Simply told, the story of tPA or ‘tissue-type plasminogen activator’ begins like this. In 1952, Danish biochemists isolated a fibrinolytic substance from normal animal tissues. It had been detected a few years before and was known to act through activation of plasminogen. In the mid-1970s, groups in New York and Leuven found that several malignant tumours also secrete a plasminogen activator, generally in proportion to their degree of invasive behaviour. The Leuven group obtained a highly productive melanoma cell line from an American patient and eventually purified it. The Belgian group also observed that the human plasminogen activator had a specific affinity for fibrin: in the absence of fibrin it had only a weak-activating effect on plasminogen, through competitive binding and ‘silencing’ of tPA by α2-antiplasmin. This selective action represented a great advantage—at least theoretically—over the non-specific plasminogen activators urokinase and streptokinase that were available at the time. Indeed in the early 1980s, animal experiments in Leuven and St Louis showed that tPA could dissolve thrombi of venous and arterial origin without systemic fibrinolysis. The first humans to be treated with tPA (from the melanoma cell line, in hindsight with low doses) were two patients in Rotterdam with thrombosis of renal and iliofemoral veins in the region of a transplanted kidney (Weimar et al., 1981); one of them survives to this day. The clinical experiment resulted from a chance conversation at a scientific meeting between a nephrologist from Rotterdam and a biochemist from the Leuven group. After pilot experiments in patients with coronary thrombosis had proved successful, a proper clinical trial was on the agenda but would require far greater quantities of the substance than could be extracted from the tumour cell line. In collaboration with the Leuven group, scientists at the Genentech corporation cloned the complementary DNA for the sequence of 527 amino acids, which in 1982 took a full year. The recombinant molecule (rtPA), first expressed in Escherichia coli and then in mammalian cells, proved indistinguishable from the naturally occurring substance, structurally as well as functionally. It was named alteplase, brand name Activase® (Genentech) or Actilyse® (Boehringer Ingelheim). Later a few similar drugs would be developed by other companies, but that plays no part in the current story. A preliminary study of rtPA in patients with myocardial infarction started in 1984 and was followed by several full-scale clinical trials, with doses of ∼100 mg, often depending on body weight, and with a variety of administration schedules. The drug was approved by the US Food and Drug Administration (FDA) for use in patients with myocardial infarction in 1987. Yet its superiority over streptokinase or other agents did not fully emerge until the early 1990s, when it became clear that two additional conditions should be met: two-thirds of the dose should be given over the first 30 min; and intravenous heparin administered at the same time.
Recombinant tPA for cerebral ischaemia
The preceding episode forms the core of Chapter 2 in the book by two authors from the USA—Justin A. Zivin, a neurologist turned research physiologist; and John Galbraith Simmons, a science writer (their very first chapter is predictably devoted to the stroke of an attractive girl, about to be married but now rushed to hospital). Their account of the synthesis of rtPA largely parallels an article by two key persons from the Leuven group (Collen and Lijnen, 2009), though the Belgian version is more complete and more coherent. The book really takes off where the paper stops, since its main subject is the use of rtPA for the treatment of ischaemic stroke. Chapter 3 recounts Zivin's animal experiments for this purpose. Having obtained rtPA from Genentech—not without some difficulty—he demonstrated that the drug could indeed lyse small blood clots in the carotid circulation of rabbits (Zivin et al., 1985). Readers can readily understand the elation this investigator must have felt at the success of his attempts, a mood that apparently prompted memories of Pasteur deflating the theory of spontaneous generation and of Rutherford concluding from his gold foil experiment that the positive charge of an atom is not randomly distributed but concentrated in a single nucleus. On the other hand, to think that a drug capable of lysing a clot in coronary arteries might do the same in cerebral arteries did not require a momentous leap of imagination. Sadly, after this high note, the remainder of the book is mostly drenched in Dr Zivin's frustration: at the extension of the pivotal National Institute of Neurological Disorders and Stroke (NINDS) clinical trial (see below) beyond its originally planned term (‘study leaders … sat on highly positive data’; Chapter 4); at the sceptical reaction of ‘Boston neurologists’ and their reluctance to put the results of the study into practice (‘neurology attracted brainy but not hands-on kinds of people’; Chapter 5); at the outright hostility of leading emergency physicians (through ‘global mistrust of pharmaceutical industries, exclusive focus on epidemiology and cultivated pessimism’; Chapter 6); and at the ‘anaemic marketing’ of rtPA for acute cerebral infarction by Genentech as well as at the inadequate reimbursement by most insurance companies (Chapter 7). Apparently some of this frustration was relieved by Dr Zivin acting as a witness in court on behalf of patients who might have been treated but were not (Chapter 8). The last part of the book returns to a more positive note. Chapter 9 documents the belated acceptance of the treatment in the USA and elsewhere, while the penultimate chapter tells how the two writers met and refers to reanalyses of the original NINDS trial supposedly showing that the results were even more positive than originally believed. In the epilogue (Chapter 11), the authors decry the persisting lack of public awareness about stroke and the small proportion of eligible patients actually receiving rtPA treatment—8% in a single urban region is the highest proportion receiving the drug in the USA. Oh yes, and what became of the attractive girl? Indeed, she largely recovered. The ambulance driver who initially took her not to the nearest hospital but to a hospital where they gave rtPA was at the wedding; and all those present had tears in their eyes.
