Author + information
- Received January 14, 2013
- Revision received April 29, 2013
- Accepted April 29, 2013
- Published online August 1, 2013.
- Chohreh Partovian, MD, PhD∗,†∗ (, )
- Shu-Xia Li, PhD†,
- Xiao Xu, PhD†,‡,
- Haiqun Lin, PhD†,§,
- Kelly M. Strait, MS†,
- John Hwa, MD, PhD∗ and
- Harlan M. Krumholz, MD, SM∗,†,§⋮
- ∗Section of Cardiovascular Medicine, Department of Internal Medicine, Yale University School of Medicine, New Haven, Connecticut
- †Center for Outcomes Research and Evaluation, Yale-New Haven Hospital, New Haven, Connecticut
- ‡Department of Obstetrics, Gynecology, and Reproductive Sciences, Section of Comparative Effectiveness Research, Yale University School of Medicine, New Haven, Connecticut
- §Department of Health Policy and Management, Yale School of Public Health, New Haven, Connecticut
- ⋮Robert Wood Johnson Clinical Scholars Program, Department of Internal Medicine, Yale University School of Medicine, New Haven, Connecticut
- ↵∗Reprint requests and correspondence:
Dr. Chohreh Partovian, Yale University School of Medicine, Department of Internal Medicine/Cardiology, 1 Church Street, New Haven, Connecticut 06510.
Objectives This study sought to determine hospital patterns of change in use of nesiritide over a 6-year period after publications of safety concerns in 2005 and to identify hospital characteristics associated with these patterns.
Background The changing nature of medical evidence often requires a change in practice. Nesiritide was commercialized in 2001 for early relief of dyspnea in patients with decompensated heart failure. In 2005, concerns about its safety led to recommendations to restrict its use. Little is known about how hospitals responded to this information.
Methods We analyzed data from the Premier database, including 403 hospitals contributing 813,783 hospitalizations with heart failure from 2005 to 2010. We applied a growth mixture modeling approach to hospital-level, risk-standardized, quarterly use rates of nesiritide to distinguish hospital groups on the basis of their patterns of change in use.
Results The proportion of hospitalizations using nesiritide declined from 15.4% in 2005 to 1.2% in 2010. The level and speed of change varied markedly among hospitals. After adjusting for differences in patient characteristics across hospitals and years, we identified 3 distinct groups of hospitals: “low users,” “fast de-adopters,” and “slow de-adopters.” In multivariate regression analysis, these groups did not differ in traditional hospital characteristics, such as size, urban setting, or teaching status.
Conclusions We identified 3 distinct hospital groups characterized by their patterns of change in nesiritide use. These trajectory curves can provide hospitals with important feedback on how fast and effectively they react to new information compared with other hospitals. Uncovering factors that promote organizational learning requires further research.
The changing nature of medical evidence often requires a change in practice. Studies have described the challenges of translating new information into practice, which may take decades, as it did with the BHAT (Beta-Blocker Heart Attack Trial) (1). To our knowledge, no studies have evaluated longitudinal patterns of change in practice at the hospital level. Nesiritide (Natrecor, Scios, Inc., Fremont, California) provides a good case study of how hospitals changed practices in response to new information. Nesiritide was approved by the Food and Drug Administration in 2001 for early relief of dyspnea in patients with acutely decompensated heart failure (HF), but once on the market, it was widely prescribed and used beyond its original indication (2). In Spring 2005, 2 meta-analyses of small randomized trials raised concerns regarding renal toxicity (3) and higher mortality associated with nesiritide (4). These publications resulted in a Food and Drug Administration–mandated revision of prescribing information in the “Adverse Reactions/Effects on Mortality” section. A panel of experts recommended in June 2005 that nesiritide be used only in patients with acutely decompensated HF who had dyspnea at rest and not be used for improvement of renal function, enhancement of diuresis, intermittent outpatient infusion, or scheduled repetitive use (5). To physicians planning the use of nesiritide to relieve symptoms, the panel recommended considering the use of alternative therapies. In 2011, the results of a large randomized trial, the ASCEND-HF (Acute Study of Clinical Effectiveness of Nesiritide in Decompensated Heart Failure), showed that nesiritide had no effect on dyspnea, renal function, mortality, or readmission but was associated with increased rates of hypotension, and it was concluded that nesiritide could not be recommended for routine use in patients with acute HF (6).
