Cancers of female reproductive organs (cervical, endometrial, ovarian), contribute substantially to the cancer burden in women. While these cancers originate from a continuous epithelial lining, etiology and natural history differ across the sites with important implications for risk prediction and prevention.

Based on the excellent understanding of cervical cancer natural history and the critical role of HPV, it is now possible to estimate risk of cervical precancer with high accuracy to identify who requires treatment to prevent cervical cancer. Recent guidelines use a risk-based approach for cervical screening and management recommendations. In contrast, population-based screening is currently not recommended for endometrial cancer. Several strong risk factors have been included in endometrial cancer risk prediction models that can predict clinically meaningful absolute risk. It is critical to expand risk models to include clinical factors such as history of abnormal bleeding and to evaluate whether risk prediction performs well for the aggressive serous subtype which is common in Black women. Population-based screening has not reduced ovarian cancer mortality, but preventive bilateral salpingo-oophorectomy is highly effective at preventing ovarian cancer in high-risk populations. Improved ovarian cancer risk prediction models may allow identification of populations at increased risk of ovarian cancer for preventive interventions or targeted early detection approaches but are challenged by the rarity and heterogeneity of the disease.

In summary, gynecological cancers can illustrate the full range of promises, successes, limitations, and challenges of using cancer risk prediction for clinical and public health applications.

Cancer screening plays a vital role in cancer control and reducing cancer mortality by increasing the likelihood of identifying cancers earlier when there is a greater opportunity for therapeutic intervention. Cancer screening tests for early disease detection, such as mammography, currently focus on detection of a single cancer type, and screening recommendations for these tests are guided by risk of developing that single cancer. Recommended cancer screening tests are available only for a few cancers that collectively account for approximately one-third of cancer-related deaths, leaving opportunity for enhanced screening that would potentially include additional cancer types. Newer technologies that are under development have the potential to reduce the overall cancer burden because they are focused on the early detection of multiple cancers in a single screening test (called multi-cancer early detection, or MCED, tests). Given existing screenings are cancer-specific, the introduction of MCED tests, which simultaneously target detection of several cancers, will require a new approach to risk assessment and screening guidelines development. Specifically, MCED screening guidelines will require that we understand the populations who will most benefit from using MCED tests while minimizing potential harms; identify the risk factors, other than age, that are associated with risk of developing any cancer; and determine the frequency of screening intervals. In this presentation, we will explore strategies for how to consider risk assessment related to MCED tests to begin building the evidence base to inform potential future multi-cancer screening guidelines.

With advancing therapeutics, lung cancer (LC) survivors are rapidly increasing in number. Although mounting evidence suggests LC survivors have high risk of second primary lung cancer (SPLC), there is no validated risk-prediction model available for risk-stratified SPLC surveillance. We aim to develop and validate a risk-prediction model for SPLC in lung cancer survivors and explore the potential of improving efficiency in detecting SPLC through risk-stratified surveillance strategy. Using data from 6325 ever-smokers in the Multiethnic Cohort (MEC) study diagnosed with initial primary lung cancer (IPLC) in 1993-2017, we developed a prediction model for 10-year SPLC risk after IPLC diagnosis using cause-specific Cox regression. We validated it using 2 population-based data—PLCO and NLST, as well as a real-world hospital-based cohort from the Mayo Clinic. Our prediction model demonstrated a high predictive accuracy (Brier score=2.9% [2.4-3.3]) and discrimination (AUC=0.82 [0.78-0.86]) in MEC. External validation using data from PLCO, NLST, and Mayo Clinic showed a Brier score of 3.3-5.2% and an AUC of 0.72-0.81. When the patients from Mayo Clinic were stratified by a 10-year risk threshold of ≥5.6% (i.e., 80th percentile from the SPLC-RAT development cohort), the observed SPLC incidence was elevated in the high-risk vs. low-risk subgroup (13.1% vs. 1.1%, p<1×10^–6). The risk-based surveillance through SPLC-RAT (≥5.6% threshold) outperformed the NCCN guidelines with higher sensitivity (86.4% vs. 79.4%) and specificity (38.9% vs. 30.4%). We proposed a validated a SPLC prediction model that can help identify high-risk LC patients for SPLC and can be incorporated into clinical decision making for SPLC surveillance.

Absolute risk models estimate an individual's future disease risk over a specified time interval by using data on known risk factors from healthy individuals in a given population. Applications utilizing server-side tooling to build, validate, and apply absolute risk models encounter serious limitations in portability and privacy. These limitations stem from their need for specialized computational setups during model development and validation, and from the necessity to circulate user data on remote servers when delivering these models to end-users. Server-side software encompasses packages reliant on technologies unsuitable for native web operations, including R, Python, SAS, and STATA. The R-based Individualized Coherent Absolute Risk Estimation (iCARE) package is similarly affected by these issues. To overcome these issues, we adapted iCARE for web use through the new WebAssembly technology. We refactored the original R-based iCARE into a Python package (Py-iCARE) and then compiled it to WebAssembly (Wasm-iCARE), creating a portable web module that operates entirely within the privacy of the user’s device.

In this presentation, we will demonstrate the portability and privacy of Wasm-iCARE through two use-cases: 1) enabling researchers to create reproducible model validation workflows that can also be replicated with other related validation cohorts; and 2) empowering developers to build and deliver risk models as privacy-preserving consumer-facing applications. Special attention will be given to how Wasm-iCARE supports adherence to the FAIR (Findable, Accessible, Interoperable, Reusable) principles. Wasm-iCARE aims to foster accessible and privacy-preserving risk tools, accelerating their validation and delivery.