Comparative analysis of patient-derived organoids and patient-derived xenografts as avatar models for predicting response to anti-cancer therapy

Abstract

Patient-derived xenografts (PDX) and organoids (PDO) are widely used to model cancer and predict treatment response in matched patients. However, their predictive accuracy has not been systematically studied nor compared. We conducted a systematic review and meta-analysis of studies using PDX or PDO from solid tumors treated with identical anti-cancer agents as the matched patient, identifying 411 patient-model pairs (267 PDX, 144 PDO). Overall concordance in treatment response between patients and matched models was 70%, with no significant differences between PDX and PDO. Sensitivity, specificity, positive and negative predictive value were also comparable. Patients whose matched PDO responded to therapy had prolonged progression-free survival. For PDX, this association held only when analyses were restricted to patient-model pairs with low risk of bias after applying a bias assessment metric. Together, these findings suggest that PDO perform similarly to PDX in predicting matched-patient response, while potentially offering lower financial and ethical burdens.

Keywords: PDO; PDX; avatar; organoid; xenograft.

Figure 1 –. Consort diagram.

Following screening, full-text review, and extraction, 411 patient-model pairs were extracted from 73 publications.

Figure 2 –. Demographic overview of cohort and concordance.

A) Distribution of PDX and PDO patient-model pairs across grouped cancer types. ^#Gastroesophageal PDX = 8 gastroesophageal, 1 esophageal, 1 gastric; PDO = 15 gastric, 2 esophageal, 1 gastroesophageal. ^&Other GI PDX = 3 cholangiocarcinoma, 3 hepatocellular, 2 duodenal, 2 anal; PDO = 3 gastroenteropancreatic neuroendocrine, 1 cholangiocarcinoma, 1 hepatocellular, 1 small bowel. ^†Genitourinary PDX = 4 renal, 3 urothelial, 2 prostate, 1 bladder, 1 testicular; PDO = 2 neuroendocrine prostate, 1 testicular. B) Percent of patient-model pairs with concordant responses compared to matched patients for all (left) and across cancers with at least 3 pairs per type (right). P-values calculated with Fisher’s exact test. Abbreviations: PDX; patient-derived xenograft, PDO; patient-derived organoid, NSCLC; non-small cell lung cancer, GI; gastrointestinal.

Figure 3 –. Predictive statistics comparing PDX and PDO.

A) Sensitivity, specificity, PPV and NPV between PDX and PDO. Point estimates, 95% CI, and adjusted P-values are shown. B) Predictive statistics between PDX and PDO, for high reliability (HR) and C) low reliability (LR) patient-model pairs. P-values were calculated with Fisher’s exact test and corrected with the Benjamini-Hochberg procedure to calculate adjusted P-values. D) Contingency tables highlighting concordance rates between high HR and E) LR pairs. P-values calculated with Fisher’s exact test. Abbreviations: PDX; patient-derived xenograft, PDO; patient-derived organoid, PPV; positive predictive value, NPV; negative predictive value, CI; confidence interval.

Figure 4 –. Responsiveness in model is associated with prolonged progression-free survival in matched patient.

Comparison of PFS in patients from whom associated model did or did not respond to treatment across A) entire cohort, B) PDX and C) PDO. Comparison of PFS in patients from high reliability patient-model pairs and from whom associated model did or did not respond to treatment across D) entire cohort, E) PDX, and F) PDO. One sided P-values calculated using log-rank test (Methods). Below each Kaplan-Meier curve are multivariate Cox proportional hazard models assessing the association between variables and patient PFS after removing categories with insufficient sample size (Methods). Abbreviations: PDX; patient-derived xenograft, PDO; patient-derived organoid, NSCLC; non-small cell lung cancer, PFS; progression-free survival.