Thrombolysis for Stroke, Part II: Where the Decision Fails

The clock is what the pleadings argue about. A complaint describes a time interval, the defense describes a workflow, and the argument runs on minutes. Delay is a real failure mode and the easiest to reconstruct from a chart, which explains much of its prominence. It is not the failure mode the registries record most often. A thrombolysis decision can go wrong in two opposite directions, and a third failure sits underneath both and prevents the decision from being made at all. Which of the three a case actually involves determines which documents carry it, and the three do not share a theory of harm.

Part I covered the medicine, and the account of the drug, what it costs and how the window moved is there. In brief: thrombolysis is a clot-dissolving drug given for ischemic stroke, the kind caused by a clot blocking an artery rather than by bleeding, and its own principal harm is bleeding. What follows is the decision.

Four Outcomes, Not One Question

The decision resolves into four outcomes rather than one. A patient either received thrombolysis or did not; they either met the eligibility criteria in force on that date or did not. Two of the four combinations represent correct care. The other two are considered failures and are not variants of each other. Treating a patient who met no exclusions is the intended outcome; withholding treatment from a patient who met an exclusion is equally correct and is the single most common reason a chart shows no thrombolysis at all.

The two failure combinations run in opposite directions. In one, an eligible patient was never treated. In the other, an excluded patient was. The first is an omission; the harm is the lost benefit quantified in Part I, and the causation argument runs through what treatment would probably have achieved. The second is a commission; the harm is bleeding, and the causation argument runs through a risk that was documented as knowable before the drug was given. An expert who reaches for omission language in a commission case has misidentified the injury.

Delay is not a fifth category. It is a patient who moved from the eligible group to the ineligible one while the clock was running, which is why delay and omission tend to appear in the same file and why the pleaded theory should specify which of the two it is.

Beneath all four sits a prior question. Every one of them assumes the patient was recognized as having a stroke. Where that recognition failed, the eligibility question was never reached, and the decision under review is not a treatment decision at all.

What a Careful Decision Leaves in the Record

A carefully made thrombolysis decision leaves a specific trail, whichever way it went. What the trail omits is evidence in its own right.

• The last known well time, and its source. Part I established that the clock starts at the last moment the patient was observed at baseline rather than when symptoms were noticed, so the value is only as good as the person who supplied it. A chart recording an onset time without recording who gave it has documented a conclusion rather than a fact.
• The pre-treatment scan, and who read it. The non-contrast CT (a plain CT scan with no injected dye; a study sensitive to showing an acute bleed) is the gate to the one error someone rarely recovers from, and it is an interpretation rather than an observation. A record that reports the finding without naming the reader has dropped the variable that the evidence shows moves the error rate most.
• The deficit, described as well as scored. The NIHSS (the National Institutes of Health Stroke Scale, which is what a chart will actually call it) is a 15-item bedside examination scored from 0 to 42, where 0 is normal and higher is worse. Scores of 5 or below are conventionally called mild. It is a number derived from an examination, and the examination is what establishes whether the deficit was disabling for this particular patient, independent of the raw number.
• The exclusion list that was actually applied. Trial entry criteria, drug license criteria, and guideline contraindications are three different lists, and they disagree. Which one a clinician was working from determines whether a deviation was from a safety rule or from a recruitment rule.
• The blood pressure before the drug, and what was done about it. The threshold of 185/110 is a hard stop until the pressure is brought below it, and the response to a pressure above it is a decision in its own right. Whether antihypertensive treatment was given, withheld, or never considered is a separate documented act.
• The conversation, or the reason there was none. The American Academy of Neurology position statement on consent in acute stroke states that a patient with capacity, or an available surrogate, should be told the diagnosis and rationale, the prospects with and without treatment, and the risks including bleeding into the brain and angioedema (sudden swelling of the tongue and mouth), and that verbal consent or refusal should be documented by the treating physician.¹ Consent under time pressure is abbreviated, not waived. A chart with neither a conversation nor a reason has left the question open.

The Untreated Eligible Patient Is the Largest Category

Two denominators are in play, and conflating them is the most common briefing error in the area.

