Missing cancer: from individual error to system failure

Aneel Bhangu, Surgical Data Institute, Department of Applied Health Sciences, University of Birmingham, Birmingham, UK

DOI: 10.5281/zenodo.20799276

Every doctor who sees patients will miss cancers; I have missed cancers. The more patients a doctors see across a career, the more cancers they will miss. I have also seen patients whose cancers were missed before they reached me, sometimes by months, sometimes by years. The experience is one of the hardest in clinical practice, and its falls heaviest on the patient's family, then on the patient, and then on the clinician responsible. This is the starting point for understanding why missed cancer cannot be reduced to a question of individual competence, and why the systems within which clinicians work are the source of both fault and improvement.

At the point of first presentation in primary care, a GP faces a patient whose symptoms may be non-specific, whose examination is unremarkable, and whose overall risk appears low. Cancer is one of thousands of diagnostic possibilities at that moment, competing with symptom complexity, cognitive load, and time. An NHS Resolution thematic review of 105 settled claims relating to delayed cancer diagnosis in general practice found that 40% of diagnoses were made following routine referrals or emergency attendance rather than urgent suspected cancer pathways, and that 65.7% of patients were diagnosed with stage 3 or 4 disease¹. Missed colorectal cancer was the most common type and the window for early intervention had already closed by the time of the subsequent diagnosis.

Each step in the diagnostic pathway carries its own miss rate: clinical examination, endoscopy, and imaging all have a cancer miss rate due a combination of technical issues and then physical, social, and environmental factors associated with the clinician doing or reporting the scan². When these miss rates compound, the Swiss cheese model of system failure applies directly, where each layer has its own perforation and when the perforations align, the cancer passes through undetected. Responsibility for a missed cancer rarely sits cleanly with one clinician at one moment, rather it distributes across referral decisions, investigation choices, reporting thresholds, and the safety-netting mechanisms that were or were not in place. Transparent documentation of decision-making at each stage matters not only for accountability but because it is the only mechanism through which causation can be traced and learning fed back into the system.

The response to a missed cancer begins with candour, which requires that patients and families are told when something has gone wrong and that an honest account is provided. In most cases, understanding and acknowledgement is what they ask for most urgently. Alongside candour, rapid clinical correction is essential, since the interval between recognition of the miss and initiation of appropriate treatment is a direct determinant of outcome.

The opportunity to reduce miss rates across the diagnostic pathway is increasing. Computer-aided detection systems in endoscopy have been evaluated across multiple large randomised trials, and a meta-analysis of 28 trials involving nearly 24,000 participants demonstrated a 20% increase in adenoma detection rate and a 55% reduction in adenoma miss rate with AI-assisted colonoscopy compared with standard examination³. Similar gains are emerging in radiology, where AI systems assist in the detection of pulmonary nodules and breast lesions that readers may overlook under conditions of high volume or fatigue. AI does not eliminate miss rates as it shifts and reduces them. An AI system embedded in an endoscopy workflow has its own sensitivity and specificity characteristics and its own conditions under which performance degrades. For AI to fulfil its potential, it must be integrated into systems that monitor its performance and report its miss rate with the same transparency expected of human clinicians.

Beyond these single diagnostic encounters, AI in cancer diagnosis may resolves missed and delays in patient admission and tracking. Systems that integrate primary care records, investigation results, imaging reports, and follow-up data could identify patients whose symptom trajectory warrants reassessment, flag investigations requested but not completed, and alert clinicians when risk has changed. The missed cancer that presents as an emergency two years after a normal investigation is often not a failure of that investigation alone as it is a failure to re-evaluate risk as time and symptoms evolve. Technology that supports that longitudinal view can reach a category of missed diagnosis that no improvement in clinical investigation will achieve; who is responsible when the AI fails (industry, clinician, hospital manager) is a problem for the future; it is likely to fall onto the shoulders of a responsible clinician.

So, the individual clinician still matters enormously. The quality of a consultation, the thoroughness of a clinical assessment, the decision to pursue an atypical presentation remain the key moments in a diagnostic trajectory. The clinician who operates within an increasingly robust system that supports good decision-making, monitors outcomes, learns from its misses, and deploys technology as an additional detection layer will miss fewer cancers than one working without that infrastructure.

Conflict of interest statement: None declared.

Corresponding author: Professor Aneel Bhangu, Director, Surgical Data Institute, University of Birmingham, UK. a.a.bhangu@bham.ac.uk

References

1. NHS Resolution. Thematic review: delayed cancer diagnosis in general practice. London: NHS Resolution; 2025. Available from: https://resolution.nhs.uk/2025/10/22/nhs-resolution-report-reveals-pattern-in-missed-cancer-diagnoses (accessed June 2026).
2. Li Y, Xiong H, Liang T et al. Missed colorectal cancer diagnosis by screening colonoscopy based on the PLCO cancer screening trial. Int J Colorectal Dis 2025; 40: 206. doi:10.1007/s00384-025-04952-4
3. Lami M, Bhatt DL, Marya NB et al. Use of artificial intelligence improves colonoscopy performance in adenoma detection: a systematic review and meta-analysis. Gastrointest Endosc 2025; 101: 32-43. doi:10.1016/j.gie.2024.08.031