Joint Modelling of Intermediate Longitudinal Biomarkers to Predict Overall Survival in Patients with Solid Tumors
Linsell L1, Paracha N2, Grossman J3, Bokemeyer C4, Garcia-Foncillas J5, Italiano A6, Vassal G7, Chen Y8, Torlinska B9, Abrams K10
1Visible Analytics, Oxford, OXF, UK, 2Bayer Pharmaceuticals, Basel, Switzerland, 3Bayer Pharmaceuticals, Westerville, OH, USA, 4University Medical Centre Hamburg, Eppendorf, Hamburg, Germany, 5University Cancer Institute and the Department of Oncology, Madrid, Spain, 6Institut Bergonié Comprehensive Cancer Centre, Bordeaux, France, 7Gustave Roussy Comprehensive Cancer Centre, Villejuif, France, 8Visible Analytics, Oxford, UK, 9Visible Analytics, Oxford, Oxfordshire, UK, 10Warwick University, Coventry, WAR, UK
OBJECTIVES: Joint modelling (JM) of longitudinal and time-to-event data simultaneously can provide an estimated biomarker profile (adjusted for informative dropout due to death) or predictions of the time-to-event outcome conditional upon the longitudinal biomarker profile. This application explores the association between tumor burden and overall survival (OS) in patients with solid tumors, compared to a traditional parametric approach.
METHODS: Data were pooled from three phase I/II open-label trials evaluating the safety and efficacy of Larotrectinib in adults and pediatric patients with TRK fusion cancer (data-cut 20/07/2021; n=196). Tumor burden was measured as the sum of diameters of target lesions (SLD). Bayesian joint modelling was used to obtain patient-specific predictions of OS using individual-level SLD profiles up to the time at which the patient died or was censored, using alternative assumptions for the association parameter. These were compared to predictions from a standard Weibull model. All models were adjusted for age, ECOG status >1, metastatic progression at treatment initiation and tumor type.
RESULTS: Median follow-up was 32.4 months (range 0.4 to 71) and 58/196 deaths (29.6%) were observed. The JM using a common association parameter across primary tumor type was the best fitting model. The restricted mean survival time was 8.9 years (95% credible interval (CrI): 6.3 to 11.5) compared to 8.0 years (95% CrI: 5.0 to 13.3) in the Weibull model. 10-year OS predictions from treatment initiation were also similar, with less uncertainty in the JM; 26.4% (95% Crl: 18.0% to 34.8%) compared to 27.7% (95% Crl: 12.8% to 40.4%) in the Weibull model.
CONCLUSIONS: JM can offer an alternative approach to traditional survival modelling and may improve survival predictions from limited follow-up data. This approach allows complex hierarchical data structures, such as patients nested within tumor types. It can also incorporate multiple longitudinal biomarkers in a multivariate modelling framework.