Missed Subarachnoid Hemorrhage: The Clock the Record Has to Establish

The contest in a missed subarachnoid hemorrhage case rarely reflects what the scan showed. Both sides usually agree on that, and the argument is about what a negative scan was worth. Aneurysmal subarachnoid hemorrhage is bleeding into the fluid-filled space around the brain from a ruptured aneurysm, a weak outpouching of an artery wall that can leak or burst. The pathway the 2023 American Heart Association and American Stroke Association guideline sets out runs on two tests, both scored against a clock that starts the instant the headache began. The guideline splits its recommendation on that clock: a head CT (an X-ray study that rebuilds the brain in cross-sections and shows fresh blood as bright white) obtained inside six hours of onset is treated differently from one obtained later, and a lumbar puncture (a needle placed between two vertebrae in the lower back to draw off spinal fluid) is directed after a negative scan in the later presentations.¹ That interval is nowhere in the images and nowhere on the laboratory report. It exists only if someone wrote down when the headache started.

What a Diagnosis Buys

Intracranial aneurysms occur at the branching points of arteries at the brain's base, and their distribution is uneven. Most ruptures happen at specific sites, mainly the anterior communicating artery, the posterior communicating artery's origin, and the middle cerebral artery's bifurcation. When an aneurysm ruptures, arterial blood is forced into the space surrounding the brain. The primary danger to a patient who survives a rupture is the risk of re-bleeding from the same aneurysm if no intervention is taken.

Medical management does not close an aneurysm; instead, it is secured either by placing a metal clip across its neck through an opening in the skull or by filling it with platinum coils inserted via a catheter threaded from an artery in the groin. That is what a diagnosis buys, and it is the whole of the causation question in these cases. A patient whose hemorrhage is recognized enters the pathway to securing; a patient sent home with a diagnosis of migraine does not. In a cohort of 482 patients followed from the onset of the illness, 56 were not correctly diagnosed at first contact, and 22 of those 56 suffered a neurological complication before the correct diagnosis was made, 12 of them a second bleed.²

What no study in this literature does is randomize earlier diagnosis against later. Every counterfactual about what recognition would have bought a particular patient is therefore modeled, and the modeling assumptions are where the argument actually happens.

The History Does Not Discriminate

Two misdiagnosis figures circulate here and are routinely fused into a single range. They are not two estimates of one quantity. The higher comes from the cohort above: 56 of 482 patients, 12 percent, missed at first contact with any medical professional, where failure to obtain a CT was the most common error and migraine or tension headache the most common substitute diagnosis.² The lower rate comes from every admission for non-traumatic subarachnoid hemorrhage across every Ontario hospital over four years, counting a miss whenever a related emergency department visit appeared in the two weeks before admission. That rate was 5.4 percent. The same study found a miss roughly two and a half times likelier at low triage acuity, the urgency level a nurse assigns at the door.³ Two disease definitions, two denominators, two ways of finding the cases. Quoted as a range, the pair describes nothing.

The teaching is that the discriminating question is whether the headache came on suddenly, and the phrase carrying that teaching is the worst headache of the patient's life. Neither performs as advertised. Of 3,132 patients scanned across eleven Canadian teaching hospitals to rule this out, 82.1 percent described the worst headache of their life, and 7.7 percent had the hemorrhage.⁴ A phrase offered by four of every five patients under investigation cannot pick out one in thirteen. The systematic review that anchors this literature reviewed 22 studies and found no single feature of the history or examination adequate to rule the diagnosis in or out; the strongest were neck pain and neck stiffness, and both shift the odds rather than settling them.⁵

The patient who looks well is the patient who gets missed. Among the patients who were mentally clear at first contact in that cohort, 19 percent were misdiagnosed, compared with 12 percent overall. Misdiagnosis among them was associated with worse outcomes at three months and at a year.²

What exists in place of a discriminating symptom is a rule-out instrument. The Ottawa Subarachnoid Hemorrhage Rule covers alert patients older than fifteen with a new severe headache peaking within one hour, and calls for further investigation on any of six criteria: age 40 or older, neck pain or stiffness, witnessed loss of consciousness, onset during exertion, instantly peaking pain, or limited neck flexion on examination.¹ In the cohort that produced it, the rule caught every hemorrhage and cleared 15.3 percent of the patients who did not have one.⁶ The 2023 guideline grades its use may be reasonable, the weakest of the three affirmative rungs the document uses.¹ The rule tells a clinician whom not to investigate, and it clears very few people.

