Determining the efficacy of thrombolytic therapy in acute ischemic stroke

Determining the efficacy of thrombolytic therapy in acute ischemic stroke: An analysis of the recent stroke literature.

The first part of this manuscript will start off with a critique of the present-day methodology of determining the therapeutic efficacy of tissue plasminogen activator (TPA) therapy -- specifically the statistical methodology that uses an adjusted analysis, rather than an unadjusted analysis. I believe that the present-day methodology of only using adjusted ORs is biased and inaccurate and not truly reflective of reality. In the second section of the manuscript, I will use "evidence" from recent research findings to demonstrate the complexities involved in analysing TPA-for-stroke trials. The subject is enormously complex because there are many confounding variables which can impact, and potentially inflate, or deflate, TPA's "apparent" therapeutic effect in any individual stroke patient -- site of clot location, degree of vessel occlusion, type of occluding clot (fibrin-rich versus platelet-rich), the volume of the central infarct core zone, the size and viability of the ischemic penumbra zone, the efficacy of the collateral circulation which is responsible for maintaining viability of the ischemic penumbra zone prior to recanalisation, the degree and speed of vessel recanalisation, the frequency and degree of vessel re-occlusion following partial or complete recanalisation, the local blood pressure and local metabolic conditions (eg. hyperglycemia), and the degree of reperfusion injury following delayed restoration of blood flow. I personally believe that it is not possible to accurately determine the "true" efficacy of TPA therapy using a randomised controlled trial design (RCT) if one either does not ensure that these confounding variables are balanced between the treated and placebo patients, or if one does not apply appropriate post hoc "corrective" adjustments to correct for any significant imbalance in those critically important prognostic variables.

I think that Anthony Furlan is correct. I think that the problem of stroke heterogeneity has been grossly underestimated in TPA-for-stroke trial design and result interpretation, and that problems relating to stroke heterogeneity is the source of many of the "incorrect" estimations of TPA's clinical efficacy, that are derived from TPA-for-stroke RCTs. I also believe that it is a mistake to trust statisticians to make sense of it all using complex statistical techniques of uncertain validity. I agree with James Penston's attitude regarding the statistical analysis of a RCT's results. He stated the following in his insightful book [2] regarding large-scaled RCTs-:

I personally think the stroke research community should design and interpret TPA-for-stroke RCTs in such a way that they are not dependent on statisticians to determine whether TPA is actually efficacious in acute ischemic stroke. The RCT's results should be presented in a clear, transparent and easily understandable manner.

How does TPA work, and why do we expect acute ischemic stroke patients to benefit from intravenous TPA therapy? Basically, an acute ischemic stroke is due to a clot occlusion of an arterial vessel supplying blood to the brain. The most common sites of an occluding clot are in arterial blood vessels supplying the anterior cerebral circulation eg. the internal carotid artery, the carotid siphon area at the circle of Willis, or the middle cerebral artery or its branch vessels. The two most common types of clots are i) embolic clots (cardio-embolic strokes) that usually arise from a cardiac source or from a clot in a proximal extracranial vessel, and ii) atherothrombotic clots that usually arise de novo in large atherosclerosed arterial vessels (large vessel strokes). Embolic clots are usually fresh, fibrin-rich (red) clots, and those red clots are most likely to respond to TPA therapy because the drug is essentially a fibrinolytic agent. Atherothrombotic clots are usually platelet-rich (white) clots and those white clots are obviously less likely to respond to a fibrinolytic agent.

How does one know whether TPA is effective in dissolving a blood clot in an intracerebral blood vessel? Theoretically, one first has to establish that a clot is present in an arterial vessel by using an appropriate imaging technique, and then demonstrate that recanalisation occurred following TPA therapy. Recanalisation may be partial or complete, and present-day stroke researchers currently use the TIMI system for grading the degree of recanalisation. TIMI 0 recanalisation implies a total failure to restore any blood flow while TIMI 3 recanalisation implies the complete restoration of blood flow. TIMI 1 and 2 recanalisation implies partial restoration of blood flow to varying degrees. At the time the NINDS study was performed, imaging technology that would allow the stroke researchers to establish that a vessel clot was definitely present and that recanalisation actually occurred following TPA therapy, was not routinely available. The NINDS stroke researchers used a clinical endpoint as the trial's primary endpoint -- the rate of favorable stroke outcome at 24 hours and 3 months. The obvious ' apriori' assumption was that if TPA resulted in partial/complete recanalisation, then the patient would more likely recover fully from the stroke.

In designing the NINDS study, the investigators based their diagnosis of acute ischemic stroke on the clinician's assessment of the patient's clinical presentation, and they excluded a non-ischemic stroke (eg. hemorrhagic stroke) and other intracranial pathology (eg. brain tumors) by using a CT scan prior to administering TPA therapy. They also incorporated inclusion/exclusion criteria in their trial design to minimize the recruitment of TIA and stroke-mimic patients. After administering intravenous TPA therapy to stroke patients within 3 hours of stroke onset, they then re-evaluated the stroke patient 24 hours later to determine whether the patient's neurological status had improved. They used the NIHSS neurological scoring system to quantify the severity of the stroke at baseline (prior to administering TPA therapy) and again at 24 hours. Although some stroke patients showed a dramatic recovery within hours of receiving TPA therapy, they could not demonstrate an "overall" benefit at 24 hours, and there was not a clinically significant difference between treated and placebo patients at 24 hours. They also studied the patients at different time points (eg. 7 days and 3 months) after TPA therapy. The NINDS investigators never officially presented their 7-day results in their 1995 NEJM article [1], and their original NEJM report mainly demonstrated a clinically significant difference at 3 months between treated and placebo patients using a clinical stroke outcome scoring system as the primary endpoint. What clinical stroke outcome scoring system did they use as a clinical endpoint at 3 months? They used a multiplicity of stroke outcome scoring systems -- NIHSS, modified Rankin scale, Barthel Index, Glasgow, and their "self-created" Global Test statistic. The Global Test statistic scoring system was apparently designed by the NINDS statisticians using a statistical methodology that is unspecified (or incompletely specified) in the mainstream medical literature.

These are their 3 month favorable stroke outcome results as presented in their original 1995 NEJM report [1].

Note that they plotted their favorable stroke outcome results using 5 different stroke outcome scoring systems (Global, Barthel Index, Modified Rankin Scale, Glasgow Outcome Scale, NIHSS) and two different statistical methods (Odds Ratio and Relative Risk). I don't want to bore the reader with the details of the different scoring systems, but it is my present understanding that the two most useful stroke outcome scoring systems in clinical use are the Modified Rankin scale and Barthel Index, and that by defining a favorable stroke outcome as a mRS≤1 or BI >95, one is best able to difference between patients who have a favorable versus unfavorable stroke outcome result (see reference 3 for further details on the comparative value of different stroke outcome scoring systems). I mainly used the Modified Rankin Scale stroke outcome scoring system when evaluating the "apparent" efficacy of TPA therapy, because it is the stroke outcome scoring system most frequently used by trialists when reporting their results in the medical literature, and I accepted the traditional definition of a favorable stroke outcome as being a mRS≤1.

What I particularly like about table 4 is the fact that the NINDS study's results are presented in an unadjusted (unadulterated) manner, which is clear and transparent and totally free of statistical manipulations. Note that the NINDS investigators simply reported what percentage of their treated and placebo patients had a favorable stroke outcome, and that they then calculated the OR and RR values without indulging in any statistical manipulations. Note that the OR for a favorable stroke outcome for patients treated in 0-90 minutes was 1.7 (mRS≤1) compared to 2.4 (mRS≤1) for patients treated between 91-180 minutes. Those unadjusted OR results suggest that TPA was more effective for patients treated later rather than earlier within the three hour time window. That surprising result contradicted the NINDS investigators 'a priori' expectations, and I am not entirely surprised that they arbitrarily decided to create their own stroke outcome scoring system - Global Test - which is apparently based on an arbitrary congealment of the other four stroke outcome scoring systems. Note that use of the Global Test stroke outcome scoring system resulted in an OR of 1.9 for patients treated early, as well as later. That identical OR result of 1.9 suggested that TPA was equally effective for patients treated between 91-180 minutes as it was for patients treated in <90 minutes, and that there was no time-to-treatment interaction for patients treated in <3 hours.

