David Bertoch, Vice President

1
Using the Pediatric Health Information System and
Administrative Data for Research

• David Bertoch, Vice President
• Matt Hall, Senior Statistician

2
Presentation Objectives

• To provide an overview of available
  administrative data sources, including CHCAs
  Pediatric Health Information System
• To describe how to use these data sources for
  research

3
Presentation Content

• Administrative Data Overview
• Pediatric Health Information System (PHIS)
• Other Administrative Data Sources
• Reliability of ICD-9 Codes
• Types of Research
• Analytic Considerations
• Resources/Contacts

4
What is administrative data and where does it
come from?
Provider performs assessment and treatment of
patient
Provider documents some component of their
actions and thoughts
Patient arrives at hospital and demographic info
is collected and entered into IDX
Trained coders review all of medical record
including documentation
As treatments are provided and ordered, charge
codes are captured in IDX
Data pulled to send bills, quality and outcomes
reporting, operations/finance, PHIS/NACHRI
Diagnoses and procedures are entered into
patients administrative record (IDX)
Using national required coding guidelines,
ICD-9-CM diagnoses and procedure codes are
assigned
Flowchart designed by the Childrens Hospital of
Wisconsin
5
Common Administrative Data Sources

• State Databases
• HCUP
• KID 2003
• NIS 2006
• CHCA Pediatric Health Information System (PHIS)
• Thomsons National Pediatric Discharge Database
  (NPDD)

6
Pediatric Health Information System (PHIS)
7
PHIS Data Overview
PHIS By The Numbers

• Participating Hospitals 40
• Inpatient Cases 2.2 million
• Inpatient Days 13.1 million
• ED encounters 6.7 million
• Total Charges 90.7 billion
• Total ICD-9 Codes 33.6 million
• Pharmacy Transactions 116.8 million
• Physicians 297,250

All data submitted electronically (no manual
entry) on a quarterly basis
Since 2002, does not include available archived
data back to 1992
8
CHCA North Americas Leading Childrens Hospitals
Seattle - Childrens Hospital and Regional
Medical Center Orange Childrens Hospital of
Orange County San Diego Childrens Hospital
and Health Center Columbus Childrens
Hospital Los Angeles Childrens Hospital Los
Angeles Oakland Childrens Hospital Oakland
Memphis Le Bonheur Childrens Medical Center
Palo Alto Lucile Packard Childrens Hospitals
Dayton The Childrens Medical Center Fresno
/ Madera Childrens Hospital Central California
Milwaukee Childrens Hospital of
Wisconsin Buffalo Childrens Hospital of
Buffalo Denver The Childrens Hospital
Akron Childrens Hospital Medical Center of
Akron Norfolk Childrens Hospital of The
Kings Daughters Health System Washington D.C.
Childrens National Medical Center Pittsburgh
Childrens Hospital of Pittsburgh Little Rock
Arkansas Childrens Hospital Philadelphia
The Childrens Hospital of Philadelphia St.
Louis St. Louis Childrens Hospital St
Petersburg All Childrens Hospital Listed In
Order of Membership Submit Data into PHIS
Kansas City The Childrens Mercy Atlanta
Childrens Healthcare of Atlanta Chicago The
Childrens Memorial Hospital Birmingham The
Childrens Hospital of Alabama Dallas
Childrens Medical Center of Dallas New Orleans
Childrens Hospital Cincinnati Childrens
Hospital Medical Center Miami Miami
Childrens Hospital Corpus Christi Driscoll
Childrens Hospital Houston Texas Childrens
Hospital Ft. Worth Cook Childrens Medical
Center Boston Childrens Hospital Boston
Omaha Childrens Healthcare Services
Memphis St. Jude Childrens Research Hospital
Minneapolis Childrens Hospitals and Clinics
Phoenix Phoenix Childrens Hospital Detroit
Childrens Hospital of Michigan Nashville -
Vanderbilt Childrens Hospital Hartford
Connecticut Childrens Medical Center Toronto
The Hospital for Sick Children New York
Childrens Hospital of New York
Presbyterian Indianapolis Riley Childrens
Hospital/Clarian Health Partners

9
PHIS Hospitals 41 CHCA Hospitals Submitting Data
Minnesota Omaha Kansas City
Dayton Columbus Cincinnati Akron Buffalo
Milwaukee Chicago St. Louis Detroit Indianapolis
Boston Hartford New York Philadelphia DC Norfolk P
ittsburgh
Seattle Oakland Palo Alto Madera Los
Angeles Orange San Diego
Memphis Nashville Atlanta St. Petersburg Miami
Phoenix Denver Dallas Fort Worth Corpus Christi
Little Rock New Orleans Birmingham
Houston
10
PHIS Features

