SANTA CRUZ, CA--The movement to computerize patient records in a growing number of hospitals is paving the way for the use of sophisticated statistical methods to assist doctors' decision making. The National Institutes of Health has provided $1.35 million to a team of researchers working to develop new statistical approaches that could dramatically improve the care for severely ill newborn babies. These new methods could eventually be applied to all patients, radically changing the face of hospital care, the researchers said.
"We are at the forefront of a new wave in medicine," said David Draper, professor of applied mathematics and statistics at the University of California, Santa Cruz.
Draper is providing statistical expertise for the project, which is led by Dr. Gabriel Escobar, a pediatrician at Kaiser Permanente Medical Center in Walnut Creek and director of the Perinatal Research Unit at Kaiser Permanente's Division of Research in Oakland.
The researchers will develop methods that can be incorporated into automated medical records to monitor very ill and critically ill newborns. Their hope is that these methods will eventually take advantage of Kaiser Permanente's decision to deploy an automated medical record system, Escobar said. Electronic records will make it possible to perform instant computer analyses of an individual patient's medical data, while also giving researchers access to an entire database of medical records for similar patients.
In particular, Escobar's team wants to improve treatment strategies for newborn babies potentially at risk for infection. Their approach would permit a bedside computer to estimate a newborn's risk by combining the baby's data with information from previous newborn patients. Based on the baby's current condition and how other babies fared in the past, computer algorithms could provide physicians with accurate estimates of the probabilities of a variety of outcomes, including death or the need for highly invasive treatments. Doctors could provide early intervention if the analysis indicated a high probability of an adverse outcome.
Researchers envision that in the future such algorithms could even suggest treatment options to doctors. Draper, who has been working on statistical approaches to medicine for 20 years, said this is "by far the most exciting development that's come along in quite awhile for making optimal use of medical information to improve clinical care."
Draper and one of his former Ph.D. students, Milovan Krnjajic (now at Lawrence Livermore National Laboratory), are working on the statistical tools and algorithms that drive the entire project. Their methods use Bayesian statistics, an approach Draper predicts will become "the dominant statistical paradigm for the 21st century."
The Bayesian approach provides a rigorous mathematical framework for combining new information with existing knowledge to assess a given situation. These tools will enable the system to calculate a likely prognosis for a newborn based on the past history for that baby and the medical histories of other babies. Armed with the experience of an entire database of case histories, doctors will be able to make better-informed decisions.
According to Escobar, the statistical methodology proposed by Draper was a key factor in enabling them to win funding for this project.
"Working with him has had a tremendous impact not only on this project, but on the entire way I approach my research," Escobar said.
Quantitative methods, like the Apgar score developed in 1952, have revolutionized newborn care. But there are still major shortcomings, Draper said. Existing approaches to quantitative outcome prediction are limited in that they use only a fraction of the available information and only for fixed time durations (usually either 12 or 24 hours). Generally, current algorithms examine a few variables in the time frame, select the worst result for each variable, and generate a single probability estimate. This severely limits their use in actual clinical situations, which constantly evolve, he said.
"Many sick babies die or become critically ill during the first 12 to 24 hours after birth," Draper said. "It would be far better to use information as it arrives."
The power of Bayesian statistical tools lies in their ability to readily take into account changing conditions. For example, current approaches only incorporate one value of heart rate and ignore whether it is rising, falling, or staying the same. The new methods will be dynamic, accounting for changes over time in variables such as heart rate or respiratory rate as they produce a statistical gauge of a baby's health.
Embedded in an electronic medical record, these methods could help a doctor decide whether to take certain risks when treating a severely ill baby. According to Escobar, only half of the hospitals in the United States are equipped to care for a severely ill newborn. Transporting the baby to another hospital is not only expensive, but it also poses inherent risks. On the other hand, failure to move a baby that needs specialized care could lead to brain damage or death.
"Should the doctor drop everything and get the baby out of there, or can he or she assume things are going well? That's what we need to know," Escobar said.
Newer statistical approaches will give physicians important information in making such critical decisions, and, as a result, could save lives, he said.
Bayesian methods can also prevent unnecessary and painful tests by giving a doctor a more accurate description of how sick the baby is and whether tests are really needed, Escobar said. Another goal of the project is to help scientists see how well common screening tests--such as counting white blood cells--work, he said. While the project is aimed at treating newborns, these methods can be applied toward all patients.
"The beauty of this project is that the approach we're going to be using could be generalized to other conditions," Escobar said.
Researchers will spend the first several years of the study period collecting and analyzing paper and electronic data from 340,000 newborns in 14 hospitals in northern California and Boston. They expect their statistical methods will be up and running in three to five years and could be embedded in automated hospital records in five to 10 years. Such systems will improve over time as the size of the database increases, generating more accurate estimates based on growing amounts of properly analyzed data, Draper said.
"Developing these methods is challenging, but it is likely that more and more hospitals will employ them in the future," Escobar said. "It used to seem like science fiction, but I don't think it is anymore."
Other members of the research team are coprincipal investigator Dr. Thomas B. Newman of UCSF, and Dr. Ellice Lieberman, Dr. John Zupancic, and Dr. Karen Puopolo of Harvard University. The project is funded by a three-year grant from the National Institute of General Medical Sciences.
Last reviewed: By John M. Grohol, Psy.D. on 21 Feb 2009
Published on PsychCentral.com. All rights reserved.