A fix for overcrowded emergency rooms

Photo by mah_​japan via Flickr

Last month, Atul Gawande had a pop­ular article in the New Yorker ques­tioning whether the hos­pital industry could learn a thing or two from the likes of the Cheese­cake Fac­tory. In the article, he points to qual­i­ties like man­age­ment over­sight and stan­dard­iza­tion common to large-​​scale restau­rant chain oper­a­tions as areas for hos­pi­tals to work on. Now, I’m not sure, but the Cheese­cake Fac­tory may also have some idea of how many cus­tomers will show up on a given night. Things like time of year, day of the week, hol­i­days, local pop­u­la­tion, etc., could all have an impact and I bet CF knows it.

This idea of fore­sight is cer­tainly some­thing that guides pro­duc­tion flow in large-​​scale man­u­fac­turing set­tings, according to North­eastern indus­trial engi­neering pro­fessor James Ben­neyan. In a recent article in Aca­d­emic Emer­gency Med­i­cine, Ben­neyan and his col­leagues apply the idea to the emer­gency depart­ment set­ting, as a way to improve patient flow and min­i­mize crowding. “Man­u­fac­turing set­tings improve flow by starting some pro­duc­tion early based on pre­dicted demand, rather than waiting for all orders to be placed,” write the authors.

The article com­pares three pre­dic­tive mea­sures — two com­pu­ta­tional models and expert opinion — to deter­mine whether patients walking into the emer­gency depart­ment will be admitted to the inpa­tient unit. If they can give the clin­i­cians in the latter depart­ment ample time to pre­pare, patients will more readily move out of the emer­gency depart­ment, scaling back a major problem that plagues hos­pi­tals across the country.

In con­trast to sim­ilar studies, which typ­i­cally pre­dict bed need based on indi­vidual patient admis­sions, the two com­pu­ta­tional models devel­oped here aggre­gate the indi­vidual patient pre­dic­tions into a “sum­ma­tive mea­sure” of near-​​future bed demand.

Of the three pre­dic­tive mea­sures, expert opinion was the least accu­rate. For the study, nurses per­forming triage were asked to cat­e­go­rize patients on their like­li­hood of admit­tance, based on a 6 point scale from “def­i­nitely yes” to “def­i­nitely no.” They tended to under­es­ti­mate whether a person would be admitted, and by exten­sion the need for inpa­tient beds.

The model that worked the best took into account patient age, pri­mary com­plaint, bed type des­ig­na­tion and mode of arrival (ie., by stretcher, wheel­chair or ambulance).

The authors account for the major lim­i­ta­tions of the study, noting that the data used to develop the models and test them were taken from only one hos­pital. Thus a clear direc­tion for future work, they write, is to see if the approach works sim­i­larly across a range of hos­pi­tals with varied needs and pop­u­la­tions. They have already begun this work.

Ben­neyan has long researched the strate­gies deployed in indus­trial man­u­fac­turing set­tings to health­care. He’s done the same thing for ski resorts, air­lines and other industry.

Just dri­ving slop and waste out of our health­care system would save a tril­lion dol­lars,” he said to me back in March. “But it’s a com­plex problem and it neces­si­tates a variety of approaches.”