However, the NINDS investigators were obviously not happy with their 1995 interpretative conclusion, and they decided to re-interpret the same raw data at a later date. In December 2000, they published a paper [4] that demonstrated that TPA was more effective if given earlier rather than later. Consider this figure from their article.

Note, from this graph, that the NINDS investigators estimated there is a significant time-to-treatment interaction, and they estimated that TPA was much more effective if given early (<90 minutes) rather than late (>91 minutes). In fact, if one looks at the time period 60-90 minutes, the NINDS investigators estimated that the OR is approximately 2.8-3.6. That figure is much higher than their unadjusted OR figure of 1.9 (I am presuming they are using the same stroke scoring outcome system in the graph -- the Global Test). What would cause the adjusted OR value to be so much higher than the unadjusted OR value? In their paper, the authors claimed that they adjusted for baseline NIHSS stroke severity +/- other covariates, but I could find no "information" in their paper that could explain why the adjusted OR figure would be so much higher than the unadjusted OR figure. I personally believe that if the NINDS investigators think that an adjusted OR value is more accurately reflective of reality (TPA's "true" treatment effect) than an unadjusted OR value, then they have to provide much more details regarding their pathophysiological assumptions and statistical manipulations. Until the NINDS investigators provide complete details of their statistical methodology -- in a manner that is totally transparent, easily understandable and entirely plausible -- then I think that one should regard their adjusted OR figures as being scientifically unacceptable.

What further decreases my confidence in the NINDS investigators' scientific objectivity is their propensity to change their adjusted OR figures with the passage of time, without supplying adequate background information on their reasoning and their statistical manipulations. Compare the adjusted OR figures of that graph to the following graph published in the Lancet [5].

This graph was produced from the pooled patient-level data from the ECASS, ECASS II, ATLANTIS and NINDS trials (2,775 patients). However, virtually all the patients treated in <90 minutes were derived from the NINDS study. That allows one to compare the adjusted OR results for patients treated between 60-90 minutes from the first graph (figure 2) to this graph (figure 3). Note that the OR for patients treated between 60-90 minutes was approximately 2.8-3.6 in graph 2 compared to 2.1-2.2 in graph 3. That is huge difference! How does one account for that large difference considering that one is interpreting the same raw data from essentially the same patients. The NINDS investigators could claim that they used a different Global Test stroke outcome scoring system and different covariates in their statistical manipulations, but that only proves that the adjusted OR figures have an "arbitrary" quality that may not be scientifically valid. We still have no proof that the "arbitrary" adjusted OR figures are more reflective of TPA's "true" treatment effect than the unadjusted OR figures.

Also, compare the adjusted OR figures for patients treated between 91-180 minutes using the OR results from graph 2 and graph 3. In graph 2, the estimated adjusted OR for patients treated between 91-180 minutes was approximately 1.2-2.8, while in graph 3 the estimated OR figure for patients treated between 91-180 minutes is 1.7-2.1. That is a significant difference! The main effect of that difference is that it allows the NINDS investigators to now claim that TPA still has a substantial therapeutic effect at 180 minutes ("new" adjusted OR 1.7 compared to a "previous" adjusted OR figure of 1.2), and that TPA is even effective beyond 180 minutes. Are these new "arbitrary" adjusted OR figures credible?

This is graphical representation of the pooled results from the 2,775 treated patients -- based on their baseline NIHSS stroke severity scores.

Consider the OR results from patients treated between 91-180 minutes. Note the wide disparity in the OR results for the different stroke severity subgroups. Are those adjusted OR results pathophysiologically plausible?

Note that the ATLANTIS-ECASS-NINDS rt-PA study group investigators estimated that TPA was particularly effective in patients with a baseline NIHSS stroke severity score >20 (very severe stroke) and that the OR is approximately 4.2. Is an OR of 4.2 pathophysiologiocally plausible? Is an OR of 4.2 for very severe stroke patients consistent with "expected" reality? Consider the basic facts. In the NINDS study, patients with a baseline NIHSS stroke severity score >20, who were treated in <90 minutes, had a 6% rate of a favorable stroke outcome at 3 months ((mRS≤1) -- compared to 3% for placebo patients -- and the unadjusted figure OR figure for those patients is 1.9. If there is no substantial evidence to suggest that TPA was very efficacious in very severe stroke patients (baseline NIHSS >20) treated in <90 minutes in the NINDS study, why would one expect TPA to be much more effective in very severe stroke patients treated after 90 minutes?

Note, also, that the ATLANTIS-ECASS-NINDS rt-PA study group investigators estimated that TPA had a OR of approximately 2.6 for very mild stroke patients (baseline NIHSS score 0-5) who were treated between 91-180 minutes, and according to the plotted results in figure 2, TPA was even more efficacious for patients with very mild strokes than it was in patients who had intermediate severity strokes. Is that pathophysiologically plausible? Also, consider the actual unadjusted results from the NINDS study. In the NINDS study, treated patients with a baseline NIHSS stroke severity score of 0-5, who were treated between 91-180 minutes, had a 83% rate of a favorable stroke outcome (mRS≤1) -- compared to 86% for placebo patients -- and the unadjusted OR figure is <1.0. I realize that there were different numbers of patients, and therefore different end-results, in the pooled-data study compared to the NINDS study, but how can one explain an adjusted OR of 2.6 for the pooled data when the unadjusted OR is <1.0 for the NINDS study?

I personally find the ATLANTIS-ECASS-NINDS rt-PA study group investigators' adjusted OR results for some of the stroke severity subgroups inexplicable and pathophysiologically implausible.

Consider another example. Consider a direct comparison between the unadjusted NINDS study results and the adjusted OR results for the pooled patients treated in 0-90 minutes -- based on baseline stroke severity (I presume that all the pooled patients actually came from the NINDS study thereby making this direct comparison particularly pertinent).

NINDS study: 0-90 minute arm. Rate of a favorable stroke outcome at 3 months (mRS<1)
Baseline NIHSS	Treated patients	Placebo patients	Unadjusted OR	*Adjusted OR
0-5	9/13 (69%)	7/9 (78%)	0.64	0.7
6-10	23/30 (77%)	15/37 (40%)	4.8	4.8
11-15	17/39 (44%)	10/31 (32%)	1.6	2.9
16-20	12/40 (30%)	8/37 (21%)	1.5	3.3
>20	2/35 (6%)	1/31 (3%)	1.9	1.7
All patients	63/157 (40%)	41/145 (28%)	1.7	2.8

Note that there was no significant difference between the adjusted and unadjusted OR results for the baseline NIHSS 0-5, 6-10, and >20 stroke severity subgroups. However, there was a significant difference between the adjusted and unadjusted OR results for the baseline NIHSS 11-15 and 16-20 stroke severity subgroups. Do you understand why there is that selective difference for certain subgroups? Could you discover any "information" in the statistical methodology section of the ATLANTIS-ECASS-NINDS rt-PA study group investigators' Lancet article that fully explains why there is a selective difference between the adjusted and unadjusted OR values in those two stroke severity subgroups?

I think that it is scientifically negligent to only present adjusted OR values when performing an analysis of the pooled data from multiple TPA-for-stroke RCTs. I think that an objective researcher should also present the unadjusted OR values for the same data set, and then demonstrate why the adjusted OR values are likely to be more representative of TPA's "true" treatment effect than the unadjusted OR values. Otherwise, the entire interpretative process will appear to be politicised, and TPA-contrarians will believe that the adjusted OR values are arbitrary and deliberately biased.

