Introduction
In 2015, endovascular treatment (EVT) was shown to be effective in improving functional outcome in patients with ischaemic stroke due to intracranial large vessel occlusion [
1]. Secondary outcome analyses of the randomised clinical trials also indicated significantly smaller infarct volumes at follow-up imaging in patients who were allocated to the intervention group [
2,
3]. Studies have suggested that follow-up infarct volume (FIV) could be a useful early outcome measure [
4‐
6].
FIV as a surrogate outcome is a well-quantifiable measure and therefore less sensitive to interobserver variability compared to clinical assessment of functional outcome such as the modified Rankin Scale score at 90 days [
7,
8]. Also, FIV measurements can be assessed relatively easily and semi-automatically after treatment on non-contrast computed tomography (NCCT) or magnetic resonance imaging (MRI) scans [
9,
10]. A moderate correlation between FIV and clinical outcome has been demonstrated [
11]. FIV has been suggested as a primary endpoint in late phase II clinical trials, which are intended to demonstrate an indication of therapeutic effect in promising novel treatments. Assessment of functional outcome as a clinical endpoint requires prolonged follow-up. An early surrogate marker could therefore be more feasible in clinical trials, and limit loss to follow-up [
12,
13].
A recent post hoc study demonstrated that the beneficial effect of EVT on functional outcome could be explained by preventing progression of FIV, suggesting that the effect of intervention on functional outcome is mediated by FIV [
6]. Formal testing of such a mechanism requires a causal mediation model to estimate the extent to which the treatment effect is explained by a mediator [
14]. This is usually expressed as a proportion of the original treatment effect. In the context of testing a mediator as a surrogate marker, the Prentice criteria have been proposed to formally test for a causal relation between surrogate and clinical endpoints [
15]. This analytical approach for estimating the causal effect of FIV on functional outcome has not yet been fully reported for EVT in acute ischaemic stroke and the extent to which the beneficial effect of intervention on functional outcome can be explained by difference in FIV is not yet known [
16]. Understanding the causal pathway of this relation may provide further insight and may help in developing surrogate markers of functional outcome after EVT, and shed further light on outcome predictors that can be used for future stroke trials. The aim of this study was to assess whether and to what extent FIV on NCCT at 5–7 days’ follow-up is a mediator of the effect of intervention on functional outcome in acute ischaemic stroke patients.
Discussion
In this study, we tested with mediation analysis whether the beneficial effect of intervention for acute ischaemic stroke on functional outcome could be explained by FIV. We found that FIV on NCCT at 5–7 days was affected by treatment, and was related to functional outcome, but only explained a modest part of the effect of intervention on functional outcome at 90 days measured by the modified Rankin scale in patients with acute ischaemic stroke. This implies that FIV on NCCT only partially explains the effect of intervention on functional outcome and should therefore not be used as an early surrogate imaging marker for clinical endpoints in trials.
A previous study found a significant association between volume of FIV on NCCT and three different functional outcome measurements at 3 months. However, a moderate correlation between infarct volume and all functional outcome measures was found [
11]. The study did not report the commonly used mRS score as a functional outcome measurement. Another study, which also included ischaemic stroke patients undergoing intervention, demonstrated that FIV was an important determinant of functional outcome at 3 months [
4]. However, this study used imaging (NCCT or MRI) in a broad time window between 24 h and 2 weeks after stroke. Our conclusion also differs from a previous study on this topic, which concluded that FIV explains the effect of intervention on functional outcome [
6]. However, in that study only the first and second step of mediation analysis were performed, and not the third step. This implies that no definite conclusion on mediation could be drawn, which explains the discrepancy with our findings. No other studies that reported an association between FIV and functional outcome did not perform a full a causal mediation analysis. In the REVASCAT study a mediation analysis was carried out, with similar results, but the proportion of explained treatment effect was not estimated [
16]. Our study is the first full mediation analysis to analyse the pathway from intervention to FIV to functional outcome and report the proportion of explained treatment effect mediated by FIV.
