Given the prevalent confusion about screening for a cancer—among presumed experts and, secondarily, individuals in the public concerned to take informed decisions about it—a brief (re)exposition of the fundamentals likely is in order.
Underlying any practice of screening for a cancer must be the premise that the cancer’s treatment before its overt manifestations (in symptoms and/or clinical signs) is more commonly curative than later treatment. This premise implies that such early treatment, because of its greater curative effectiveness, is more likely to prevent the cancer’s fatal outcome (which is the principal concern about any cancer for people at large). This premise obviously is clinical, rather than epidemiological, in nature.
From that premise, regarding a particular type of cancer, flows a principle of potential clinical action, given that a person is concerned (perhaps on account of his/her doctor’s input) about the possibility of already having a latent but detectable, potentially fatal case of the cancer: That concern should lead to consideration of actually pursuing, without delay, the cancer’s detection, its latent-stage rule-in diagnosis, that is. This potentially multiphasic process, if opted for, likely would stop after the initial test, on account of its negative result, leaving the person with no diagnosis (incl. rule-out dgn.).
The routine initial (and usually final) phase in the pursuit of a cancer’s latent-stage diagnosis (mammography, say) has no more essential a status in the process than any other phase (reading a biopsy specimen, say). In consequence, the natural clinical concept of the pursuit of a cancer’s latent-stage diagnosis—which pursuit, just like the premise about treatment (above), obviously is a clinical, rather than epidemiological, matter—is this entire process, to whatever stopping point in it.
Given the absence of any special status to the initial testing, the pursuit of latent-stage diagnosis in its entirety is screening for a cancer from the natural—clinical—vantage (insofar as that term is to be used at all in this clinical context). And relegating the initial testing to community medicine is very unnatural and unjustifiable from the vantage of clinical medicine, just as also would be the case with any other element in the potentially multiphasic diagnostic testing.
Even more unnatural, and actually unthinkable, would be relegation of individuals’ counseling about the screening, toward their decisions about it, to the community doctor (epidemiologist; ).
Pertaining to a client’s decision about a round of screening at the time of the counseling, the doctor’s principal need is to know—and counsel the client—about the probability with which the screening (on this person at this time) would have the intended consequence of serving to prevent death from the cancer. And besides, the doctor should know—and counsel the client—about the timing of the death that with this probability would be prevented, were a round of screening carried out now. The extent to which the potential screening on this person at this time would, with that probability, reduce the future incidence of death from the cancer, in the person’s community, naturally is of no concern to the doctor nor to the client seeking the counsel.
That probability is, with a proviso, the product of three component probabilities having expressly to do with the screening (in the natural clinical meaning of this, on this person at this time): the probabilities of the screening actually resulting in diagnosis of the cancer, the diagnosis actually representing malignancy of the lesion (rather than misdiagnosis of its nature), and the screen-diagnosed genuine malignancy actually being curable by early treatment but not if diagnosed (later) in the absence of screening.
The proviso has to do with prospects for survival unrelated to the cancer, to the now potentially present case of it. It is that curative early treatment indeed would be critical for preventing death from the cancer, as this death would not be prevented by intercurrent death from another cause (any more than by later treatment). The probability of otherwise surviving, along with the three above, naturally bears on a rational decision about the screening.
The doctor needs to teach the client about the product of all four of these probabilities, as constituting the probability that screening now would have the consequence of preventing death from the cancer. This probability is client-specific, since the first and last (fourth) of the component probabilities eminently are.
This teaching the doctor needs to supplement with information about the timing of the death that with the already-specified probability would be prevented by the process that would begin (and likely end) with the initial test in the particular regimen of screening that is being contemplated. For, death which, with a given probability, looms at a distance of, say, 15 years in the future is of very different significance at 40 years of age relative to 70.
Implicit in all of this is an outline of what is to be addressed in clinical research on screening for a cancer and, specifically, once preparatory clinical research has led to at least one regimen of screening-and-early-treatment to be addressed in the research; but the nature of the component types of clinical study is not a concern here.
What here needs to be addressed, and persuasively communicated, is this: The results of such randomized ‘intervention’ trials as now are being viewed as essential in research on screening for a cancer are seriously misleading—leading, because of their seriously flawed designs, to substantially downward-biased perceptions of what can typically be accomplished, for individuals, by screening them for a particular type of cancer (breast cancer, say) at a particular age.
If a clinical trial were to be deployed to study (a particular regimen of) screening for a cancer, it naturally would address the fundamental premise of the screening, namely that treatment immediately upon latent-stage diagnosis is (appreciably) more effective—more commonly curative—than treatment once the cancer has progressed to a clinically manifest stage (cf. above). In such a trial, patients with screen-diagnosed cases of the cancer would be randomly allocated either to undelayed, early treatment or to treatment delayed until the cancer becomes clinically manifest (if ever). And with death from the cancer the endpoint of interest (as in the trials that now are viewed as essential), the two subcohorts might have to be followed for up to 20 years, or perhaps even longer, in the case of breast or prostate cancer, to be able to address the full effect of early treatment, relative to late treatment, in preventing death from the cancer.
