J-shapedness: an often missed, often miscalculated relation: the example of weight and mortality.

We present three considerations in analysing the association between weight and mortality, as well as other relations that might be non-linear in nature. First, authors must graphically plot their independent and dependent variables in a continuous manner. Second, authors should assess the shape of that relation, and note its shape. If it is non-linear, and specifically, J-shaped or U-shaped, careful consideration should be given to using the 'best' statistical model, of which multivariate fractional polynomial regression is a reasonable choice. Authors should also refrain from truncating their data to avoid dealing with non-linear relations. Published by the BMJ Publishing Group Limited. For permission to use (where not already granted under a licence) please go to http://group.bmj.com/group/rights-licensing/permissions.