Critical Decisions in the Emergency Department

University of Pennsylvania:

Brendan G. Carr, MD MS

Sage Myers, MD MS

Scott Lorch, MD MS

Patrick Reilly, MD

Dylan Small, PhD

Charles C. Branas, PhD

Agency for Healthcare Research & Quality

Ryan Mutter, PhD

How do we design and measure the emergency care system?

(Trauma as a case study)

University of Pennsylvania:

Brendan G. Carr, MD MS

Sage Myers, MD MS

Scott Lorch, MD MS

Patrick Reilly, MD

Dylan Small, PhD

Charles C. Branas, PhD

Agency for Healthcare Research & Quality

Ryan Mutter, PhD

Disclosures

• Federal research funding AHRQ, NICHD, CDC, NINDS www.traumamaps.org www.strokemaps.org

• AHA research funding NRCPR/GWTG

• National Quality Forum RECS Steering Committee

• HHS/ASPR Senior Policy Advisor I am not appearing in this role today

Conceptual Framework

• Ambulatory Care Sensitive Conditions Conditions for which good outpatient care can

potentially prevent the need for hospitalization, or for which early intervention can prevent complications or more severe disease.

• Emergency Care Sensitive Conditions Conditions for which rapid diagnosis and early

intervention in acute illness or acutely decompensated chronic illness improves patient outcomes

Background:The Volume-Outcome relationship

• 12 surgical proceduresCABG, AAA, TURP, etc.

• 1500 hospitals

Procedures Mortality=

0

0.2

0.4

0.6

0.8

1

1.2

Low Medium High

Cath Volume

Adj

uste

d O

R D

eath

Adjusted Oddsof Death

Sta

nd

ardi

zed

Mo

rtal

ity

70%

65%

60%

55%

50%

20 40 60 80 100 120

45%

Cardiac Arrest Patients Admitted to ICU/year

Hospitals that treated over 50 pts/year had

lower mortality

Cardiac Arrest Mortality

Background: The time-outcome relationship

Sepsis

STEMI

STEMI

30+% of STEMI patients get no reperfusion therapy

3% of ischemic strokes treated at TJC certified centers

3-8.5% receive rt-PA

Sepsis

Sepsis

5-7% of EDs perform EGDT

26% of physicians have used hypothermia (ever)

Volume

Time

• the organization of a system for the delivery of health care within a region to avoid costly duplication of services and to ensure availability of essential services.

– Mosby’s medical dictionary

What is regionalization?

Regionalized Trauma Care

Prehospital triage

The Trauma Model – Inventory

Access to trauma care

Trauma care outcomes

Trauma Model. All success?• 27,130,283 injuries treated in US hospitals in 2006

32% in trauma centers 68% in non-trauma centers

• Severely injured patients (ISS>15) - More likely to be treated in trauma centers (51.3% TC vs. 48.7% nTC, p<0.001)

• Critically injured patients (ISS>25) - More likely to be treated in non-trauma centers (41.6% TC vs. 58.4% nTC, p<0.001)

Research questions with policy implications

• Have we improved population outcomes for injury? 1. In a nationally representative analysis

– Do trauma centers save lives? 2. What is the relationship between

access to trauma care and injury outcomes? (supply and demand)

Research questions with policy implications

• (What can understanding population outcomes for trauma teach us about examining other systems created to focus on unplanned critical illness?)

StrokeSTEMICardiac arrest…

Q1. In a nationally representative analysis – Do trauma centers save lives?

• Population: All injured patients treated at trauma centers and non-

trauma centers in the US • Data:

Nationwide Emergency Department Sample (HCUP) Trauma Center Level (American Trauma Society)

• Geography Patient location, hospital location (US census, ArcGIS)

• Prehospital transport time estimates - empirically derived & arcGIS network analyst

Q1. In a nationally representative analysis – Do trauma centers save lives?

• Analysis Logistic regression

Survey weightsConfounders

– Age, injury severity, comorbid conditions, region, insurance, hospital size, teaching status, hospital ownership, (prehospital time)

Sub groups– Severely injured, penetrating, blunt, age > 55,

only patients surviving to admission

Characteristics of hospitals with ED encounters for injury - 2009

Characteristic PercentLevel 1 Trauma Center 12.96Level 2 Trauma Center 12.29Level 3 Trauma Center 9.63Non-trauma Center 65.12Public Hospital 15.73For-profit Hospital 14.24Not-for-profit Hospital 70.03Teaching Hospital 35.27Large Hospital 53.66Medium Hospital 28.82Small Hospital 17.52Urban Hospital 72.96

Characteristics of injured patients - 2009

Variable Level 1 Level 2 Level 3 Non-trauma

Demographics        Age (average) 41.36 44.40 43.75 44.11Male (percent) 57.64 52.64 51.38 50.66Medicare (percent) 14.77 20.34 20.57 20.33Medicaid (percent) 14.77 12.41 11.97 13.26Private insurance (percent) 34.76 37.57 38.23 37.11Uninsured (percent) 25.70 19.67 19.11 19.73Other payer (percent) 10.00 10.01 10.12 9.57Comorbidities        Has no comorbidities (average) 71.24 69.37 71.01 75.24Has comorbidities (average) 28.76 30.63 28.99 24.76Injury Characteristics        Injury Severity Score (average) 3.62 3.20 2.71 2.54Severe injury, ISS > 15 (percent) 3.69 2.29 1.02 0.64Blunt trauma (percent) 54.96 57.70 57.78 52.98Penetrating trauma (percent) 11.56 11.32 11.33 11.11Self-inflicted (percent) 1.16 0.88 0.69 0.57Assault (percent) 9.45 5.51 4.41 3.98

