Read a single success story or a calculator

One person's remarkable outcome shows that something is possible, not how often it happens or what caused it. A calculator turns population data into one number, which is only as good as its inputs and validation.

Before you start: Case reports and N-of-1 studies explains what a single timeline can support. Population versus individual risk explains what a model's probability means.

Where this step sits

This is the last step of Read trial claims and survival statistics. Lessons 1–3 read comparisons between groups. This lesson reads the two things families meet most often outside a trial: a story and a calculator.

What a story can and cannot establish

A case report, a peer's journey or a company's patient testimonial is a timeline: what was given, in what order, and what was measured. It can establish that an outcome occurred and suggest a question worth testing. It usually cannot establish three things:

1. Cause. Other treatments, the disease's own course and the timing of tests can all explain the change.
2. Frequency. Remarkable outcomes are more likely to be shared and published. Without the number who tried the same thing, one success says nothing about the chance of success.
3. Comparison. Nobody knows what would have happened to that person on another path.

Ask the denominator question every time: how many people received this, under what conditions, and what happened to all of them? A story's value rises when it comes with a full timeline, a measurement at more than one time and, ideally, a link to a study that counted everyone.

A worked story

A fictional blog describes a person whose metastatic tumor shrank after an off-label drug added to standard chemotherapy. The post includes before-and-after scan reports.

• Established: the tumor shrank during the combined course, on those scans.
• Not established: that the added drug caused it, how often it works, or whether it is safe.
• Next question: Is there a trial or case series that counted everyone given this combination, with a comparison?

Check a calculator before its number

A calculator can make uncertain inputs look settled. Population versus individual risk explains what a model can estimate and what a probability cannot promise. Keep this record beside its output:

• Identity: exact model, software version and any recalibration; a familiar name is insufficient.
• Inputs: required values, units, definitions, specimen dates and the model's documented handling of missing values. Unknown is not automatically negative.
• Setting: treatment era, received treatments and intended population. Older treatment context needs an applicability check.
• Question: endpoint, time origin and prediction horizon.
• Validation: independent evaluation in relevant populations. Calibration checks probabilities; discrimination checks ranking. Retain the cutpoint and how it was chosen when a score becomes a category or proposed action.

TRIPOD+AI describes the reporting needed to inspect prediction models. It covers regression and machine-learning methods. A checklist or published calculator is not proof of usefulness for every setting. Do not adjust its output by multiplying unrelated prognostic hazard ratios or adding benefits from other trials.

What can go wrong at this step

• "It worked for her, so it will work for me." One outcome is not a frequency or a personal forecast.
• "Many people online say it helped." A collection of successes without the failures has no denominator.
• "The calculator says 85%, so that is my chance." The number is an estimate for people like those in the model's data, under its inputs.
• "I'll adjust the calculator for my extra treatment." Multiplying in hazard ratios from other trials builds a model nobody validated.

Try it

A fictional company video features three patients whose cancer disappeared on its therapy. What single question most changes how much the video tells you?

Answer: How many patients received the therapy in total, and what happened to all of them? Three successes out of 5 and three out of 500 mean very different things.

Calculator card: A fictional model requires a residual-tissue marker after surgery. Only a pretreatment biopsy result is available. Can the reader substitute it?

Answer: No. Check the specified input and missing-value rules. An earlier specimen does not supply the required postoperative value, and an improvised hazard-ratio adjustment cannot fix that mismatch.

Explain it back

"A single story can show ___, but it cannot show ___ without ___."

One possible answer: "That an outcome is possible; how often it happens or what caused it; a count of everyone treated and a comparison."

Takeaway

Ask for the denominator behind every story and the validation behind every calculator.

Next: Apply the framework to a test or vendor claim, or practise on cell-therapy evidence.

Sources and scope

Source-checked November 7, 2026. General evidence appraisal; expert and learner review pending. All stories, models and numbers are fictional.

• Gagnier and colleagues 2013: consensus case-report guidelines — timeline and attribution reporting.
• Collins and colleagues 2024, TRIPOD+AI: reporting prediction-model inputs, outcomes and evaluation.