1 of 22

EFFECTS OF RESIDENTS ON EFFICIENCY IN AN EMERGENCY DEPARTMENT

J. Silberholz, D. Anderson, M. Harrington, Dr. Jon Mark Hirshon, Dr. Bruce Golden

1

2 of 22

Overview

2

Broad Healthcare

Landscape

-Health Care Reform Bill, 2010

-Americans spent $2.3 trillion on health care in 2007

-Hospitals are one of the least efficient sectors

​

University of Maryland

Medical Center (UMMC)

UMMC

UMMC ED

700 beds

1,182 doctors

742 residents

55 beds

20% admission rate

46,000 patients/year

3 of 22

Residency Model

3

4 of 22

Research Question

What effects do residents have on the efficiency of the emergency department?

    • Total throughput
    • Patient waiting time

​

Residents are in the hospital to learn

​

One conjecture is that the teaching of residents takes time away from patient care and negatively impacts efficiency

4

5 of 22

Resident Seminars

  • Residents absent every Wednesday morning for a seminar
  • No replacement workers hired
  • Wednesday mornings provide a representative sample of all emergency department activity
    • Wide range of arrival rates
    • All types of patients and severities
    • Congestion levels vary as well

5

6 of 22

Simulation Model Overview

6

Patients Generated

    • Poisson process with varying rates

Triage Nurse

    • Severity and treatment parameters assigned

Waiting Room

    • Patients held until called back

Bed and Treatment

    • Patients called back according to multiple logistic regression
    • If the patient has not left before being seen, they are taken to a bed to be treated
    • Treatment time drawn from empirical distribution for this patient’s category

Discharge

    • Patient either discharged or admitted as an inpatient
    • Once the bed has been cleaned, a new patient is chosen by the logistic regressions

7 of 22

Patient Arrivals

  • Nonhomogenous Poisson Process

7

8 of 22

Patient Attributes

  • Sent to ambulatory zone?
  • Severity Score: 1(high) to 5(low)
  • Admitted to inpatient ward?
  • Triage time
  • Labs needed (Yes/No)?
  • Drawn from historical data
    • (database data Oct 2009 – Jan 2010)

​

8

Patient

1

2

Sent to AZ?

No

No

Severity?

2

4

Admitted?

Yes

No

Triage Time?

4 min.

8 min.

Labs Needed?

Yes

No

9 of 22

Patient Bed Selection

  • Discrete choice problem: which patient do we select?
    • Large choice set – 16 patient types
    • Multinomial logistic regression infeasible
    • Multiple logistic regression models used
    • Probabilities calculated for each patient type
    • Roulette selection used

9

10 of 22

Abandonment

  • Patients sometimes leave before they are called back
  • Simulation determines abandonment probability based on a function of severity score and time in waiting room

​

10

11 of 22

Patient Categories for Treatment Time

11

Yes

No

Yes

No

12 of 22

Model Validation

  • Compared patients per bed per day, abandonment rate, average time until first bed and total time in system statistics from simulation to those from historical hospital data

​

​

​

​

​

​

  • Also Kolmogorov-Smirnov test comparing total time in system distributions found no differences (p = .18)

​

12

Metric

Historical Mean

Simulation Mean

p-value

Patients per bed per day

2.38

2.39

.4413

Abandonment rate (in percent)

8.02

7.76

.4611

Time to first bed placement

(in minutes)

80.32

81.12

.8650

Total time in system (in minutes)

550.15

549.28

.9134

13 of 22

Experimental Design

  • Alter percent of both high and low priority patients seen by residents to test how decreasing resident activity affects the efficiency of the system

​

  • Performance metrics used
    • Time to Bed
    • Total Time in System
    • Throughput

​

13

14 of 22

Effect on Throughput

​

​

​

​

​

​

​

​

​

​

​

14

Increasing resident presence from 0 to 100% increases throughput by 6%

15 of 22

Effect on Total Time In System

15

Increasing resident presence from 0 to 100% decreases total time in system by 16%

16 of 22

Effect on Waiting Time

16

Increasing resident presence from 0 to 100% decreases waiting time by 35%

17 of 22

High Severity vs. Low Severity

17

Vertical contour lines imply that percent of high severity treated is the driving factor in efficiency gains – most of the patients treated are high severity

​

18 of 22

Effects on Time in System

18

Again, the vertical contour lines imply percent of high severity patients seen is the driver for gains in efficiency

19 of 22

Different Patient Populations

19

Residents have a bigger impact on more severe populations

20 of 22

Results

  • Residents do have an impact on the efficiency of the ED
  • Contrary to commonly-held intuition, residents add efficiency to the system
  • Strong linear relationships found between percent of patients seen by residents and throughput metrics

​

20

21 of 22

Discussion

  • Strong linear trends showing increasing efficiency with residents present
  • Most important for high severity patients to be seen by residents
  • Decreases patient service times slightly, which leads to significant decrease in time to bed for low priority patients, as there is more slack in the system
  • Contradicts the notion that residents themselves are a source of inefficiency – we do not compare them to other healthcare workers

21

22 of 22

Future Work

  • Quantify effect of each additional healthcare worker and compare nurses and nurse practitioners to residents
  • Model doctor decisions explicitly, show how they move through ED
  • Identify bottlenecks in the system
  • Include data gathered in person to help model doctor movement

22