The central distinction is between bounded perceptual work, such as flagging an abnormality in an image or during an endoscopic procedure, and open-ended reasoning, which requires clinicians to synthesize histories, examinations and test results into competing diagnoses and a management plan. In the Swedish Mammography Screening with Artificial Intelligence trial, AI-supported reading increased breast-cancer detection from 5.0 to 6.4 per 1,000 women, a relative increase of about 29%, while reducing screen-reading workload by 44.2%. The completed trial also found a non-inferior interval-cancer rate: cancers arising between screening rounds numbered 1.55 per 1,000 participants with AI support and 1.76 per 1,000 with standard double reading. Sensitivity was higher with AI support, at 80.5% versus 73.8%, while specificity was 98.5% in both groups. The analysis did not measure breast-cancer mortality or overdiagnosis, leaving those outcomes unknown.

Real-time computer-aided detection produced a similar pattern during colonoscopy. In a multicenter randomized trial involving 685 patients, the adenoma detection rate rose from 40.4% without AI assistance to 54.8% with it, corresponding to a relative risk of 1.30. The increase was concentrated among smaller adenomas, including those no more than 5 millimeters and those measuring 6 to 9 millimeters. A second randomized trial involving 1,058 patients also reported an increase, from 20.3% to 29.1%. Neither colonoscopy trial assessed whether the higher detection rate translated into lower colorectal-cancer incidence or mortality, and the differences in baseline detection rates mean that the absolute gains should not be treated as interchangeable estimates.

Open-ended diagnostic reasoning produced a notably different result. In a randomized trial, 50 attending and resident physicians in internal, family and emergency medicine worked through six complex clinical vignettes with conventional digital resources alone or with the same resources plus GPT-4. Blinded experts assessed their performance with a validated diagnostic-reasoning rubric. The median score was 76% with access to the model and 74% without it, an adjusted difference of two percentage points. The confidence interval allowed a modest benefit and therefore represented a null finding within a bounded range, not proof that the approaches were equivalent. GPT-4 achieved a median score of 92% when used alone on the same vignettes, 16 percentage points above physicians using conventional resources. The trial did not directly test why the model’s standalone performance was not reproduced in collaborative use.

That gap may reflect how the systems are used, although the review presents that explanation as a hypothesis rather than an established mechanism. During mammography and colonoscopy, the algorithm directs attention to a specific location that a clinician can inspect at modest cost. A language model may instead supply a fluent diagnostic conclusion whose verification requires nearly the same reasoning needed to reach the conclusion independently. The review proposes that calibrated use involves forming one’s own hypotheses first, treating the output as one piece of evidence and checking disagreement, whereas uncalibrated use allows the algorithm to anchor judgment. These are proposed poles on a continuum, not clinician categories directly observed in the studies.

Experimental studies indicate that incorrect advice can sharply reduce performance. In a simulated mammography reader study involving 27 radiologists, correct BI-RADS assignments fell when the system provided incorrect suggestions. Accuracy declined from 79.7% to 19.8% among inexperienced readers, from 81.3% to 24.8% among moderately experienced readers and from 82.3% to 45.5% among very experienced readers. The design compared readers with their own earlier performance and stratified results by experience, but it deliberately included incorrect advice in 12 of 40 test cases, a much higher error rate than would be expected from a deployed system. It establishes susceptibility under uncertainty rather than the likely size of harm in routine care.

Missing alerts can also alter behavior. In a small eye-tracking study, 10 breast-screening readers interpreted 60 examinations with and without AI prompts. On cases the AI failed to flag, median sensitivity fell from 71% without assistance to 39% with it, and readers’ gaze patterns changed only when prompts were incorrect. The test set was deliberately enriched for errors, so the findings remain a signal rather than a realistic effect estimate. A separate randomized trial involving 44 physicians who had completed 20 hours of AI-literacy training found that deliberately erroneous language-model advice reduced reasoning accuracy. Because every participant was trained and there was no untrained comparison group, the experiment shows that this training did not prevent the observed effect, but it does not establish that training itself is ineffective.

The review also identified a possible threat to skills retained without AI. In a retrospective pre–post study of 1,443 unassisted colonoscopies performed by 19 experienced endoscopists in four Polish centers, adenoma detection fell from 28.4% before computer-aided detection was introduced to 22.4% afterward, an absolute decrease of six percentage points and a relative reduction of about 21%. Exposure to AI remained associated with lower detection in multivariable analysis. The finding is consistent with clinicians substituting automated prompts for their own vigilance, but the design cannot exclude changes in case mix, temporal trends or seasonal variation. It also does not show that any decline is lasting or occurs in other specialties.

