Development of a Clinical Screening Tool for Undiagnosed Diabetes in a Hospital Observation Unit
Background: While diabetes mellitus (DM) is typically diagnosed in the outpatient setting, 30% of patients in the U.S.A remain undiagnosed, prompting novel approaches to screening. There are 2 million Observation Unit (OU) stays annually in the U.S.A. The OU allows prolonged care for Emergency Depar...
Saved in:
| Published in: | Diabetes (New York, N.Y.) Vol. 67; no. Supplement_1 |
|---|---|
| Main Authors: | , , , , , , , , , , |
| Format: | Journal Article |
| Language: | English |
| Published: |
01-07-2018
|
| Online Access: | Get full text |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background: While diabetes mellitus (DM) is typically diagnosed in the outpatient setting, 30% of patients in the U.S.A remain undiagnosed, prompting novel approaches to screening. There are 2 million Observation Unit (OU) stays annually in the U.S.A. The OU allows prolonged care for Emergency Department (ED) patients who are not sick enough to be hospitalized. A previous study found 9% of OU patients have undiagnosed DM and therefore our OU practice is to obtain HbA1c on patients without a history of DM. A typical OU stay is 24 hours, allowing adequate time for screening, education and initiation of treatment if warranted. To conserve resources, we developed a screening tool to allow selective HbA1c testing in the OU.
Methods: We reviewed OU admissions at a tertiary hospital and included patients with no known history of DM and who had HbA1c assessed during their OU stay. Those with HbA1c ≥ 6.5 were categorized as newly diagnosed DM. Previously validated clinical variables were tested in the model; also included were clinical variables readily available in the OU. A model was developed on 2/3 of the cohort and validated on the remaining 1/3.
Results: A total of 2425 patients met inclusion criteria. The final model included: BMI, race, family history of DM, age, serum sodium and potassium, glucose on ED arrival and initial systolic blood pressure. Predictive assessment of the model was strong with a C-statistic of 0.823. The model was validated and findings remained strong with a C-statistic of 0.788. Performance at the optimal cut-point yielded a sensitivity of 0.750 and specificity of 0.771.
Conclusions: This screening tool identifies 75% of previously undiagnosed DM patients, while excluding 77% of nondiabetic patients from testing. The model can be used at the bedside or embedded in the electronic medical record to selectively screen for DM. There is potential to identify large numbers of previously undiagnosed DM in the OU and the typical 24-hour stay allows time for education and treatment to be initiated.
Disclosure
R.M. Noonan: None. W.J. Castillo: None. G.M. Saffran: None. C.B. Montgomery: None. G. Phayal: None. R.H. Seemangal: None. N. Kohn: None. A. Tiberio: None. E. Wolff: None. R. Schulman: None. R. Silverman: None. |
|---|---|
| ISSN: | 0012-1797 1939-327X |
| DOI: | 10.2337/db18-1323-P |