Baseline stroke risk and efficacy of dual antiplatelet therapy: a post-hoc analysis of the POINT trial

Abstract

Background-

High-risk transient ischemic attacks (TIA) and minor ischemic strokes are followed by a variable risk of ischemic stroke. We aimed to determine how baseline stroke risk modified the efficacy of clopidogrel-aspirin (referred to here as dual-antiplatelet therapy, or DAPT) for TIA and minor ischemic stroke.

Methods:

We performed an unplanned secondary analysis of the POINT (Platelet-Oriented Inhibition in New TIA and Minor Ischemic Stroke) trial. We first evaluated the associations of the CHA₂DS₂-VASc and SPI-II scores with the risk of incident ischemic stroke and major hemorrhage (ICH or major systemic hemorrhage). We then tested for heterogeneity of the relative and absolute treatment effect of DAPT relative to aspirin across low and high-risk patient subgroups.

Results-

4,841 trial participants were included in this analysis, with 2,400 participants assigned to treatment with short-term DAPT and 2,430 participants to treatment with aspirin and placebo. The dichotomized SPI-II score, but not the CHA₂DS₂-VASc score (P=0.18), was associated with risk of incident ischemic stroke. A high-risk SPI-II score (>3) was associated with greater risk of incident ischemic stroke (hazard ratio (HR) of incident ischemic stroke relative to low-risk SPI-II score 1.84, 95% confidence interval 1.44–2.35, P<0.001) and numerically greater risk of major hemorrhage though not meeting statistical significance (HR 1.80, 95% CI 0.90–3.57, P=0.10). The relative risk reduction with DAPT was similar across SPI-II strata (P_interaction=0.31). The absolute risk reduction for ischemic stroke with DAPT compared to aspirin was nearly four-fold higher (2.80% versus 0.76%; number needed to treat 31 versus 131) in the high-risk SPI-II stratum relative to the low-risk stratum. The absolute risk increase for major hemorrhage with DAPT compared to aspirin was three-fold higher (0.84% versus 0.30%; number needed to harm 119 versus 331) in the high-risk SPI-II stratum relative to the low-risk stratum.

Conclusion-

Stratification by baseline stroke risk identifies a patient subgroup that derives greater absolute benefit from treatment with DAPT.

Clinical trial registration:

URL: https://www.clinicaltrials.gov; Unique identifier: NCT03354429.

Graphical Abstract

Introduction

The high incidence of ischemic stroke following a transient ischemic attack (TIA) or minor ischemic stroke has motivated the search for pharmacologic therapies that can reduce this risk. In selected stroke and TIA populations, one such effective therapy is a short-term course of dual antiplatelet treatment (DAPT).^1,2 In the Platelet-Oriented Inhibition in New TIA and Minor Ischemic Stroke (POINT) trial, the relative risk for prevention of ischemic stroke of DAPT over aspirin was 0.72, with an overall absolute risk reduction (ARR) of 1.7% which corresponds to a number needed to treat of 58.¹

The ARR of an intervention reported in clinical trials is a group average that reflects the baseline risk in the overall trial population for the disease outcome and thus may not accurately estimate the average benefit for any given subgroup. In most trials, the distribution of event risk across the population is such that a small proportion of high-risk patients account for the majority of clinical outcomes.³ Thus, estimating ARR based on the baseline risk of a disease outcome across patient subgroups may be more relevant to high-risk patients.³ The most efficient strategy to estimate this baseline risk is to apply a multivariable approach that pools several clinical risk factors and that can be calculated for every participant in a clinical trial.³ This strategy is in contrast with the less efficient approach of performing numerous underpowered subgroup analyses using single variables that do not fully capture baseline risk.³ Applying this paradigm to clinical trial data facilitates a move beyond treatment effects based on population averages to more tailored estimates of benefit, which is a key principle of precision medicine.³

In the present study we applied this analytic paradigm to better understand which patient sub-populations in the POINT trial¹ derived the greatest absolute benefit from treatment with DAPT after TIA or minor ischemic stroke.

Methods

Data availability and IRB review

The authors will make the data, detailed methods, and all other study materials available to researchers who wish to reproduce the analysis in this article. We performed a secondary analysis of deidentified data and so the study was exempt from institutional review board review.

