Session Type: Abstract Submissions (ACR)
Background/Purpose: The electronic medical record (EMR) is increasingly used as a primary source of retrospective research data from large ‘virtual cohorts’ of patients. The accuracy of EMR data is not guaranteed. Importantly, when performing case-control studies, the cases and controls may be imperfectly defined by EMR data, particularly for syndromic conditions such as SLE. This study was undertaken to ascertain the accuracy of EMR diagnoses of SLE made by specialists at an academic medical center.
Methods: Cases of specialist-diagnosed SLE were defined as individuals who had billing diagnoses using the ICD-9 CM code 710.0 applied by a faculty rheumatologist, nephrologist, or dermatologist two or more times in a twelve month period. These patients’ EMR were reviewed by a single rheumatologist trained in SLE diagnosis and 10% of these charts were independently reviewed by a second rheumatologist for verification. Each patient’s EMR was categorized for its support for the diagnosis of SLE using both the 1997 ACR and 2012 SLICC criteria. “Certain” diagnoses had > 4 criteria found in the chart; “Likely” diagnoses had < 4 criteria but had serology findings specific for SLE (anti-DNA or anti-Sm) and 1 typical clinical feature (e.g., malar rash, arthritis, cytopenia); “Possible” diagnoses had > 1 criterion suggesting SLE and no better explanation given in the chart; “Unlikely” diagnoses included a positive ANA and no other findings, or a single clinical finding with temporally inconsistent serology; “Not SLE” diagnoses were documented if another diagnosis for their condition was firmly established.
Results: From a database of almost 4 million patient records, a total of 139 outpatient charts meeting the definition of specialist-diagnosed SLE were identified. 99 (71.2%) were felt to have Certain lupus; 19 (13.7%) were Likely; 14 (10%) were Possible; 4 (2.8%) were Unlikely; 3 (2.1%) were Not SLE. Combining the Certain patients (who meet classification criteria) and the Likely patients who might have ‘early’ or ‘incomplete’ lupus, or may only lack proper documentation for SLE, the positive predictive value of repeated ICD-9 710.0 codes is approximately 85%. Conversely, about 15% of patients who carry diagnoses of SLE lack sufficient, easily accessible data in their EMR to support that finding. SLE features that had significantly different prevalence between the Certain/Likely and Possible/Unlikely/Not groups were both lupus-specific: anti-DNA (55.9% vs 19.0%, p=0.002), anti-Sm (30.5% vs 0%, p=0.002), and nephritis (51.7% vs 0%, p<0.0001), as well as non-specific: hypocomplementemia (57.6% vs 9.5%, p<0.0001) and lymphopenia (71.2% vs 28.6%, p=0.0003).
Conclusion: This study underscores the challenges of using ICD-9 billing codes to identify research cohorts from EMR data. Even specialist physicians persistently label patients incorrectly while their notes and data do not support the lupus diagnosis. Surprisingly, lymphopenia, an inexpensive and widely obtained laboratory parameter, differentiates patients who truly have lupus from patients incorrectly labeled with the disease, and can be used to develop “decision support” features in the EMR to assist with the diagnosis of SLE.
N. J. Olsen,
D. R. Karp,
« Back to 2013 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/the-accuracy-of-the-icd-9-code-710-0-to-identify-a-cohort-of-sle-patients-from-the-electronic-medical-record/