We used data from Premier's Perspective Comparative Database, a large US hospital clinical and economic database developed for quality and utilization benchmarking. This database includes patient-level data on each admission from approximately 800 acute care hospitals across the United States, providing nationally representative information on nearly 5 million annual hospital discharges at both rural and urban hospitals. All data are organized by discharge month. We use the term patient to refer to a discrete admission; an individual patient may be included in the database more than once.
To avoid any possible issues regarding the use of protected health information in the analyses, dates of admission and discharge were reported by month and year; day-of-service detail was provided using chronological days; and the age of patients older than 89 years were assigned an age of 89 years. The Saint Louis University Institutional Review Board approved the study and waived the requirement for patient informed consent.
Data were obtained from several periods to allow for examination of changes in prescribing patterns following the release of the 2 sentinel articles challenging the safety of nesiritide (March 29 and April 20, 2005). The specific periods were selected to allow for before-and-after publication comparisons and a reference point before nesiritide was introduced. Specifically, we examined the period January to April 2005, which includes the 4 months just prior to the publication of the mortality article28
and the subsequent 8 months in the same calendar year (May-December). To compare on a year-over-year basis, we divided the previous year into the same periods (January-April 2004 and May-December 2004). In addition, we obtained data for the time frame of January to August 2001, representing the period prior to the approval and introduction of nesiritide to the market (September 2001) to examine secular trends in the utilization of vasoactive therapies. The number of hospitals (n=491) varied slightly over time, contributing to differences in the number of patients during each period.
Variables in the PREMIER database include patient demographic information (eg, age, sex, and race based on UB92 coding), admission and discharge dates by month and year, concurrent background heart failure therapy (ACE inhibitor, angiotensin-receptor antagonist, β adrenergic antagonist, digoxin, diuretic, aldosterone antagonist), type of admission, length of stay (including days in intensive care), intravenous drug used (including day of initiation and discontinuation of therapy), patient discharge status, primary and secondary International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes (limited to 1 secondary code in the 2001 sample), hospital characteristics (size, geographic location, and teaching status), payer type, and the specialty of both the admitting and attending physician.
Inclusion criteria included (1) a primary ICD-9 diagnosis of heart failure (a detailed list is available on request), (2) age older than 18 years, and (3) acute care inpatient status. Standardized charge codes were used to identify drugs administered during the hospitalization. We defined intravenous vasoactive therapy as any one of the following: nesiritide, nitroglycerin, sodium nitroprus-side, dobutamine, dopamine, or milrinone. We also recorded the use of all 3 available intravenous loop diuretics (furosemide, bumetanide, and torsemide). We defined cardiology care if either the admitting or attending physician was coded as a cardiologist.
Secondary analyses were performed in which all admissions with a secondary ICD-9 code for heart failure were considered. Additionally, we were interested in the use of nesiritide for patients who did not have a primary or secondary ICD-9 code for heart failure to understand the scope of use that might fall outside the approved labeling for the drug and the associated underlying primary diagnoses in this cohort. We also separately analyzed the use of intravenous drugs in patients treated in hospitals that contributed patient data throughout 2004 and 2005; the number decreased modestly from 341 in the January-April 2004 period to 320 from March to December 2005.
A χ2 analysis was used to compare frequency of drug use across the periods of observation. One-way analysis of variance was used to test for overall differences in total length of stay, accumulative days receiving therapy, and day of initiation of drug across the 4 periods examined in 2004 and 2005. Tukey post hoc tests were then conducted to determine exactly which periods differed significantly from one another.
We used logistic regression to identify characteristics associated with physician use of nesiritide among patients with heart failure. The following characteristics were included: age (19-64, 65-74, 75-84, ≥85), race (black, other), sex, use of background heart failure medications as defined above, hospital location (urban or rural) and geographical region, hospital teaching status, hospital size (defined by 0-100, 101-400, 401-600, >601 beds), physician (cardiologist, other), and payer type. Odds ratios (ORs) were calculated to evaluate risk. Time interaction variables for before and after publication of the survival article were used to evaluate risks based on patient and hospital characteristics for both periods, as well as change in risk between periods. Data management and analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, NC). Differences were considered statistically significant at a 2-sided P<.05 level.