About Worth the Test

Worth the Test is written and maintained by one person: a physician who practices emergency medicine. It is published without a name, an employer, or a city attached, and that is a deliberate choice. Nothing here asks you to trust an author. Every result is either a calculation you can reproduce by hand or a figure linked to a source you can open and check for yourself, so the identity behind it should not decide whether you believe it. Keeping the byline empty keeps the attention on the arithmetic instead of on a résumé.

Why this site exists

A test result rarely means what it first appears to mean. Tell someone that a screening test is "90 percent accurate" and that they have tested positive, and most people conclude that the person almost certainly has the condition. For a rare condition that conclusion is usually wrong: the healthy group is so much larger than the sick group that false alarms can outnumber true positives several times over. Overlooking how common a condition is before the test — its base rate — is the base-rate fallacy, and it is not a beginner's error. When researchers put the classic version of the problem to physicians and staff at a teaching hospital, most answers were wrong by more than an order of magnitude (Casscells, 1978); a replication a generation later found the same mistake alive and well (Manrai, 2014).

The calculator exists to make that hidden arithmetic visible. You set a population, how common a condition is, how good the test is, and how much a downstream treatment helps or harms — then watch a single cohort move from population to test to treatment to outcome, with true and false positives, predictive value, and the count of people helped, harmed, or left unchanged all updating together. There was no free, interactive tool that laid out that whole chain — benefit and harm side by side — which is why this one exists. The worked example is a good place to start: a 1% condition and a 90%/90% test turn 108 positive results into only 9 real ones.

Editorial principles

These are the rules the whole project is built on.

• Numbers are cited, not invented. Every example figure offered beside a control links to a primary source — a task-force recommendation, a Cochrane review, or a clinical trial. Test-accuracy figures are taken from that literature and re-checked against it; treatment estimates that are weaker or indirect are labeled as such rather than dressed up as firm. How each figure was chosen and verified is written out in the methodology.
• The tool takes no side. It is argument-agnostic. A benefit is never shown without the matching harm on the same denominator, and the site never tells you to get, or to skip, any particular test. What counts as a worthwhile trade-off is a judgment for you and your clinician; the site supplies only the counts.
• Simplifying assumptions are labeled. A teaching model leaves things out. Where it does — treating "helped" and "harmed" as separate groups, or treating repeated tests as statistically independent — the simplification is stated plainly next to the result it affects, so a convenience is never mistaken for a claim about the world. The glossary defines the terms as the tool uses them. This is an educational model, not medical advice, and it can say nothing about your own case.

How the math is tested

The statistical engine is ordinary, testable code kept entirely separate from the interface: a set of pure functions that take numbers in and return numbers out, with no dependence on the page around them. That separation is what makes it checkable, and the formulas and assumptions behind it are documented in the methodology. The engine ships with 39 automated tests. Some pin exact results — the canonical 1%-prevalence, 90%-by-90% example must always reproduce the same counts down to the person. Others are property-based "fuzz" tests: they generate thousands of random scenarios — six thousand for the main cascade alone — and assert that the invariants hold every time, for instance that the buckets always sum back to the population and that no combination of inputs can leave more people helped-plus-harmed than were ever treated. A change that broke one of these would fail the tests before it could reach the site.

Privacy and funding

The site currently carries no advertising, and the default build loads no third-party advertising or tracking scripts at all — no ad code, no analytics library that profiles you. The typefaces are self-hosted, so reading a page does not call out to a font network, and visits are counted with a single cookieless analytics beacon that stores nothing in your browser and does not follow you across other sites. The tool needs no account and sends nothing you type to a server — every calculation runs in your own browser. If advertising is ever enabled it will be disclosed plainly, and the privacy policy governs exactly what data is and is not collected.

Contact

Corrections are welcome and taken seriously; a factual error is the first thing worth fixing. If a number looks wrong, a link has moved, or a source is being read incorrectly, email [email protected] with the specifics. To keep the math on this site honest, that kind of message is prioritized above everything else.