Researchers have developed and validated a new algorithm to use with electronic health records (EHRs) that can accurately identify patients with systemic lupus erythematosus (SLE; Barnado A, et al. Arthritis Care Res [Hoboken]. 2017;69:687-693). Distinct from previous methods of identifying patients with SLE, the new algorithm is the first to incorporate laboratory values and medications with the SLE International Classification of Diseases, Ninth Revision (ICD-9) codes.
“Previous studies mainly used 1 or 2 instances of the SLE ICD-9 code to identify SLE patients; however, our study demonstrated that this method did not identify SLE patients accurately in a general population,” stated lead author of the validation study, April Barnado, MD, MSCI, Research Director, Vanderbilt Lupus Center, and Instructor in Medicine, Division of Rheumatology & Immunology, Vanderbilt University Medical Center, Nashville, TN, in an interview with Value-Based Care in Rheumatology.
EHRs have gained firm footing in the management of patients with medical conditions, facilitating research on these conditions and ensuring that physicians who treat a particular patient have all the relevant medical information needed to make good decisions. For less common diseases, such as SLE, EHRs can be an efficient and cost-effective way to study many patients from different settings.
However, identifying patients with SLE has been challenging. Many studies have relied on SLE ICD-9 codes, but this method is not well-validated. Of note, a recent study found a positive predictive value (PPV) of 50% to 60% with the use of ICD-9 codes in general populations.
“We developed and validated multiple algorithms that incorporate ICD-9 billing codes, laboratories, and medications to identify SLE patients accurately in the EHR, with PPVs ranging from 89% to 95%,” Dr Barnado explained.
“These algorithms are helpful for researchers and clinicians who are trying to identify SLE patients from a large pool of records in the EHR or a database to study outcomes in SLE or even perform a quality improvement project. Since different algorithms incorporate different types of data, such as medications or labs, researchers and clinicians can select which algorithms would be easiest to use in their EHR or database,” stated Dr Barnado, when asked how the new algorithm should be incorporated into clinical practice.
The algorithm development and validation study relied on an EHR database with approximately 2.5 million patients. One or more ICD-9 codes for SLE were used to identify 5959 potential cases of SLE. Two hundred of the potential cases were selected for chart review as a training set. A subject was deemed a case if their SLE diagnosis was made by a rheumatologist, nephrologist, or dermatologist.
Of the 200 potential cases, 90 were defined as “true” SLE by chart review. Of the remaining 110 subjects, 76 were defined as not having SLE (19 had an unconfirmed diagnosis of SLE, and 15 had missing clinical data). The 19 unconfirmed cases were pooled with the 76 subjects not classified as having SLE, for a total of 95 subjects deemed as not having SLE. The 15 subjects with missing data were excluded from the analysis.
Of the 90 confirmed SLE cases, 7 had a secondary or overlap autoimmune disease in addition to SLE, and these patients all had ICD-9 codes for SLE and the second autoimmune diseases (eg, Sjögren’s syndrome, Behçet’s syndrome, or rheumatoid arthritis).
The SLE cases and non-SLE cases were mostly among women. Compared with non-SLE cases, patients with SLE were significantly younger (aged 61 years vs aged 53 years, respectively; P = .001), and less likely to be white (82% vs 68%, respectively; P = .04).
The investigators determined the PPV and sensitivity of code counts of the SLE ICD-9 code, a positive antinuclear antibody (ANA), and ever use of medications (eg, disease-modifying antirheumatic drugs [DMARD], steroids, antimalarial drugs). Three algorithms were found to have the highest PPV and sensitivity, and each of these was validated on 100 randomly selected subjects who were not part of the training set.
The 3 highest performing algorithms were:
- Three or more SLE ICD-9 codes plus positive ANA of ≥1:40 plus ever DMARD and ever steroid use (PPV 95% in the training set, 91% in the validation set)
- Four or more SLE ICD-9 codes plus positive ANA of ≥1:160 (PPV 89% in the training set, 94% in the validation set)
- Three or more SLE ICD-9 codes plus ever antimalarial use (PPV 91% in the training set, 88% in the validation set).
The investigators confirmed that incorporating pertinent medications and a positive ANA with the SLE ICD-9 code increased the PPV of the EHR algorithms.
“These 3 algorithms had high PPVs. Since 1 algorithm incorporates ANA positivity and medications with the SLE ICD-9 code, another uses only the ANA and the SLE ICD-9 code, and the third uses medications and the SLE ICD-9 code, investigators can select which algorithm is best suited to their EHR or administrative database,” Dr Barnado said.
“These algorithms represent powerful tools for clinical and translational researchers to identify and study patients with SLE efficiently and accurately,” she concluded.