From Data to insights through Machine Learning

Estimated reading time: 3 minutes

The world of medicine knows a lot and learns quickly about the human body. However, it can be complex to interpret the many variables that say something about a patient's health and status. These then need to be converted into usable and reproducible insights. In this blog post, we’ll tell you how AI and machine learning are used to turn data into actionable insights to improve outcomes for patients and heallth providers.

Getting Value from Data Through Model Development (Machine Learning)

‍Insights from data are obtained through a model. A model can take the form of a decision tree with many branches, where the branches are determined by the data and eventually lead to a prediction. A model is formed by learning from past data, meaning the branches of the decision tree are not explicitly programmed.

‍Machine learning is a branch of artificial intelligence (AI) in which computer systems are developed that can learn from data and improve without being explicitly programmed. Instead of following specific instructions, these systems learn patterns and insights from data to make predictions or decisions. .

From Decision Tree to Predictions

‍
Predictions follow from the decision tree. An example: the effects of a drug for a specific patient population varies between patient groups with different combinations of heart rate and blood pressure. These correlations form the basis of the decision tree, which can then be used to help future patients. You can make a prediction based on a decision tree by using the data of the patient for whom you want to make a prediction and following the branches. For the mentioned example, the likely effects of the drug can be predicted by measuring the heart rate and blood pressure of current patients, enabling a healthcare professional to anticipate.

Refining Accuracy and Performance

‍The accuracy of a model is tested with both historical and new data. This yields a performance metric called the Area Under the Curve (AUC). This metric indicates how well the model performs: the higher the AUC, the better the model is at making correct predictions. ‍

Enhancing Healthcare and Processes Through Insights: a case example of a Dutch hospital

‍A carefully designed medical model can help foresee risks in time and translate them into appropriate care for specific patients. A case example of decision support, where Pacmed has been involved in the past:
‍
In a Dutch hospital, a patient with a severe infection was in the ICU. The patient appeared stable, and the doctor was planning to discharge them. However, Pacmed’s AI decision support indicated a high risk of readmission or death. The healthcare provider decided to keep the patient in the ICU. That night, the patient had to be put back on a ventilator. If this deterioration had occurred in the general ward, an (emergency) readmission to the ICU would have been necessary. Emergency readmissions are not beneficial for either patients or healthcare providers.

What to Watch Out for in AI Model Development and Application

‍This is how we create powerful models to support healthcare and processes! However healthcare is vital for patients lives and quality of life. It is therefore of utmost importance to strictly monitor the functioning of these models.

In the next blog post, we’ll discuss the risks of applying AI in healthcare. Read more in: AI Basics: AI Risks and Solutions.

‍