Methodological Issues in
Comparing Hospital
Performance: Measures, Risk
Adjustment,
and Public Reporting
Harlan M. Krumholz, MD
Yale University School of Medicine July 31, 2015
Comparison of health care spending in Taiwan and
the World’s major countries
NHI public satisfaction
ratings 1995-2014
It would be great if Prof. Krumholz
could share the US experiences in
developing hospital report cards
regarding:
1. How to select appropriate
indicators/performance metrics to
better represent the performance of
hospitals.
• Published “Evaluating the Quality of Medical
Care” in 1966
• Separates quality measures into 3 categories: • Structure
• Process • Outcome
Types of Quality Measures Structure Assumes “good medical care will follow” Process Outcome
Types of Quality Measures Structure Assumes “good medical care will follow” Process Assesses “whether ‘good’ medical care has been applied” Outcome
Structure Measures – Pros & Cons
Pros
• Measure important foundational concepts • Easy to measure and compare
Structure Measures – Pros & Cons
Pros
• Measure important foundational concepts • Easy to measure and compare
Cons
• Relationship to quality is assumed
• Improving performance may not improve outcomes
Structure Measures
Define the presence or absence of specific care resources Include:
• Facilities and equipment • Number of beds
• EHR capabilities
• Professional and organizational resources • Licensure and certification
• Joint Commission accreditation • Policies and procedures
Process Measures
Assess whether or not specific care is performed Include:
• Clinical and preventive care • CMS “Core” measures
Process Measures
Examples
Applicable to hospitals:
• Aspirin administration to heart attack patients • Timing of antibiotic initiation for pneumonia
Process Measures – Pros & Cons
Pros
• Evolve from evidence-based practice guidelines • Combine consensus-based information with clear
Process Measures – Pros & Cons
Pros
• Evolve from evidence-based practice guidelines • Combine consensus-based information with clear
instructions for improving care
Cons
• Can be burdensome to collect
• Exclude some patients from measurement • May not correlate well with outcomes
Outcome Measures
Assess what actually happens to the patient • Survival
• Unintended treatment effects • Symptom relief
Include:
• Acute clinical events • Health status
Outcome Measures
Examples
Applicable to hospitals:
• 30-day all-cause mortality and readmission following AMI, heart failure and pneumonia
Outcome Measures – Pros & Cons
Pros
Outcome Measures – Pros & Cons
Pros
• Measure what matters most to patients
Cons
• Must be influenced by care
• Requires adequate risk adjustment to account for differences in case mix
• Influenced by many factors • More difficult to measure
What makes a good outcome measure? Well thought out measure concept
Considering potential applications in design
A well thought out
measure concept
Addresses measurement and performance gaps Measurement gap: Not currently measured
Performance gap: variation in performance not explained by variation in patient case mix
Considering potential
applications
in design Clinical care Quality improvement Accountability
Measure
specifications
Population to be measured (cohort)
Result to be measured (outcome)
Outcome attribution (who is responsible)
Measure
specifications
Population to be measured (cohort)
Result to be measured (outcome of interest)
Outcome attribution (who is responsible)
Measure
specifications
Population to be measured (cohort) Result to be measured (outcome)
Outcome attribution (who is responsible)
Measure
specifications
Population to be measured (cohort) Result to be measured (outcome)
Measure Cohort
Reliably identifiable Clinically coherent
Measure Outcome
Reliably identifiable
Outcome attribution
Ownership of care
2. The mechanism to select risk factors
for each outcome indicators,
especially to get/build consensus from
medical professionals.
Outcome
Hospital
Care
Patient Status on PresentationOther
Effects
Patient
Status on
Presentation
Expected
Outcome
Expected
Outcome
Observed
Outcome
Purpose of
risk adjustment
To help define EXPECTED OUTCOME Level the playing field
Account for patient-level factors that influence outcome but do not reflect care quality
Avoid risk factors that
Represent complications of care
Represent healthcare system attributes Will potentially mask disparities
3. The statistic models used for risk
adjustment, and their strength and
weakness.
Approach to risk adjustment
Account for patient case mix
Account for clustering of patients at measurement unit (hospital, group practice, etc.)
Avoid risk factors that
Represent complications of care
Represent healthcare system attributes (e.g., discharge disposition)
Approach to risk adjustment
Account for patient case mix
Account for clustering of patients at measurement unit (hospital, group practice, etc.)
Avoid risk factors that
Represent complications of care
Represent healthcare system attributes (e.g., discharge disposition)
4. The issues regarding quality of data,
especially when claim data is the
major source for measuring
performance.
5. The policy/approach that help
medical professionals and consumers
to understand report card, especially
the result of adjusted outcomes.
Aspects of outcome measures specific to accountability: Higher threshold of scientific reliability and validity Must be fair to measured entities
Should assign performance categories with high degree of confidence
Aspects of outcome measures specific to accountability:
Higher threshold of scientific reliability and validity
Must be fair to measured entities
Should assign performance categories conservatively
Aspects of outcome measures specific to accountability:
Higher threshold of scientific reliability and validity Must be fair to measured entities
Should assign performance categories conservatively
Aspects of outcome measures specific to accountability:
Higher threshold of scientific reliability and validity Must be fair to measured entities
Should assign performance categories with high degree of confidence
Aspects of outcome measures specific to accountability:
Higher threshold of scientific reliability and validity Must be fair to measured entities
Should assign performance categories with high degree of confidence
CMS’s hospital AMI 30-day all-cause mortality measure Reported as risk-standardized mortality rate (RSMR) Measures all-cause mortality after AMI admissions
Captures deaths within 30 days of admission Publicly reported on HospitalCompare.gov
Risk-standardized mortality rate (RSMR) = Hospital’s “predicted” deaths
Hospital's “expected” deaths
“Predicted” = number of deaths within 30 days predicted on basis of hospital’s performance with its observed case mix
“Expected” = number of deaths expected on basis of the nation’s performance with that hospital’s case mix
national unadjusted mortality rate
Performance categories for public reporting
Hospitals categorized as “Better”, “Worse”, or “No Different” than the national rate
Use 95% interval estimate (like 95% confidence interval) to define categories