(37c) Mathematical Modeling and Data Analytics Using WBC Populations for the Prognosis and Diagnosis of Acute Coronary Syndrome

The complete blood count (CBC) is one of the most common clinical tests, integral to the diagnosis, treatment, and monitoring of almost all diseases because it provides a simple high-level assessment of the health of a patient’s hematologic and immunologic systems by reporting an estimate of the current number of each type of blood cell circulating per unit volume blood. Increases or decreases in the counts of different cell types may indicate anemia, infection, malignancy or more. Most routine CBCs involve high-resolution and high-throughput single-cell measurements of the morphology of tens of thousands of blood cells, providing single-cell details of morphology, protein concentration, or other characteristics. This additional high-throughput information is not utilized for clinical purposes, which results in the squander of precious data. These single-cell characteristics reflect states of maturation, activation, production, destruction, and the perturbation of those processes in different disease conditions. If we can infer these states and their rates of change from routine blood counts, we can diagnosis disease earlier and more precisely.

Here we develop a mathematical model of white blood cell (WBC) population dynamics inspired by cellular mechanisms for this purpose. We first show that this model can be useful to distinguish healthy individuals from those with a range of acute disease processes, and we then show how the model can improve the risk-stratification of patients being evaluated for acute coronary syndrome. Instead of using one model parameter for the purpose of diagnosis and disease stratification, we use advanced machine learning techniques to build a robust, cross-validated classifier that can help with the classification of disease etiology in patients having the same external manifestation of symptoms. We can show how patients indistinguishable based on the CBC indices alone can be classified using the parameters depicting the population dynamics of WBC morphology. Our study demonstrates how mechanistic modeling of existing clinical data can realize the vision of precision medicine in a cost-effective way.