Number needed to screen for acute revascularization trials in stroke: Prognostic and predictive imaging biomarkers

Abstract

Objective

To systematically assess imaging biomarkers on CT-based multimodal imaging for their being predictive versus prognostic biomarkers for intravenous and endovascular (IA) revascularization therapy, and for their prevalence.

Methods

Our retrospective study included patients suspected of acute ischemic stroke with admission work-up including a non-contrast head CT, perfusion CT, and CT angiography. Modified Rankin scores at 90 days were used as outcomes. For each imaging biomarker, the effect size of the test of interaction between the presence of the biomarker and the treatment effect was calculated, allowing the inference of a total sample size. The total sample size required was combined with the prevalence of the biomarker to determine the number needed to screen.

Results

In the 0–4.5-h time window, the two predictive biomarkers associated with the smallest number needed to screen were perfusion CT penumbra ≥ 20% (404 NNS) and CT angiography collateral score ≥ 2 (581 NNS). In the 3–9-h time window, the four predictive biomarkers associated with the smallest number needed to screen were clot burden score (CBS) on CT angiography (1181 NNS), clot length ≥ 10 mm (1924 NNS), CBS and clot length ≥ 10 mm (1132 NNS), and CBS and perfusion CT penumbra ≥ 100% (1374 NNS). Perfusion CT ischemic core was a prognostic biomarker in both time windows.

Interpretation

Predictive biomarkers need to be differentiated from prognostic biomarkers when being considered to select patients for a trial, and their prevalence should be assessed to determine the number needed to screen and overall feasibility of the trials.

Keywords

Stroke imaging tissue-type plasminogen activator endovascular treatment biomarker revascularization

Introduction

The distinction between prognostic and predictive imaging biomarkers is very relevant to the design of clinical trials of acute revascularization therapy in stroke. A prognostic biomarker is associated with a specific clinical outcome, whether the patient is treated or not.¹ A predictive biomarker associated with a differential treatment effect, i.e. a more favorable clinical outcome exclusively in the treatment group¹ (Figure 1). In order to be successful, a clinical trial of acute revascularization therapy needs to use a predictive, not prognostic, biomarker in order to identify patients more likely to respond favorably to the tested therapy.² Compared to a trial that does not use any selection biomarker, a clinical trial that uses a predictive biomarker should allow to magnify the effect size, and therefore reduce the sample size required to demonstrate a positive effect of the tested therapy on clinical outcome. This is, however, only true if the prevalence of the predictive biomarker is high enough that the enrollment rate into the trial is not hampered, and that the duration of trial is not unnecessarily increased because of the need to screen a very large number of stroke patients to find only a few positive for the predictive biomarker.

Figure 1.

Plots of outcome versus present/absent biomarker in treated and untreated patients, for an ideal prognostic biomarker and an ideal predictive biomarker. An ideal prognostic biomarker is associated with a favorable clinical outcome, whether the patient is treated or not. An ideal predictive biomarker is associated with a differential treatment effect, i.e. a more favorable clinical outcome exclusively in the treatment group.

In the present study, we systematically assessed all currently investigated imaging biomarkers for their being predictive versus prognostic biomarkers for revascularization treatment in acute stroke. We also determined the theoretical sample size required for a clinical trial using such biomarker to be powered appropriately considering the effect size of the biomarker, and the number of patients needed to screen considering the prevalence of the biomarker. Because an imaging biomarker may behave differently depending on the type of therapy considered, we tested the biomarkers separately for intravenous (IV) and endovascular (IA) therapy.

Methods

Study population

Our retrospective study involves a repository of clinical and imaging data in acute stroke patients treated at two institutions between January 2008 and June 2013: Center Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland; and the University of Pittsburgh Medical Center, Pittsburg, PA, where similar imaging protocols were used to work up acute stroke patients. All data collection and analysis were approved by the institutional review boards of each institution involved.

All consecutive patients in this registry who met the following criteria were included in our study:

Suspicion of acute ischemic stroke with admission work-up including a non-contrast head CT (NCT), perfusion CT (PCT), and CT angiography (CTA);

No subacute or chronic/remote infarct seen on baseline NCT;

No hemorrhage on baseline NCT;

National Institutes of Health Stroke Scale (NIHSS) score measured on admission and modified Rankin score (mRS) measured at 90 days after initial event.

