Modeling and Analytics for Cancer Diagnostics

This Research Program details a sequence of projects for two technologies that are generating intense current interest with wide-ranging practice implications and serious evidence gaps:

  • Multi-cancer early detection testing, and PSMA-PET/CT for newly diagnosed and recurrent prostate cancer. The MCED work will deepen our understanding of performance characteristics, provide guidance regarding a defensible test confirmation strategy, project benefits and harms of different MCED strategies and offer new ideas for shortcutting the typically lengthy process of cancer screening trials.
  • The PSMA-PET/CT work will develop an approach for updating treatment benefit estimated derived from trials that included a mixture of patients with unknown PSMA status and will project lives saved of treatment reallocation on the basis of PSMA-PET.CT result.

The tools and processes developed for modeling these technologies will be applicable to other new diagnostics that emerge during the lifetime of the Research Program. The modeling work will be accompanied by a sequence of real-world analytics projects to assess dissemination of and disparities in uptake of novel diagnostics and their consequences for healthcare utilization and costs.

Modeling and Outcomes Research in Prostate Cancer

Our team’s work focuses on the modeling of prostate cancer progression, detection, and outcomes. We use a combination of mathematical, statistical, and computer simulation modeling. Our models were originally designed to interrogate patterns of prostate cancer incidence and mortality in the US population. Prostate screening was adopted in the population long before results of randomized screening trials became available. Thus, the population has served as an “uncontrolled experiment” offering information on prostate cancer natural history as well as the potential benefits and harms of screening. Modeling provides a sophisticated and coherent approach for unlocking this information and making reasonable inferences from population data.

We have used our models to make inferences about overdiagnosis due to prostate cancer screening and study the role of screening versus treatment in explaining declines in prostate cancer mortality. More recently we have used our models to study the comparative benefits and harms of many competing PSA screening policies. The goal of this work is to provide quantitative information on the tradeoffs of competing policies to screening policy panels as they deliberate their recommendations regarding PSA screening. We have carefully studied the evidence used by the Task Force in developing their recent D recommendation against PSA screening and have suggested, using results from modeling studies, that the Task Force decision was based on evidence that underestimated benefit and overestimated harms of screening.

In our recent articles, our national presentations, and our ongoing work with national policy panels we are actively involved in the great policy debate about prostate cancer screening. New initiatives include using modeling to study the comparative effectiveness of active surveillance and to identify optimal surveillance policies, and to evaluate the harm-benefit tradeoffs of early treatment of recurrent disease.

Active Projects

Modeling and Analytics for Novel Cancer Diagnostics: Traversing the Data-Evidence Divide

The field of cancer diagnostics is in a rapidly expanding growth phase that goes hand in glove with the precision medicine revolution. However, the rapid pace at which new technologies are entering the marketplace makes rigorous evaluation via the standard clinical trials-based pipeline infeasible for all but a relative few. This means that while we typically have some data about diagnostic test performance, we frequently lack evidence regarding the outcomes that drive clinical and policy decisions. Two technologies in particular are generating intense current interest with wide-ranging practice implications and serious evidence gaps: Multi-cancer early detection (MCED) testing, and advanced imaging tests such as PSMA-PET/CT for newly diagnosed and recurrent prostate cancer. The Etzioni Lab is tackling this data evidence divide using the tools of modeling and analytics, with the goal of improving our understanding of how these technologies can be used wisely and equitably to improve care for all cancer patients.

This work is funded by the National Cancer Institute (R35 grant to Dr. Etzioni).

CISNET Prostate

The Cancer Intervention and Surveillance Modeling Network (CISNET) was initiated in 2000 to explain the drivers of trends in cancer incidence and mortality in the United States. At this time, prostate cancer mortality had been declining for a number of years after peaking in the early 1990s. The defining goal of CISNET Prostate was to quantify the roles of screening and changes in primary treatment in driving these mortality declines. The Prostate CISNET group consists of three modeling groups, at Fred Hutch Cancer Center, Erasmus Medical Center, and the University of Michigan. Our overarching goal remains to determine the population impact of changing strategies for prostate cancer control, by linking trends in disease incidence and mortality with trends in screening and treatment. Our methods combine simulation models and maximum likelihood analysis to shed light on two of the most active controversies in prostate cancer research: the value of PSA screening versus advances in prostate cancer treatment, and the link between disparities in care and differences in prostate cancer outcomes between Black and White men in the population. In addition to being a prostate modeling site (with modeling led by Roman Gulati), the  Fred Hutch also serves as the co-ordinating center for CISNET Prostate.

