Research Signal – Source date 2021-02-24
A diagnosis code can point to a health problem without showing why it happened. That gap matters when researchers use routine records to count possible side effects of cancer immunotherapy: the same cough or bowel symptoms may have several causes, and the clinical reasoning that distinguishes them may live in notes rather than billing data.
A retrospective study of 1,063 people who received immune checkpoint inhibitors compared treating-team classifications in medical records with searches based on diagnosis codes. Among 174 patients whose charts identified colitis, hepatitis or pneumonitis, 80 had no code evidence reflecting the reaction. Among a separate 748 patients whose charts identified no immune-related adverse event, 459 had codes suggesting a possible one. The two counts show mismatches in both directions; they do not establish a universal error rate for codes or an independently confirmed count of reactions.
What the chart review counted
The reviewers examined patients’ clinical records and classified them as having colitis, hepatitis, pneumonitis, another immune-related adverse event, or no such event. Their assessment drew on the treating team’s documentation, including clinical notes, lab results, imaging and pathology. The team’s reasoning could take account of whether a problem began after immunotherapy, whether pathology supported a diagnosis, whether other causes had been ruled out, and whether the patient improved with treatment directed at an immune-related reaction.
This is a richer record than a diagnosis-code search. A code can identify a documented condition or symptom, but by itself it may not preserve the sequence of events or the clinician’s reasoning about cause. The researchers used chart classifications as their comparison point, so their results describe agreement with those classifications—not an independent determination that every patient was classified correctly.
The 174 patients in the main comparison had one of the three specified reactions: pneumonitis, colitis or hepatitis. The study also classified 159 additional patients as having other immune-related adverse events, while 748 had no such event identified in chart review. Those groups matter: the 459 patients with suggestive codes came from the 748-person no-event group, not from the 174-person group with one of the three specified reactions.
Dates change what a code can mean
Of the 174 patients whose charts identified one of the three reactions, 49 had no diagnosis code reflecting it. Another 31 had a relevant code documented before immunotherapy began. Together, those groups account for the reported 80 of 174 mismatch.
A code that predates treatment cannot, on its own, document a new reaction that began after treatment started. A code recorded later has a different timeline, but its date still does not establish that immunotherapy caused the condition. The researchers grouped codes by whether they appeared before treatment, during the first 89 days after it began, or 90 or more days afterward. This timing helps explain how a code relates to treatment chronology; it does not replace the clinical assessment of what caused a symptom.
Similar symptoms can have different causes
The paper describes why a broad code search can flag conditions that the treating team does not classify as immunotherapy-related. A person with lung cancer may have cough or shortness of breath because of the cancer. A patient who is already unwell or hospitalized may develop pneumonia. These problems can resemble pneumonitis, an inflammation of the lungs that may be related to immunotherapy.
In the paper’s example, a billing record might use a general code for breathing difficulty or acute respiratory failure while clinicians investigate whether pneumonia or immunotherapy-related pneumonitis explains the symptoms. The chart can preserve that investigation and its outcome. A suggestive code alone does not tell the researcher which explanation the team ultimately favored—or whether the condition was caused by treatment.
That ambiguity also affects the meaning of a “positive” code search. Among the 748 patients whose charts identified no immune-related adverse event, 459 had codes that suggested a possible event. The code may point to a symptom or condition worth examining, but the study’s comparison shows that it is not equivalent to a treating-team classification of an immune-related reaction.
Broader code lists catch more possibilities, not more context
The researchers tested a broad set of symptom and disease codes intended to represent possible colitis, hepatitis and pneumonitis. They also examined a narrower set of codes considered more specific to these reactions. A broad list can include more potential manifestations; a narrower list can reduce which records are flagged. But selecting codes cannot recover the clinical narrative that explains when a symptom began, what alternatives were considered, or how the treating team reached its diagnosis.
The comparison was retrospective and relied on records from one study population of cancer patients treated between 2011 and 2017. It therefore cannot tell researchers that the same proportions would apply in other institutions or treatment settings. Nor is chart review a perfect, independently verified reference: the authors discuss challenges in consistently identifying immune-related reactions, including differences in how clinicians interpret symptoms and classify them. The reported mismatch is evidence of a problem for this comparison, not a definitive measure of every code-only search.
How fuller records could sharpen immunotherapy safety research
If clinical-note context can be captured and checked reliably across care settings, routine-record surveillance could detect possible immunotherapy reactions missed by code searches and avoid treating every suggestive code as a treatment reaction. That could help researchers track safety patterns in everyday cancer care with fewer misleading counts. This study did not validate an automated way to capture that context or show that better surveillance improves patient outcomes.
Key Sources
- Primary source: Comparative assessment of manual chart review and ICD claims data in evaluating immunotherapy-related adverse events.
DOI or Journal Page · Review or Meta-Analysis · 2021-02-24
Keep exploring