Emergency department (ED) visits in the United States have risen dramatically in the recent past, from 88.5 million in 1991 to 120.8 million in 2006, for a utilization rate of 351 visits/1,000 persons to 401 visits/1,000 persons.1 During this time, the number of U.S. hospitals and EDs has declined from 5,108 to 4,587, increasing the utilization at the remaining departments. ED directors have generally increased their staffing to accommodate these changes, but they have little information on selecting the best staffing model for money spent.
Traditionally, EDs have relied on attending physicians for their staffing needs. Other options include midlevel providers (MLPs) (i.e., nurse practitioners and physician assistants), who have had shorter and less complete training but work for lower salaries than physicians do. At academic institutions, emergency medicine (EM) residents can provide service, but they spend a large proportion of their time on non-ED rotations as well as on formal education. For a limited number of academic sites, fellows in advanced training are also available.
Other specialties have looked at the issue of academic productivity and staffing as well.2–8 The models of care provision in fields outside of EM do not necessarily match EM’s approach, and their results are not necessarily comparable. For example, care provision in primary care settings such as those of family practice is more similar, but it is less so in internal medicine and pediatrics, with their larger components of inpatient and consultation services, and it is even further afield in cardiology, anesthesiology, and radiology. There is no general consensus about the impact of teaching on productivity and costs in any of these specialties, and even within family practice, the literature offers different results.2,3 It is possible that educational costs in EM would be lower than the value of service provided, but this is by no means a foregone conclusion.
In 2005, Rhode Island Hospital opened a new ED, creating a situation that, as predicted, increased the patient volume by more than 20%, which meant that more staff would be needed. In planning for this staff increase, we sought to develop a software-based model to determine which combination of attendings working with or without residents and/or MLPs was most cost-efficient for incremental ED staffing. The rest of this report describes that process. Other factors, such as job satisfaction derived from teaching, patient satisfaction, or research, were not included, but the model can easily be expanded to incorporate such elements. While the model we created is designed to address academic institutions, it can also be adapted for use for nonacademic EDs by removing the equations with residents from the model.
In this study, we created a cost analysis, from the perspective of the hospital or faculty group (whichever entity is paying for the cost of physician staffing), of various staffing models for an ED. Much of the data for this model are modeled predominantly on the Society for Academic Emergency Medicine (SAEM) salary surveys.9,10 Information on physician, MLP, and resident salaries, hours, and productivities was sought from published literature and organizations such as SAEM, the Council of Residency Directors, and the American Association of Physician Assistants. Other literature was used where available and appropriate.11–20 When we found no clearly relevant literature, we used our experience.
Development of the model
As a cost analysis, the decision was made to produce results in dollars/patient seen as the best reflection of staff costs. Only direct/marginal costs were analyzed; it was assumed that the additional staff necessary did not require sufficient administrative support, overhead, or space to require structural changes in the nonclinical costs associated with care. The large costs associated with administrative overhead for a residency program’s educational efforts and faculty support were assumed to be already present; we did not model the costs of creating a residency program from scratch that would meet residency review committee requirements. Similarly, it was assumed that nursing staffing/staff costs would not be affected by whatever staffing model was chosen and would not contribute to the equations.
Dollars/patient were derived from the multiplication of variables relating to salary (dollars/year), hours worked (hours/year), and productivity (patients/hour). The attending-alone model is shown as
salMD/(hrMD × prodMD)
In the expression above, salMD = annual salary of the attending, hrMD = annualized clinical hours worked, and prodMD = patients seen per clinical hour worked by the attending physician.
In the other options, where there are contributions to the cost and productivity by residents and/or MLPs, adding the salary of the nonattending staff to the attending salary and using the productivity for the new combination of staff incorporated these variables. As an example, the scenario of one attending and one resident is expressed as
(salMD + salR)/(hrMD × prodMDR)
Here, salR = annual salary of the resident, and prodMDR = the combined patients per hour seen by the two together.