The population-wide figure is small: fewer than one in ten stroke patients receives the drug at all.² That number is dominated by people who arrived long after any window closed. It is not a failure rate and cannot be used as one, and an opinion that opens with it has not yet said anything about the case.

The litigation-relevant denominator is narrower: patients who reached the door in time with no documented contraindication. In a national quality-improvement registry, a quarter of those patients were not treated inside three hours.³ Two features of that finding matter more than the number. It fell by two thirds across the registry's eight years, so the applicable benchmark moves with the year of the event and an expert opining on a 2004 case cannot borrow a 2011 expectation. And the participating hospitals volunteered, which makes a quarter a floor rather than a central estimate.

The dominant documented reason for withholding treatment is that the deficit looked mild or was improving, and that is the reason the category is litigated: roughly three in ten of those patients left the hospital unable to walk unaided or not going home at all.⁴ Improvement itself predicted later worsening.⁵

One finding in that literature does more work than its size suggests. Among patients not treated because they looked too good, a visible arterial blockage was present in a fifth of them, some in a major vessel. The only thing that independently predicted the decision to withhold was the NIHSS.⁵ The decision tracked the number rather than the anatomy, and vessel imaging is where a reviewer sees that.

The counterweight is real and belongs in the same breath. PRISMS found no benefit from alteplase in deficits judged not disabling, alongside excess bleeding into the brain, but stopped at a third of its planned enrollment and says in its own conclusion that the early termination precludes definitive conclusions.⁶ ARAMIS, adequately powered, found dual antiplatelet therapy (two clot-inhibiting drugs given together) non-inferior to the drug in the same population.⁷ Between them the two trials move the decisive question off the score, though both used a score threshold to decide who could enroll. What they move it to is whether the deficit was disabling for this patient, which is a judgment made at the bedside and recorded, or not, in the chart.

Access is the other axis, and it is measurable. In the 2014 cross-section of a national Medicare analysis, treatment at a certified stroke center was associated with roughly twice the odds of receiving thrombolysis (adjusted odds ratio 1.96, 95 percent confidence interval 1.68 to 2.29), and arrival by ambulance with higher odds still (2.67, 2.38 to 3.00), while treatment at a rural hospital was associated with lower odds (0.88, 0.77 to 1.00, an interval whose upper bound touches unity and therefore cannot exclude no effect).² Where a patient landed changes what happened to them, which bears directly on whether a community hospital's conduct is measured against a community standard or an academic one.

The Treated Patient Who Should Not Have Been

This failure mode has a pedigree reaching back to the first phase 3 trial of the drug. ECASS I randomized 620 patients and excluded 109 of them from its target population, a rate the trial reported as 17.4 percent, because they had been enrolled despite major protocol violations.⁸ Sixty-six of the 109 breached the same criterion, an admission scan already showing early ischemic change across more than a third of the middle cerebral artery territory.⁹ A later review reports mortality among treated protocol violators of 33 percent, rising to 40 percent where the scan criterion was the one breached, against 14.6 percent in the treated target population.⁹ That breakdown and those mortality figures come from the review rather than from the trial report itself, and an opinion that leans on them should say so.

Modern registry data separates deviation from danger, and the distinction is the one most often collapsed. A Berlin registry analysis of 2,881 off-label treatments found no worse functional outcome than on-label treatment (common odds ratio 0.89, 0.78 to 1.02) and better outcomes than no treatment, with no signal for symptomatic hemorrhage or mortality.¹⁰ The criteria breached there were license criteria, meaning the terms of the drug's marketing approval: age above 80, NIHSS below 5 or above 25, treatment beyond 4.5 hours, prior stroke combined with diabetes. They were not the hard safety exclusions. Off-label and contraindicated are not synonyms, the reassuring result attaches only to the first, and an expert who extends it to the second is stating something the study does not support.

Three hard exclusions have retrievable evidence behind them.

Blood already on the scan. The catastrophic error is treating a hemorrhage (a bleed) as an infarct (tissue dying from a blocked artery), and the scan that prevents it is a reading rather than a bright line. In a nine-reader study of plain CT, one illustrative case of bleeding was missed by six of the nine, two board-certified radiologists and a neuroradiologist among them, and accuracy fell further among non-radiologist physicians.¹¹ That establishes that the reading fails at a measurable rate and that the rate rises as reader expertise falls. It does not establish how often patients with hemorrhage are actually treated, and it should not be offered for that. Who read the scan is therefore a fact worth having.

