skip to Main Content

Understanding patient’s health trajectories

In the fight against cancer, prevention is one of the most potent weapons. Not only the analysis of current and past data are crucial for the diagnosis and treatments of patients, but also the estimation of their value over time can provide key information and support. Among the main estimations of patient data over time, of main interest for supporting clinical decisions, are patients’ health trajectories.

In this article, we explain how the patient’s health trajectories can improve diagnosis, treatments, and prognosis. But first of all, what is a patient’s health trajectory?

Imaging, we want to conduct a study to estimate the life expectancy of breast cancer patients. So, we start the trial by recruiting cancer patients; let’s say that we reached 100 patients. On the one hand, we can confidently say that the probability that the patient in such a group is alive at the beginning of the study is one –the maximum. On the other hand, the likelihood that the patient in the same group lives forever is zero –the minimum. A patient’s health trajectory then, or survival trajectory, is the value of such probability over time –between the two extreme points. In other words, the estimation of the patient’s survival probability during the study and beyond.

Why patient trajectories analysis is needed?

The patients in the trial can be grouped for types of treatment, cancer stage, and many other covariates, which helps physicians and healthcare professionals clinicians to compare and analyse their trajectories fairly.

The analysis of such trajectories highlights the patient’s response to a particular treatment, the relapse probability based on time, and the prognosis. In addition, patients’ health trajectories enable identifying cohorts of patients according to their risk level. Therefore, clinicians can leverage such a support decision tool based on the patient’s trajectories and cohorts.

How we can compute patients’ health trajectories?

There are several statistical models to estimate a patient’s health trajectory. For instance, Kaplan-Meier (KM) is one of the most used to assess group health trajectories. Cox Proportional Hazard (CPH) is another statistical model that provides individual trajectories and the respective feature importance. Finally, advanced AI-based models, particularly deep learning models enhance the result’s accuracy. However, deep learning models can perform poorly in terms of explainability. Besides the choice of the model, the quality and quantity of the data remain the main priority to ensure the quality of patients’ health trajectories.

Author: Gaetano Manzo, Applied Machine Learning Post-Doc Researcher at HES-SO (University of Applied Sciences and Arts Western Switzerland)

Back To Top