Problems can arise even when individual clinicians never see an obvious error. An external validation of the Epic Sepsis Model across 38,455 hospitalizations found an area under the receiver operating characteristic curve of 0.63, with 33% sensitivity, 83% specificity and a positive predictive value of 12%. About two-thirds of sepsis episodes were missed and roughly 88% of alerts were false positives. A separate audit of a population-health algorithm used expenditure as a proxy for illness, systematically under-identifying Black patients with the same clinical burden because less was historically spent on their care. In the authors’ simulation, replacing expenditure with a direct measure of illness increased the share of Black patients selected for additional support from 17.7% to 46.5%. That was a modeled counterfactual rather than the measured result of deploying a corrected system.

The evidence base remains uneven. The review was based on searches of Scopus, PubMed/MEDLINE, SciELO and LILACS in English, Spanish and Portuguese, followed by purposive selection by one reviewer and appraisal of nine anchor studies. It included randomized and prospective evidence, reader experiments, pre–post comparisons and external validation, but no meta-analysis, duplicate screening or exhaustive search. Most evidence of benefit came from large trials measuring intermediate outcomes, while evidence of harm came mainly from small or design-limited studies. The reviewer also identified no eligible study of human–AI diagnostic reasoning in Latin America within the search, although this was not a proof that no such research exists. Local validation and a broader regional review would be needed before the true regional evidence base could be characterized.

The practical implication is not that AI should be treated uniformly as either a cognitive amplifier or a source of deskilling. Evaluation must directly compare assisted and unassisted clinician performance, measure whether clinicians disagree with or override the system and track performance without assistance. The review’s proposed safeguards—including periodic audits of unassisted competence, subgroup monitoring and documentation of why clinicians accept or reject algorithmic advice—remain research proposals rather than established standards. Likewise, explainable AI is not automatically a safeguard: post-hoc explanations can be unstable or contradictory and may increase trust without helping clinicians distinguish correct output from incorrect output. Ultimately, the studies support a conditional conclusion: AI can improve clinicians’ detection when it provides a bounded, verifiable cue, but a model’s impressive standalone performance is not evidence that physicians will reason better with it, and apparent safety cannot be inferred from the scarcity of studies demonstrating harm.

Subject of Research: Effects of artificial intelligence assistance and overreliance on physicians’ clinical reasoning and diagnostic performance

Article Title: Augmentation or erosion? Artificial intelligence and the transformation of clinical reasoning: a critical narrative review

Article References: Sánchez, J. D. (2026). Augmentation or erosion? Artificial intelligence and the transformation of clinical reasoning: a critical narrative review. Frontiers in Digital Health, 8, Article 1949323. https://doi.org/10.3389/fdgth.2026.1949323

Image Credits: AI Generated

DOI: 10.3389/fdgth.2026.1949323

Keywords: artificial intelligence, clinical reasoning, diagnostic accuracy, automation bias, large language models, mammography screening, computer-aided colonoscopy, medical AI, clinician deskilling, algorithmic bias, Augmentation, erosion

The numbers behind the study are sobering. Among the 4,004 patients diagnosed between 2010 and 2019, 23 percent died within six months of diagnosis. That figure, the authors argue, is not simply a reflection of aggressive biology. It is also a signal of inequities in how health care is delivered and accessed. Patients diagnosed with de novo metastatic disease are especially vulnerable to disparities in health equity and access to care, and most previous work on this question has come from the United States, leaving a gap in Canadian data that the research team set out to fill.

Breast cancer is the most common cancer among women in Canada, with roughly one in eight females receiving a diagnosis during their lifetime. In Ontario alone, about 3,000 patients are diagnosed annually, and de novo metastatic disease accounts for 3 to 6 percent of all breast cancer diagnoses. Although advances in systemic therapy have extended survival for many people with metastatic disease, the benefits have not been distributed evenly. The study asked whether differences in access and social circumstances could help explain why some patients die so soon after learning they have cancer.

Methodologically, the researchers drew on ICES, an Ontario-based research institute holding linked health administrative data covering the province’s population. They applied the Ontario Marginalization Index, a validated tool that quantifies dimensions of marginalization across neighborhoods, including housing stability, income, age, and employment status. By linking these sociodemographic measures to clinical outcomes, the team could test whether marginalization was independently associated with early mortality after adjusting for other factors.

