Variable Validity of Computer Extracted Problem Lists for Diabetic Retinopathy and other Co-Morbidities within the Greater Los Angeles Veterans Health Administration

Abstract

Purpose : Electronic health data in the form of International Classification of Disease, Ninth Revision (ICD-9) codes is routinely used for clinical research, yet the accuracy of specific diagnoses is largely unknown. The purpose of this study is to assess the validity of the computer extracted problem list for diabetic retinopathy (DR) and other complications of diabetes mellitus (DM) within the Greater Los Angeles Veterans Health Administration.

Methods : The study population consisted of patients at the VA Greater Los Angeles Health System with an ICD-9 diagnosis of DM between Jan 1^st 1999 and March 22^nd 2016, with baseline and follow-up visits to the eye clinic at least 10 years apart. 50 patients with an ICD-9 diagnosis of DR at the follow-up visit, and 50 patients without a diagnosis of DR were randomly selected from eligible patients. The accuracy of ICD-9 codes for DR, as well as related co-morbidities such as hypertension, hyperlipidemia, coronary artery disease (CAD), and cerebrovascular accident (CVA), were assessed at both time points through manual chart review.

Results : A total of 3193 patients met our inclusion criteria. Of the 50 patients with an ICD-9 diagnosis of DR at the follow-up time point, 15 had an incorrect diagnosis, with a positive predictive value (PPV) of 0.7. For 50 patients without a ICD-9 diagnosis of DR at the follow up time point, 5 had an incorrect diagnosis, with a negative predictive value (NPV) of 0.9. Of the other co-morbid medical conditions, NPV ranged from a low of 63% for obesity to a high of 98% for CVA and CAD (Table 1).

Conclusions : Validity of ICD-9 diagnoses of diabetic complications in this VA population varied considerably, with DR demonstrating moderate agreement, obesity being the most under-documented, and CVA and CAD being the most consistently documented. These discrepancies should be considered when using billing codes for research purposes.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Table 1. Sensitivity, Specificity, PPV, and NPV of ICD-9 codes for DR and associated risk factors. 95% confidence intervals in parentheses.