Sylvia Plevritis: Better cancer treatment through data

Episode Summary

Executive Summary: The episode explores how big data and computational modeling are reshaping cancer care, from multi-omic tumor profiling and drug-combination design to screening policy and patient decision tools. Sylvia Plevritis explains how heterogeneous tumors require personalized, data-driven strategies, how simulation can inform screening guidelines and policy, and how online tools can help high-risk patients make informed choices.

Main Topics: Big data and multi-omic cancer profiling (Priority: 5/5): Plevritis describes how longitudinal clinical records and deep molecular profiling—genomics, transcriptomics, epigenomics, proteomics, and post-translational modifications—are transforming cancer research and care. Tumor boards and computational decision-making (Priority: 4/5): Traditional multidisciplinary tumor boards are expanding to include molecular data and computational scientists, improving coordination and enabling more sophisticated interpretation of tumor behavior. Personalized drug combinations for heterogeneous tumors (Priority: 5/5): Her research focuses on using single-cell and signaling data to predict the most effective minimum drug combinations, accounting for cell-type heterogeneity and resistance mechanisms. Screening models and policy evaluation (Priority: 4/5): The conversation covers how simulation models estimate the impact of screening on mortality and help policymakers assess tradeoffs between screening frequency, age, cost, and treatment advances. Personalized screening and false positives (Priority: 4/5): Plevritis emphasizes that screening should be individualized based on genetics, family history, and behavior, and that false-positive burden is highly patient-specific. Patient-facing risk tools for BRCA carriers (Priority: 4/5): She discusses a Stanford decision-support tool for women with BRCA1/BRCA2 mutations that helps compare screening, prophylactic procedures, or watchful waiting.

Key Arguments: Cancer treatment increasingly depends on integrating large-scale molecular and clinical data rather than relying only on traditional pathology or imaging. Tumors are heterogeneous; the same cancer type can contain different cell populations with different drug sensitivities and resistance pathways. Mutation profiles are clinically actionable, but they do not capture all resistance mechanisms; epigenetic and post-translational effects also matter. Computational models can reduce the number of drug-combination hypotheses that need to be tested in patients by prioritizing the most promising options. Screening is a population-level intervention whose value depends on treatment effectiveness, disease incidence, and false-positive burden. Policy decisions about screening should be informed by simulation models that compare counterfactual scenarios, not only by clinical trials. Patients vary in how they experience false positives, so screening policies and counseling should not assume one universal tolerance level. Decision tools can help high-risk patients understand tradeoffs among screening, prophylactic intervention, and delayed action. Clinical translation is underway through proof-of-concept studies and collaborations with institutions testing combinations in real time.

Data Points: Years in consortium: 20 years - Plevritis has been part of the CISNET modeling consortium for two decades. Estimated users of BRCA decision tool: About 45,000 users - Usage of the Stanford BRCA risk-management tool to date. Short-term stimulation window: 30 minutes to 1 hour, sometimes a couple of hours - Duration used in single-cell signaling experiments on cultured hematologic malignancy samples before and after drug exposure. Patient subgroup needing different drug: About 10% - In the pediatric ALL proof-of-concept study, a small subgroup appeared best matched to a different drug than the main two-drug pattern. Main drug pattern in archived samples: About half of patients: two drugs; almost the other half: one of those two drugs - The computational model suggested a dominant combination pattern across archived pediatric ALL samples. Consortium acronym: CISNET - The cancer intervention and screening modeling consortium used for policy-facing simulation work.

Pivotal Quotes: "Cancer sucks." — Russ Altman: Opening framing of the episode, emphasizing the severity and emotional weight of the disease. "It is too rich of a source of inspiration, innovation, and discovery for new things in medicine." — Russ Altman: On why data must be protected but not closed off, especially for medical innovation. "We need to understand that complexity to understand the mechanisms of drug resistance and then to think about how to combat those mechanisms of drug resistance." — Sylvia Plevritis: Explaining why mutational data alone is insufficient for modern cancer treatment.

Implications: Cancer care is moving toward highly personalized, data-driven decisions across treatment and screening. Expect more computational tools in clinics, more nuanced screening guidelines, and greater reliance on molecular profiling to choose therapies and patient-specific risk strategies.