At many levels of medicine, it’s important for physicians to make predictions about their patient’s future. Will their disease or condition worsen? Will this treatment or that treatment be more effective in curing them? How much longer does a patient have to live? Such decisions are especially important for pediatricians in the NICU, the neonatal intensive-care unit, where infants born prematurely can be put in incubators and given oxygen for the early days of their life. In the gray area of extremely premature infants born between 22 and 25 weeks of pregnancy, important decisions need to be made by physicians and parents at birth and in the following days about whether to pursue all possible medical interventions for a child that may not survive. Reliable predictions are crucial for these decisions.
William Meadow, professor of pediatrics at the University of Chicago Medical Center, is dedicated to improving the accuracy of that important early-life prognosis. In a recent editorial, Meadow and colleagues from Children’s Memorial Hospital, the Medical College of Wisconsin, and Children’s Mercy Hospital found an unexpected guiding analogy for that effort: the world of sports betting. To illustrate the fluctuating odds of a premature infant’s survival, Meadow chose the competitions on the European football or soccer pitch, where life-or-death stakes are mere hyperbole.
“For premature babies, 24 weekers, we know that many do badly. But which “many” do badly? We need an algorithm to tell us this baby is probably going to do pretty well, this baby is probably going to do pretty badly,” Meadow said. “If we can envision an algorithm like that, then the world would be a better place.”
In sports gambling, the original odds are initially set using all the information available before the game begins. The record of the two teams, relevant statistics, recent streaks and momentum, injured players, and many more factors are part of the calculations of who is the favorite, and by how much. (Betting lines also take into account the influence of a team’s popularity on betting behaviors, but the authors put that factor aside for this analogy.) Similarly, neonatologists use a core set of factors at the time of birth to determine a premature infant’s chances of living, including gestational age, size, steroid levels, and sex. But oddsmaking, in both the sportsbook and the NICU, doesn’t need to end with the initial “betting line.”
“Both in the NICU and in football betting, the pregame odds of a particular outcome only get you so far,” the authors write.
In the boom industry of online sports gambling, the practice of in-game betting has grown more popular. As the game plays out and points go on the scoreboard, gamblers can bet on the eventual winner as odds change in real time. To test how the quality of predictions change as the game progresses, Meadow and colleagues (including his son, Xander Meadow), performed a common-sense experiment, testing how well the score at various intervals of a game predict the final outcome. Originally, they used data from more than 400 baseball games to test the premise. But when the British journal Acta Pediatrica agreed to publish the editorial, they asked that the analysis be repeated in a sport more familiar to their culture.
For both sports, as you might expect, the deeper you go into the game, the easier it is to pick the result. Two-thirds of the way into a match, the researchers determined that the score has become a better predictor of the outcome than the pre-game betting line. By the 80th minute of a 90-minute soccer match (or the 8th inning of a baseball game), the team that’s in front wins 94% of the time. What might be a “duh” result in sports can be instructive in predicting the outcomes for premature infants in the NICU, the authors write: “at some point during the NICU stay, we can know more about the likely outcome of that particular baby than we could ever have known before birth.”
But while the status of a game can be measured easily in terms of runs or goals, the status of a baby in the NICU is not judged so easily. Illness severity scores that collate several different measures into one number are often used, but a 2008 study published in Pediatrics by Meadow and colleagues found that they become less, rather than more, predictive of a baby’s outcome over time. Surprisingly, a much softer measure – the intuition of NICU nurses and physicians as to whether the baby would ultimately live or die – was more predictive of death or severe neurologic disability two years after discharge (but terrible at predicting whether the baby would die in the NICU).
Finding a more effective prognostic measure that improves, rather than decays with time, would greatly benefit decision-making in the NICU, Meadow said. Rather than making a final decision on how intensively to treat a premature infant at the moment of birth, a better prognostic system could help update a baby’s chances of living a full, healthy life as their status changes over the critical first few days and weeks. While what constitutes “winning” or “losing” in an infant my be more complex than in a soccer match, better information will hopefully make difficult decisions easier to bear for parents – and give some premature babies additional chances to live a long life.
“If, after a week or two, we could know with much more certainty (say 90% probability as opposed to 45% probability) that a particular baby was going to have a ‘bad’ outcome (e.g. death, survival with severe impairment), then we could direct the parents of that particular baby towards palliative care,” the authors write. “And, conversely, if we learned within a week or so that a different baby did not look any worse prognostically than the newborn in the adjacent crib who was two gestational weeks older, would not the parents feel good about having given their kid a chance to play?”
Meadow, W., Meadow, X., Tanz, R., Lagatta, J., & Lantos, J. (2011). The value of a trial of therapy – football as a ‘proof-of-concept’ Acta Paediatrica, 100 (2), 167-169 DOI: 10.1111/j.1651-2227.2010.02113.x
Meadow W, Lagatta J, Andrews B, Caldarelli L, Keiser A, Laporte J, Plesha-Troyke S, Subramanian M, Wong S, Hron J, Golchin N, & Schreiber M (2008). Just, in time: ethical implications of serial predictions of death and morbidity for ventilated premature infants. Pediatrics, 121 (4), 732-40 PMID: 18381538