Imaging-based pregenetic screening for NOTCH3 p.R544C mutation in ischemic stroke in Taiwan

Study design and participants

The FSGC is a prospective, multicenter stroke registry that recruits stroke patients in Taiwan.⁹ The R544C screening score was developed from 1734 patients with ischemic stroke or transient ischemic attack (TIA) enrolled from the FSGC and validated by a second group of 235 patients enrolled from the stroke center of National Taiwan University Hospital. Subjects who presented with acute ischemic stroke or TIA within 10 days of onset were enrolled from the FSGC during April 2005 to July 2015 and screened for the NOTCH3 p.R544C mutation as described previously.⁵ TIA was defined as a transient focal neurologic deficit of ischemic origin that resolved within 24 hours.¹⁰ Acute ischemic stroke was defined as the acute onset of a neurological deficit with signs or symptoms persisting longer than 24 hours with or without acute ischemic lesion(s) on brain CT or with acute ischemic diffusion-weighted imaging lesion(s) on MRI that corresponded to the clinical presentations.¹⁰

Enrolled subjects received a structured questionnaire for clinical characteristics and blood sampling. The stroke subtypes were classified according to the Trial of Org 10172 in Acute Stroke Treatment (TOAST) criteria.¹¹ Among patients harboring the NOTCH3 p.R544C mutation (N = 41), 36 had brain MRI for the index ischemic event available and were included for imaging analysis. Among those subjects who did not harbor the NOTCH3 p.R544C mutation, the subgroup of subjects (n = 673) recruited from two medical centers, National Taiwan University Hospital and Shuang Ho Hospital, were selected for imaging analysis. The validation cohort was composed of 235 stroke patients enrolled from the stroke center of the National Taiwan University Hospital, using the same inclusion and exclusion criteria as that for the FSGC, but different recruiting periods from the derivation group.

Brain MRI acquisition and visual rating for imaging characteristics

Magnetic resonance imaging (MRI) of the brain was performed on a 1.5T MRI scanner. The scanning protocol included axial T2-weighted-fluid-attenuated inversion recovery (T2-FLAIR), fast spin-echo T2-weighted sequences, and a coronal T1-weighted image (T1WI) or a 3D T1WI. Imaging characteristics, including infarct size, infarct location, extent of white matter hyperintensity (WMH), white matter lesions involving the anterior temporal pole and external capsule, and the severity of mesial temporal atrophy (MTA), were evaluated with semiquantitative and qualitative methods. The infarct size was classified into three groups: less than 1.5 cm, 1.5 to 4.0 cm, or larger than 4.0 cm in the largest diameter. The infarct location was classified into eight areas: cortical, white matter or subcortical, basal ganglia, thalamus, brainstem, cerebellum, borderzone, and multiple infarctions. The extent of WMH in the periventricular white matter (PVWM) and deep white matter (DWM) were rated separately using the Fazekas score.¹² The WMH from the T2-FLAIR series of brain MR images was rated from 0 to 3 on an axial view, with a higher score indicating a more extensive WMH. Each side of the hemisphere was rated separately, and the average score of the two sides was used to represent the extent of WMH for each subject. White matter lesions involving the anterior temporal pole and the external capsule were coded as dichotomous variables. The severity of MTA was rated on a coronal view of the T1WI of the brain MRI scan using the scoring system proposed by Scheltens et al.¹³ Each side of the medial temporal lobe and the hippocampus was scored from 0 to 4. A higher MTA score indicated more advanced medial temporal lobe atrophy. The average score of the two hemispheres was used to represent the severity of MTA for each subject.

To calculate the interrater reliability of visual scoring, 30 MRI scans were rated independently by two experienced neurologists (SCT and YWC). Cohen’s kappa and the linearly weighted kappa were calculated for binominal and ordinal variables, respectively. There was strong interrater agreement for white matter lesions involving the anterior temporal pole (kappa = 0.805, 95% CI = 0.593-1.017), lesions involving the external capsule (kappa = 0.860, 95% CI = 0.707-1.013), and WMH of the DWM (weighted kappa = 0.825, 95% CI = 0.730-0.920). The interrater agreement for WMH of the PVWM was moderate (weighted kappa = 0.677, 95% CI = 0.539-0.816).

