1. University of South Carolina University of California, Davis
JUST A MINUTE:  The Effect of Emergency Department Wait Time on the Cost of Care
Lindsey Woodworth
University of South Carolina
James F. Holmes
University of California, Davis

2. BACKGROUND Wait times in U.S. emergency departments (EDs) are long
In 2013, over 17% of ED patients waited more than one hour to reach the physician
Long ED wait times are associated with adverse clinical outcomes (e.g., Fee et al. 2007)
If patients’ conditions deteriorate with time, then longer ED wait times could exacerbate treatment costs
Fee, C., Weber, E. J., Maak, C. A., & Bacchetti, P. (2007). Effect of emergency department crowding on time to antibiotics in patients admitted with community-acquired pneumonia. Annals of Emergency Medicine, 50(5),
Sun, B. C., Hsia, R. Y., Weiss, R. E., Zingmond, D., Liang, L. J., Han, W., ... & Asch, S. M. (2013). Effect of emergency department crowding on outcomes of admitted patients. Annals of Emergency Medicine, 61(6),

3. BACKGROUND
Annual U.S. healthcare expenditures exceed $3T
Highest estimates suggest ~10% of total healthcare expenditures exhausted in emergency care (Lee, Schuur & Zingk 2013)
The policy complication is that reducing healthcare expenditures generally compromises patients’ outcomes
Maybe targeting timeliness of care could both reduce costs and improve patients’ outcomes…

4. QUESTION
What is the causal effect of emergency department (ED) wait time on the total cost to care for a patient?

5. Costi = β0 + β1Waiti + β2X’ + εi
QUESTION
Empirical Challenge:
Costi = β0 + β1Waiti + β2X’ + εi
Acuity captured by εi and correlated with Waiti
In OLS models, β1 is persistently negative

7. AMBULANCE

8. TRIAGE
STATIONS

9. TRIAGE
STATIONS 1 5

10. TRIAGE WAITING
STATIONS ROOM 1 5

11. TRIAGE WAITING
STATIONS ROOM 2 4 1 5

12. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 2 4 1 5

13. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 2 4 1 5

14. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 2 4 5

15. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 2 4 5 3

16. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 2 4 5 3

17. TRIAGE WAITING
STATIONS ROOM
PHYSICIAN 3 2 4 5

19. TRIAGING
A one level change in a patient’s triage level can have a big effect on their wait time
How exact are triage level assignments?
Patients are perpetually bumped back as long as other patients arrive with lower levels

20. TRIAGING

21. TRIAGING NOTES: There is room for inter-nurse variation in triaging
Two triage nurses are always working concurrently
Patients’ acuities overlap between nurses
Triage nurses only assign triage levels
1 – And a one level change can have a huge impact on wait time

22. IDENTIFICATION
Instrumental variables, using the stringency of a patient’s triage nurse as the instrument for their wait time.
Assumptions:
Triage nurse stringency affects patients’ wait times
The stringency a patient is exposed to is effectively random
Exposure to a stringent triage nurse moves patients’ wait times monotonically

23. DATA
Medical Records
Presented at large urban academic medical center’s ED, July ‘08 – April ‘13 (ages 18-89)
Time stamps
Triage nurse ID
Triage level
Demographics
Billing Records
Reflecting all utilization throughout the course of patient’s stay in the health system
Aggregate charges‡
Aggregate payments
Final diagnosis
‡ Multiplied by hospital’s cost-to charge ratio

24. DATA Mean Wait Time 1hr 23min Total Charge $39,754 Total Payment
Descriptive Statistics
DATA
Descriptive Statistics
Mean
Wait Time
1hr 23min
Total Charge
$39,754
Total Payment
$6,334
Total Cost*
$6,395
Male
50.4%
Age
45.3 years
Triage Assignment:
Level 1
17.8%
Level 2
30.4%
Level 3
36.5%
Level 4
14.6%
Level 5
0.7%
N=187,149

