<p>Noise (synonyms: variability, imprecision) in clinical data is underappreciated. All measurements are affected by noise and bias. Research studies enrol many patients to average out noise and use randomisation and double-blinding to counter bias. These mitigation techniques are often not available in clinical medicine, where individual patients typically have a variable measured only once.</p><p>The example of echocardiographic assessment of aortic stenosis severity illustrates how the conventional presentation of data as point estimates masks wide variability in noise between variables, which affects utility. Adding aortic valve area to the quantification of aortic stenosis may result in poorer (noisier) estimates.</p><p>Every clinical variable reported as a point estimate comes with an invisible cloud of uncertainty (noise) of varying size across variables. Being mindful of how this noise varies across variables should facilitate optimal clinical decisions.</p> Graphical abstract <p></p>