Do Real-World Data Sets Accurately Predict Clinical Outcomes? Researchers Investigate

By: Joshua D. Madera, MD
Posted: Monday, August 12, 2024

Because of the problems and inherent biases associated with real-world data sets, a study published in JCO Clinical Cancer Informatics aimed to investigate these issues in patients with breast cancer tested for BRCA1 and BRCA2 variants. Thales C. Nepomuceno, PhD, of the H. Lee Moffitt Cancer Center and Research Institute, Tampa, Florida, and colleagues identified common issues associated with data entry and provided suggestions to reduce the extent of data loss in hopes of maintaining the clinical significance of variants.

The real-world clinical data from 12,243 patients with breast cancer who were tested for BRCA1 and BRCA2 variants were analyzed. All patients were diagnosed with breast cancer between 2000 and 2022. Patient data were subjected to cleaning and harmonization and were subsequently cross-referenced with additional public databases. Functional assays and variable reassessments were also performed on the data.

The study findings revealed a significantly increased number of White and Black patients in the cohort compared with non-White Hispanic and Asian patients. In addition, the largest reason for data loss was identified as missing or incorrect variant designations. This increased number of missing or incorrect variant designations remained, despite the efforts of manual curation to amend the incorrect designations. Furthermore, clinical significance assessments were accurate, regardless of the increased number of nonreported patients with clinical significance. When these nonreported patients were appropriately reassessed, there was a notable improvement in the quality of the data.

“The issue of missing data could be mitigated by more extensive training of personnel entering the data, better integration with other software from which data are derived, recovery of original testing reports, and additional safeguards during manual data entry (eg, pull-down menus instead of free entry),” the authors commented.

Disclosure: Dr. Nepomuceno reported no conflicts of interest. For full disclosures of the other study authors, visit ascopubs.org.

JCO Clinical Cancer Informatics

---