Evaluating disease activity scores and CBC-derived indices in relation to osteoporosis risk in rheumatoid arthritis: a cross-sectional analysis
TL;DR
Disease-activity scores provided modest incremental discrimination beyond established clinical factors, whereas CBC-derived indices added little once conventional inflammatory and disease-activity information was included.
Abstract
Background Osteoporosis (OP) is a frequent comorbidity in rheumatoid arthritis (RA). This study compared the performance of RA disease-activity scores with complete blood count (CBC)-derived indices for identifying prevalent OP. Methods We retrospectively included 226 consecutive patients with RA who underwent dual-energy X-ray absorptiometry (DXA). Predictors comprised age, sex, body mass index (BMI), disease activity (DAS28-ESR, DAS28-CRP, CDAI, and SDAI), ESR, CRP, and six CBC-derived indices (NLR, PLR, MLR, SII, SIRI, and AISI). Logistic and linear regression, receiver-operating-characteristic (ROC) analysis with paired DeLong tests, decision curve analysis (DCA), and four-knot restricted cubic spline (RCS) analyses were performed. Results Older age, female sex, lower BMI, and higher CDAI were associated with OP. The five-covariate clinical reference model had an AUC of 0.833. Adding CDAI increased the AUC to 0.892 (ΔAUC, 0.059; P = 0.005), and adding SDAI increased it to 0.877 (ΔAUC, 0.044; P = 0.018), although the ROC curves overlapped substantially. In parallel fully adjusted models, none of the six CBC-derived indices was independently associated with OP (all P≥0.142; apparent AUCs, 0.892–0.895). Adding CRP after DAS28-ESR changed the AUC from 0.863 to 0.869 (ΔAUC, 0.006; P = 0.491). In the extended discrimination model, adding any CBC-derived index changed the AUC by no more than 0.0014 (all P≥0.467). DCA showed small, threshold-dependent differences in net benefit rather than consistent separation between models. Four-knot RCS analyses showed evidence of nonlinearity for DAS28-ESR (P for nonlinearity=0.007) and SDAI (P for nonlinearity<0.001), but not for ln(AISI) (P for nonlinearity=0.391) or ln(SIRI) (P for nonlinearity=0.718). Conclusions Disease-activity scores provided modest incremental discrimination beyond established clinical factors, whereas CBC-derived indices added little once conventional inflammatory and disease-activity information was included. These data do not support calculating CBC-derived indices solely to identify prevalent OP or using them as substitutes for guideline-based clinical risk assessment, FRAX, or DXA. The findings do not establish a new screening or treatment threshold.