Prior work by Hauptman et al. (7) has shown that the overall use of nesiritide decreased by 66% (from 16.6% to 5.6%) between March and December 2005 (7). Their study focused on overall change in use immediately before and after the publications of safety concerns. Our current study was designed to extend prior work by evaluating the patterns of change among hospitals between 2005 and 2010. We hypothesized that amid a continuing general decrease in nesiritide use, there would be a marked heterogeneity in the level and speed of de-adoption across hospitals, revealing various institutional responses to new information. We also sought to determine what hospital characteristics would be associated with these distinct hospital groups.
We used data from a voluntary, fee-supported database developed by Premier, Inc., Charlotte, North Carolina, for measuring quality and health care use. Containing >330 million discharges from 620 geographically diverse hospitals, the database represents 1 in every 5 discharges from U.S. hospitals. In addition to the information available in the standard hospital discharge file, the Premier database contains a date-stamped log of all billed items at the individual patient level, including medications and laboratory, diagnostic, and therapeutic services. We used data from calendar years 2005 to 2010 for our analysis.
Patient data are de-identified in accordance with the Health Insurance Portability and Accountability Act, and a random hospital identifier assigned by Premier is used to identify individual hospitals. The Yale University Human Investigation Committee determined that this study is not considered to be Human Subjects Research as defined by the Office of Human Research Protections.
Heart failure cohort
All hospitalizations from January 1, 2005, to December 31, 2010, were included in the study cohort, with a principal diagnosis of HF as defined by the International Classification of Diseases-Ninth Revision-Clinical Modification (ICD-9-CM) codes 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, and 428.xx or a principal diagnosis of respiratory failure (ICD-9-CM code 518.81) with a secondary diagnosis of congestive HF (ICD-9-CM code 428.0). We excluded patients who were less than 18 years of age at the time of admission or those whose physicians were pediatricians, because our focus was not on congenital disease. A patient could contribute more than one hospitalization to the study cohort.
Patient and hospital characteristics
Patient characteristics available in our dataset included age, sex, race/ethnicity, insurance status, and comorbidities. We used the Healthcare Costs and Utilization Project software provided by the Agency for Healthcare Research and Quality to classify comorbidities from the standard hospital discharge file on the basis of methods described by Elixhauser et al. (8).
For each hospital, the Premier database contains information, collected from the American Hospital Association database, on bed count, teaching status, geographic location, and whether it serves an urban or rural population. In addition, we derived the following measures about each hospital's characteristics by pooling its patient-level hospitalization data across 2005 to 2010: average number of HF hospitalizations each year, proportion of the attending physicians being a cardiologist, proportion of patients with Medicaid as the primary payer, whether the hospital had any cardiology intensive care unit, and capability of performing a number of procedures, including ventricular assist device or heart transplant, percutaneous coronary intervention, and implantable cardioverter-defibrillator (ICD) insertion.
Descriptive statistics (frequencies and percentages) were calculated to assess sample characteristics and drug use. We assessed the proportion of nesiritide use at hospitalization-level (denominator being all hospitalizations with HF across all hospitals) and compared it with the use of potential alternative therapies, including other vasodilators (intravenous [IV] nitroglycerin and sodium nitroprusside) and positive inotropic agents (dobutamine, dopamine, and milrinone).
