The ABCs and the P-D-S-A of the Intensive Care Unit Queue

Abstract

Department of Neurology, Albert Einstein College of Medicine, Montefiore Medical Center,Bronx, New YorkGlobally, there is significant variability inthe availability of intensive care unit (ICU)resources. Although the United States hasa high ratio of ICU beds to hospital bedsoverall, there are hospitals even in thiscountry where ICUs operate at or beyondfull capacity (1, 2). An unfortunate result ofthe high occupancy state at those hospitalsis recurrent delays in ICU admission fornewly critically ill patients, delays that havebeen shown to negatively affect clinicaloutcomes (3–5). Hospitals faced with thesedelays have implemented numeroussolutions to improve ICU throughput [e.g.,intermediate care units (6), telemedicineservices (7)] and increase the effectivenumber of ICU beds [e.g., creating “ICUswithout walls” (8) and medical emergency/rapid response teams (9)]; some havesimply built more ICU beds (10). Eachpossible solution, however, is resource-intensive. Moreover, data are lacking thatany of them consistently improves ICUthroughput. The optimal solution likelydiffers from hospital to hospital. Knowingbefore actual implementation whichsolution can best optimize ICU throughputin a given hospital setting would beimmensely helpful.In this month’s issue of AnnalsATSMathews and Long (pp. 886–894) report onthe development of a mathematical modelbased on queuing theory, which simulatesseveral solutions to address medical ICUovercapacity based on empiric data andcare practices specific to a single tertiary-care hospital (11). Queuing theory allowsfor the modeling of processes in whicha queue of customers (e.g., critically illpatients) are waiting for a service (e.g., carein an ICU bed) with limited capacity (e.g.,fixed ICU size) and, thus, serving a newcustomer (e.g., admitting a new patient tothe ICU) may require completion of serviceto an existing customer (e.g., discharge ofan existing patient from the ICU). Theconstraints of the model include thecapacity of the service (e.g., ICU size), aswell as how customers start and completethe service (e.g., the sickest patient comes tothe ICU first, the rate of ICU dischargedepends on momentary hospital census,etc.). Once the model is built, theconstraints can be easily altered to assessthe effect of changes (e.g., ICU structureand triage practices) on flow through thesystem.The objectives of Mathews and Longwere threefold: to describe the methodologyfor constructing such a model, todemonstrate its ability to simulate real-world experiences, and to use the model toevaluate and compare the effect of varioussolutions to improve ICU throughput attheir hospital. Bolstered by a detailed datasupplement, the authors succeeded fully indescribing how to construct a queuing-theory-based model with constraintsderived from historical data. They createda recipe others can adapt for application atother hospitals. They then demonstratedhow their model accurately simulates thestatus quo in their institution: wait time forICU admission (Figure 1), service timein the ICU (Figure E3a in the onlinesupplement), and time to transfer from theICU (Figure E3b). Such validation lendscredence to any additional results theirmodel will generate.Mathews and Long then used theirmodel to evaluate the effect of severalsolutions to improve ICU throughput.Specifically, they made changes toconstraints to allow for a differentbreakdown of ICU versus stepdown (SDU)beds, the ability to reserve the last availableICU bed for an acute patient, an increasein ICU bed number, and the ability totransfer any patient out of the ICU within1 hour. Building additional ICU beds, asexpected, results in shorter wait times forICU admission and lower overall ICUoccupancy. As the authors highlight, theability todecrease time totransfer outof theICU to 1 hour improves delays in ICUadmission to the same degree as buildingone to two new ICU beds does. Similar toICUexpansion, however,decreasing timetotransfer is accompanied by lower ICUoccupancy.Theseresultsmaynotbesurprising,yetthrough them one begins to understand thepotential power of the model: to uncoverfeatures of the system that can informnew potential solutions to improve ICUthroughput that can then be easily testedusing the model in an iterative, “plan-do-study-act” (P-D-S-A) (12) fashion(Figure 1). For example, in studying themodel results, we see a correlation ofincreased effective ICU bed number anddecreased ICU occupancy that suggests thatin this current hospital/ICU environment,there is not always a critically ill patientawaiting ICU admission when one is readyto be discharged. As lower ICU occupancywill likely lead to decreased revenue withlittle effect on cost (13), solutions with thisconsequence may not be financially viable.In acting on this information, it becomesclear that any solution may need tospecifically address the effect of improvedICU throughput on ICU occupancy.Planning a new solution of activatinga rapid time to transfer of 1 hour only whena patient is awaiting ICU admission may