Queuing Theory Misplaced in Hospitals
A carpenter with a nail gun can build a house. A nail gun in my hands predicts a medical emergency. I am not qualified to use a nail gun and the one time I tried, I nearly nailed my hand to a board.
One of my best clients was about to nail his hand to his desk with a queuing theory model. He summoned me to talk about the demand for ICU beds coming from the OR. He wanted help extracting 'controlled variation' from his data so that he could apply a queuing theory (QT) model to determine the number of ICU beds he needed. My first reaction was, "What!" Then, "Why????"
It seems my client had just spent a couple of days with IHI and Eugene Litvak, Ph. D., a professor of healthcare and operations management at Boston University and director of its program for the Management of Variability in Healthcare Delivery. He also is an adjunct professor at the Harvard School of Public Health. Dr. Litvak is a great proponent of the use of such quantitive approaches as the queuing theory model to address pressing hospital problems. (Click here to read about Dr. Litvak and Queuing Theory.)
As a manufacturing systems engineer, many might think I would be an advocate of such methodology. However, I believe, in most cases, queuing theory is misplaced and even misleading Here's why:
1. Complex tools require practice and expertise. Queuing theory derives its power from rigor and discipline. Today, even operations professionals with 20 years' experience struggle with queuing theory. To be effective, these tools require years of practice. A tool that works for a Ph. D. in operations research is not always a good tool for the typical hospital administrator.
2. QT has limited application. Queuing models are famously restrictive in their content and often are incapable of precisely modeling real-world situations. As Peter Drucker so aptly wrote, "…Hospitals are the most complex human enterprise…" Answers culled from queuing theory may not be particularly useful because they can't model the hospital's natural complexity. I find that, since hospitals have many scenario-specific considerations, simulation modeling is a more appropriate analytical tool.
3. QT has a limited useful life. These are capacity planning/decision tools. When administrators are unable to quantify how many beds they need, Dr. Litvak tells them they need queuing theory to determine the number. How often are you going to need that question answered? Let us say you have a ten-bed ICU and a queuing model suggests that a better number is 12 or 15; what is the administrator to do? Repeatedly asking the question will result in the same answer. This is because QT offers an extremely limited view of the alternatives. Hospital administrators need to look at other options to make the best decision for their hospital.
4. Applying the theory doesn't always work. Here again, the usefulness of the rigor is diminished. Hospitals add capacity in chunks. For example, rarely would it make sense to simply add one ICU bed, they open a whole unit. How do they determine how many units to add? They add as many as they can staff! But that doesn't mean this is an effective approach, even when the queuing theory says otherwise. Because the hospital operation is complex, adaptive, and interdependent on many variables, a broader look is necessary to find the right balance among efficiency, patient safety, quality care, and cost control.
5. Demand on the system is complex and interdependent. Hospital demand is enormously complex. You have to factor into the equation the biological variation, social/economic factors, the overall capacity of the market, the physician preference/patterns, payer mix, quality of nursing care in the units, availability of step-down, skill(ed) nursing units, and much more. This complexity does not lend itself to queuing theory, because the theory can't accommodate all of them, and all of them must be taken into account.
6. Service time in the system is complex and interdependent. Data on length of stay for most hospitals is largely inaccurate; the variance between actual and recorded times is generally off by several hours. In a typical hospital stay, service time is impacted by ease of diagnosis, the physician's course of action, ease of placement, use of provisional radiology/labs, availability of insurance coverage, availability for care at home, timeliness of a ride home, and even meal times. These interdependent variables make life for queuing theory analysts quite troublesome. I find my own abilities to define service time for QT models just about impossible in these circumstances.
Look beyond queuing theory, if you want solutions that really work. There are many quantitative approaches that can bolster your performance. For some ideas about what might work best for your hospital, e-mail me at john.dalesandro@gmail.com.
