Hand a group of men in their late fifties and sixties a standard prostate biopsy report and ask them one question: does this say I have cancer? In a randomized study of 2,238 adults published in JAMA in early 2025, 39 percent of those who read a standard academic report answered correctly (1). Handed a patient-centered version containing the identical findings, 93 percent got it right.

The reflex is to call this a health literacy problem, but that puts the problem in the wrong place. The pathology report was written by physicians for physicians, using a vocabulary developed over a century to communicate precisely with surgeons, oncologists, tumor registrars, and other pathologists. Nobody drafting the phrase "prostatic adenocarcinoma, Gleason 3+3=6, involving 15 percent of one core" expected the patient to encounter those words before speaking with a clinician. For most of the history of our specialty, that assumption was reasonable because patients rarely saw the report directly.

In the US, the information blocking provision of the 21st Century Cures Act changed that arrangement almost overnight. Beginning in April 2021, results that had traditionally passed from pathologist to treating clinician and then to patient became available immediately through electronic portals. At one large cancer center, three quarters of results were being viewed by patients before the ordering clinician had opened them, nearly double the proportion before implementation of the Cures Act. The audience for the pathology report changed by regulation, while the document itself remained essentially the same.

We have spent much of the period since debating whether pathologists should become more visible to patients, even as patients have acquired direct access to the work we produce.

When patients read their reports

Some of the most useful data on what this experience means for patients arrived this summer. Sheena Bhalla and colleagues at UT Southwestern surveyed 2,412 people diagnosed with cancer in 2024 (2). Most had learned their diagnosis from a clinician, whether in person, by video, or by telephone – but 7 percent learned it from the portal. Among that group, 71 percent were at home when they found out and 59 percent were alone. Nearly half went directly to the internet, and more than a third of them spent five hours or longer searching before contacting anyone on their care team.

Five hours is a substantial interval to spend alone with a cancer diagnosis that one may not understand. The anguish only increases when the document providing the diagnosis was never designed to explain itself to the person reading it. During that period, the pathology report may be the only communication from the health system available to the patient, which means that the clinician's intention to communicate directly with the patient has become largely irrelevant.

The same study also makes clear that patients do not all want the same thing. Seventy-five percent of the cohort said they would prefer to hear a cancer diagnosis from a clinician. Yet, among those who had actually received their diagnosis through the portal, a slight majority said they would choose that route again. Patients value access to their information, while many also want the opportunity to understand what that information means. Several states, including California, Kentucky, and Texas have legislated around this tension by permitting delayed portal release of some cancer-related results. But delaying a report does not address what happens when the patient eventually reads it.

Artificial intelligence and the pathology report

When patients cannot readily reach us and cannot understand the language we use, they increasingly turn to tools that will interpret it for them. NPR reported last September on patients taking screenshots of MyChart results and submitting them to ChatGPT, Claude, or Gemini while waiting to hear from their physicians (3). A 2024 KFF poll found that 56 percent of people who use artificial intelligence had little confidence that what a chatbot tells them about health is accurate (4). Yet, the behavior persists because an immediate explanation, even one regarded cautiously, may be preferable to waiting without one.

The pathology report was written for one reader. It now has three: the clinician it was addressed to, the patient who often opens it first, and the language model asked to explain it to them. That third reader is already at work outside the governance of pathology departments, health systems, professional societies, and, in many cases, the clinicians caring for the patient.

The evidence on how well artificial intelligence performs this function is sufficiently mature to make its potential and limitations clear. A prospective controlled trial published in Radiology in November 2025 gave 200 patients either standard oncologic CT reports or versions simplified by a large language model with mandatory radiologist review (5).

With the simplified reports, comprehension improved sharply, with an adjusted odds ratio above 13, cognitive burden fell, and reading level decreased from approximately thirteenth grade to ninth grade. However, despite physician review, factual errors remained in 6 percent of simplified reports and content omissions in 7 percent, with approximately 4 percent of each judged severe.

A narrative review of 49 studies of large language models in radiologist–patient communication similarly found improvements in readability of two to six grade levels, while professional review was required in as many as 80 percent of outputs in controlled settings, against fewer than 10 percent in observational ones (6).

These findings support a model in which artificial intelligence can help translate medical information for patients, provided that responsibility for the resulting communication remains with a clinician. The difficulty is that patients are already using the same technology without that supervision, often precisely because a clinician is not available when the result arrives.

Real-world solutions

Different organizations have begun responding in different ways. Stanford Health Care incorporated the model into the clinician workflow in January 2025, generating plain-language interpretations that physicians review before patients see them (7). In March 2026, Quest Diagnostics introduced an AI Companion that allows patients to request explanations of up to five years of laboratory results (8). One approach keeps interpretation under clinical authorship; the other makes it a consumer product. Both respond to a need that medicine has left unmet.

Pathology has an additional complication because the most machine-tractable documents our specialty produces are the synoptic cancer reports built on CAP protocols refined over more than thirty-five years (9). CAP appropriately protects that intellectual property and restricts incorporation of customized or derivative versions into artificial intelligence systems without written authorization. The practical consequence is that the structured content most amenable to a safe and auditable patient-facing explanation requires deliberate agreement and governance. Meanwhile, general-purpose language models can already interpret our free-text diagnoses without either. That mismatch deserves attention if we want interpretation of pathology reports to remain connected to the expertise that produced them.