A Negative Test Reports an Interval

This literature does not describe a curve that decays gently by the hour. It describes a step at six hours, and the step is steep.

The cohort that produced the six-hour rule enrolled neurologically intact adults whose headache peaked within an hour, across eleven Canadian emergency departments. Of the 953 patients scanned inside six hours, the scan found all 121 hemorrhages. Of the 2,179 scanned later, it missed 17 of 119, a sensitivity of 85.7 percent, which is the share of diseased patients a test catches.⁴ Pooling the studies on both sides of that boundary produces the same shape: essentially perfect inside the window, roughly 89 percent beyond it.⁵

Nothing in that is a twelve-hour or a twenty-four-hour figure, and there is a reason. The most recent systematic evaluation looked for accuracy at under twelve hours and at under twenty-four, found three studies for each band, and published no pooled estimate for either.⁷ The intermediate numbers that circulate in teaching are not measurements: a precise miss rate attached to an eighteen-hour negative scan is a figure this literature has not produced, in whichever direction it is offered.

The guideline reads the same data as two populations rather than as one curve. Beyond six hours from onset, or with a new neurological deficit, a head CT and, if it is negative, a lumbar puncture should be performed. Inside six hours without a deficit, a head CT on a high-quality scanner interpreted by a board-certified neuroradiologist, one with subspecialty training in imaging the nervous system, is reasonable, which is permissive rather than mandatory. The guideline then attaches the limit that matters most: none of that applies to patients who present atypically, with neck pain, fainting, a seizure, or a new deficit, and the absence of a classic presentation should still prompt imaging and workup.¹ MRI sits outside that pathway, which is a fact about the pathway rather than about the scanner: blood-sensitive sequences detect bleeding at least as well as CT early and better after the first days. What a negative MRI lacks is a place in the accuracy literature the six-hour boundary rests on.

When the scan is negative and suspicion holds, the second test is spinal fluid, and it carries a clock of its own. The finding that does the work is xanthochromia, a yellow tint produced when hemoglobin released from red cells is converted to bilirubin, the pigment that also colors a fading bruise. Because the pigment has to be manufactured, it is not present at the moment of the bleed. It takes up to twelve hours to develop.⁵ A needle at three hours can find red cells but cannot attribute them.

The second clock carries a second condition, and it is the one American records usually fail to capture. The tint can be assessed through spectrophotometry, which measures how much light the fluid absorbs at bilirubin's wavelength, or it can be visually estimated. In a 2001 survey of 1,952 hospital laboratories, 99.7 percent reported visual inspection, and no comparable resurvey has been published since.⁸ The substitution is not free: in the largest prospective cohort of patients punctured after a normal scan, visual inspection alone identified seven of the fifteen aneurysmal hemorrhages.⁹ Xanthochromia is the specific finding in the fluid, not the sensitive one.

The other question in the fluid is whether blood was present in the tube beforehand or was introduced by the needle. The heuristic for distinguishing them is a decline in the red-cell count from the first tube to the last. The group that built the operative rule declined to test that clearance at all, for a structural reason: a traumatic tap and a subarachnoid hemorrhage are independent events and can coexist.⁹ What replaced the heuristic is a two-element rule. Fewer than 2,000 red cells per microliter in the last tube drawn, together with absent xanthochromia, excluded aneurysmal hemorrhage in that cohort with no misses. Neither element carries the rule alone: of the fifteen hemorrhages the cohort diagnosed by lumbar puncture, seven were caught by the xanthochromia and eight by the cell count.⁹

The Sentinel Headache Is a Prior Encounter

A sentinel headache is a smaller leak from the same aneurysm, days to weeks before the major rupture. The defensible figure is a range rather than a point: across the good-quality studies, 10 to 43 percent of patients with an aneurysmal hemorrhage report one, and that is the range the 2023 guideline itself prints.^10,1 The two ends are two study designs, and an opinion quoting either as the incidence has picked a design without saying so.

