Introduction: Blood-based multi-cancer early detection (MCED) tests in development may substantially expand the number of “screenable” cancers. Defining an optimal approach for tissue of origin (TOO) localization in individuals with cancer-suspected MCED results is critical. Two recent prospective trials employed differing and distinct approaches for conducting TOO evaluation. The PATHFINDER study conducted molecular TOO testing for localization and DETECT-A used upfront neck-to-thigh imaging-first to rule-in or rule-out and then to localize the cancer. Using mathematical modeling, we examined the diagnostic burden of each approach. Methods: Building upon a published quantitative framework, a mathematical expression for diagnostic burden was derived as a function of positive predictive value (PPV) of the detection module, TOO accuracy of the localization module, and number of procedures associated with each diagnostic outcome; the latter were estimated from established clinical guidelines with diagnostic procedures narrowly selected for molecular TOO calls. We then explored the relationship between PPV, TOO accuracy, and diagnostic burden. Imaging and molecular TOO strategies were compared by estimating absolute difference in diagnostic burden across a range of performance levels. Results: With a molecular TOO strategy we estimate 2.1 procedures to reach diagnostic resolution for correctly-localized true positives, 4.4 for incorrectly-localized true positives, and 4 for false positives; with imaging TOO, 2.75 procedures for true positives and 2.4 for false positives. Across the entire spectrum of performance for both detection and localization, a strategy with molecular TOO resulted in a mean diagnostic burden of 3.6 (Var 0.198); with imaging TOO, 2.6 (Var 0.010). The breakeven curve defined by zero absolute difference between strategies shows that molecular TOO has lower burden for only 4.5% of all possible PPV and TOO accuracy combinations and that 79% PPV is required for a 90% accurate molecular TOO strategy to be less burdensome than imaging TOO. Conclusions: This analysis demonstrates that imaging TOO is more efficient than molecular TOO across 95.5% of all possible PPV and TOO accuracy combinations. Molecular TOO can reduce diagnostic burden when localization accuracy is very high or overall test PPV is very high. This analysis highlights the need for ongoing innovation in cancer detection and cancer localization as separable parts of the MCED test. In summary, the use of advanced imaging first to determine TOO as part of an MCED screening approach is likely less burdensome than a broad molecular TOO approach, although a nuanced approach to molecular TOO based on probability and type of cancer may contribute to diagnostic efficiency. Citation Format: Christopher Tyson, Elizabeth K. O'Donnell, Elliot Fishman, Vijay Parthasarathy, Tomasz M. Beer. Evaluating the diagnostic burden of tumor localization strategies for multi-cancer early detection tests [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 769.

No abstract

e22508 Background: Public health goals often articulate aspirations to improve cancer survival by some pre-specified percentage. Such survival improvements may be achieved through a combination of medical interventions and lifestyle change. Early detection of cancer can contribute to cancer survival improvements. Impact of early detection programs is realized both through innovation driving advances in screening technologies and increased access and adherence with screening tests. We develop a quantitative approach to characterize the magnitude of early detection across multiple cancer types that corresponds to a specified percent improvement in survival. Methods: A matrix equation is developed relating survival and stage shift. The percent of survival improvement is defined a priori as 20% in this case but can be set to other percentages. We populate the matrix equation with incidence and cause-specific survival for 15 cancers by stage at diagnosis from the National Cancer Institute’s Surveillance, Epidemiology, and End Results database. We then use linear programming to solve the matrix for all specified cancers simultaneously given a set of constraints formulated to steer the solution towards the least stage shift required to achieve the objective and to partially compensate for length-time bias. Results: Based on our target survival improvement of 20%, three common trends emerged across 14 of the 15 cancer types: first, we see that the bulk of survival improvement can be achieved by detecting most disease prior to stage 4; second, that the remaining survival improvement can be achieved by detecting most cancers just one stage earlier; and third, that stage 1 diagnosis is generally not necessary to achieve reasonable survival improvement goals. The solution revealed lung cancer as the one type that required more aggressive earlier detection than others, an expected result given that lung cancer has both very high incidence and very poor survival. Conclusions: The mathematical framework we develop is very flexible and can be helpful for public health officials and innovators to characterize the contribution that earlier cancer detection can make towards improved cancer survival. Our results suggest that detecting cancer prior to the development of distant metastases, even after progression beyond stage 1, has potential for significant public health benefit. This approach may also be extended to assess the impact that emerging treatment landscapes, coupled with early cancer detection, may have on improved cancer survival.

10634 Background: Population-level cancer registries report observed (screening or clinically detected) incident cancer cases. However, the underlying true cancer incidence may be higher than observed. We estimate the unobserved cancer incidence by stage for eight different cancers. Methods: Using the CDC’s National Program of Cancer Registries (NPCR) and the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) combined incidence databases, we first estimated observed incidence rates by cancer type and stage. Newly observed cancers in later stages must have existed at a previous point in time at an earlier stage (and age). Considering that, we used backward induction approach to estimate the unobserved incidence rate of cancer. We also accounted for mortality from competing factors before cancer was detected. Cancer progression between stages was defined by dwell times, which were synthesized in an ensemble approach from published literature and empirical estimates. We estimated unobserved annual incidence cases in early-stage (stages I and II) for the age 50+ population and compared them with observed incidence cases for each cancer type between 2015-2019. Results: The below table shows the annual unobserved and total (unobserved plus observed) cancer incidence by stage for the eight most prevalent cancers. We estimated that the top three cancers with the highest proportion of unobserved early-stage annual incident cases were lung (157,400 cancers unobserved out of 226,400 total), with 70% unobserved; pancreatic (34,700 annual unobserved out of 52,500 total), with 66% unobserved, and non-Hodgkin’s lymphoma (46,400 unobserved out of 73,600 total), with 78% unobserved. In contrast, the cancers with the lowest proportion of unobserved early-stage incident cases were breast (29,100/203,400 [14%]), urinary bladder (27,000/81,800 [33%]), and prostate (110,600/258,900 [43%]). Cancers with a high screening rate have lower unobserved early stage incidence than cancers with a low screening rate or without a screening test. Conclusions: We estimated a large undiagnosed incidence of cancer in early stages. This represents an opportunity for novel early detection technologies to detect cancer in earlier stages. [Table: see text]