Blood pressure. The threshold of 185/110 is a line drawn across a continuous gradient rather than a cliff. The counterintuitive finding, from a registry of 11,080 treatments, is that the highest-risk group was not the highest-pressure group: it was patients with known hypertension whose blood-pressure medication had been withheld, at roughly double the bleeding risk.¹² Lowering pressure harder does reduce bleeding, without improving function at 90 days.¹³ What remains unanswered is the question a chart most often raises, which is whether a patient can safely be brought down from above the threshold in order to become eligible.

Anticoagulation. Guidelines advise against the drug within 48 hours of a dose of a direct oral anticoagulant (a newer blood thinner taken by mouth that needs no routine monitoring) unless clotting tests are normal. The largest registry found no excess bleeding in patients recorded as recent users.¹⁴ The caveat is load-bearing and cuts both ways: that study counted any dose within seven days, not within the 48 hours the guideline addresses, so it does not show that treating inside the window is safe, and it does not show that a deviation harmed anyone either.

Several validated scoring models, SEDAN among them, estimate bleeding risk before treatment from values already recorded on the chart: blood sugar, age, severity, blood pressure, and findings on the qualifying scan.^15,16,17 They stratify groups rather than predicting individuals, and none was built as an eligibility test, so a point total neither authorizes nor forbids treatment. What they establish for a reviewer is narrower and more useful. The information needed to characterize this patient's bleeding risk was on the chart before the drug was given, which makes was the risk knowable a checkable question rather than an argumentative one. Which model applies, and what it yields on a particular record, is work for an expert.

The professional standard on consent connects the two failure modes. Where the balance of risks and benefits is uncertain and the patient has neither capacity nor an available surrogate, the American Academy of Neurology statement directs the neurologist to adhere more closely to guideline-based inclusion and exclusion criteria.¹ The statement puts the point most explicitly of endovascular treatment rather than of thrombolysis, and in those terms: the further outside published criteria a patient's condition falls, the less justified emergency treatment becomes without patient or surrogate consent.¹ A record of treatment well outside the criteria and no record of a conversation is therefore two findings, not one.

Underneath Both, the Stroke That Was Never Called

Roughly one stroke in eleven is missed at first emergency department contact, and the rate climbs steeply, in some studies past half, when the presenting complaint is mild, non-specific, or transient.¹⁸ Those studies differ enormously from one another, so the pooled figure is a central tendency rather than a rate.

A two-hospital study puts the medico-legal consequence in the sharpest available terms. Of 465 ischemic strokes, 22 percent were initially misdiagnosed, and a third of those missed patients had presented while a treatment window was still open.¹⁹ The omission was a live lost opportunity, not a theoretical one. Nausea, vomiting and dizziness were the presentations that drove the misses.

Specialist involvement did not resolve it. At the academic hospital in that study, more than a third of the missed strokes had been seen by a neurologist in the emergency department before the diagnosis was missed, and nearly half of those had arrived inside the thrombolysis window.¹⁹ A record showing that neurology was consulted is not a record showing the question was answered.

Two mechanisms explain most of the pattern, and both are checkable. The first is anatomical: posterior circulation strokes (those in the arteries supplying the brainstem, cerebellum, and back of the brain) were missed at more than twice the rate of anterior ones (the carotid-fed territory covering most of the cerebral hemispheres).¹⁹ The second is instrumental, and Part I named it: gait and truncal ataxia (loss of balance of the trunk when sitting or standing), difficulty swallowing, and cough are all absent from the NIHSS, so a posterior stroke scores roughly half what an anterior one does in the same study.¹⁹ The NIHSS was never built as a screening test, and a low score is not evidence of a mild stroke.

Imaging carries its own limit. A normal head CT does not exclude an acute stroke: in patients presenting with dizziness and nothing else, plain CT caught about half the strokes that MRI found, and the bedside examination did not separate the two groups either.²⁰ That cuts against a defense that the examination was reassuring and equally against a plaintiff theory that the diagnosis was obvious. The timing data splits the difference too. In interhospital transfers, the scan is generally obtained on time; the hours accumulate afterward, between the images and the patient moving.²¹ Pleading a delayed scan is usually pleading the step that was performed on time.

