MedStar Washington Hospital Center researchers have developed a mathematical model for predicting which hospital patients will survive longer and which will not.

The model which will be presented Saturday at the American Institute of Bioengineering (AIBL) 2019 meeting in San Diego creates a combined database of clinically relevant hospital-like disease risk factors for each patient-including age sex hospital settings and viral load-to better understand track and compare outcomes of patients with the same disease for the first time.

The research team is using the Hospital Compare Plus a population-based hospital-delivered project that has been the subject of research in multiple hospitals globally. In this study they sought to determine the strength and directionality of the data and the progress of the disease process in 500 patients. Data included data on three specialties: cardiac radiology and orthopaedics.

The Hospital Compare Plus data which included demographic information age and sex was then linked to critical outcomes-most of them being categorized as critical mildly severe or moderate or very destructive.

The study team found that the mathematical model possessed a very large and robust predictive value of carrying predictive value. They demonstrated the model to complete 16 clinical trials. Six of the trials used this mathematical model showing the applicability to accurate clinical decision-making.

Recognizing that performing a given clinical trial involves carefully assessing results with a very large set of data the researchers developed a computational method to integrate this data and its use in the Hospital Compare Plus which is a publicly available dataset that is mined by patients hospitals researchers and doctors within one query.

This was a massive undertaking said Dr. Joseph Pulica lead author of the study and MedStar Washington hospital associate director for Translational Health Science at the Hospital Compare Plus.

Five years ago we needed to provide over 58 million in data to conduct this study. Now we have nearly 20 million in data that is more representative not only of our entire population but of the hospitals across our Healthcare System Pulica said.

The Hospital Compare Plus data was shared with the hospitals institutional network in December 2020 through the National Center for Computational Modeling Interpretation and Modeling Infrastructure (NCMMI). Within this NCMMI computer models can be shared typically with hospital systems. Despite the effort data sharing still took longer than anticipated Ku the program manager of the Hospital Compare Plus said. Once the unlimited data was made available the team was able to pool it and show a remarkable example of how software programs can use data gained from the study to answer real-world epidemiological questions.

It is like when you have to share your refrigerator because your refrigerator might not work anymore Pulica said. Now we cannot do this because if we do that it will create demand for the equipment that refreshes the information on the disease-related data. So we begin with the computational model do a rigorous analysis of all the data and if it works we share the results. If the model does not no data remains. The researchers looked at updates to their computational model (too-cute) to determine if the human-unique fully-robotic model had a better predictive value than the computer-generated model.

Our model is susceptible to errors Pulica said. And why shouldnt the model do better than the human-unique model? The replacement of our own model does not in any way impacted our models survival.