Several assumptions must be made to perform an unbiased causal mediation analyses [
21]. First, there is no unmeasured confounding between treatment and outcome; this assumption is automatically satisfied in our study due to randomisation of treatment. Secondly, no unmeasured confounding between mediator and outcome should be present. This is true for our study as the observers were blinded with respect to clinical information during imaging analysis [
9]. Third, there should be no unmeasured confounding between the treatment and mediator. This requirement is also satisfied in our study due to randomisation and the fact that FIV measurements were assessed after baseline. This is also confirmed by the consistency of the results of the adjusted and unadjusted analyses.
A limitation of our study is the exclusion of deceased patients in the first week after onset and therefore missing FIV measurements at 5- to 7-day follow-up (n=60 (12%)). In our sensitivity analysis, we tried to overcome this by imputing FIVs in these deceased patients. Results of the sensitivity analysis did not change the conclusions of our paper and effect sizes are comparable. Although factors other than FIV possibly play a role in early death, it is likely that the more severely affected patients with potentially large FIV will be over-represented among patients who died early [
4,
6]. In our study, no FIV measurements on NCCT in 99 patients could be assessed at 5- to 7-day follow-up mostly because of logistic reasons; this could result in a distortion of the results. We therefore used imputation techniques to adjust for this potential bias [
22,
23]. The estimates of the mediator (FIV measurements) must be reliable and valid. Our automated, observer-checked estimation method has been shown to be reliable [
17]. Overestimation of infarct size due to oedema may occur. The randomised assessment of treatment effect will reduce this bias. In a sensitivity analysis, we showed that use of 24-h NCCT FIV for missing FIV did not increase the explained proportion, probably because FIV measurement is less precise, and hypodense areas may yet increase in size.
Another limitation is our relatively small sample size. The different pathways in our mediation model (EVT-FIV, FIV-functional outcome and EVT- functional outcome) are all frequently studied and confirmed in multiple datasets. However, the proportion of the effect of EVT on functional outcome has never been calculated before. Our relatively wide CI expressed the uncertainty in this estimate. Therefore, our findings need to be replicated in other randomised control trials performed on EVT.
Our study made use of follow-up NCCT to assess FIV, because this is the most widely available and used modality. It would be of interest to also study effect mediation by FIV measured with MRI. Care should be taken, however, that selection bias in assessment does not distort the comparison between MRI and CT.
A large proportion of the treatment effect on functional outcome remains unexplained, suggesting FIV alone cannot be used as an early proxy of functional outcome. Effects of other pathways may play a role in determining functional outcome such as infarct location. Previous studies have demonstrated that certain brain regions are more sensitive than others to hypoperfusion, which may interact with FIV regarding functional outcome, given that the relevance for functional outcome varies by regional eloquence [
24‐
26]. In our study, patients had an occlusion of the middle cerebral artery supplying eloquent brain regions. Small lesions in eloquent regions may have a larger destructive effect on functional outcome than larger infarcts in non-eloquent regions.
Further studies should address the question whether combining FIV with a measure of eloquence can increase the predictive value for functional outcome [
26,
27]. Taking eloquence into account might improve the proportion of explained mediated effect. The best method to combine eloquence, location and infarct volume is not yet known. Infarcts do not only affect the cortical regions but also white matter tracts. Small infarcts in eloquent cortical regions or important white matter tracts might result in severe strokes. This type of analysis, which takes into account the location of infarct in mediation models, requires larger datasets. In our study, we were mainly interested in FIV as a surrogate imaging biomarker as a first step, because it has been used in several studies [
28]. However, for further understanding of the pathophysiological mechanisms relating infarct volume to functional outcome, taking location into account is the obvious next step [
26]. Another approach could be to combine the FIV measurement with assessments of specific stroke symptoms and stroke severity. NIHSS is currently increasingly used for assessment of initial stroke severity in clinical practice [
29‐
31], but it might also be an interesting intermediate outcome measurement [
16]. Also, in our primary analysis, we assessed FIV at 5–7 days. FIV on NCCT in other time windows might also be of interest as a surrogate marker [
32]. Finally, the use of more advanced imaging modalities such as MRI to determine FIV would be interesting for future studies.
In conclusion, we confirmed that intervention prevents progression of FIV on NCCT, but this only partly explains the beneficial effect of intervention on functional outcome.