Patients were grouped based on the time window they were admitted in:

#10–4.5-h time window: patients in this time window who were eligible based on their NCT ASPECT score with a cut-off of 7 received IV recombined tissue plasminogen activator (r-tPA) (treatment – ASPECT score 7–10; controls – ASPECT score<0–6);

#2. 3–9-h time window: patients in time window who were eligible based on their NCT ASPECT score with a cut-off of 7 and the presence of an M1 clot on CTA received endovascular treatment (treatment – ASPECT score 7–10 and M1 clot present on CTA; controls – ASPECT score<0–6 or no M1 clot present on CTA).

Clinical data

Age, sex, time from symptom onset to baseline imaging, time from symptom onset to successful treatment (for IA), NIHSS score on admission, mRS at 90 days after initial event were recorded.

Imaging biomarkers

Alberta Stroke Program Early CT (ASPECT) score³ and presence or absence of a dense artery sign⁴ were assessed on the baseline NCT.

Clot involvement of the terminal internal carotid artery and M1 segment of the middle cerebral artery, clot length,⁵ relative clot density,⁶ clot burden score,⁷ and modified M1-Boston Acute Stroke (M1-BASIS) score⁸ were assessed on the baseline CTA. For characterizing the CTA Collateral Score circulation on baseline CTA images, we used the scoring system described by Tan et al.⁹: a score of 0 represents complete absence of CTA Collateral Score flow, and scores of 1–4 represent the ≤50%, >50% but <100%, =100%, and >100% (compared with the normal side) filling of the territory of the occluded artery, respectively.

Baseline brain ischemic lesion size was measured automatically on baseline PCT, using the deconvolution-based Philips Brain Perfusion software (version 4.5.2, Cleveland, OH, USA). The volumes of ischemic core and ischemic penumbra thresholds were calculated using previously validated thresholds: MTT > 145% of the contralateral side values and CBV < 2.0 mL/ g for ischemic core; MTT > 145% of the contralateral side values and CBV > 2.0 mL/ g for ischemic penumbra.¹⁰ The volumes of ischemic core and penumbra were normalized to the total volume of the MCA territory to correct for the variable coverage of the PCT scans.¹¹ Different subgroups of patients were evaluated based on different cut-offs for the volume of ischemic core (more or less than 50 mL, more or less than 70 mL, more or less than 100 mL) and the volume of ischemic penumbra (more or less than 20%, more or less than 50%, more or less than 100%).

All the image interpretations were done by two neuroradiologists (one with 15 years of experience, one with 5 years of experience) in consensus in a blinded fashion with respect to the treatment groups and clinical outcome.

Biomarker analysis

The biomarkers used clinically to make treatment decisions (ASPECT score in 0–4.5-h time window – Group #1, ASPECT score and presence of an M1 clot on CTA in the 3–9-hour time window – Group #2) were not considered in the biomarker analysis as, only those patients with positive values for these biomarkers were treated (IV for Group #1, IA for Group #2).

On the other hand, the other biomarkers evaluated, which were not used to make treatment decision, were determined for each patient as being positive or negative (Supplemental Online Table I), and the likelihood of a positive clinical outcome (mRS of 0–2 at 90 days) was plotted for “positive” and “negative” values of the biomarker considered, and this both in the treatment group (IV for Group #1, IA for Group #2) and in the untreated group, in the corresponding time windows. These plots were compared to the typical plot appearance of predictive and prognostic biomarkers (Figure 1).

Statistical analysis

Summary statistics

Continuous scaled data (age, time from symptom onset to baseline imaging, time from symptom onset to successful treatment, baseline infarct volume, etc.) were summarized by the median and interquartile range (IQR) of the distribution, and the two sample non-parametric Wilcoxon Rank Sum test was utilized to compare continuous scaled variables between the treated (T = 1) and the untreated patients (T = 0). Frequency data (e.g. perfusion PCT Penumbra on baseline PCT, clot burden score, CTA Collateral score, etc.) were summarized by counts or percentages, and the Fisher’s exact test was utilized to compare frequencies between the treated (T = 1) and the untreated patients (T = 0).

Prevalence calculation

The proportion of patients in our data registry who were positive for the biomarker (B + ) was utilized as the estimator for the prevalence of the biomarker (B+) in the target population.