Statistical analysis

To compare the demographic and imaging characteristics between groups, the independent Student’s t-test or the Mann-Whitney U test was used for continuous variables, and the Chi-square or Fisher’s exact test was used for categorical variables. To identify key predictors for subjects harboring the NOTCH3 p.R544C mutation, a forward stepwise multivariable logistic regression model was applied to estimate the odds ratio (OR) and 95% confidence interval (CI). Clinical and imaging characteristics that significantly differed between the NOTCH3 p.R544C mutation carriers and noncarriers were selected as independent variables for the regression model. A P-value < 0.05 for entry and a P-value > 0.10 for removal from the model were set for the stepwise selection method. The discriminative ability of the prediction models was tested by calculating the area under the receiver operating characteristic (ROC) curve, and the calibration was calculated using the Hosmer-Lemeshow goodness-of-fit test. Based on the estimated odds ratio for each clinical and imaging characteristic, we developed a simplified R544C screening score to predict a positive NOTCH3 p.R544C mutation identification. For each scale, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Youden's J statistic were calculated to determine the optimal cutoff point to predict a positive NOTCH3 p.R544C mutation. The area under the ROC curve, PPV, and NPV were calculated in the validation cohort to test the performance of the R544C screening score. The clinical usefulness of the final prediction score was demonstrated using decision curve analysis.^{14, 15} Data were analyzed using the statistical software SPSS for Windows, version 25.0, and R version 4.0.2 (for decision curve analysis).

Standard protocol approvals, registrations, and patient consents

The study was approved by the local ethical committees of the participating hospitals. All investigations were conducted according to the principles expressed in the Declaration of Helsinki. All the participants or their proxies provided written informed consent before enrollment.

MRI characteristics between p.R544C mutation carriers and noncarriers

The comparison of imaging characteristics between the NOTCH3 p.R544C mutation carriers and noncarriers are shown in (Table 2). Subjects harboring the NOTCH3 p.R544C mutation more frequently had a stroke located in the white matter (55.6% vs 14.3%), had more extensive WMH in both the PVWM (Fazekas score 2.58 vs 1.53, P < 0.0005) and DWM (Fazekas score 2.25 vs 1.17, P < 0.0005), had more advanced mesial temporal atrophy (Scheltens sale 1.46 vs 1.08, P = 0.034), and more frequently had white matter lesions involving the external capsule (69.4% vs 14.3%, P < 0.0005) or the anterior temporal pole (22.2% vs 3.6%, P < 0.0005). The infarct size did not differ between the two groups.

The imaging characteristics in different age groups are shown in (Fig. 2). In addition to its association with the carrier status of the NOTCH3 p.R544C mutation, the extent of WMH increased with advanced age (Fig. 2A). Among the NOTCH3 p.R544C carriers, the Fazekas score in either the PVWM or DWM reached the maximum score (3.0) for most of the subjects aged more than 60 years. Additionally, the frequency of white matter lesions involving the anterior temporal pole increased with advanced age among the NOTCH3 p.R544C mutation carriers, whereas it remained less than 5% among the noncarriers (Fig. 2C). Regardless of age, the frequency of white matter lesions involving the external capsule was much higher among the NOTCH3 p.R544C carriers than the noncarriers (Fig. 2D). MTA was more advanced in the NOTCH3 p.R544C mutation carriers in all age groups (Fig. 2B). In addition, MTA was associated with advanced age in both the NOTCH3 p.R544C mutation carriers and noncarriers.

