Diagnostic Accuracy of CT for Subarachnoid Hemorrhage
Diagnostic Accuracy of CT for Subarachnoid Hemorrhage
Search Strategy: Using the PUBMED Clinical Queries diagnosis/broad filter you conduct a search for “subarchnoid hemorrhage” yielding 7121 citations which you then combine with a search for “computed tomography” yielding 2305 citations. Because you are most interested in the newest generation CT scanners, you add limits for English-only and published in the last 5-years to yield 535 citations (see http://tinyurl.com/7wmbend) which contains all of the manuscripts below.
“Doctor, this is the worst headache of my life!” moans the 33 year old mother of three on a blustery winter evening. She reports that the “stabbing” left-sided frontal headache began at 4:03PM while driving her daughter to tennis practice. The headache is unremitting over the subsequent two-hours and she notes associated nausea. She denies any prior history of headache disorders (no migraines or tension headaches) and cannot remember any recent or remote head trauma. She takes birth control pills, but denies any other daily medication use. She denies any recent travel history or sick contacts. She has no family history of aneurysms. On review of symptoms, she denies any other symptoms, including no fever, cough, or sore throat. On physical exam you note an extremely uncomfortable and tearful adult without any obvious trauma. Her vital signs are unremarkable and a detailed neurological exam demonstrates no focal deficits.
With this presentation, you are highly suspicious for subarachnoid hemorrhage. A Bayesian diagnostician, you develop the following list of diagnostic possibilities (with your gestalt-based pre-test probabilities): new-onset migraine (50%), tension headache (25%), subarachnoid hemorrhage (15%), cerebral venous sinus thrombosis (5%), CNS infection (3%), non-CNS infection (2%). In discussing these possibilities with your patient and her husband, you mention that a lumbar puncture will be essential to definitively exclude subarachnoid hemorrhage, even if a head CT is non-diagnostic/normal. Since she is not a big fan of long needles in her back, your patient tells you that she will not consent to a lumbar puncture for any reason. As you order an intravenous anti-emetic to provide her symptom relief from her headache and nausea, you contemplate on the diagnostic accuracy for subarachnoid hemorrhage of CT alone.
Population: ED headache patients with suspected subarachnoid hemorrhage
Intervention: Cranial CT alone
Comparison: Cranial CT + lumbar puncture
Outcome: Diagnostic accuracy, procedural morbidity
First years: Determining the sensitivity of computed tomography scanning in early detection of subarachnoid hemorrhage, Neurosurgery 2010; 66: 900-903. (http://pmid.us/20404693)
Second years: Is the combination of negative computed tomography result and negative lumbar puncture result sufficient to rule out subarachnoid hemorrhage? Ann Emerg Med 2008; 51: 707-713. (http://pmid.us/18191293)
Third years: Sensitivity of noncontrast cranial computed tomography for the emergency department diagnosis of subarachnoid hemorrhage, Ann Emerg Med 2008; 51: 697-703. (http://pmid.us/18207607)
Fourth years: Sensitivity of computed tomography performed within six hours of onset of headache for diagnosis of subarachnoid haemorrhage: prospective cohort study, BMJ 2011; 343: d4277. (http://pmid.us/21768192)
Article 1: Determining the sensitivity of computed tomography scanning in early detection of subarachnoid hemorrhage, Neurosurgery 2010; 66: 900-903.
Article 2: Is the combination of negative computed tomography result and negative lumbar puncture result sufficient to rule out subarachnoid hemorrhage? Ann Emerg Med 2008; 51: 707-713.
Article 3: Sensitivity of noncontrast cranial computed tomography for the emergency department diagnosis of subarachnoid hemorrhage, Ann Emerg Med 2008; 51: 697-703.
Article 4: Sensitivity of computed tomography performed within six hours of onset of headache for diagnosis of subarachnoid haemorrhage: prospective cohort study, BMJ 2011; 343: d4277.
Headache accounts for ~2% of presenting complaints in the ED and about 1% of these headache patients (1% of 2% = 0.02% of ED patients) will have subarachnoid hemorrhage (SAH). Spontaneous (atraumatic) SAH is caused by one of three disorders: cerebral aneurysms (most common), arteriovenous malformations, and perimesencephalic hemorrhages. The challenge is that migraine headaches outnumber SAH by 50-to-1 and can mimic SAH. The six-month mortality of aneurysmal SAH decreased from 1960 to 1995, but mortality was still 25%-50% with substantial regional variability. A third of survivors have neurological deficits affecting their daily living. Although 40% (Duffy 1983, Leblanc 1987, Verweij 1988, Bassi 1991) of SAH patients will have a small warning bleed (“sentinel leak”) a short time before their devastating bleed, missed diagnosis is the main culprit for delayed referral to a neurosurgeon (Adams 1980, Kassell 1985, Sved 1995, Mayer 1996, Jakobsson 1996, Neil-Dwyer 1997) and 12%-53% of SAH are misdiagnosed on their initial presentation. Misdiagnosis of SAH can produce less optimal outcomes and consequent litigation. Fortunately, early surgical management of cerebral aneurysms significantly reduces the mortality of SAH.