Cohort description

The POINT trial recruited patients who had experienced either a high-risk TIA (ABCD² score of four or higher) or minor ischemic stroke (with a National Institutes of Health Stroke Scale score of 3 or less) as previously described.¹ Patients were excluded if they were candidates for a carotid intervention, anticoagulation, or intravenous thrombolysis. Patients were enrolled within twelve hours of their last known normal time and were randomly assigned to treatment with either a 90-day course of DAPT, consisting of clopidogrel (75mg daily after a loading dose) and aspirin (at a dose of 50mg-325mg), or a 90-day course of aspirin and placebo. The present analysis used only intention-to-treat assignments. Patients were followed for up to 90 days and clinical outcomes were adjudicated by an independent clinical-event committee.

Risk scores

We reviewed available recurrent stroke risk scores⁴ and selected scores for which most variables were available in the POINT dataset and that could be applied to the entire cohort so as to maximize statistical power (e.g. not specific to TIA such as the ABCD² score⁵, but rather applicable to both TIA and stroke patients). We selected the Stroke Prognosis Instrument II (SPI-II)⁶ and CHA₂DS₂-VASc⁷ risk scores. The variables used to define these scores are presented in Table 1.

Table 1. Elements of the SPI-II and CHA₂DS₂-VASc risk scores.

POINT: Platelet-Oriented Inhibition in New TIA and Minor Ischemic Stroke; TIA: transient ischemic attack.

Variable	Available in POINT dataset	SPI-II points	CHA2DS2-VASc points
Congestive heart failure	Yes	3	1
Diabetes	Yes	3	1
Prior stroke	No (omitted from scores)	3	1
Age	Yes	2 (>70)	<65 – 0 65–74 - 1 ≥75 – 2
Stroke for the index event (not TIA)	Yes	2	NA
Hypertension	Yes	1	1
Ischemic heart disease (qualifies as ‘vascular disease history’)	Yes	1	1
Female Sex	Yes	NA	1

The SPI-II score was derived from four cohort studies⁶ and is associated with recurrent stroke risk in the clinical trial setting.⁸ The SPI-II score includes a ‘prior stroke’ variable which was not available in the POINT database and so this item was not included in the present analysis (Table 1). Although the CHA₂DS₂-VASc score was designed to stratify stroke risk amongst patients with atrial fibrillation, this score is also associated with recurrent ischemic stroke risk in patients without atrial fibrillation.^9,10 The CHA₂DS₂-VASc score includes a ‘prior stroke or TIA’ variable, but since all participants in POINT had either a prior stroke or TIA, we also dropped this item from this score (Table 1). Participants with missing data on any of the variables used in the risk scores listed in Table 1 were excluded (n=40). Each trial participant’s risk score was modelled both as a continuous variable and as a binary variable dichotomized at the median. Of note, given that the two risk scores are ordinal variables, a median split resulted in similar but not equally sized subgroups.

Clinical outcomes

The primary efficacy outcome for the analysis was incident ischemic stroke within 90 days of randomization. Ischemic stroke was operationalized in the POINT trial as the presence of either a new neurological deficit that persisted for greater than 24 hours or presence of infarct on brain imaging.¹¹ The primary safety outcome was ‘major hemorrhage,’ which was operationalized as a composite of: symptomatic intracranial hemorrhage, bleeding resulting in loss of vision, bleeding resulting in hospitalization or prolongation of existing hospitalization, bleeding resulting in transfusion or two or more units of red blood cells, and bleeding resulting in death.¹¹

Statistical analysis

Across all analyses we used either univariable or multivariable Cox proportional hazards models to estimate hazard ratios and 95% confidence intervals. In the first stage of the analysis, we evaluated the associations between each risk score and incident ischemic stroke and major hemorrhage using univariable Cox proportional hazards models. We modeled the risk scores separately as continuous variables and as dichotomous variables split at the median score, with patients above this cutoff considered to be ‘high-risk.’ Associations of the risk scores with ischemic stroke were visualized using Kaplan-Meier survival curves. We assessed model discrimination by calculating the concordance statistic (C-statistic) for each Cox model.¹² Dichotomized scores that were significantly associated with incident ischemic stroke were carried forward for further analysis.