Power calculation

Using the mRS category data (i.e. mRS ≤ 2 versus mRS > 2) in the data registry, as well as the biomarker status (B−, B+) and treatment status (T = 0, T = 1) information in the data registry, for each biomarker, the effect size of the test of interaction between biomarker status (B−, B+) and the treatment (i.e. T = 0 versus T = 1) was estimated via multivariate logistic regression model (MLRM) adjusting for age, NIHSS score, and ASPECTS (as well as presence of M1 clot for Group #2). The effect size indicates the strength of the interaction between a predictive biomarker and a treatment. Based on the effect size associated with interaction term of the MLRM, Monte Carlo Simulation (MCS) was then used to estimate the total sample size that would be required in order to have at least 0.80 statistical power to reject the null hypothesis that there is no differential treatment effect between patients who are positive for the biomarker (B+) and patients who are negative for the biomarker (B−) for a study design in which the total sample size is equally divided between the two treatment arms (i.e. T = 1 and T = 0) and an equal number of biomarker positive (B+) and biomarker negative patients (B−) are assigned to each treatment arm. The required total sample size (N) was derived at by using an MCS search method, in which the total sample size specified in the MCS was systematically increased until the total sample size was sufficiently large enough to meet the threshold of having at least 0.80 statistical power to detect the empirical derived differential treatment-effect between biomarker positive (B+) patients and biomarker negative (B−) patients. Details regarding how the MCS power analyses were conducted are provided in the accompanying Supplemental Materials under Supplemental Methods.

Projected screening requirement

Based on the estimate for the total sample size (N), and the based on empirical estimate for the prevalence of the biomarker in the target population, we estimated an upper bound for the total number of patients that would have to be screened (N_screened) in order to fulfill the total sample size requirement for a study design in which the total sample (N) is divided equally between the two treatment-arms (i.e. T = 1, T = 0) and an equal number of biomarker positive (B+) and biomarker negative (B−) patients are assigned to each treatment-arm. Details regarding how the number needed to screen was calculated are provided in the accompanying Supplemental Materials under Supplemental Methods.

Results

Group #1 included 162 patients, 113 of which received IV rt-PA and 49 of which did not. Group #2 included 173 patients with M1 occlusion, 107 of which received endovascular therapy and 66 of which did not. Patients who received IV rt-PA followed by endovascular treatment were excluded from both study groups.

Group #1 – 0–4.5-h time window: Patients in this time window who were eligible based on their NCT ASPECT score received IV r-tPA

The different imaging biomarkers evaluated in Group #1 (patients in the 0–4.5-h time window) (Table 1) differed in terms of being predictive (e.g. PCT penumbra 20%) versus prognostic (e.g. PCT ischemic core 50 mL) (Supplemental Online Figure I). Based on the effect size of the interaction between biomarker presence and IV r-tPA treatment, we calculated a required sample size for the trial designed to show this biomarker as efficient to select acute stroke patients for IV r-tPA treatment. We took into account the biomarker prevalence to determine how many patients should be screened in order to find the patients with the appropriate biomarker and achieve the required sample size (Table 2). Some combinations of biomarkers required a smaller sample size, but the screening size was large because of the rarity of finding these biomarkers in combination in stroke patients. The two biomarkers that were associated with the smaller number needed to screen were the PCT penumbra ≥ 20% (404 needed to screen) and the CTA collateral score ≥ 2 (581 needed to screen).

Table 1.

Demographics and imaging biomarkers in Group #1 (patients in the 0–4.5-h time window)

	IV r-tPA (N = 119)	No IV r-tPA (N = 38)	p
Age (y) [IQR]	.68.0 [60.0, 76.0]	65.0 [53.0, 75.0]	0.343
Gender (F)	56 (47%)	24 (63%)	0.096
Time from onset to baseline imaging (h) [IQR]	2.3 [1.8, 3.0]	3.0 [2.0, 3.5]	0.041
NIHSS upon admission [IQR]	16.0 [11.0, 19.0]	16.0 [6.0, 18.0]	0.190
mRS on 90 day [IQR]	2.0 [1.0, 4.0]	2.0 [1.0, 4.0]	0.532
NCT ASPECT score [IQR]	8.0 [7.0, 9.0]	5.0 [4.0, 6.0]	0.082
NCT dense artery sign (yes)	71 (60%)	15 (39%)	0.039
CTA M1-BASIS major	90 (76%)	28 (74%)	0.831
CTA M1-BASIS minor	29 (24%)	10 (26%)	0.831
CTA visible clot (yes)	102 (86%)	30 (79%)	0.319
CTA ICA involvement (yes)	91 (76%)	25 (66%)	0.227
CTA M1 involvement (yes)	87 (73%)	24 (63%)	0.306
CTA clot burden score (CBS) [IQR]	7.0 [5.0, 9.0]	6.5 [5.0, 9.0]	0.942
PCT ischemic core volume (mL)	100.7 [27.5, 168.5]	68.1 [29.0, 137.1]	0.145
PCT ischemic penumbra (%) [IQR]	0.4 [−0.4, 3.6]	0.9 [−0.0, 2.7]	0.348
CTA collateral score [IQR]	2.0 [1.0, 2.0]	1.5 [1.0, 2.0]	0.906

Table 2.

