Comparison of ImmunoCAP™ and NOVEOS clinical performance
We performed ROC analysis to compare clinical performance of ImmunoCAP™ and NOVEOS. We also selected sIgE optimal thresholds allowing optimal discrimination between food allergic and non-allergic individuals. As presented in Table 2 and Figure S2, areas under ROC curves (AUC) were comprised from 0.79 for hazelnut extract (ImmunoCAP™ f17) to 0.97 for ovomucoid nGal d 1 (NOVEOS F233). However, the mean value of AUC was higher for NOVEOS than for ImmunoCAP™ (p=0.03). Next, and for setting the optimal threshold for the 10 food allergens, Youden’s index that combines optimal sensitivity and specificity was calculated and the optimal cut-off values for sIgE reported in Table 2. Cut-off values for ImmunoCAP™ and NOVEOS were similar (mean: 6.7 ± 3.8 kUA/L for ImmunoCAP™ and 4.6 ± 2.1 kUA/L for NOVEOS) and the difference was not significant. The most divergent cut-off values were between hazelnut extract (ImmunoCAP™ cut-off 16.7 kUA/L versus NOVEOS cut-off 3.6 kUA/L) and cow’s milk (ImmunoCAP™ cut-off 6.9 kUA/L versus NOVEOS cut-off 3.3 kUA/L). Sensitivity, specificity, positive (PPV) or negative (NPV) predictive values were also comparable without significant difference (mean sensitivity and specificity: 81% and 84% for ImmunoCAP™ and 84% and 86% for NOVEOS). The highest PPV was 95% except for egg white extract (ImmunoCAP™ and NOVEOS: highest PPV=90%) (Table 2). Next, sIgE cut-offs established for the 10 food allergens were further used to discriminate allergic from non-allergic individuals by calculating OR, RR, Cohen’s kappa coefficients, and percentages of agreement between each method and patients’ status. As presented in Figure 3, a strong association between sIgE results and clinical status was reported for the 10 allergens. The overall values of RR (4.1 for ImmunoCAP™ versus 4.6 for NOVEOS), OR, kappa and agreement were higher for NOVEOS than for ImmunoCAP™ when data from all 10 allergens were pooled, as well as when only allergen extracts were considered (Table S1). However, these differences were not significant. For the four MA, OR and RR were higher for ImmunoCAP™ than for NOVEOS, while agreement and kappa indexes were identical with both techniques (Table S1). In addition, both techniques were associated with better OR, RR, agreement and kappa indexes for MA than for allergen extracts (Table S1). Considering individual allergens (Table S2), cow’s milk sIgE results (extract and casein nBos d8) were associated with the highest RR (8.2), OR (>100), percentages of agreement (90-92%) and kappa indexes (0.80-0.83). The lowest values were obtained with ImmunoCAP™ peanut extract (RR: 2.7; kappa: 0.49; agreement: 75%) and hazelnut extract (RR: 2.6 for NOVEOS).
Cohen’s Kappa index showed a “good” (0.61-0.80) or “very good” (0.81-1) degree of association for 6 out of 10 allergens tested for ImmunoCAP™ as compared to 8/10 for NOVEOS (Table S2, p=0.035, Wilcoxon test). In addition, percentages of agreement were above 80% for 9 out of 10 allergens tested with NOVEOS while this was true for 7 out of 10 allergens tested with ImmunoCAP™ (Table S2). Overall, comparing kappa indexes and agreement for 10 allergens (i.e. 20 comparisons), values were more frequently higher for NOVEOS (11/20) than for ImmunoCAP™ (2/20).
Finally, thirty-one non-allergic patients present with a class “1” or class “2” sIgE results (0.35 to 3.5 kUA/L) with ImmunoCAP™, but have a class “0” (≤0.35 kUA/L) NOVEOS result (p<0.0001, Figure 2B), while there is no significant difference between the two methods for allergic patients with low sIgE results (Figure 2C).