For this study, 309 consecutive patients with arthralgias of the wrist- and/or finger joints were screened for the presence of an RA diagnosis, via clinical, laboratory, OST, and US examinations (examined joints: n = 6,798). The diagnosis of RA was made in 94 patients, 22 of whom were excluded due to initiated glucocorticoid therapy, previous to study screening. Of the 215 remaining non-RA patients, 53 were diagnosed with non-inflammatory arthralgia (fibromyalgia and/or osteoarthritis) and were used as a second control group (next to the healthy control group). The other 162 patients were diagnosed with further rheumatologic diseases (i.e. various inflammatory arthritides, crystal arthropathies, and connective tissue diseases) and were excluded from the study.

1,584 joints of 72 RA patients (65.3% female), 2,200 joints of 100 healthy control subjects (80% females), and 1.166 joints of 53 non-inflammatory arthralgia patients (86.8% female) were examined by OST. OST scores of the RA group were statistically significantly higher compared to the healthy control group (16.8 ± 5.5 vs. 10.8 ± 4.0) and the non-inflammatory arthralgia group (16.8 ± 5.5 vs. 11.4 ± 4.4), respectively (both; p < 0.001) (Table 1). Further descriptive characteristics of the 3 groups are presented in Table 1.

Table 1 Descriptive characteristics by groupAssociations of OST score with continuous variables

Except for the OST score and DAS28-ESR, further examined variables were not normally distributed (Shapiro-Wilk-test p < 0.05). Thus, we calculated Spearman’s rank correlation for all bivariate associations. Among RA patients, Spearman’s analyses showed a moderate correlation between OST and swollen joint counts (SJC-rho = 0.355,p = 0.002), VAS scale (rho = 0.383,p = 0.001), DAS28-ESR (rho = 0.361,p = 0.002), and age (rho = 0.385,p = 0.001) (Table 2). Moreover, a weak association between OST and hand size (rho = 0.267,p = 0.023) was found.

Table 2 Association between patient characteristics and OST in RA group and control group

Among control subjects, OST correlated moderately with hand size (rho = 0.465,p < 0.001) and weakly with BMI (rho = 0.258,p = 0.01) (Table 2).

Receiver operating characteristics (ROC) analysis and cut-off values

ROC analyses were done to test the diagnostic performance of OST. Because of the substantial difference in OST scores between males and females (Fig. 2), the ROC analysis was additionally stratified by sex. To test the diagnostic performance of OST, ROCs were performed twice, once comparing the whole RA group and once an RA-subgroup [patients with ≥ 1 acutely swollen wrist of finger joint(s)] with the two control groups, respectively.

Fig. 2

Top left. ROC of OST in females between RA-group and control group (reference). Top middle. ROC of OST in males between RA-group and control group (reference). Top right. ROC of OST overall (male and female together) between RA-group and control group (reference). OST-AUC for females (0.848; 95% CI 0.780–0.917); males (AUC 0.696; 95% CI 0.537–0.855); overall (AUC 0.81; 95% CI 0.746–0.873) ROC: Receiver-operating characteristic; OST: optical spectral transmission; RA: rheumatoid arthritis; AUC: area under the curve

In the comparison between the whole RA-group vs. healthy control group, the overall diagnostic performance of OST was found to be excellent by an AUC of 0.810 (95%CI: 0.746–0.873) and further improved during the comparison of the RA subgroup with ≥ 1 swollen wrist or finger joint vs. the healthy control subjects [AUC 0.841 (95%CI: 0.773–0.908)]. Interestingly, the diagnostic performance of OST was higher in the female group, by an AUC of 0.848 (95%CI: 0.780–0.917), compared with the male group (AUC 0.696; 95%CI: 0.537–0.855). These values were also found to be higher in the comparisons between the RA subgroup with ≥ 1 swollen wrist or finger joint(s) and the control group for both females [AUC 0.87 (95%CI; 0.797–0.943)] and males [AUC 0.78 (95%CI; 0.606–0.954)], respectively.

In the case of comparing RA- with non-inflammatory arthralgia patients as an additional control group, similar results were observed by a very good overall diagnostic performance [AUC of 0.788 (95%CI: 0.709–0.867)]. Also in this case, diagnostic performance further improved in the comparison between the RA subgroup with ≥ 1 swollen wrist or finger joint(s) and the non-inflammatory arthralgia patients by an AUC of 0.822 (95%CI: 0.740–0.90) [females: AUC: 0.823 (95%CI: 0.727–0.918) and males AUC 0.688 (95%CI; 0.441–0.934)].

