Walk into any tumor board and you can watch the problem resolve itself for one hour a week. A radiologist, a pathologist, an oncologist, and a surgeon each hold a piece of the same patient. It is one of the most valuable hours a hospital spends, and one of the few moments when the system concedes that a single physician cannot see an entire case.

Outside that room, we still build as though one can.

Diagnostic error reveals the limits of that assumption. The latest national estimate suggests that diagnostic errors may contribute to roughly 909,000 deaths or cases of permanent disability in the United States each year, although the figure is modeled rather than directly counted. A recent review of hospitalized patients who died or required intensive care also found a diagnostic error in 23 percent of cases.

The scale cannot be explained by individual competence alone. Diagnosis now depends on information and expertise distributed across clinicians, tests, technologies, and care settings, while responsibility for assembling them still often falls to one physician.

Medicine built its operating model around individual clinical judgment

Modern medicine took shape when a physician could still obtain much of the available clinical evidence directly. The doctor took the history, examined the patient, formed a diagnosis, and selected from a relatively narrow range of treatments. Even as hospitals and supporting professions grew, professional authority remained concentrated in the individual physician.

For decades medicine managed growing complexity through specialization. AI changes the equation again. Every new model becomes another contributor whose findings must be evaluated alongside human expertise. The problem is no longer just coordinating specialists. It is coordinating specialists, algorithms, historical data, and clinical context inside one decision.

That model matched the medicine available more than a century ago. Before laboratory science, medical imaging, molecular diagnostics, and hundreds of specialties, diagnosis relied mainly on observation, physical examination, and a small set of instruments. The physician’s advantage lay in personally knowing and interpreting more than the people around them.

Medical science then expanded beyond that model. By the late nineteenth century, physicians already viewed specialization as necessary because no individual could master the profession’s growing domains. Since then, medicine has added layers of tests, evidence, technologies, and specialist expertise.

Specialization turned diagnosis into collective work

Modern diagnosis rarely comes from one source. A suspected cancer may require laboratory tests, imaging, and biopsy to identify and characterize the disease. Different specialists then interpret each layer and decide what the combined findings mean for treatment.

That pattern extends beyond oncology. The National Academies describes diagnosis as a dynamic, team-based process involving treating clinicians, radiologists, pathologists, nurses, patients, and other contributors. It specifically calls for closer collaboration among pathologists, radiologists, other diagnosticians, and treating clinicians throughout test selection, interpretation, communication, and subsequent decisions.

Medicine’s collective knowledge keeps expanding, while each clinician necessarily commands only part of it. The bottleneck now lies in finding the right expertise, connecting it to the patient, and synthesizing the findings in time for a decision.

Other complex fields built coordination into the work

Software development, aviation, engineering, and large-scale science each abandoned the idea that one person can hold an entire system. Aviation makes the clearest case. After investigators traced crashes to breakdowns in crew communication rather than deficits in flying skill, the industry built crew resource management: structured coordination, explicit handoffs, and a shared picture no individual owns alone. Pilots did not become less skilled. The complexity outgrew the boundaries of solo work.

Medicine cannot copy that model directly. Clinical decisions unfold across longer periods, several organizations, and professionals governed by different scopes of practice. In many team-based care models, a physician also retains final authority for diagnosis or treatment, so wider collaboration cannot leave accountability unclear.

Medicine is trying to coordinate collaborative care

Medicine already uses several forms of coordinated expertise. Formal referrals, second reads, multidisciplinary tumor boards, and diagnostic management teams bring additional specialists into a case. Informal curbside consultations serve a similar need, although they often rely on incomplete records and may leave little documentation.

Some of these approaches produce promising but uneven results. A systematic review of breast cancer tumor boards found that multidisciplinary review changed diagnoses in all four studies that measured diagnostic impact and changed treatment plans in every study that assessed treatment decisions. Its pooled analysis also associated tumor board participation with a 14 percent lower mortality risk. A 2023 meta-analysis of multidisciplinary care for non-small cell lung cancer discovered that patients received treatment about 12 days sooner, were 36 percent more likely to receive complete staging, and had better overall survival.

Evidence outside oncology points in the same direction. In a 2024 study covering 257,000 patient encounters, cases reviewed by a coagulation diagnostic management team were six times more likely to receive an established, scientifically supported diagnosis than cases reviewed without one.

The evidence is not uniform. A 2026 systematic review of ninety-seven tumor board studies found that multidisciplinary teams can improve coordination and support more systematic, evidence-based discussion. But studies used inconsistent quality measures, many hospitals faced time and staffing constraints, and few studies could isolate the tumor board’s effect from other factors that may have influenced patient outcomes.

These models show that medicine can coordinate distributed expertise, but results depend on how each structure is designed, staffed, and integrated into clinical work. They remain partial solutions rather than a reliable system-wide capability.

Medicine needs to make collective expertise routine

Medicine has already shown that structured collaboration can improve clinical decisions. The remaining problem is scale.

Tumor boards, diagnostic management teams, second reads, and specialist review still tend to appear only in selected specialties, institutions, or scheduled meetings. Complex cases outside those settings often depend on ad hoc calls, referrals, and individual effort.

The next step is to make coordination a standard clinical capability rather than an exception. Systems should help clinicians identify the right expertise, preserve context across handoffs, compare interpretations, and reach a decision before delay creates harm.

Medicine will continue to produce more knowledge. The harder task is making that knowledge work together around the patient.

Yaroslav Dokuchaev is the founder and CEO of DICO, a company developing collaborative infrastructure for radiology and diagnostics assisted by artificial intelligence (AI), and an independent researcher in radiology workflow innovation and health care AI. His work focuses on how health care workflows are evolving as AI becomes an active participant in clinical decision-making.

Rather than viewing AI as a replacement for physicians, he explores how distributed expertise, workflow orchestration, and transparent collaboration can improve diagnostic quality and reduce cognitive burden on clinicians. He is also the creator of Atlas of Radiology, a free interactive CT and MRI anatomy platform built from real Digital Imaging and Communications in Medicine (DICOM) studies.

His writing focuses on radiology, health care AI, diagnostic systems, and the future of collaborative medicine, with work appearing in Unite.AI, HackerNoon, The European Business Review, and Grit Daily. A representative sample includes essays on the structural workflow bottlenecks limiting radiology AI, the radiology shortage as a coordination problem, and the diagnostic reasoning work that comes after DICOM. He shares updates on LinkedIn.