Data science – at the intersection of revolutionary medicine and technology – is transforming healthcare. BlueGranite’s team is at the forefront of this innovation. Collaborating with a noted pediatric neurosurgeon, whose expertise spans hydrocephalus, congenital disorders, cranial tumors, and epilepsy in children, our data scientists designed a machine learning analysis model to predict intracranial pressure spikes far enough in advance to allow time-critical therapeutic intervention.
Increased intercranial pressure can cause brain damage or even death. Brain swelling, bleeding into the brain, and fluctuation from fluids around the brain and spine can all increase intercranial pressure. Accurately predicting when these potentially life-threatening pressure spikes will occur can allow medical staff the time it needs to intercede.
Until recently, bedside monitoring hasn’t allowed doctors adequate time to respond prior to brain pressure surges; technology only accommodated reactive therapy. But medical intervention before a predicted spike can delay or even prevent the need for surgery, as well as prevent or minimize injury associated with these pressure surges.
BlueGranite worked with one of the top U.S. children’s healthcare organizations, and the affiliated pediatric neurosurgeon, to design and implement the brain pressure prediction model. The hospital system, with 638 beds, three hospitals, and almost 30 neighborhood locations, had a prediction prototype in place, but it only looked at data 15 minutes out, and there were issues with operationalizing the model.
Our team was able to build out a model that analyzes real-time patient metrics – continuous intracranial pressure, vitals, and medications over two hours – to predict, with extreme accuracy, intracranial pressure surges in children, 30 minutes, and up to an hour, in advance.
We used Apache Kafka to process streaming patient data through an Azure Databricks API we built, trained, and deployed. The successful, real-time solution is turning reactive therapy into proactive therapy. It’s also opening the door to future care improvements and deeper study.
BlueGranite’s successful proof of concept looks at quantitative, or numerical data. But now that the pipeline is in place, the neurosurgeon and his research partners hope to also soon begin exploring qualitative, or discovery-oriented, data (such as information gleaned from patient observation and interviews). Thanks to recent rapid advances, machine learning and natural language processing technologies can now assist researchers in coding textual data, making this type of research possible.