Relation between treatment at a level 1 or 2 trauma center and death

Population UnadjustedBeta

(t-statistic)

Adjusted Adjusted with Instrument

All injuries .00512**

(12.19)

.00109**

(4.81)

-.00194**

(-3.07)

Injuries with ISS > 15 .05859**

(17.05)

.00964**

(2.88)

-.04705*

(-2.54)

Blunt trauma .00583**

(12.85)

.00135**

(4.66)

-.00294**

(-3.08)

Penetrating trauma .00913**

(8.30)

.00225**

(4.16)

-.00009(-0.07)

Aged > 55 .01087**

(11.00)

.00339**

(4.64)

-.00556**

(-2.75)

Only patients who survived to be admitted & with ISS > 15

.02880**

(6.86)

.01160*

(2.09)

-.06370*

(-2.00)

* P < 0.05, ** p < .01

Relation between treatment at a level 1 or 2 trauma center and death

Population UnadjustedBeta

(t-statistic)

Adjusted Adjusted with Instrument

All injuries .00512**

(12.19)

.00109**

(4.81)

-.00194**

(-3.07)

Injuries with ISS > 15 .05859**

(17.05)

.00964**

(2.88)

-.04705*

(-2.54)

Blunt trauma .00583**

(12.85)

.00135**

(4.66)

-.00294**

(-3.08)

Penetrating trauma .00913**

(8.30)

.00225**

(4.16)

-.00009(-0.07)

Aged > 55 .01087**

(11.00)

.00339**

(4.64)

-.00556**

(-2.75)

Only patients who survived to be admitted & with ISS > 15

.02880**

(6.86)

.01160*

(2.09)

-.06370*

(-2.00)

* P < 0.05, ** p < .01

Unmeasured confounders?

• Have not fully controlled for case mix?• Have not fully controlled for injury severity?

(no physiologic data)• The system is intentionally (and effectively)

regionalized the sickest and most complex patients are

taken to the highest tier centers

• Ideally, we would conduct a trial in which we randomize to treatment at a trauma center

But we can’t randomize…so… IV

Relation between treatment at a level 1 or 2 trauma center and death

Population UnadjustedBeta

(t-statistic)

Adjusted Adjusted with Instrument

All injuries .00512**

(12.19)

.00109**

(4.81)

-.00194**

(-3.07)

Injuries with ISS > 15 .05859**

(17.05)

.00964**

(2.88)

-.04705*

(-2.54)

Blunt trauma .00583**

(12.85)

.00135**

(4.66)

-.00294**

(-3.08)

Penetrating trauma .00913**

(8.30)

.00225**

(4.16)

-.00009(-0.07)

Aged > 55 .01087**

(11.00)

.00339**

(4.64)

-.00556**

(-2.75)

Only patients who survived to be admitted & with ISS > 15

.02880**

(6.86)

.01160*

(2.09)

-.06370*

(-2.00)

* P < 0.05, ** p < .01

Final model examining impact of trauma center on mortality*

Variable Coefficient t-statisticPatient Characteristics    Age .00005** 10.92Female -.00065** -8.50Medicare (private insurance reference) -.00003 -0.19Medicaid .00043** 4.38Uninsured .00087** 7.01Other payer -.00001 -0.13Injury Characteristics    Probability of death .74719** 44.32Intent – self-harm .00828** 8.27Intent – assault -.00218** -9.19Penetrating trauma -.00225** -11.93Hospital Characteristics    Not-for-profit (public ownership reference)

-.00038 -1.49

For-profit -.00005 -0.15Teaching .00129** 4.20Medium hospital (small size reference) .00361 1.83Large hospital .00106** 4.64Northeast region (West region reference) .00043* 1.99Midwest region .00008 0.37South region .00054* 2.49Trauma center -.00194** -3.07

* With IV

Question 2. Population outcomes for trauma

• Data Sources (trauma system - supply)US Census Data

– Location of residence at the level of the block group and county

CDC, American Trauma Society, Penn Cartographic Modeling Lab– Trauma center access

• Data sources (injury death - demand)National Center for Vital Statistics

– Multiple Cause of Death (MCOD) Data

Question 2. Population outcomes for trauma

• Methods Supply Side – Access to trauma care

Access to level 1/2 trauma center within an hour Demand Side – Injury Deaths

ICD codes to identify injury death location Population data to calculate injury death rate

• Analysis Examine relation between injury death rates and

access to trauma care using poisson distribution

Question 2. Population outcomes for trauma

• Results Supply

60 minute access to trauma care–84.7% of US residents–46.4% of US counties

Mean time to care = 43 minutes +/- 22 Demand

152,766 injury deaths in 2005 27,964 in counties without access within 60 min 124,802 in counties with access to care within 60 min

Counties without access to trauma care within 60 minutes had higher rates of injury death when compared to counties

with access to trauma care within 60 minutes (OR 1.24, 95% CI 1.18-1.30)

The relative risk of death increased at a rate of 3.4% for each 10 minute increase in time to trauma care

(95% CI 2.4% - 4.4%).

Q1. Next steps & remaining questions

• We have a dichotomous outcome variable but are using linear regression…

• Hard to estimate the strength of the instrument given survey design of NEDS Is differential distance unrelated to outcome?

• It would be nice to generate point estimates• The direction of effect flips – do you believe it?

Q2. Next steps & remaining questions

• What is the right geographic unit to sum outcomes to? (we used counties)

• Should we be targeting counts or rates of death?• Should we adjust for injury severity and case mix?

(is there systematic variability in severity by geography?)

Questions?