Peter van Mourik

147 Personalized treatment of CF using rectal organoids r=0.575, P=<0.001, Fig. 1i) and SCC (n=33, r=-0.708, P=<0.001, Fig. 1j), but a statistically significant relation between ppFEV 1 and SCC was not observed. (Fig. 1h,k). People with rare mutations who were selected by organoids prior to treatment showed a median increase of 10% in ppFEV 1 (n=7, p=0.058) and a reduction of 39 mmol/L in SCC (n=6, p=0.028). Collectively, these data demonstrates that in vitro CFTR modulator responses in organoids correlate with two important therapeutic endpoints. Prediction of clinical responses using organoids Next, we generated ROC-curves to examine the predictive potential of different organoid-based thresholds for identifying clinical responders. We dichotomized both the ppFEV 1 and SCC response into changes that are generally considered clinically significant and beyond the test variability (changes in ppFEV1>5%, or SCC>20mM or a combined change in ppFEV 1 >5% and SSC>20mM) and changes that are not 33 . The area under the ROC-curve provides a general measure for test accuracy and was 0.837 (95% CI 0.661 – 1.000) for predicting responders in ppFEV 1 and increased towards 0.938 (95% CI 0.830 – 1.000) for predicting responders in either SCC or SCC and ppFEV 1 (Fig. 2a). When repeated measurements were taken into account, the area under the ROC-curve did not change. A Youden index was used to select an organoid cut-off point with the most optimal combination of sensitivity and specificity in an unbiased fashion 25 . The selected cut-off value to identify responders in both SCC and ppFEV 1 had a sensitivity of 0.80 and a specificity of 1.00 with a corresponding Youden index of 0.8 for identifying responders and non-responders in both ppFEV 1 and SCC. The associated positive and negative predictive values were 100% and 80%, respectively. Since data driven selection of the Youden index might cause over-estimation of both sensitivity and specificity, we performed a leave- one-out cross validation to further validate our findings 26 . This additional analysis showed a sensitivity of 0.70 and specificity of 1.00, with a corresponding Youden index of 0.70. For patients that started with a ppFEV 1 <40% or >90% the ROC-curve had an area under the curve between 0.694 and 0.767 (Fig. 2b). For the total group of patients that was treated, the area under the ROC-curve varied between 0.783 and 0.869 (Fig. 2c). Because of the small sample size we did not calculate ROC-curves for the group of patients that had at least one rare CFTR mutation. In conclusion, the organoid-based test displayed excellent accuracy (AUC of ROC- curve > 0.9) for identifying clinical responses defined by changes in SCC and ppFEV 1 or only SCC, while good accuracy (AUC of ROC-curve between 0.8 and 0.9) was observed for identifying clinical responses defined only by ppFEV 1 34 . 7