Using data analytics to proactively treat or even prevent infections in premature babies
Data analytics can support the doctor and nurse to make the best possible decisions. Daniel Vijlbrief Neonatologist UMC Utrecht
Better patient outcomes
UMC Utrecht champions data-driven clinical decision support to provide patients with the optimal care
Around 10 percent of all infants are born premature (<37 weeks of gestational age). Although the survival of premature infants continues to improve, their neurodevelopmental outcome remains a major concern. When a baby is born premature, it is admitted to the Neonatology Department (NICU) of a hospital. During this period, the baby is very vulnerable and prone to infections. To keep track of the health of the baby, it is connected to many devices. No other department in the hospital collects more data than the department of neonatology.
Having collected 10 years of patient data, Universitair Medisch Centrum (UMC) Utrecht, a leading international university medical center, felt the need to put the collected data to good use. Nowadays, machine learning and artificial intelligence (AI) are becoming more and more important for the healthcare sector. That is why UMC Utrecht started the Applied Data Science in Medicine (ADAM) project. With ADAM, the UMC Utrecht uses the well-known approach of large-scale data analysis to ensure that it chooses the treatment that is most beneficial to a specific patient.
The aim is to develop multiple models in SAS in order to better inform parents, provide the best possible healthcare for babies for a better long-term neurodevelopmental outcome and to eventually apply it to other intensive care departments. Manon Benders Professor and Head of Neonatology UMC Utrecht
Preparing the big data for analysis
Big Data for Small Babies was one of four projects within the ADAM program, initiated by the Board of Directors of UMC Utrecht. This particular project proposed the following question: Is it possible to proactively treat or even prevent an infection in premature babies using data analytics? This question needed to be answered within three months.
To ensure that the technical side of the project ran smoothly and timely, UMC Utrecht called in various data and IT specialists. The multidisciplinary Big Data for Small Babies team consisted of two engineers from IT consultant Finaps, SAS consultants and Daniel Vijlbrief, neonatologist and pediatrician at UMC Utrecht. They used SAS Enterprise Guide and SAS Enterprise Miner to make the data accessible (and anonymous). With SAS Visual Analytics they could analyze and visualize the data.
UMC Utrecht – Facts & Figures
of babies had been treated unnecessarily with antibiotics, a study revealed
accuracy in forecasting the presence of the bacteria that causes sepsis, using SAS
An algorithm that prevents unnecessary use of antibiotics
There were many different sources where data was stored. For this purpose, data analysts helped transfer and provide the data to Finaps. This was the first time data was extracted from NICU devices in a retrospective way. As a result, the team developed a smart algorithm, a statistical model with SAS that can support or deny the suspicion of an infection in premature babies, such as sepsis (blood poisoning).
This algorithm can prevent unnecessary use of antibiotics in premature babies. After all, sepsis is an important clinical problem, and UMC Utrecht had limited knowledge of the course of the infection, including the peak moment of the infection. In retrospect, 60 percent of the babies were treated unnecessarily with antibiotics, with all consequences that treatment entails, such as dependence to the medicine and financial implications. Moreover, every baby with sepsis has to be hospitalised one week longer.
Artificial Intelligence Technologies
SAS delivers artificial intelligence through these primary technologies: machine learning, natural language processing, computer vision, forecasting and optimization. Want to learn more about how these technologies play a role in AI?
These outcomes required further investigation. If the algorithm-driven model indicated a 75 percent certainty that the result of the test will be negative (or positive), doctors were prepared to consider it. Depending on the error margin, the model that was developed has an accuracy of 90 percent in forecasting the presence of the bacteria that causes sepsis. This is significantly higher than when doctors make predictions based on examination of the patient and present symptoms. In that case, the accuracy is 40 percent.
The new insights give UMC Utrecht the underpinnings it needs to base its decisions on. Prior to the SAS model, it was mainly based on the patient’s examination combined with gut feeling.
“This project shows that analytically driven solutions are capable of solving complex problems in the health sector,” says Nathan van der Lei, Business Engineer at Finaps. “In our opinion, the next step in healthcare is data-driven clinical decision support.”
What is a nice fun fact for data scientists can have huge implications for the sector. Tim Pijl Data Engineer Finaps