Daniel Glazar1, Heather Jim1, Matthew Schabath1, Aasha Hoogland1, Renee Brady-Nicholls1
1Moffitt Cancer Center & Research Insitute
Background: Patient-reported outcomes (PROs) are defined by the FDA as “any report of the status of a patient’s health condition that comes directly from the patient, without interpretation of the patient’s response by a clinician or anyone else.” [1] While PROs have traditionally been assessed in oncology in the context of quality of life, intriguing recent data suggests that they may be early indicators of clinically important events. For example, two studies reported that some undiagnosed lung and pancreatic cancer cases can be identified early through symptoms entered as web search terms. [2,3] Moreover, a recent series of preliminary studies suggests that patient-reported symptoms may be sensitive early indicators of cancer progression. [4-6] A small randomized trial based on these data successfully used PROs to detect lung cancer progression five weeks earlier than usual care, extending median survival by seven months. [4] These preliminary findings provide a compelling rationale for further investigation into whether PROs are reliable early indicators of cancer progression. In addition, PROs are promising investigative biomarkers for cancer response and progression due to their being non-invasive and low-cost as well as their ease and frequency to be administered to patients, in contrast to other, more standard biomarkers, such as biopsies, blood draws, and imaging studies. In this study, we seek to elucidate the relationship between tumor size (TS) and PRO dynamics and leverage this information to infer patient-specific tumor growth inhibition (TGI) parameters, including initial TS, growth rate, maximum tumor kill rate, and rate of evolution of resistance. Objectives: • Develop a joint model describing longitudinal tumor size (TS) and patient-reported outcomes (PRO) • Test the hypothesis that PRO data can infer patient-specific TGI parameters Methods: We developed a joint model describing longitudinal tumor size (TS) using the Claret tumor growth inhibition (TGI) model with mixed effects [7] and PROs using a time-continuous non-homogeneous Hidden Markov model (HMM) [8]. A link between the two data types was made by including TS as a time-dependent covariate in the transition rate matrix of the HMM. After setting some nominal population parameters, we then simulated 10 in silico patients by sampling individual parameters, TS every 28 days, and a single PRO item every 7 days. To test how well the simulated PROs inform TGI parameters, we performed Bayesian inference by sampling from: 1) a prior distribution (no data as input); 2) a posterior distribution considering only simulated TS data; 3) a posterior distribution considering only simulated PRO data; and 4) a posterior considering both TS and PRO data. Sampling was performed using the Hamilton Markov Chain with No-U-Turn sampling (HMC-NUTS) implemented in Stan. [9] Results: Just by considering PRO data, precision (1 / standard deviation) of patient-specific parameters dramatically increased 1.9-fold (1.4–2.6) across all TGI model parameters relative to the prior distribution. As expected, considering TS data alone improved precision even more than PRO data alone with a 2.8-fold (1.2–6.2) increase relative to the prior. Considering both data types in tandem further improved precision only marginally by 1.1-fold (0.7–1.8) relative to TS alone. By contrast, accuracy (1 / root mean squared error) marginally improved across all TGI model parameters by a 1.4-fold (0.4–4.5) increase relative to the prior distribution. TS alone led to an expected larger 2-fold (0.6–6.4) improvement relative to the prior. Finally, considering both data types in tandem further improved accuracy 1.27-fold (0.5–3.5). Conclusions: These results suggest that PROs have the promise to be leveraged as a minimally invasive and inexpensive biomarker to inform patient-specific TGI parameters. Future research directions include exploring the effect of sampling frequency, number of patients, number of PROs, and model misspecification, as well as application on a clinical dataset of 63 NSCLC patients treated with immune checkpoint inhibitors.
[1] U. S. Department of Health Human Services, F. D. A. Center for Drug Evaluation Research, U. S. Department of Health Human Services, F. D. A. Center for Biologics Evaluation Research, U. S. Department of Health Human Services, F. D. A. Center for Devices Radiological H. Guidance for industry: patient-reported outcome measures: use in medical product development to support labeling claims: draft guidance. Health Qual Life Outcomes. 2006;4:79. [2] Paparrizos J, White RW, Horvitz E. Screening for Pancreatic Adenocarcinoma Using Signals From Web Search Logs: Feasibility Study and Results. J Oncol Pract. 2016. [3] White RW, Horvitz E. Evaluation of the Feasibility of Screening Patients for Early Signs of Lung Carcinoma in Web Search Logs. JAMA Oncol. 2016. [4] Denis F, Lethrosne C, Pourel N, et al. Overall survival in patients with lung cancer using a web-application-guided follow-up compared to standard modalities: Results of phase III randomized trial. J Clin Oncol 2016;34:abstr LBA9006. [5] Denis F, Viger L, Charron A, Voog E, Letellier C. Detecting lung cancer relapse using self-evaluation forms weekly filled at home: the sentinel follow-up. Support Care Cancer. 2014;22(1):79-85. [6] Denis F, Yossi S, Septans AL, et al. Improving Survival in Patients Treated for a Lung Cancer Using Self-Evaluated Symptoms Reported Through a Web Application. Am J Clin Oncol. 2015. [7] Laurent Claret et al. Model-Based Prediction of Phase III Overall Survival in Colorectal Cancer on the Basis of Phase II Tumor Dynamics. JCO 27, 4103-4108(2009). https://doi.org/10.1200/JCO.2008.21.0807 [8] Kendall, E. B., Williams, J. P., Hermansen, G. H., Bois, F., & Thanh, V. H. (2024). Beyond Time-Homogeneity for Continuous-Time Multistate Markov Models. Journal of Computational and Graphical Statistics, 1–15. https://doi.org/10.1080/10618600.2024.2388609 [9] Stan Development Team. 2024. Stan Reference Manual, 2.36.0. https://mc-stan.org
Reference: PAGE 33 (2025) Abstr 11446 [www.page-meeting.org/?abstract=11446]
Poster: Methodology - New Modelling Approaches