Selection bias due to delayed comprehensive genomic profiling in Japan

Patients with advanced cancer undergo comprehensive genomic profiling in Japan only after treatment options have been exhausted. Patients with a very poor prognosis were not able to undergo profiling tests, resulting in a selection bias called length bias, which makes accurate survival analysis impossible. The actual impact of length bias on the overall survival of patients who have undergone profiling tests is unclear, yet appropriate methods for adjusting for length bias have not been developed. To assess the length bias in overall survival, we established a simulation‐based model for length bias adjustment. This study utilized clinicogenomic data of 8813 patients with advanced cancer who underwent profiling tests at hospitals throughout Japan between June 2019 and April 2022. Length bias was estimated by the conditional Kendall τ statistics and was significantly positive for 13 of the 15 cancer subtypes, suggesting a worse prognosis for patients who underwent profiling tests in early timing. The median overall survival time in colorectal, breast, and pancreatic cancer from the initial survival‐prolonging chemotherapy with adjustment for length bias was 937 (886–991), 1225 (1152–1368), and 585 (553–617) days, respectively (median; 95% credible interval). Adjusting for length bias made it possible to analyze the prognostic relevance of oncogenic mutations and treatments. In total, 12 tumor‐specific oncogenic mutations correlating with poor survival were detected after adjustment. There was no difference in survival between FOLFIRINOX (leucovorin, fluorouracil, irinotecan, and oxaliplatin) or gemcitabine with nab‐paclitaxel‐treated groups as first‐line chemotherapy for pancreatic cancer. Adjusting for length bias is an essential part of utilizing real‐world clinicogenomic data.