Building Capacity for Supply-Side Modeling, Simulation, and Research: An Example Using HCUP Data to Improve Timeliness of Estimates September 21, 2011 Claudia Steiner, M.D, M.P.H.

What is HCUP? HCUP is HCUP is – Longitudinal Multi-Year and All-Payer, Inpatient, Emergency Department, and Ambulatory Surgery Databases based on Hospital Billing Data. 2

Demographic Data DiagnosesProceduresCharges 3 The Foundation of HCUP Data is Hospital Billing Data

The HCUP Partnership 4 State Federal Industry

5 Partnership: HCUP Database Participation By State

Three state-level databases Three state-level databases State Ambulatory Surgery Databases (SASD) State Inpatient Databases(SID) State Emergency Department Databases (SEDD) 6 HCUP Has Six Types of Databases

Three nationwide databases Three nationwide databases HCUP Has Six Types of Databases Nationwide Inpatient Sample(NIS) Kids’ Inpatient Database(KID) Nationwide Emergency Department Sample (NEDS) 7

8 Included in HCUP Inpatient care State Inpatient Databases (SID) Nationwide Inpatient Sample (NIS) Kids’ Inpatient Database (KID) Emergency Department State Emergency Department Databases (SEDD) Nationwide Emergency Department Sample (NEDS) Ambulatory Surgery State Ambulatory Surgery Databases (SASD) Not Included in HCUP Physician office visits Pharmacy Labs/Radiology What Types of Care Are and Are Not Captured by HCUP?

9 Source: American Hospital Association (AHA), % (N=805) Typically not included in HCUP data Included in HCUP data 86% (N=5,010) HCUP data is mostly from community hospitals Where Do We Get HCUP Data?

What Are Community Hospitals? 10 IncludedExcluded Multi-specialty general hospitalsLong-term care OB-GYNPsychiatric ENT Alcoholism/Chemical dependency OrthopedicRehabilitation PediatricDoD / VA / IHS Public Academic medical centers American Hospital Association Definition: Non-Federal, short-term, general, and other specialty hospitals, excluding hospital units of other institutions (e.g., prisons)

Accelerating HCUP Data and Information Need for timely projections on trends – – Provide analysts and policy makers timely information that can be used to evaluate the impact of quality improvement efforts – – HCUP Nationwide Inpatient Sample (NIS) typically lags the current calendar year by 17 months 2009 NIS available in June 2011 Demonstrate feasibility of producing gap-year national estimates Demonstrate feasibility of collecting and processing quarterly data 11

Which HCUP Partners Collect Quarterly Data? A total of 40 of 44 States (91%) reported that they collect data at more frequent intervals than annually: A total of 40 of 44 States (91%) reported that they collect data at more frequent intervals than annually: – 23 States collect quarterly data (AR, CT, FL, GA, HI, IA, IL, IN, KY, MA, ME, MD, MI, MN, MO, MT, NC, NE, NM, NY, OH, OR, PA, RI, TN, TX, UT, VA, VT, WI & WY) – 4 States collect monthly data (NJ, SC, WA & WV) – 3 States collect both quarterly and monthly data (CO, NH & NV) – 2 State collects semi-annual data (AZ, CA) Four of the 44 States do not collect data more frequently than annually: Kansas, Louisiana, Oklahoma, and South Dakota. Four of the 44 States do not collect data more frequently than annually: Kansas, Louisiana, Oklahoma, and South Dakota. 7

HCUP Data for Timely National Projections Factors that contribute to success of the initiative: – – Longitudinal nature of the HCUP databases 1988 forward – – Breadth of data across 44 states 295 million inpatient discharges from the 2001 to 2009 – – Capacity of states to produce early quarterly data – – Modeling expertise at AHRQ and contract staff – – Availability of SAS Econometric Time Series Software – – Leveraging of report technology developed under the NHQR 13

Selected HAIs and Outcomes Eight HAIs selected; six reported separately for adults and pediatrics The HAIs reported in this study may have originated from either inpatient or outpatient health care services HAIs are identified by a principal or secondary diagnosis on an inpatient stay Indication that the diagnosis was present on admission (POA) could not be considered because POA is not available in historical SID Approach provides nationwide, population- based prevalence instead of the hospital-based performance or accountability measures 14

Five Outcomes of Interest Projections focus on hospital utilization and outcomes: – – Number of inpatient discharges – – Rate per 1,000 discharges – – Average total charge (includes hospital services, no professional fees, not inflation-adjusted) – – Average length of stay – – In-hospital mortality rate 15

Postoperative Sepsis (Adult) Population at risk: Elective, non-maternal, adult, surgical discharges with a length of stay >= four days, excluding discharges with any diagnosis of immunocompromised state, discharges with any diagnosis of cancer, and discharges with a principal diagnosis of infection 16

Postoperative Sepsis (Pediatric) Population at risk: Non-neonatal, pediatric, surgical discharges with a length of stay >= four days, excluding discharges with a principal diagnosis of infection or a DRG indicating surgery for likely infection 17

Clostridium Difficile Infections (Adult) Population at risk: Non-maternal, adult discharges 18

Clostridium Difficile Infections (Pediatric) Population at risk: Pediatric discharges 19

HCUP Data for Timely National Projections HCUP projections in newest report are based on: – – 295 million inpatient discharges from the 2001 to 2009 HCUP SID “Early” 2010 data from 14 selected HCUP States that submitted data by July 2011 Ten cardiovascular / cerebrovascular conditions and procedures selected – – Each stratified by adult age (18-44, 45-64, 65+) and gender 20

Five Outcomes of Interest Projections focus on hospital utilization and outcomes: – – Number of inpatient discharges – – Average total cost (includes hospital services, no professional fees, not inflation-adjusted) – – Average length of stay – – In-hospital mortality rate 21

Acute Myocardial Infarction (Adult Age Group) 22

Acute Myocardial Infarction (Gender) 23

HCUP Data Mining Purpose: Use early 2010 State Inpatient Data to identify diagnoses and procedures for which observed outcomes in 2010 digressed substantially from those outcomes predicted for 2010 using historical data from Purpose: Use early 2010 State Inpatient Data to identify diagnoses and procedures for which observed outcomes in 2010 digressed substantially from those outcomes predicted for 2010 using historical data from Method: Analyze normalized residuals to identify the 2010 residuals that were statistical outliers compared with residuals observed during the baseline period. These outlier residuals indicate potentially radical changes to the established trend for the outcome under consideration. Method: Analyze normalized residuals to identify the 2010 residuals that were statistical outliers compared with residuals observed during the baseline period. These outlier residuals indicate potentially radical changes to the established trend for the outcome under consideration. 24

Procedure Categories with Substantial Deviations Between Actual vs. Expected 25

Questions? 26

27 Healthcare Cost and Utilization Project (HCUP)