Patient-derived xenografts (PDX) versus patient-derived organoids (PDO) as predictors of clinical response to anti-cancer therapies.

e15169 Background: PDX and PDO are two of the most frequently applied avatar model systems used to predict treatment response to anti-cancer therapies. Despite their frequent use, and the significant financial and ethical costs associated with developing these models, there has never been a systematic assessment of their ability to predict matched-patient treatment response. We sought to define and compare the efficacy of PDX and PDO in predicting matched-patient response to treatment. Methods: We performed a systematic review and meta-analysis in accordance with PRISMA guidelines. MEDLINE and EMBASE were queried. Inclusion criteria: PDX or PDO derived from adult solid cancer patients treated with identical systemic anti-cancer agents as the matched patient, with response assessment performed for both patient and model. Fisher’s exact test and Kaplan-Meir estimator with log-rank test were used for statistical comparisons. A 6 criteria quality assessment method based on Newcastle-Ottawa scale was applied to patient-model pairs. Results: 21565 abstracts were screened. 274 were eligible for data extraction, with 411 patient-model pairs included (N = 267 PDX, N = 144 PDO). The most common cancer types were colorectal (N = 102, 25%) and ovarian cancers (N = 77, 19%). Most common treatment modalities were chemotherapy (244, 59%) and targeted therapy (122, 30%). 55% of models were responsive to therapy (N = 227). Overall concordance in treatment response between patient and matched models was 70%, with no difference between PDX and PDO ( Table ). No significant differences in sensitivity, specificity, positive- and negative predictive value (PPV and NVP) were observed (Table). 196 pairs (48%) and significantly more PDX had high quality data reporting (56% of PDX vs 33% of PDO, P < 0.001). Of pairs with high quality data reporting, PDX had higher sensitivity, while PDO had higher specificity ( Table ). Patients whose matched PDO responded to therapy had longer median progression-free survival (mPFS; responders: 11.3 vs non-responders: 3.4 months P < 0.01). For PDX this only remained true for high quality data pairs (mPFS; responders: 9.5 vs non-responders: 6 months P < 0.01). Conclusions: This is the first study to systematically assess the utility of PDX and PDO as patient avatars. Together, these results suggest that PDO generally perform similarly to PDX as predictors of matched-patient response despite a lower ethical and financial burden. Performance metrics for PDX and PDO as predictors of matched-patient treatment response. All PDX (N = 267) PDO (N = 144) P-Value Concordance (%) 71 69 0.9 Sensitivity (%) 87 85 0.81 Specificity (%) 58 60 0.69 PPV (%) 64 52 0.1 NPV (%) 84 89 0.37 High Quality Data PDX (N = 149) PDO (N = 47) P-Value Concordance (%) 72 81 0.57 Sensitivity (%) 95 73 <0.01 Specificity (%) 54 88 <0.01 PPV (%) 61 84 0.07 NPV (%) 94 79 0.07