Creating a Hybrid Database by Adding a POA Modifier and Numerical Laboratory Results to Administrative Claims Data

Michael Pine, M.D., M.B.A.
Michael Pine and Associates, Inc.
mpine@consultmpa.com

Overview

• Alternative databases for performance monitoring
• Comparative performance of alternative databases
• Present-on-admission coding
• Numerical laboratory data
• Vital signs and other clinical data
• The bottom line

Data for Monitoring Clinical Performance

• Claims Data – from HCFA Mortality Reports and HealthGrades.com to HCUP Quality and Patient Safety Indicators
• Clinical Data – from APACHE, Pennsylvania Health Care Cost Containment Council and Cleveland Health Quality Choice to Specialty Society Registries (e.g., STS, ACC)

Claims Data Versus Clinical Data

• Data serves as the basis for:
• public reporting
• reimbursement
• quality improvement initiatives
• Must balance the need for data to support
• accurate measurement of risk-adjusted clinical performance
• ease and cost of data collection

diagram

Relative Ease of Data Collection

Data fall on a spectrum of type of data collection, with automated collection at one end of the spectrum, and manual collection at the other end of the spectrum.

Data fall on a spectrum of type of data collection, with automated collection at one end of the spectrum, and manual collection at the other end of the spectrum.

Manual collection: Numerical laboratory, vital signs, and other clinical data, which together are classified as clinical data.

end of diagram

Efficient Use of Clinical Data

diagram

Hemoglobin: Low analytic power, low cost to collect.
FEV1: Low analytic power, high cost to collect.
Albumin: High analytic power, low cost to collect.
Mental status: High analytic power, high cost to collect.

end of diagram

Enhancing Claims Data

• Present-on-Admission Coding – from the Mayo Clinic, New York State’s SPARCS database, and California’s OSHPD database to the UB-04 and CMS’s new coding requirements
• Numerical Laboratory Data – from Michael Pine and Associates to the Agency for Healthcare Research and Quality (AHRQ)
• New Hybrid Databases – AHRQ's Pilot Projects

diagram

Creating a Hybrid Database
Claims Data are: Standard claims and Present-on-Admission.
Clinical Data are: Numerical laboratory, vital signs, and other clinical data.
Hybrid Data are: Standard claims, Present-on-Admission, and numerical laboratory.

end of diagram

Potential Benefits of Enhancing Claims Data

• Better distinguish between comorbidities and complications
• Add objective findings to more subjective diagnostic designations
• Provide finer definition of progression of disease and underlying pathophysiology than do diagnostic codes alone

Comparative Performance of Alternative Databases

Inpatient Quality Indicators (Mortality)

• Medical Conditions – Acute Myocardial Infarction; Cerebrovascular Accident; Congestive Heart Failure; Gastrointestinal Hemorrhage; Pneumonia
• Surgical Procedures – Abdominal Aortic Aneurysm Repair; Coronary Artery Bypass Graft Surgery; Craniotomy

Patient Safety Indicators (Complications)

• Elective Surgical Procedures
• Complications – Physiologic / Metabolic Abnormalities; Pulmonary Embolus / Deep Vein Thrombosis; Sepsis; Respiratory Failure

Data Used in CLAIMS Models

• Age and sex
• Principal diagnosis
• Secondary diagnoses only infrequently acquired during hospitalization
• Selected surgical procedures

Data Used in HYBRID Models

• All data used in CLAIMS models
• Additional secondary diagnoses when clinical data establish that they were present on admission
• Numerical laboratory data (e.g., creatinine, white blood cell count) generally available in electronic form

Data Used in CLINICAL Models

• All data used in HYBRID models
• Vital signs and laboratory data not in HYBRID models (e.g., blood culture results)
• Key clinical findings abstracted from medical records (e.g., immunocompromised)
• Composite clinical scores (e.g., ASA class)

diagram

Bias Due to Suboptimal Risk-Adjustment

Measured performance:

• Average performance falls within plus or minus 2 standard deviations of the mean. Good and poor performance fall outside of 2 standard deviations of the mean.

