How this check works

What it does

You answer 16 dropdown questions — age, sex, and 14 symptoms people commonly report. The tool compares your pattern against 251 unique patient profiles from a diabetes-hospital questionnaire (520 rows with exact copies removed) where the diagnosis is known, and tells you how closely your answers match the patients who had diabetes.

It's a pattern-matcher, not a doctor: it can't look at you, run a blood test, or know your history. What it can do is tell you whether your symptom pattern is common enough among diabetes patients to make a blood test worth asking for.

How it decides

Inside your browser, five small statistical models each read your answers their own way — one looks for simple overall trends, one for strict probability math, one compares you to the most similar patients on record, and two learn from decision-tree patterns. A final layer weighs their opinions into one score.

Why five? Each has different blind spots: single models are either overconfident or too crude. Combining them — and checking that behaviour with cross-validation on unique profiles — gives a more trustworthy estimate than any one model alone. The full comparison scores are on The numbers page.

Privacy

The entire model is a small file downloaded once when the page loads. Your answers are scored locally; nothing is uploaded, stored, or shared. Close the tab and everything is gone. One honest detail: the "most similar patients" expert carries the training table inside that downloaded file — de-identified questionnaire rows, no names — because comparing you to past patients is literally what it does.