The severe consequences of missing SAH amongst headache patients, coupled with the widespread availability of definitive diagnostic tests (CT, lumbar puncture) have produced a standard of care whereby a liberal CT followed by LP if the CT is negative is the most defensible position. However, the role of LP after a negative CT has been questioned due to an increasingly low yield, high clinician variability in performing the LP, and lab inconsistencies in the analysis of CSF (Foot 2001). Furthermore, LP is only ~50% specific for SAH due to the problem of traumatic LP’s which represent up to 20% of LP’s (Marton 1986, Shah 2003). Inability to visualize or palpate the spine are predictors of a traumatic LP which can be reduced using fluoroscopy-guided LP.
Various methods have been proposed to differentiate traumatic LP from SAH, but none has been validated and none is well accepted. These methods include the use of D-dimer, decreasing number of RBC’s between Tubes 1 and 4, opening pressure, and visual xanthochromia. Spectrophotometry for xanthochromia is not available in 97% of North American hospitals. No study has demonstrated that spectrophotometry is superior to visual inspection for the detection of xanthochromia. Xanthrochromia may require 6- to 12-hours to develop following the sentinel bleed, too (Duffy 1985, Vermeulen 1996, van Gijn 2001). Xanthochromia can also develop in CSF tubes if there is a delay in the time to analyze the specimen (Edlow 2000, Noguchi 2000, Graves 2004). LP can yield false negative results in patients with anemia (hematocrit < 27% for CT or <22% for MRI, see Noguchi 2000). The risks of LP to discuss with patients include iatrogenic meningitis, post-LP heaches, back pain, and epidural hematoma or infection (Evans 1998, Roos 2003). Because of the risks of LP, some have recently advocated for CT angiography instead for the subset with a negative non-contrast CT (Carstairs 2006, McCormack 2010, Westerlaan 2011), but this approach also holds unintended consequences (Edlow 2010): Are the aneurysms identified causative or incidental?
The current studies indicate that newer generation CT scanners have significantly improved the sensitivity of CT alone. Perry et al. 2008 revealed that in ED patients with headache or syncope-associated headache suspicious for SAH, the prevalence (pre-test probability) of SAH is ~10%, while the most common alternative diagnoses are benign headache (46.5%) or migraine (26.4%). Assuming that 1/60 of those without follow-up (~10% were lost to follow-up) had SAH the positive likelihood ratio (LR+) for subarachnoid blood on CT of >5 rbc in last tube of CSF is 2.98 (95% CI 2.6-3.4) and the negative likelihood ratio (LR-) is 0.024 (95% CI 0.0-0.17). An abnormal CT or > 5 rbc’s in the CSF would not confirm the diagnosis of SAH (but it would increase the probability from 10% to 25%), but negative results on CT and LP would significantly reduce the probability of SAH from 10% to 0.27% (95% CI 0-1.9%). Perry et al 2011 confirm the high sensitivity of CT within 6-hours of headache onset adding to decades of older and much smaller CT diagnostic data (Wijdicks 1988, Harling 1989, Markus 1991, Linn 1999, Landtblom 2002). On the other hand, Byyny et al 2008 advise caution in a CT-only approach to rule-out SAH without LP since the sensitivity in cognitively intact headache patients was only 91% (95% CI 82%-97%).
Bayesian clinicians can also use Perry’s data to estimate test-treatment thresholds via Pauker’s formula:
Using the most conservative (sensitivity analysis-derived) estimates of sensitivity of CT for SAH, one obtains the following
- Ppos/nd = 1-spec = 0.33
- Pneg/nd = spec = 0.67
- Ppos/d = sen = 0.98
- Pneg/d = 1-sen = 0.02
Then one needs to ascertain the risk of the test (Rt), benefit of treatment in someone with disease (Brx) and risk of treatment in someone without disease (Rrx). In the era of CT, the risk of post-LP herniation is exceedingly rare. Other complications are not so rare including Post-LP headache (40% – McSwiney 1995, Lybecker 1990, Lybecker 1995), spinal or epidural hematoma (Roos 2003), or backache (35%, Evans 1998). To focus on the more serious complications of LP (headache for up to 1-year, prolonged backache, nerve root injury, or iatrogenic meningitis) we will use the figure of 0.3% (Evans 1998). Hillman et al demonstrated that 72% of past or future patients with early interventions had a good recovery versus 50% of those with delayed surgery. In the absence of a randomized controlled trial, we will use these values to estimate an absolute risk reduction of 22%. The risk of clipping or coiling a cerebral blood vessel that does not have an aneurysm would be in 8% intracranial hemorrhagic complication (Bodily 2011), although for a patient without a cerebral aneurysm to proceed to coiling would require a series of false positives: CT, LP, angiography. Therefore, Rrx = 0.08, Rt = 0.003, Brx = 0.22 which yields Test Threshold = 0.12 and Treatment Threshold = 0.87.
Further diagnostic research using contemporary CT scanners and explicit descriptors of STARD criteria will further define the diagnostic accuracy of CT and LP for SAH. In addition, the diagnostic accuracy of history, physical exam, and physician gestalt should be explored, perhaps in constructing a SAH clinical decision rule. Finally, future diagnostic research evaluating CSF RBC’s should treat this lab value as a continuous variable and report interval likelihood ratios.