Spectrum of bias:

• OK bias falls within within plus or minus 0.5 standard deviations of the mean. Problematic bias falls outside plus or minus 0.5 standard deviations of the mean.

end of diagram

diagram

Bias Due to Suboptimal Data (IQIs)

Line chart of type of data (raw, claims, and hybrid) graphed by the percent of data exceeding the upper threshold for bias in standard deviations.

Data values are not shown in the chart, so percentages are listed in ranges.

Raw Data60% to 70%30% to 40%20% to 30%10% to 20%Claims Data40% to 50%20% to 30%0% to 10%0% to 10%Hybrid Data10% to 20%0% to 10%0% to 10%0% to 10%

Percent Exceeding Upper Threshold	Upper Threshold for Bias in Standard Deviations
	0.5	1.0	1.5	2.0

end of diagram

Beginning of diagram

Bias Due to Suboptimal Data (PSIs)

Line chart of type of data (raw, claims, and hybrid) graphed by the percent of data exceeding the upper threshold for bias in standard deviations.

Data values are not shown in the chart, so percentages are listed in ranges.

Raw Data50% to 60%20% to 30%10% to 20%0% to 10%Claims Data30% to 40%10% to 20%0% to 10%0% to 10%Hybrid Data10% to 20%0% to 10%0% to 10%0% to 10%

Percent Exceeding Upper Threshold	Upper Threshold for Bias in Standard Deviations
	0.5	1.0	1.5	2.0

end of diagram

POA Coding

New Information Derived from POA Coding

• In the past, difficult to determine whether coded secondary diagnoses described:
• Comorbid conditions present on admission
• Complications that occurred in hospital.
• Newly mandated POA distinguishes between:
• Comorbidities that increase the likelihood of adverse outcomes and higher costs
• Inpatient complications possibly due to suboptimal care.

General Guidelines for POA Coding

• With rare exceptions, a POA modifier must be assigned to each principal and secondary diagnosis code on a hospital claim.
• A diagnosis should be coded as present on admission if it is present at the time the order for inpatient admission occurs.
• All POA coding must be supported by medical record documentation by a qualified healthcare practitioner.

Valid POA Codes

• Blank, 1, or E = diagnosis exempt from POA reporting
• Y = present at time of order to admit
• N = not present at time of order to admit
• W = practitioner unable to determine if Y or N
• U = insufficient information to determine if Y or N after good faith attempt to resolve uncertainty with qualified practitioner

Rules for POA Coding (1)

• Chronic conditions are coded as POA=Y regardless of when they are diagnosed.
• A diagnosis of an acute condition is coded as POA=Y when:
• documented as present, suspected, or impending at the time of or shortly prior to admission even if the definitive diagnosis is made during hospitalization
• signs or symptoms of the diagnosis are documented as present on admission.

Rules for POA Coding (2)

• An acute exacerbations of a chronic condition is coded as POA=Y only when the acute exacerbation is present on admission.
• A diagnosis is coded as indeterminate (W) only when a qualified practitioner documents that s/he cannot determine if diagnosis was present on admission.
• A diagnosis is coded as unknown (U) only when a coder cannot obtain information needed to assign another POA modifier.

Rules for POA Coding (3)

• For obstetrical codes, POA assignment:
• based on relation of pregnancy-related diagnoses to admission
• not affected by whether or not the patient delivers.
• If obstetrical code includes more than one diagnosis, POA=Y only if all diagnoses are present on admission.

Rules for POA Coding (4)

• For newborns, admission occurs at the time of birth. Therefore, POA=Y for all congenital conditions and anomalies, all in uteroconditions, and all complications that occur during delivery.
• For accidents (i.e., E codes), POA codes are based on the relation of the time of injury to the time of admission. Therefore, POA=Y only when injury occurs prior to admission.

Rules for POA Coding (4)

• Accurate coding requires expertise and teamwork
• Inaccurate coding may affect performance assessments and reimbursement.
• Chart reviews to detect coding errors are expensive.
• Well-designed screens can detect problems efficiently.

POA Quality Screens

• Developed using New York State hospital discharge data from 2003 through 2005.
• Screens high-risk conditions, elective surgical procedures, and inpatient childbirth.
• Employs 12 screens for inconsistent and implausible coding.
• Provides composite scores and performance profiles.

