Background CT has excellent sensitivity for subarachnoid haemorrhage (SAH) when performed within 6 hours of headache onset, but it is unknown to what extent patients with more severe disease are likely to undergo earlier CT, potentially inflating estimates of sensitivity. Our objective was to evaluate which patient and hospital factors were associated with earlier neuroimaging in alert, neurologically intact ED patients with suspected SAH.
Methods We analysed data from two large sequential prospective cohorts of ED patients with acute headache undergoing CT for suspected SAH. We examined the time interval from headache onset to CT, both overall and subdivided from headache onset to hospital registration and from registration to CT.
Results Among 2412 patients with headache, 194 had SAH, with 178 identified on unenhanced CT. Of these, 91 (51.1%) were identified by CT within 6 hours of headache onset and 87 after 6 hours. Patients with SAH had a shorter time from headache onset to hospital presentation (median 4.5 hours, IQR 1.7–22.7 vs 9.6 hours, IQR 2.8–46.0, p<0.001) and were imaged sooner after headache onset (6.4 hours, IQR 3.5–27.1 vs 12.6 hours, IQR 5.5–48.0, p<0.001) compared with those without SAH. The median time from in-hospital registration to CT scan was significantly shorter in those patients with SAH although this difference was less than 1 hour (1.9 hours, IQR 1.2–2.8 vs 2.5 hours, IQR 1.5–3.9, p<0.001). Arrival by ambulance (OR 3.1, 95% CI 1.94 to 4.98, p<0.001) and higher acuity at triage (OR 1.39, 95% CI 1.02 to 1.88, p=0.032) were among the factors associated with having CT imaging within 6 hours of headache onset.
Conclusions Time from headache onset to imaging is moderately associated with positive imaging for SAH. Delay to hospital presentation accounts for the largest fraction of time to imaging, especially those without SAH. These findings suggest limited opportunity to reduce lumbar puncture rates simply by accelerating in-hospital processes when imaging delays are under 2 hours, as diagnostic yield of imaging decreases beyond the 6-hour imaging window from headache onset.
- emergency department
Statistics from Altmetric.com
What is already known on this subject?
In patients with rapid onset, severe headache, cranial CT performed within 6 hours of headache onset and deemed negative by a qualified radiologist essentially excludes the diagnosis of subarachnoid haemorrhage (SAH), challenging traditional teaching to nevertheless perform LP.
It is unclear whether this higher sensitivity associated with earlier CT may in part be attributed to a spectrum effect in which patients with readily identifiable SAH on CT are more likely to present and be imaged earlier, thus questioning the merits of processes aimed at accelerating the time to imaging in all patients with headache.
What this study adds?
Time from headache onset to CT imaging is moderately associated with positive imaging for SAH and the largest fraction of time to imaging is the delay to hospital presentation, especially in those without SAH.
These findings suggest limited opportunity to reduce lumbar puncture rates simply by accelerating in-hospital processes when door-to-imaging intervals are under 2 hours, as diagnostic yield of imaging decreases markedly beyond the 6-hour imaging window from headache onset.
Subarachnoid haemorrhage (SAH) is a life-threatening cause of headache, and is usually the chief diagnostic consideration in patients with a rapid onset, severe headache. Traditional teaching has required lumbar puncture (LP) despite negative CT to exclude this diagnosis.1 Our group, however, has reported near perfect sensitivity of CT, when performed within 6 hours of headache onset and interpreted by qualified radiologists, in patients who are awake and neurologically intact,2–4 allowing these patients to safely forego LP.
The higher reported sensitivity compared with historical data is widely assumed to be due to improved scanner technology and reduced time for the blood to diffuse into the cerebrospinal fluid (CSF). Many have suggested expedited imaging might, therefore, reduce the need for LP. A 6-hour window for neuroimaging from headache onset is also comparable with the current emphasis on timely imaging in ischaemic stroke. However, it is unclear whether the higher sensitivity associated with earlier CT may in part be attributed to spectrum effect, that is, patients with readily identifiable SAH on CT presenting and being imaged earlier.5–7 In other words, patients with SAH who present for care within 1 hour of a sudden headache are likely to differ substantially from those who present 1 or 2 days later. Such potential confounders might include pain intensity, perceived severity of illness, age, arrival by ambulance and sex. As such, we wondered to what degree factors independent of imaging technology and time elapsed may have contributed to the higher sensitivity of CT within the first 6 hours post headache onset in our cohort of patients. Evidence of spectrum effect would question the external validity of our findings in healthcare systems where imaging delays are substantially shorter, and the merits of altering practice to accelerate neuroimaging to avoid performing LP in patients with lower pretest likelihood of SAH. Prioritising neuroimaging as ‘next on the table’ for patients with sudden headache could also delay imaging for other patients with time-sensitive conditions.
The objective of this study was to evaluate which patient and hospital factors were associated with earlier neuroimaging in alert neurologically intact ED patients with suspected SAH. We also estimated the impact of accelerating in-hospital processes on diagnostic efficiency and yield.
