Using Predictive Analytics and Simulation to Improve Therapeutic Outcomes

Problem:

Peripheral neuropathy is a condition caused by chronically high blood sugar and diabetes. It leads to weakness, numbness, and pain in hands, feet, and other body parts. About 60% of all people with diabetes eventually develop this disease. To make sustainable treatment decisions and provide personalized care strategies, scientists, doctors, and insurance companies use tools for in silico clinical trials. With these tools, they make personalized computer simulations of patient treatment and predict the responses.

Pfizer, one of the world's largest pharmaceutical companies, asked Fair Dynamics, in collaboration with Health Services Consulting Corporation, to develop a platform that would help the company’s researchers test a new drug for patients with painful diabetic peripheral neuropathy. The platform would be based on previous clinical studies and act as a decision support tool, which could assess a patient’s personal parameters, prescribe drug dosage, and predict possible outcomes. The platform also needed to be flexible and have a user-friendly interface to allow inexperienced users to work with it.

Solution:

To create a predictive analytics platform, engineers needed to process raw data from different sources and categorize it. For this purpose, they integrated SAS data files and machine learning algorithms in an AnyLogic model. The algorithm grouped the data with patient profiles into six clusters with clustering variables, such as gender, age, depression history, and others. These parameters were essential when completing patient treatment programs.

To include new patients in the model, engineers used an AnyLogic agent-based modeling approach. It allowed users to set up a patient with predefined parameters similar to those in the clusters. The patient would then fall into one of the identified clusters depending on these parameters. AnyLogic capabilities for parallel computation also offered simulation of scenarios with multiple patients using the parameter variation experiment.

Following categorization, the treatment process of a patient was simulated in the model with multiple instances. It was based on the data from the previously clustered patient profiles. To validate the model, the 4-6 weeks treatment for each patient was simulated.

Doctors were finally presented with the optimal treatment scenario and dosage for a patient. For each patient or cluster, users could export dynamically created reports.

As the model was supposed to be used by inexperienced people, engineers used Java technologies, supported by Anylogic, to complete the convenient interface.

video

Outcome:

In this project, AnyLogic acted as a software tool for integrating various datasets, machine learning algorithms, and simulation capabilities. Altogether, they allowed the processing of diverse historical data and its regrouping into unique clusters. With AnyLogic agent-based modeling, engineers managed to complete an easily configurable predictive model and simulate personalized treatment processes with great precision. The model helped doctors make informed decisions on drug dosage for every patient and see how he or she would respond to the treatment. With Java-based design elements, the model’s interface became more intuitive and could be easily understood by new users.

video

Project presentation by Luigi Manca, Fair Dynamics

Similar case studies

More case studies