Distribution of Hospital Scores

Score	Hospitals (#)	Hospitals (%)
> 90%	65	39.4%
> 80% to 90%	41	24.8%
> 70% to 80%	26	15.8%
> 60% to 70%	19	11.5%
60% or lower	14	8.5%
Total Scored	165	100%
>% Unknown	22	n/a

Screening and Improvement of POA Coding

diagram

Circular diagram of process of screening and improvement of POA coding. Steps 1 through 6 repeat.

1. POA screening
2. Performance evaluation
3. Process analysis
4. Identification of opportunities for improvement
5. Plan for improvement
6. 1Plan for improvement

Numerical Laboratory Data

Types of Data in HYBRID IQI Models

15.6 data elements are: Standard claims and Present-on-Admission.

11.1 data elements are: Numerical laboratory.

Hybrid Data are: Standard claims, Present-on-Admission, and numerical laboratory.

end of diagram

Types of Data in HYBRID PSI Models

21.8 data elements are: Standard claims and Present-on-Admission.

6.5 data elements are: Numerical laboratory.

Hybrid Data are: Standard claims, Present-on-Admission, and numerical laboratory.

end of diagram

Numerical Laboratory Data

• 22 Laboratory Tests Enter At Least 1 Model
• 14 of These Tests Enter 4 or More Models
• pH (11)
• Prothrombin Time (10)
• Sodium (9)
• White Blood Count (9)
• Blood Urea Nitrogen (8)
• pO2 (8)
• Potassium (7)
• SGOT (7)
• Platelet Count (7)
• Albumin (5)
• pCO2 (4)
• Glucose (4)
• Creatinine (4)
• Creatinine (4)

Recommended Chemistry Data

• Aspartate Aminotransferase
• Albumin
• Alkaline Phosphatase
• Amylase
• Bicarbonate
• Bilirubin (Total)
• B Natriuretic Peptide
• Calcium
• C-Reactive Protein
• Creatine Kinase
• Creatine Kinase MB
• Creatinine
• Glucose
• Lactic Acid
• Potassium
• Pro-B Natriuretic Protein
• Sodium
• Troponin I
• Troponin T
• Urea Nitrogen

Other Recommended Lab Data

Blood Gas

• Arterial O² Saturation
• Arterial pCO²
• Arterial pH
• Arterial pO²
• Base Excess
• Bicarbonate
• FIO² (if electronic)

Hematology

• Hemoglobin
• International Normalized Ratio
• Neutrophil Bands
• Partial Thromboplastin Time
• Platelet Count
• Prothrombin Time
• White Blood Count

Vital Signs and Other Clinical Data

diagram

Types of Data in CLINICAL IQI Models

15.6 data elements are: Standard claims and Present-on-Admission.

11.1 data elements are: Numerical laboratory.

9.0 data elements are: Vital signs and other clinical data.

Hybrid Data are: Standard claims, Present-on-Admission, and numerical laboratory.

end diagram

Types of Data in CLINICAL PSI Models

diagram

15.6 data elements are: Standard claims and Present-on-Admission.

11.1 data elements are: Numerical laboratory.

9.0 data elements are: Vital signs and other clinical data.

Hybrid Data are: Standard claims, Present-on-Admission, and numerical laboratory.

end diagram

Vital Signs, Other Lab Data, Scores

• All Vital Signs Enter 4 or More Models
• Pulse (8)
• Temperature (6)
• Blood Pressure (6)
• Respirations (5)
• Ejection Fraction and Culture Results Each Enter 2 Models
• Both Composite Scores Enter 4 or More Models
• ASA Classification (6)
• Glasgow Coma Score (4)

Abstracted Key Clinical Findings

• 35 Clinical Findings Enter At Least 1 Model
• Only 3 Enter More Than 2 Models
• Coma (6)
• Severe Malnutrition (4)
• Immunosuppressed (4)
• 14 Have Corresponding ICD-9-CM Codes
• Coma
• Severe Malnutrition

The Bottom Line

Risk-Adjusted Mortality in CABG Surgery

diagram

Chart of risk-adjusted mortality in percent for CABG surgery

Chart shows confidence interval, predicted percent risk-adjusted mortality, observed percent risk-adjusted mortality, and p-value of predicted to observed percent risk-adjusted mortality for CABG surgery by hospital. Chart does not include data values, so range of values are estimated.

