Better Care. Faster A new predictive algorithm for aiding clinical decision-making in lung cancer Y Kogan, I Sela, Y Kheiffetz, M Kleiman, Z Agur, Optimata Ltd., O Liran, N Peled, Sheba hospital, I Lazarev, Y Dudnikov, S Ariad, Soroka hospital, Israel
Agenda Introduction Background lung cancer (LC) Aims Algorithm for Individualized Medicine LC-Predicare TM Features Technology Retrospective Clinical validation Prospects
Background LC is the most common human malignancy Most patients are diagnosed at advanced stages treatable only by systemic chemotherapy Efforts to personalize therapy validated only two predictive mutations. Physicians still choose the treatment based on statistical evidence from clinical trials (i.e., NCCN guidelines). Long-term prognosis is poor and five-year survival is 15%. Available individualization algorithms are insufficient; they can only slot a patient to one of a few characteristic patient subpopulations. 3
Aim Develop a new approach to treatment individualization Realize it in a clinically implementable algorithm for predicting treatment response and prognosis in individual LC patients. Validate the algorithm 4
Approach Several categories of knowledge must be integrated for predicting a patient s response: Response in the patient population (statistics of large datasets) Clinical and biological characteristics of the patient (clinical tests) Disease dynamics as affected by the particular drug regimen (biomathematics) How to integrate all the above in an efficient predictor. 5
Method Improve a statistical/mathematical integrative methodology to encompass the population scale, reflected in the statistical distributions of clinical parameters in the population the individual patient scale, reflected in the dynamic pathology/pk/pd model of the patient This integration enables to create personal models Which describe drug-patient interactions 6
Features Web-based (Cloud): Easily Accessible Fast update Input Report
Algorithm Oncologist Select from the list of available treatments: T 1, T 2,,T N Select from the list of available endpoints: E 1, E 2,,E M Personal Model for Patient X Input all available individual data Computer Construct the personal model Simulate individual response to treatment T 1 Simulate individual response to treatment T 2 Simulate individual response to treatment T N Evaluate selected endpoints E 1,,E M for treatments T 1,,T N Generate report on predicted outcomes of the selected treatments in patient X
Retrospective Validation Patients: Advanced LC patients (38) treated by cisplatin and pemetrexed. Data collection: Radiological evaluations of individual lesion size at diagnosis, during and post-treatment (1-3 lesions per patient, 2-6 measurements per lesion). Drug administration schedules per patient, Pre-treatment data, including age, gender, BSA, medical history, TNM staging, tumour histology and genetics. 9
Retrospective Validation Patients can be clustered into responders/non-responders, based on pretreatment data; actual response of a lesion depends on several factors and can be predicted 10
Retrospective Validation Good 6 months RECIST predictions Based on pretreatment data 11
Retrospective Validation Algorithm-predicted vs. observed lesion sizes, using individual pre-treatment data 12
Summary We have developed an algorithm (LC-Predicare TM ), which, based on pre-treatment information, predicts the response of individual patients to treatment with the 1 st line chemotherapy combination, cisplatin & pemetrexed Extension to all other 1 st line and 2 nd line drug therapies in LC patient is underway Technology is in place; prediction success depends on availability of sufficient patient files for training and validating the algorithm Prospective validation for affirming clinical applicability is mandatory PC-Predicare (in collaboration with Mayo Clinic) is under development
Optimata is interested in Thank partnerships You with providers of NSCLC patients data bases, which include tumor size measurements Professor Zvia Agur Chair & CSO Optimata Ltd agurz@optimata.com 14
15