Of course, one could use the oft-repeated argument that one should not even selectively look at individual subgroup results because there is no guarantee that the subgroups (which are non-randomised) are balanced for important prognostic variables. That is the standard argument, but I think that the standard argument makes no sense, because there is also no evidence that randomisation actually resulted in balanced groups in the entire patient group in those four TPA-for-stroke studies anyway. In fact, the prime example of a very unbalanced RCT is the 91-180 minute arm of the NINDS study where the treated group had far more patients with a very mild stroke (baseline NIHSS 0-5) and far less patients with a very severe stroke (baseline NIHSS >20) than the placebo group. If one corrects for that imbalance post hoc [6], then one notes that there is little time-to-treatment difference in the unadjusted OR results between patients treated early (<90 minutes) compared to late (91-180 minutes), and this interpretative conclusion is not dependent on arbitrary stroke outcome scoring measuring tools (Global Test) and/or arbitrary statistical models. I previously argued [6] that the standard interpretation of the unadjusted results of the 91-180 minutes arm of the NINDS study is scientifically invalid, because the groups were not balanced for baseline stroke severity. In fact, that specific imbalance is only the tip of the iceberg! What we also do not know is whether they were balanced for other important prognostic variables -- site of clot occlusion, stroke subtypes, size of central core infarct zone, efficacy of the collateral circulation and degree of viability of the ischemic penumbra zone tissue, rates of re-occlusion following partial/complete recanalisation -- and any significant imbalance in those confounding variables could also significantly influence a TPA-for-stroke RCT's overall results. The difference in the unadjusted OR results for the different stroke severity subgroups in the NINDS study could be due to imbalances in those important prognostic variables. The lack of basic knowledge about the influence of imbalances in confounding variables is a major conundrum in TPA-for-stroke trials -- we simply have no idea what would happen to a comparable group of untreated stroke patients if they were not treated with TPA therapy, because we have no means of determining whether randomisation actually resulted in an equal balance of those important prognostic variables between treated and placebo patients. Randomisation may result in reasonably balanced groups in large clinical trials (>10,000 patients), but it certainly cannot guarantee an adequate balance in smaller trials (few hundred patients) [7]. The pooled data from the 2,775 patients is merely a conglomeration of results from a multiplicity of smaller unbalanced subgroups, and if the unadjusted/adjusted results of many of the subgroups are invalid, then the internal validity of the entire interpretative enterprise is questionable.

I think that the sheer complexity of the "effect" of many confounding variables on any tPA-for-stroke RCT's results makes it very difficult to accurately interpret a RCT's final results. To understand how complex the situation really is, read the next section dealing with confounding variables.

In this section, I will be analysing recent research papers that significantly increase our understanding of what happens to stroke patients who receive thrombolytic therapy. The two major advances that have occurred during the past few years is that stroke researchers are now able to identify the site and degree of clot occlusion, and the speed and degree of recanalisation. Those advances allows one to evaluate two phenomena-: i) the degree of TPA's efficacy in producing recanalisation; and ii) the degree of correlation between the rate of recanalisation and the rate of clinical recovery at 3 months.

To make the following discussion more understandable for the uninitiated, I will first describe recent changes in our understanding of what likely happens to a stroke patient following clot occlusion of a middle cerebral artery.

Our understanding of what happens following clot occlusion of a middle cerebral artery has been significantly advanced by MRI and MRA studies, which have demonstrated that there are probably two zones of ischemic brain tissue. The best review article on the subject that I have read is by Kidwell [8]. According to the Kidwell MRI model, clot occlusion of a middle cerebral artery results in a central zone of infarcted tissue (MRI diffusion imaging abnormality). The central infarct zone is surrounded by an ischemic penumbra zone (MRI perfusion imaging abnormality zone), which represents viable, but threatened brain tissue. Surrounding the ischemic penumbra zone, is a peripheral zone of benign oligemia.

Note from the diagram, that the Kidwell model suggests that the central core infarct zone consists of two separate infarct zones. The most central core zone consists of a central core of irreversibly infarcted tissue (core), and it is surrounded by another central zone of potentially “salvageable” infarct tissue (MRI diffusion abnormality). Early thrombolytic therapy may occasionally salvage some of the “salvageable” infarct core zone tissue if the thrombolytic drug is given expeditiously and recanalisation occurs promptly. The final size of the core infarct zone is also affected by the degree of recruitment of ischemic penumbra zone tissue (MRI perfusion abnormality) into the central infarct tissue zone with the passage of time. Untreated stroke patients, who do not undergo spontaneous recanalisation, experience significant growth of the central infarct zone into the ischemic penumbra zone during the first 24 hours, as the ischemic penumbra zone fails to survive. Thrombolytic therapy, by restoring blood flow to the threatened, but viable, ischemic penumbral zone tissue, can limit the extent of growth of the central core infarct zone into the ischemic penumbra zone, and thereby limit the final size of the core infarct zone. The therapeutic benefit of thrombolytic therapy partially depends on the speed and degree of recanalisation, but the final size of the core infarct zone also depends on other factors such as the degree and efficacy of the collateral circulation, the pattern of clot fragmentation and the chance-event occurrence of downstream embolisation resulting in distal occlusion of branch vessels in the same vascular territory, the degree of reocclusion of the original clot site following successful recanalisation, the degree of local tissue edema, the degree of reperfusion injury, the "local" blood pressure and other local metabolic conditions (eg. hyperglycemia).

Although I have been critical of the official interpretation of the NINDS study, I have never asserted that TPA doesn't work in acute ischemic stroke. It obviously works! One only has to review case reports where angiography demonstrates an abrupt clot occlusion of a basilar artery in a patient with a profound neurological deficit, and note how angiography demonstrates complete vessel recanalisation following timely intraarterial thrombolytic therapy, which is then subsequently followed by complete neurological recovery, to realise how useful TPA therapy can be for the individual stroke patient. However, what we really need to understand is how useful TPA therapy can be for a sample population of stroke patients, and that requires a deeper understanding of how confounding variables can influence a patient's response to TPA therapy.

To start exploring the potential value of TPA therapy in a sample population of acute ischemic stroke patients, consider this paper by Felberg [9]. The researchers studied patients with a middle cerebral artery stroke syndrome due to a clot occlusion of the terminal internal carotid artery or middle cerebral artery, who subsequently received intravenous TPA therapy. They noted that 22% of the TPA-treated patients had a dramatic recovery, which they defined as an improvement of 10 NIHSS points or an improvement to 3 NIHSS points at the end of the hour-long infusion. They used serial or continuous transcranial Doppler monitoring to quantify the degree of recanalisation and restoration of blood flow. They were interested in delineating the differences between stroke patients who had a dramatic recovery (DR) versus stroke patients who did not have a dramatic recovery (non-DR). What can one learn for their study? Is there a direct correlation between the speed and degree of recanalisation and the likelihood of a dramatic recovery within a few hours of thrombolytic therapy? Is there a direct correlation between the degree of recanalisation and the likelihood of a favorable stroke outcome at 24 hours and at 3 months?

This is what the researchers found. There was no correlation between the likelihood of a DR and the site of vessel occlusion -- terminal internal carotid artery, proximal or distal middle cerebral artery. There was only a slightly greater likelihood of a DR if the patients had a cardioembolic stroke (67% of DR strokes, 51% of non-DR strokes) rather than a large vessel stroke (25% of DR strokes, 20% of non-DR strokes). There was no significant difference in time-to-treatment between DR and non-DR stroke patients (114 minutes for the DR group versus 127 minutes for the non-DR group). That result is surprising because it demonstrates that speedier thrombolytic therapy doesn't necessarily increase the chance of a DR, despite the authors' exaggerated claim-: "In summation, dramatic recovery during IV-TPA infusion is a relatively frequent occurrence in patients with acute MCA strokes and occurs because of the early restoration of flow during tPA infusion."

So what specifically causes patients to have a dramatic recovery if it is not the promptness of thrombolytic treatment? Is it related to the degree of vessel recanalisation? In fact, there is a definite association between the likelihood of a DR and the degree of recanalisation, but the association is not clear-cut.