If the proportions with fatal outcome in the early- and late-treated subcohorts would turn out to be P 1 and P 0, respectively, then the corresponding estimate of the proportion of deaths from the cancer prevented by early treatment as the replacement for late treatment—of the probability of this prevention—would be P = P 0 − P 1.
If at issue were the fundamental premise of screening, specifically, for breast cancer, the 20-year result from such a trial on the cancer’s early treatment presumably would be something like P = 30% − 10% = 20%. With only 10 years of follow-up it would be something quite different and misleading, perhaps P = 10% − 7% = 3%. (Early treatment in place of late treatment is less effective in preventing death from the cancer relatively early after the cancer’s early diagnosis.)
If those 20-year results for P 1 and P 0 were not only available (they aren’t, nor will they be) but also used, as such, as estimates of the respective probabilities, then, for a woman typical of those in the trial as for life expectancy unrelated to the case of breast cancer, the probability that screening—a single round of it—would serve to prevent death from breast cancer would be estimated to be that 0.20 (above) multiplied by the probability that a round of the screening now would result in diagnosis of the cancer (possibly a misdiagnosis, as in the trial). If the latter probability were set at 0.20%—for a woman in her mid-40s—then the probability that the round of screening would serve to prevent death from the cancer, already latently present yet detectable, would be estimated to be 0.20 × 0.0020 = 0.04%.
Keeping those numbers in mind, let us now think of the kind of trial that is being revered as ‘the gold standard’ of scientific assessment of the life-saving consequence of screening for a cancer. In such a trial on screening for breast cancer, women of a given (range of) age, with no overt indication of presence of this cancer and no recent screening for it, are randomly assigned either to mammography, say baseline and five annual repeat mammographies, or to lack thereof. The women are followed for, say, 10 years to document deaths from breast cancer. The essential results are analogous to—but very different in magnitude from—those from the trial (hypothetical) on early treatment, say P 1′ and P 0′. For the ‘proportional reduction in mortality’ from the cancer the estimate is (P 0′ − P 1′)/P 0′, and for the ‘number needed to screen’—six times each—to avert one death from the cancer, the estimate is 1/(P 0′ − P 1′). The result for P 0′ − P 1′ is prone to be downward-biased on account of incompleteness of adherence—for years—to the assigned ‘interventions.’
In this trial, if enrolled were women at age 45, the 10-year cumulative rates of incidence for death from breast cancer—the proportions with this outcome in 10 years—might be P 1′ = 0.30% and P 0′ = 0.35% in the screened and unscreened subcohorts, respectively. These would translate into 14% as the estimate for the ‘proportional reduction in mortality’ from breast cancer, as a consequence of the screening, and into 2,000 as the estimate of the ‘number needed to screen,’ with up to five annual repeat screenings following the baseline screening, to prevent one death from the cancer (occurring in the absence of the screening).
Now, the probability 0.04% that one round of screening on a woman 45 years of age would provide for averting her dying from breast cancer (at age 60, perhaps) means that 2,500 such screenings would typically have the consequence of getting one death prevented. By contrast, that ‘gold-standard’ trial’s corresponding result is 6 × 2,000 = 12,000—a number almost five-fold relative to that from the trial (hypothetical) on the effectiveness of early treatment.
This large discrepancy, even when allowing for the hypothetical, though not arbitrary, nature of the input numbers, illustrates the seriously misleading nature of the ‘gold-standard’ trial on screening for a cancer. Involved are two serious mistakes of thought, one explicit and the other implicit. The explicit mistake is to view, and test, screening as an intervention; and associated with this is, implicitly, the idea that the full effect of the ‘intervention’ would be manifest within just a few years as of any given round of the ‘intervention’s’ application in a cohort of people—that, for example, the consequence of the fifth annual repeat screening for breast cancer would be manifest, and in full, in the ensuing 5 years of follow-up. In the face of the bias arising from these conceptual mistakes, the influence of nonadherence to the multi-year regimens (incl. that of no screening) is of relatively minor consequence, even though substantial nonadherence generally and naturally does attend trials in which—in the spirit of community medicine—the contrast is between people merely invited and not invited for the screening.
Both of these trials, as described above, address prevention of death from the cancer with an implicit proviso (noted above): at issue is prevention of death from the cancer insofar as this prevention is not provided by intercurrent death from another cause (incl. a subsequent case of the same type of cancer). To distinguish between these two types of prevention and to focus on the intended consequence of the screening, directly studied should be the death’s incidence density (rather than its cumulative incidence) over the time span of follow-up. In the trial (hypothetical) on early vs. late treatment, the treatment-specific temporal patterns of this could be translated into the corresponding cumulative rates of incidence, allowing estimation of the typical probability of an early-diagnosed cancer’s curability by early treatment when not curable by late treatment; and estimation of this would allow individualization of the prognosis in respect to death from the cancer in the light of individualized prospects of death from other causes. In the ‘gold-standard’ trial, correspondingly, the focus ought to be on the death’s incidence density in the period of follow-up in which its ratio for the contrast is at its lowest—in which it would reflect in full the curability gain from early treatment if the ‘interventions,’ with full adherence to the assignments, were of sufficiently long duration .