Modeling Multicancer Screening

There are already many models and metrics in the public domain for evaluation of screening for a single cancer, some focusing on description of test accuracy, others on the balance of favorable and unfavorable effects of the screening (1–4). With the prospect of blood tests for multiple cancers, there is a gap for similar models and metrics for multicancer early detection (MCED) tests used in screening (5).

In this article, Jiao and colleagues develop a basic arithmetic framework for estimating the likely benefit in terms of cancers successfully detected and lives saved, and the number of further investigations prompted in subjects who transpire not to have cancer, which is both a human and a resource cost of screening (6). The formulation is attractive, being based on simple, closed-form algebra. It uses as fundamental building blocks, cancer-specific prevalence, marginal sensitivities, and anticipated proportional mortality reductions associated with early detection. The method enables calculation from these of the likely numbers of cancers detected, cancer-related deaths prevented, and investigations in those who do not have cancer. Incidentally, the latter are referred to as “unnecessary tests”, on the basis that in the absence of the screening test, they would not have taken place. However, it could be argued that further investigation in a person identified by a screening test as having a significant probability of cancer being present is necessary, whatever the result of the further investigation. The authors derive estimates for the situation of testing for breast cancer and a number of cancers at other anatomic sites, and find, based on a number of published estimates and inevitably some assumptions, that there is indeed potential for a multicancer test to improve on the single-cancer screening tests which we currently use.

The authors rightly acknowledge the assumptions made, and note a number of qualifications to the model and the reported estimates. For example, the effect of such a screening test in terms of cancer-related deaths prevented will be dependent not only on stage-specific sensitivity but also on sensitivity to those biological cancer types which are potentially lethal and for which early detection materially improves the prospect of successful treatment. The issue is related to both overdiagnosis – detecting early-stage cancers whose lethal potential is far less than that of symptomatic early-stage cancers – and finding cancers born-to-be-bad – detecting seemingly early-stage cancers that nevertheless have poor prognosis and may be lethal despite early treatment.

The authors' model is a welcome development. Colleagues may disagree with some of the prevalence and mortality reduction figures used in their worked examples. It is likely, for example, that prevalence might amount to more than 5 years of incidence for some (on average) slowly developing cancers such as prostate, but to less than 2 years for more rapidly developing cancers such as ovarian or pancreatic cancers. However, the authors supply an online calculator, where users can derive outputs from their own preferred inputs.

Given the basic model, how might it be developed further?

In the first place, the approach could be enhanced to incorporate some of the complexities of diagnostic workup. Specificity may vary cancer by cancer, in the sense that false positives with an indication for cancer at one anatomic location might be more frequent than false positives with an indication at another location. Moreover, some confirmatory investigations are more expensive or have higher human costs than others. A mammogram plus ultrasound is less expensive and better tolerated than a colonoscopy, a skin biopsy is better tolerated than a lung biopsy, and so on. Quantification of the burden of additional investigation specific to the cancer is a likely aim for future development.

The model could be developed to possess greater flexibility with respect to the probabilities of multiple outcomes, including:

• Modeling the effect of a cancer signal but at the wrong site of origin (so a subject might have further investigation but at the wrong anatomic site and still have their cancer missed, or the cancer is found early but only after investigation at several sites);

• True positive cancer signal and investigation at the correct site of origin, but still missed during work-up (either due to imperfect follow-up or because the cancer is too small to be detected by conventional diagnostic workup);

• The presence of multiple primary cancers (admittedly a rare event), with only one site of origin being picked up (this could be a big problem if, for instance, the test finds common indolent cancers with many years of lead-time which mask the signal for a rare aggressive cancer found with just a few months of lead-time);

• Expanding the model to more than one sensitivity parameter per cancer, for example, for different stages of disease;

• More explicit consideration of the fact that the specificity is likely to get worse if more sites are included in the test;

• Explicit modeling of the marginal harms and benefits of adding an additional cancer site to an existing MCED test.

Another important task for the future is to assess how the expected outcomes (including the diagnostic workup) compare with the current context. In relation to this, the benefits and harms of multi-cancer testing will depend on whether the multi-cancer test replaces existing programs, such as mammography, or whether it is applied as a supplement to them. Also, it would be of interest to assess the benefits and harms not only in comparison with current screening programs but with routine and urgent symptomatic referrals, since it seems likely that some MCED tests used in screening will have a better positive predictive value than do symptoms that lead to urgent referral.

It would be too much to expect from this paper, but future developments may wish to consider how an MCED test might be tailored to individuals taking into account their particular risk of developing different cancers over the next 10 years. How might a genetic predisposition to one particular cancer affect the benefits and harms of screening with a multi-cancer test?

From the above, it can be seen that we owe the authors a vote of thanks for this development, but we look forward to further enhancements to incorporate some of the potential complexities of early detection of cancer at multiple sites. As multi-cancer screening becomes an increasingly hot topic, the task of quantifying its effects will become more and more important.

P. Sasieni reports grants and personal fees from GRAIL. INC; grants from Cancer Research UK; and grants from National Institute for Health Research outside the submitted work. No disclosures were reported by the other authors.