When Two Diagnoses Disagree, Compare the Reasoning

October 30, 2026 | The Psychiatric Record | Clinical Assessment

Two diagnoses in the same record create a practical problem. Before changing a label or defending the earlier one, the next clinician needs to know where the accounts diverge. Did both assessments describe the same experience? Did they cover the same period? Did they have the same information?

A recent argument about diagnostic reliability makes those questions timely. The memorable number is about 55% agreement. Its meaning depends on the task that produced it.

In a paper published in July 2026, Boberg and colleagues reported 1,902 assessments from 1,038 doctors working in adult psychiatry across 19 countries. Participants assigned ICD-10 diagnoses to written vignettes drawn from nine cases. The sample included psychiatrists, residents and other doctors. They selected from 30 diagnostic options; this was not a study of two clinicians interviewing the same patients in routine practice. [1]

Overall chance-adjusted reliability was Krippendorff's alpha 0.48. Observed pairwise agreement was about 55%. These are different measures: alpha 0.48 does not mean that 48% of diagnoses were correct. The paper separately reported 66% concordance with its reference diagnoses, which the authors described as informed comparisons rather than definitive ground truth. Agreement asks whether raters give the same answer. Reference concordance asks whether they match a specified answer. [1]

Accuracy was not spread evenly. Concordance with the reference diagnoses was 37% to 54% for the four schizophrenia-spectrum cases and 88% to 94% for the bipolar, depressive and obsessive-compulsive cases, so the overall 66% partly reflects which cases were chosen. [1]

The voluntary sample, selected cases and written format limit generalization, and the study cannot supply a universal error rate for psychiatric practice. The authors argue the result is more likely an upper bound: every doctor read identical, preselected information, and they conclude that agreement in routine settings is likely to be lower. They add that grouping related diagnoses and limiting the choice to 30 options may have raised agreement. That is their interpretation; the study did not measure live assessments. [1]

Standardizing an interview does not make every reliability question disappear. Xie and colleagues' 2026 systematic review included 57 studies of standardized adult diagnostic interviews. Its quantitative meta-analysis used 46 articles and found pooled test-retest reliability of kappa 0.69 across mental and substance use disorders (0.65 for mental disorders alone), with substantial heterogeneity. [2]

That review concerns consistency across repeated interviews. The vignette study concerns agreement among doctors given written material. The estimates cannot be read as a contest between two assessment methods, and kappa 0.69 is not a 69% accuracy rate. The review supports care in choosing and interpreting instruments; it does not establish that an unstructured interview is superior. [2]

For a clinician facing conflicting diagnoses, the immediate task is more specific than deciding whether psychiatry as a whole is reliable. It is to identify what would explain the disagreement in this record.

Consider three questions when reviewing a diagnostic change. These are editorial prompts for examining the record, not a validated intervention shown to improve agreement.

Are the clinicians describing the same phenomenon? A shorthand word such as "paranoid" may conceal the detail that needs examination. Rather than compare labels alone, look for what the person described, the clinician's observations and the reasoning connecting those findings to the diagnosis. If that account is missing, the next step may be further inquiry rather than a vote between two labels.

Are the accounts about the same period? One assessment may address an acute presentation and another the person's longer course. The useful comparison identifies when the relevant experiences began, what changed and which information was available at each assessment. A changed diagnosis deserves an explanation of what changed in the evidence or its interpretation.

What information could distinguish the competing accounts? Name the unresolved question before ordering another assessment or requesting collateral. "Clarify the timeline of the reported experiences" gives the next clinician a task. "Please clarify diagnosis" leaves the disagreement largely intact. The appropriate inquiry still depends on the individual, consent, urgency and setting.

In the first-episode psychosis section of NICE's CG178, recommendation 1.3.3.1 calls for a comprehensive multidisciplinary assessment of people with psychotic symptoms in secondary care. Its named domains include psychiatric (with risk and alcohol and drug history), medical, developmental, psychological and psychosocial (including trauma), occupational and educational, and quality of life. This is UK guidance for that population and setting, not proof that a particular documentation template resolves disagreement. [3]

A useful reconciliation should make the current conclusion understandable to the patient and to the next clinician. It can describe the findings supporting the working diagnosis, the plausible alternative and the information still needed. Where uncertainty changes today's plan, explain how. Where it does not, explain that too.

That does not require pretending that every disagreement can be settled immediately. Nor should unresolved classification postpone assessment of urgent safety or medical concerns. The record can distinguish the question still being investigated from the needs being addressed now.

If a patient brings the 55% headline to a visit, the answer need not dismiss the study or apply its estimate to their personal chances of being misdiagnosed. Explain the written-case task and its limits, then return to their own assessment: what supports the current view, what remains uncertain and what would lead to reconsideration.

The next consultation should leave more than a replacement code. It should leave a reason the current diagnosis is being used and a clear question for what happens next.

Educational Disclaimer: The Psychiatric Record provides general educational information for psychiatric and mental-health professionals. Content does not constitute medical, legal, regulatory, compliance, billing, or other professional advice; does not establish a standard of care; and is not a substitute for independent professional judgment. Appropriate assessment and treatment depend on the individual, setting, and current authoritative guidance. Patients should coordinate treatment decisions with their treating clinicians.