Because all patients were assumed to be seen by the attending, salary variables for residents and MLPs were normalized by hours worked compared to hours worked by the attending to derive “effective” salary for the combination for insertion into the equation. During the sensitivity analyses, the calculations would automatically be updated with a range of possible values.
Only direct economic variables were included in this model. If other variables are desired, the model allows the introduction of these. As an example, for EDs where MLP cases receive only partial payment, a variable for revenue received could be included, such as
(salMD + salP)/(hrMD × prodMDP) × (partial payment fraction)
Here, salP = annual salary of MLP.
Noneconomic issues such as educational impacts, patient satisfaction, or hospital political issues can be addressed as long as they can be reduced to a factor that has an impact on costs, revenue, or productivity.
Variable development and assumptions
The first and most important assumption was that there was sufficient patient volume to keep any number of staff, in whatever combination, fully occupied with clinical activities during their time in the ED. These patients were assumed to be “average,” that is, corresponding to the overall mix of patients seen in the ED. The corollary to this assumption is that adding staff would not otherwise affect throughput, overcrowding, or other ED factors, as in the situation where a section of the ED that was not used 24 hours a day, 7 days a week was opened for additional time, along with a “sudden” addition of new patients to the existing stream of patients. The second major assumption was that resident positions “under the cap” (i.e., positions for which the institution or faculty group would receive direct payments from the government to offset salary costs) were unavailable; additional resident positions would be purchased at full cost. If “free” resident positions were available within the cap and approved by the residency review committee, it was assumed that this would be the first option explored. Finally, revenues to the institution/faculty group from government sources for education and other indirect payments would not be altered regardless of which model were used to provide staffing. Our ED’s statistics are similar enough to the average results found in the SAEM salary survey—approximately 65,000 visits per year, 22% admission ratio, average LOS 278 minutes for all patients and hospital size of 453 beds—to use our results for variables that are not discussed in the literature.
One major cost that was not included in this model was malpractice insurance. If the costs of various options were different or were not included in the literature on salary structures, that could have an impact on the outcomes of this model. Each area of the country has sufficiently different malpractice trends that it is difficult to address this globally. Our ED uses a model where malpractice is determined solely by the number of patients, not by the number or mixture of providers; the choice of staffing mixture would not affect our malpractice costs. Other EDs may face different structures, and each could incorporate the structure it faces as a reduced/enhanced payment similar to the partial payment example listed above. We did not model the situation where MLPs could evaluate and manage patients without attendings’ oversight. In such a situation, the MLPs would function effectively just as attendings do. It is a common practice in low-acuity areas of EDs for MLPs to evaluate and manage patients independently, but this is not necessarily the case for higher-acuity sections. Because we were modeling an aggregate volume of average acuity, not just a high- or low-acuity section, we did not think this assumption (i.e., that MLPs could function independently) was accurate. If the expectation was that MLPs would have additional productivity beyond that seen with the attending, this can be added to the composite productivity as:
(salMD + salP)/([hrMD × prodMDP] +[prodPonly])
Here, prodPonly = the incremental productivity seen and released without attendings’ involvement.
Staff productivity and cost factors are listed in Table 1. We could find no literature that explored the relationship to total productivity as the numbers of supervisees increased; we assumed that each additional resident would only provide two thirds of the incremental productivity of the previous resident, because the faculty would be able to see fewer patients independently as the numbers of supervisees grew.21,22
The final outcome for these analyses was the direct staffing costs of physician and/or resident/MLP time for each patient seen. It was assumed that the attending had direct patient-care involvement with each patient. (Note: Because the use of fellows is not a typical solution for most EDs, we did not formally investigate that option.)
Data were inserted into Data 4.0 (TreeAge Software, Williamstown, Mass) to create a decision analysis tree (see Figure 1). All outcomes are expressed as dollars/patient seen.
In the base case scenario, the combination of one attending and one resident had the lowest cost/case at $74.91/patient. Most other options were within $5 of this choice. See Table 2 for full details. Using sensitivity analysis for the variables of hours worked, salary, and productivity, the breakpoints (where the cheapest option switches from one choice to another) between options are very close to the base case scenario. As an example, when varying attending salary, the breakpoint between one attending with one resident and one attending with one resident and one MLP is $211,000—just slightly above the base figure of $208,259. At an all-new attending’s salary, there are several choices within pennies of each other, while at the experienced attending’s salary the differences are greater.