The standard objection is recall bias, the tendency to remember an event differently once its outcome is known, and it runs the wrong way. Two case-control studies inside that review put apparent sentinel headaches in patients without an aneurysmal bleed at around 5 percent, loose enough to run as high as 16, which bounds how much of the signal can be memory.¹⁰

For a record review the operative fact is not how common the event is but what happened when one reached a clinician. The sentinel visit is the encounter at which a normal scan and a normal puncture can both be true and the aneurysm still be sitting there, because neither test looks for it. The diagnostic sequence the guideline sets out runs from the scan to the puncture and only then to imaging of the vessels themselves.¹ An opinion reading a normal sentinel workup as proof the diagnosis was available has to name the test performed that day that would have shown the aneurysm. An opinion reading the same workup as complete has to answer the vascular imaging nobody ordered.

What the Record Has to Establish

A chart has to establish seven things before any opinion about the workup can rest on it. Where one is missing, the absence is itself an analytical fact.

• The onset timestamp. Not the arrival time and not the triage time, but the hour the headache began and the interval to peak, with who supplied that history and when it was written down. Every threshold here runs from that moment: six hours for the scan, twelve for the fluid, one hour to peak for the decision rule.¹ A history of present illness reading "headache, four hours" has recorded a duration reported at an unstated moment, not an onset time.
• The interval from onset to scan, and which scan time it was measured to. A chart does not carry one scan time. It carries an order time, a time the patient reached the scanner, an acquisition time in the imaging metadata, a preliminary-read time, and a final-report time, and in a busy department those sit hours apart. Any interval offered against the six-hour boundary has to name the endpoint it was computed to. A negative scan whose interval cannot be reconstructed at all cannot be assigned a sensitivity from any study in this literature.
• The reader and the equipment. The guideline conditions its permissive sub-six-hour recommendation on a high-quality scanner and a board-certified neuroradiologist.¹ Neither half is in the radiology report. The scanner sits in the imaging metadata and the department's equipment inventory; the reader's credential sits in the signature block and the hospital's privileging file. Charts also carry a preliminary read, on which the decision to admit or discharge is usually made, and a final signed report, sometimes issued the next day. What that record contains and how to request it is its own subject.
• The lumbar-puncture decision and its stated reasoning. Whether a puncture was recommended, performed, declined by the patient, deferred, or replaced by CT angiography (a scan timed to a contrast injection so the arteries themselves are visible); what reasoning sits beside that decision; and, where angiography was substituted, what followed a negative or inconclusive result.
• The fluid result as the laboratory actually produced it. Whether xanthochromia was assessed by eye or by instrument, what the red-cell count was in the last tube collected rather than the first, and how many hours after onset the needle went in. An American laboratory report prints a result and not a method, so the method usually has to be obtained from the laboratory's procedure manual rather than from the chart.
• The prior encounter. Whether anyone went looking for a visit in the preceding weeks, at that facility or another, for a headache that resolved. In a sentinel-headache case this is the highest-yield document in the file, and it is often held by a facility other than the one that produced the record under review.
• The discharge instructions and return criteria. What the patient was told the headache probably was, whether the instruction named the events warranting a same-day return rather than a clinic appointment, and whether a follow-up interval was named. In a disease whose second presentation is the rupture, this is the only part of the index visit that keeps operating after the patient leaves the building.

What the Evidence Will and Will Not Carry

Three calibrations matter, and each stops somewhere an opinion has to stop with it.