Sample size and number needed to screen for imaging biomarkers considered in isolation and in combinations in in Group #1 (patients in the 0–4.5-h time window)

Biomarker/combination of biomarkers	Effect size test of interaction n	Total sample size	Prevalence of Positive biomarker (in %)	Number needed to Screen point estimate	Number needed to screen upper 95% cl
PCT-penumbra≥20%	4.11	320	52.47	337	404
CTA-collateral score≥2	3.31	440	54.32	482	581
CTA-collateral score≥2+NCT-dense artery sign	4.52	280	25.93	540	723
PCT-ischemic core<100 mL+CTA-collateral score≥2+NCT-dense artery sign	5.27	220	18.52	594	856
PCT-penumbra≥20%+NCT-dense artery sign	4.61	260	20.37	638	899
PCT-penumbra≥20%+CTA-collateral score≥2+NCT-dense artery sign	6.30	180	14.20	634	976
Clot burden score (CBS)>6	2.36	840	50.62	851	1013
PCT-ischemic core<100mL+ CTA-collateral score≥2	2.78	640	36.42	879	1103
PCT-penumbra>=50%+CTA-collateral score≥2+NCT-dense artery sign	5.70	200	13.58	736	1148
PCT-penumbra≥100%+CTA-collateral score>=2	3.03	480	25.93	926	1239
PCT-penumbra≥20%+CTA-collateral score≥2	2.66	700	33.33	1050	1339
PCT-ischemic core<100 mL+NCT-dense artery sign	2.77	640	28.40	1127	1482
NCT-dense artery sign	2.07	1160	45.06	1287	1557
PCT-penumbra≥50%+NCT-dense artery sign	3.11	480	19.14	1254	1793
PCT-penumbra> =100%+PCT-ischemic core<100 mL+CTA-collateral score	2.68	680	24.69	1377	1862
PCT-penumbra≥20%+PCT-ischemic core<100 mL+CTA-collateral score>=2+NCT-dense artery sign	4.74	260	10.49	1239	2086
PCT-penumbra≥50%+	4.74	260	10.49	1239	2086
PCT-ischemic core<100 mL+CTA-collateral score≥2+NCT-dense artery sign
PCT-penumbra≥100%+CTA-collateral score>=2+NCT-dense artery sign	4.74	260	10.49	1239	2086
PCT-penumbra≥20%+PCT-ischemic core<100 mL+NCT-dense artery sign	3.28	440	14.81	1485	2261
PCT-penumbra≥50%+PCT-ischemic core<100 mL+NCT-dense artery sign	3.28	440	14.81	1485	2261
PCT-penumbra≥50%	1.82	1800	46.91	1918	2305
PCT-penumbra≥50%+CTA-collateral score≥2	2.05	1280	30.25	2116	2748
PCT-penumbra≥20%+PCT-ischemic core<100 mL+CTA-collateral score≥2	2.03	1280	27.16	2356	3125
PCT-penumbra≥50%+PCT-ischemic core<100 mL+CTA-collateral score≥2	2.03	1280	27.16	2356	3125
PCT-penumbra≥100%+	3.30	400	9.26	2160	3791
PCT-ischemic core<100 mL+CTA-collateral score≥2+NCT-dense artery sign
PCT-ischemic core<100 mL+CTA-collateral score+M1	2.33	880	16.67	2640	3901
PCT-ischemic core<70 mL	1.54	3400	44.44	3825	4639
CTA-collateral score≥2+M1	1.74	2000	19.75	5062	7183
Visible clot on CTA	1.97	1640	16.67	4920	7269
PCT-penumbra≥100%+NCT-dense artery sign	1.78	1720	14.81	5805	8840
PCT-penumbra≥20%+PCT-ischemic core 100 mL	1.38	6000	39.51	7594	9397
PCT-ischemic core<50 mL	1.43	5600	36.42	7688	9651
PCT-penumbra≥50%+PCT-ischemic core<100 mL	1.38	6400	39.51	8100	10,023
PCT-penumbra≥100%+PCT-ischemic core<100 mL +NCT-dense artery sign	1.82	1680	12.96	6480	10,236
CTA-M1-BASIS classification(minor)	1.50	4200	24.69	8505	11,499
PCT-penumbra≥20%+	1.71	2000	11.73	8526	13,868
CTA-Clot in M1
PCT-ischemic core<100 mL	1.25	12400	54.32	13,573	16,384
PCT-penumbra≥100%+CTA-collateral score>=2+CTA-clot in M1	1.92	1320	6.17	10,692	22,004
PCT-penumbra>=100%+PCT-ischemic core<100 mL+CTA-collateral score≥2+CTA-clot in M1	1.92	1320	6.17	10,692	22,004
PCT-penumbra≥100%+PCT-Ischemic Core<100 mL	1.20	18,800	35.80	26,255	33,059
PCT-ischemic core<100 mL+CTA-clot in M1	1.23	16,000	23.46	34,105	46,605
PCT-penumbra≥20%+PCT-ischemic core<100 mL+CTA-clot in M1	1.24	14,000	9.88	70,875	121,710
PCT-penumbra≥50%+PCT-ischemic core<100 mL+CTA-clot in M1	1.24	14,000	9.88	70,875	121,710
PCT-Penumbra≥100%+CTA-clot in M1	1.16	26,000	8.64	150,429	270,532
PCT-penumbra>=100%+	1.16	26,000	8.64	150,429	270,532
PCT-ischemic core<100 mL+CTA-clot in M1
PCT-penumbra≥100%	1.01	200,000	38.27	261,290	325,135
CTA-clot in M1	1.07	Too high to report	29.63	Too high to report	Too high to report
PCT-penumbra≥50%+CTA-clot in M1	1.04	Too high to report	11.11	Too high to report	Too high to report
PCT-penumbra≥20%+CTA-collateral score≥2+CTA-clot in M1	1.02	Too high to report	8.64	Too high to report	Too high to report
PCT-penumbra ≥50%+CTA-collateral score≥2+CTA-clot in M1	1.02	Too high to report	8.64	Too high to report	Too high to report
PCT-penumbra≥20%+PCT-ischemic core<100 mL+CTA-collateral score≥2+CTA-clot in M1	1.07	Too highto report	7.41	Too high toreport	Too high to report

