Author + information
- Eliano Pio Navarese, MD⁎ (, )
- Fausto Castriota, MD and
- Stefano De Servi, MD
- ↵⁎Catholic University of Rome, Italy, Largo A. Gemelli, 8, 800168 Rome, Italy
We read with interest the article by Freeman et al. (1). The authors investigated the relationship between hospital implantable cardioverter-defibrillator (ICD) implantation volume and procedural complications.
They found that patients who have an ICD implanted at a high-volume hospital are less likely to have an adverse event associated with the procedure; patient data were collected from 1,201 different hospitals. An appropriate statistical method is critical when investigating the impact of procedure volume on clinical outcome. One-level hierarchical logistic regression to demonstrate the relationship between clinical outcome and procedures (quartiles of increased hospital annual ICD implantation volume) has been used as a statistical approach.
However, the choice of such an analysis may have provided biased results. Clustering of patient populations and hospital characteristics at individual hospitals influences regression analysis in ways that must be accounted for (2). In other words, when dealing with data from different institutions, patients in the same hospital are more likely to experience similar outcomes than patients treated in another hospital with the same volume because of differences in technique, skill, or supportive care, operators that statistically translate in observations within a center are correlated and those in different centers are independent. Therefore, during the analysis, it is necessary to account for clustering of outcomes that occur within a hospital.
A multilevel modeling (mixed-model) analysis applied to the logistic regression, 2-level in the present study, represents the proper choice; in this context, the hospital would be treated as a random effect, allowing for the differential influence of hospitals so that a variable (e.g., procedure volume) has an outcome at any given hospital.
Another issue of concern is the investigation of the relationship between volume of ICDs implanted and outcome. The authors investigated this using quartiles of ICD volumes. However, such an approach may produce several common mistakes (3). First, the use of quartiles may lose the greater explanatory power of treating volume as continuous outcome, affecting the results of regression (4). Second, the choice of this categorization leads to arbitrary classifications; conversely, treating volume as continuous outcome or using precise estimates derived from receiver-operating characteristic curves may prevent the statistical confusion provided by quartiles related to volume categories of varying sizes.
In conclusion, we believe that the statistical approach used in this study may have produced seriously flawed results due to the oversimplified view of the relationship between procedure volume and outcome.
- American College of Cardiology Foundation
- Freeman J.V.,
- Wang Y.,
- Curtis J.P.,
- Heidenreich P.A.,
- Hlatky M.A.
- Altman D.G.,
- Lausen B.,
- Sauerbrei W.,
- Schumacher M.