2.1. Data Source and Study Population

The Virtual International Stroke Trials Archive (VISTA) is an international, prospective data repository whose acute subsection (VISTA-Acute) has collated data from 38 RCTs in AIS to date, the methods of which have previously been described [15]. VISTA centralises high-quality individualised, patient-level data that has proved to be a valuable resource for numerous studies investigating thrombolysis [16–21].

Patients that were enrolled in RCTs whose investigative treatments reported no significant effect on AIS outcomes or that were placebo-treated in positive trials were eligible. Screening for eligible subjects yielded 3187 patient records that documented all of age; gender; prestroke mRS score; 90-day mRS score; IVT; onset-to-treatment time (OTT); baseline National Institutes of Health Stroke Scale (NIHSS) score; 90-day NIHSS score; and medical history including hypertension, myocardial infarction, diabetes mellitus, prior AIS, and smoking status. Adverse event records were extracted and screened for ICH—assessed per the trials’ respective diagnostic protocols—and all-cause mortality, within 90 days.

The resultant patient pool was divided on the basis of prestroke mRS score. Patients with a prestroke mRS score of 0 were previously asymptomatic while those with a prestroke mRS score of 1 were previously symptomatic, experiencing at least minor handicap in routine activities due to symptoms. VISTA vastly underrepresented patients with slight-to-severe prestroke disability (mRS ≥ 2), with only three (0.1%) mRS 2 patients and nine (0.3%) mRS 3–5 patients.

Asymptomatic and symptomatic patient groups were in turn subdivided on the basis of IVT status: IVT-treated or IVT-withheld. The proportion of IVT-treated patients was lower in VISTA than in typical observational datasets, with 18% of mRS 0 patients and 11% of mRS 1 patients receiving IVT. This is because the decision to administer IVT in VISTA’s contributing RCTs—with the notable exception of neuroprotectant trials [21]—was not made independently of the trial’s inclusion and exclusion criteria.

2.2. Outcome Variables

The primary safety outcome was 90-day mortality. The secondary safety outcome was ICH, reports of which were retrieved by screening adverse event records of RCTs per their diagnostic protocols. The primary efficacy outcome was favourable functional status at 3-month follow-up, defined as a 90-day mRS score of 0 or 1. This corresponds to one arm of the “excellent” outcome dichotomisation scheme of the mRS scale, that is, a score of 0 or 1 for patients without prestroke disability and a return to prestroke mRS score for patients with prestroke disability. The secondary efficacy outcome was significant neurological improvement, defined as a decrease of ≥ 8 points on the 90-day NIHSS score relative to baseline, or a decrease to 0 [22–24].

2.3. Statistical Analysis

3.1. IVT-Withheld vs. IVT-Treated Symptomatic Patients

Of the 721 previously symptomatic patients, IVT-withheld patients were older (median of 79 [69–83] years vs. 72 [61–78] years, p < 0.001) and had a more frequent history of previous stroke (59.8% vs. 41.8%, p = 0.003). No significant differences were found for the other baseline parameters (see Table S1).

3.2. Symptomatic vs. Asymptomatic IVT-Treated Patients

Of the 530 IVT-treated patients, the symptomatic group had a higher proportion of females (51.9% vs. 38.4%, p = 0.032). Moreover, previously symptomatic patients had a more frequent history of previous strokes (41.8% vs. 26.2%, p < 0.001). No significant differences were found for other baseline parameters (see Table S2).

3.3. Supplementary Analysis

In order to reduce the potentially confounding effect that pre-existing symptomaticity associated with previous stroke could have on outcomes among symptomatic patients, we conducted a supplementary analysis to compare symptomatic patients that received IVT for an index event (negative stroke history) versus a recurrence (positive stroke history). Of the 79 previously symptomatic patients that received IVT, 33 had a positive stroke history while 46 had a negative stroke history. No significant differences (p < 0.05) and no trends (p < 0.1) were found for any of the outcomes of interest with PSA. Regression analysis was not performed as the size of the dataset was not amenable to logistic regression modelling.

4.1. IVT-Withheld vs. IVT-Treated Symptomatic Patients

We found that risk of mortality was not significantly higher in IVT-treated mRS 1 patients compared to IVT-withheld mRS 1 patients (24.1% vs. 27.8%, p = 0.717). While the likelihood of favourable functional status was not significantly higher among the IVT-treated group, significant neurological improvement was significantly more common among treated patients (40.5% vs. 24.1%, p = 0.041). There were numerically more instances of ICH in the IVT-treated group; however, the difference failed to reach statistical significance in propensity score and multivariate logistic regression analyses.

To date, only two studies have compared IVT-treated versus IVT-withheld AIS patients with prestroke disability [26, 27]. The findings of these studies were not pooled in a recent meta-analysis [28] because one examined outcomes at 90-day follow-up [26] while the other did so at discharge [27]. In keeping with our findings, Merlino et al. reported that mortality and ICH did not differ between the two groups and that significant neurological improvement was more common among IVT-treated patients [26]. In contrast, Gumbinger et al. found that the odds ratio (OR) for mortality was not statistically significant among patients with prestroke mRS 1–5 but was significantly lower for IVT-treated patients with a prestroke score of 0 [27]. Unlike our analysis, both studies reported that IVT-treated patients had higher rates of favourable functional status than their untreated counterparts. In fact, Merlino et al. found that IVT-treated patients had more than seven times the odds of returning to prestroke mRS than untreated ones (aOR 7.26, 95%CI = 2.51–21.02, p = 0.001), with two in three disabled patients returning to their prestroke mRS when treated with IVT. Conversely, 12.7% of our disabled but fully functional subjects returned to prestroke mRS with IVT. A cause for this discrepancy could be that our analysis comprised few patients and presented a lower rate of IVT treatment in the subgroup with the higher grade of dependency (18% of mRS 0 vs. 11% of mRS 1 patients received IVT), potentially making our analysis prone to selection bias. However, Merlino et al.’s findings should be interpreted conservatively given that the study did not match treated cases and controls, making it difficult to make a judgement on how effectively covariates were adjusted for. Meanwhile, the declarative character of data reported in electronic case report forms (eCRFs) could have introduced bias in Gumbinger et al.’s study.

Although we do not report higher incidence of favourable functional status among IVT-treated patients, we did observe a higher rate of significant neurological improvement in this group. This suggests that 90-day NIHSS provides insights into therapeutic effects that are not captured by 90-day mRS alone, underscoring the importance of using NIHSS as an outcome measure alongside mRS. Although acute stroke trials typically use disability scales as their primary endpoint, it has been claimed that NIHSS could allow the therapeutic effects of drugs to be more readily identified [29]. Notably, the NIHSS 1 dichotomisation scheme has been shown to retain statistical power at 62% of the sample size required when using Barthel index score 95 (BI 95), previously established to be one of the best available disability endpoints alongside mRS 1 [29]. Our study provides incremental evidence in favour of the NIHSS enhancing statistical power. This may prove particularly useful when exploring the safety and efficacy of treatments in underrepresented cohorts where data remain scarce.

4.2. Symptomatic vs. Asymptomatic IVT-Treated Patients

We found that the proportion of patients exhibiting favourable functional status and major significant neurological improvement, or suffering mortality and ICH, did not differ significantly between the two groups.

To date, six studies have compared the outcomes of disabled versus nondisabled AIS patients receiving IVT [7, 8, 11, 30–32]. Of these, five defined disability as a prestroke mRS score of 3–5 and were pooled in a recent meta-analysis [7, 8, 30–32]. The exception was the study conducted by Zhang et al., which compared patients with prestroke mRS scores of 2–4 to patients with scores of 0–1 [11]. Of the studies pooled, four defined favourable functional status per the aforementioned good outcome dichotomisation scheme. The meta-analysis found that 41.2% of patients with premorbid disability achieved favourable functional status compared with 55.7% of patients without disability, corresponding to a statistically significant unadjusted OR of 0.53 (95%CI = 0.40–0.71). However, statistical significance did not persist after adjusting for age, sex, and other stroke risk factors using data available from three of the studies.

Although pooled analysis of the five studies found numerically more instances of symptomatic intracranial haemorrhage (sICH) among patients with premorbid disability, this trend did not achieve statistical significance after adjustment for demographic and stroke risk factors (aOR = 1.03, 95%CI = 0.87–1.23) [28]. Mortality, on the other hand, was revealed to be significantly higher among patients with premorbid disability and remained significant after adjustment (aOR = 2.27, 95%CI = 1.81–2.85) [28].