Productivity of staff is a key element of the assumptions, with the least known about how the various parts interact. Figure 2 shows the sensitivity analysis for attending productivity, where the cut point between one attending with one resident and one attending with one resident and one midlevel is just below the base case of 2.1 patients/hour. At higher levels of attending productivity (above 2.5 patients/hour), physician staffing alone is the cheapest model. Figures 3 and 4 show sensitivity analyses for residents and MLPs. Again, the cut points are very close to that of the base case. Using two-way sensitivity analyses (not shown)—varying two items concurrently—the results remain the same, showing that the top choices are almost exclusively one attending with one resident or with one resident and one MLP. Only at the extremes of variable ranges do other options become cheaper, but only by pennies/case.
In this article, we report our analysis of various options for increasing staffing in a large, urban, academic ED. The major assumptions were that additional residents were not paid for by the government, there were enough patients to keep any number of providers busy, and only economic factors were important to making the decision on how to staff. In our model, the best choices for adding staffing to an ED were either to pay for additional residents in equal proportion to additional attendings hired or to pay for residents and MLPs in combination with attendings. All other options were within a few dollars/patient of each other except those of three residents or three MLP/attendings or two MLP/attendings. Regardless of these calculations, if the hospital/residency program were “under the cap” (able to increase number of residents and still receive full federal funding), increasing the number of residents clearly would be the best option. The salary or productivity levels that would change the optimal solution are very close to many of our base case assumptions; therefore, the optimal solution may in fact be one of the other options.
It is up to each ED director to determine whether the difference in cost/patient is “worth it” to choose between options. With the typical ED as noted in the SAEM salary survey seeing >60,000 patients/year, a difference of only a few dollars/patient may make a large total difference in the budget.
While this model explicitly uses assumptions that are valid for a large ED, any ED can use this approach in determining the best strategy for increasing staffing. Additional elements that may be important to each ED can also be added to the model, such as quality, patient satisfaction, education, and others. The dollar cost of choosing one option over any other is explicitly defined, allowing for better decision making for all ED directors and managers.
Results from a decision analysis are entirely dependent on the assumptions the model is based on and the values used for the variables. The major limitation to this analysis is the lack of literature on productivity. We could find no literature on MLP productivity outside of an independent practitioner model; there is no literature on attending productivity in academic settings outside of aggregate SAEM data, and even the SAEM data do not indicate the typical number of residents supervised by an attending.10 All of this created the need to use our own experience or make general assumptions. Our experience provides the data for these assumptions—at the time of the study, our faculty saw roughly 3 to 3.25 patients/hour in our main site (supervising an average of 2 residents/attending) and roughly 2 to 2.25 patients in our other site with minimal resident coverage. SAEM data show typical productivity of 2.86 to 3.34 patients/hour, consistent not only with our experience but with our assumptions on supervision.9,10
The literature on residents’ productivity matches our experience—most residents see roughly one patient per hour, with little variation across years of the resident’s experience.12,16–20 There are few data on relative value unit productivity for residents, but residents are expected to have graduated responsibilities. Taken with the little variation in patients/hour across year of training, this might suggest that more senior residents are seeing more complex cases or those more likely to require procedures and collaboration with consultants or admission to the hospital. We chose not to adjust for this because we assumed that all providers (attendings, residents, and MLPs) were “generic” and did not differ within their respective categories.
We did not model the ease of implementing any of these staffing solutions. We also assumed that all providers are equally easy to recruit. Both ease of implementing and ease of recruiting can significantly influence which approach might best be taken. This model also does not account for any teaching roles that residents might undertake, which decreases their clinical productivity (but may increase the attending’s productivity). Finally, we did not address the issue of MLPs’ seeing patients and releasing them independently of attendings’ supervision. Many hospitals allow MLPs to function in this manner for some or all of their patients, which would create higher productivity for the combination of MLPs and attendings.
