A Japanese clinical trial (JGOG3016) showed that dose-dense weekly paclitaxel in combination with carboplatin extensively prolonged overall survival (OS) in patients with advanced ovarian cancer. However, in other clinical trials, dose-dense paclitaxel regimens were not superior to triweekly paclitaxel regimens. In this study, causal tree analysis was applied to explore subpopulations with different treatment effects of dose-dense paclitaxel in a data-driven approach. The 587 participants with stage II-IV ovarian cancer in the JGOG3016 trial were used for model development. The primary endpoint was treatment effect in terms of 3-year OS in patients receiving dose-dense vs. conventional paclitaxel therapies. In patients <50 years, the 3-year OS was similar in both groups; however, it was higher in the dose-dense group in patients ≥50 years. Dose-dense paclitaxel showed strong positive treatment effects in patients ≥50 years with stage II/III disease, BMI <23 kg/m2, non-CC/MC, and residual tumor ≥1 cm. In contrast, although there was no significant difference in OS; the 3-year OS rate was 23% lower in dose-dense paclitaxel than conventional paclitaxel in patients ≥60 years with stage IV cancer. Patients in this group had a particularly lower performance status than other groups. Our causal tree analysis suggested that poor prognosis groups represented by residual tumor tissue ≥1 cm benefit from dose-dense paclitaxel, whereas elderly patients with advanced disease and low-performance status are negatively impacted by dose-dense paclitaxel. These subpopulations will be of interest to future validation studies. Personalized treatments based on clinical features are expected to improve advanced ovarian cancer prognosis.
Keywords: carboplatin; heterogeneous treatment effect; machine learning; ovarian cancer; paclitaxel.
© 2024 The Authors. International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC.