Causation questions: AI finds the evidence, you form the opinion
Short version: use a record tool to pin down what the patient's condition was at each decision point, and a literature tool (Consensus, Elicit, OpenEvidence, then PubMed) to find outcome data for earlier versus later treatment. Read every paper yourself before you cite it. The causation opinion, stated to whatever level of medical probability your jurisdiction requires, is yours alone.
Limits: We read each product's own web pages on 2026-11-12 and did not upload records to any of them, ours included. Prices, plans and features change, and nothing here is legal advice or a substitute for your own clinical judgment.
The causation question, split into parts AI can help with
A causation opinion in a delay case usually answers 1 question: if the right thing had been done at the time it should have been, would the outcome more likely than not have been different? That splits into parts. Some are record facts. Some are literature. 1 is judgment.
| Part | Type | Best tool | Your check |
|---|---|---|---|
| What was the patient's state at the point care should have changed? | Record fact | Page-cited record tool: vitals, labs, exam at that hour | Read the flowsheet and notes for that hour yourself |
| What would the right action have been, and how fast would it have taken effect? | Clinical knowledge | Your training; a literature tool for the time course | Specialty and setting match the case |
| How do outcomes differ by time to treatment for this condition? | Literature | Consensus, Elicit, OpenEvidence, PubMed | Every paper exists, is read in full, and fits the patient |
| Did something else cause or contribute to the harm? | Record fact plus judgment | Record tool: comorbidities, prior imaging, other providers | Pre-existing disease and its course before the event |
| Does the difference reach the required degree of probability? | Judgment | None | Your opinion, in the standard your jurisdiction uses |
Literature tools for causation, as of 2026-11-12
These cite the papers behind their answers, which is the minimum for causation work. What each maker says, read on 2026-11-12:
| Tool | What the maker says | Use it for |
|---|---|---|
| Consensus | Searches 250M+ research papers; can narrow to about 50,000 clinical guidelines and 8M articles from top medical journals | A first pass on "does time to treatment change outcome for X?" |
| Elicit | Searches 138 million papers and 545,000 clinical trials; sentence-level citations | Pulling effect sizes and study details into a table you can check |
| OpenEvidence | Evidence answers for clinicians; says covered entities may send PHI under its standard BAA | Quick clinical framing with references |
| ChatGPT for Clinicians | Free for verified US clinicians; clinical search with citations; individual BAA available | Framing the question and drafting search terms |
| PubMed | The NLM's free index of biomedical literature | Confirming every paper exists and reading the abstract and full text |
Source links and capture dates for each are on the tools page. None of them knows your patient; describe the clinical question without identifiers.
Check every citation before it enters a report
Chatbots invent references. A 2023 study asked ChatGPT for 84 short literature reviews and checked all 636 citations: 55% of the GPT-3.5 citations and 18% of the GPT-4 citations were fabricated, and many real ones carried errors (Walters and Wilder, Scientific Reports 2023). Tools that search a paper index first are built to avoid this, but a wrong page number or a misread finding still gets through. Run this check on every paper:
- Find it in PubMed or on the publisher's site by title, then by DOI.
- Read the abstract and the results table that the AI summary relied on.
- Confirm the population matches your patient: age, setting, severity, era of treatment.
- Confirm the paper existed before your opinion date, and note whether it post-dates the care. Causation can draw on later science; the standard of care cannot.
- Keep a copy in your file.
Causation words differ by jurisdiction
"More likely than not", "reasonable degree of medical probability", "reasonable medical certainty", proximate cause and loss of chance carry different meanings in different states. A proximate cause is an actual cause that is also legally enough to support liability (Cornell LII, proximate cause). Ask the retaining attorney which causation standard applies, and whether your state treats a lost chance of a better outcome as its own claim, before you write. Do not ask an AI tool to choose the standard for you; a general model will blend the rules of several states into 1 answer.
Prompts that keep causation questions answerable
- "Summarize outcome data for [condition] by time from [finding] to [treatment], in adults in [setting]. List each study with year, design, size and the outcome measure."
- "Which studies report outcomes separately for patients who were [state at decision point]?"
- "What is known about the natural course of [condition] without treatment over [hours]?"
- To the record tool: "List every finding between [time] and [time] that bears on [condition] severity, with page."
Never paste the opinion you want and ask the tool to support it. That is the fastest route to a cited paper that says something else.