The near-ceiling sensitivity inside six hours is real, it is conditional, and the conditions are separately checkable in a chart. In the recent systematic evaluation, a third of the included studies did not state who read the scan at all, and most were at high risk of being distorted by their design.⁷ The guideline nonetheless conditions its recommendation on a board-certified neuroradiologist, a stricter reader than the evidence base that produced the number. An opinion importing the pooled figure into a community department has imported a reader standard the record may not meet. The setting is measurable in the other direction too, though less firmly than it is usually quoted: the one population-scale study put the adjusted odds of a miss about twice as high at non-teaching hospitals, with neither case volume nor scanner availability explaining the gap, but the finding sits close enough to the edge that no difference at all remains consistent with the data.³

The pathway substituting CT angiography for the puncture rests on less than it is asked to carry. The guideline records that no study has evaluated angiography against puncture as the next step after a normal or non-diagnostic scan; that angiography does not evaluate for subarachnoid blood at all, only for disease of the vessels; and that its sensitivity for ruptured aneurysms under 3 millimeters has been estimated at 61 percent, which is to say that roughly two of every five ruptured aneurysms of that size are not seen.¹ A record in which angiography replaced the puncture has substituted one question for another, and that substitution should appear in the chart as a decision.

Population yield and individual causation are different questions, and each side has a characteristic way of confusing them. Whether recognizing every sentinel headache would move outcomes across a national population is a health-services question. Whether recognition would have changed the course of one patient with one aneurysm at one point in its history is a clinical one, answered against that patient's imaging, that aneurysm's location and size, and the interval between the missed presentation and the second bleed. An opinion for the plaintiff treating a population-scale benefit as the measure of one patient's lost chance has used the wrong denominator. An opinion for the defense citing the same population figure to argue that recognition would not have changed this patient's course has used the identical wrong denominator.

An opinion in this disease is only as durable as the timestamps under it, and the timestamps are the first thing the other side will test.

Where the Workup Goes Wrong

The cases that reach review fail in a small number of recurring ways.

• The onset time is never fixed. The complaint is recorded as a duration rather than a time, the interval to peak is never asked, and no note in the visit states the hour the headache began. Every threshold downstream becomes unenforceable, in both directions.
• A negative scan is read as a negative test. The scan is negative, the interval is past six hours, and nothing in the chart marks the difference between a test that found every hemorrhage inside the window and one that missed roughly one in seven outside it.⁴
• The well-appearing patient is triaged low and stays low. Low triage acuity carried roughly two and a half times the odds of a miss in the population study, and a fifth of the patients who were mentally clear at first contact were misdiagnosed in the cohort study.^3,2
• The sentinel encounter is never retrieved. The index visit is reconstructed in detail, and the visit three weeks earlier, at which a headache was called something else, is never requested.
• The puncture is performed before the marker exists. A needle at three hours can find red cells but cannot attribute them, because the pigment that does the attributing has not yet formed.⁵
• Xanthochromia is read by eye and reported as absent. Visual inspection alone found seven of the fifteen aneurysmal hemorrhages in the largest prospective cohort, and nearly every American laboratory surveyed reported using it.^9,8
• Tube-to-tube clearance is used to dismiss the blood. A traumatic tap and a subarachnoid hemorrhage are independent events that can coexist, which is why the group that built the operative rule declined to test clearance at all.⁹

These are not separate failure modes. They compound. An onset time that was never fixed produces a scan whose interval cannot be scored; a scan whose interval cannot be scored produces a decision to stop that has no stated basis; a decision to stop with no stated basis produces a discharge with no return criteria; and a discharge with no return criteria produces a second presentation whose delay the patient is left to explain.

Reading a Missed Subarachnoid Hemorrhage Case in Context

A defensible review of a case in which a missed subarachnoid hemorrhage is central, plaintiff or defense, works through a small set of specific questions put to the record rather than to the parties.

