Risk Prediction Models And ML Models For Cancer Prediction

Risk prediction models for Cancer are available in clinical practice in various parts of the world but it has been a challenge to adopt these models clinically. Multivariate prediction models use multiple variables to predict the risk of event or outcome of event. Till date, a lot of risk prediction models have been developed for a defined population, incorporating a variety of risk factors. The most identified drawbacks of these risk-prediction models are lack of external validation, differences in the measurement of factors, lack of calibration curves. Machine-learning (ML) models are more data driven, computation intensive and have better and more accurate prediction abilities. ML models have an enormous capacity to store the data related to individuals and analyze this data to predict the risk of an event in a specific individual or specific population. ML based prediction models are different from traditional statistical models in which it uses extensive clinical data, including comorbidity, medication, and many variables being considered. Owing to the rising numbers of cancer and other diseases, it is the need of hour to overcome the inactivity surrounding the use of these prediction models and machine algorithms. Research to develop novel ways of implementing these models into clinical practice, is being undertaken worldwide. An ideal prediction model or ML should predict the risk of developing cancer at least 3 years prior to the actual development i.e., the prediction should be able to provide enough time to modify the risk and the outcome of the risk over the time. Tree based models have been extensively studied and used and have been known to outperform other models like neural network, reason being that the tree-based models can analyze both linear and non-linear relationships between the variables of models. Critical Validation and many controlled clinical trials can be helpful in developing these prediction models better and the adoption of these models as well as ML models into clinical practice.