Health Information Technology and Patient Outcomes

AHRQ Sponsored Evidence and Next Steps

Stephen T. Parente, Ph.D., University of Minnesota

Jeffrey McCullough, Ph.D., University of Minnesota

Jean Abraham, Ph.D., University of Minnesota

Martin S. Gaynor, Ph.D., Carnegie Mellon University

September 28, 2010

Presentation Overview Policy & industry context for

research

Data & empirical methods

Results & interpretation

Policy Prescription for more HIT and better (or at least more) clinical effectiveness data

What do We Know about National IT Impact Measured by Data in the US Actually, very little.

Studies generally extrapolate from case examples in a set of national sites.

Very little mainstream health insurer success stories.

The Issue With Regional Insurer EMR Cases Applied to the Nation… Hi, I’m a

PPO design and I have 85%+ market. I also rule the FEHBP market and TRICARE.

Hi, I’m a HMO design and I have -15% market. Oh, and I’m the model for EMR success stories in Colorado, West Coast, North Central PA and Massachusetts. I’m so ACO ready!!!

What if No One Wants a Trojan Rabbit?

Sir Bedivere the Wise: “Now once we have gotten all the physicians to buy a Stimulus Bill-financed Dell computers from Wal Mart for an EMR install, we can distribute the software to them to place more data entry onto their existing workflow and then pay them less when we use the system to tell them they are under-performing in their new ACO/medical home.”

Bonus Film Points: Also from 1975: Chouinard A. Shall I not ask for whom the [electronic] medical record is kept? CMAJ 1975. Start of SNOMED

Conceptual Model The conceptual model for our proposed

analysis is an economic model of technical production.

We assume hospitals produce a number of different outputs: quantity (Q) and quality (Z) subject to the following technical production relation, as

L: Labor K: Capital IT: IT Systems patient attributes affecting efficiency (e.g.,

severity) hospital specific factors 1st derivatives with respect to outputs are positive 1st derivatives with respect to inputs are negative 2nd derivatives with respect to inputs are positive

0),,;,,,,( ITKLZQF .

( 0, ZQ FF )

( 0,, ITKL FFF )

Data for Empirical Investigation We measure HIT value by combining hospital-

and patient-level data during 1997-2007. Sources:

Medicare inpatient admissions during our study period – the 100% MedPar inpatient Medicare claims data file. These data provide patient-specific outcomes and severity adjustment measures.

The Healthcare Information and Management Systems Society (HIMSS) Analytics Database provides detailed hospital IT adoption data for a variety of applications including: electronic medical records (EMR), nurse charts, and picture archiving communications systems (PACS).

HIMSS Analytics comprises a near census of acute care, urban, nonfederal US hospitals.

American Hospital Association’s (AHA’s) annual survey which describes hospital characteristics.

Econometric Approach - 1 We regress patient-level PSIs on a set of HIT variables,

patient-level controls, and hospital fixed effects. Each of the PSIs is a binary variable equal to 1 if an

adverse event occurred and zero otherwise. Control variables include patient age, gender

(female=1, else=0), race (non-white=1, else 0), risk score, and year of admission.

HIT variables are a set of three binary indicators for the presence of EMR, nurse charting, and PACS.

HIT variables were lagged by one year to reflect anecdotal evidence and expert interviews indicating that HIT value is realized one or more years subsequent to adoption.

Econometric Approach - 2 HIT value may change with time through

unobserved learning and innovation. We include a set of nine HIT-by-year interaction terms

allowing HIT to have a different affect in each year. interactions of the binary HIT application variables with

binary indicators for the years 2000, 2001, and 2002 respectively.

Finally, we control for unobserved hospital attributes by including hospital-specific fixed effects. Creates over 2,700 binary variables, one for each hospital in

the study. These fixed effects control for hospital attributes that are stable across time such as bed size and patient case load described

This design controls for unobserved time-invariant quality differences. Effectively, this specification controls for some types of selection in the HIT adoption process.

Descriptive Statistics

Hospital descriptive statistics, 1999 values

Variables Sample Average Early adopters Late adopters Non-adoptersBedsize 187 233*** 197 179Visits 129,619 173,272** 136,549 122,467COTH Membership 8% 18%*** 8% 7%Multihosp. System 69% 67% 72% 68%Nonprofit 71% 87%*** 71% 68%**For-profit 15% 1%*** 15% 16%Government 15% 12%** 14%* 15%% Medicare 47% 46% 46% 47%% Medicaid 18% 17% 18% 18%

Number 2,846 247 522 2,077* denotes significance at p=0.10, ** at p=0.05, and *** at p=0.001

BY EMR adoption

Health IT Adoption

0%

10%

20%

30%

40%

50%

60%

70%

80%

1999 2000 2001 2002

Figure 1. HIT Application Diffusion

EMRPACSNurse Chart

Descriptive Stats of Medicare Hospital Admissions in 1999, by type of IT Invested

Electronic Imaging Nursing IT No EMR/PACSPatient Attributes Medical Records PACS Systems Nursing IT

Age 73.4 73.4 73.8 74.0Female Gender (%) 56.5% 57.2% 57.2% 57.8%Non-White (%) 15.6% 10.8% 15.2% 16.4%Risk Score 2.39 2.38 2.30 2.18