The determination of cut-off-values was based on the Youden index and the comparison between the RA group and healthy controls. The OST cut-off value in the female subgroup was 11.17 with a sensitivity of 85.1% and specificity of 71.2% (Youden index: 0.563). In comparison, in the male subgroup, a cut-off of 16.05 with a sensitivity of 72% and specificity of 65% (Youden index: 0.37) was found. Overall sensitivity/specificity values improved when the RA subgroup with ≥ 1 swollen wrist or finger joint(s) was compared with the control group (cut-off of 11.17; sensitivity 93%, specificity 71.2% for females, and cut-off of 18.21; sensitivity 67%, specificity 90% for males).

Prediction models for the risk of RA

To predict the probability of a positive RA diagnosis, based on the measured OST score and the presence of possible OST influencing factors, we have pre-selected 4 different logistic regression models (Table 3, suppl. material). In the first model, OST (as the main variable), age, and gender were included. In the second model, smoking was added to the variables to rule out an influencing effect through blood-flow restriction. Models 3 and 4 included hand size and BMI respectively, which have been shown to correlate with OST in the past and the current study.

These 4 models were tested against each other using an LR test to find the optimal prediction model for the presence of RA. The four models, as presented in Table 3, were analyzed and compared using logistic regression. All models were first estimated for all patients, followed by a gender-specific estimation.

Statistic model regarding all patients

According to the omnibus test, all four models were statistically significant (p < 0.001, Table 3, suppl. material). Compared with model 1, the second model including smoking status as an additional variable, did not yield a significantly better risk estimation of RA (LR-Test: 1.804, p = 0.179). The sum of the percentage of overall correct predictions was 80.2% for model 2 and 78.5% for model 1. Adding hand size and BMI in models 3 and 4 led to no significant improvement in the prediction of RA compared to model 1 (model 3 vs. 1: LR-test: 1.996, p = 0.158; model 4 vs. 1: LR-test: 2.264, p = 0.132), with the sum of the percentage of overall correct predictions being the same for models 3 and model 4: 80.2%.

Thus, model 1 proved to be the optimal one for the prediction of RA for all patients. Among the analyzed predictors, only OST score (OR = 1.296 95%-CI: 1.163–1.444, p < 0.001) and age (OR = 1.067, 95%-CI: 1.033–1.103,p < 0.001) were significant. The sum of the percentage of overall correct predictions was 78.5% (RA-patients: 70.8%). The prediction equation for the risk of RA for all patients can be found in Table 3. An additional analysis showed a significant interaction effect between gender and OST score for all patients. For increasing OST values above the respective mean, the risk for RA was higher in females versus males. For decreasing OST values below the respective mean, the risk for RA was higher in males versus females (Fig. 3).

Fig. 3

Plot showing the association of predicted probabilities for RA and mean-centered OST-values stratified by sex.Respective lines showing the association for men and women based on a logistic regression model with predictors: sex, mean-centered OST-values, interaction term of sex and OST-values. RA: rheumatoid arthritis; OST: optical spectral transmission

Stratified for gender

For females and males separately, the first model including OST values and age proved to be the optimal prediction model for RA (females: Chi-Square: 62.47, p < 0.001, males: Chi-Square: 17.01,p < 0.001). Adding smoking, hand size, and BMI did not significantly improve the risk prediction (LR-test not significant for model 2, 3, 4 vs. model 1). The sum of the percentage of overall correct predictions for females was 81.9% (RA-patients: 70.2%) and for males 73.3% (RA-patients: 84.0%). OST score was shown to be a significant predictor for RA in females (OR = 1.411; 95% CI: 1.225–1.624,p < 0.001) but not in males (OR = 1.411; 95% CI: 1.225–1.624,p < 0.323). Moreover, age was a significant predictor for RA both in females (OR = 1.059, 95% CI: 1.018–1.101,p = 0.004) and males (OR = 1.091, 95% CI: 1.027–1.160, p = 0.005).

The association of predicted probabilities for RA and mean-centered OST values stratified by sex is shown in Fig. 3 and the prediction equation for the risk of RA stratified by gender is in Table 3.

Table 3 Selected prediction models for RA in male and female group. P(Y): the estimated risk for RA with the parameter