BACKGROUND: Analyses of secondary data (e.g., insurance claims) to generate real-world evidence often entail use of algorithms to identify, define, and study key measures (e.g., diagnosis [dx] date; receipt of therapy).
OBJECTIVE: To systematically assess the validity of claims data-based algorithms used to identify multiple myeloma (MM) dx and treatment-related measures.
METHODS: A retrospective, observational study was conducted using claims data from Geisinger Health (GH), an integrated health care delivery system in Pennsylvania. In a cohort of patients with MM from January 2004-November 2016, algorithms were used to identify and define measures related to dx and ensuing therapy. Measures from claims data were adjudicated against a medical record review (MRR) to evaluate the validity of the claims-based algorithms. Validity of claims-based study measures were evaluated by assessing positive predictive values (PPV) and proportions. PPV was calculated as the number of true-positives divided by the sum of true- and false-positives.
RESULTS: Of 352 patients with ≥2 MM dx claims ≥30 days apart, MRR was conducted for 177 patients who met other selection criteria (e.g., ≥12-month pre-index period without MM dx). Most, (68.9%) were ≥65 years old and 54.8% were male. Of the patients reviewed, 131 had a confirmed MM dx, per the MRR, with a PPV of 74.0% (95% CI: 67.6–80.5%). Among these, 84.7% (95% CI: 78.6-90.9%) of patients had an initial MM dx date from claims within 30 days of the initial MM dx date ascertained from the MRR. From the MRR, 89.3% of patients were confirmed to receive first-line (1L) therapy versus 82.4% of patients identified using the claims-based algorithms. The resulting PPV was 94.4% (95% CI: 90.1-98.8%). Based on the claims-based algorithms, 56.5%, 22.2%, and 21.3% of patients received doublet, steroid only, and triplet regimens, respectively. The MRR showed that 63.3%, 26.5%, and 6.8% received doublet, triplet, and steroid only regimens, respectively. The proportion having the same 1L regimens identified from claims and MRR was 66.7% (95% CI: 57.8-75.6%); those with the same bortezomib- or lenalidomide-based 1L regimens was 77.8% (95% CI: 68.2-87.4%). Among patients receiving 1L therapy, 45.4% and 48.7% had evidence of progression from claims and MRR, respectively (resulting PPV: 63.3% [95% CI: 49.8-76.8%]).
CONCLUSIONS: Using GH data, the claims-based algorithms had a PPV/proportion of >70% for identifying true MM dx, receipt of 1L therapy, and 1L bortezomib- or lenalidomide-based regimens. Previous validation studies have reported similar PPVs.