Finally, and curiously, Dr Zivin adds a personal postscript in which he once more vents his displeasure at the conduct of the NINDS trial, from which he seems to have been disconnected from the half-point onwards.
The NINDS trial
The conduct of this landmark clinical trial has been somewhat unconventional (NINDS rt-PA Stroke Study Group, 1995). Zivin and Simmons hint at key issues and take sides but fail to give a proper explanation of what was at stake, which makes the reader feel as if in Plato's cave where one sees no real objects but only their shadows. Besides, is it not possible to explain the basic rules of clinical trials even in a book aimed at a general readership?
The NINDS trial started in 1991. It was a courageous undertaking, because it was already known that thrombolysis in ischaemic stroke might be followed by two different complications, both much more serious than when it was used for myocardial infarction. First, symptomatic haemorrhagic transformation of the infarct was likely to occur more often than without thrombolysis, eventually borne out by the failed clinical trials of streptokinase for ischaemic stroke, which ran more or less parallel with the NINDS study (Sandercock, 1995). Also, most trials in patients with myocardial infarction had shown a small but definite increase in the rate of cerebral haemorrhage after treatment. So, for rtPA the question was again in what proportion would the excess of haemorrhages lead to clinical deterioration and would these complications be outweighed by the benefits. Secondly, neurologists in Rotterdam had started an open pilot study with rtPA as early as 1987—with the haematologist who had been involved in the treatment of the two nephrological patients a few years before—but reported fatal brain swelling in the first 2 of the 10 patients they planned to include (Koudstaal et al., 1988). In retrospect this must have been a piece of bad luck, but the Dutch group was sufficiently discouraged to stop the project.
The American group was not. Remarkably, some of the principal investigators were not academic clinicians, but employees of the funding body. Had the study sponsor been a pharmaceutical industry, commercial interests might have led to serious bias. Now, the rather unique situation existed that the study was not only funded, but to some extent also devised and led, by a government agency, NINDS. The study was placebo controlled and double blind. Patients were eligible for inclusion if they had an ischaemic stroke with a clearly defined time of onset, could be treated within 3 h of onset and haemorrhage had been excluded by a CT scan of the brain. Contraindications were specified. The dose of rtPA given to the experimental group was 0.9 mg/kg (maximum 90 mg), 10% of the dose as an intravenous bolus injection, the remaining 90% as an infusion >1 h.
The pre-planned outcome event was an improvement on the National Institutes of Health Stroke Scale (NIHSS), 24 h after stroke onset (complete resolution of deficits or improvement by at least 4 points, the maximum deficit score being 42, in 11 domains). The trial was first analysed after some 2 years, when the planned number of approximately 280 patients had been randomized. The rtPA group was not statistically different from the placebo group in stroke severity scores after 24 h, but they were after 3 months, according to one of the three scales for functional independence on which the patients had also been rated at this time point.
The study leaders then decided to continue the trial for another 2 years and recruit an additional 320 patients. I have no inside information, but I can think of at least three sound methodological reasons for this decision that so much antagonized Dr Zivin: one mainly statistical, the second mainly medical and the third mainly practical. The first reason has to do with the quirks of chance. Given that the investigators had decided on a given outcome event, in this case improvement on the NIHSS impairment scale after 24 h, they had to stick to it and accept the ‘draw’ that actually emerged on that count with analysis of the first 291 patients. Even if the result had been favourable for rtPA at that point, the accepted rule of statistical significance would have implied a risk of up to 1 in 20 that the results were spurious. Looking for more favourable results according to another measure of outcome would have greatly increased that risk, more with each new test. Let me for a moment compare clinical trials with soccer games. According to convention, the game is decided by comparing the number of referee-approved goals scored by each team after 90 min. Now it sometimes happens that the result is a draw, despite one of the teams having been in possession of the ball for far >50% of the time, or more corner kicks than the other team, or more free kicks, shots on goal, throw-ins or whatever. The team that had the upper hand will therefore be frustrated at the draw, but they will not challenge the result. Otherwise the winner of a match result might often be changed afterwards, according to some or other new-found measure. So if the soccer match ending in a draw is a cup tie, it will have to be played again—at least in the UK; and that is what the NINDS investigators did.