In alignment with these findings, we reported no significant difference in the incidence of ICH between the two groups. Moreover, we found that previously symptomatic subjects were as likely to exhibit favourable 90-day outcomes, that is, an mRS score of 0 or 1 and major neurological improvement on the NIHSS, as asymptomatic patients. However, we observed that the rate of mortality did not differ between the two groups.

Our definition of disability limits the direct comparability of our results to those of existing studies. However, Karlinski et al. performed a prestroke mRS-stratified analysis that affords insight into how mRS 1 patients fared relative to mRS 0 patients [7]. Analysing the Safe Implementation of Treatments in Stroke-Eastern Europe (SITS-EAST) registry, the association between mRS 1 score and sICH failed to reach statistical significance. However, increased likelihood of 90-day mortality was observed, as well as decreased likelihood of favourable mRS outcome per the good outcome dichotomisation scheme. Importantly, though the lower rate of favourable mRS outcome among mRS 1 patients reached statistical significance (p < 0.047), the CI included 1 (aOR = 0.80, 95%CI = 0.65–1.00). In corroboration with our findings, there may be reason to believe that this is a trend and not a correlation.

The principal strength of our study is that it is the first to perform analyses of both IVT-withheld versus IVT-treated previously symptomatic patients and symptomatic versus asymptomatic IVT-treated patients on the same population. The first approach—IVT-withheld versus IVT-treated—is the less well studied of the two. By adopting this approach and having control cases, that is, IVT-withheld mRS 1 patients, we enhanced internal validity. The second key strength is our use of PSA as primary analysis, which was motivated by the ambition to measurably minimise bias. Applied to the high-quality data collated and maintained by VISTA which was conceived to facilitate data compatibility and validity of analyses, PSA has particular credibility herein. Third, performing regression analysis alongside PSA added robustness to our study as these methods complement one another. PSA mimics an RCT and enabled us to readily evaluate the degree to which matching reduced systematic baseline differences between treatment groups. It is much more difficult to determine the degree to which regression adjustments minimise such differences [33]. On the other hand, it can be difficult to ascertain whether the assumptions underlying the validity of PSA are satisfied. Fourth, by placing emphasis on significant neurological improvement as favourable outcome parameter, our analysis was made more resistant to potential ceiling effects associated with mRS in patients with low prestroke mRS score but very severe stroke (NIHSS ≥ 25), of which there were 39 (5.4%) and 18 (3.4%) in Analyses 1 and 2, respectively. Fifth, our dataset was composed of patient records retrieved from AIS RCTs, many of which employed mRS as primary or secondary outcome measure such as European Cooperative Acute Stroke Study I (ECASS I) [34], European Cooperative Acute Stroke Study II (ECASS II) [35], National Institute of Neurological Disorders and Stroke (NINDS) [36], Tinzaparin in Acute Ischaemic Stroke Trial (TAIST) [37], Eurostat (ESTAT) [38], IMAGES [39], and more [15]. Raters of 90-day mRS were thus blinded to prestroke mRS, minimising bias introduced by erroneous use of the scale (e.g., reporting prestroke mRS that reflects patient status at admission rather than before onset of symptoms, or temporary disability due to recent injury or exacerbation of chronic conditions). After all, it is known that the assessment of prestroke mRS is challenging and interpretative [40]. An analogous argument can be made for NIHSS. Finally, while the external validity of our findings is difficult to assess, it is encouraging that VISTA has been used to identify and validate several clinically relevant prognostic factors [41].

