Methods

We identified 100% of Medicare fee-for-service beneficiaries (≥65 years) using 1999 to 2010 inpatient claims data from the Centers for Medicare and Medicaid Services (CMS) who had participated for at least 1 month in fee-for-service and resided and were hospitalized for PE from January 1, 1999 to December 31, 2010, in the United States. Hospitalizations for PE were defined as discharged from an acute-care hospital for a principal discharge diagnosis of PE according to the following the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes: 415.11 (Iatrogenic PE and infarction), 415.13 (saddle embolus of pulmonary artery), and 415.19 (Other PE and infarction). These ICD-9-CM codes are consistent with those used in previous studies and have been shown to have high sensitivity and specificity (positive predictive value 96% to 98%). ICD-9-CM codes in the secondary position were not used in the primary analyses because of low positive predictive value. Institutional review board approval was obtained through the Yale University Human Investigation Committee.

The demographic and clinical characteristics of patients hospitalized with PE were examined across years and stratified according to demographic factors of age (65 to 74, 75 to 84, and ≥85 years), gender, and race (white, black, and other). Race was determined using patient-reported data from the Social Security Administration, as indicated in the Medicare Denominator File. We selected 21 clinical co-morbidities based on the method used by CMS for profiling hospitals for cardiovascular conditions, which are included in Table 1 . Co-morbidities from diagnosis codes consisted of principal and secondary diagnosis codes of all hospitalizations for any cause up to 12 months before the index PE hospitalization and those found in the index admission, which were not a complication of PE.

Our outcomes included PE-specific hospitalization, all-cause mortality, and all-cause readmission rates. We calculated the PE hospitalization rate separately for each year by dividing the number of PE hospitalizations (numerator) by the corresponding person-years of fee-for-service Medicare beneficiaries for that year (denominator). Person-years were calculated for each beneficiary to account for new enrollment, disenrollment, or death during an index year. Mortality outcomes included in-hospital 30-day and 6-month mortality rates, calculated from the date of index PE admission. To calculate 30-day all-cause readmission, we identified all rehospitalizations for any cause occurring within 30 days from the date of discharge. In addition, we calculated the length of stay based on admission and discharge dates. For mortality, readmission, length of stay, and major discharge disposition analyses, we restricted our sample to unique patients. Specifically, if a patient had more than 1 admission in a given year, one hospitalization was selected at random as to minimize sample selection bias.

We used the Cochran–Armitage test to evaluate the statistical significance of trends in patient characteristics and outcomes across years. We fitted a linear mixed-effects model with a Poisson link function and state-specific random intercepts to assess the annual trends in hospitalization rates for PE, adjusted for age, gender, and race. We fitted a linear mixed-effects model with a logit link function and hospital-specific random intercepts to assess annual trends in the rate of 30-day and 6-month mortality, adjusted for age, gender, race, and co-morbidities. Change in 30-day readmission was assessed using a Cox proportional hazards model with death as a censoring event and was also adjusted for age, gender, race, and co-morbidities. The proportional hazards assumption was satisfied, demonstrating the adequacy of the Cox regression model. We incorporated analysis of trends for the outcomes by including indicator variables representing each year in the models. Using the baseline year (1999) as the referent, we calculated the adjusted incidence rate ratio (IRR) for hospitalization, adjusted odds ratio (OR) for mortality, and hazard ratios (HR) for readmission outcomes. ORs were converted to risk ratio estimates and were then multiplied for each year by the outcome rate in the baseline year to calculate the risk-adjusted mortality and readmission rates across years. All analyses were performed with SAS, version 9.3 64-bit (SAS Institute, Cary, North Carolina). All statistical tests were 2 sided at a significance level of p <0.05.