Hidradenitis Suppurativa and Diabetes: Big Data Bias Masks a True Association

Abstract

Phan et al present an updated meta‐analysis of the association between Hidradenitis Suppurativa (HS) and Type 2 Diabetes Mellitus (DM) which demonstrates a statistically significant elevation in the odds ratio (OR) for DM in HS Patients compared with healthy controls (OR=2.17 95% CI 1.85‐2.55 p<0.001). They astutely highlight that the risk difference of 0.4% between HS patients and controls (16.1% vs 15.7%) implies that this result is clinically non‐significant, quite at odds with clinical experience, data from existing studies and the remainder of the article discussing the importance of DM awareness in HS patients and possible mechanistic links between DM and HS. The reason for this statistical‐common sense disconnect is the lack of acknowledgement of external bias in Phan's analysis. Whilst the methodology is sound, it does not eliminate all bias and residual confounding despite the high reported I2 (which as a point estimate is misleading in the setting of large heterogeneity). Although weighting of studies addresses 'within study variance' and heterogeneity (I2) assesses methodological (internal) heterogeneity, no consideration of external biases (including population bias and outcome bias) is provided by a statistical program, it requires the use of common sense on behalf of the statistician and authors often by plotting and visualising the data. We replicated Phan et al's analysis with the additional step of assessing studies for external bias and threats to generalisability. Visualisation of data in a L'Abbe plot demonstrates lack of clustering on the x axis‐ indicating violation of the assumption of one underlying baseline risk.

This article is protected by copyright. All rights reserved.

https://ift.tt/2Nr4n5F