The second argument for extending the trial is the choice of a more clinically relevant outcome event. Improvement in self-care after 24 h is difficult to assess in patients who are still in a hospital bed, with few exceptions. Therefore, one must rely on neurological examination, by using instruments that measure impairment, such as the NIHSS scale. This means that points are given for each possible deficit, representing such diverse functions as level of consciousness, language, visual fields, sensation and, separately for each limb, muscle power. The sum score of these arbitrary points was supposed to represent the severity of the overall deficit. But in the early 1990s, neurologists were gradually coming round to the idea that what actually counts for patients is the condition in which they come home, if at all; in other words, what they can do and to what extent they are dependent on others. As a consequence, outcome should be assessed not after 24 h but after 3 months, and the measure should represent their independence and performance at home and in society, not penalty points recorded at the bedside or on the neurologist's couch (van Gijn and Warlow, 1992). In short, this kind of end-point was chosen for the second part of the NINDS trial, as well as for the two parts combined. Almost, I should add, because in fact the new primary measure of outcome was a ‘global test’ that combined three different functional scales after 3 months with the old measure of improvement on the NIHSS, but now at 3 months. To amalgamate the four constituents into the ‘global test’, some kind of weighting was applied for each of them, a method that was not explained in the paper and that many people found difficult to understand. But as it turned out, all four submeasures on their own showed a statistically significant difference in favour of rtPA, for part 1 as well as for part 2 and for all patients treated within 90 min as well as for all those treated between 90 and 180 min after the onset of stroke. Yet the inclusion of the first part of the trial in the new analysis would cause some raising of eyebrows later on, since to some extent ‘the rules were changed after the game had ended’.
A third reason that may have led to a continuation of the trial may to some extent have been opportunistic, because it has to do with the hope that the FDA would eventually approve the drug as a regular prescription for stroke, as had earlier been the case for myocardial infarction. Of course, the stakes were even higher for the Genentech company, which supplied drugs for the study, but they had no say in the running of the trial. It is—and was—well known that if the effect of a new drug is not spectacularly life saving but more modest, even if significant in statistical terms, licensing authorities tend to remain on the safe side until the results are reproduced in a further trial. To what extent officials from the FDA were involved in the decision to add on the second part of the NINDS trial is not known to me, but Zivin and Simmons suggest they were; on this, as well as other points, the authors interviewed some 50 people for their book, mostly by telephone.
Are the sceptics castigated with reason?
In the end, the results of the NINDS trial were resoundingly positive (see below), but nevertheless—as the authors tirelessly recount—the reception was somewhat guarded. Some concerns were prompted by the concerns outlined above: first, a new and somewhat complicated measure of outcome had been chosen; secondly, there indeed was an excess of cerebral haemorrhages; and, thirdly, not everyone was convinced these were really two trials (after all, the patients were different, but the physicians and the geographical areas were the same).
The main issue that gave food for discussion, however, was how robust were the results. In other words, would the difference in favour of the rtPA group that the trial showed be the same in future patients? Take, for example, the modified Rankin scale. This is a 6-point functional scale, where 0 means no symptoms at all and 6 means death. The five grades in the middle represent increasing difficulty in self-care and general activity. The NINDS study group dichotomized the scale between grades 1 and 2 and regarded all other outcomes as ‘poor’. Other physicians might prefer to split the scale elsewhere, for example between 3 and 4, a division which roughly indicates the split between being able to return home or not. The proportion of patients in grade 3 or less at 3 months was 60% in the NINDS rtPA group, against 51% in the placebo group; both groups consisted of 312 patients. This represents an absolute difference of 9%; in other words, 11 patients had to be treated to allow a single patient to return home instead of being cared for in a nursing home (number needed to treat). The mistake the authors of the book make is to regard this number as the absolute truth from the very beginning. It is not—it is only a sample of the truth. Other samples might produce other results. In fact, a simple method—the 95% confidence interval—allows us to calculate what the absolute difference between the two groups might have been if the trial had been done a hundred times instead of just once. This arithmetic shows that 95 out of those hundred times the difference might be as big as 17% (number needed to treat: 6) or as small as 1% (number needed to treat: 100). The other five times, the difference would be even bigger or smaller than the numbers contained in this ‘95% confidence interval’. It may seem ridiculous that this principle should have to be explained in a scientific journal, but apparently it had not been made sufficiently clear to the authors of this book. In fact, they try to make ‘the truth’, as they perceive it, look even better by moving the goalposts, in referring to more sophisticated methods of analysing the outcome scales.