Note that 92% of DR patients had partial or normal blood flow restoration compared to 55% of non-DR patients. That fact suggests that DR is more likely to happen if restoration of blood flow occurs following thrombolytic therapy. However, note that 41% of DR patients had a dramatic clinical recovery within hours of the TPA infusion despite having complete/partial occlusion at the end of the infusion. Why did those 41% of DR patients have such a dramatically good clinical result if recanalisation was incomplete? That phenomenon suggests that there is another confounding variable in play that allowed many stroke patients to have a DR even though recanalisation was incomplete. What confounding variable? The only variable that I can think of that would allow such a good clinical response in the absence of complete recanalisation is the variable of collateral circulation efficiency.

Recall that according to the Kidwell model, there is a zone of ischemic penumbra surrounding the central core infarct zone, and that clinical recovery is due to resoration of blood flow to the ischemic penumbra zone. If blood flow is not restored to the ischemic penumbra zone within a finite period of time, it will fail and become part of the central core infarct zone. In patients who have a DR without complete vessel recanalisation, one could presume that i) despite a significant MCA occlusion there must have been a very effective collateral circulation that minimised the size of the central core infarct zone and ii) that even partial (or even minimal) restoration of blood flow is enough to maintain viability of the ischemic penumbra zone and prevent growth of the small core infarct zone into the ischemic penumbra zone. Can I prove that those DR patients had a small core infarct zone and little subsequent growth of the core infarct zone with the passage of time? I cannot, because those patients did not have a MRI (or cerebral blood flow study) performed, and we have no knowledge of the size of the core infarct/ischemic penumbra zones in those patients.

Another interesting fact that can be learnt from the Felberg paper is that stroke patients who had a DR also had a good clinical response at 24 hours and at 3 months. That fact is not really surprising because, from a pathophysiological perspective, one would expect stroke patients, who have a small core infarct zone and sufficient restoration of blood flow following thrombolytic therapy to prevent growth of the core infarct zone, to do well clinically. What is really surprising from my perspective, is that the non-DR patients had a poor clinical recovery at 3 months despite receiving TPA therapy (their average mRS score at 3 months was 4 compared to an average mRS score of 1 for DR patients). Why am I surprised that the non-DR recovery patients did so poorly? Note that 55% of the non-DR group had restoration of blood flow following thrombolytic therapy (41% partial, 14% complete), and one would expect a significant number of those patients to have a favorable stroke outcome (mRS≤1) at 3 months (more on the relationship between restoration of blood flow and clinical recovery in a later section). What could have prevented a favorable clinical recovery in those non-DR patients who had partial or complete recanalisation? One theoretical possibility is that they were unlucky enough to have a poor collateral circulation that resulted in a large central infarct zone despite partial/complete restoration of blood flow. In other words, I am postulating that the efficacy of the collateral circulation is a major confounding variable that can significantly affect a stroke patient's chance of a favorable clinical recovery following thrombolytic therapy. I suspect that stroke patients who are lucky enough to have a good collateral circulation are much more likely to have a small core infarct zone, a DR and a good clinical recovery at 3 months.

I previously suggested that stroke patients, who have a complete MCA occlusion, probably have different core infarct zone volumes, presumably due to varying degrees of efficacy of the collateral circulation. Do I have EBM evidence to support that belief?

Consider this paper by Fovin [10]. The researchers studied patients with middle cerebral artery occlusions. Many of the patients received intravenous or intraarterial thrombolytic therapy, or both. The researchers used a xenon cerebral blood flow imaging technique to define the extent of the core infarct zone and the ischemic penumbra zone as a percentage of the middle cerebral artery territory. They noted that there was strong correlation between the size of the central core infarct zone and an unfavorable clinical outcome. Patients with a large central core infarct zone had a poor prognosis, irrespective of whether the ischemic penumbra zone was salvaged, or not. Patients with a large central core infarct zone were also more likely to experience hemorrhagic transformation and edema within the core infarct zone, and those two pathophysiological phenomena would obviously increase the likelihood of an unfavorable stroke outcome. I think that the results of their study are pathophysiologically plausible. What most interested me about their study was the large variability in the size (volume) of the central core infarct zone in the 36 patients studied.

Consider this figure demonstrating the size of the core infarct zone and ischemic penumbra zone in all the 36 studied patients.

Note the large variation in the size of the core infarct zone in those 36 studied patients -- varying from 7% to 65% of the MCA territory. The large degree of variability is important because it likely represents a significant confounding variable in TPA-for-stroke RCTs. If a RCT by chance has a disproportionately greater number of stroke patients with a large core size (volume) in the treatment group relative to the placebo group, then that imbalance would significantly deflate TPA's "apparent" therapeutic effect, because those patients would be less likely to respond to any therapy that recanalises occluded vessels. That imbalance would be equivalent to stacking the deck against TPA therapy! In the previous section, I described the Felberg study which demonstrated that most non-DR patients had a poor clinical recovery at 3 months even though 55% of those patients had partial/complete restoration of blood flow, and I suspect that a poor collateral circulation causing a large core infarct zone was a major causal factor responsible for the poor clinical response. One would not normally expect such a poor clinical recovery in a sample population of stroke patients if the rate of successful recanalisation is 55%. How do I know what's the "average" likelihood of a favorable stroke outcome in stroke patients who experience an approximately 55% rate of partial/complete recanalisation following an acute ischemic stroke? Consider the article by Neumann-Haefelin in the next section.

The relationship between the degree of recanalisation and a favorable stroke outcome.

Neumann-Haefelin [11] studied patients with a middle cerebral artery occlusion, to determine the frequency and relevance of complete versus incomplete recanalisation. He used MRI imaging to determine the size of the central core infarct zone and ischemic penumbra zone, and MRA imaging to determine the degree of and frequency of recanalisation. The degree of recanalisation was quantified as follows -- TIMI 0 = No recanalisation. TIMI 1 = Minimal recanalisation TIMI 2 = Partial recanalisation TIMI 3 = Complete recanalisation. Most (52) of the study patients (87) received intravenous thrombolytic therapy.

Note that 38% of the patients who received thrombolytic therapy achieved partial/complete recanalisation (TIMI 2/3). What is interesting is that 24% (1-in-4) of the patients who didn't receive thrombolytic therapy also achieved partial recanalisation (TIMI 2). Therefore, there was not a significantly greater number of patients who achieved TIMI 2/3 flow in the thrombolytic group compared to the untreated group, and only 17% of the thrombolytic patients in this study achieved complete recanalisation.

What is also interesting is that there was not a linear relationship between the degree of recanalisation and the rate of favorable stroke outcome at 3 months.

Consider the following table which shows the rates of a good stroke outcome at 3 months depending on the degree of TIMI recanalisation (whether they received thrombolytic therapy, or not).

First of all, note the enormous variability in the baseline NIHSS score -- despite all the study patients having an identical site of clot occlusion (M1 occlusion = MCA main stem occlusion). I can only presume that the large variability is due to the fact that there was a large interindividual variability in the efficacy of the collateral circulation. Note that a good stroke outcome was achieved in 60% of the patients who had a TIMI 2/3 recanalisation, and that there was not a significant difference in the rate of a good stroke outcome at 3 months between patients who had TIMI 3 recanalisation compared to TIMI 2 recanalisation (60% versus 63%). Note, surprisingly, that 36% of TIMI 0 patients had a good stroke outcome at 3 months even though they achieved zero recanalisation. There must be other pathophysiological mechanisms in play that enabled 36% of TIMI 0 patients to have a good stroke outcome at 3 months. One possible explanation is that those patients had an excellent collateral circulation that minimised the core infarct volume even though early recanalisation did not occur (the authors did not mention whether there was a direct correlation between a low baseline NIHSS score and a good clinical outcome in that particular subgroup).

