Educational content only. This module is not CME, not accredited, not medical advice, and not patient-specific guidance. The Certificate of Completion is not a professional credential.

Module 1 - The Oncology AI Landscape

Level: Beginner Time to complete: About 10 minutes

What you should know by the end

  • What people usually mean when they say "AI" in oncology.
  • The most common places AI shows up in a cancer practice.
  • The three questions to ask before you trust an AI output.

Big idea

AI is software that finds patterns, creates drafts, or helps move work through a process. It is not magic. It is not a clinician. It does not know whether it is right in the way a trained person knows.

Most AI tools used today are narrow tools. They are built for a specific task, such as drafting a note, flagging an image, matching a patient to a trial, or summarizing a document. A tool that is good at one task may be unsafe or useless for another task.

For a beginner, the first skill is not model science. The first skill is naming what the tool is doing.

Common AI categories in oncology

1. Documentation and communication

These tools listen, transcribe, summarize, or draft notes and messages. The output should be treated as a draft until a responsible person reviews it.

2. Imaging and pathology support

These tools look for patterns in scans, pathology slides, or other clinical data. They may flag a finding or measure something. Because errors can affect diagnosis or treatment, these tools need strong validation and clear human review.

3. Trial matching

These tools compare patient information with trial criteria. They can help find trials that might otherwise be missed. They can also miss eligible patients if the chart data is incomplete or hard to read.

4. Administrative automation

These tools help with prior authorization, referrals, forms, scheduling, or appeals. The work sounds administrative, but delays and mistakes can affect care.

5. Knowledge retrieval

These tools search or summarize guidelines, policies, articles, or internal documents. They are useful when they show sources. They are risky when they give confident answers without showing where the answer came from.

6. Workflow agents

Agents do more than answer a question. They may take several steps, such as collecting information, drafting a form, sending a message, or updating a task list. Agents need more guardrails because one early mistake can affect later steps.

Three beginner questions

When you see any AI tool, ask:

  1. What information goes in? Is it a note, image, lab result, message, claim, or patient record?
  2. What comes out? Is it a draft, a flag, a recommendation, a score, or an action?
  3. Who checks it before it matters? Is there a clinician, staff member, supervisor, or automated stop?

These questions are simple, but they prevent many bad decisions. A draft note and a treatment recommendation should not be governed the same way.

Simple oncology example

A tool summarizes a long oncology visit and drafts patient instructions.

  • The input is visit audio or text.
  • The output is a draft summary.
  • The reviewer should be a clinician or trained team member before it is sent or placed in the chart.

The tool may save time. It may also omit an important detail, misunderstand a drug name, or write something more certain than the visit allowed. The output is useful only if review is real.

Remember

  • Start by naming the task, not the technology.
  • Most current AI is narrow and task-specific.
  • Drafts, flags, recommendations, and actions carry different levels of risk.
  • The safest beginner question is: "Who reviews this before it affects a patient?"
← Back to course overviewCourse enrollment