Abstract
Purpose :
To understand if using ICD-9, CPT and medication codes from billing databases are an accurate reflection of the charted exam, testing and procedures.
Methods :
Diabetic retinopathy (ICD-9 362.01-362.07) patients were selected randomly from 5 retina specialists across 3 practices from 2011-2015. The associated ICD-9, CPT and medication codes were obtained for the 75 patients across all their visits during a 2-year time interval. Billing data and a chart review gathered the demographic and clinic data from each visit including exam findings, procedures and ancillary testing performed. The billing and chart data was compared for 13 predetermined categories for procedures, medications and ancillary testing. The clinical chart was considered the ground truth. Positive and negative predictive values for billing data to predict the chart data were the main outcomes.
Results :
Seventy-five patients were included representing 345 visits. Eight categories of codes had enough data for analysis. Of the 91 encounters billed for intra-vitreal injection (IVI), 88 had clinical charts reflecting the procedure performed with 96.7% PPV and 96.3% NPV. The 50 billed encounters for IVI ranibizumab had a 85.7% sensitivity, 94.5% specificity, 67.9% PPV and 98% NPV while the 32 billed encounters for IVI bevacizumab had 92% sensitivity, 96.3% specificity, 65.7% PPV and 99.4% NPV (aflibercept did not meet the threshold for analysis). Focal laser (FL) and pan-retinal photocoagulation (PRP) each had 91.6% sensitivity, 100% specificity and 100% PPV and 99.7% and 98.7% NPV respectively. Fundus photography billed events had 78% sensitivity, 99.4% specificity, 84.6% PPV and 99.1% NPV. Fluorescein angiography had 88.9% sensitivity, 99.4% specificity, 92.3% PPV and 99.1% NPV. The optical coherence tomography code 92134 used from 2011 onward had 89.9% sensitivity, 93.9% specificity, 91.8% PPV and 92.5% NPV.
Conclusions :
Administrative billing data can accurately identify procedures performed in the care of patients with diabetic retinopathy. The PPV and NPV for all the procedures reported could be considered valid for use in research studies on administrative databases. Using 5 retina physicians from 3 local practices suggests a high likelihood of generalizability of using large billing databases to provide a new approach to study diabetic retinopathy management and outcomes.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.