Digital Pathology can potentially be a useful test to triage cases that get e.g. MSI testing or not. (But note recent papers on whether digital pathology prediction of biomarkers may suffer confounding which affects apparent accuracy).
See a Linked In blog on the topic of dig path screening prior to genomics, by Katherina von Loga at WAIV former OWKIN DX.
###
Andrew Vickers at MSKCC has done great work on the statistics of this decision theory problem.
For example, he models insertion of kallikrein biomarkers to REDUCE the number of physical prostate biopsies. This probably models similarly to using Dig Path to REDUCE the number of physical MSI genomics.
###
Yes — the person is Andrew J. Vickers, PhD, of Memorial Sloan Kettering. He is an attending research methodologist/biostatistician at MSK, with a particular focus on prostate cancer detection, prediction models, and decision-making. MSK’s own profile highlights his work on baseline PSA, PSA velocity, MRI targeting, Gleason grading, and active surveillance, which fits your recollection very well. (Memorial Sloan Kettering Cancer Center)
The core methodological idea he is most associated with is decision curve analysis (DCA). In his 2006 paper, Vickers argued that the right question is not merely whether a model has a better ROC curve or AUC, but whether using it in practice would actually do more good than harm. DCA does this by calculating net benefit across a range of threshold probabilities — in other words, across the range of risk levels at which a clinician or patient would reasonably choose biopsy or treatment. That is exactly the framework for judging an intermediate test that sits between a screening marker and a definitive but invasive procedure. (PubMed)
In prostate cancer, this became especially important because a test like PSA or a multivariable marker panel is not the end point; the practical question is whether it improves the decision to proceed to needle biopsy. Vickers and colleagues showed that a four-kallikrein panel could predict biopsy outcome in men with elevated PSA and that its use could substantially reduce unnecessary biopsies, with most missed cancers tending to be lower-stage, lower-grade disease rather than the clinically most dangerous tumors. (PubMed)
That body of work is closely linked to what became the 4Kscore. A later European Urology paper explicitly described the four-kallikrein statistical model as commercially available as the 4Kscore and examined its value across biopsy subgroups, again squarely in the setting of deciding when an elevated PSA should lead to biopsy. (PubMed)
So, in practical terms, Vickers’ contribution was to formalize a clinically realistic way to ask: Does this intermediate test improve the biopsy decision enough to justify using it? Not “is it statistically associated with cancer,” but “does it improve decisions, reduce unnecessary procedures, and preserve detection of important disease?” That is why his work has been so influential far beyond prostate cancer. The 2006 DCA paper has been cited extensively, and his later tutorial paper was written precisely because the method became widely used across medicine. (PubMed)
If helpful, I can next give you a one-paragraph explanation of decision curve analysis in plain English, or a short list of the 3–5 classic Vickers papers on this topic.
###
Plain-English one-paragraph explanation:
Decision curve analysis asks a practical question: not merely whether a test is statistically associated with cancer, but whether using it would lead to better decisions for real patients. In the prostate setting, that means asking whether an intermediate test such as PSA or a four-kallikrein panel/4Kscore helps decide who should go on to biopsy in a way that produces more benefit than harm. Vickers’ framework does this by calculating net benefit across a range of clinically reasonable risk thresholds—the point at which a patient or doctor would say, “Yes, this risk is high enough that biopsy is worth it.” That made it possible to judge intermediate tests by their clinical utility—how many unnecessary biopsies they avoid while still preserving detection of important cancers—rather than by AUC alone. (PubMed)
Here are 5 core Andrew Vickers papers on this theme, with PMIDs and brief summaries:
1) Vickers AJ, Elkin EB. “Decision curve analysis: a novel method for evaluating prediction models.” Med Decis Making. 2006. PMID: 17099194.
This is the foundational paper. It introduces decision curve analysis and the concept of net benefit, showing how to evaluate models in terms of whether they improve decisions, not just whether they improve statistical fit or discrimination. This is the key methods paper behind later work on PSA, kallikreins, and biopsy decision-making. (PubMed)
2) Vickers AJ, Cronin AM, Elkin EB, Gonen M. “Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers.” BMC Med Inform Decis Mak. 2008. PMID: 19036144.
This is the important follow-on methods paper. It expands the original DCA framework and shows how it can be used more broadly for diagnostic tests, prediction models, and molecular markers—very much the category into which intermediate screening tests fall. If the 2006 paper gave the concept, this paper made it easier to use in real biomarker research. (PubMed)
3) Vickers AJ, Cronin AM, Aus G, et al. “A panel of kallikrein markers can reduce unnecessary biopsy for prostate cancer: data from the European Randomized Study of Prostate Cancer Screening in Rotterdam, Netherlands.” Cancer Epidemiol Biomarkers Prev. 2008. PMID: 18611265.
This is one of the classic early kallikrein papers. It showed that combining multiple kallikrein forms in blood could improve prediction of biopsy outcome in men with elevated PSA, and that a multivariable model could help determine which men should proceed to biopsy and which could reasonably defer biopsy and continue screening. In other words, it made the intermediate-test idea operational. (PubMed)
4) Vickers AJ, Cronin AM, Roobol MJ, et al. “Reducing unnecessary biopsy during prostate cancer screening using a four-kallikrein panel: an independent replication.” J Clin Oncol. 2010. PMID: 20421547.
This is a major replication/validation paper and one of the most directly relevant to your recollection. It reported that the four-kallikrein panel could predict biopsy outcome in men with elevated PSA and that using it would dramatically reduce biopsy rates. The tradeoff was that some cancers would be missed initially, but those missed cases were said to be predominantly low-stage, low-grade disease, supporting the clinical value of the test as an intermediate triage step. (PubMed)
5) Vickers AJ, Cronin AM, Aus G, et al. “A four-kallikrein panel predicts prostate cancer in men with recent screening: data from the European Randomized Study of Screening for Prostate Cancer, Rotterdam.” Cancer Prev Res (Phila). 2010. PMID: 20400522.
This paper focused on previously screened men, an especially important population because biopsy decisions become trickier after prior PSA testing. It found that the statistical model based on the four-kallikrein panel could substantially decrease unnecessary biopsy in this setting as well. Together with the other 2010 paper, it helped establish the practical case for what later became the 4Kscore-type approach. (PubMed)
A useful later teaching paper, though not prostate-specific, is Vickers AJ, van Calster B, Steyerberg EW. “A simple, step-by-step guide to interpreting decision curve analysis.” Diagn Progn Res. 2019. PMID: 31592444. It is probably the best short “how to read a DCA graph” explainer. (PubMed)
If you want, I can next turn these into a mini reading guide in your usual memo style: “start here, then read these two prostate papers, then the tutorial.”
