Copyright © 2003 by the European Society of Cardiology.
Editorial
How should we analyse hospitalizations in clinical trials?
1 University of Bergen, Cardiology Division, Central Hospital in Rogaland, Stravanger, Norway
2 Merck Research Laboratories, West Point, PA, USA
See doi:10.1016/S1095-668X(02)00384-6for the article to which this editorial refers.
Heart failure is the fourth leading cause of hospitalization in the United States and the most frequent cause in the population over 65 years of age.1 Due to the rising prevalence of heart failure in the aging population, morbidity and mortality from this disease is increasing in epidemic proportions.2 While it is impossible to measure perfectly the burden of heart failure on patients, their families and society, both the acute and chronic symptoms of heart failure eventually translate into increased hospitalization time.3 Admission to hospital due to heart failure is a clinically important event and is associated with a striking increase in the risk of death.4 However, assessment of the overall impact of hospitalization is not straightforward, and one of the challenges involved in designing and performing clinical trials in these patients is to determine the mostappropriate method.
Hospitalization data does not represent reporting of surrogate endpoints. Surrogate endpoints relate to biologic mechanisms involved in the disease process. In patients with heart failure these include haemodynamic measurements, arrhythmias, exercise capacity, neurohumoral profile, ejection fraction and autonomic nervous system markers.5 True endpoints should be biologically sensible, readily identifiable, reliably measurable, and of clinical importance. Hospitalization undoubtedly fulfils these criteria. Due to the large number of events, hospitalization data will usually permit greater confidence in the conclusions. Taken together, analysis of mortality and morbidity is a measure of the coherency of the data.
In this issue, Metcalfe et al. focus on the analysis of reported hospitalization data from 130 clinical studies in patients with heart failure published during 1999 and 2000. The authors present a critical analysis of the statistical methodology employed in reporting hospitalization as outcome data in the studies reviewed.6 They found that over 70% of the papers report, analyse and graph data based solely on the first hospitalization for heart failure. Few of the studies provide data on the number of admissions or the actual amount of time spent in hospital. The criteria used to determine the primary admission diagnosis are rarely defined and the data are seldom adjudicated. Clearly, both the frequency and duration of admissions are important indicators of the disease burden for the patient and reflect the demands placed on health service resources. Limiting the reporting to data on first admissions results in loss of clinically important information.
Despite this, it is not surprising that the majority of papers focus on the first hospitalization only. There are standard methods for time-to-event analyses that produce an easily understood measure of the treatment effect (the risk ratio), a valid statistical test (the logrank test or the Cox model, for example), and a clear graphical representation of the results (the Kaplan–Meier curve). While there are better measures of the total disease burden, such as the total number of hospital admissions, the rate of admissions, and the total number of days hospitalized, these measures are not without problems. For example, the number of hospital admissions can be biased by differential censoring between groups, and the rate of hospital admissions is a meaningful measure only when it isrelatively constant over time.
Another potential advantage associated with analyses of multiple hospitalizations is an increase in statistical power. However, there are many alternative models for the analysis of recurrent events that differ in important ways and can be complicated to perform and interpret. Of course, if the gain in power is substantial then the additional complexity is worth the effort, but it is not clear how much gain to expect.
The authors raise the issue of the trade-off between analyses based on total hospitalization, an endpoint which is relatively easy to measure, and heart failure hospitalization, which is more specific but requires adjudication. This is analogous to the choice between evaluation of two cardiovascular drugs on the basis of total mortality or cardiovascular mortality. One possible compromise is to use information on all hospitalizations, but account for the level of certainty in the diagnosis of heart failure as part of the analysis.7
What are the practical implications of these findings for the conduct of clinical trials? Trials should be designed to capture reliable information that will permit an adequate analysis of hospitalization data. Patients with heart failure are often elderly with substantial co-morbidity and will tend to be hospitalized for multiple reasons. This will be reflected in the all-cause hospitalization data. The criteria used to define the primary cause of a hospitalization must be clearly defined. The case report forms should be developed to ensure adequate reporting. The duration of follow-up for each patient should be provided. This permits calculation of the proportion of patients with admissions, the mean number of admissions and the mean number of days spent in hospital. Thesesimple summary statistics are sensitive indices of total disease burden.
The standard first-event analysis should still play an important role, and should probably remain the primary statistical approach for determining the significance of the treatment effect. To avoidthe bias associated with censoring patients at the time of death, the analysis should be based on a composite of heart failure hospitalization and death. It seems reasonable to do supportive analyses based on models for recurrent endpoints. In time one of these models might become an accepted standard. Additional descriptive analyses of the total disease burden, such as those recommended by the authors, should also play a more prominent role than they have in the past. We agree that widespread adoption of these practices would provide clinically important information about the efficacy of a therapeutic intervention as well as increase the relevance and impact of data collected in epidemiological studies. The ability of a treatment to shorten total time spent in hospital may well be as important an indicator of its efficacy as its ability to prolong life.
References
- Krumholz H, Chen Y, Wang Y. Predictors of readmission among elderly survivors of admission with heart failure. Am Heart J. 2000;139:72–77[Web of Science][Medline]
- Heart National Lung, and Blood Institute. Congestive heart failure in the United States: a new epidemic. National Heart, Lung, and Blood Institute Data Fact Sheet, 1996.
- Yusuf S, Negassa A. Choice of clinical outcomes in randomised trails of heart failure therapies: Disease-specific or overall outcomes? Am Heart J. 2002;143:22–28[CrossRef][Web of Science][Medline]
- Dickstein K, Snapinn S. Prediction of mortality and morbidity following myocardial infarction. J Am Coll Cardiol. 2002;39:194
[Abstract/Free Full Text] - Anand I, Florea V, Fisher L. Surrogate endpoints in heart failure. J Am Coll Cardiol. 2002;39:1414–1421
[Abstract/Free Full Text] - Metcalfe C, Thompson S, Cowie M, et al. The use of hospital admission data as a measure of outcome in clinical studies of heart failure. Eur Heart J. 2003;23:105–11212
- Snapinn SM. Survival analysis with uncertain endpoints. Biometrics. 1998;54:209–218[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
Related articles in EHJ:
- The use of hospital admission data as a measure of outcome in clinical studies of heart failure
- C Metcalfe, S.G Thompson, M.R Cowie, and L.D Sharples
EHJ 2003 24: 105-112.[Abstract] [FREE Full Text]
This article has been cited by other articles:
|
|
 |
|
  R. Roccaforte, C. Demers, F. Baldassarre, K. K.Teo, and S. Yusuf Effectiveness of comprehensive disease management programmes in improving clinical outcomes in heart failure patients. A meta-analysis Eur J Heart Fail, December 1, 2005; 7(7): 1133 - 1144. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||