In the second stage of the analysis, we tested for heterogeneity of relative risk between each stratum of risk. We first tested the association of treatment with DAPT with incident ischemic stroke separately within the ‘low-risk’ and ‘high-risk’ strata. Kaplan-Meier curves were generated showing associations of DAPT therapy with the outcome stratified across risk score strata. We then performed a fixed-effects meta-analysis of these two point estimates for efficacy of DAPT relative to aspirin and calculated Cochran’s Q for heterogeneity. A P value less than 0.05 was considered to constitute evidence for heterogeneity of relative risk reduction across the risk strata. We also performed conventional analyses testing for statistical interaction within the whole cohort by including risk score, treatment effect, and the interaction between risk score and treatment effect as variables in a Cox regression model.

In the third stage of analysis, we compared absolute risks of ischemic stroke and major hemorrhage within each risk score stratum⁸. This was accomplished by calculating the percentage of participants who developed the primary outcome within each stratum, with the numeric difference in the absolute risk (expressed as a percentage) between the ‘high-risk’ and ‘low-risk’ groups constituting the absolute risk reduction (ARR). The number needed to treat (NNT) to prevent an ischemic stroke was calculated by taking the inverse of the percent absolute risk reduction. Parallel analyses were undertaken using the outcome of major hemorrhage to calculate the number needed to harm (NNH).

Secondary analyses

We performed two secondary analyses. First, we modeled the risk scores that were carried forward to the second stage of analysis using tertiles to create ‘low-risk,’ ‘intermediate-risk,’ and ‘high-risk’ groups. We then tested for association of these strata with incident ischemic stroke and major hemorrhage, and quantified the ARR, ARI (absolute risk increase), NNT, and NNH (number needed to harm) across the risk score strata. Second, we carried out analyses using risk scores carried forward to the second stage of analysis that were limited to 21 days of follow-up by censoring participants who had not experienced an event by 21 days. This analysis was motivated by the finding that most of the benefit of DAPT in the POINT trial was observed in the first 21 days, and that treatment beyond this duration is associated with a small, but steadily accruing risk of major hemorrhage.^13,14

Results

A total of 4,841 trial participants were included in the analysis, with 2,400 participants assigned to treatment with DAPT and 2,430 to treatment with aspirin and placebo. Cohort characteristics stratified across risk score medians are described in Table 2. Counts for each stratum of the risk scores are provided in Supplementary Table 1. The median SPI-II score was 3 and the median CHA₂DS₂-VASc score was 2. The SPI-II (P=0.41) and CHA₂DS₂-VASc scores (P=0.24) were similarly distributed across the two treatment arms.

Table 2. Baseline characteristics and outcome for patients above and below the median of the SPI-II and CHA₂DS₂-VASc risk scores.

Binary variables are reported as n (%) and continuous variables as mean (standard deviation (S.D.)). P values were calculated using analysis of variance (ANOVA) for continuous variables and the chi-squared test for categorical variables.

