Supplementary Material for: Immunological Biomarkers Improve the Accuracy of Clinical Risk Models of Infection in the Acute Phase of Ischemic Stroke
datasetposted on 28.02.2013 by Salat D., Penalba A., García-Berrocoso T., Campos-Martorell M., Flores A., Pagola J., Bustamante A., Quintana M., Giralt D., Molina C.
Datasets usually provide raw data for analysis. This raw data often comes in spreadsheet form, but can be any collection of data, on which analysis can be performed.
Infection is an independent risk factor for adverse outcome in stroke patients. The risk of developing an infection in this setting is partly related to a stroke-induced immunodepression, in which a shift to a predominant Th2 (immunosuppressive) phenotype has been postulated to play a major role. Our aim was to study whether clinical variables or changes in plasma cytokine expression can predict poststroke infections. Patients and Methods: Medical records of 92 stroke patients were reviewed, and the baseline concentration of cytokines from the Th1/Th2 system was determined. Clinical and serological predictors of incident infections and their prognostic significance were sought by means of univariate and multivariate analysis, and two predictive models for developing an infection were constructed by combining independent predictors (strictly clinical in one, and both clinical and serological in the other) for this outcome. The improvement conferred by the addition of immunological markers to the clinical model was assessed by comparing their respective ROC curves and by improvement (Net Reclassification Index and Integrated Discriminator Improvement) analysis. Results: Nineteen patients (20.7% of the study sample) developed an infection. Ongoing antiplatelet therapy at symptom onset (OR 0.02, 95% CI 0.001-0.23, p = 0.001), diabetes mellitus (OR 9.96, 95% CI 1.32-75.29, p = 0.03), IL-13 level <33 pg/ml (OR 84.16, 95% CI 2.53-2795.18, p = 0.01) and interferon-γ level >8.4 pg/ml (OR 60.17, 95% CI 1.78-2037.23, p = 0.02) were independently associated with the development of infections during hospital admission. The combined regression model predicted infection with an accuracy of 93.4%, an improvement in the predictive capacity of 17% (p < 0.001). Infection was associated with a worse neurological status at hospital discharge (median NIHSS score 11 (6-18) vs. 4 (1-11.5), p = 0.014). Conclusions: This study shows that bloodstream biomarkers are useful to improve the accuracy of clinical prognostic models for infection in the acute phase of stroke. The clinical predictors of infection in the acute phase of stroke are relatively well established in the medical literature, but further research to identify the optimal combination of biomarkers (possibly inflammatory and stress markers) to be included in a clinically useful model is needed. Such a model could be subsequently used in clinical trials to assess the effect of prophylactic and/or early antibiotic therapy in this setting, a currently controversial issue in this field.