Development and performance of the NOTCH3 p.R544C mutation risk score

(Table 3) shows three models of multiple logistic regression analysis for the clinical and imaging characteristics associated with a positive NOTCH3 p.R544C mutation status. Model 1 was constructed using the stepwise logistic regression method. A younger age of stroke (OR = 0.92, 95% CI = 0.88-0.97), the SVO stroke subtype (OR = 3.01, 95% CI = 1.19-7.64), a family history of stroke or TIA in siblings (OR = 3.42, 95% CI = 1.07-10.95), white matter lesions involving the external capsule (OR = 4.18, 95% CI = 1.33-13.13), and a higher DWM Fazekas score (OR = 4.35, 95% CI = 2.14-8.86) were identified as key factors associated with the NOTCH3 p.R544C mutation status (Table 3, model 1). The interaction term between DWM Fazekas score and age did not reach statistical significance (P = 0.263), thus was not included in Model 1. To simplify the scoring system, we classified age of stroke into four strata and the DWM Fazekas score into two groups (Table 3, model 2). An age of stroke younger than 50 years was significantly associated with the NOTCH3 p.R544C mutation carrier status. The cutoff of the DWM Fazekas score was determined based on the distribution of the score in different age groups (Fig. 2A). The significant DWM Fazekas score for age was defined as an average Fazekas score> 1 for subjects at or younger than 60 years and an average Fazekas score > 2 for subjects older than 60 years. Model 3 was simplified from model 2 and used only five dichotomous variables to predict the p.R544C mutation status. The performance of model 3, measured by the area under the ROC curve, was comparable to that of models 1 and 2 (Fig 3A,C). The areas under the ROC curve for models 1, 2, and 3 were 0.920, 0.902, and 0.900, respectively.

Based on the estimated odds ratio in model 3, we developed the R544C screening score. The scoring items are shown in (Table 4). The sum of the scores ranged 0 to 8, with a higher score indicating a greater likelihood of a positive NOTCH3 p.R544C mutation test result. The ROC curve for the R544C screening score is shown in (Fig. 3). The area under the ROC curve was 0.867 (95% CI = 0.840-0.892, P < 0.01), indicating good performance. To be applied as a screening tool that could lead to further genetic testing, adequate sensitivity is required. On the other hand, adequate specificity is also important for the screening score to be applied efficiently in the clinical setting. A cutoff score of 5 was chosen to maximize Youden's J statistic, and the associated sensitivity, specificity, PPV, NPV, and accuracy were 0.750, 0.880, 0.131, 0.993, and 0.877, respectively (Table 5). The performance parameters of individual variables in the R544C screening score are summarized in supplementary table (Table S1). Although the excessive DWM hyperintensity for the age alone is associated with a sensitivity comparable to the full R544C screening score, the addition of the other four clinical parameters will increase the specificity to predict a positive p.R544C mutation test result from 0.823 to 0.880.

The demographic data and imaging characteristics of the validation cohort are shown in (Table S2). Six of the 235 subjects in the validation cohort harbored the NOTCH3 p.R544C mutation. The ROC curve of the R544C screening score in the validation cohort is shown in (Fig. 3B). The area under the ROC curve was 0.957 (95% CI = 0.916-0.997). Using a cutoff score of 5, the sensitivity, specificity, PPV, NPV, and accuracy were 0.833, 0.895, 0.161, 0.996, and 0.894, respectively (Table 5).

The clinical usefulness of the R544C screening score was further shown by decision curve analysis (Figure 4). The decision curve analysis showed the net benefit of using the screening score to guide the NOTCH3 p.R544C genetic testing over a range of risk threshold for genetic testing. Due to minimal risk and relatively low cost associated with genetic testing for a single-point mutation, we considered the clinically reasonable risk threshold to be less than 0.2 (i.e., number-needed-to-test more than 5). Over the relevant range of risk threshold, the R544C screening score provides a larger net benefit compared with stratifying patients by advanced DWM hyperintensity, external capsule involvement, or SVO stroke subtype alone.