	SPI-II score			CHA2DS2-VASc score
	Low risk	High risk	P	Low risk	High risk	P
	n=2764	n=2077		n=2782	n=2059
Demographics
Age (S.D.)	60.80 (12.50)	69.61 (12.17)	<0.001	57.89 (10.76)	73.62 (10.27)	<0.001
Female Sex (%)	1212 (43.8)	962 (46.3)	0.093	856 (30.8)	1318 (64.0)	<0.001
Race (%)			0.561			0.094
Asian	73 (2.6)	70 (3.4)		81 (2.9)	62 (3.0)
Black	536 (19.4)	420 (20.2)		585 (21.0)	371 (18.0)
White	2038 (73.7)	1488 (71.6)		1988 (71.5)	1538 (74.7)
Ethnicity (%)			0.821			0.799
Hispanic or Latino	216 (7.8)	170 (8.2)		228 (8.2)	158 (7.7)
Not Hispanic or Latino	2414 (87.3)	1812 (87.2)		2422 (87.1)	1804 (87.6)
Clinical variables
Stroke as Qualifying Event	1306 (47.3)	1442 (69.4)	<0.001	1652 (59.4)	1096 (53.2)	<0.001
Randomization to Clopidogrel (%)	1362 (49.3)	1049 (50.5)	0.414	1353 (48.6)	1058 (51.4)	0.062
Heart failure (%)	1 (0.0)	123 (5.9)	<0.001	17 (0.6)	107 (5.2)	<0.001
Ischemic heart disease (%)	78 (2.8)	418 (20.1)	<0.001	68 (2.4)	428 (20.8)	<0.001
Hypertension (%)	1594 (57.7)	1762 (84.8)	<0.001	1483 (53.3)	1873 (91.0)	<0.001
Type 2 Diabetes (%)	50 (1.8)	1283 (61.8)	<0.001	385 (13.8)	948 (46.0)	<0.001
Smoking (%)			<0.001			<0.001
Never (%)	1427 (51.6)	1099 (53.0)		1343 (48.3)	1183 (57.5)
Past (%)	665 (24.1)	660 (31.8)		687 (24.7)	638 (31.0)
Present (%)	671 (24.3)	315 (15.2)		751 (27.0)	235 (11.4)

Stage One – Associations of risk scores with incident ischemic stroke and major hemorrhage

A one-unit increase in the SPI-II score was associated with increased risk of incident ischemic stroke (hazard ratio of ischemic stroke per unit increase in risk score 1.15, 95% confidence interval (CI) 1.09–1.21, P<0.001 Table 3). An SPI-II score above the median of 3 (‘high-risk’) was associated with increased incident ischemic stroke risk (1.84, 95% CI 1.44–2.35, P<0.001; Figure 1). The score had a concordance statistic of 0.60 (standard error 0.02). The SPI-II score was not significantly associated with incident major hemorrhage (Table 3).

Table 3. Hazard ratios for associations of SPI-II and CHA₂DS₂-VASc risk scores with incident ischemic stroke and with incident major hemorrhage in the POINT trial (n=4,841).

Point estimates reported from models that include the risk scores as continuous variables reflect the hazard ratio for the outcome per unit increase in the risk score. Point estimates reported from models that include the risk scores as dichotomized variables reflect the hazard ratio of the outcome in patients above the median score relative to those at or below the median score. CI: confidence interval; HR: hazard ratio

Risk score	Model	HR for ischemic stroke [95% CI]	P value	HR for major hemorrhage [95% CI]	P value
SPI-II	Continuous (unit increase)	1.15 [1.09–1.21]	<0.001	1.11 [0.95–1.29]	0.16
	Dichotomized (≤3 vs >3)	1.84 [1.44–2.35]	<0.001	1.80 [0.90–3.57]	0.10
CHA2DS2-VASc	Continuous (unit increase)	1.14 [1.05–1.24]	0.003	1.34 [1.06–1.70]	0.01
	Dichotomized (≤2 vs >2)	1.18 [0.93–1.50]	0.18	2.34 [1.15–4.75]	0.02

Figure 1. Kaplan-Meier (KM) curves for cumulative ischemic stroke event rates across strata of the modified SPI-II and CHA₂DS₂-VASc risk scores (n=4,841).

(A) KM curves depicting cumulative ischemic stroke event rates for POINT participants at or below the median SPI-II score of three (light grey, n=2,764), or above the median SPI-II score of three (black, n=2,077). The log-rank test was statistically significant (P<0.001). (B) KM curves depicting cumulative ischemic stroke event rates for POINT participants at or below the median modified CHA₂DS₂-VASc score of 2 (light grey, n=2,782), or above the median CHA₂DS₂-VASc score of 2 (black, n=2,059). The log-rank test was not statistically significant (P=0.18).

A one-unit increase in the CHA₂DS₂-VASc score was associated with increased risk of incident ischemic stroke (1.14, 95% CI 1.05–1.24, P=0.003); however, the dichotomized score was not associated with incident ischemic stroke (1.18, 95% CI 0.93–1.50, P=0.18; Figure 1). The score had a slightly weaker concordance statistic of 0.55 (standard error 0.02). The continuous and dichotomized CHA₂DS₂-VASc scores were associated with increased risk of major hemorrhage (Table 3). We thus carried forward the SPI-II score for further analysis.