Patient Safety Indicator (PSI) admission rate per 1,000 patientsAggregate PSI 8.628 8.407 7.831 6.794Complications of Anesthesia 0.191 0.135 0.244 0.204Infection due to Medical Care 3.061 2.893 2.804 2.389Post-operative Hemorrage or Hematoma 2.665 3.006 2.498 2.415Postoperative Pulmonary Embolism or Deep Vein Thrombosis 13.547 12.768 12.812 12.708Post-operative Wound Dehiscence 4.201 3.499 4.011 3.876

Hospital Attributes of Patients AdmittedNumber of Hospital Beds 368.5 340.0 331.6 293.4Number of Nurses on Staff 575.3 515.0 469.2 405.4% Medicaid Admissions 13.3% 13.2% 13.9% 15.9%% Medicare Admissions 47.7% 47.8% 49.3% 48.9%

Admissions in 1999 879,723 167,115 5,118,437 2,375,527

Based on a all Medicare patients treated at 2,707 US hospitals

Change in Patient Safety by TechnologyChange in Patient Safety by Technology (per 1,000 admissions)

All YearsElectronic Medical Records Infection due to Medical Care -0.490 *

Post-operative Hemorrage or Hematoma -0.240 *Postoperative Pulmonary Embolism or DVT -1.000 *

PACS Infection due to Medical Care -0.610 *

Post-operative Hemorrage or Hematoma -0.072Postoperative Pulmonary Embolism or DVT -2.000 *

Clinical/Nusing IT Infection due to Medical Care -0.390 *

Post-operative Hemorrage or Hematoma -0.009Postoperative Pulmonary Embolism or DVT 0.300

Notes: * denotes a reduction/increase in clinical error with t-statistic significant at p<.001

IT Total and Year-Specfic EffectsYear-specific and system-specific effect of Health IT on patient safety indicator, adjusted by patient risk and hospital specific attributes (per 1,000 admission)

IT in allYears 2000 2001 2002

Infection due to Medical Care Electronic Medical Records -0.2912** -0.0134** -0.0391** -0.0685**

PACS -0.3056 -0.1010 0.0533 0.0512Clinical/Nusing IT -0.0053 -0.0331 -0.0632 0.1479**

Post-operative Hemorrage or Hematoma Electronic Medical Records 0.0503 0.0866 0.1916 -0.1513

PACS -2.2799 -0.0079 -0.3930 -0.0645Clinical/Nusing IT -0.0213 0.0717 -0.0245 -0.0113

Postoperative Pulmonary Embolism or DVT Electronic Medical Records 0.0870 0.3299 0.9670** 0.5225

PACS 2.3452 0.3784 0.9017 0.8966Clinical/Nusing IT 0.1933 -0.2254 -0.2466 0.0188

*denotes significance at p=0.10, ** at p=0.05, and *** at 0.001Note: Health IT effects are lagged one year

Year-specific IT Impact

Main Findings EMR investments improve patient

safety by reducing infections due to medical care

Others (PACS & Nurse Charting) Health IT Systems are not as effective as EMR

EMR’s affect on patient safety grows with time

We find limited evidence wide-spread HIT value

Contribution & Policy Implications Demonstrates the use of large scale claims

data analysis to study health IT impact. More could be done: More recent years Other populations besides Medicare

Evidence suggest savings will not be quite as big as projected.

For new initiatives that are part of health reform, it will be critical for them to show their value using nationally generalizable data, since it is available for analysis and the fiscal stakes have not been higher.

Going Forward Use National Data to track how IT investments

are influencing measure outcomes with claims data linked to available clinical data TODAY (e.g., lab results and Imaging URLs) for appropriate CPT codes.

If Phase IV Post-launch drugs and devices can use claims data for monitoring effectiveness of treatments and avoidance of adverse events – why not the federal government with Medicare Parts A, B and D data.

Zhan & Miller (2003) set a great precedent for AHRQ to publish the code to measure medical errors. There needs to be far more efforts in this direction to gauge national impact of health IT.

Physicians

Congress Main Street Medical Technology

Courts

Federal Government

<90% Income

Insurers/Banks 99% Income 91-99% Income

Big Business

Hospitals

Today’s World What $30 billion better build

If Government Really Wants HIT Acceleration - Consider:

What: Federal/State health benefits require providers to pay using a national health card technology platform.

Government Why: Want clinical data attached to claim for de-identified comparative effectiveness data pipeline. Bonus – technology platform to mitigate prevent fraud as ‘pay for’.

Provider Why: I’ll get paid in 4 days and under for 90% claims

The Big How: Augment Medicare Administrative Contracts (MACs) for

2011-12 to include card use and require linked clinical data for approximately 100 HCPCS/CPT codes as pilot – more later.

Augment TRICARE contracts to do the same. States put contracts out for competitive bid following

TRICARE model. FEHBP buts this required specification as well.

Next Steps to Dead End:Integrated Health Care Demonstration Project R-

18 Proposed a Trial Integrated Health Card (combine

clinical data with claims transmission for new locations: University of Minnesota employees and dependents (>35,000

lives) Minnesota Care (Minnesota’s variant of state Medicaid

program) Additional support

Metavante – Issues card technology / payment hub (without clinical link) for >30K lives including all of Minnesota’s public programs and other state programs – would provide all demonstration technology and consulting support gratis.

Comments from Study Panel Reviewers (paraphrased): 1) Claims data is inferior for measuring outcomes and should

be not be encouraged as a platform because it is not consistent with an ACO.

2) The researchers are well regarded but the technology (e.g., Medavante - >$10 billion firm) does not exist.

3) Great idea – go for it – Best ebayer ever A+++++++ My question – Other than building a company and

evaluating it (my current hobby), could this ever be expedited for demonstration funds given the stakes at end.