Longitudinal Data to Enhance Dynamic Stroke Risk Prediction

Figures

• 

  Figure 1. Flowchart of the observation timeline, where n indicates the number of individuals.

• 

  Figure 2. Receiver operating characteristic (ROC) curves for dynamic prediction, with incremental number of measurements. The x-axis is the false positive rate, calculated as (1 − specificity) and the y-axis is the true positive rate, calculated as sensitivity. Black dots indicate the optimal threshold of ROC curve, with sensitivity and specificity in the bracket. (a) ROC curve with one measurement (2008) in 2008; (b) ROC curve with two measurements (2008, 2011) in 2011; and (c) ROC curve with three measurements (2008, 2011, and 2014) in 2014.

• 

  Figure 3. ROC curve for patients who were not identified as having had a stroke in 2014. The x-axis is the false positive rate, calculated as (1 – specificity), and the y-axis is the true positive rate, calculated as sensitivity. (a) ROC curve with one measurement in 2014; (b) ROC curve with two measurements in 2011 and 2014; and (c) ROC curve with three measurements in 2008, 2011, and 2014.

• 

  Figure 4. Dynamic risk of stroke prediction for two representative cases. (a) Personalized dynamic prediction plot for Subject 1, who was identified as having had stroke between 2014 and 2017; (b) personalized dynamic prediction plot for Subject 2, who was not identified as having had a stroke until the end of study (2017). Follow-up years of 0, 3, 6, and 9 equate to 2008, 2011, 2014, and 2017, respectively. The left y-axis is the three-year risk of stroke. The right y-axes are values of longitudinal risk factors: sysBP—systolic blood pressure (mmHg); diaBP—diastolic blood pressure (mmHg); cho—total cholesterol (mmol/L); hdlc—high-density lipoprotein cholesterol (mmol/L); rbc—red blood cell count (1012/L); and plt—platelet count (109/L). Black dots indicate the three-year risk of stroke at each measurement. Connected plain black lines represent the risk change trend. The vertical interval of each point is the confidence interval calculated by repeated 10-fold cross-validation. The vertical red dotted line is the time point of being diagnosed with a stroke and the vertical green dotted line is the end of the observation time.

• 

  Figure 5. Comparison of the confidence interval using (a) bootstrap repetition and (b) repeated 10-fold cross-validation on Subject 1. Follow-up years of 0, 3, 6, and 9 equate to 2008, 2011, 2014, and 2017, respectively. The left y-axis is the three-year risk of stroke. The right y-axes are values of longitudinal risk factors: sysBP—systolic blood pressure (mmHg); diaBP—diastolic blood pressure (mmHg); cho—total cholesterol (mmol/L); hdlc—high-density lipoprotein cholesterol (mmol/L); rbc—red blood cell count (1012/L); and plt—platelet count (109/L). Black dots indicate the three-year risk of stroke at each measurement. Connected plain black lines represent the risk change trend. The vertical interval of each point is the confidence interval calculated by repeated 10-fold cross-validation. The vertical red dotted line is the time point of being identified as having had a stroke.

• 

  Figure 6. Stroke risk comparisons with seven models based on Subject 1, who was identified as having had a stroke between years 6 and 9. The solid black line and black dots are based on the model used in this paper. Other dotted lines with hollow shapes, in different colors denote the prediction results produced by other stroke risk calculators. Details can be found in the legend and in Table 4.