This type of TPA-for stroke study demonstrates how stroke patient heterogeneity makes it extremely difficult to interpret a study's results. What lessons can be learnt from this study's results. The authors concluded that this study demonstrated that "incomplete recanalization on day 1 is a frequent MR finding after MCA main stem occlusion and is predictive of a favorable clinical course. Even a TIMI 1 flow grade on MRI 24 hours after symptom onset was found to be associated with less lesion growth (P<0.05) than persistent occlusion. MR indicators of early recanalization could be useful surrogates of efficacy in acute stroke studies." Do you believe that this study's results suggest that MR indicators of recanalisation can be useful surrogates of TPA's efficacy in acute ischemic stroke? First of all, there was no linear correlation between the use of thrombolytic therapy and the likelihood of complete versus incomplete recanalisation in this study. Secondly, there was no linear correlation between the degree of recanalisation and the likelihood of a good clinical recovery at 3 months -- in fact, there was no "extra" benefit at 3 months in patients who had complete recanalisation (TIMI 3) versus partial recanalisation (TIMI 2). Also, 36-45% of patients in this study had a good stroke outcome at 3 months even though they had no/minimal recanalisation (TIMI 0/1). How do you make sense of the study's data? Surely, there must be other major confounding variables in play that cause the relationship between the rate/degree of recanalisation and the final clinical outcome at 3 months to be so ill-defined and so irregular. One confounding variable that I have already discussed is interindividual variability in the efficacy of the collateral circulation. That confounding variable probably played a highly significant role in this small study, because one notes that at baseline the patients had a marked variability in their baseline NIHSS stroke severity score (median of 13, but varying from 1-26) and also in their core infarct zone volume (median of 22ml, but varying from 1-187 ml) despite having the same site of vessel occlusion (M1 MCA occlusion). What other confounding variables could be affecting the study's result? What about the confounding variable of re-occlusion following partial or complete recanalisation? How often does re-occlusion occur and could a chance occurrence of more re-occlusion events in one subgroup bias a TPA-for-stroke study's results against that subgroup? Consider the study by Alexandrov [12].

I previously stated that Felberg studied stroke patients who had a dramatic recovery in response to thrombolytic therapy to better understand why they responded so well. However, I was intrigued by the fact that 55% of the non-DR patients had a poor clinical response at 24 hours and at 3 months despite partial/complete restoration of blood flow following thrombolytic therapy. I previously wondered whether the non-DR patients didn't respond because they may have had a large central core infarct zone due to an inefficient collateral circulation. However, could there have been other confounding variables that limited the clinical response to TPA therapy following partial/complete recanalisation? Alexandrov [12] specifically studied MCA stroke patients who had partial or complete recanalisation following intravenous TPA therapy, but who didn't have a good clinical response following the restoration of blood flow. He studied stroke patients with a MCA occlusion (M1 or M2 occlusion), who received standard dose intravenous TPA therapy. Non-responders were defined as those patients with early recanalisation (ER = partial or complete recanalisation within 2 hours of TPA therapy) who had no change or a decrease in NIHSS score by 4 points at 2 and 24 hours after stroke onset. Responders were patients who improved by >4 NIHSS points at 2 hours (immediate) and 24 hours (early). He noted that of 73 MCA stroke patients who had ER, 27 were non-responders (37%). Why did 37% of the study's patients who recanalised early following thrombolytic therapy not have a good clinical response at 24 hours? Interestingly, there was no difference in the median baseline NIHSS score, time-to-treatment, and time-to-recanalisation between non-responders and responders who had early recanalisation. One factor that was of particular relevance was the rate of re-occlusion, which occurred in 41% of non-responders and 22% of responders. Therefore, one can conclude that a differential increase in the rate of re-occlusion could significantly increase the likelihood of a stroke patient not responding clinically despite early recanalisation. Because the study did not measure cerebral blood flow or perform MRI imaging, one cannot compare the degree to which re-occlusion caused a poor clinical response in non-responders compared to the previously discussed confounding variable of a disproportionately large core infarct zone due to an ineffective collateral circulation. To complicate matters further, the study authors suspected that additional confounding variables also had to be responsible for the poor clinical response in non-responders -- and they theorised about the possibility of reperfusion injury. It is obviously difficult to know to what degree reperfusion injury is a confounding variable in TPA-for-stroke studies, and whether it is a significant confounding variable.

First of all, there is no doubt that TPA works extremely well in a minority of acute ischemic stroke patients. Those dramatic responders represent stroke patients who have an excellent clinical response within 24 hours of stroke onset, and who also have a very favorable clinical response at 3 months. The reason why they have such a good response is probably due to a multiplicity of interlinking pathophysiological mechanisms. MCA stroke patients who are dramatic responders are probably lucky enough to have an excellent collateral circulation, which can maintain viability of the ischemic penumbra brain tissue while they await the occurrence of thrombolytic-induced recanalisation. That means that those patients probably only develop a very small core infarct zone during the first few hours after stroke onset. The dramatic responders are presumably also lucky enough to experience either complete, or significant partial, recanalisation following timely TPA therapy. In other words, early partial/complete recanalisation is a necessary pre-requisite in most dramatic responders, but complete/partial recanalisation doesn't automatically guarantee a dramatic response. Interestingly, there is no evidence of a time-to-treatment interaction that differentiates dramatic responders from non-dramatic responders. I also presume that dramatic responders must be lucky enough not to experience vessel re-occlusion or a subsequent reperfusion injury.

One of my central themes is that a dramatic response to thrombolytic therapy is very dependent on a good collateral circulation, and that a favorable clinical response would not occur if recanalisation is optimised, but the collateral circulation is poor. Do I have EBM evidence to support my belief that the state of the collateral circulation is as important as the rate/degree of recanalisation in producing a favorable clinical stroke outcome? Consider this informative article by Kuchinski [13].

Kuchinski studied patients with occlusions in different arteries who then received either local or intravenous thrombolytic therapy. The arterial occlusions were either i) extracranial carotid artery stenosis at its bifurcation accompanied by embolism into the MCA (ICA & MCA); ii) terminal bifurcation of the carotid artery at the carotid siphon - carotid T-occlusion (CTO); iii) middle cerebral artery occlusion (MCA); iv) single or multiple distal occlusions (distal). He studied the degree/rate of TIMI recanalisation and also the state of the collateral circulation. Leptomeningeal collateral blood supply to the ischemic MCA territory was assessed on the basis of the number of branches filling retrogradely up to the M2 segment of the distal main stem, and collateralisation was classified as "good"' or "poor".

Note that a good collateral blood supply was only present in patients with a MCA occlusion (78% versus <46% in other types of occlusions) and that those MCA stroke patients had a 40% favorable outcome rate at 3 months (BI>90%). The patients who had the worst collateral circulation were patients with distal occlusions (only 30% had good collateralisation) and those patients had a poor clinical outcome (18% favorable outcome rate at 3 months). The reason for the poor collateral circulation is that those patients usually had multiple distal occlusions (presumably from multiple emboli from proximal clots) which occluded many of the distal vessels that could cause retrograde filling of the MCA territory. Patients who had carotid T-occlusions had a very poor recanalisation rate (only 27% versus 60% in MCA patients) and they had a poor clinical outcome (3% favorable outcome rate) despite a fair degree of collateralisation (46% rate of good collateralisation). Interestingly, and surprisingly, patients with ICA & MCA occlusions had the best favorable outcome rate at 3 months (47% favorable outcome rate) despite having suboptimal collateralisation and suboptimal recanalisation rates. That surprising result shows that there is not necessarily a clear-cut relationship between certain favorable pathophysiological phenomena and a favorable clinical outcome at 3 months (possibly because of other confounding variables -- varying time-to-treatment, varying re-occlusion rates, varying reperfusion injury rates).

When Kuchinski performed a multivariate analysis to determine which factors predicted a favorable clinical outcome, these were his results.

Note that the most powerful predictor of a favorable clinical outcome was good collateralisation (OR 5.9).

