Researchers refine a mechanical approach to a medical problem
More than 500 premature infants are cared for each year at the MU Children’s Hospital Neonatal Intensive Care Unit (NICU), carefully monitored by a fleet of doctors, nurses and specialized equipment attuned to the most sensitive conditions of a child’s early life.
Neonatal care is a labor divided between man and machine, but orchestrated overall by the sharp intuition of nurses who must promptly recognize the precise needs of perhaps a dozen children at a time. It’s a mission famously taxing.
In a development that promises to relieve some of the constant attention demanded of neonatal care providers, University of Missouri mechanical engineers have worked with a neonatologist to develop an innovative method for modeling, monitoring and adjusting a crucial component of an infant’s early development: blood-oxygen saturation.
Because the lungs only fully develop during the final weeks of pregnancy, blood-oxygen levels in premature infants can become dangerously low if left unassisted, a condition known as neonatal respiratory distress syndrome (RDS). Doctors can compensate for the low saturation by delivering oxygen through a nasal cannula, but that comes with the risk of delivering too much oxygen, the consequences of which can be equally damaging to a baby’s health.
“Babies have to be screened for this all the time,” said neonatologist Ramak Amjad, one of the project’s principal investigators. “If they aren’t treated, they run into all kinds of complications with their development — blindness being the biggest problem.”
An MU adjunct professor of research in child health, Amjad currently serves as director of the neonatal intensive care unit at Sacred Heart Children’s Hospital in Pensacola, Fla.
As a complication of RDS, an infant’s circulatory system effectively overdevelops, forming an excess of blood vessels that can strain surrounding tissue. In the eye, it’s enough to detach the retina from the inside, resulting in a form of blindness called retinopathy of prematurity (ROP).
The disease is responsible for blindness in up to eight percent of premature newborns, the most famous case being that of jazz musician Ray Charles.
“One of the ways to reduce the risk of ROP is to monitor a baby’s oxygen levels,” Amjad explained. “It’s a pretty continual process and involves a lot of strategic adjustments on the part of the nurse.”
During pregnancy, oxygen is delivered to the baby through the umbilical cord, from regions of high saturation (the mother) to low saturation (the baby). While a healthy adult will maintain close to 100 percent blood-oxygen saturation, children in the womb develop at around 90 percent, only reaching full saturation after taking their first real breaths at birth.
Infants born prematurely are outfitted with a device called a pulse oximeter, an optical sensor attached to the foot that measures blood-oxygen saturation. Typically, a simple valve regulates airflow between canisters of pure oxygen and sterile medical air. Nurses then interpret pulse-oximeter readings and adjust the valve manually.
“And every baby is a little different,” said Amjad. “But, when a nurse is taking care of a patient, they [the baby] tend to fall into certain patterns that are unique. They shift in and out of the safe zone sort of predictably, and nurses get good at recognizing those patterns. But, maybe a new nurse comes on shift and has to pick up where the previous nurse left off without knowing the infant’s patterns as well as the first.”
Blood-oxygen saturation is a problem of consistent care. In an attempt to help automate what is normally an exercise in a nurse’s good judgment, Amjad teamed up with MU Mechanical and Aerospace Engineering Associate Professor Roger Fales and a graduate student working in his lab, Bradley Krone.
Their paper, “Model of neonatal blood oxygen saturation,” presented at the 2011 Dynamic Systems and Control Conference in Arlington, Va., described a new set of models, developed by Krone and Fales, that guide the operation of a new control box designed to keep infants at their optimal oxygen saturation levels.
“Ultimately, this research has resulted in a new device,” said Fales. “Now that we have the models, we can implement them on an automatic control system for the oxygen, which hasn’t really been tried before.”
Amjad, Fales and Krone gathered oxygen saturation data from over 40 NICU patients, charting their oxygen levels and responses to change over the course of several hours. Using a set of genetic algorithms to compile the data, the engineers developed a virtual model that closely mimics the patterns and uncertainties of an infant’s respiratory cycle.
“We use the models to design control systems that we can test offline,” Fales said. “If we can develop a model where we understand the dynamics — which has been hard to do for infants, because they’re difficult to gather data on — we can design a control system that will respond to those dynamics without putting a patient at risk.
“Since we actually have a reliable model now, and have analysis of the uncertainty of that model, we can design automated control systems that will adapt under lots of different conditions,” said Fales.
In short, neonatal oxygen levels are not a simple game of input versus output. Current control systems already have the ability to respond immediately to minute oxygen changes, but those quick adjustments can be jarring to an infant. The researchers had to come up with a system that could discern when, if and at what rate new oxygen levels needed to be introduced. They needed a dynamic model.
“Every baby is a little bit different, and as they mature their dynamics change over time, which is what we tried to capture with our clinical data,” said Fales. “Reacting to those changing conditions and being able to have the control system reconfigure itself for that particular situation — that’s the hard part.”
A control box, guided by the computer model, operates a servomechanism that balances inputs between oxygen and medical air supplies.
Daniel Quigley, an MU graduate with a degree in mechanical engineering, joined the project in 2011 for his master’s thesis. He said a control system like this is tricky because it needs to take more of a biological approach to the problem.
“That was what really interested me: how do we make a machine solve a problem like a doctor or nurse would, as opposed to more of a mechanical approach engineers are used to?”
Quigley, a Mizzou track athlete at the time, had an interest in the physiology of oxygen consumption. His research helped marry the engineering process to what doctors and nurses recognize as the best practices for neonatal infants.
“A lot of times, in engineering, we get caught up in trying to make a control system work as quickly and as perfectly as possible, but in this case, having a fast response wasn’t necessarily a benefit to the baby,” Quigley explained. “If we’re trying to regulate the infant’s blood oxygen saturation in this really quick way, we found that we would run into a lot of physiological problems and could actually be increasing the stress on the infant. Adjustments have to be more gradual and patient-specific.”
Previous attempts at designing oxygen control systems such as this actually resulted in an increase in the number of times that neonatal infants would find themselves out of the oxygen safe zone. Although the data might have looked better in terms of time spent within the safe range, the number of drastic fluctuations increased, which medical research suggests is just as harmful.
“The model is incomplete,” said Quigley. “And sometimes when engineers are working on a ‘black box’ of sorts, we think of it in terms of input and output and compensate for some of these unknowns.”
Despite the volume of information the research team has collected, there are still plenty of unknowns that cannot be accounted for. Many respiratory processes are considered immeasurable because collecting that data would be dangerously invasive.
“The model is incomplete, but we do have a system that is automatic and, in a way, smarter,” said Fales. “So, in the end, even under certain circumstances the machine can’t do quite as well as a human, but the fact that it’s always working makes sure that it does a better job of never letting the error get out of hand.”
The rest, the researchers say, is still a matter of intuition.
“Nurses are typically very good at this, but it’s a heavy workload,” said Fales. “If you can automate this, you can take away some of that constant attention to the simple stuff and focus your efforts on the more unusual things that might be going wrong.”