Oncology’s AI frontier: crafting survival prediction tools
In 2020, cancer was responsible for around one in six deaths worldwide according to the WHO. Cancer’s high mortality rate can be explained in part by its heterogeneity: it’s not a single disease but a complex group of diseases, each with its own unique diagnostic profile and treatment plan. There are many types of cancer, all varying significantly in their manifestation, progression, and response to oncology therapies.
The genetic alterations that lead to malignant tumors are caused by many factors, both internal (genetic) and external (smoking, pollution, ultraviolet rays, etc.). Several statistical methods have been developed to predict survival in oncology. They rely on analyzing extensive patient data from medical follow-ups to identify factors affecting survival or mortality. Integrating these variables allows for the construction of a predictive survival curve for a patient, taking into account specific data such as age, weight, cancer type and stage, nationality, and socio-economic status. However, these models are limited to predicting the patient’s chances of survival until the end of their medical treatment.
Given that the survival of cancer patients depends on numerous factors, it is challenging for oncology researchers to accurately predict a patient’s post-treatment health trajectory. Mantu brand Amaris Consulting is harnessing its data analysis and artificial intelligence expertise to test new AI predictive modelling methods and apply them to extrapolate the long-term survival curves in oncology.
Artificial Intelligence is accelerating and innovating oncology research
Artificial Intelligence is not just a tool but a gamechanger in oncology research, offering a beacon of hope for personalized medicine. Developing a reliable predictive model for long-term survival in the field of oncology is exceptionally complex due to the vast diversity of factors that can influence outcomes. These factors can include the type and stage of cancer, patient’s genetic makeup, health history, treatment responses and lifestyle, among others. Each of these elements contributes to the complexity of creating a universal model that can predict accurate survival rates across the broad spectrum of oncological diseases.
This is where AI plays a crucial role. With its vast computational power, AI can process and integrate a much larger number of variables than traditional methods. This AI-enhanced approach uses medical extrapolation to predict a patient’s long-term survival prospects, extending predictions well beyond the period of their medical follow-up. The accuracy of these survival curves is crucial due to their far-reaching implications, such as their use in assessing the economic value of a treatment through cost-effectiveness analysis. However, current methods of extrapolating survival curves are limited, often yielding only approximate results because of the large number of factors involved.
AI, with its advanced computational power, holds promise for enhancing the clinical plausibility and precision of these methods.
Amaris’ teams, using a small well-documented patient sample, are compiling extensive data to create a matrix for developing predictive curves. This will benefit other patients lacking such comprehensive medical follow-up. By employing methods such as the CoxBoost regression model and the Random Survival Forests decision tree, Amaris Consulting is cross-analyzing data from multiple simulations to detect patterns of recurrence. These patterns will inform the identification of new variables for more accurate extrapolation of survival curves.
In short, Amaris Consulting’s project has two concrete applications. Firstly, it will facilitate the discovery of new prognostic markers for mortality or survival. Secondly, by producing reliable long-term survival curves it will significantly influence healthcare economics. Enhanced prediction accuracy will expand the application of these methods in evaluating the cost-effectiveness of treatments, ultimately benefiting a large number of patients.