Predicting prognosis after stroke

Abstract

Patients with a severe neurologic deficit after acute ischemic stroke, as measured by the NIHSS, have a poor prognosis. During the first week after acute ischemic stroke, it is possible to identify a subset of patients who are highly likely to have a poor outcome. These findings require confirmation in a separate study.

Association of Serum Bilirubin with Ischemic Stroke Outcomes

“…All these aforementioned associations have been identified in prior work. [12][13][14][15] Our study has some limitations. This was a single center study, in which patients were not randomized, and we did not collect data on the exact timing of bilirubin levels or stroke severity assessment.…”

“…To evaluate the role of possible confounding factors, other potential determinants of incident stroke severity were also analyzed based on prior reports in the literature: age, sex, history of atrial fibrillation, history of hypertension, hyperlipidemia, diabetes, smoking status, admission glucose, premorbid antithrombotic use, premorbid statin use, and premorbid functional status. [12][13][14][15][16][17] Potential baseline covariate predictors of functional status at hospital discharge included all the aforementioned potential determinants of stroke severity and admission NIHSS score. 12 The pool of potential covariates (listed above) were then selected using backward elimination at P =.2.…”

“…[12][13][14][15][16][17] Potential baseline covariate predictors of functional status at hospital discharge included all the aforementioned potential determinants of stroke severity and admission NIHSS score. 12 The pool of potential covariates (listed above) were then selected using backward elimination at P =.2. Covariates were retained if any of them remained in the model.…”

Explicit inclusion of treatment in prognostic modeling was recommended in observational and randomized settings

“…Methods to account for treatments in the development of a prognostic model to be used for predicting the health course of individuals in the absence of treatment include simply ignoring treatment, 5 restricting the development set to untreated individuals, 6 censoring observations after treatment has started, 7 and explicit modeling of the treatment. 8 Also, in the TRIPOD statement, there is an item on the reporting of treatment received among participants of a study developing or validating a multivariable prediction model for diagnosis or prognosis.…”

“…5 If treatments are effective in reducing the risk of the predicted outcomes, simply ignoring those treatments in the development of a prognostic model may result in incorrect predictor-outcome associations and hence incorrect risk predictions of the natural history when used in new individuals. 6 Even though predictions are correct for those among whom the model was developed (the 'derivation set'), they may not generalize to future individuals who may be treated differently. In other words there is a danger of risk predictions being confounded by treatment: risk predictions appear low because of treatment, but in future patients the true risk might be substantially higher if they remain untreated.…”