PCT: perfusion CT; CT angiography; NCT: non-contrast head CT.

Group #2 – 3–9-h time window: Patient in time window who were eligible based on their NCT ASPECT score and the presence of an M1 clot on CTA received endovascular treatment

The different imaging biomarkers evaluated in Group #2 (patients in the 3–9-h time window) (Table 3) differed in terms of being predictive (e.g. clot burden score) versus prognostic (e.g. PCT ischemic core 100 mL) (Supplemental Online Figure II). Based on the effect size of the interaction between biomarker presence and endovascular treatment, we calculated a required sample size for the trial designed to show this biomarker as efficient to select acute stroke patients for endovascular treatment. We took into account the biomarker prevalence to determine how many patients should be screened in order to find the patients with the appropriate biomarker and achieve the required sample size (Table 4). Some combinations of biomarkers required a smaller sample size, but the screening size was large because of the rarity of finding these biomarkers in combination in stroke patients. The four biomarkers that were associated with the smaller number needed to screen were the clot burden score (CBS) on CTA (1181 needed to screen), the clot length ≥ 10 mm (1924 needed to screen), the CBS and clot length ≥ 10 mm (1132 needed to screen), and the CBS and PCT penumbra ≥ 100% (1374 needed to screen).

Table 3.

Demographics and imaging biomarkers in Group #2 (patients in the 3–9-hour time window)

	Endovascular treatment (N = 37)	No endovascular treatment (N = 36)	p
Age (y) [IQR]	66.0 [51.0, 75.0]	63.0 [53.0, 76.0]	0.982
Gender (F)	19 (51%)	20 (55%)	0.644
Time from onset to baseline imaging (h) [IQR]	4.5 [2.3, 6.0]	3.5 [2.0, 5.8]	0.264
Time from onset to treatment (h) [IQR]	6.0 [4.6, 8.3]	–	–
NIHSS on admission (h) [IQR]	15.0 [11.0, 20.0]	17.0 [13.5, 21.0]	0.177
mRS on 90 day [IQR]	2.0 [1.0, 4.0]	4.0 [2.0, 4.0]	0.067
NCT ASPECT score [IQR]	8.0 [7.0, 9.0]	5.5 [4.0, 6.0]	<0.001
NCT dense artery sign (yes)	26 (70%)	25 (69%)	1.000
CTA ICA involvement (yes)	7 (19%)	17 (47%)	0.013
CTA clot length (cm) [IQR]	1.4 [0.8, 2.1]	1.8 [0.7, 2.5]	0.323
CTA relative clot density [IQR]	1.3 [1.2, 1.4]	1.2 [1.1, 1.3]	0.053
CTA clot burden score (CBS) [IQR]	6.0 [6.0, 8.0]	5.5 [3.0, 6.0]	0.001
PCT ischemic core volume (mL) [IQR]	79.6 [46.7, 136.4]	103.0 [70.5, 191.9]	0.103
PCT ischemic penumbra (%) [IQR]	1.3 [−0.0, 3.8]	0.6 [−0.3, 1.9]	0.136
CTA collateral score [IQR] number needed to screen point estimate	2.0 [1.0, 3.0] number needed to screen upper 95% Cl	1.0 [1.0, 2.0]	0.011

