BETHESDA, MD—Real-world data holds the promise of enabling a better understanding of how medications work in the type of patients typically seen in clinics and hospitals. In the case of cancer, the patients who require treatment in day-to-day care often have co-morbidities, might be older, frail and not entirely adherent to recommended protocols and are more likely to experience adverse effects that require reduced dosages, treatment interruptions and discontinuations.
Unfortunately, real-world data is a mixed bag. For some patients in some facilities, the data in the electronic record might include everything a researcher desires with sufficient detail to identify treatment courses, issues and outcomes. In many others, however, key data may be lacking, leaving researchers to exclude many patients with incomplete records or to impute values for the missing variables.
In addition, excluding deaths at time zero, as is often done in real-world studies, can distort findings, particularly in terms of survival with aggressive cancers.
Researchers at the Walter Reed National Military Medical Center in Bethesda, MD., analyzed the difference that exclusion, imputed data and death at time zero had on survival analysis in multiple myeloma in a study presented at the 2023 American Society of Clinical Oncology (ASCO) annual meeting in Chicago, June 2-6, 2023.1
Drawing on ASCO’s CancerLinQ Discovery Multiple Myeloma dataset of 34,234 patients diagnosed with multiple myeloma (MM) from 2009 to 2021, the team, led by presenter Karen J. Shou, DO, first analyzed the impact of missing data. Ten percent of the study population lacked a recorded date of diagnosis. For these patients, the date of MM treatment initiation was used as a surrogate for the date of diagnosis as most MM treatment starts at diagnosis. In comparing overall survival (OS) in patients with recorded diagnosis dates versus those with presumed dates of diagnosis, the team found a statistically significant difference, with median overall survival of 115 months in the known group vs. 45 months in the presumed group (HR 2.54, 95% CI 2.41-2.69, p< 0.001).