Pathology explanation clinics

The human version of this work is older and better studied than many pathologists realize. Lija Joseph developed a structured pathology consultation program after a patient asked to see what her cancer looked like, and by 2019 her department at Lowell General Hospital was running a weekly pathology clinic. The first paper from that program, with Adam Booth as lead author, took its title from a patient's words: "Please help me see the dragon I am slaying"(10). A subsequent four-institution quality improvement study led by Rachel Jug and Thomas Cummings found that all 67 participating patients were satisfied, found the visit useful, and would recommend it.

More importantly, these encounters can affect care. In a series of 59 oncology patients who met with their pathologist at Tel Aviv Sourasky Medical Center, half were referred for additional workup, 42 percent were counseled toward genetic sequencing, one in five had their case sent for internal pathology revision, and three patients had their treatment plan changed (11).

In my own department, Cathryn Lapedis and colleagues studied a pathology explanation clinic for men with newly diagnosed localized prostate cancer, interviewing ten patients before the visit, immediately afterward, and again at one and six months (12). The pathologist reviewed the slides with the patient and then documented the discussion and remaining questions for the urologic oncologist. Oncologists reported that these patients arrived better prepared and asked more focused questions about prognosis and treatment. Outside biopsy slides were also re-reviewed, diagnostic changes could be communicated across the clinical team, and in several cases the discussion supported a patient's consideration of active surveillance rather than treatment.

Patient interest is not the obstacle. An earlier Michigan survey found that 85 percent of patients with cancer were interested in meeting the pathologist who made their diagnosis (13). When the same group subsequently asked treating clinicians about pathology explanation clinics, a recurring perception was that relatively few patients would want such a service. The limited uptake of patient-facing pathology reflects what clinicians assume patients want rather than what patients themselves report.

The infrastructure is beginning to catch up. ASCP has released a Pathology Clinics Certification Program addressing the administrative, billing, and communication requirements of establishing a clinic (14). Michele Mitchell, an ASCP Patient Champion and breast cancer survivor who never saw her own pathology report when she was diagnosed in 2006, proposed the program and contributed to several of its sections, including the module addressing the Cures Act. Her involvement is important because patient-facing pathology should not be designed entirely by pathologists deciding among ourselves how patients ought to receive information.

The workforce constraint

Any discussion of expanding direct patient communication has to contend with the pathology workforce. HRSA projects a 7 percent decline in pathologist supply against a 16 percent increase in demand by 2037, with particularly severe shortages projected in rural areas, and ASCP estimates that the United States will need approximately 3,000 additional pathologists over that period (15). A model that simply adds patient consultations to already full clinical workloads will not scale regardless of how valuable individual encounters may be.

Universal pathologist–patient consultation is neither necessary nor realistic. Published experience suggests that demand is concentrated among patients with newly diagnosed cancers, ambiguous or borderline diagnoses, particularly complex reports, substantial distress, or cases in which understanding the pathology may materially influence a clinical decision. A workable system would identify the patients most likely to benefit and create a route for them to reach pathology without requiring every report to generate a separate encounter.

The barriers described in the Jug study are instructive because they are largely operational. Electronic records did not allow patients to message pathologists, scheduling systems lacked a mechanism for pathology appointments, and billing processes were poorly defined. These problems are less intellectually interesting than another position paper about the importance of patient engagement, but solving them is more likely to determine whether patient-facing pathology becomes routine practice.

The role of digital pathology

Digitization changes the scale at which this work can occur. Once a slide becomes a digital file, the same image can be available to the pathologist, the tumor board, the treating clinician, and the patient without moving the physical specimen. Computational pathology has also moved into regulated clinical practice, with the FDA granting De Novo authorization in August 2025 to ArteraAI Prostate (16). This represents the first AI-powered software cleared to prognosticate long-term outcomes in non-metastatic prostate cancer. Primary diagnosis in most American laboratories still occurs on glass, but the transition remains early enough for us to decide how patient communication fits within it.

Making the pathology report understandable is the most immediate opportunity, but digital pathology allows us to go further by showing patients the evidence behind the diagnosis. Pathologists can show a patient the tumor, the margin, the involved lymph node, or the difference between normal and abnormal tissue rather than relying entirely on textual descriptions. Michele Mitchell printed the digital image of her own invasive ductal carcinoma and keeps it on her bedside table. In another reported encounter, a patient with atypical hyperplasia sat at a double-headed microscope and watched her own cells change across six years of biopsies. Those experiences demonstrate something pathology can contribute that no rewritten report or chatbot can fully reproduce: direct access to the tissue on which the diagnosis was made.

Patient-facing pathology is sometimes treated as a service that well-resourced academic centers can provide once the diagnostic work is finished. My experience with diagnostic capacity in low- and middle-income settings makes me think the argument runs in the opposite direction. Where access to pathologists and specialists is most limited, the pathology report may be the only durable artifact of the diagnostic encounter that a patient carries forward. The fewer opportunities a patient has to obtain subsequent explanation, the more important it becomes that the original document can be understood.

For most of our history, the pathology report worked because another clinician completed the communication process. We made the diagnosis and relied on someone else to explain it to the patient. Immediate portal access has disrupted that sequence, and artificial intelligence is rapidly occupying the space between receiving a result and speaking with a clinician.

Patients are already reading our reports, including reports we never wrote for them, and when they cannot understand what they read they are increasingly asking a machine to interpret our words. We can continue to regard what happens after the report is released as someone else's part of the diagnostic process, but that distinction is becoming difficult to defend. If the pathology report is our work, then making sure the patient can understand what we have said is part of our work as well.

Newsletters

Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.

I have read and understand the Privacy Notice*

Subscribe