• Which of the six Ottawa criteria appear anywhere in the record, and which were neither asked nor recorded?
• What triage acuity level was assigned at the door, by whom, and was it revised before the patient left?
• Which scan timestamps does the record carry, and how far apart do the order, the acquisition, the preliminary read, and the final signed report sit?
• On what scanner was the study performed, who signed the final report, and does anything in the file establish either?
• If a puncture was performed, did the needle go in before the twelfth hour, does the report state how xanthochromia was assessed, and if it does not, what does the laboratory's procedure manual say?
• If CT angiography was substituted for the puncture, what was it recorded as ruling out, and what followed a negative or inconclusive result?
• Was this a teaching department or a community one, and does the opinion under review carry figures measured in the other setting?
• Between the two presentations, what changed; and after the second, was the aneurysm secured, by what route, and how long after arrival?

Most of these cases resolve once those answers are laid side by side. Some that look strong on the complaint dissolve against a record that fixes the onset time, scans inside the window, documents the reasoning for stopping, and names return criteria the patient did not act on. Others that look strong on the defense collapse against a chart in which a normal-appearing patient was triaged low, scanned at hour nineteen, discharged without a puncture, and never asked about the headache three weeks earlier that had already been called a migraine somewhere else.

The film shows blood or it does not. What the film is worth is decided by a clock the images do not carry, and that clock runs in the history.

References

Footnotes

1. Hoh BL, Ko NU, Amin-Hanjani S, et al. 2023 Guideline for the Management of Patients With Aneurysmal Subarachnoid Hemorrhage: A Guideline From the American Heart Association/American Stroke Association. Stroke. 2023;54(7):e314–e370. doi:10.1161/STR.0000000000000436 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

2. Kowalski RG, Claassen J, Kreiter KT, et al. Initial misdiagnosis and outcome after subarachnoid hemorrhage. JAMA. 2004;291(7):866–869. doi:10.1001/jama.291.7.866 ↩ ↩² ↩³ ↩⁴

3. Vermeulen MJ, Schull MJ. Missed diagnosis of subarachnoid hemorrhage in the emergency department. Stroke. 2007;38(4):1216–1221. doi:10.1161/01.STR.0000259661.05525.9a ↩ ↩² ↩³

4. Perry JJ, Stiell IG, Sivilotti MLA, et al. Sensitivity of computed tomography performed within six hours of onset of headache for diagnosis of subarachnoid haemorrhage: prospective cohort study. BMJ. 2011;343:d4277. doi:10.1136/bmj.d4277 ↩ ↩² ↩³

5. Carpenter CR, Hussain AM, Ward MJ, et al. Spontaneous Subarachnoid Hemorrhage: A Systematic Review and Meta-analysis Describing the Diagnostic Accuracy of History, Physical Examination, Imaging, and Lumbar Puncture With an Exploration of Test Thresholds. Acad Emerg Med. 2016;23(9):963–1003. doi:10.1111/acem.12984 ↩ ↩² ↩³ ↩⁴

6. Perry JJ, Stiell IG, Sivilotti MLA, et al. Clinical decision rules to rule out subarachnoid hemorrhage for acute headache. JAMA. 2013;310(12):1248–1255. doi:10.1001/jama.2013.278018 ↩

7. Gillespie CS, Hanrahan JG, Mahdiyar R, et al. Diagnosis of subarachnoid haemorrhage: Systematic evaluation of CT head diagnostic accuracy and comparison with the 2022 NICE guidelines. Brain Spine. 2025;5:104200. doi:10.1016/j.bas.2025.104200 ↩ ↩²

8. Edlow JA, Bruner KS, Horowitz GL. Xanthochromia. Arch Pathol Lab Med. 2002;126(4):413–415. doi:10.5858/2002-126-0413-X ↩ ↩²

9. Perry JJ, Alyahya B, Sivilotti MLA, et al. Differentiation between traumatic tap and aneurysmal subarachnoid hemorrhage: prospective cohort study. BMJ. 2015;350:h568. doi:10.1136/bmj.h568 ↩ ↩² ↩³ ↩⁴ ↩⁵

10. Polmear A. Sentinel headaches in aneurysmal subarachnoid haemorrhage: what is the true incidence? A systematic review. Cephalalgia. 2003;23(10):935–941. doi:10.1046/j.1468-2982.2003.00596.x ↩ ↩²