Risk of death is a big question for anyone admitted to a hospital. People tend to base their decision on which hospital to go to on its reputation, but this is an imprecise metric at best.
Despite the complex array of factors involved in a patient's risk of death, a new tool that uses only routinely collected data can accurately predict this risk in the year following hospital admission, according to a study in the Canadian Medical Association Journal. The tool could help medical researchers assess the efficacy of treatments and help patients make more informed decisions about where to go for the best care.
"You can compare this prediction to the observed [rate of survival] to figure out if that hospital or region or physician or whatever is performing better, worse, or as expected when you measure survival of those patients," lead author Dr. Carl van Walraven, a researcher at the Ottawa Hospital and the University of Ottawa, told Tech Times.
Researchers tested the model using data from the hospital records of over 3 million people, mainly Canadians. The information used to generate the predictions comes from health administrative data, which includes sex, age, severity of illness, number of illnesses and other routinely collected information.
Such information is not very detailed, but there is a ton of it because it is collected from every patient. The researchers found that, taken together, this data provided an accurate picture of a patient's risk of death under a given set of conditions.
"I like to tell people that these are big, stupid databases," says van Walraven. "But despite the lack of detail, we were able to use [these] data to come up with a model that accurately predicts the probability that a person will be alive within a year of their admission to the hospital."
The new tool could also help medical researchers parse out the effects of treatments.
"This model would make it very easy to control for all of these other things that might influence death risk, and after you control for those other factors, you can more accurately determine the association of your treatment with survival," van Walraven says.
While the model shows promising results so far, further research is needed before it can be implemented in a clinical setting, according to van Walraven.
"I believe that we'll be able to use this model in routine care, but I can't say that for sure because this study did not test that hypothesis," he says.
Photo: Bill McChesney | Flickr
