(143) Assessment of the Application of a 90-Day-Gap Rule for Mitigating Frontline Treatment Sequence Misclassification in Electronic Health Records in the United States

Background: Defining first-line therapy (1L) using electronic health record (EHR) data is not straightforward, as treatment started outside the EHR network may not transfer into the structured data in the record. A 90-day-gap rule (90DGR) was developed as a method to identify patients who may be missing treatment data before the start of 1L. The 90DGR excludes patients with no structured EHR activity within 90 days following clinical eligibility date (e.g., date of advanced diagnosis).

Objectives: The objective was to evaluate the 90DGR’s ability to flag patients with treatment misclassification due to missing 1L treatment information before the start of structured activity.

Methods: Using de-identified data from the Flatiron Health Research Database, 14 cohorts of roughly 275 randomly-sampled patients with a derived 1L regimen were created in multiple myeloma (MM), chronic lymphocytic leukemia (CLL), advanced (adv) non-small cell lung (advNSCLC), small cell lung (SCLC), metastatic (met) prostate, adv melanoma, met breast, adv urothelial, met renal cell carcinoma, ovarian, adv gastric (advGastric), adv head and neck, hepatocellular carcinoma (HCC), and adv endometrial (advEndo) cancers.

90DGR performance was evaluated by first assessing whether a patient’s true 1L line of therapy (LOT; based on extensive review of structured/unstructured data in the chart) matched the 1L regimen derived using Flatiron’s oncology-defined LOT algorithm, and cross-checking this against those patients flagged by the 90DGR. Sensitivity (SE), specificity (SP), positive predictive value (PPV), negative predictive value (NPV), and overall accuracy (OA) of the 90DGR were examined, as was patient attrition from and performance of varying gap lengths (15–180 days) as alternatives to 90 days.

Results: Except for advEndo (88.8%) and CLL (82.6%), ≥90% of the patients in each dataset evaluated had a correctly identified 1L (range 90.9–97.8%) without application of the 90DGR. SE ranged from 50.0% (advGastric) to 92.0% (MM), while SP ranged from 65.8% (CLL) to 98.9% (SCLC). PPV ranged from 8.1% (HCC) to 57.1% (SCLC), and NPV ranged from 95.2% (advEndo) to 99.2% (advNSCLC). OA was lowest for CLL (69.2%) and highest for SCLC (97.8%).

Extending the gap length > 90 days did not elicit a meaningful impact on OA, and gap length < 90 days led to notable increase in attrition.

Conclusions: Removing the 90DGR as an exclusion criterion can increase the number of patients available for analysis. While doing so may increase the prevalence of patients with LOT sequence misclassification, the prevalence is low and unlikely to introduce substantial misclassification bias. In light of these findings, the 90DGR is no longer recommended as an analytic tool for identifying patients with misclassified frontline treatment.