We also assessed nesiritide use at the hospital level (the denominator being all hospitalizations with HF in a given hospital). Hierarchical generalized linear modeling was used to calculate hospital-level, risk-standardized use rates of nesiritide (9). The model included patient demographic characteristics (age groups, sex, race/ethnicity), comorbidities, and a hospital random effect for each calendar quarter. This model specification takes into account within-hospital correlation of use patterns while adjusting for differences in case mix both across hospitals and over time. The full list of risk variables included in the hierarchical generalized linear models with their estimated odds ratios and corresponding 95% confidence intervals (CIs) are reported in Online Table 1.
We applied a growth mixture modeling approach to hospital risk-standardized use rates via a SAS macro Proc Traj (10). This approach assumes there are clusters or groupings of distinctive patterns of change in a population (11). All hospitals that contributed HF hospitalizations in at least one calendar quarter were included in the analysis. Models with different number of trajectory groups were estimated, and the optimal number of distinct trajectory groups was determined by comparing the Bayesian Information Criteria index across these models. Our final analysis used a 3-group model that had the most favorable Bayesian Information Criteria index. Each hospital was assigned to a trajectory group on the basis of the estimated posterior probability of its group membership (i.e., following a maximum posterior probability assignment rule) (11).
Chi-square and Kruskal-Wallis tests were used to assess whether there were any significant associations between individual hospital characteristics and identified trajectory groups. Multivariate multinomial logistic regression analysis was also performed to examine the association between hospital characteristics and trajectory group membership. Stepwise selection algorithm was used to choose the variables included in the final multivariate model. Estimates with p < 0.05 were considered statistically significant. Analyses were conducted with SAS version 9.2 (SAS Institute Inc., Cary, North Carolina), and Figures 1 to 3⇓⇓⇓ were created with R version 2.11.1 (12).
Use of nesiritide at hospitalization level
Between 2005 and 2010, there were 813,783 hospitalizations with HF. Among these hospitalizations, the proportion using nesiritide decreased from 15.4% (5,508 of 35,769) in the first quarter of 2005 to 1.2% (429 of 35,872) in the last quarter of 2010 (Fig. 1). The sharpest decrease in use occurred between the second and third quarters of 2005 when the odds ratio for being treated by nesiritide (compared with the last quarter of 2010) decreased from 15.4 (95% CI: 11.9 to 19.8) to 8.7 (95% CI: 6.7 to 11.3) (Online Table 1). During the same period, the proportion of hospitalizations including IV nitroglycerin remained stable between 6% and 8%: 6.5% (2,325 of 35,769) in the first quarter of 2005 and 7.3% (2,612 of 35,872) in the last quarter of 2010. The proportion of hospitalizations using sodium nitroprusside was less than 1% throughout the 6-year period: 0.6% (208 of 35,769) in the first quarter of 2005 and 0.4% (136 of 35,872) in the last quarter of 2010. The proportion of hospitalizations with a positive inotropic agent was 12.1% (4,328 of 35,769) in the first quarter of 2005, 12.3% (4,787 of 38,978) in the first quarter of 2006, and then decreased progressively to 9.8% (3,516 of 35,872) in the last quarter of 2010 (Fig. 1).
Use of nesiritide at hospital level
Between 2005 and 2010, a total of 403 hospitals contributed data on patients with HF to the database. These were mainly urban, nonteaching, small- and medium-sized hospitals. The key characteristics of these hospitals are summarized in Table 1.
There was a wide variation across hospitals in the proportion of patients with HF treated with nesiritide. In the first quarter of 2005, the risk-standardized rates ranged from a minimum of 1% to a maximum of 65.9% (median: 11.4%, interquartile range [IQR]: 5.6% to 20.8%). In the last quarter of 2010, the adjusted rates ranged from a minimum of 0.3% to a maximum of 19.2% (median: 0.7%, IQR: 0.5% to 1%) (Fig. 2).
