. 2021 Jun 12;22:534. doi: 10.1186/s12891-021-04403-5

Table 1.

Candidate variables

Domains	Variables^a
Demographics and lifestyles^b	(1) age, (2) BMI, (3) gender, (4) smoker, always or never/seldom, (5) drinking, always or never/seldom
Medical history^b	(6) hypertension, (7) diabetes mellitus, (8) coronary heart disease, (9) hyperuricemia, (10) dyslipidemia, (11) femoral neck necrosis
Medication history of RA patients^b	(12) chronic NSAIDs usage, (13) chronic GC usage, (14) types of cDMARDs recently taking, (15) TNFi usage in the last one year, (16) overall anti-osteoporotic medication history^b, (17) disease duration
Laboratory	(18) serum calcium level, (19) serum phosphate level, (20) sCr, (21) sUA, (22) CRP level^b, (23) ESR^b, (24) rheumatoid factor concentration^b, (25) anti-CCP concentration^b (26) serum 25(OH)D3 level^b, (27) C3 level, (28) C4 level, (29) CH50 level, (30) ANA titer, (31) rheumatoid factor positive, (32) anti-CCP positive, (33) AKA positive, (34) RA33 positive

BMI Body mass index, NSAIDs Non-steroidal anti-inflammatory drugs, GC Glucocorticoid, cDMARDs Conventional disease-modifying anti-rheumatic drugs, TNFi Tumor necrosis factor-α inhibitor, sCr Serum creatine level, sUA Serum uric acid level, CRP C-reactive protein, ESR Erythrocyte sedimentation rate, anti-CCP Anti-cyclic citrullinated peptide antibodies, AKA Antikeratin antibodies, RA33 Anti-RA33 antibodies, C3 Complement component 3, C4 Complement component 4, CH50 Serum total complement activity, ANA Antinuclear antibodies

^aAll variables were included in statistics-driven (LASSO) and random forest model. Except from variable 1,2,16–29, all variables were categorial or dichotomous

^bFactors selected for the clinical knowledge-preselected model