• Comparable hospitals
• Largest childrens hospitals in the US
• Unblinded peer selection
• Select hospitals with whom you want to compare
• Ease of networking
• Physicians, clinicians, quality leaders, analysts
  
• Direct access to data
• Control over report specifications

11
Every hospital uses PHIS in different ways
physician
Credentialing
Meet JCAHO requirement
Make better decisions
Improve coding
Manage utilization
Access to database with 6 million pt encounters
Increase revenue
Reduce manual MR review
Develop new service lines
12
Increased Use for Research

• 23 articles published with at least 3 pending and
  many more in development
• 25 posters/presentations at Pediatric Academic
  Societies meeting since 2003
• 9 in 2007
• Wide variety of topics
• See handouts

13
Typical Data Submission Process

• Sample for October data load

October 7
August 15
August 29
September 26
Monthly or Quarterly Submission
October 18
October 18
Annual (2nd Qtr)
October 21
As Needed
14
Practical PHIS Data Quality Resources

• Data Quality and Completeness Report Card
• Data Quality Alerts
• Web Cast Validating Your Data
• Significant Issues List

15
PHIS for Research Resources

• Subset of PHIS main web site
• Includes
• Standard PHIS methodology text
• Existing articles by topic
• Data Quality resources
• Data content resources
• Register at www.chca.com and select PHIS as
  requested site

16
Level 1 and Level 2 Breakout
LEVEL 1 Patient Abstract and ICD-9 Coding
Patient Abstract
Procedures (ICD-9)
Diagnoses (ICD-9)
LEVEL 2 Billed Transaction/Utilization Data
(all items/services billed to the pt)
Imaging/ Radiology
Lab
Clinical
Pharmacy
Other Room/Nursing Surgical Svcs Other misc
Supplies
Other
17
Level 1 Patient Abstract

• Patient Identification
• Demographics
• Episode of Care
• Physician Profiles
• Dx/Px Profiles
• Clinical Classification (Groupers)
• Payer Source
• Charge Summaries

18
Patient Abstract

• Episode of Care
• LOS
• Admit Date/Month/Year
• Discharge Date/Month/Year
• Infection Flag
• Surgical and Medical Complication Flags
• Disposition
• Pre-Op LOS
• Post-Op LOS

• Demographics
• Gender
• Birthweight (gms)
• DOB
• Pediatric Age Group
• AAP Age Code
• Age (based on age at admission)
• Age in Years
• Age in Months (if less than 2 yrs)
• Age in Days (if less than 30 days)
• Race/Ethnicity

19
Patient Abstract

• Physician Profiles
• Attending Physician
• Attending Physician Sub-specialty
• Principal Px Physician
• Principal Px Physician Sub-specialty

• Dx/Px Profiles
• Principal Dx
• Principal Px

• Clinical Classification (Groupers)
• Major Diagnostic Category (MDC)
• CMS (HCFA) DRG
• APRDRG
• Version 15
• Version 20
• Version 24

20
Patient Abstract

• Payer Source (Principal Payer)
• Medicare
• Medicaid
• Title V
• Other govt
• Workers comp
• Blue Cross
• Other Ins Co
• Self Pay
• Other

• Charge Summaries
• Pharmacy Charges
• Supply Charges
• Lab Charges
• Imaging Charges
• Clinical Charges
• Other Charges
• Unmapped Charges

21
Pt Abstract Data (Level 1)
Patient Abstract

• Value
• Compare LOS
• Readmission Rates
• Stratify patients based upon your criteria
• Physician Profiling
• Severity Adjust

We will follow one patient visit through
different sections of PHIS Discharge ID
142006763
22
Diagnosis Codes (ICD-9) (Level 1)
Diagnoses (ICD-9)

• Value
• Go beyond the principal dx
• Specific inclusion/ exclusion of patients

23
Procedure Codes (ICD-9) (Level 1)
Procedures (ICD-9)

• Value
• Go beyond the principal px
• Pre vs Post Op LOS
• Analysis by surgeon

24
Charge Master Comparability
Hospital B
Hospital A
6561447 Tablet 125 mg Vancomycin
CTC Code 124133.1011552

• 12 ? Anti-infectives (Drug Class 12)
• 124 ? Misc antibiotics (Therapeutic Cat 124)
• 124133 ? Vancomycin (Generic Drug124133)
• 12413310 ? oral (Route of Administration10)
• 1241331011 ? tablet (Dosage Form11)
• 124133101155 ? 55 (Strength125)
• 1241331011552 ? mg (Unit of Measure2)