Hospital groups based on patterns of change in nesiritide use
Application of the growth mixture modeling to hospital risk-standardized use rates led to the emergence of 3 distinct groups of hospitals based on their patterns of change in use over time: “low users,” “fast de-adopters,” and “slow de-adopters” (Fig. 3). The approach took into account both level and speed of change in use over the entire 6-year period; however, for the sake of simplicity, only the most dominant attribute was used to name the groups. The “low-user” group included 302 hospitals (75% of hospitals, together accounting for 69% of all hospitalizations) with an average risk-standardized rate of 9% in the first quarter of 2005, which decreased to approximately 2% at the beginning of 2006 and plateaued at approximately 1% from 2009. The “fast de-adopter” group included 82 hospitals (20% of hospitals, together accounting for 25% of all hospitalizations) with an average initial risk-standardized use rate of 26% that decreased to 10% at the beginning of 2006, 5% at 2009, and 3% at the end of 2010. The remaining 19 hospitals (5% of hospitals, together encompassing 6% of all hospitalizations) were classified as the “slow de-adopters.” They had the highest initial risk-standardized use rates and a slower rate of decrease in use over time than the other hospitals. They started with an average use rate of 38%, which decreased to 26% at the end of 2005, 20% at 2007, and 10% at the beginning of 2010 (Fig. 3).
The average posterior probability of group membership was >0.98 for each of the groups, indicating excellent performance of the model in distinguishing the different trajectory patterns.
Association between hospital characteristics and distinct hospital groups
We investigated what hospital characteristics were associated with different nesiritide de-adoption trajectory groups. Table 2 shows the hospital characteristics by trajectory group. The 3 groups differed significantly in hospital size, annual volume of HF hospitalizations, regional location, percutaneous coronary intervention and ICD capability, proportion of cardiologist as attending physician, and proportion of Medicaid patients (Table 2). However, in multivariate regression analysis, none of the hospital characteristics differed significantly between the slow de-adopters group and the other 2 groups. The fast de-adopters were more likely to be located in the Midwest and the South, to have ICD capability, and to have a higher proportion of Medicaid patients in comparison with low users (Table 3).
In this study, we used data from a large network of hospitals to characterize longitudinal patterns of change in nesiritide use after publications raising concerns about its safety. The results showed a continued reduction in the use of this medication between 2005 and 2010, with an initial sharp decrease immediately after the publications followed by a more gradual decrease between 2006 and 2010. However, the overall average change obscures that there was marked variation in nesiritide use across hospitals. When taking into account both level and speed of change in use over the 6-year period, the hospital trajectories coalesced around some specific patterns leading to the emergence of 3 distinct groups of hospitals. Because the use rates already adjusted for differences in case mix across hospitals and across years, these 3 groups depict mainly the heterogeneity of organizational response to new information. These trajectory curves can provide crucial feedback to hospitals about how fast and effectively they react to new information in comparison with other hospitals.
We chose to use hospitals as our unit of analysis for several reasons. First, patients with HF are usually seen by multiple physicians, and it is not always possible to identify the prescribing physician. Second, revealing variation at the hospital level rather than individual physician level is consistent with an emerging appreciation of team-based care, systems of care, and the impact of hospital internal environment on performance (13–16). Third, medical decision making is influenced by various organizational characteristics, such as team composition (number and type of specialists on the team, inclusion of a pharmacist), internal culture (quality and frequency of communication and collaboration between team members), regulatory context (drug formularies), and availability and use of clinical decision support systems for the practice of evidence-based medicine (13,17,18). However, one of the limitations of our study and the currently available healthcare databases in general is the lack of information on these characteristics.
Our results suggested that there may be common underlying factors among hospitals within each trajectory group. However, when we examined the association between hospital characteristics available in our database and various trajectory groups, none was significantly associated with a hospital's likelihood of being in the slow de-adopter group compared with the other 2 groups. This could be due to the small number of hospitals in this group or to the data limitations (i.e., lack of measures reflecting team composition, communication, internal culture, regulations, and restrictions). There is a need for further qualitative and mixed method research to identify additional factors, both internal to the organization and external. For example, one of the unmeasured factors that may explain the significant difference in regional location observed between hospital groups could be the prevalence of pharmaceutical marketing across regions.