The results pointed clearly in one direction. Among patients who died within six months, older age, the presence of chronic illness, and higher levels of housing and income marginalization were all associated with early death. In practical terms, people living in unstable housing situations or in low-income neighborhoods were more likely to be diagnosed with disease so advanced that survival measured in months rather than years, and once diagnosed, they died faster than their more advantaged counterparts.

Access to primary care emerged as one of the most consequential and modifiable factors. Patients who were linked to a family physician had 26 percent lower odds of early death, and they were also more likely to have received a mammogram within the two years preceding their diagnosis. The authors note that access to a primary care provider is a growing challenge in Canada, with one in four Canadians lacking regular access to primary care. Without a consistent provider, symptoms may go unexamined, screening may lapse, and the path from first concern to diagnosis and treatment can stretch out, allowing disease to progress to an incurable stage.

The mechanism is not mysterious. Primary care serves as the entry point to the entire cancer care system. A family physician orders screening mammograms, evaluates new symptoms, coordinates referrals to specialists, and manages the chronic conditions that complicate cancer treatment. Lack of access to primary care may negatively affect the timeliness of diagnosis and also lead to delays in starting important anticancer therapies, the authors write. In metastatic disease, where every week of untreated progression matters, those delays can translate directly into lost survival time.

The study also identified an association between rural residence and early death. The researchers suggest this may reflect the logistical burden of travelling for treatment, which in Ontario is often coordinated through larger urban cancer centres. Patients living far from these hubs face longer journeys for chemotherapy, radiation consultations, and follow-up visits, which can be particularly taxing for people already weakened by advanced disease or constrained by income and employment. The authors call for future studies to determine effective ways of optimizing rural delivery of cancer care and integrating virtual care, which could reduce travel burdens and connect remote patients with oncology expertise.

Beyond diagnosis and treatment, the authors emphasize that primary care plays a pivotal role across the entire cancer continuum, from prevention and screening through treatment support, adverse-effect management, symptom control, and psychosocial and palliative care. The demand for cancer care is rising exponentially as the population ages and treatment options expand, and specialist oncologists are unlikely to be able to meet these care needs alone. Strengthening the primary care workforce and ensuring that every patient has a regular provider is therefore not merely a convenience but a structural determinant of cancer survival.

The study’s implications extend to policy. Optimizing cancer prevention and mammography screening in marginalized communities may reduce the likelihood of presenting with metastatic disease in the first place, the authors conclude, and the findings highlight the need for better primary care access across the province. As Ontario and other Canadian jurisdictions grapple with a growing shortage of family physicians, the research offers a concrete warning: when people fall through the cracks of primary care, cancers are found later, and some patients die within months of a diagnosis that earlier detection might have transformed. The research was supported in part by donor funding through London Health Sciences Foundation.

Subject of Research: Sociodemographic factors and primary care access associated with early death from de novo metastatic breast cancer in Ontario, Canada

Article Title: Breast cancer: marginalization, lack of primary care provider increase risk of early death from newly diagnosed metastatic cancer

Article References: Breast cancer: marginalization, lack of primary care provider increase risk of early death from newly diagnosed metastatic cancer. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: breast cancer, metastatic cancer, primary care, health equity, marginalization, Ontario Marginalization Index, early mortality, mammography screening, rural health, CMAJ, health administrative data, cancer disparities

The research, published in JNCI, the Journal of the National Cancer Institute, is a meta-analysis that brings together data from all eight randomised controlled trials of mammography screening ever conducted: the New York Health Insurance Plan trial, the Malmö trial, the Two-County trial, the Edinburgh trial, the Canadian National Breast Screening Study, the Stockholm trial, the Gothenburg trial and the UK Age trial. Rather than treating each trial as an isolated source of truth, the team reanalysed the pooled evidence against a real-world benchmark, using two Danish regional screening programmes as a reference population. The comparison included both invasive breast cancer and ductal carcinoma in situ, the non-invasive condition often picked up through calcifications on mammograms.

The central finding is striking in its simplicity. When the researchers accounted for the temporal dynamics of screening, the excess number of breast cancer cases detected in the randomised trials closely matched what would be expected in a Danish screening population, where overdiagnosis is estimated to be below five percent. In other words, the apparent surplus of cancers in the trials, which has long been cited as proof that overdiagnosis is a substantial problem, looks far more like the ordinary lead-time effect of finding cancers earlier than they otherwise would have been found.