The error also runs in the other direction, and the evidence there is counterintuitive enough to be worth stating plainly. Patients given the drug who turn out not to have had a stroke, known as stroke mimics, are common, running from under 2 percent at experienced centers to 16 percent in a trial population.^22,23,24,25 The most frequent are functional and conversion disorders (neurological symptoms produced without a structural lesion), complicated migraine, and seizure.²⁴ What the safety data show is that these patients do not bleed at the rate the intuition assumes: across every retrieved series they bled less than patients with confirmed strokes, not more.^22,23 Treating a mimic is a diagnostic error. On the available evidence it is usually not, by itself, the cause of an injury, and a theory that treats it as harm per se will not survive the literature.

Five Ways the Decision Fails, and How They Chain

• Eligible patient, never treated. The largest category by volume, and the one whose benchmark moves fastest with the year of the event.
• Excluded patient, treated anyway. The smallest by volume and the gravest per case, and the one where the risk was usually documented before the drug was given.
• Eligible patient, treated too late. Not a separate error so much as the first one caught partway, which is why it so often shares a file with it.
• Stroke never recognized. The failure that prevents the decision from being reached, and the one that leaves the least behind in the chart.
• Mimic treated as a stroke. A diagnostic error whose harm, on the retrieved evidence, is usually not the treatment.

These are not separate failure modes. They compound, and the compounding is what makes the resulting chart look defensible. A posterior circulation stroke presents with dizziness and vomiting, the two complaints most associated with a missed diagnosis. The NIHSS, which omits the findings that this stroke produces, returns a low score. The low score supports the impression that the deficit is mild, which is the single most common documented reason for not treating an eligible patient. The clock runs while an alternative diagnosis is worked up. By the time the stroke is recognized, the window has closed, and the chart now records that the patient was outside the window, which is true, and which reads as the reason rather than as the consequence. An unrecognized stroke has become an untreated eligible patient and then a delay, and only the delay is visible in the record.

What the Evidence Will and Will Not Carry

Three calibrations matter.

The omission evidence establishes its premise and stops short of causation. The registry finding that 28.3 percent of untreated mild-stroke patients did not go home is a discharge disposition, not a 90-day outcome, and the study includes no treated comparison group. It supports the proposition that a mild deficit is not a benign one. It does not support the proposition that treatment would have prevented those outcomes, and the authors' own conclusion goes no further than suggesting a trial in the population may be warranted.⁴ An opinion that cites it for causation has borrowed weight the paper does not carry.

The reassuring off-label data covers only part of the exclusion list. The Berlin registry's reassurance attaches to license criteria and says nothing about hemorrhage on the scan, uncontrolled pressure, or active anticoagulation, and it is observational, with the obvious confound that clinicians chose which off-label patients to treat.¹⁰ No retrieved source measures how often patients with hemorrhage already visible on the qualifying scan are thrombolysed in practice. The reader-error data establishes that the reading fails at a measurable rate; the frequency of the resulting treatment error has not been quantified, and an expert who supplies a number for it is supplying it from somewhere other than this literature.

The benchmark is dated, and it is dated twice over. Eligibility criteria moved across the period Part I described, and separately, the rate at which eligible patients went untreated fell by two thirds within the span of a single registry.³ An opinion about an omission in 2004 measured against 2011 practice, or about a 2011 omission measured against current practice, makes the same error as an opinion applying today's eligibility rules to an older chart. The date of the event fixes both the criteria and the expectation.

Reading a Thrombolysis Decision in Context

A defensible review of a case in which a thrombolysis decision is central, plaintiff or defense, begins by establishing which of the failure modes is actually alleged and then works through a small set of specific questions.

• Was the patient recognized as having a stroke at first contact, and if not, what were the presenting complaints and how long did recognition take?
• Was the last known well time recorded, and is its source identified in the chart, or was it inferred later from the ambulance record?
• Where treatment was withheld, does the record state the reason, and is that reason an exclusion criterion or a clinical impression that the deficit was mild?
• If the deficit was judged non-disabling, is that judgment documented against the patient's actual function and occupation, or only against the NIHSS?
• Was vascular imaging obtained before the decision, and did it show an occlusion?
• Where treatment was given, which exclusion list was checked, and does the chart show the pre-treatment blood pressure and what was done about it?
• Were the elements of a documented bleeding risk present on the chart before the drug was given?
• Is there a documented conversation with the patient or a surrogate, and does its content match how far outside the published criteria this patient fell?