Before its safety concerns were published in 2005, nesiritide was widely prescribed (7,19). The proportion of HF hospitalizations using nesiritide almost doubled those with the main alternative vasodilator, IV nitroglycerin, despite the fact that nesiritide was approved only for specific indications and was more expensive. After the publications, the rate of nesiritide use declined dramatically, but we did not observe a “substitution” effect, such as a sudden or substantial increase in use of other vasodilators or positive inotropic agents. These results could suggest a case of nesiritide overuse before Spring 2005.
Our study further revealed that this initial, short-term strong response to new information was followed by a steady decrease in use over subsequent years, although at a more gradual level and speed. This pattern is consistent with what has been observed in many other studies of the adoption of innovations. Those studies have suggested that adoption decisions of organizations are a function of both internal factors and external and social factors, but the relative importance of these factors change over time as information diffuses among potential adopters (20–22).
First, hospitals included may not be a representative sample of all hospitals in the United States. Nevertheless, the Premier database contains approximately 20% of annual nationwide acute care hospitalizations. Second, a patient could contribute more than 1 hospitalization to the study cohort, introducing correlation in data between the multiple hospitalizations. However the impact was likely small because only 9% of patients had more than 1 hospitalization per quarter, of whom the majority had only 2 hospitalizations (median: 2; IQR: 2 to 2). Third, our risk-adjustment model relied on claims data only. However, our earlier work of profiling hospital performance for the Centers for Medicare and Medicaid Services has demonstrated that administrative data can provide estimates similar to models using richer clinical data (23,24). Finally, as previously mentioned, we lack data on a number of characteristics that might have affected drug use, such as formularies, other hospital restrictions, and marketing factors.
This study establishes that amid a general decrease in nesiritide use, there were important variations across hospitals revealing distinct hospital groups based on their patterns of change in practice in response to new information. These trajectory curves can provide hospitals with important feedback on their “learning rates” or how fast and effectively they react to new information. The study also highlights the need for additional mixed-methods research to uncover the factors that foster or impede organizational learning.
This work was supported by Grant DF10-301 from the Patrick and Catherine Weldon Donaghue Medical Research Foundation in West Hartford, Connecticut, and by Grant UL1 RR024139-06S1 from the National Center for Advancing Translational Sciences in Bethesda, Maryland. Dr. Krumholz is supported by Grant U01 HL105270-03 (Center for Cardiovascular Outcomes Research at Yale University); is the recipient of a research grant from Medtronic, Inc., through Yale University; and is chair of a cardiac scientific advisory board for UnitedHealth. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- heart failure
- implantable cardioverter-defibrillator
- International Classification of Diseases-Ninth Revision-Clinical Modification
- interquartile range
- Received January 14, 2013.
- Revision received April 29, 2013.
- Accepted April 29, 2013.
- American College of Cardiology Foundation
- Sackner-Bernstein J.D.,
- Skopicki H.A.,
- Aaronson K.D.
- Braunwald E.,
- Burnett J.C.,
- Colucci W.S.,
- et al.
- Jones B.L.,
- Nagin D.S.
- Nagin D.S.
- Team R.
- Fonarow G.C.,
- Heywood J.T.,
- Heidenreich P.A.,
- Lopatin M.,
- Yancy C.W.
- Burns L.R.,
- Wholey D.R.
- Rogers E.
- Young G.J.,
- Charns M.P.,
- Shortell S.M.
- Keenan P.S.,
- Normand S.L.,
- Lin Z.,
- et al.
- Krumholz H.M.,
- Wang Y.,
- Mattera J.A.,
- et al.