DOI: 10.1097/01.CCM.0000199041.26426.1E
,
,
PMID: 16424770
Issn Print: 0090-3493
Publication Date: 2006/02/01
Excerpt
I read with great interest the article by Dr. Uchino and colleagues (1) comparing different scoring systems to predict mortality in acute renal failure. All scoring systems had areas under the receiver operating characteristic curves <0.7 and performed substantially worse than when described initially by their authors. Dr. Uchino and colleagues discussed several factors that they thought contributed to the poor performance of the scoring systems. However, the major limitation in developing a worldwide predictive model is the limited data set employed. Constructing models from demographics, vital signs, comorbidities, and laboratory values ignores processes of care. Differences in processes of care ranging from nurse/patient ratios, nursing education, and physician staffing and availability, to hemodynamic, transfusion, and ventilator management may explain most of the differences in outcome between good and bad hospitals and swamp the information provided by demographics, vital signs, comorbidities, and laboratory values. Models created in one center or even from one geographical, economic, and health care area like North America would be based on patients with very similar processes of care, which would act like a mathematical constant in a logistic regression and not be needed for an accurate predictive model. Including patients from very diverse hospitals and countries, as the authors did (and should be congratulated for), begs the need for including processes of care in the predictive models. Including these in the scoring system should improve their predictive performance. Importantly, it would also allow identification of the processes of care that contribute to mortality or survival. I urge the authors to analyze their database to include processes of care, even if only simple ones like economic status of the country and hospital-wide nurse/patient ratio are available.
The author has no financial interests to disclose.
Facebook
Twitter
LinkedIn
Email
