHRs?
- Can researchers use the database to see how often patients are prescribed different medicine classes for type 2 diabetes?
- How much data are missing from the database?
- Can researchers use the database to look for differences in which patients are prescribed which medicine classes?
What were the results?
Can researchers identify patients with type 2 diabetes using information from EHRs? The team found that the new method identified patients with type 2 diabetes 96 percent of the time. The team compared the information in the PCORnet shared database to reviewing patient EHRs by hand at four different medical centers.
Can researchers use the database to see how often patients are prescribed different medicine classes? The team was able to identify more than 613,000 patients with type 2 diabetes and see how many classes of medicine each patient was prescribed, and which combinations of medicine were most common when patients had several prescriptions.
How much data are missing from the database? Some types of data were missing in the database more often than others. For example, 85 percent of patient records had data on blood pressure, but only 40 percent had data on cholesterol levels.
Can researchers use the database to look for differences in which patients are prescribed which medicine classes? The team was able to identify differences among patients who received different types of medicine. For example, patients prescribed metformin were younger on average than patients who received other types of medicine to treat diabetes.
Who was in the project?
The project included data from the PCORnet shared database from 613,203 patients with type 2 diabetes from 2012 through 2017. All patients had a diagnosis of type 2 diabetes or a blood test that showed type 2 diabetes or were taking medicine used to treat type 2 diabetes. Of these patients, 65% were white, 20% were black, 11% were another race, 5% were missing race, and 9% were Hispanic or Latino. The average age was 64 and 50% were female.
What did the project team do?
The project team used database information about type 2 diabetes diagnoses, blood sugar levels, and prescriptions to identify people with type 2 diabetes. The team also looked for patterns in patient and prescription data.
What were the limits of the project?
The database only includes information on medicines that are prescribed. Patients’ actual use of the medicines may differ.
Future projects can see how well prescribing data predicts whether patients actually fill a prescription.
How can people use the results?
Research teams can use the methods in this study to plan future studies using the PCORnet database.
Professional Abstract
PCORI funded the development of PCORnet®, the National Patient-Centered Clinical Research Network, to make research faster, easier, and less costly to conduct. PCORnet is made up of Partner Networks of healthcare systems, patients and communities, and health plans that harness the power of large amounts of health data.
PCORI supports brief, descriptive projects to assess the feasibility of conducting research using data gathered and shared securely through PCORnet. This project is one of several designed to test the network while addressing priorities identified by PCORI and its stakeholders.
Objective
To assess the feasibility of using computable phenotypes to identify patients with type 2 diabetes and to identify patient characteristics, laboratory values, medication prescriptions, and missing data for these patients.
Project Design
| Design Element | Description |
|---|---|
| Design | Retrospective descriptive analysis project |
| Data Sources and Data Sets | EHR data from 2012 to 2017 for adult patients with type 2 diabetes from 33 sites in four PCORnet networks |
| Analytic Approach | Review of EHR data |
| Outcomes | Aim 1: Accuracy of new method to detect patients with type 2 diabetes in PCORnet data at 4 sites Aim 2: Frequency of type 2 diabetes prescriptions, organized by medicine class Aim 3: Missingness of PCORnet data Aim 4: Differences in patient demographics, comorbidities, and other non-diabetes medications by class |
The team had four aims:
- Aim 1. Assess the accuracy of a new algorithm to detect patients with type 2 diabetes using electronic health record (EHR) data.
- Aim 2. Expand type 2 diabetes identification to examine the frequency of prescriptions for different classes of medications used to treat type 2 diabetes.
- Aim 3. Examine how much and what kind of data was missing in the data sets.
- Aim 4. Determine differences in patient demographics, comorbidities, and other non-diabetes medications prescribed among patients who were prescribed different medication classes for type 2 diabetes.
The project team first studied definitions for type 2 diabetes at four data marts, then expanded and reviewed data retrieved through 33 PCORnet data marts. Data marts are collections of data that can be queried and that return outputs from sites that are part of PCORnet-affiliated networks. The first step included validation of electronic health record (EHR) data from four sites and included 1,600 patients. Patients with type 2 diabetes were identified through having a billing code for type 2 diabetes, having laboratory testing consistent with diabetes, or having been prescribed medication to treat type 2 diabetes. The project team then evaluated the prevalence of type 2 diabetes from 2012 through 2017 across 33 sites using the PCORnet common data model (CDM). The CDM organizes data into a standard structure for use by researchers. Of the more than 613,000 patients identified, 65% were white, 20% were black, 11% were another race, 5% were missing race, and 9% were Hispanic or Latino. The average age was 64 and 50% were female.
Results
Aim 1. The team reviewed 1,600 records across the first four data marts and found that the new method had a positive predictive value for detecting the presence of type 2 diabetes of 96.2%.
Aim 2. Among the 613,000 patients identified at the 33 data marts, the project team found that 42% of patients were prescribed no type 2 diabetes medicines, 42% were prescribed a single class of medicine, and 15% were prescribed two or more classes. Among patients prescribed a single class of drug, metformin was the most common (50%), followed by insulin (21%), and sulfonylureas (16%).
Aim 3. Missing clinical and laboratory data varied across the 33 data marts and between patients with different prescribed medication regimens. The clinical variables with the most available data were blood pressure (85% of patients had at least one reported value), followed by body mass index (BMI, 79%), renal function (52%), hemoglobin A1c (54%), and cholesterol (40%).
Aim 4. Patient demographics, comorbidities, and prescribed non-diabetes medications differed among patients prescribed the most common type 2 diabetes medication classes. For example, patients prescribed metformin (alone or in combination with another medicine) had a lower median age (metformin only: 62 years; metformin-sulfonylurea: 61 years; metformin-insulin: 59 years) compared with those receiving insulin, dipeptidyl peptidase-4 inhibitors (DPP4), or sulfonylurea regimens only (64, 65, and 68 years, respectively).
Limitations
The EHR data only track medicines that are prescribed. Patients’ actual use of medicines may differ substantially.
Conclusions and Relevance
The project team was able to use the PCORnet CDM to identify patients with type 2 diabetes and understand prescription patterns for different medicines to treat type 2 diabetes. Other researchers may be able to use PCORnet data for future observational studies.
Future Needs
Future projects can validate prescribing data against prescription fill data to determine patient use of medicine for type 2 diabetes.