25. DATA Balance in Mean Patient Characteristics Stringent=1
Stringent=1
(Treatment Group)
Stringent=0
(Control Group)
Share Male
50.5%
50.3%
Mean Age*
45.0 years
46.1 years
Final Diagnosis:
Chest pain, unspecified (ICD-9: )
3.4%
3.3%
Headache (ICD-9: 784.0)
1.7%
Abdominal pain, other specified site (ICD-9: )*
1.6%
1.4%
Urinary tract infection, site not specified (ICD-9: 599.0)
1.5%
Unspecified septicemia (ICD-9: 038.9)*
1.3%
N=134,884
N=52,265
* Denotes statistically significant difference in means

26. ESTIMATION

27. ESTIMATION
2SLS identifies LATE (estimates apply to patients at thresholds)
To capture effect heterogeneity, stratify sample by neighboring triage levels
L1 L L L L5

28. ESTIMATION
2SLS identifies LATE (estimates apply to patients at thresholds)
To capture effect heterogeneity, stratify sample by neighboring triage levels
To ensure monotonicity,
restrict to times without upfront care
L1 L L L L5

29. First-Stage Estimates
Wait Time (hours)
A. Level 1 & 2 Patients (n=89,789)
Stringent Nurse
0.0612
***
(0.0128)
F-Stat: 55.77
Mean Wait: 59min
B. Level 2 & 3 Patients (n=124,700)
0.1450
(0.0150)
F-Stat: 80.59
Mean Wait: 1hr 38min
C. Level 3 & 4 Patients† (n=17,433)
0.0642 *
(0.0376)
F-Stat: 17.17
Mean Wait: 1hr 54min
D. Level 4 & 5 Patients (n=28,520)
0.0287
(0.0299)
F-Stat: 28.49
Mean Wait: 1hr 21min
† Omits 6am-9:59pm arrivals (so Level 4 patients are never susceptible to upfront care).

30. X 2SLS Estimates ln(Cost) A. Level 1 & 2 Patients (n=89,789)
ln(Cost)
A. Level 1 & 2 Patients (n=89,789)
Wait Time in Hours
0.3045
(0.1444) **
[0.1795] *
Mean Cost: $10,822
B. Level 2 & 3 Patients (n=124,700)
0.2135
(0.0527)
***
[0.0714]
Mean Cost: $5,315
C. Level 3 & 4 Patients† (n=17,433)
0.1446
(0.2884)
[0.3107]
Mean Cost: $2,560
D. Level 4 & 5 Patients (n=28,520) X
Mean Cost: $819
† Omits 6am-9:59pm arrivals. (SE unclustered) [SE clustered by “shift”]

31. ROBUSTNESS CHECKS Sensitivity to controls
Descriptive Statistics
ROBUSTNESS CHECKS
Sensitivity to controls
Sensitivity to different measure of “cost” (i.e., payment)

32. EXTERNAL VALIDITY
30%
21%
L1 L L L L5

33. EXTERNAL VALIDITY
30%
21%
L1 L L L L5

34. EXTERNAL VALIDITY
Calculate back-of-the-envelope approximations for cost-savings from reducing wait times in EDs with different distributions of patient acuity
If reduce each patient’s wait time 1/2 hour…
Costs ~7% ↓ in the average U.S. ED
Costs ~10% ↓ in our relatively high-acuity ED
■8(−(0.30/2) $𝟏𝟓,𝟐𝟐𝟓 〖𝑆ℎ𝑎𝑟𝑒〗_1−(0.21/2) $𝟖,𝟐𝟑𝟗 〖𝑆ℎ𝑎𝑟𝑒〗_1+$𝟖,𝟐𝟑𝟗 〖𝑆ℎ𝑎𝑟𝑒〗_2+$𝟐,𝟖𝟖𝟑 〖𝑆ℎ𝑎𝑟𝑒〗_3+$𝟖𝟑𝟖 〖𝑆ℎ𝑎𝑟𝑒〗_4+$𝟒𝟎𝟒 〖𝑆ℎ𝑎𝑟𝑒〗_5 )
The higher an ED’s share of high-acuity patients, the greater the opportunity for cost-savings
Starting from a mean wait time of ~1/2 hour for Level 1 patients

