Improving Administrative Data for Public Reporting

Slide 1

Improving Administrative Data for Public Reporting

Anne Elixhauser
Joe Parker
Michael Pine
Roxanne Andrews
September 9, 2008

Slide 2

Outline

• Background and rationale.
• Summary of two prior studies:
  • Potential safety events present on admission?
  • Adding clinical information to administrative data.
• Problems in present on admission (POA) coding—California example.
• Screens for detecting these problems.
• Supporting the enhancement of administrative claims data through state pilots.

Slide 3

Administrative, or Billing Data

The slide shows a sample copy of a UB-92 (UB-04) Billing Form.

• Patient demographics (age, sex).
• Diagnoses & procedures.
• (ICD-9-CM, diagnosis-related group [DRG]).
• Expected payer.
• Length of stay.
• Patient disposition.
• Admission source & type.
• Admission month.
• Charges.

Slide 4

12 States Use AHRQ Quality Indicators (QIs) for Hospital Reporting to the Public

The slide shows a map of the United States with the following states highlighted:

• Vermont
• New York
• Massachusetts
• Ohio
• Kentucky
• Florida
• Wisconsin—part of state
• Iowa
• Texas
• Colorado
• Utah
• Oregon

Slide 5

Limitations of Administrative Data

• Lack clinically important information.
  • Limited to ICD-9-CM diagnosis codes.
• Do not distinguish between diagnoses present on admission (POA) and those that originate during the hospital stay.
• Questions regarding use of only administrative data for hospital-specific reporting.
  • Inadequate risk adjustment—additional data needed to predict individual patient's risk of mortality.
  • Concern about penalizing providers with the sickest patients.

Slide 6

Tension Between Value of Data and Cost of Obtaining the Data

• New York and California provide POA coding for diagnoses—now required for Medicare patients and many states will collect for all
• Pennsylvania hospitals provided chart-abstracted clinical detail
  • Hospital concern about costs of medical record abstraction
• Electronic medical records not yet poised to provide data efficiently
  • Exception: Lab data

Slide 7

How Often are Potential Patient Safety Events Present on Admission?

• Study aimed at using POA information to determine what effect it will have on AHRQ Patient Safety Indicators.
• Examined face validity of POA coding in two states—California (CA) and New York (NY).
• Study reported in...
  • Houchens R, Elixhauser A, Romano P. How often are potential "patient safety events present on admission?" Joint Commission Journal on Quality and Patient Safety, March 2008.

Slide 8

Percent of patient safety events remaining after POA diagnoses were removed*

The bar graph measures the percentage for "Anesthesia cx," "Respir. failure," "Accidental puncture," "Hemorrhage," "Metab. Derang.," "Sepsis," "Pneumothorax," "Infection," "Foreign body," "Transfus. rxn," "PE/DVT," "Hip fx," and "Decub. ulcer." The graph shows "Anesthesia cx" with the highest percentage at 100%. The percentages for the others in the list slowly decrease, with "Decub. ulcer" showing the lowest percentage at approximately 12%.

Slide 9

Impact of Adding Clinical Data to Administrative Data

• Assess impact of incrementally adding:
  • POA codes for diagnoses.
  • Lab values on admission.
  • Increased number of diagnosis fields.
  • Improved documentation (ICD-9-CM codes).
  • Vital signs.
  • More difficult to obtain clinical data.

Slide 10

Study Reported in...

• Pine M, Jordan HS, Elixhauser A, et al. Enhancement of claims data to improve risk adjustment of hospital mortality. JAMA 2007; 267(1):71-76.
• Jordan HS, Pine M, Elixhauser A, et al. Cost-effective enhancement of claims data to improve comparisons of patient safety. Journal of Patient Safety 2007; 3(2):82-90.
• Fry DR, Pine M, Jordan HS, et al. Combining administrative and clinical data to stratify surgical risk. Annals of Surgery 2007 246(5):875-885.
• Pine M, Jordan HS, Elixhauser A, et al. Modifying claims data to improve risk-adjustment of inpatient mortality rates. Medical Decision Making (forthcoming).

Slide 11

Indicators Studied

• Mortality:
  • Indicators.
  • Abdominal aortic aneurysm (AAA) repair.
  • Coronary artery bypass graft (CABG) surgery.
  • Craniotomy.
  • Acute Myocardial Infarction (AMI).
  • Ccongestive heart failure (CHF).
  • Cerebrovascular accident.
  • Gastrointestinal (GI) hemorrhage.
  • Pneumonia.
• Post-operative patient:
  • Safety events.
  • Pulmonary embolism/deep vein thrombosis.
  • Physiologic/metabolic abnormalities.
  • Respiratory failure.
  • Sepsis.

Slide 12

Data Used in Incrementally More Complex Models

The table shows the "Types of Data Elements" for each "Model."

• ADM-8: Age, sex, principal diagnosis, up to 8 secondary diagnoses, selected surgical procedures.
• POA-8: ADM-8 + diagnoses present on admission.
• POA-24: Increased secondary dx to 24.
• POA-ICD: POA-24 + secondary dx present on admission in clinical database, but not reported using ICD codes.
• LAB: POA-24 + laboratory data.
• LAB-ICD: POA-ICD + laboratory data.
• FULL: LAB-ICD + vital signs + lab data not in LAB (e.g., blood culture results) + key clinical findings abstracted from medical records + composite clinical scores (i.e., ASA Classification).

Slide 13

C-Statistics for Mortality Models

The slide shows a graph with a vertical axis, average C-statistics, going from 0.76 to 0.90 and a horizontal axis citing various models: ADM-8; POA-8; POA-24; POA-ICD; LAB; LAB-ICD; and FULL.

Slide 14

Numerical Lab Data

• Results of 22 lab tests entered at least one model.
• Results of 14 of these tests entered four or more models:
  • pH (11).
  • PTT (10).
  • Na (9).
  • WBC (9).
  • BUN (8).
  • pO2 (8).
  • K (7).
  • SGOT (7).
  • Platelets (7).
  • Albumin (5).
  • pCO2 (4).
  • Glucose (4).
  • Creatinine (4).
  • CPK-MB (4).

Slide 15

Vital Signs and Other Clinical Data

• All vital signs entered four or more models.
  • Pulse (8).
  • Temp (6).
  • Blood pressure (6).
  • Respirations (5).
• Ejection fraction and culture results entered two models.
• Composite scores entered four or more models.
  • ASA classification (6).
  • Glasgow Coma Score (4).

Slide 16

Abstracted Key Clinical Findings

• 35 clinical findings entered at least one model.
• Only three findings entered more than two models.
  • Coma (6).
  • Severe malnutrition (4).
  • Immunosuppressed (4).
• 14 of these clinical findings have corresponding ICD-9-CM codes (e.g., coma, malnutrition).

Slide 17

Summary of Analyses

• For some measures, POA coding has a significant impact on whether conditions are considered patient safety events.
• Administrative data can be improved at relatively low cost by:
  • Adding POA modifiers.
  • Adding numerical lab data on admission.
  • Improved ICD coding.

Slide 18

Other Enhancements

• Link to vital statistics.
• Link across settings.
• Readmissions.
• Episodes of care.

• Today's focus: POA and lab data.