25
Pharmacy Data (Level 2)
Pharmacy

• Value
• Compare drug utilization by drug, class, and/ or
  category
• Compare when drugs were given (by day)
• Compare route of administration (IV, PO, etc)

26
Lab Data (Level 2)
Lab

• Value
• Compare lab tests/pt
• Revenue Enhancement Opportunity

27
Room/Nursing (Level 2)
Room/ Nursing

• Value
• Compare LOS by type of room (med/surg vs ICU)
• Measure Return to ICU/Direct Admit to ICU
• Analyze resource utilization by room type (eg.
  drugs while in NICU

28
Issues/Measures by CTC Area
29
Issues/Measures by CTC Area
30
PHIS Direct Access User Agreement

• Key Components of Agreement
• PHIS data, other than the PHIS Member Hospitals
  own data, cannot be shared verbally, in written
  form or electronically with any individual or
  group not acting on the sole behalf of the PHIS
  Member Hospital without prior consent from the
  PHIS External Use of Data committee.
• The confidentiality statement applies, but is not
  limited to, the following situations
• the publishing of research results in an external
  publication
• using PHIS data in a promotional/advertising
  campaign
• giving PHIS data in any format to a managed care
  organization
• making an external presentation with PHIS data
  displayed either verbally or visually in a
  handout
•  

31
HCUP KID / Thomson NPPD
32
Healthcare Cost and Utilization Project
(HCUP)Kids Inpatient Database (KID)

• Current Data 2003 updated every three years
• Short-term, general, non-federal hospitals
• Stratified systematic sample
• 2.9 million discharges weighted to 7.4 million
• i.e. one row represents multiple discharges
• Data elements similar to Level I data in PHIS
• Purchase for 200 per year from
  http//www.hcup-us.ahrq.gov
• Free web tool http//hcupnet.ahrq.gov/

33
Thomsons National Pediatric Discharge Database
(NPDD)

• Similar to HCUP KID, but updated twice annually
• Uses datasets available to Thomson
• State hospital associations
• Public state data
• Individual hospitals contracting with Thomson
• Various hospitals systems
• 2.2 million discharges are weighted to 7.1
  million
• Not as widely published as HCUP KID
• Accessible through CHCA

34
Reliability of ICD-9 Codes
35
Reliability of ICD-9 Codes

• Epidemiology, Outcomes, and Costs of Invasive
  Aspergillosis in Immunocompromised Children in
  the United States, 2000
• Zaoutis, Heydon, Chu, Walsh, Steinbach
• Pediatrics. 2006 Apr 117(4) e711-e716
• In general, health services researchers believe
  that the use of ICD-9-CM codes to identify cases
  in administrative databases has high specificity
  (eg, few instances in which patients did not in
  fact receive a diagnosis of the condition) but
  may be lower in sensitivity (ie, the
  administrative diagnosis may fail to detect all
  true cases).

36
Reliability of ICD-9 Codes

• Differences in Admission Rates of Children With
  Bronchiolitis by Pediatric and General Emergency
  Departments
• Johnson, Adair, Brant, Holmwood, Mitchell
• Pediatrics. 2002 Oct 110(4)e49
• Spec Of 3,091 charts coded as having a discharge
  diagnosis of bronchiolitis (ICD-9 code 466.1),
  3,054 cases (99) met clinical definition
• Sen Additional 43/377 (11) should have been
  coded as bronchiolitis

37
Reliability of ICD-9 Codes

• Use of Active Surveillance to Validate
  International Classification of Diseases Code
  Estimates of Rotavirus Hospitalizations in
  Children
• Hsu, Staat, Roberts et al.
• Pediatrics. 2005 Jan 115(1)78-82
• Spec Discharge coded as rotavirus very specific
  marker for true rotavirus disease 98 of
  discharge records coded specifically as rotavirus
  had a laboratory-confirmed diagnosis.
• Sen Discharge records were coded as rotavirus in
  less than half of the confirmed rotavirus
  infections

38
Types of Research Using Administrative Data
39
1. Epidemiology / Population Estimates

• Population estimates difficult with PHIS
• Convenience sample, lacks denominator
• Possible with specific quaternary diagnoses or
  procedures
• HCUP KID Thomsons NPDD
• Weighted for national estimates
• Potential research topics
• What is the prevalence of a disease in the
  population
• How frequent is a px done in a population

40
1. Epidemiology / Population Estimates (Example
1)