PCT: perfusion CT; CT angiography; NCT: non-contrast head CT.

Table 4.

Sample size and number needed to screen for imaging biomarkers considered in isolation and in combinations in in Group #2 (patients in the 3–9-h window)

Biomarker/combination of biomarkers	Effect size test of interaction	Total sample size	Prevalence of positive biomarker (in%)	Number needed to Screen point estimate	Number needed to screen upper 95% CL
CTA-clot burden score (CBS)>6+CTA-clot length>=10 mm	5.95	240	18.67	643	1132
CTA-clot burden score (CBS)>6	2.84	540	33.3	810	1181
PCT-penumbra≥100%+CTA-clot burden score(CBS)>6	3.82	320	20.0	800	1374
CTA-clot length ≥10 mm	2.41	880	33.3	1320	1924
PCT-Penumbra ≥100%+PCT-ischemic core <70 mL	1.90	1400	32.0	2188	3227
PCT-ischemic core <70 mL	1.83	1600	34.7	2308	3327
CTA-clot length ≥8 mm	1.97	1520	32.0	2375	3503
PCT-penumbra ≥100%	1.26	12,000	46.7	12,857	17,118
CTA-clot length ≥20 mm	1.17	24,000	58.7	29,032	39,900
PCT-penumbra ≥100%+CTA-clot length >=10 mm	2101.72	8600	17.33	24,808	44,955
PCT-ischemic core <70 mL+CTA-clot length ≥10 mm	4553.72	7460	12.0	31,083	66,174
PCT-penumbra >=100%+PCT-ischemic core <70 mL +CTA-clot length ≥10 mm	4553.72	7460	12.0	31,083	66,174
PCT-penumbra ≥100%+CTA-clot burden score(CBS)>6 + CTA-clot length≥10 mm	4553.72	7460	12.0	31,083	66,174
PCT-ischemic core <70 mL+CTA-clot burden score(CBS)>6	1.20	16,800	16.0	52,500	98,243
PCT-penumbra ≥100%+	1.20	16,800	16.0	52,500	98,243
PCT-ischemic core <70 mL+CTA-Clot Burden Score(CBS)>6
PCT-Ischemic Core <70 mL+CTA-clot burden score (CBS)>6 + CTA-clot length>=10 mm	2032.68	8740	9.33	46,821	113,939
PCT-penumbra ≥100%+PCT-ischemic core <70 mL+CTA-clot burden score(CBS)>6 + CTA-clot length>=10mm	2032.68	8740	9.33	46,821	113,939
PCT-penumbra ≥20%	1.00	1,000,000	62.7	Too high to report	Too high to report
NCT-dense artery sign	0.92^a
CTA-clot in ICA	0.74^a
PCT-ischemic core <50 mL	0.74^a
PCT-penumbra ≥50%	0.63^a
CTA-relative clot density >1.3	0.57^a
PCT-ischemic core <100 mL	0.42^a
CTA-collateral score ≥2	0.31^a

Test of interaction less than 1 indicates wrong interaction direction, so discarded.

PCT: perfusion CT; CT angiography; NCT: non-contrast head CT.

We obtained similar relative ranking of the biomarkers when we repeated our analyses using an mRS 0–1 outcome.