Unfortunately, all acute ischemic stroke patients are not sufficiently lucky to have a good collateral blood supply and a good recanalisation rate, and they are not likely to become dramatic responders. Many MCA stroke patients have an inadequate response to thrombolytic therapy -- presumably because i) they have a poor collateral circulation which results in a large volume core infarct zone; and/or ii) they have inadequate restoration of blood flow following incomplete recanalisation (which may be related to the site of occlusion [large proximal clot rather than a small distal clot] +/- type of clot [large vessel atherothrombotic clot rather than cardioembolic clot] +/- low perfusion pressure at the proximal end of clot); and/or iii) they have a higher incidence of re-occlusion; and/or iv) they have a higher incidence of reperfusion injury. On average, only approximately 40-50% of MCA stroke patients experience partial/complete recanalisation following intravenous thrombolytic therapy, and 30-50% of those patients experience re-occlusion. Therefore, if stroke interventionalists hope to improve the individual stroke patient's clinical response to thrombolytic therapy, additional therapeutic modalities will have to be utilised eg. mechanical or ultrasound-induced clot defragmentation to augment thrombolytic-induced clot lysis, and/or the use of anticoagulants/antiplatelet agents to prevent re-occlusion.

A frustrating conundrum for stroke researchers involved in RCTs is that one cannot accurately predict which patients in a sample population of stroke patients will become responders versus non-responders. Molina [14] developed a five-item prediction model for MCA occlusions (MOST score, which is based on five factors - the degree of recanalisation, site of occlusion, early ischemic CT scan changes, severity of clinical presentation, and presence of moderate hypertension), but the MOST score only has a 80% accuracy rate for MCA stroke patients. In his study, 25% of the patients had an excellent recovery despite no recanalisation, while 20% of the patients had no recovery despite early recanalisation. I personally think that one cannot hope to accurately determine what percentage of the sample population will become responders -- if one cannot accurately quantify the degree of efficacy of the collateral circulation, the rate/degree of recanalisation, the rate/degree of re-occlusion, and the rate/degree of reperfusion injury in that sample population of stroke patients. Those confounding variables plague TPA-for-stroke RCTs because some trialists often incorrectly assume that randomisation should result in an equal balance of those critically important prognostic variables between treated and placebo patients. However, simple common sense suggests that it is unlikely to happen in TPA-for-stroke RCTs that only recruit a few hundred patients.

I also think that stroke researchers would gain much more information from their raw data if they didn't limit their analysis to a comparison of single summary statistics. They should also use scattergrams to examine their raw data. For example, consider this table (which I previously described) from Neumann-Haefelin's paper.

The authors used this table to demonstrate the relationship between the degree of recanalisation and the short-term clinical outcome (NIHSS score at 7 days) and the long-term likelihood of a good stroke outcome (mRS≤2 at 90 days). I previously stated that the relationship was irregular. Part of the reason could be due to the fact that there is a striking degree of stroke severity heterogeneity in the stroke patients at baseline even though all the patients had a main stem MCA occlusion. Note the marked variability in their baseline NIHSS score, which varied >8-fold from 3 to 26. Those MCA stroke patients cannot be presumed to be a homogeneous population of stroke patients, and by simply using a single summary statistic (median) to analyse their results, one will be involuntarily distorting the studied relationships. I think that additional information could be gained if the authors also used scattergrams to supplement their analysis. An example of a scattergram is this sample from my personal analysis of the NINDS study [6].

If Neumann-Haefelin plotted the relationship between the likelihood of an excellent stroke outcome (mRS≤1) against the baseline NIHSS stroke severity score for each TIMI recanalisation subcategory (TIMI 0,1,2,3), one could possibly determine whether any established relationship between the degree of recanalisation and the likelihood of a favorable stroke outcome is consistent throughout the stroke severity range. It could well be that the baseline NIHSS score is a surrogate marker for the degree of efficacy of the collateral circulation when one is primarily dealing with MCA stroke patients, and a scattergram may provide more information than just relying on a single summary statistic eg. median baseline NIHSS score.

It should now be obvious to the reader that analysing TPA-for-stroke RCTs, with the intention of determining TPA's "true" treatment effect, is a complex task. Many confounding variables influence the clinical response to TPA therapy -- site of vessel occlusion; efficacy of the collateral circulation and consequent size of the central core infarct zone following clot occlusion; rate and degree of spontaneous or thrombolytic-induced recanalisation; rate and degree of re-occlusion following partial/complete recanalisation; rate and degree of reperfusion injury -- and unless one ensures that the treatment and placebo groups are balanced for those confounding variables, one cannot hope to accurately estimate TPA's "true" treatment effect.

I am extremely dismayed by the interpretative approach that the ATLANTIS-ECASS-NINDS rt-PA study group investigators recently adopted with respect to analysing the pooled data from the 2,775 patients from the major TPA-for-stroke RCTs. In the Lancet article [5] the ATLANTIS-ECASS-NINDS rt-PA study group investigators did not even bother to supply the unadjusted OR values for the pooled sample of 2,775 patients. Therefore, we have no unadjusted OR values to use as an unbiased "measuring stick" when judging the wisdom of choice of any proferred adjusted OR values. The ATLANTIS-ECASS-NINDS rt-PA study group investigators apparently based their "arbitrary" adjusted OR values on an "arbitrary" stroke outcome scoring system (Global Statistic) and an "arbitrary" choice of covariates that do not necessarily have a large prognostic influence on the rate of a favorable stroke outcome. They chose factors such as age, baseline glucose, baseline diastolic blood pressure and a history of previous hypertension as covariates worthy of inclusion in their adjusted OR estimations. I am not arguing that those factors have little power to influence TPA's treatment effects. I merely think that their power to influence TPA's treatment effect pales in significance when compared to pathophysiologically plausible factors that are much more likely to powerfully impact TPA's treatment effect -- site of vessel occlusion, degree of vessel occlusion, type of clot (fibrin-rich versus platelet-rich), degree of efficacy of the collateral circulation in minimizing the size of the core infarct zone prior to recanalisation, rate of recanalisation, speed of recanalisation, degree of recanalisation, rate and degree of re-occlusion following partial/complete recanalisation, and degree of reperfusion injury. I believe that those pathophysiological factors are the major factors that cause variability in response to thrombolytic therapy, and I believe that one cannot rationally expect randomisation to result in a reasonably equal balance of those confounding variables between treated and placebo patients, if each of the four TPA-for-stroke RCTs only enrolled a few hundred patients.

I think that Anthony Furlan is right when he made the following statement in his Feinberg lecture.

I think that the problem of a "confused vascular analysis" plagues the present-day interpretation of TPA-for-stroke RCTs, and that we will never be able to accurately determine TPA's "true" treatment effect unless stroke researchers change the way they design and analyse stroke trials.

In the Lancet article, the ATLANTIS-ECASS-NINDS rt-PA study group investigators delineated potential confounders that they thought could influence the correct interpretation of the pooled data from the 2,775 patients from the major TPA-for-stroke RCTs. This is their list.

I am sceptical of the significance of some of those confounders with respect to their power to have a powerful prognostic influence on TPA's "true" treatment effect. Do you really believe that a history of angina or CHF, or a history of previous atrial fibrillation has a powerful prognostic influence in TPA-for-stroke RCTs? I suspect that any "association" between those factors and TPA's treatment effect is purely coincidental, and not necessarily causal. In terms of prognostic power, those potential confounders must pale in significance when compared to unmeasured confounders that are more pathophysiologically likely to affect a stroke patient's response to TPA therapy eg. site of vessel occlusion, degree of vessel occlusion, type of clot (fibrin-rich versus platelet-rich), degree of efficacy of the collateral circulation in minimizing the size of the core infarct zone prior to recanalisation, rate of recanalisation, speed of recanalisation, degree of recanalisation, rate and degree of re-occlusion following partial/complete recanalisation and rate and degree of reperfusion injury.