35. CONCLUSION
There is concern that long ED wait times might exacerbate costs
It is difficult to identify the effect of ED wait time on costs because patients are seen in order of acuity
We exploit the quasi-random assignment of patients to triage nurses (who differ in stringency) to identify causal effects
Average ED:
1-3% patients L1 (2%)
20-30% patients L2 (25%)
30-40% patients L3 (40%)
20-35% patients L4-5 (30% L4, 3% L5)
Our ED:
17.8% patients L1
30.4% patients L2
36.5% patients L3
14.6% patients L4
0.7% patients L5

36. CONCLUSION
Evidence that long ED wait times exacerbate costs, with effect intensification as acuity heightens
From a policy perspective, minimizing wait times could potentially both improve patients’ health and lower healthcare expenditures
Our results reiterate the importance of triaging; EDs could focus on precise triaging as a first step for cost-minimization
Average ED:
1-3% patients L1 (2%)
20-30% patients L2 (25%)
30-40% patients L3 (40%)
20-35% patients L4-5 (30% L4, 3% L5)
Our ED:
17.8% patients L1
30.4% patients L2
36.5% patients L3
14.6% patients L4
0.7% patients L5

37. Thank You Average ED: 1-3% patients L1 (2%) 20-30% patients L2 (25%)
20-35% patients L4-5 (30% L4, 3% L5)
Our ED:
17.8% patients L1
30.4% patients L2
36.5% patients L3
14.6% patients L4
0.7% patients L5

38. Descriptive Statistics

39. Sensitivity to Controls ln(Cost) A. Level 1 & 2 Patients (n=89,789)
ln(Cost)
A. Level 1 & 2 Patients (n=89,789)
M A I N S P E C I F I C A T I O N
Wait Time in Hours
0.3078
0.1856
0.2806
0.2347
0.4040
(0.1681) *
(0.1680)
(0.1513)
(0.1593)
(0.1575)
***
[0.2000]
[0.2279]
[0.2086]
[0.1950]
[0.1972] **
Mean Cost: $10,822
B. Level 2 & 3 Patients (n=124,700)
0.1925
0.3224
0.3860
0.1804
0.2404
(0.0579)
(0.0573)
(0.0640)
(0.0475)
(0.0516)
[0.0740]
[0.0879]
[0.0983]
[0.0642]
[0.0691]
Mean Cost: $5,315
C. Level 3 & 4 Patients† (n=17,433)
0.1884
0.0539
0.3364
0.0027
0.0824
(0.3306)
(0.2149)
(0.4426)
(0.1608)
(0.2596)
[0.3560]
[0.2366]
[0.5040]
[0.1702]
[0.2741]
Mean Cost: $2,560
Controls:
Time of Presentation ü
Clinical Characteristics
Payer
Replace $0 with $0.01 (affects ~1%) No
Yes
<1% of sample has $0 charges.
† Omits 6am-9:59pm arrivals. (SE unclustered) [SE clustered by “shift”]

40. X Alternative Measure of “Cost” (i.e., Payment) First-Stage 2SLS
First-Stage
2SLS
Stringent → Wait
Wait → ln(Payment)
A. Level 1 & 2 Patients (n=71,516)
0.0492
***
0.3020
(0.0137)
(0.2420)
F-Stat: 19.13
[0.2728]
Mean Wait: 59min
Mean Payment: $11,367
B. Level 2 & 3 Patients (n=91,496)
0.1304
0.1407
(0.0170)
(0.0773) *
F-Stat: 24.72
[0.0884]
Mean Wait: 1hr 38min
Mean Payment: $4,722
C. Level 3 & 4 Patients† (n=11,440)
0.0366 X
(0.0452)
F-Stat: 6.01
Mean Wait: 1hr 54min
Mean Payment: $1,847
D. Level 4 & 5 Patients (n=16,598)
0.0085
(0.0384)
F-Stat: 8.13
Mean Wait: 1hr 21min
Mean Payment: $484
† Omits 6am-9:59pm arrivals. (SE unclustered) [SE clustered by “shift”]