Abstract
Health systems and caregivers are overloaded and the resources are limited. These pain points are expected to increase, as the population ages and continues to grow. Moreover, the guidelines regarding the need for treatment and the selection of proper treatment are generic and cover only the minority of the population. New treatments are continuedly being developed and it is required to actively update with best research and practices, at the point of care. Therefore, there is a need to use new technologies such as Artificial Intelligence (AI) and Machine Learning (ML) in order to provide a decision support tool for caregivers, with regard to the appropriate treatment. In addition, to predict the efficacy of treatments, there is a need for a sufficient data set. However, in many cases, the relevant data is not present in medical records or elsewhere, or it is partial, and insufficient for concluding the treatment's efficacy. The American Cancer Society estimates that in 2021 about 21,250 new cases of chronic lymphocytic leukemia (CLL) were diagnosed, with estimated deaths of 4,410 persons.
In this work, a clinical decision support system is being developed for the decision-making regarding the question of whether treatment of Chronic Lymphocytic Leukemia/Small Lymphocytic Lymphoma (CLL/SLL) is justified and for prediction of treatment efficiency, using ML methods which have been developed by Seresnus. AI. Serenus.AI has developed an AI-based system (Serenus .AI system is registered as a patent). that is based on the most updated medical guidelines and practices, updated research, experts' knowledge and ML algorithms. Due to the lack of relevant data in historical files needed for the system's training, a computer program that is capable to generates endless relevant virtual scenarios that reflect the distribution of the disease in the population, has been developed. The training process of the system is done in several graduated stages: 1. Creation of a baseline database by defining major and minor factors, the complex relations between these factors, and the estimated dynamic impact of each of them. 2. Simulated Expert Labels. These labels are fed into a range of learning tools from various regression models. This stage has served as a guiding tool for subsequent data collection and algorithmic development. Using active learning techniques, the scope of samples needed to improve the model quality can be defined. In addition, the AI system can reveal correlations between patient profiles, treatment protocols, and patient outcomes. The system works by inputting all the important factors regarding the specific patient who is facing a CLL treatment. This is done by providing factors to an interactive intelligent chatbot via a computer or by extracting data from medical files. The system then integrates the patient's anonymous clinical profile with the medical guidelines and research, practices, experts labeling, and ML algorithms, outputting a detailed report regarding the need for treatment, the reasoning, and the alternative pathway. The report presents all the factors that influenced the decision and their relative impacts. The system improves significantly the quality of patients' history records which can be used for prospective studies and to reveal treatments' efficiency as well as allows a better data flow and supervision in the medical institution.
Conclusion: An automated AI-based method and system that can provide real-time support for CLL treatment pathways has been developed and presented. Critically, the system also provides meaningful structured data that will allow prospective machine learning methodologies to reveal correlations between patient indicators, patient outcomes, and the effectiveness of treatment protocols. It is emphasized, that the system does not intend to replace the traditional discretion of professionals but rather to empower them with all the necessary information to improve patients' outcomes and reveal treatment efficiencies. In the future, we intend to deploy the system, as clinical decision support system for treatment of CLL/SLL. Collection of prospective data will allow to implement further research and find new correlations between patients' anonymous profiles, treatment protocols and patients' outcomes to reveal treatment efficiencies.
Disclosures
No relevant conflicts of interest to declare.
Author notes
Asterisk with author names denotes non-ASH members.