A prime example of a poor quality interpretative analysis using questionable prognostic factors is the interpretative analysis by Brown [15]. The authors retrospectively studied the NINDS study's patient-level raw data. This is what the authors stated in their paper's preamble-:

The authors first defined a major neurological improvement (MNI) as an improvement in NIHSS score of 8 points or a NIHSS score of 0 at 24 hours. They then assessed whether there was a relationship between the MNI and a favorable stroke outcome at 3 months. They then used a multivariable logistic regression statistical technique to design a model that would predict MNI.

When discussing their final results, the authors stated-: "Overall, the model performed only moderately well in predicting MNI, suggesting the possibility that other variables not captured in our model may also affect outcome." Wow! Are you surprised that a statistical model that only used those baseline variables was only moderately effective in predicting MNI at 24 hours?

I personally think that this type of retrospective analysis has no scientific/educational value because it simply looks for an "association" between certain baseline clinical variables and a clinical endpoint without considering the pathophysiological plausibility, and without ensuring that other important confounding variables do not confound the analysis. In the "Confounding variables in TPA-for-stroke trials" section of this manuscript, I provided ample EBM evidence to refute their basic assumptions -- i) that one can predict which stroke patients will have a dramatic improvement within 24 hours; ii) that there is a linear corrleation between the rate of early improvement and a favorable clinical outcome at 3 months. One cannot rationally expect that certain baseline clinical variables will predict MNI at 24 hours when research evidence has amply demonstrated that we cannot reliably predict which TPA-treated patients will have a MNI at 24 hours, because there are so many likely confounding pathophysiological variables in play. A MNI at 24 hours is dependent on so many confounding pathophysiological variables that were not studied (and therefore unknown) in the NINDS study eg. site of vessel occlusion, type of clot, degree of vessel occlusion, efficacy of the collateral circulation and size of the core infarct zone, rate/degree of recanalisation, rate/degree of re-occlusion, and rate/degree of reperfusion injury. There is an irregular relationship between those variables and the likelihood of a clinical recovery at 24 hours, and at 3 months -- presumably because there is a varying degree of interplay between varying combinations of those confounding variables. That is the fundamental, and exasperating, problem that makes this type of superficial interpretative analysis near-worthless.

Another 'a priori' attitude that handicaps some stroke researchers is a fixed 'a priori' belief that there must be a time-to-treatment interaction even though the evidence from recent research studies does not really support a time-to-treatment interaction. Why do so many stroke researchers empirically believe that there must be a time-to-treatment interaction? I think that it could stem from knowledge of the EBM evidence from TPA-in-AMI studies. TPA has a significant time-to-treatment interaction in AMI patients, because it prevents infarction of cardiac muscle in inverse relationship to the time-to-treatment ("sooner-the-better"). Cardiac muscle can survive many hours of "no blood flow" before it becomes irreversibly infarcted, and timely TPA therapy will likely prevent irreversible infarction if recanalisation occurs and re-occlusion is prevented by heparin/antiplatelet therapy. However, and by contrast, brain tissue is much more sensitive to ischemia and it cannot survive a state of "no blood flow" beyond a few minutes. Brain tissue infarction is likely to occur within 60 minutes after a total clot occlusion of an intracerebral vessel, although the initial size of the core infarct zone likely depends on a major uncontrollable factor (efficacy of the collateral circulation). The maximum benefit that TPA therapy can realistically offer an anterior circulation stroke patient (if administered >60 minutes after stroke onset) is a minimisation of growth of the core infarct zone into the ischemic penumbra tissue zone -- presuming, of course, that partial/complete recanalisation occurs in a timely manner, and presuming that re-occlusion or reperfusion injury doesn't subsequently occur. Because TPA is essentially rescuing ischemic penumbra tissue, rather than reversing infarction, there may not be a significant time-to-treatment interaction in the first 3 hours after stroke onset, if the stroke patients have sufficient collateral circulation to maintain viability of the ischemic penumbra tissue while awaiting partial/complete recanalisation. My personal analysis of the NINDS study [6] suggests that TPA produced an unadjusted absolute risk difference of 12% in patients treated in <90 minutes, and 9% in patients treated between 91-180 minutes. That unadjusted estimate of TPA's treatment effect suggests the absence of a strong time-to-treatment interaction. The future will determine whether my personal interpretation of the NINDS study's raw data using an unadjusted analysis is more accurately reflective of reality than the ATLANTIS-ECASS-NINDS rt-PA study group investigators's recent adjusted analysis [5] that suggests a strong time-to-treatment interaction.

I personally think that the ATLANTIS-ECASS-NINDS rt-PA study group investigators should make the patient-level raw data from the 2,775 pooled patients publically available, so that other analysts can independently analyse the data using alternative methodological techniques. Science is always well-served when scientific information is openly shared among all members of the scientific community.

Can one accurately assess whether alternative methods of administering thrombolyic therapy is better than the standard intravenous approach -- an analysis of the EMS study.

In the April 2004 issue of the "Stroke" journal, there is a report on a pilot study of combined IA/IV TPA therapy - the Interventional Management of Stroke (IMS) Study [16].

Why do some stroke researchers think that there is a need to use combined IA/IV TPA therapy rather than standard intravenous TPA therapy? Consider what Scott Kasner wrote in an accompanying editorial [17] in the same issue of the journal (arbitrary selection of comments)-:

First of all, note that many stroke researchers generally agree that IV TPA therapy is not enough, and that combination therapy might offer a better outcome. I agree, but I don't know how the stroke researchers are going to be able to accurately quantify the difference in treatment efficacy between alternative treatment strategies if they have not developed a methodology of correcting for imbalances in important confounding variables (*noise). I believe that the noise level of TPA-for-stroke trials is so high, and the resultant signal/noise ratio so low, that one cannot be confident in a RCT's interpretative conclusion.

Sackett [18], the guru of evidence-based medicine, stated that confidence in a RCT's interpretative conclusion is directly related to the trial's signal/noise ratio and the square root of the sample size. This is Sackett's basic physiologic formulae-:

An increase in a RCT's noise (due to variations in confounding variables which can either inflate or deflate the treatment effect) can markedly decrease the RCT's signal/noise ratio and decrease one's confidence in the RCT's conclusions. How high is the noise level in the IMS study, which is actually a comparative study using historical controls and not a RCT?

In the IMS study, stroke patients, who had a baseline NIHSS score >10, received combined IV/IA therapy. The stroke patients first received IV-TPA therapy (0.6mg/kg to a maximum of 60mg, with 15% of the dose as a bolus over 1 minute with the remainder administered over 30 minutes) followed by immmediate cerebral angiography. If a thrombus was identified in an appropriate artery, IA TPA was administered. Sixy two patients out of 80 patients received combined IV/IA therapy. The researchers then assessed the efficacy of the combined IV/IA approach by comparing the IMS study group's results to the placebo and treatment arms of the NINDS study. This is how the researchers presented their results.

Note that there was a similar rate of favorable stroke outcome using the mRS≤1 stroke outcome scoring system between the IMS patients and the NINDS treated patients (30% versus 32%). The OR of a favorable stroke outcome slightly favored the EMS study group's patients if other stroke outcome scoring systems were used, and if the OR was also adjusted by baseline NIHSS, age, gender and time-to-treatment. The no/small difference in the rate of favorable stroke outcome means that the IMS study's signal is very small. I personally think that this type of comparison is scientifically invalid because there is no guarantee that the IMS study patients and the NINDS treated patients are balanced for important prognostic confounding variables -- site of occlusion, type of clot, efficacy of the collateral circulation and size of the core infarct zone, rates of TIMI 2/3 recanalisation, rates of re-occlusion, and rates of reperfusion injury -- and a significant imbalance in those confounding variables can create so much *noise that one cannot be confident in the study's comparative conclusion (because a high noise level would override the small signal and markedly reduce the signal/noise ratio).

Consider just one confounding variable -- site of arterial occlusion. Consider the locations of arterial occlusion in the IMS patients.

