In this current study, all patients were selected from Changhua Christian Hospital with a retrospective chart review conducted by the Department of Neurology. Participants were selected based on the confirmed diagnosis of first acute ischemic stroke and underwent intravenous thrombolysis therapy with recombinant tissue plasminogen activator (rtPA), followed by intra-arterial thrombectomy. Patients were selected based on the current indication for intravenous thrombolysis of presenting within 4.5 hours of the onset of symptoms1 and did not have any contraindications to receive rtPA. CT angiography perfusion scan was followed in these patients to screen for candidates to receive intra-arterial thrombectomy.
Additional inclusion criteria are neuroimage confirmed anterior circulation obstruction, above eighteen of age, and were followed for at least one year. Patients with intracerebral hemorrhage, aneurysm rupture, cerebral arteriovenous malformation, and recurrent stroke were excluded. A total of 92 patients was selected fulfilling the above criteria between 2015 and 2017.
Pre-intervention data include baseline patient demographics such as age, body mass index, systolic and diastolic blood pressure, total cholesterol, HbA1c, and other comorbidities were documented. Additionally, in-hospital complication such as pneumonia was also recorded.
As part of the protocol, all patients underwent carotid doppler examination for evaluation of the status of blood flow in large vessels including the common carotid artery (CCA), internal carotid artery (ICA), external carotid artery (ECA), and vertebral artery (VA), and the presence of artherosclerotic plaque. Multiple parameters were extrapolated, including peak systolic velocity (PSV), end-diastolic velocity (EDV), resistance index (RI), and pulsatility index (PI).
Primary outcome in this study was to identify any significant change in the functionality of patients who underwent treatment, which is done by assessment of NIHSS score2, mRS score3, and Barthel index4 both upon admission to the hospital and at 1-year follow up. Additionally, using statistical methods, described in the following section, to determine whether any of the pre-intervention variables have a significant impact on the functional outcome.
Data was collected from 92 stroke patients with a total of 68 independent pre-interventional indexes and 4 outcome variable (Appendix 1) were recorded. Authors use principal component analysis (PCA) to integrate the prognostic variables into an aggregative index, prognosis, which is an index that takes into account all four outcome variables and their contribution to the overall outcome. Then using the forward stepwise regression model to determine which, if any, of the sixty-eight independent variables had a significant effect on the prognosis. Authors end up with only eleven variables and only seven of which had statistically significance.
Classify the Variables
To facilitate the analysis, the data is first divided into pre-interventional and outcome variable, which are classified into categories, orders, and continuations. The outcome variables include the following: Follow up CT cerebral bleeding, MBD NIHSS, MBD mRS, MBD Barthel Index, and the remaining 68 variables are classified as pre-interventional independent variables.
Use Principal Component Analysis on the prognostic variables
The authors use principal component analysis (PCA) to integrate the outcome variables into an aggregative index, prognosis. According to equation (1), each principal component recombines the original variables into a new set of several independent variables. The coefficient of Xi in the linear combination is its eigenvector which is obtained by maximizing the explanatory variation of the corresponding principal component5. Therefore, authors can determine the relationship between the principal component and the original variable.
PCA = ϕ1X1 + ϕ 2X2 +...+ ϕ nXn (1)
In the current study, prognosis is an outcome variable that is influenced by the need of follow up head CT in case of suspected intracranial bleed, the NIHSS score, mRS score, and Barthel index. The relationship is shown in the below equation (2):
Prognosis = 0.218 Follow.up.CT + 0.566 NIHSS + 0.528mRS - 0.595B.I (2)
Exploratory Data Analysis
The exploratory data analysis section was performed with the ggplot2 package in R. Authors describe the statistics of variables in order to understand the simple information of the data.
(1) Multiple Regression Model
Multiple regression is to explore the correlation between the independent variable and the dependent variable and to build a regression model6. The equation (3) shown below.
y = b0 + b1x1 + b2x2 +...+bnxn (b0...bn: Regression coefficients) (3)
In our research , authors regarded all variables in pre-intervention as independent variables, and prognosis as a dependent variable, and built a multiple regression model. Authors find that only a few variables are significant, however, with Adj-R2 being only 0.3106, the model has a relatively low explanatory power. In order to improve the model, forward stepwise regression model is then used.
(2) Forward Stepwise Regression Model
The forward stepwise regression begins with an empty regression, adding variables one by one to select the best performing model according to the Akaike information criterion (AIC). AIC is a standard for assessing the complexity of statistical models and measuring the superiority of statistical model fit the data. The model with the lowest AIC value should be given priority when selecting the model. This is done by adding the independent variables one by one until the additional contribution of any one of the variables does not provide any statistical significance. Finally, the authors identified eleven variables with only seven of them are significant (table 3). This method improved the Adj-R2 of our model from 0.3106 to 0.4588, with p-value less than 0.05, and therefore, the model setup by using forward stepwise regression provides more explanatory power.
(3) Segmented linear regression
Segmented linear regression is a method when the independent variables clustered into different groups7. That is, there are different relationships between the variables in these regions. The following equation (4) is a threshold equation model:
η = α1 + α2Tz + β1 (x - e)+ + yx (4)
In this equation ‘e’ is the threshold parameter, and ‘x’ is the predictor with threshold effect, ‘z’ denotes additional predictors.
Authors estimate a segmented linear regression model to determine if there is a threshold for independent variable age, that would mark as a boundary for poor prognosis with statistical significance. The package authors utilized is chngpt package in R and uses the formula built in forward stepwise regression.