Some cases that look strong on the complaint dissolve against a chart that records a deficit described rather than scored, an occlusion looked for and absent, and a documented refusal. Others that look strong on the defense collapse against an NIHSS used as though it were an eligibility test, a scan read by a clinician the record never names, and no note of any conversation at all.

The clock is the part of a thrombolysis case that reconstructs itself. Everything that decides the case sits in what the chart says was considered before it ran out.

References

Footnotes

1. Sattin JA, Chiong W, Bonnie RJ, et al. Consent issues in the management of acute ischemic stroke. Neurology. 2022;98(2):73-79. doi:10.1212/WNL.0000000000013040 ↩ ↩² ↩³

2. Meng T, Trickey AW, Harris AHS, et al. Lessons learned from the historical trends on thrombolysis use for acute ischemic stroke among Medicare beneficiaries in the United States. Front Neurol. 2022;13:827965. doi:10.3389/fneur.2022.827965 ↩ ↩²

3. Messé SR, Khatri P, Reeves MJ, et al. Why are acute ischemic stroke patients not receiving IV tPA? Results from a national registry. Neurology. 2016;87(15):1565-1574. doi:10.1212/WNL.0000000000003198 ↩ ↩²

4. Smith EE, Fonarow GC, Reeves MJ, et al. Outcomes in mild or rapidly improving stroke not treated with intravenous recombinant tissue-type plasminogen activator: findings from Get With The Guidelines-Stroke. Stroke. 2011;42(11):3110-3115. doi:10.1161/STROKEAHA.111.613208 ↩ ↩²

5. Smith EE, Abdullah AR, Petkovska I, Rosenthal E, Koroshetz WJ, Schwamm LH. Poor outcomes in patients who do not receive intravenous tissue plasminogen activator because of mild or improving ischemic stroke. Stroke. 2005;36(11):2497-2499. doi:10.1161/01.STR.0000185798.78817.f3 ↩ ↩²

6. Khatri P, Kleindorfer DO, Devlin T, et al. Effect of alteplase vs aspirin on functional outcome for patients with acute ischemic stroke and minor nondisabling neurologic deficits: the PRISMS randomized clinical trial. JAMA. 2018;320(2):156-166. doi:10.1001/jama.2018.8496 ↩

7. Chen D, Cui Y, Zhou Z, et al. Dual antiplatelet therapy vs alteplase for patients with minor nondisabling acute ischemic stroke: the ARAMIS randomized clinical trial. JAMA. 2023;329(24):2135-2144. doi:10.1001/jama.2023.7827 ↩

8. Hacke W, Kaste M, Fieschi C, et al. Intravenous thrombolysis with recombinant tissue plasminogen activator for acute hemispheric stroke: the European Cooperative Acute Stroke Study (ECASS). JAMA. 1995;274(13):1017-1025. doi:10.1001/jama.274.13.1017 ↩

9. Lewandowski C, Lotfipour S. Lessons learned from multicenter randomized clinical trials with intravenous thrombolysis for acute ischemic stroke. J Stroke Cerebrovasc Dis. 2002;11(3-4):125-136. doi:10.1053/jscd.2002.129961 ↩ ↩²

10. Mohrbach D, Wagner L, Nolte CH, et al. Off-label thrombolysis in acute ischemic stroke patients: frequencies and outcome compared to on-label and no treatment. Eur Stroke J. 2025;10(4):1328-1336. doi:10.1177/23969873251329875 ↩ ↩²

11. Yun TJ, Choi JW, Han M, et al. Deep learning based automatic detection algorithm for acute intracranial haemorrhage: a pivotal randomized clinical trial. NPJ Digit Med. 2023;6(1):61. doi:10.1038/s41746-023-00798-8 ↩

