846P - Artificial intelligence-driven identification of onco-hematology patients who may develop severe COVID-19

Background

The widespread implementation of COVID-19 vaccines achieved a great success in the general population. However, an unmet medical need exists for onco-hematological and other immunocompromised patients for whom vaccines may not provide optimal protection. Identifying these patients to consider additional layers of protection, including passive immunization with monoclonal antibodies, has been a challenge. Herein, we propose an artificial Intelligence (AI) driven methodology to aid the identification of patients at risk of severe COVID-19 via a minimal number of variables.

Methods

A multi-centric training (340 patients) and an external testing (103 patients) retrospective cohort with homogeneous population characteristics were built. Inclusion referred to hematological or lung cancer patients with COVID-19 infection and absence – except vaccination – of any COVID-19 preventive treatments. The E-CRF included 70+ clinical and demographic variables. Patients were classified into two different COVID-19 severity populations, mild and moderate/severe, based on oxygen support. We sought to identify (i) the minimal set of independent variables (signatures) characterizing these populations, and (ii) the best AI solution acting on these signatures to predict the severity of infection via an end-to-end integration of statistical importance and AI methods.

Results

Using the training cohort we identified 7 variables (age and cancer stage ranked as most important) enabling the prediction of moderate/severe cases with a balanced accuracy of 72% (AUC 0,76) on the internal cohort. The model performed equally well on the external testing cohort yielding a similar balanced accuracy of 73% (AUC 0,77). Most importantly the model reached a 85% performance on the severe cases.

Conclusions

The conducted analysis leads to promising insights towards identifying onco-hematological patients who are most vulnerable to develop severe COVID-19. This may help to offer additional prevention strategies to protect them from COVID-19 and can potentially be further extended to other immunocompromised populations.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

TheraPanacea, Paris.

Funding

AstraZeneca.

Disclosure

J-Y. Blay: Financial Interests, Institutional, Invited Speaker: MSD, PharmaMar; Financial Interests, Institutional, Advisory Board: Bayer, GSK, Roche; Financial Interests, Personal, Advisory Board: Deciphera; Financial Interests, Personal, Other, member of the supervisory board. No remunerations in 2021 and 2022.: Innate pharma; Financial Interests, Personal, Member of Board of Directors: Transgene; Financial Interests, Institutional, Funding: MSD, BMS, Deciphera; Financial Interests, Institutional, Research Grant: AstraZeneca, Roche, Bayer, GSK, Novartis, OSE pharma. P-E. Heudel: Financial Interests, Personal, Advisory Board: AstraZeneca, Pfizer, Novartis, Seagen, Lilly, MSD, Gilead; Financial Interests, Personal, Ownership Interest: GeodAIsics. C. Bigenwald: Financial Interests, Personal, Advisory Board: Janssen. L. Albiges: Financial Interests, Institutional, Other, Consulting: Astellas, BMS, Eisai, Ipsen, Janssen, MSD, Novartis, Pfizer, Roche, Merck; Financial Interests, Personal, Other, Honoraria: Novartis; Non-Financial Interests, Principal Investigator, Clinical trial steering committee: Pfizer, BMS, AVEO, AstraZeneca, MSD; Non-Financial Interests, Principal Investigator: Ipsen; Non-Financial Interests, Other, Clinical trial steering committee: Roche, Exelixis; Non-Financial Interests, Member: ASCO; Non-Financial Interests, Other, Medical Steering Committee: Kidney Cancer Association; Non-Financial Interests, Other, Member of the Renal Cell Carcinoma Guidelines Panel: European Association of Urology (EAU). F. André: Financial Interests, Personal, Advisory Board: Lilly France; Financial Interests, Institutional, Advisory Board: AstraZeneca, Daiichi Sankyo, Roche, Lilly, Pfizer, Owkin, Novartis, Guardant Health; Financial Interests, Institutional, Advisory Board, I'm member of the advisory board of Relay Therapeutics. The compensation is going to Gustave Roussy: relay therapeutics; Financial Interests, Institutional, Research Grant: AstraZeneca, Lilly, Novartis, Pfizer, Roche, Daiichi Sankyo, Guardant Health, Owkin; Other, Founder: Pegacsy. All other authors have declared no conflicts of interest.