Transcript RTI ppt template - American Statistical Association

Panel Survey of Hospitals: Sampling
and Estimation Challenges
Darryl V. Creel
Third International Conference on Establishment Surveys
Montreal, Quebec, Canada June 18-21, 2007
RTI International is a trade name of Research Triangle Institute
6110 Executive Blvd. ■ Suite 902
Phone 301-770-8229
■
Rockville, Maryland, USA 27709
e-mail [email protected]
Outline
 Purpose of Survey
 Survey Design
 Frame Development
 Substitution
 Estimation
Purpose of the Survey
 Level of Drug Use and Abuse
 Monitor Drug Use and Abuse Changes over Time
Survey Design

Complete Survey Redesign (New Beginning 2004)
 Start of a New Panel – Maximize Overlap with Previous Panel
 Change in Stratification Scheme
 Geography
 Ownership
 Size
 Stratified SRS WOR of Hospitals
 Review All Emergency Department Visits
 Change in Case Eligibility
 Annual Sample Updates to Account for New Hospitals

Change in Contractors
Survey Design
 Possible Modification to the New Design
 Could Not Review All Emergency Department Visits in Some
Hospitals
 Added a Second Stage of Sampling to the New Design
 Select Days within Month to Review Emergency Department
Visits for Eligible Cases
 Retain Single-stage Design
 Standard Errors – Worst Case Single-Stage SE 97% of TwoStage SE
 Computational Time – Single-Stage ~ 300 Observations
Took 12 Minutes, Two-Stage ~ 70,000 Observations Took
31.5 Hours
Frame Development
 American Hospital Association (AHA) Annual Survey Database
 Good Coverage
 Inaccuracies in the File
 Incorrect FIPS Information Some for Counties in FL
 Nonexistant Codes
 Inaccuracies in the Documentation
 Counts for Changes in the Population not Consistent
 Only Use a Few Variables from AHA
 Stratification
 Eligibility Determination
Frame Development

Changes in Populations of Statistical Units (Struijs and Willeboordse)
 Change of Characteristics
 Change of Existence
 Birth
 Death
 Change of Structure
 Concentration
 Merger
 Takeover
 Deconcentration
 Break-Up
 Split-Off
 Restructuring

Track for Allocation Emergency Department Visits Counts
Substitution
 Hospitals are Difficult to Recruit
 Use Responding Not Sampled Hospitals from the Previous
Panel for Nonresponding Sampled Hospitals in Current Panel
until Number of Nonresponding Hospitals in Current Panel
Decreases
 Substitution Only Occurs within a Sampling Stratum, i.e., Same
Geography, Ownership Category, and Size Category
 Match Number of Emergency Department Visits as Closely as
Possible
Estimation
 Main Focus on Analysis of Trends
 Drug Classification Scheme Changes Constantly
 Apply Most Recent Classification Scheme
Retrospectively for Trends
Estimation
 Actual Zero Eligible Cases
 Some Hospitals Not on Micro-data Set, i.e., No Eligible
Cases Found at Hospital During Charts Review
 Hospitals on Micro-data Set May Not Have Specific Drugs,
e.g., PCP
 Carries Through to Specific Analysis
 Generate Data with a Zero Value
 Statistician Involvement Across All Aspects of the Survey
 SUDAAN – SAMPCNT Statement
Summary
 Purpose of Survey
 Survey Design
 Frame Development
 Substitution
 Estimation
 Statisticians Involved