Skip to main content

© 2021 Excellerate. All Rights Reserved

The Client

The client is one of the nation’s premier children’s hospitals, committed to identifying and implementing innovative technologies for enhanced patient care.

The Challenge

Healthcare organizations are seeking predictive analytics technology to identify patients at high risk of chronic conditions, complications, or expensive treatments. Predictive technology provides enhanced insights to healthcare providers in order to attain accurate diagnoses earlier for more effective, targeted interventions.

Almost 12% of all infants are premature. An alarming one in four premature newborns has congenital heart disease and almost 30% of infants are vulnerable to neurological damage. These statistics are indicators for critical care attention and are at high risk of developing periventricular leukomalacia (PVL). PVL is characterized by death of white matter brain cells close to the area called brain ventricle.

Our clients’ request was to develop a unique predictive analytics accelerator for PVL. Our predictive algorithm identifies vulnerable infants and alerts doctors to the proactive critical care interventions specific to each patient.

At the time of engagement, PVL detection was treated as a diagnosis problem. But PVL may not be apparent until months following a diagnosis. Each infant may experience symptoms differently. The most common symptom of PVL is a form of cerebral palsy characterized by tight, contracted muscles, especially in the legs. The symptoms of PVL may resemble other conditions, making diagnosis difficult.

One way to detect PVL is by comparing MRI images performed one week apart. At this stage it is very late to intervene, and the timing of the second MRI is crucial for intervention and appropriate care. An alternative method is a cranial ultrasound looking for cysts or hollow spaces in the brain tissue. However, these symptoms may appear immediately, complicating diagnosis. The objective was to integrate the nonlinear systems modeling, physiological knowledge, clinical expertise, and machine learning techniques into a unified rational framework that would be capable of contextually processing and analyzing the results of each of these methods to provide a more accurate and informative decision support.

The Solution

Excellarate developed a patient-specific predictive analytics accelerator for clinical decision support systems, which positively identified hidden patterns in complex clinical data. This provided insight into targeted intervention timelines to prevent or reduce the effects of PVL. Our innovative and proprietary algorithm combines machine learning, physics, non-linear dynamics, concepts of time scale, and clinical expertise enabling organizations to significantly improve care interventions and long-term outcomes.

The client requested a solution to identify hidden patterns to positively intervene when infants are at high risk of neurological damages such as PVL. Excellarate engineers successfully identified the complexity of the patho-physiological condition under study. We have the tools to model, simulate and analyze complex diseases.

Excellarate’s predictive analytics algorithm provides organizations the ability to identify patients who are at risk of complications by identifying timely interventions before chronic conditions associated with PVL occur. A new predictive analytics framework has been developed based on the vital sign measurements, blood gas, and lab results to predict the PVL occurrence after neonatal heart surgery. The solution applies wavelet transform on the data to extract time-frequency information from data, we also use nonlinear dynamics analysis to extract valuable information from the data. We then feed this data into a machine learning classifier which is capable of predicting the outcome.

The Outcome

Excellarate’s predictive analytics engine has been able to unlock targeted timelines of critical interventions  to prevent the onset of PVL. Our innovative and proprietary algorithm combines machine learning, physics, non-linear dynamics, concepts of time scale, and clinical expertise enabling organizations to significantly improve care interventions and long-term health outcomes.

The predictive algorithm identifies crucial data within the first twelve hours after the surgery. Furthermore, this advanced predictive accelerator provides actionable information for the physician to review and intervene. This capability has a success rate of 92% accuracy. Once implemented the solution is not only capable of producing precise predictions it also improves patient care through providing actionable information. Drastically reducing complications associated with PVL, chronic conditions, expensive treatments, and risk for readmissions.