What is CTCAE and why does it matter?

In oncology, precision is everything. We measure tumors down to the millimeter and titrate drugs down to the milligram. But when it comes to measuring how those drugs make a patient feel or the harm they might be causing, we rely on a different kind of ruler: the Common Terminology Criteria for Adverse Events (CTCAE).

Maintained by the NCI, CTCAE is the standard dictionary for describing adverse events (AEs). It translates a patient's experience ("I have a rash," "I'm exhausted," "My labs look weird") into a standardized language of Grades 1 through 5.

• Grade 1: Mild; asymptomatic or mild symptoms.

• Grade 2: Moderate; minimal, local, or noninvasive intervention indicated.

• Grade 3: Severe or medically significant but not immediately life-threatening.

• Grade 4: Life-threatening consequences; urgent intervention indicated.

• Grade 5: Death related to AE.

Why does this matter?

For the patient, accurate grading dictates their immediate care: whether a dose is held, reduced, or if treatment is stopped entirely. For clinical trials, these grades are the currency of safety data. If a site consistently under-grades toxicity, a dangerous drug might look safe. If they over-grade, a life-saving therapy might be deemed too toxic to approve. In theory, it is a standardized, objective system. In practice, as any research nurse or oncologist knows, it is anything but.

The Reality: Where grading breaks down

If you are an oncologist, a CRC, or a data manager, you know that the "standard" ruler is often bent by human factors, time pressure, and fragmented data. While the definitions in the CTCAE handbook are static, the clinical reality is fluid. The friction occurs in the translation layer: moving from the patient's narrative and the electronic health record (EHR) into the structured Case Report Form (CRF). Here is where the process breaks, introducing risk to patient safety and trial integrity.

1. The Trap of Inter-Rater Variability

Give the same patient chart to three different oncologists, and you might get three different CTCAE grades for a subjective symptom like fatigue or neuropathy. Consider a patient reporting "trouble with daily tasks." • Clinician A interprets this as "Instrumental ADL" (Activities of Daily Living) impact, marking it Grade 2. • Clinician B views it as a "Self-care ADL" limitation because the patient mentioned trouble buttoning a shirt, marking it Grade 3. In a multi-center trial, this variability is a statistical nightmare. It creates noise that obscures the true safety profile of a therapeutic agent, leading to endless query loops between the sponsor and the site to adjudicate the "truth."

2. The Narrative vs. Structured Gap

Physicians think in narratives; databases think in codes. A doctor's progress note might read: "Patient presents with increasing dyspnea on exertion, likely pneumonitis related to immunotherapy, managed with steroids." To the clinician, the story is clear. But for the data manager or the eventual automation tool, this sentence is a minefield. Is the dyspnea the AE? Or is it the pneumonitis? Did the steroids resolve it (implying Grade 2) or was hospitalization required (Grade 3)? The manual burden of mapping these rich, unstructured narratives into rigid drop-down menus is immense. It forces highly trained staff to spend hours acting as data entry clerks rather than clinical analysts.

3. Missed Signals and Drift

In high-volume clinics, "alert fatigue" is real. When a patient has complex comorbidities, distinguishing between a symptom of the disease (e.g., cancer-related pain) and a toxicity of the drug (e.g., drug-induced neuropathy) is cognitively taxing. We frequently see Grade 3+ signals missed simply because they were buried in a PDF lab report scanned from an outside hospital, rather than appearing in the structured lab feed. If the grading workflow relies solely on what is convenient to access, safety signals will be missed.

4. Timing and Context Drift

Grading is not just about severity; it is about attribution and timing. A Grade 3 diarrhea event is critical, but knowing exactly when it started relative to the infusion date is what determines if it is a Dose Limiting Toxicity (DLT). Manual workflows often capture the "what" but smudge the "when," leading to retrospective queries that rely on memory rather than documentation.

A Better Way: The Role of Responsible AI

We cannot fix these issues by simply asking clinicians to work harder or memorize more of the CTCAE dictionary. The cognitive load is already too high. This is where purpose-built technology enters the equation. The goal is not to replace the oncologist. The goal is to create a Clinician-in-the-Loop architecture that acts as a highly skilled safety net.

How the workflow evolves

We move from a model of manual data hunting to one of verified review. The goal is not to automate the physician out of the loop, but to automate the rote work of gathering context.

The Strategic Advantage

• Defensible Data: Decisions are documented and traceable, moving away from "black box" adjudication to audit-ready outputs.

• Enterprise-Grade Privacy: The architecture is designed to meet strict compliance standards without exposing sensitive PHI to open models.

• Standardization: It establishes a consistent baseline across sites, reducing the variability that typically plagues multi-center trials.

Moving toward a verified workflow

The future of oncology data isn't about removing human judgment; it's about empowering it with better tools. By automating the extraction and mapping of toxicity data, we free up clinical teams to focus on what matters most: interpreting the data and caring for the patient.

Are you looking to reduce query burdens and improve safety signal detection? We are currently working with select oncology partners to refine this workflow. If you're interested in seeing how assisted grading works in practice, we'd love to show you.

Marc Saint-jour, MD

Chief Medical Officer

BurnaAI

marc@burna.ai