Incremental staffing for an ED with a combination of attendings and residents in a 1:1 ratio has the lowest cost/patient. Other options are somewhat more expensive but the additional cost may be justified by ease of implementation. ED directors can use this model to determine the optimal staffing approach for their environment, especially when the economic issues are explicitly identified.
An earlier version of this manuscript was presented at the ACEP Research Forum, San Francisco, California, October 17, 2004.
2 Franzini L, Monteiro FM, Low MD. A cost-construction model to assess the cost of a family practice residency program. Fam Med. 1999;31:159–170.
3 Pauwels J, Oliveira A. Three year trends in the costs of residency training in family medicine. Fam Med. 2006;38:408–415.
4 Franzini L, Berry JM. A cost construction model to assess the total cost of an anesthesiology residency program. Anesthesiology. 1999;90:257–268.
5 Otero HJ, Ondategui-Parra S, Erturk SM, et al. Financing radiology graduate medical education: Today’s challenges. J Am Coll Radiol. 2005;3:207–212.
6 Jamadar DA, Carlos R, Caoili EM, et al. Estimating the effects of informal radiology resident teaching on radiologist productivity: What is the cost of teaching? Acad Radiol. 2005;12:123–128.
7 DeLia D, Cantor JC, Duck E. Productivity vs. training in primary care: Analysis of hospitals and health centers in New York City. Inquiry. 2002;39:314–326.
8 Franzini L, Chen SC, McGhie AI, et al. Assessing the cost of a cardiology residency program with a cost construction model. Am Heart J. 1999;138:414–421.
9 Kristal SL, Randall-Kristal KA, Thompson BM. 2001–2002 SAEM emergency medicine faculty salary and benefits survey. Acad Emerg Med. 2002;9:1435–1444.
10 Kristal SL, Randall-Kristal KA, Thompson BM. The Society for Academic Emergency Medicine’s 2004–2005 emergency medicine faculty salary and benefit survey. Acad Emerg Med. 2006;13:548–558.
11 Graff LG, Radford MJ. Formula for emergency physician staffing. Am J Emerg Med. 1990;8:194–199.
12 French D, Zwemer FL, Schneider S. The effects of the absence of emergency medicine residents in an academic emergency department. Acad Emerg Med. 2002;9:1205–1210.
13 Reeder TJ, Burleson DL, Garrison HG. The overcrowded emergency department: A comparison of staff perceptions. Acad Emerg Med. 2003;10:1059–1064.
14 American Academy of Physician Assistants. 2005 Salary Survey. Alexandria, Va: American Academy of Physician Assistants; 2006.
15 Welch SJ, Allen TL. Data-driven quality improvement in the emergency department at a level one trauma and tertiary care hospital. J Emerg Med. 2006;30:269–276.
16 DeBehnke D, O’Brien S, Leschke R. Emergency medicine resident work productivity in an academic emergency department. Acad Emerg Med. 2000;7:90–92.
17 Langdorf MI, Strange G, Macneil P. Computerized tracking of emergency medicine resident clinical experience. Ann Emerg Med. 1990;19:764–773.
18 Deveau JP, Lorenz JE, Hughes MJ. Emergency medicine resident work productivity and procedural accomplishment. J Am Osteopath Assoc. 2003;103:291–296.
19 Jasani NB, O’Connor RE. An analysis of emergency medicine resident productivity by patient load, relative value units and level of training. Acad Emerg Med. 1999;6:409.
20 Stone CK, Stapczynski JS, Thomas SH, Koury SI. Rate of patient workups by non-emergency medicine residents in an academic emergency department. Acad Emerg Med. 1996;3:153–156.
21 DeBehnke DJ. Clinical supervision in the emergency department: A costly inefficiency for academic medical centers. Acad Emerg Med. 2001;8:827–828.
22 Gerbeaux P, Ledoray V, Liauthaud H, et al. Medical student effect on emergency department length of stay. Ann Emerg Med. 2001;37:275–278.