As for shortcomings, we first recognise that our investigation was a retrospective observational study with a fairly small sample size. The negligible number of patients with slight-to-severe prestroke disability (mRS ≥ 2) signifies that our analysis could only compare nondisabled patients (mRS 0) and patients functionally impaired only insofar as prestroke symptomaticity (mRS 1). For this reason, we could not employ the “good” outcome dichotomisation scheme of the mRS scale—wherein an mRS of 0, 1, or 2 qualifies as a positive outcome for patients without prestroke disability—because even minor residual deficit poststroke, by definition, will increase an mRS 0 patient to mRS 1 (though the same deficit will not cause an escalation from mRS 1 to 2). This would have favoured patients with prestroke mRS 1 as they would have been more likely to achieve favourable functional status, thereby introducing bias. Admittedly, the limited count of mRS 1 subjects treated with IVT (n = 79) constrains our statistical power. Second, we were unable to distinguish between sICH definitions according to Safe Implementation of Treatments in Stroke (SITS), ECASS II, and NINDS because the dataset lacked sICH data. Our treatment of the most feared complication of IVT was thus somewhat crude, increasing risk of bias associated with interrater variability. This may explain why the regression models for ICH exhibited the lowest predictive powers, with F1 scores of approximately 0.1 and 0.2. Third, outcomes may have been biassed by nonrandom sampling since patients were definitely selected for thrombolysis on the basis of baseline and clinical characteristics. For example, the causes of pre-existing dependency may have played a role in the decision to administer or withhold IVT. Indeed, it has been shown that patients with disability not related to previous stroke may be slightly more likely to suffer sICH [7]. Unfortunately, causes of prestroke disability were not recorded. While a supplementary analysis was performed to compare previously symptomatic patients that received IVT for an index event versus a recurrence, this does not amount to a substitute for this variable. Other confounders may also have been missed due to incomplete documentation; in particular, endovascular treatment data was unavailable. Missing medication data for > 80% of the patient population also meant that we could not identify those patients that were IVT-withheld due to absolute contraindication owed to the use of anticoagulants. These limitations reiterate the importance of performing logistic regression analyses, since the validity of PSA rests on two assumptions: (1) that all confounding variables are measured and (2) that each participant has a nonzero probability of being treated, meaning no participant has an absolute contraindication to either the active or control treatment [42]. Fourth, since we reported significant neurological improvement without functional gain among IVT-treated mRS 1 patients, it is important to reiterate the caveats of NIHSS: the scale does not consider symptoms such as postural instability, gait disturbance, and dysphagia [43]. As a result, significant neurological improvement can coexist with debilitating symptoms that have a greater impact on patient quality of life. Importantly, handicapping symptoms may have made patients with more severe sequelae less likely to return for 90-day NIHSS evaluation, which may have introduced systematic bias. However, it is more likely the missing 90-day NIHSS data resulted from deliberate RCT design. VISTA’s foundational document acknowledges that RCTs routinely failed to make optimal endpoint choices [15], and some contributing RCTs such as Surgical Treatment for Ischemic Heart Failure (STICH) [44] and AccesS to Kidney transplantation (ASK) [45] did not explicitly set out to measure 90-day NIHSS in their protocols. As VISTA does not disclose the trials from which individual patients were sourced, it was not possible to determine whether records lacking 90-day NIHSS differed nonrandomly from complete cases. However, this appears unlikely since no patients were lost to 90-day mRS follow-up evaluation. Finally, the fact that none of the contributing trials investigated IVT presents an added challenge in understanding the possible role of confounding by indication. Our analysis would be more robust if we could confirm that the included trials had been conducted simultaneously in regions where alteplase was routinely used and others where it was not yet adopted. While we have limited information on this point, it is encouraging to note that many contributing trials predate the widespread use of alteplase.

Overall, demographic ageing of the global population demands that neurologists become better equipped to make decisions about IVT treatment in AIS patients with prestroke disability. The prevalence of disability increases with age [46], as does the incidence of AIS which has been found to double with each decade after the age of 45 years [47]. Further, the increasing incidence of AIS in the younger population will add to the large number of survivors that live with stroke-related disabilities for many years [48]. High-quality evidence is imperatively needed for this heretofore neglected patient population. Recent post hoc analyses of Head Position in Stroke Trial (HeadPoST)—an international, cluster-randomised crossover trial of lying-supine versus sitting-up head positioning in acute stroke care—examined the impact of including patients with prestroke disability in an acute stroke trial [49]. The HeadPoST investigators reported that there was no heterogeneity of trial treatment effect by level of premorbid function. Optimistically, this will promote the recruitment of patients with pre-existing disability. In the meantime, pragmatic and transparent decision-making that considers prestroke mRS score and baseline NIHSS in tandem—against the broader backdrop of social and lifestyle factors—may produce the best patient outcomes.