Hospital Number	Confidence Limit (Lower)	Confidence Limit (Upper)	Predicted percent risk-adjusted mortality	Observed percent risk-adjusted mortality	P-value of predicted to observed percent mortality
Hospital 6	1.5% to 2.0%	4.5% to 5.0%	3.0% to 3.5%	1.5% to 2.0%	p = 0.001 to 0.01
Hospital 7	2.0% to 2.5%	4.0% to 4.5%	3.0% to 3.5%	2.0% to 2.5%	p = 0.01 to 0.05
Hospital 18	1.0% to 1.5%	7.5% to 8.0%	4.0% to 4.5%	2.0% to 2.5%	p = 0.01 to 0.05
Hospital 13	1.5% to 2.0%	5.0% to 5.5%	3.5% to 4.0%	2.5% to 3.0%	p is greater than 0.05
Hospital 1	0.5% to 1.0%	5.0% to 5.5%	3.0% to 3.5%	1.5% to 2.0%	p is greater than 0.05
Hospital 10	0.0%	4.0% to 4.5%	2.0% to 2.5%	1.0% to 1.5%	p is greater than 0.05
Hospital 14	0.5% to 1.0%	4.5% to 5.0%	2.5% to 3.0%	2.0% to 2.5%	p is greater than 0.05
Hospital 5	1.0% to 1.5%	6.5% to 7.0%	3.5% to 4.0%	3.0% to 3.5%	p is greater than 0.05
Hospital 3	0.0% to 0.5%	4.5% to 5.0%	2.5% to 3.0%	2.0% to 2.5%	p is greater than 0.05
Hospital 12	0.0%	3.5% to 4.0%	1.5% to 2.0%	1.5% to 2.0%	p is greater than 0.05
Hospital 11	0.0% to 0.5%	4.5% to 5.0%	2.5% to 3.0%	2.5% to 3.0%	p is greater than 0.05
Hospital 17	0.5% to 1.0%	4.0% to 4.5%	2.5% to 3.0%	2.5% to 3.0%	p is greater than 0.05
Hospital 16	0.5% to 1.0%	5.0% to 5.5%	2.5% to 3.0%	3.0% to 3.5%	p is greater than 0.05
Hospital 15	0.0%	3.0% to 3.5%	1.0% to 1.5%	2.0% to 2.5%	p is greater than 0.05
Hospital 9	1.5% to 2.0%	4.0% to 4.5%	2.5% to 3.0%	3.0% to 3.5%	p is greater than 0.05
Hospital 8	1.0% to 1.5%	3.5% to 4.0%	2.0% to 2.5%	2.5% to 3.0%	p is greater than 0.05
Hospital 2	1.0% to 1.5%	4.5% to 5.0%	3.0% to 3.5%	3.5% to 4.0%	p is greater than 0.05
Hospital 4	0.5% to 1.0%	5.5% to 6.0%	3.0% to 3.5%	4.5% to 5.0%	p = 0.01 to 0.05

end of diagram

Bias in Measurement of PSIs

Line chart of observed vs predicted rates of true complications, bias due to failure to risk-adjust true complication rates, and bias due to misclassifying comorbidities as complications, graphed by the percent of data exceeding the upper threshold for bias in standard deviations.

Data values are not shown in the chart, so percentages are listed in ranges.

Percent Exceeding Upper Threshold	Upper Threshold for Bias in Standard Deviations
0.5	1.0	1.5	2.0
Observed vs predicted rates of true complications	80% to 90%	60% to 70%	40% to 50%	20% to 30%
Bias due to failure to risk-adjust true complication rates	40% to 50%	10% to 20%	0% to 10%	0% to 10%
Bias due to misclassifying comorbidities as complications	40% to 50%	20% to 30%	10% to 20%	0% to 10%

end of diagram

Carpe Diem!