Discussion

The distinction between prognostic and predictive biomarkers is critical to the design of clinical trials, not only for cancer, but also for stroke.¹ Patient selection using a biomarker for a clinical trial can be successful only if the considered biomarker is predictive of the effect of the treatment tested in the trial, but will be less successful if the biomarker is prognostic of the clinical outcome independently of whether the patients are in the treatment or control groups. In our systematic review of the imaging biomarkers studied in the field of stroke, we found that some are prognostic, others predictive and a lot of them in a mixed or intermediary situation. PCT ischemic core is typically a prognostic biomarker in the sense that a large ischemic core is associated with an unfavorable outcome, and this independently of whether the patient receives revascularization therapy or not. Others, such as the PCT ischemic penumbra, are predictive biomarkers, because, as demonstrated previously,¹² a large penumbra is only associated with a favorable outcome if there is early recanalization and this penumbra is salvaged. If there is persistent occlusion or late recanalization, the large penumbra becomes a large infarct associated with an unfavorable outcome. In MR RESCUE, the rate of recanalization was relatively low,¹³ both in the treatment and control groups, perhaps explaining why penumbra did not show any significant treatment differential effect. Instead, the penumbra appeared in this setting, stripped of its predictive component, as a prognostic biomarker likely linked to the amount of collaterals, negatively related to the final infarct size and associated with a favorable outcome independently of treatment.

When using a predictive biomarker to enrich a clinical trial population with patients more likely to respond favorably to the tested treatment, one increases the likelihood of success of the clinical trial. At the same time, the generalizability of the tested treatment decreases as only those patients with the imaging biomarkers are eligible to be enrolled in the trial. The prevalence of the biomarker needs to be considered because it will affect the number of patients needed to be screened to find those with the biomarker present who will be eligible for the trial. If the biomarker has an excellent predictive value but is extremely rare, the clinical trial, while likely to be successful, may not be feasible because of the too large number of patients needed to screen in order to enroll the few who will demonstrate the superiority of treatment over control. The biomarkers we found to be predictive of treatment effect and sufficiently prevalent to be associated with “feasible” number needed to screen were the PCT penumbra and the CTA collateral score for IV thrombolysis and the clot burden score, the clot length, and the PCT penumbra for endovascular revascularization. The sample size calculation yielded by our analysis was consistent with the sample sizes of trials that used the different biomarkers listed above (Table 5).

Table 5.

Sample sizes of representative acute stroke revascularization randomized clinical trials

	Imaging modality used and therapy tested	Sample size
NINDS trial¹⁴	NCT/r-tPA	624 patients
ECASS III¹⁵	NCT/r-tPA	821 patients
PROACT¹⁶	NCT/pro-urokinase	105 patients
MR CLEAN¹⁷	NCT-CTA/endovascular	500 patients
ESCAPE¹⁸	NCT-CTA/endovascular	316 patients
SWIFT PRIME¹⁹	NCT-CTA-PCT/endovascular	196 patients
tenecteplase trial²⁰	NCT-CTA-PCT/tenecteplase	75 patients
EXTEND IA²¹	NCT-CTA-PCT/endovascular	70 patients

We acknowledge a number of limitations to our study. Our results are restricted to CT analysis. Another limitation is that, ideally, the treatment decision would need to be essentially random and independent of the biomarkers tested. In our study, some imaging biomarkers (ASPECT score in 0–4.5-h time window – IV tPA Group #1, ASPECT score and presence of an M1 clot on CTA in the 3–9-h time window – endovascular treatment Group #2) were used to make treatment decisions. These biomarkers were not considered in the biomarker analysis as only those patients with positive values for these biomarkers were treated. Only those biomarkers not used for making treatment decisions were considered for this study. As long as the tested biomarkers are not correlated too strongly to the imaging biomarkers used in treatment decision, the validity of our study should be preserved. This seems to be the case as illustrated by the range of PCT ischemic core despite all patients having an ASPECT score ≥ 7. The results described above apply to our registry data, which carries a risk of selection bias as all registries, and may not be generalizable to the stroke population in general. However, our reported numbers should have internal validity and, while the absolute numbers needed to screen may be different in other study populations, their ranking relative to each other should hold even in different populations. It would be interesting to test an approach similar to the one used in this study in the pooled data from randomized clinical trials, in order to test the external validity and generalizability of our results. Another limitation is the possible interference of both clinical parameters or etiology of the infarct (not available). Finally, we did not have a sufficient number of patients with bridging r-tPA and endovascular therapy to conduct our analyses in this clinically important group.

In conclusion, predictive biomarkers need to be differentiated from prognostic biomarkers when being considered to select patients for a trial, and their prevalence should be assessed to determine number needed to screen and overall feasibility of the trials. We identified PCT penumbra and CTA collateral score as “feasible” predictive biomarkers for IV thrombolysis, and clot burden score, clot length, and PCT penumbra as “feasible” predictive biomarkers for endovascular revascularization.

Footnotes

Authors’ contribution

PM, TJ, and AE acquired the data.

QH performed the data processing and analysis.

JTP and WX performed the statistical analysis.

MW provided oversight for the study.

