1067 — Comparing Two Methods to Track Hospitalizations for Hypoglycemia Using Electronic Health Records in Veterans Health Administration Diabetes Patients

Soroka O, VA New Jersey Health Care System; Tseng CL, VA New Jersey Health Care System; Pogach L, VA New Jersey Research Office;

Objectives:
Hospitalizations for hypoglycemia (HYPO-H) associated with uses of insulin/sulfonylurea and have been recognized as public health issues. The Agency for Healthcare Research and Quality Health care Cost and Utilization Project (HCUP) is utilized to estimate national adverse drug event incidence, including hospitalizations caused by insulin and hypoglycemic medications (Weiss, 2013). Lipska (2014) used a validated algorithm (Ginde, 2005) to report trends of HYPO-Hs in national Medicare populations. Our objective was to evaluate uses of HCUP (Weiss, 2013) and Ginde (2005, 2014) coding schemes in estimating the rate of HYPO-H among Veteran Health Administration (VHA) diabetes patients.

Methods:
We used VHA diabetes patients in 2004 and 2011 to evaluate HYPO-H incurred in VHA settings comparing Ginde and HCUP methods. Additionally, we compared the two methods separately for VHA and Medicare settings in 2004 using VHA and Medicare data; Medicare Health Maintenance Organization enrollees were removed because unavailability of health information . Principal ICD-9-CM diagnosis codes were used to determine HYPO-H: (1) HCUP: 962.3, 995.23, 996.57, and E932.3; (2) Ginde: 250.8 (absence of selected co-diagnosis), 251.0, 251.1, 251.3, 970.3, and 962.3. We calculated the rates of HYPO-Hs (number of persons with HYPO-H per 100,000 persons) and percent of change in the rate for each method.

Results:
There were 902,566 and 1,211,616 VHA diabetes patients with diabetes in 2004 and 2011, respectively. There were 1,938/94 HYPO-Hs in 2004/2011 based on Ginde/HCUP. Most HYPO-Hs (95%/91% in 2004/2011) were identified through Ginde only; 5%/9% of HYPO-Hs were identified based on HCUP. Of the HYPO-Hs identified by HCUP, the majority (97%/90% in 2004/2011) had ICD-9-CM code 962.3. The rate for HYPO-H from 2004 to 2011 decreased by Ginde (27%; 1.9 to 1.4) but increased by HCUP (35%; 0.10 to 0.14). Analysis of 871,433 diabetes patients using 2004 VHA-Medicare data showed: 1,748 (95%) of VHA HYPO-Hs and 2,764 (94%) of Medicare HYPO-Hs were identified through Ginde only.

Implications:
Using additional ICD-9-CM codes increased the estimate of the HYPO-H rate 9 to 19 folds in VHA and Medicare national populations.

Impacts:
Surveillance of HYPO-Hs may consider applying a validated algorithm to better estimate magnitude of serious hypoglycemia.