Note the marked variability in the sites of arterial occlusion. Surely, there is little chance-likelihood that there could be an equal balance of different sites of arterial occlusion between the IMS patients and the NINDS treated patients, considering the small size of the study samples. Also, note how few patients had a particular site of arterial occlusion. For instance, only 16 patients in the IMS study had a MI MCA occlusion (the most common site of occlusion). Based on the previous evidence that I presented from a number of MCA stroke studies, what is the chance-likelihood that those 16 patients had the same core infarct size (which is presumably inversely proportional to the efficacy of the collateral circulation), same recanalisation rates, same re-occlusion rates, and same reperfusion injury rates as the comparable group of MCA stroke patients from the NINDS study? If there is a significant imbalance in those confounding variables between the two MCA subgroups, the noise level would be so high that it would decrease the signal/noise ratio to such a low level that one could not be confident in the comparison's conclusion. The frustrating problem of confounding pathophysiological variables is significantly compounded when one also incorporates stroke patients with posterior circulation strokes and/or arterial dissections in the study group. I imagine that it would require a RCT of >10,000 stroke patients to be reasonably confident that randomisation would result in a reasonable balance of important confounding variables between standard intravenous TPA treated patients and combined IV/IA treated patients. The stroke outcome results of smaller RCTs would probably be so severely handicapped by noise from a likely imbalance in those confounding variables, that one could not be confident in the RCT's interpretative conclusion. However, there is no indication that the stroke research community is fully cognizant of this problem. In his editorial, Scott Kasner suggested that a randomised controlled trial of 400 patients would be required to compare standard IV versus combined IV/IA TPA therapy. I personally think that he markedly underestimates the sample size requirement because he does not appear to be sufficiently aware of the problem of confounding pathophysiological variables in stroke RCTs that recruit a heterogeneous population of stroke patients.

In conclusion, I think that an imbalance in confounding pathophysiological variables between treatment groups is the Achilles heel of TPA-for-stroke RCTs, and that stroke researchers will never be able to determine the "true" efficacy of standard intravenous TPA therapy, or combined IV/IA thrombolytic therapy, in acute ischemic stroke if they do not solve this particular problem. A more aggressive, and resource-intensive, thrombolytic approach may be warranted in patients with acute ischemic stroke, but it should be based on scientifically valid research evidence, which is dependent on RCTs having a high signal/noise ratio.

I think that James Penston is right, and I believe that one cannot ignore the potential methodological weaknesses of randomised controlled trials if one is searching for the scientific truth.

1. The National Institute of Neurological Disorders and Stroke rt -PA Stroke Study Group. Tissue Plasminogen Activator for Acute Ischemic Stroke. NEJM 1995;333:1581-1587.

2. Penston James. Fiction and Fantasy in Medical Research: The Large-scaled Randomised Trial. (London Press - ISBN 0954463617).

3. Fiona B. Young, BSc; Kennedy R. Lees, MD, FRCP; Christopher J. Weir, PhD; for the Glycine Antagonist in Neuroprotection (GAIN) International Trial Steering Committee and Investigators. Strengthening Acute Stroke Trials Through Optimal Use of Disability Endpoints. Stroke. 2003;34:2676-2680.

4. Marler, J R. MD. Tilley, B. C. PhD. Lu, M. PhD. Brott, T.G. MD. Lyden, P. C. MD. Grotta, J. C. MD. Boderick, J. P. MD. Levine, S. R. MD. Frankel, M.P. MD. Horowitz, S. H. MD. Haley, E. C. Jr. MD. Lewandowski, C. A. Kwiatkowski, T. P. MD. for the NINDS rt-PA Stroke Study Group *. Early Stroke Treatment Associated With Better Stroke Outcome: The NINDS rt-PA Stroke Study. Neurology 55 (11) 1649 - 1655, December 12, 2000

5. ATLANTIS-ECASS-NINDS rt-PA study group investigators. Association of Early Stroke Outcome with Early Stroke Treatment: Pooled Analysis of ATLANTIS, ECASS, and NINDS rt-PA stroke trials. Lancet 2004;363:768-74.

Available online at http://www.homestead.com/guidemaps/JeffMannEMguidemaps.html (in the soapbox section). If the url address hyperlink doesn't work -- use Google's search engine "Jeff Mann EM guidemaps".

7. Krause MS Howard KI. What randomisation does, and does not do. Journal of Clin Psychology. 59 (7): 751-766 (2003).

8. Chelsea S. Kidwell, MD; Jeffry R. Alger, PhD; Jeffrey L. Saver, MD. Beyond Mismatch: Evolving paradigms in Imaging the Ischemic Penumbra With Multimodal Magnetic Resonance Imaging. Stroke. 2003;34:2729-2735.

9. Robert A. Felberg, MD; Nicholas J. Okon, MD; Ashraf El-Mitwalli, MD; W. Scott Burgin, MD; James C. Grotta, MD; Andrei V. Alexandrov, MD. Early Dramatic Recovery During Intravenous Tissue Plasminogen Activator Infusion: Clinical Pattern and Outcome in Acute Middle Cerebral Artery Stroke. Stroke. 2002;33:1301-1307.

10. Tudor G. Jovin, MD; Howard Yonas, MD; James M. Gebel, MD; Emanuel Kanal, MD; Yue Fang Chang, PhD; Stephen Z. Grahovac, MD; Steven Goldstein, MD; Lawrence R. Wechsler, MD. The Cortical Ischemic Core and Not the Consistently Present Penumbra Is a Determinanant of Clinical Outcome in Acute Middle Cerebral Artery Occlusion. Stroke. 2003;34:2426-2435.

11. T. Neumann-Haefelin, MD; R. du Mesnil de Rochemont, MD; J.B. Fiebach, MD; A. Gass, MD; C. Nolte, MD; T. Kucinski, MD; J. Rother, MD; M. Siebler, MD; O.C. Singer, MD; K. Szabo, MD; A. Villringer , MD ; P.D. Schellinger , MD ; for the Kompetenznetz Schlaganfall Study Group. Effect of Incomplete (Spontaneous and Postthrombolytic) Recanalization After Middle Cerebral Artery Occlusion. Stroke. 2004;35:109-115.

12. Andrei V. Alexandrov, MD; Christiana E. Hall, MD; Lise A. Labiche, MD; Anne W. Wojner, PhD; James C. Grotta, MD. Ischemic Stunning of the Brain. Early Recanalisation Without Immediate Clinical Improvement in Acute Ischemic Stroke. Stroke. 2004;35:449-452.

13. Kuchinski T, Eckert B, Becker V, Kromer H, Hessen C, Gryzska U, Greitage HJ, Rother J, Zeumer H. Collateral Circulation is an Independent Radiological Predictor of Outcome After Thrombolysis in Acute Ischemic Stroke. Neuroradiology. 2003;45:11-18

14. Carlos A. Molina, MD; Andrei V. Alexandrov, MD; Andrew M. Demchuk, MD; Maher Saqqur, MD;Ken Uchino, MD; José Alvarez-Sabín, MD, PhD; for the CLOTBUST Investigators. Improving the Predictive Accuracy of Recanalization on Stroke Outcome in Patients Treated with Tissue Plasminogen Activator. Stroke. 2004;35:151-157.

15. Devin L. Brown, Karen C. Johnston, Douglas P. Wagner, and E. Clarke Haley, Jr. Predicting Major Neurological Improvement With Intravenous Recombinant Tissue Plasminogen Activator Treatment of Stroke. Stroke 2004;35:147 - 150.

16. The IMS Study Investigators. Combined Intravenous and Intra-Arterial Recanalization for Acute Ischemic Stroke: The Interventional Management of Stroke Study. Stroke 2004;35:904-911.

17. Scott Kasner. Editorial Comment—More Than One Way to Lyse a Clot. Stroke. 2004;35:911-912.

18. Sackett, David L. Why randomized controlled trials fail but needn't: 2. Failure to employ physiological statistics, or the only formula a clinician-trialist is ever likely to need (or understand!) CMAJ. 165(9):1226-1237. October 30, 2001.