We read with great interest the report by Otomo et al (Otomo K, Atsumi T, Amengual O, Fujieda Y, Kato M, Oku K, et al. Efficacy of the antiphospholipid score for the diagnosis of antiphospholipid syndrome and its predictive value for thrombotic events. Arthritis Rheum 2012;64:504–12). The authors describe the derivation of a new quantitative index, the antiphospholipid score (aPL-S), as a way to predict clinical manifestations of antiphospholipid syndrome (APS) using the antiphospholipid antibody (aPL) assay profiles of patients with autoimmune diseases.
In the study, each phospholipid assay was weighted according to a formula based on an exponential transformation of the odds ratio (OR) of having clinical manifestations of APS (thrombosis and/or pregnancy morbidity). Then, the aPL-S for individual patients was calculated as the sum of weights for each aPL test.
While Otomo and colleagues have derived an original tool for scoring the risk of clinical events in patients with aPL, we nevertheless believe that a few points should be raised. First, the statistical rationale for deriving the weights for each aPL test, i.e., 5 × exp([OR] − 5)/4, is unclear. Second, the maximum weight for each test was arbitrarily set at 20, i.e., ORs >10.55 were weighted as if they were 10.55. Consequently, IgG anti–β2-glycoprotein I antibodies (anti-β2GPI) were given the same weight as IgG anti–phosphatidylserine/prothrombin complex (anti-PS/PT) and IgG anticardiolipin antibodies (aCL), whereas the former was associated with a much greater risk of clinical events (OR 19.3 for IgG anti-β2GPI versus 11.1 and 11 for IgG anti-PS/PT and IgG aCL, respectively). Third, the summation of weighted ORs obtained through a bivariate analysis may not be considered a statistically valid way to construct a score for predicting the probability of an event. Indeed, each OR should be adjusted in relation to the others through a multivariate regression model, because the relationship between the various aPL assays is truly a critical issue. Fourth, through their analysis of 2 sets of patients with various autoimmune diseases, Otomo et al show that the aPL-S is higher in patients with thrombosis/pregnancy morbidity than in those without manifestations of APS. We believe this is a tautology, since the aPL-S rates the presence of aPL while clinical manifestations of APS are associated with the presence of such antibodies. Therefore, we believe that an analysis that is limited to patients with aPL (i.e., with a higher pretest probability of an event) would dramatically increase the positive predictive value and clinical interest of the aPL-S.