Pacific Northwest Specialized Program of Research Excellence (SPORE)

The Pacific NW Prostate Cancer SPORE is a multi-project research program of clinical, translational and population sciences research aimed at ascertaining drivers of metastatic disease risk and developing interventions to prevent and treat high-risk disease. Dr Etzioni is the Principal Investigator and Mr. Gulati is the chief statistical consultant on the Biostatistics Core for the Pacific NW Prostate Cancer SPORE. The Biostatistics Core develops strategies for study design, data collection, measurement, and analysis to validly and rigorously address the critical hypotheses and questions of SPORE projects, reduce systematic bias, and ensure a high likelihood of detection of biologically meaningful effects. The Biostatistics Core also identifies and implements quantitative methods to address scientific questions of interest and develop the evidence needed to support study hypotheses.

Modeling Outcomes of Reclassifying Low Grade Prostate Cancer

Overdiagnosis and overtreatment of low-grade prostate cancer has long been a concern. The Etzioni group was the first to quantify the frequency of overdiagnosis in prostate cancer screening and has contributed materially to the methodology for estimating overdiagnosis from screening trials and observational cohorts. In recent years, there have been calls to relabel low-grade (Gleason 6) lesions in the prostate as non-cancer, largely due to their ubiquity among men over 50 and their relative clinical indolence in men undergoing active surveillance. However, the population health effects of such a relabeling remain unclear. The Etzioni Lab is collaborating with colleagues from the University of Chicago and Erasmus Medical Center to model the harm-benefit tradeoffs of implementing such a policy, relative to current management practices. 

This work is funded by a contract from the Centers for Disease Control.

Early Detection Research Network Data Management and Coordinating Center

Over the past 20 plus years the Early Detection Research Network (EDRN) Data Management and Coordinating Center (DMCC), located at Fred Hutch Cancer Center, has served a vital role in co-ordinating network research activities, providing statistical support for study design and analysis, and developing innovative statistical methods for cancer biomarker development. The recently renewed DMCC will add a modeling capability for translation of novel biomarker performance to predicted clinical utility and outcomes. Dr Etzioni is one of the multi-PI’s for the renewal. 

Modeling to Reduce Detection Bias in Cancer Risk Prediction Studies

Detection bias occurs when the predicted risk of disease due to a specific risk factor is distorted by the association between cancer detection and that risk factor. Detection bias can be due to differential screening or biopsy across subgroups defined by a risk factor (e.g. family history) or to differential performance of an existing screening modality (e.g mammography performance for dense versus non-dense breasts). This project aims to use disease natural history modeling fit to data from a screened cohort to estimate the association between a risk factor and the risk of disease onset rather than between the risk factor and disease diagnosis. We examine natural history as a risk factor for prostate cancer using data from the SELECT prostate cancer prevention trial and race/ethnicity and breast density as risk factors for breast cancer using data from the Breast Cancer Surveillance Consortium. 

Prostate Modeling to Identify Surveillance Strategies (PROMISS)

Widespread PSA screening in the US has created an epidemic of low-risk prostate cancer, the vast majority of which is not lethal. Active surveillance (AS) is an increasingly popular option for managing low-risk prostate cancer, but no evidence-based standard exists for how to implement it. Several AS studies are ongoing but they constitute a loose collection of different approaches and their results cannot be readily compared or integrated. This Prostate Modeling to Identify Surveillance Strategies (PROMISS) research will determine an optimal approach to AS given patient characteristics and preferences.

Estimation and Communication of Overdiagnosis in Cancer Screening (Overdiagnosis)

Overdiagnosis, or the detection by screening of cases that would never become clinically diagnosed, is now recognized as the greatest potential harm of screening. Knowledge about overdiagnosis is critical for well-formed screening policies and for well-informed patient decision making. However, overdiagnosis depends on screening practices and personal factors and many published studies are biased or do not apply to populations that differ from those used for estimation. The Estimation and Communication of Overdiagnosis in Cancer Screening (Overdiagnosis) research will advance knowledge about how to validly estimate overdiagnosis and to provide concrete information about overdiagnosis associated with specific cancer screening settings to inform screening policy development and clinical decision making.

Past Projects

A Tool to Translate Intermediate Endpoints to Mortality in Cost-Effectiveness Studies (CANTRANce)

Many cost effectiveness studies comparing methods to prevent, treat, or cure cancer do not have the time or the information to evaluate how the approaches being studied affect cancer deaths. Our goal is to develop a software system to translate the results of these studies into projections of the effects of the methods being compared on deaths due to the disease. Learn more about CANTRANce.

Outcomes Based Guideline Development for Prostate Cancer Screening and Treatment (Guidelines)

Clinical practice guidelines for prostate cancer screening impact millions of men at risk of a prostate cancer diagnosis. The research aims to improve how these guidelines are produced by providing policy makers with a computerized decision support tool that will quantify the benefit-harm tradeoffs associated with candidate guidelines.