12. Ahmed N, Wahlgren N, Brainin M, et al. Relationship of blood pressure, antihypertensive therapy, and outcome in ischemic stroke treated with intravenous thrombolysis: retrospective analysis from Safe Implementation of Thrombolysis in Stroke-International Stroke Thrombolysis Register (SITS-ISTR). Stroke. 2009;40(7):2442-2449. doi:10.1161/STROKEAHA.109.548602 ↩

13. Anderson CS, Huang Y, Lindley RI, et al. Intensive blood pressure reduction with intravenous thrombolysis therapy for acute ischaemic stroke (ENCHANTED): an international, randomised, open-label, blinded-endpoint, phase 3 trial. Lancet. 2019;393(10174):877-888. doi:10.1016/S0140-6736(19)30038-8 ↩

14. Kam W, Holmes DN, Hernandez AF, et al. Association of recent use of non-vitamin K antagonist oral anticoagulants with intracranial hemorrhage among patients with acute ischemic stroke treated with alteplase. JAMA. 2022;327(8):760. doi:10.1001/jama.2022.0948 ↩

15. Menon BK, Saver JL, Prabhakaran S, et al. Risk score for intracranial hemorrhage in patients with acute ischemic stroke treated with intravenous tissue-type plasminogen activator. Stroke. 2012;43(9):2293-2299. doi:10.1161/STROKEAHA.112.660415 ↩

16. Strbian D, Engelter S, Michel P, et al. Symptomatic intracranial hemorrhage after stroke thrombolysis: the SEDAN score. Ann Neurol. 2012;71(5):634-641. doi:10.1002/ana.23546 ↩

17. Lou M, Safdar A, Mehdiratta M, et al. The HAT score: a simple grading scale for predicting hemorrhage after thrombolysis. Neurology. 2008;71(18):1417-1423. doi:10.1212/01.WNL.0000330297.58334.DD ↩

18. Tarnutzer AA, Lee SH, Robinson KA, Wang Z, Edlow JA, Newman-Toker DE. ED misdiagnosis of cerebrovascular events in the era of modern neuroimaging: a meta-analysis. Neurology. 2017;88(15):1468-1477. doi:10.1212/WNL.0000000000003814 ↩

19. Arch AE, Weisman DC, Coca S, Nystrom KV, Wira CR 3rd, Schindler JL. Missed ischemic stroke diagnosis in the emergency department by emergency medicine and neurology services. Stroke. 2016;47(3):668-673. doi:10.1161/STROKEAHA.115.010613 ↩ ↩² ↩³ ↩⁴

20. Hu X, Liu S, Wu X, et al. Prevalence of stroke and diagnostic performance of emergency MRI in acute isolated dizziness. Ann Clin Transl Neurol. 2025;12(12):2514-2522. doi:10.1002/acn3.70195 ↩

21. Royan R, Stamm B, Giurcanu M, Messe SR, Jauch EC, Prabhakaran S. Emergency department process times and door-in-door-out times in interhospital transfers after acute ischemic stroke. JAMA Netw Open. 2024;7(9):e2431183. doi:10.1001/jamanetworkopen.2024.31183 ↩

22. Tsivgoulis G, Zand R, Katsanos AH, et al. Safety of intravenous thrombolysis in stroke mimics: prospective 5-year study and comprehensive meta-analysis. Stroke. 2015;46(5):1281-1287. doi:10.1161/STROKEAHA.115.009012 ↩ ↩²

23. Zinkstok SM, Engelter ST, Gensicke H, et al. Safety of thrombolysis in stroke mimics: results from a multicenter cohort study. Stroke. 2013;44(4):1080-1084. doi:10.1161/STROKEAHA.111.000126 ↩ ↩²

24. Tsivgoulis G, Alexandrov AV, Chang J, et al. Safety and outcomes of intravenous thrombolysis in stroke mimics: a 6-year, single-care center study and a pooled analysis of reported series. Stroke. 2011;42(6):1771-1774. doi:10.1161/STROKEAHA.110.609339 ↩ ↩²

25. Kvistad CE, Novotny V, Naess H, et al. Safety and predictors of stroke mimics in The Norwegian Tenecteplase Stroke Trial (NOR-TEST). Int J Stroke. 2019;14(5):508-516. doi:10.1177/1747493018790015 ↩