The MPSMS was a national surveillance project created in 2001 under the leadership of the Centers for Medicare and Medicaid Services to determine the rates of specific medical adverse events within the hospitalized Medicare population. The MPSMS defines an adverse medical event as “an unintended, measurable harm, injury or loss more likely associated with the patient’s interaction with the health care delivery system than from any attendant disease process” [5
]. The MPSMS sample was a subset of the Hospital Payment Monitoring Program (HPMP) record sample. The HPMP sample, which represents approximately 40% of all Medicare fee-for-service payments, was randomly selected each month from the Medicare National Claims History (NCH) File by the Centers for Medicare and Medicaid Services from a pool of approximately 1 million Medicare beneficiary hospital discharges across 50 states, Washington DC, Puerto Rico, and the Virgin Islands.
The study sample was drawn from the MPSMS database that included more than 180,000 hospital discharges between January 1, 2002, and December 31, 2007. All 1809 THA patients in this sample of greater than 180,000 patients were analyzed for the MPSMS. These 1809 THA patients were treated in hospitals from all 50 states. Hospitals forwarded the selected medical records to the Clinical Data Abstraction Centers for data abstraction on a monthly basis. The overall average aggregate agreement rate across all data elements of the MPSMS data is greater than 97% [5
]. The agreement rates range from 96% to 99% for variables used to identify exposures and 94% to 99% for identification of adverse events. This analysis suggests the abstractors were accurate in their data collection. For this study, all patients included in the sample had to have a THA for degenerative arthritis during their hospitalizations.
Abstracted patient characteristics included demographics (age, gender, race) and selected common clinical characteristics and comorbidities, including congestive heart failure, chronic obstructive pulmonary disease, cerebrovascular disease, obesity, cardiovascular disease, diabetes, history or current smoking, and use of corticosteroids (Table ). Age was divided into four groups (younger than 65 years, 65–74 years, 75–84 years, and 85 years or older), gender was coded as female versus male, and race was categorized as white versus others.
The primary outcome was the adverse event (Table ). Secondary outcomes included 30-day and in-hospital mortalities, length of stay (LOS), and 30-day readmissions. The 30-day all-cause mortality was defined as any death within 30 days after the procedure and the in-hospital mortality was defined as all-cause death within the index hospitalization. LOS was calculated as a difference between dates of discharge and admission. If a patient was discharged on the same date of admission, his/her LOS was defined as 1 day. The 30-day readmission was categorized in two ways: (1) the rate of all-cause readmission that only counts the first rehospitalization within 30 days after discharge from the index hospitalization, regardless of the condition of a rehospitalization, and (2) the cause of readmission that counts as any rehospitalization within 30 days after discharge. The NCH database was used to obtain the in-hospital mortality and readmission information, and the Medicare Enrollment database was linked to obtain the 30-day mortality information.
Definitions of adverse events
Postprocedure events were captured if the specific terms were found anywhere in the medical record. Abstractors were registered nurses and nurse practitioners. The abstractors used the algorithms created by physicians specifically for this project. The physicians responsible for defining these algorithms were from a variety of specialties, including infectious disease, general internal medicine, nephrology, pulmonology, critical care, neurology, cardiology, general surgery, and orthopaedic surgery. The medical record included progress notes, nursing notes, procedure notes, consultations, history and physical examinations, discharge summaries, intraoperative anesthesia records, operating room circulating nurse notes, emergency department notes, laboratory results, transfusion records, nursing admission assessments, radiology reports, and pathology reports. The adverse events were specifically defined by the technical experts of the MPSMS (Table ).
We divided the sample into two subperiods (2002–2004 and 2005–2007) to seek changes in the patterns of adverse events and outcomes. We elected to do this to assure the validity of the statistical analysis, as the rate of two of the variables measured (age and obesity) changed from 2002 to 2007. Descriptive and bivariate analyses were conducted to compare patient characteristics, observed adverse events, and outcomes between the two periods (see Table for variables included); a chi square test was used to compare dichotomous and categorical variables and a t test to compare continuous variables. We used the hierarchical generalized linear modeling (HGLM) approach to assess the association of adverse events with the outcomes and to assess the odds of change in adverse events over time. This modeling approach was also used to assess the relationship of adverse events with patient characteristics by modeling the log-odds of adverse events as a function of patient demographic and clinical variables adjusted for year variable, which was coded as 1 to 6 denoting 2002 to 2007. All HGLMs were fitted with a random state-specific effect to account for within-state correlation of the observed adverse events and outcomes and to separate within-state variation from between-state variation. The 95% confidence interval (CI) was calculated for each estimate obtained from models. All of the statistical analyses were conducted with SAS® Version 9.1.3 (SAS Institute Inc. Cary, NC), and HGLMs were estimated using the GLIMMIX macro in SAS®.