Improving Phenotyping of Patients’ Diabetes Status from Electronic Dental Record

Objectives: In retrospective studies using Electronic Dental Record (EDR) data, patient-reported diabetes are in general, used to determine patient’s Type 2 diabetes status. However, patients’ diabetes information could be present in different sections of the EDR. For example, a patient may not report having diabetes in medical history (MH), but, may report taking diabetes medications in medication history (MED). Therefore, the objective of this study was to determine improvement in phenotyping diabetes status using MH and MED sections of the EDR.
Methods: First, we retrieved 25, 308 unique patients’ MH and MED who received comprehensive oral evaluation between January 1^st 2009 to December 31^st 2013 at the Indiana University School of Dentistry. Next, we developed a reference standard to classify patients’ diabetes status into “YES”, “NO” or “unknown” based on MH and MED. Using this reference standard, we developed named entity recognition programs to automatically classify patients’ diabetes status from MH and MED sections. We then measured completeness of diabetes information present only in one section (MH or MED) compared to both sections.
Results: We found 1,424 positive diabetes cases documented only in MH, 307 documented only in MED, and 828 documented in both sections. Similarly, we found 34 negative cases (“No”) recorded in MH, 18, 056 recorded in MED, and 224 recorded in both sections. We observed MH had 16,959, and MED had only 54 missing values (“Unknowns”). When phenotyping using the two sections, completeness increased by 5% (1,424 to 2,559) for positive cases, 81% (34 to 16,901) for negative cases and unknowns decreased by 74%.
Conclusions: Study results showed that when MH and MED sections were combined the completeness improved significantly compared to one section (MH or MED). We recommend considering at least medical and medication histories sections of the EDR when phenotyping patients’ diabetes status for retrospective studies.