Stage Two – Heterogeneity of relative treatment effects across risk score strata

The hazard ratios were consistent with a protective effect of DAPT over aspirin in both the ‘low-risk’ SPI-II stratum (0.82, 95% CI 0.57–1.20, P=0.30) and the ‘high-risk’ SPI-II stratum (0.64, 95% CI 0.46–0.88, P=0.007; Figure 2). These effect size estimates were not significantly heterogeneous (P=0.31) between strata. Interaction terms between the treatment arm and the continuous SPI-II score (P=0.78), and between the treatment arm and the binary SPI-II score (P=0.32) were not statistically significant.

Figure 2. Kaplan-Meier (KM) curves for cumulative ischemic stroke event rates across treatment arms (dual antiplatelet therapy versus aspirin) stratified on high vs. low SPI-II scores.

(A) KM curves depicting cumulative ischemic stroke event rates across treatment arms for POINT participants at or below the median SPI-II score of 3 (n=2,764). The black line depicts event rates in the clopidogrel arm, and the grey line depicts the placebo arm. Shaded areas reflect corresponding 95% CIs. The log-rank test was not statistically significant (P=0.30). (B) KM curves depicting cumulative ischemic stroke event rates across treatment arms for POINT participants above the median SPI-II score of 3 (n=2,077). The black line depicts event rates in the clopidogrel arm, and the grey line depicts event rates in the placebo arm. Shaded areas reflect corresponding 95% CIs. The log-rank test was statistically significant (P=0.007). The effect of dual antiplatelet therapy relative to aspirin did not interact with the SPI-II score (P=0.32).

Stage Three – Heterogeneity of absolute treatment effects across risk score strata

The absolute risk reduction attributable to treatment with DAPT over aspirin across all participants in the trial (n=4,841) was 1.7% which corresponds to a number needed to treat of 56 to prevent one stroke. The absolute risk reduction in the low SPI-II stratum (2,764) was 0.76% which corresponds to a number needed to treat of 131 to prevent one stroke. Finally, the absolute risk reduction in the high SPI-II stratum (2,077) was 2.8% which corresponds to a number needed to treat of 36 to prevent one stroke. A similar risk gradient was observed with respect to the number needed to harm for major bleeding (Table 4; Supplementary Table 3): The NNH was 186 in the unstratified cohort, 331 in the low SPI-II stratum, and 119 in the high SP-II stratum.

Table 4. Descriptive statistics of absolute risks, number needed to treat (NNT), and number needed to harm (NNH) for incident ischemic stroke and major hemorrhage across treatment and placebo arms stratified by high and low SPI-II scores.

Results reported from the unstratified cohort are abstracted from the original POINT trial results.

SPI-II stratum	Treatment arm (n)	Absolute stroke risk	NNT	Absolute major hemorrhage risk	NNH
Unstratified cohort	Aspirin (2,430)	6.3%	56	0.41%	186
	Clopidogrel plus aspirin (2,411)	4.6%		0.95%
Low (SPI-II ≤3)	Aspirin (1,402)	4.41%	131	0.36%	331
	Clopidogrel plus aspirin (1,362)	3.65%		0.66%
High (SPI-II >3)	Aspirin (1,028)	9.20%	36	0.48%	119
	Clopidogrel plus aspirin (1,049)	6.40%		1.32%

Secondary Analyses

Modelling the SPI-II score using tertiles yielded a similar pattern of results (Supplementary Tables 2–3, Supplementary Figure 1). Analyses limited to a 21-day follow-up period showed a similar association of the SPI-II score with ischemic stroke risk (hazard ratio of dichotomized score 1.67, 95% CI 1.27–2.18, P<0.001) but a weaker association with major hemorrhage (hazard ratio of dichotomized score 1.32, 95% CI 0.46–3.79, P=0.59; Supplementary Table 4). There was a similar pattern of absolute risks as well, with a NNT of 28 within the high-risk SPI-II stratum and 119 within the low-risk SPI-II stratum (Supplementary Table 5).