• National hospitalization impact of pediatric
  all-terrain vehicle injuries.
• Killingsworth JB, Tilford JM, Parker JG, Graham
  JJ, Dick RM, Aitken ME
• Pediatrics. 2005 Mar115(3)e316-21
• HCUP KID 1997 and 2000
• 5,292 children hospitalized with ATV-related
  injuries
• Hospitalizations increased 79.1 between 1997 and
  2000
• Rates of ATV-related hospitalization were highest
  among adolescent white male
• Total hospital charges 74,367,677 for the
  2-year study period

41
1. Epidemiology / Population Estimates (Example
2)

• Off-label drug use in hospitalized children
• Shah SS, Hall M, Goodman DM, Feuer P, Sharma V,
  Fargason C Jr, Hyman D, Jenkins K, White ML, Levy
  FH, Levin JE, Bertoch D, Slonim AD
• Arch Pediatr Adolesc Med. 2007 Mar161(3)282-90
• At least 1 drug was used off-label in 297,592
  (78.7) of 355,409 discharges
• Off-label use accounted for 270m (40.5) of the
  total dollars spent on these medications
• Factors associated with off-label use undergoing
  a surgical procedure, age older than 28 days,
  greater severity of illness, and all-cause
  in-hospital mortality.

42
2. Cost Charge Estimation

• Most data sources do not capture costs, but
  charges
• Typically use ratio of cost-to-charges
• In KID, each hospital has one ratio
• In PHIS, each hospital has 31 ratios categorized
  into drug, radiology, etc.
• Potential research topics
• Public vs. private expenditures
• Incremental charges associated with comorbidities
• Compare costs of treating with drug x versus drug
  y
• Identify factors associated with increased
  charges

43
2. Cost Charge Estimation (Example 1)

• Direct medical cost of influenza-related
  hospitalizations in children.
• Keren R, Zaoutis TE, Saddlemire S, Luan XQ,
  Coffin SE.
• Pediatrics. 2006 Nov118(5)e1321-7
• 727 patients hospitalized for community-acquired
  laboratory-confirmed influenza
• The mean total cost of hospitalization 13,159
• 39,792 pts admitted to an ICU
• 7,030 pts cared for exclusively on the wards
• Cardiac, metabolic, and neurologic/neuromuscular
  diseases and age of 18-21 were independently
  associated with the highest hospitalization costs

44
2. Cost Charge Estimation (Example 2)

• Factors associated with increased resource
  utilization for congenital heart disease
• Connor JA, Gauvreau K, Jenkins KJ.
• Pediatrics. 2005 Sep116(3)689-95
• Identify patient, institutional, and regional
  factors that are associated with high resource
  utilization for congenital heart surgery
• Some states were more likely to have high
  resource use cases
• Independent predictors of a higher odds of high
  cost
• Risk Adjustment for Congenital Heart Surgery risk
  category
• Age
• Prematurity
• Presence of other major noncardiac structural
  anomalies
• Medicaid insurance
• Admission during a weekend

45
3. Longitudinal Data Analysis

• KID and NPDD do not have unique pt identifiers
• PHIS has MRNs that can be tracked across time
  within institution (some hospitals have data back
  to 1992)
• Useful for classifying certain pts underlying
  disease
• Potential research topics
• Utilization of chronic populations
• Readmissions, if the case can be made that most
  patients dont go somewhere else
• Time-to-event analysis
• Trends in admissions or seasonality

46
3. Longitudinal Data Analysis (Example)

• A multi-center study of factors influencing
  cerebrospinal fluid shunt survival in infants and
  children
• Shah SS, Hall M,Slonim A, Hornig GW, Berry JG,
  Sharma V.
• J Neurosurg (In Press)
• 7,399 had shunt placement and at least one-year
  of follow-up
• 20.2, 7.5, and 6.9 of patients required 1, 2,
  or 3 or more shunt revisions, respectively
• In multivariable analyses, children undergoing
  shunt placement in the Northeast census region
  had a longer duration of shunt survival between
  initial placement and both the first and second
  revisions.
• Young age and a pdx of obstructive hydrocephalus
  were associated with a higher risk of failure

47
4. Utilization / Standards of Care

• Line item utilization is not available in KID or
  NPDD
• PHIS is the most robust data source
• Look for frequency of utilization (drugs,
  imaging, labs, etc.) in a population
• Common (e.g. asthma) or rare (e.g. HLHS)
• Potential research topics
• Disparities in care
• Impact of specific diagnoses on resources
  (throughput, supplies, pharmacy, etc.)
• Adherence to evidence-based guidelines
• Evaluate the effect of clinical care guidelines
  (pre vs. post)
• Impact of case volume on outcomes

