Development of a Prognostic AI-Monitor for Metastatic Urothelial Cancer Patients Receiving Immunotherapy

Abstract

Background: Immune checkpoint inhibitor efficacy in advanced cancer patients remains difficult to predict. Imaging is the only technique available that can non-invasively provide whole body information of a patient's response to treatment. We hypothesize that quantitative whole-body prognostic information can be extracted by leveraging artificial intelligence (AI) for treatment monitoring, superior and complementary to the current response evaluation methods.Methods: To test this, a cohort of 74 stage-IV urothelial cancer patients (37 in the discovery set, 37 in the independent test, 1087 CTs), who received anti-PD1 or anti-PDL1 were retrospectively collected. We designed an AI system [named prognostic AI-monitor (PAM)] able to identify morphological changes in chest and abdominal CT scans acquired during follow-up, and link them to survival.Results: Our findings showed significant performance of PAM in the independent test set to predict 1-year overall survival from the date of image acquisition, with an average area under the curve (AUC) of 0.73 (p < 0.001) for abdominal imaging, and 0.67 AUC (p < 0.001) for chest imaging. Subanalysis revealed higher accuracy of abdominal imaging around and in the first 6 months of treatment, reaching an AUC of 0.82 (p < 0.001). Similar accuracy was found by chest imaging, 5–11 months after start of treatment. Univariate comparison with current monitoring methods (laboratory results and radiological assessments) revealed higher or similar prognostic performance. In multivariate analysis, PAM remained significant against all other methods (p < 0.001), suggesting its complementary value in current clinical settings.Conclusions: Our study demonstrates that a comprehensive AI-based method such as PAM, can provide prognostic information in advanced urothelial cancer patients receiving immunotherapy, leveraging morphological changes not only in tumor lesions, but also tumor spread, and side-effects. Further investigations should focus beyond anatomical imaging. Prospective studies are warranted to test and validate our findings.

How does target lesion selection affect RECIST? A computer simulation study

“…To test the influence of each of the parameters of the simulation model on the RECIST assessment, we fix, in succession, all but one of the six input parameters to a default value, while the remaining variable is made to change within a certain range: L max between 1 and 20, O max between 1 and 10, µ between −100% and 300%; and µ, ε P , ε O and ε R each between 10 2 and 100 2 ‱. The default values of the µ, ε P , ε O and ε R , (12.8%, 44.6 2 , 11.6 2 , 36.6 2 , respectively) were estimated from a real dataset of patients with melanoma and urothelial cancer (Trebeschi et al 2019; Trebeschi et al 2021), treated with immunotherapy at the NKI-AVL hospital, by fitting a linear mixed-effects model, whereas L max and O max were set to 3 different pairs of values (5, 2), (10, 4) and (15, 8), respectively — low, medium and high disease stage, respectively.…”

“…The default values of µ, Ɛ P , Ɛ O, and Ɛ R were estimated from three real datasets: n=61 patients with melanoma treated with immunotherapy, n=44 urothelial cancer patients treated with immunotherapy, and n=37 patients with non-small cell lung cancer treated with chemotherapy, already reported in previous work (S Trebeschi et al 2019;Stefano Trebeschi et al 2021). Ɛ P has no impact on target lesion selection, and thus its impact on variability was not analyzed.…”

“…The Response Evaluation Criteria in Solid Tumors (RECIST) consists of a standardized methodology used in clinical trials to evaluate tumors' response to therapy (Seymour et al 2010), by defining endpoints surrogate for overall survival, namely progression-free survival and overall response rate (Villaruz and Socinski 2013;Mushti, Mulkey, and Sridhara 2018). The most recent version of the criteria, RECIST 1.1 (Eisenhauer et al 2009), states that a maximum of five measurable lesions, and no more than two per organ, should be selected as target lesions and measured at baseline.…”

“…Target lesion selection assumes that target lesions can be objectively and reproducibly identified and measured (Villaruz and Socinski 2013). However, inter-and intra-variability when measuring tumor size, and identification of new lesions are known factors contributing to inconsistency in the application of RECIST (Skougaard et al 2012;Abramson et al 2015;Sridhara, Mandrekar, and Dodd 2013;Beaumont et al 2018).…”

The Porto European Cancer Research Summit 2021

“…Exemplifying the promise of deep learning in clinical trial data analyses, a recently published study [ 53 ] assessed the predictive value of a prognostic AI‐Monitor (PAM) for patients with metastatic urothelial cancer receiving immunotherapy. The study investigators hypothesized that quantitative whole‐body prognostic data can be extracted by leveraging AI as a superior and complementary approach to current response evaluation methods.…”