DOI: 10.1097/01.pcc.0000436127.70886.a1
,
,
PMID: 24389716
Issn Print: 1529-7535
Publication Date: 2014/01/01
Facebook
Twitter
LinkedIn
Email
 |
Checking for direct PDF access through Ovid | |
Excerpt
We absolutely agree with Spentzas (1) that the use of unadjusted readmission rates would be inappropriate as this approach could result in unfair punishment of centers caring for high-risk patients (i.e., a high readmission rate may reflect a high-risk population rather than poor ICU management or decision making). This is not dissimilar to the concerns with using unadjusted mortality rates for comparative reporting in quality assurance. As such, this was not the intended message of our article, and we believe some clarifying points with regard to our methodology would be helpful.
Although the parenthetical statement “(of the readmission rate)” was inserted into the quoted concluding statement within the letter by Spentzas (1), we did not present data showing variation in the readmission rate among involved centers. For the reasons stated above (in line with Spentzas’ opinion [1]), we did not believe a presentation of crude readmission rates across centers would be meaningful without appropriate adjustment. Currently, there is no appropriate adjustment tool for this particular outcome, and therefore, we simply described the crude readmission rate for all centers in combination rather than by center.
What we did present in our results was an estimation of the risk of readmission associated with each center. Of course, an unadjusted risk of readmission would similarly be less informative than an adjusted risk. Therefore, the analytic method used to estimate the center-specific risk (i.e., with mixed modeling as described in the Materials and Methods section) was specifically selected to account for center variation in individual and admission characteristics. As a result, although we did not specifically describe the prevalence of different diagnoses for each center as suggested by Spentzas (1), the variation in case-mix was implicitly included in our analysis and our results. In other words, the variation in risk associated with each center shown in Figure 3 in (2) reflects the remaining risk after adjusting for variation in patient characteristics. Although not a perfect solution, this approach has been used in other studies on quality improvement measures.
With respect to Spentzas’ comment (1) on the PRISM and PIM prediction scores, we also agree that use of these scores is inherently limited for adjustment purposes. Because prior adult literature had shown an association between Acute Physiology and Chronic Health Evaluation scores and the risk of readmission, we did screen these scores as candidate variables for our multivariable model. However, we found they did not perform, as well as other variables for predicting the risk of readmission. As noted in Supplementary Table 1 in (2), the PIM-2 probability was associated with the risk of readmission on the bivariate analysis. However, it lost its significance in a fully adjusted model and only appeared in 10% of the bootstrap models. Therefore, neither score was included in the final multivariable model as described in the article.