Sure, neurologists and emergency physicians were sceptical—and rightly so. Zivin and Cummings may be justified in their comment that in many hospitals neurologists were not used to jumping out of bed or even to being standby in the emergency department. But that is by far not the only explanation for the hesitant acceptance of thrombolysis for stroke. In some countries, for example The Netherlands, neurologists were already involved in the management of acute conditions such as head trauma, spinal trauma, meningitis and aneurysmal subarachnoid haemorrhage. Yet even there not all stroke specialists were immediately convinced (van Gijn, 1996), the more so since the NINDS study was published 2 months after an unconvincing European collaborative trial that had included patients even up to 6 h after the onset of stroke (Hacke et al., 1995). Only as the years went by was it confirmed that the benefits of rtPA are greater if treatment is given earlier (The ATLANTIS, ECASS, and NINDS rt-PA Study Group Investigators, 2004); that the ‘window of opportunity’ probably does not extend beyond 4.5 h (Hacke et al., 2008); and that in general use the proportion of symptomatic intracerebral haemorrhages does not exceed that in the NINDS trial (Wahlgren et al., 2007).
The historical view
From the history of medicine, it emerges that most new discoveries gained ground only slowly. Even if new insights were quickly accepted by peer scientists, often not the case, it could still be decades before these had trickled down to the rank and file of medical practitioners. Over the same period, numerous ‘new ideas’ would also pop up but be subsequently buried and forgotten. Hindsight is highly selective. Morgagni (1761) provided overwhelming evidence that changes in organs found after death were the actual cause of disease and not the consequence of humoral imbalances, but it took >40 years before his views took hold in revolutionary Paris and from there to the medical community at large. Over the same period, many physicians put their trust in the healing properties of magnetism.
Of course, today, the speed of communication is far greater and research methods have improved. But I can only shudder at the fundamental flaw in Zivin's accusation that hordes of patients have been unjustly denied the benefits of rtPA ‘by reliance on epidemiology at the expense of science’. At the end of the 18th century, it required a royal committee with Benjamin Franklin and Antoine Lavoisier to unmask the claims of Mesmer and other magnetizers by means of empirical testing (Kaptchuk et al., 2009). And today, empirical testing is still the linchpin of medical therapeutics, no matter how successful the animal experiments or how promising the first impressions in patients are. ‘Science’ in Zivin and Simmons' minds seems to mean extrapolation of pathophysiological reasoning to clinical management of patients. The history of medicine abounds with examples of fallibility in medical reasoning. By cherry picking the famous stories of smallpox vaccination, childbed fever and antisepsis, these masters of hindsight neglect the sad stories, not only of venesection and ovariotomy in old times, but also of leucotomy, oestrogens against miscarriage and oxygen for premature babies in the past century. There are very few spectacular, life-saving interventions for which a formal trial is superfluous—penicillin is an example. Almost always the potential effects of a treatment are modest, or they can be assessed only after some time. Had medicine, surgery included, not adopted empiricism in recent decades, we would still prescribe prolonged bed rest for slipped intervertebral discs and advise extra-intracranial bypass operations for patients with symptomatic occlusion of the internal carotid artery. Thankfully, the logic of medicine is nowadays put to the test. In this way rtPA has slowly but solidly earned its place in the neurologist's therapeutic armamentarium.
Finally, there is more to thrombolysis than rtPA. The authors correctly point out that the development of rtPA was a necessary but not sufficient condition for the development of thrombolysis as a treatment for ischaemic stroke. Without the advent and general availability of CT, it would not have been possible to make a reliable distinction between haemorrhage and ischaemia in patients with a sudden cerebral deficit. Without experimental work in animals, the old belief would have persisted that ischaemic neurons are all irretrievably damaged within a few minutes. And without modern telephones, it would not have been feasible to get patients to hospital within a few hours, or at least in numbers that allow proper clinical trials.
Refinements are still necessary in the use of rtPA for stroke. For one thing, there is insufficient information about the efficacy of the drug in very mild or very severe strokes, or in those >80 years of age; these questions are now being explored in the 3rd International Stroke Trial (IST-3), which is soon to be completed. Also, doses and routes of administration may have to be fine tuned. It is unlikely that thrombolysis with rtPA will remain the only effective method for unblocking cerebral arteries. In myocardial infarction, thrombolysis has been superseded by stenting (Nabel and Braunwald, 2012), but for stroke this seems an unlikely scenario. Mechanical embolectomy for patients in whom recanalization fails to occur after intravenous rtPA has been reported in several case series, but so far without appropriate controls. Newer and perhaps safer thrombolytic drugs may well be developed sooner or later. But the greatest challenge will be to raise the public awareness of stroke without frightening people into getting rushed to hospital with a harmless numb foot.