'Comically bad' datasets used to train clinical models for stroke and diabetes

In three linesResearchers expose critically poor quality datasets used to train clinical models for stroke and diabetes prediction. Data contains systematic errors and biases that undermine the reliability of medical predictions.

Summary generated by Claude — human-verified