Screening guidelines disagree because the same trials support more than one answer

Cancer screening guidelines disagree about starting ages, and the reason is not that different committees read different studies. It is that the randomised evidence produces benefits small enough in absolute terms that reasonable people weighing them against overdiagnosis, false positives and biopsy harms arrive at different answers. Where the disagreement is sharpest — mammography in the forties — the underlying number is a fraction of one death prevented per thousand women screened over a decade.

The breast screening case, laid out

The US Preventive Services Task Force recommends biennial screening mammography for women aged 40 to 74, a B recommendation, concluding with moderate certainty that it has a moderate net benefit [s1]. It concludes the evidence is insufficient to determine the balance of benefits and harms for women 75 and older, and insufficient to assess supplemental screening with ultrasound or MRI for women with dense breasts on an otherwise negative mammogram [s1]. For context, the Task Force notes that an estimated 43,170 US women died of breast cancer in 2023, that non-Hispanic White women have the highest incidence and non-Hispanic Black women the highest mortality rate [s1].

A systematic review update commissioned to inform the Canadian Task Force on Preventive Health Care guideline searched the same literature to July 2023 and found no new randomised controlled trials at all — three new papers reporting on existing trial data and 26 observational studies [s2].

Working from that trial data, the relative reduction in breast cancer mortality with screening mammography in a general population was 15% (RR 0.85, 95% CI 0.78 to 0.93) [s2]. Expressed absolutely over 10 years, that came to 0.27 fewer breast cancer deaths per 1,000 women aged 40 to 49; 0.50 per 1,000 aged 50 to 59; 0.65 per 1,000 aged 60 to 69; and 0.92 per 1,000 aged 70 to 74 [s2]. For all-cause mortality the trial data gave a non-significant 1% relative reduction (RR 0.99, 95% CI 0.98 to 1.00), or absolute effects over 10 years of 0.13, 0.31, 0.71 and 1.41 fewer deaths per 1,000 across the same four decades [s2].

On the harm side, the review put overdiagnosis in the trial data at 1.95 more invasive and in situ cancers per 1,000 for women aged 40 to 49 — about 1 more invasive cancer per 1,000 — and 1.93 more invasive and in situ cancers per 1,000 for those aged 50 to 59 [s2]. A sensitivity analysis excluding studies at high risk of bias put the 40-49 figure at 1.57 more invasive and in situ cancers per 1,000 [s2]. Observational studies gave much larger relative mortality reductions, ranging from 29% to 62%, which is exactly the kind of gap between trial and observational estimates that committees have to adjudicate [s2].

The review's overall GRADE assessment was low or very low certainty, meaning the evidence is very uncertain about the effect of breast screening on the outcomes evaluated [s2].

Two committees can look at 0.27 fewer deaths and 1.57 to 1.95 more diagnosed cancers per 1,000 women in their forties and land in different places without either being wrong about the arithmetic.

Colorectal cancer: the disagreement is about incidence, not benefit

Colorectal screening carries stronger evidence, and the Task Force's recommendations reflect that in their grades. Screening adults aged 50 to 75 is an A recommendation, with high certainty of substantial net benefit [s3]. Screening adults aged 45 to 49 is a B recommendation, with moderate certainty of moderate net benefit [s3]. For adults aged 76 to 85 who have been previously screened, the Task Force says clinicians should selectively offer screening, a C recommendation, because the net benefit of screening everyone in that group is small [s3].

The 2021 lowering of the start age from 50 to 45 was driven by changing epidemiology rather than new trial results: incidence of colorectal adenocarcinoma in adults aged 40 to 49 increased by almost 15% from 2000-2002 to 2014-2016, and an estimated 10.5% of new colorectal cancer cases now occur in people younger than 50 [s3].

The Task Force also flagged the problem that dwarfs the start-age question. In 2016, 26% of eligible US adults had never been screened for colorectal cancer, and in 2018, 31% were not up to date [s3]. Adults who have never been screened are the ones most likely to benefit [s3].

Prostate cancer: where the harms are quantified

Prostate screening is the clearest example of a guideline declining to give a single answer. For men aged 55 to 69 the Task Force says the decision should be individual, a C recommendation; for men 70 and older it recommends against PSA-based screening outright, a D recommendation [s4].

The numbers behind that split are unusually explicit. Randomised trial evidence indicates PSA-based screening in men aged 55 to 69 may prevent approximately 1.3 prostate cancer deaths over roughly 13 years per 1,000 men screened, and approximately 3 cases of metastatic prostate cancer per 1,000 men screened [s4]. Against that, about 1 in 5 men who undergo radical prostatectomy develop long-term urinary incontinence, and 2 in 3 experience long-term erectile dysfunction [s4]. The lifetime risk of being diagnosed with prostate cancer is about 13% and of dying of it 2.5%, with a median age at prostate cancer death of 80 [s4].

The Task Force's conclusion is that the net benefit for men aged 55 to 69 is small for some men, and that how each man weighs the specific benefits and harms determines whether the overall net benefit is positive at all [s4]. It adds a line rarely quoted: clinicians should not screen men who do not express a preference for screening [s4].

What actually generates the disagreement

Three things, mostly.

Relative versus absolute framing. A 15% mortality reduction and 0.27 fewer deaths per 1,000 describe the same result [s2]. Committees that lead with the first reach different-feeling conclusions from those that lead with the second.

The weight given to overdiagnosis. Deaths prevented and cancers overdiagnosed are not commensurable quantities, and no formula converts one into the other. A committee that treats an overdiagnosed cancer as a serious harm and one that treats it as an acceptable cost of finding real ones will diverge even with identical inputs [s2] [s4].

Certainty grading. The Canadian review graded its evidence low or very low certainty, and no new randomised trials exist to raise it [s2]. The US Task Force reached moderate certainty on mammography for 40 to 74 partly by supplementing trial evidence with collaborative modelling studies [s1].

None of this tells any individual when to start screening, or whether to. These recommendations are population-level, they assume average risk and exclude people with family history or genetic predisposition, and they are explicitly framed by their own authors as inputs to a conversation with a clinician rather than as instructions.