For me, I sometimes struggle to come up with good examples of an ‘EBM’ tennet – for instance, the difference between statistical and clinical significance – which has an actual origin. Well, in a paper entitled Platelet Counts in Children With Henoch–Schonlein Purpura—Relationship to Renal Involvement I think I have hit gold.
The paper looked at 107 patients with HSP, of whom 28 had renal involvement. A multivariable logistic regression analysis showed that platelet count was significantly related to renal involvement (p=0.001). The mean counts were 483 in the renal group, and 422 in the uninvolved group.This led to an OR of 1.0 (1.00 to 1.02) for platelets. They also showed that with the optimal cut point (>565) there was a 36% sensitivity and 49% specificity for HSP nephritis (I think they got these reversed).
Yup. Read those numbers again.
LR+ 1.4, LR-0.7 — if the sensitivity and specificity are indeed written backwards — functionally useless.
And OR of 1.0 (1.00 to 1.02)? (Well actually this is poor presentation of the data — it’s ‘per platelet *10^3’ that this is done in – would make more sense to be for every 10 or 50 increase, wouldn’t it?
And then add to this the over-optimism of such data cuts.
Like I said. EB-example-gold.
Just, whatever you don, DON’T use platelet count in HSP if you’re wondering about kidneys.