48
4. Utilization / Standards of Care (Example 1)

• Racial and economic disparity and the treatment
  of pediatric fractures
• Slover J, Gibson J, Tosteson T, Smith B, Koval
• J Pediatr Orthop. 2005 Nov-Dec25(6)717-21
• Supracondylar humerus (n 2,957), femoral shaft
  (n 1,726) or radius and ulna forearm fracture
  (n 828) as their primary diagnosis
• Hispanic (78) and black (82) patients were more
  likely to receive closed reduction with internal
  fixation of supracondylar humerus fractures than
  whites (73, P 0.02)
• No other differences noted

49
4. Utilization / Standards of Care (Example 2)

• The effect of surgical case volume on outcome
  after the Norwood procedure
• Checchia PA, McCollegan J, Daher N, Kolovos N,
  Levy F, Markovitz B
• J Thorac Cardiovasc Surg. 2005 Apr129(4)754-9
• Twenty-nine hospitals and 87 surgeons performed
  801 Norwood procedures during the study period
• Survival after the Norwood procedure is
  associated with institutional Norwood procedure
  volume but not with individual surgeon case
  volume

50
4. Utilization / Standards of Care (Example 3)

• Institutional variation in ordering complete
  blood counts for children hospitalized with
  bronchiolitis
• Tarini BA, Garrison MM, Christakis DA.
• J Hosp Med. 2007 Mar2(2)69-73
• Little evidence to support the use of diagnostic
  testing, particularly complete blood counts
  (CBCs)
• 17,397 children were included in the analysis,
  and 48.2 had at least 1 CBC, whereas 7.8 had
  more than 1 CBC
• The proportion of admissions with initial
  (23.2-70.2) and repeat (0-18.6) CBCs varied
  significantly across hospitals

51
Analytic Considerations
52
Risk Adjustment

• Necessary given heterogeneity of hospitals /
  patient populations across time
• APR-DRG severity-of-illness or case mix index may
  or may not be adequate
• Charge and los weights in PHIS
• Assigned to every discharge based on APR-DRG and
  severity level assignment
• Assume resource utilization is correlated with
  severity

53
Risk Adjustment (Continued)

• Weights compare the average charge (LOS) for each
  APR-DRG / severity level combination to the
  overall average using a national dataset
• Example
• APR-DRG1 (liver transplant), Severity3 (Major)
• Charge weight is 21.2.
• This means that the average charge for patients
  in this group were 21.2 times the average charge
  for ALL pediatric discharges, regardless of their
  APR-DRG or severity level.

54
Risk Adjustment (Continued)

• Ad hoc risk adjustment may be necessary
• Create list of factors that might impact outcome
• Patient and hospital level
• Model for parsimony (remove insignificant
  predictors)
• Clearly articulate gaps in factors unavailable in
  administrative data

55
Modeling Considerations

• Data is clustered by hospital
• All models should be hierarchical
• Varying reliability of estimates across hospitals
• Consider Bayesian shrinkage estimators, also
  useful when doing risk adjustment
• Data is retrospective and observational
• Consider matching or propensity scoring to mimic
  randomized trials, be sure to verify methods
  effect on balancing covariates

56
Modeling Considerations (Continued)

• Overwhelming power
• Reduce significance
• P-values can be ineffectual, consider alternate
  presentations
• KID and NPDD is survey data
• Use software (e.g. SAS, SUDANN) to account for
  sampling frame

57
Strengths of Administrative Data in Research

• Patient level data
• Line item utilization (PHIS)
• Population size Power
• Multiple institutions for rare conditions
• National estimates
• Hospital-to-hospital variation
• Useful for designing Phase I trials

58
Limitations of Administrative Data in Research

• Retrospective and observational
• Substantial factors for risk adjustment might be
  missing
• Outcomes are limited
• Unknown Sen / Spec for many ICD-9 codes dxs and
  pxs rely on proper documentation and coding
• Charges are billed resource, not necessarily
  administered

59
PHIS for Research Resources

• Subset of PHIS main web site
• Includes
• Standard PHIS methodology text
• Existing articles by topic
• Data Quality resources
• Data content resources
• Register at www.chca.com and select PHIS as
  requested site

60
PHIS Contacts

• CHCA
• David Bertoch (david.bertoch_at_chca.com)
• Matt Hall (matt.hall_at_chca.com)
• CHOP
• Quality Improvement
• Finnah Escritor
• PHIS-related Research
• The Center for Pediatric Clinical Effectiveness -
  Theo Zaoutis or Ron Keren