Understanding why earlier estimates were so high requires a technical detour into how screening changes the timing of diagnosis. When a screening programme is introduced, the number of breast cancer diagnoses initially rises sharply, because cancers are detected earlier than they would have been without screening. Over time, this initial surge should be followed by a compensatory drop, as some of the cancers counted in the early period would otherwise have been diagnosed later in the absence of screening. If researchers fail to take this pattern into account, the temporary increase can be mistaken for overdiagnosis, inflating the estimate considerably.

Professor Elsebeth Lynge, professor emerita at the Department of Public Health at the University of Copenhagen, explained the mechanics behind the distortion. She noted that the pattern can also be affected if women in either the screening or control groups continue to undergo screening after the trials had ended, which was common in practice. Contamination of this kind blurs the boundary between the compared groups and can make the control group’s later diagnoses look like a shortfall rather than a delay. The Danish experience offered a natural laboratory for disentangling these effects, because organised screening was introduced in some Danish regions 17 years before others, allowing the researchers to observe how breast cancer diagnoses changed both when screening was introduced and over the longer term.

By comparing breast cancer incidence at identical timepoints in the trials and in Danish routine screening data, the team could examine how similar the observed patterns were and what that similarity implies for overdiagnosis estimates. The alignment between trial data and routine programme data suggests that the trials, when interpreted in their full temporal context, are consistent with overdiagnosis of less than five percent rather than with estimates approaching 50 percent. Matejka Rebolj, senior epidemiologist at Queen Mary University of London, said that some previous high estimates, which influenced screening guidelines and patient communication, were based on evidence gathered before trial data had fully matured.

The analysis also corrected for three structural differences between the trials that can skew overdiagnosis calculations. The first was whether women in the control groups were themselves screened, either after the trial ended or following the introduction of routine programmes, which erodes the contrast between screened and unscreened populations. The second was the number of screening rounds offered in each group, since more rounds provide more opportunities to detect cancers early and shift diagnoses forward in time. The third was the length of follow-up, because a longer observation period allows time for cancers detected early through screening to be diagnosed later in the control group, which is exactly what the lead-time model predicts. After accounting for these differences in screening intensity and follow-up between the groups, the results point to overdiagnosis being substantially less common than previously believed.

The definition of overdiagnosis itself deserves attention, because it is broader than many people assume. It occurs when screening detects a cancer that would never have become life-threatening or caused symptoms during a woman’s lifetime, meaning that without screening she would never have known she had it. It can also include women who die from other causes shortly after a breast cancer diagnosis, since poor health or limited life expectancy means these women were unlikely to benefit from early detection in the first place, as treating the cancer had little chance of improving their health or prolonging their lives. Distinguishing genuine overdiagnosis from cancers that were simply found earlier is therefore one of the hardest measurement problems in screening research, and it is precisely the problem the Danish reference data helped solve.

For Sisse Helle Njor, professor at the University of Southern Denmark and Lillebælt Hospital, the motivation was to bring together all the randomised evidence for a clearer picture. She observed that randomised trials have often been cited as evidence that overdiagnosis is a substantial problem, but that this interpretation is not as straightforward as it may seem. Her conclusion for women is direct: most women will not develop breast cancer, but the study offers reassurance that the benefits of detecting breast cancer early and preventing premature death will outweigh the small risk of unnecessary treatment. Accurate information about how often benefits and harms occur is essential for women weighing whether to attend screening, and the new figures suggest that the balance tilts more favourably than earlier communications implied.

The implications extend beyond the clinic into public health policy. Overdiagnosis estimates have shaped screening guidelines, invitation letters and the international debate about population-based mammography for decades, and figures as high as one in two screen-detected cancers have fuelled scepticism about organised programmes. If the matured trial data support estimates below five percent, the framework for communicating risk to invited women may need revision, and the researchers hope their analysis will provide a basis for a more realistic interpretation of the evidence. The work was supported by the Novo Nordisk Foundation and Cancer Research UK, and its publication in a leading oncology journal signals that the long-running controversy over mammography’s harms may be entering a more settled phase, one in which the fear of unnecessary treatment no longer overshadows the demonstrated capacity of screening to prevent deaths from breast cancer.

Subject of Research: Overdiagnosis in mammography screening estimated through a meta-analysis of randomised trials benchmarked against Danish screening data

Article Title: New analysis points to low overdiagnosis in breast cancer screening

Article References: New analysis points to low overdiagnosis in breast cancer screening. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: breast cancer, mammography screening, overdiagnosis, meta-analysis, randomised controlled trials, lead-time bias, Danish screening programmes, ductal carcinoma in situ, cancer epidemiology, screening guidelines, JNCI, public health