Discussion

In this secondary analysis of the POINT trial, we found that baseline stroke risk stratified the absolute but not the relative efficacy of DAPT over aspirin for reducing ischemic stroke risk after TIA and minor ischemic stroke. These descriptive statistics may be considered a resource for clinicians interested in more accurately conveying and weighing the benefits and risks of treatment with DAPT after high-risk TIA or minor ischemic stroke. These results permit several conclusions.

First, we found that the modified SPI-II score is associated with 21 and 90-day stroke risk within the clinical trial setting and thus is a valid tool to use in future analyses. Second, we found that stratification by baseline stroke risk identifies populations that derive greater absolute benefit from DAPT over aspirin. This was reflected in a nearly four-fold gradient of absolute risk reduction across low and high baseline stroke risk. Third, we found a parallel three-fold increase in major hemorrhage risk which supports the notion that ischemic stroke risk factors tend to also be risk factors for major bleeding and intracranial hemorrhage.¹⁵ Finally, we found that baseline stroke risk did not modify the relative benefit of treatment with DAPT after TIA or minor stroke. Though power to detect an interaction was limited, this is in keeping with a general trend of homogeneity of relative risk reduction across risk strata in clinical trial populations.³

These findings parallel those from a similar analysis that investigated the relationship between baseline stroke risk and benefit with pioglitazone in the Insulin Resistance After Stroke (IRIS) trial.⁸ Baseline stroke risk was estimated using both an internal model derived from IRIS data and using the SPI-II score. As in our study, this analysis identified a gradient of absolute (4.9%) but not relative risk reduction across risk strata. This absolute risk gradient was likely numerically greater in IRIS due to the longer follow-up of the trial (5 years) relative to the POINT trial (90 days).

There are several strengths from our analysis to highlight. First, we used high-quality clinical trial data with harmonized and well-defined clinical variables and adjudicated outcomes. Second, the treatment effects of DAPT are randomized and hence are not biased by confounding as would be the case in real-world observational studies. Third, the derivation cohort for the SPI-II score was distinct and so its predictive performance is not biased by potential overfitting. There are also limitations to consider. First, this was a post hoc analysis and so these results should be considered hypothesis-generating. Second, the short duration of follow-up and low event rate limited the ability to apply a more granular stratification strategy, such as by quartiles or quintiles of risk scores. This also limited power in testing heterogeneity of treatment effect. Third, we were unable to test several established stroke recurrence risk scores, some of which may have offered superior performance relative to the SPI-II score.^4,16. Moreover, one clinical element from each risk score was not available and therefore omitted from the final analysis. Indeed, despite associating with incident ischemic stroke, the SPI-II score had relatively weak discrimination. This may at least partially reflect the differences in length of follow-up used in development of the SP-II (2 years) and CHA₂DS₂-VASc (1 year) scores versus the shorter-term follow-up in the present study. This further emphasizes the need for better risk prediction tools in this context. This also contributed to the limited granularity of stratification. Finally, these results are only generalizable to patient populations that are similar to those enrolled in the POINT trial.

These limitations lend themselves to several important areas of future investigation. First, future studies should aim to identify risk factors that are associated with ischemic stroke risk but not with bleeding risk. Specialized scores created using such variables could be used to identify patient populations with more favorable risk-benefit ratios of treatment with antithrombotic agents. Second, future studies should aim to identify risk scores that discriminate incident stroke more effectively in patients with high-risk TIA or minor ischemic stroke. These multivariable risk scores should then be built into analysis plans for clinical trials. Finally, future investigations should explore the cost-effectiveness of treatment with DAPT in high vs. low-risk populations.

Conclusions

Baseline stroke risk scores identify patient subgroups that derive greater absolute ischemic risk reduction from treatment with DAPT compared to aspirin. Additional work is needed to identify the clinical scenarios in which this finding could influence clinical care.

Non-standard abbreviations

ARR

Absolute risk reduction

DAPT

Dual-antiplatelet therapy

NNH

Number needed to harm

NNT

Number needed to treat

POINT

Platelet-Oriented Inhibition in New TIA and Minor Ischemic Stroke

SPI

Stroke prognosis instrument

TIA

Transient ischemic attack