A priori, the authors decided to use as potential algorithm components the number of counts of the SLE ICD-9 code (710.0), keyword of “lupus” in the problem list, positive ANA, and ever use of medications frequently used in SLE such as antimalarials, systemic corticosteroids, and disease-modifying antirheumatic drugs (DMARDs) (see below). These components were selected based upon SLE disease criteria and management combined with data accessible in the EHR. Subjects who were not classified as true SLE cases frequently had other autoimmune diseases with ICD-9 codes for these diseases. Multiple true SLE cases had an overlap syndrome (e.g. secondary Sjögren's syndrome or RA). We examined the exclusion of individuals with ICD-9 codes for other autoimmune diseases to assess whether potential SLE subjects who were not true SLE cases were appropriately excluded and also to ensure true SLE cases with overlap syndromes were not excluded. These exclusion criteria were selected during chart review of the training set.
Positive predictive values (PPVs) and sensitivity were calculated for every combination of ≥ 1, 2, 3, and 4 counts of the SLE ICD-9 code; a positive ANA, (titer ≥ 1:40 and titer ≥ 1:160); ever use of antimalarials, systemic corticosteroids, and disease-modifying antirheumatic drugs (DMARDs); and a keyword of “lupus” in the problem list using “and” or “or” between the criteria. PPVs were also calculated excluding ICD-9 codes for systemic sclerosis (SSc) (710.1) and dermatomyositis (DM) (710.3). All algorithms included individuals with at least one count of the SLE ICD-9 code. The PPV was calculated as the number of subjects who fit the algorithm and were confirmed cases on chart review divided by the total number of subjects who fit the algorithm. Sensitivity was calculated as the number of subjects who fit the algorithm and were confirmed cases on chart review divided by total number of confirmed cases. To “fit the algorithm,” the subject had to have available data for that particular algorithm's criteria. If an ANA was not checked at Vanderbilt, it was considered missing. The F-score, which is the harmonic mean of the PPV and sensitivity ([2 × PPV × sensitivity] / [PPV + sensitivity]), was calculated for all algorithms. The F-score is commonly used in informatics because it provides a single number to compare algorithms accounting for both PPV and sensitivity.
Antimalarials included in the medication search were “hydroxychloroquine,” “plaquenil,” “chloroquine,” “quinacrine,” and “aralen.” Oral and intravenous corticosteroids included were “cortisone acetate,” “hydrocortisone,” “Cortef,” “prednisone,” “dexamethasone,” “dexamethasone Intensol,” “decadron,” “prednisolone sodium phosphate,” “Pediapred,” “prednisone Intensol,” “methylprednisolone,” “Medrol,” “Medrol Dosepak,” “prednisolone,” “Orapred,” and “Prelone.” DMARDs included were “azathioprine,” “Imuran,” “methotrexate sodium,” “methotrexate,” “Trexall,” “mycophenolate mofetil,” “CellCept,” “mycophenolic acid,” “Myfortic,” “cyclophosphamide,” “Cytoxan,” “rituximab,” “Rituxan,” “etanercept,” “Enbrel,” “Enbrel Sureclick,” “adalimumab,” “Humira,” “Humira Pen,” “infliximab,” “Remicade,” “abatacept,” and “Orencia.”
Positive predictive values (PPVs) and sensitivity were calculated for every combination of ≥ 1, 2, 3, and 4 counts of the SLE ICD-9 code; a positive ANA, (titer ≥ 1:40 and titer ≥ 1:160); ever use of antimalarials, systemic corticosteroids, and disease-modifying antirheumatic drugs (DMARDs); and a keyword of “lupus” in the problem list using “and” or “or” between the criteria. PPVs were also calculated excluding ICD-9 codes for systemic sclerosis (SSc) (710.1) and dermatomyositis (DM) (710.3). All algorithms included individuals with at least one count of the SLE ICD-9 code. The PPV was calculated as the number of subjects who fit the algorithm and were confirmed cases on chart review divided by the total number of subjects who fit the algorithm. Sensitivity was calculated as the number of subjects who fit the algorithm and were confirmed cases on chart review divided by total number of confirmed cases. To “fit the algorithm,” the subject had to have available data for that particular algorithm's criteria. If an ANA was not checked at Vanderbilt, it was considered missing. The F-score, which is the harmonic mean of the PPV and sensitivity ([2 × PPV × sensitivity] / [PPV + sensitivity]), was calculated for all algorithms. The F-score is commonly used in informatics because it provides a single number to compare algorithms accounting for both PPV and sensitivity.
Antimalarials included in the medication search were “hydroxychloroquine,” “plaquenil,” “chloroquine,” “quinacrine,” and “aralen.” Oral and intravenous corticosteroids included were “cortisone acetate,” “hydrocortisone,” “Cortef,” “prednisone,” “dexamethasone,” “dexamethasone Intensol,” “decadron,” “prednisolone sodium phosphate,” “Pediapred,” “prednisone Intensol,” “methylprednisolone,” “Medrol,” “Medrol Dosepak,” “prednisolone,” “Orapred,” and “Prelone.” DMARDs included were “azathioprine,” “Imuran,” “methotrexate sodium,” “methotrexate,” “Trexall,” “mycophenolate mofetil,” “CellCept,” “mycophenolic acid,” “Myfortic,” “cyclophosphamide,” “Cytoxan,” “rituximab,” “Rituxan,” “etanercept,” “Enbrel,” “Enbrel Sureclick,” “adalimumab,” “Humira,” “Humira Pen,” “infliximab,” “Remicade,” “abatacept,” and “Orencia.”