All authors were involved in manuscript writing and approved the final version of the manuscript.

Declaration of conflicting interests

The author(s) declared the following conflicts of interest with respect to the research, authorship, and/or publication of this article: TG Jovin has consulted for Codman Neurovascular and Neuravi, holds stock in Silk Road and Blockade; has acted as an unpaid consultant to Stryker as PI of the DAWN trial and served as an unpaid member of a Medtronic Advisory Board.

P Michel Co-investigator in ECASS-II, ECASS-III, DIAS trials, SWIFT-PRIME and BASICS.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Atkinson

Colburn

DeGruttola

. Biomarkers and surrogate endpoints: Preferred definitions and conceptual framework. Clin Pharmacol Ther 2001; 69: 89–95.

Parsons

Albers

. MR rescue is the glass half-full or half-empty? Stroke 2013; 44: 2055–2057.

Pexman

Barber

Hill

. Use of the Alberta stroke program early CT score (aspects) for assessing CT scans in patients with acute stroke. Am J Neuroradiol 2001; 22: 1534–1542.

Launes

Ketonen

. Dense middle cerebral artery sign: An indicator of poor outcome in middle cerebral artery area infarction. J Neurol Neurosurg Psychiatr 1987; 50: 1550–1552.

Shobha

Bal

Boyko

. Measurement of length of hyperdense MCA sign in acute ischemic stroke predicts disappearance after IV TPA. J Neuroimaging 2014; 24: 7–10.

Puig

Pedraza

Demchuk

. Quantification of thrombus hounsfield units on noncontrast CT predicts stroke subtype and early recanalization after intravenous recombinant tissue plasminogen activator. Am J Neuroradiol 2012; 33: 90–96.

Tan

Demchuk

Hopyan

. Ct angiography clot burden score and collateral score: Correlation with clinical and radiologic outcomes in acute middle cerebral artery infarct. Am J Neuroradiol 2009; 30: 525–531.

Torres-Mozqueda

Yeh

. An acute ischemic stroke classification instrument that includes CT or MR angiography: the Boston acute stroke imaging scale. Am J Neuroradiol 2008; 29: 1111–1117.

Tan

Demchuk

Hopyan

. Ct angiography clot burden score and collateral score: correlation with clinical and radiologic outcomes in acute middle cerebral artery infarct. Am J Neuroradiol 2009; 30: 525–531.

10.

Wintermark

Flanders

Velthuis

. Perfusion-CT assessment of infarct core and penumbra receiver operating characteristic curve analysis in 130 patients suspected of acute hemispheric stroke. Stroke 2006; 37: 979–985.

11.

van der Zwan

Hillen

Tulleken

Dujovny

. A quantitative investigation of the variability of the major cerebral arterial territories. Stroke 1993; 24: 1951–1959.

12.

Zhu

Michel

Aghaebrahim

. Prediction of recanalization trumps prediction of tissue fate the penumbra: a dual-edged sword. Stroke 2013; 44: 1014–1019.

13.

Kidwell

Jahan

Gornbein

. A trial of imaging selection and endovascular treatment for ischemic stroke. N Engl J Med 2013; 368: 914–923.

14.

Friedman

Koroshetz

Qureshi

. Tissue plasminogen activator for acute ischemic stroke. N Engl J Med 1996; 334: 1405.

15.

Hacke

Kaste

Bluhmki

. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med 2008; 359: 1317–1329.

16.

del Zoppo

Higashida

Furlan

Pessin

Rowley

Gent

. Proact: a phase II randomized trial of recombinant pro-urokinase by direct arterial delivery in acute middle cerebral artery stroke. Stroke 1998; 29: 4–11.

17.

Berkhemer

Fransen

Beumer

. A randomized trial of intraarterial treatment for acute ischemic stroke. N Engl J Med 2015; 372: 11–20.

18.

Goyal

Demchuk

Menon

. Randomized assessment of rapid endovascular treatment of ischemic stroke. N Engl J Med 2015; 372: 1019–1030.

19.

Saver

Goyal

Bonafe

. Stent-retriever thrombectomy after intravenous t-pa vs. T-pa alone in stroke. N Engl J Med 2015; 372: 2285–2295.

20.

Parsons

Spratt

Bivard

. A randomized trial of tenecteplase versus alteplase for acute ischemic stroke. N Engl J Med 2012; 366: 1099–1107.

21.

Campbell

Mitchell

Kleinig

. Endovascular therapy for ischemic stroke with perfusion-imaging selection. N Engl J Med 2015; 372: 1009–1018.