This was an a priori planned secondary analysis of two sequential multicentre prospective cohort studies that were used to derive and validate the Ottawa SAH Rule, a clinical decision rule used to identify patients with headache at high risk for SAH.2 ,8 The studies were conducted at 11 university-affiliated tertiary care hospitals in Canada from 2000 to 2010.
As described previously,2 ,8 all adults presenting to the ED with a new acute (peaking within 1 hour) headache were eligible for this study, unless accompanied by a new neurological deficit or decreased mental status. The original studies excluded patients with head trauma, multiple prior similar headaches, delay to presentation of more than 14 days, transfer from another ED with confirmed SAH, papilloedema, or prior history of SAH, cerebral aneurysm, brain neoplasm, hydrocephalus or ventricular shunt. For the current study, we pooled both cohorts but excluded those patients in whom CT was not performed.
We used the original study definitions for all variables, including a composite definition of disease positive: subarachnoid blood on unenhanced head CT as determined by an experienced radiologist (a neuroradiologist or general radiologist who regularly interprets head CT images); visible xanthochromia in the CSF as reported by the laboratory or red blood cells (>5×106/L) in the final tube of CSF collected and an aneurysm identified on cerebral angiography (whether by digital subtraction fluoroscopy, CT or MRI). For patients with a negative unenhanced CT in whom CSF was not obtained, we performed structured telephone follow-up at 14 days and searched hospital and provincial datasets to identify any missed cases.
Triage acuity was based on the initial assessment by the emergency triage nurse using the Canadian Triage Acuity Scale (CTAS), a national standardised five-level scale defined as follows: I, Resuscitation; II, Emergent; III, Urgent; IV, Less Urgent; and V, Non-Urgent.9 For analysis, we collapsed the scale into three levels based on discussion between study authors: high acuity (I–II), medium acuity (III) and low acuity (IV–V). Pain score at peak was by patient report to the treating physician at enrolment, with 10 representing the ‘worst pain imaginable’.
Severe pain at peak was defined to be a pain score of 9 or 10. We categorised time to peak headache into four levels based approximately on distribution quartiles: ≤1 s, 1 to ≤30 s, 30 to ≤300 s and 300 to 3600 s. We retained the original study binary variables (‘yes’ or ‘no’): worst headache ever, witnessed loss of consciousness, vomiting, headache onset during sexual activity, neck pain or stiffness and decreased neck flexion on examination.
We were primarily interested in the time interval from headache onset to CT. The start of this interval was determined at enrolment by the treating physician and explicitly recorded on the standardised study data collection forms. When this variable was missing, trained research personnel ascertained it by reviewing the medical record. For a small number of cases with no exact time of onset recorded, we imputed onset during ‘waking’, ‘morning’, ‘afternoon’, ‘evening’ and ‘night’ to be 06:00, 08:00, 13:00, 18:00 and 20:30, respectively, by consensus of study authors. The end of the interval was the electronic timestamp on the first CT image acquired. We prespecified an onset to imaging interval cut-point of less than 6 hours as per the original study.2 We divided this time interval into the prehospital (time from headache onset to hospital registration) and hospital phase (time from hospital registration to CT), and prespecified a hospital interval of under 2 hours as a cut-point of interest.
We compared baseline characteristics, and performed modelling on subjects with (disease positive) and without (disease negative) SAH. First, we used the Kaplan-Meier method and log-rank tests to compare differences in time intervals in patients with and without SAH. Second, we performed multivariable stepwise logistic regression modelling with variable selection using backward elimination to select predictors of interest from a set of starting variables previously shown to be associated with SAH: arrival by ambulance, worst headache of life, peak pain severity, onset during exertion, witnessed loss of consciousness, vomiting, transfer from another facility, neck stiffness with flexion, neck pain, sex, CTAS score, onset with sexual activity, age, systolic BP, time to peak headache and prehospital delay.8 We adjusted for clustering of the data by including hospital site as a fixed effect into the models. The multivariable modelling included multiple imputation for missing data, which was treated as missing at random. The multiple imputation involved three specific steps. First, we used the original dataset for observations without missing data and to this dataset, imputed data for the missing observations using the Markov Chain Monte Carlo method assuming multivariate normality (PROC MI in SAS). The missing data were filled in five times to generate five complete datasets. Second, the five complete datasets were analysed following the planned analysis strategy; in this case, logistic regression with variable selection using backward elimination10 and the estimates of the coefficients for the logistic model were derived for each dataset. Third, the results from the five complete datasets were combined for the inference. Coefficients for the logistic model were derived based on a weighted combination of the coefficient estimates from the five datasets, taking the variability of the estimates within and between the datasets into consideration (PROC MI Analyse in SAS). The variable with the highest p value was removed and the above process was repeated until no variables having p>0.05 were left in the model.
We performed separate stepwise logistic regression modelling with variable selection using backward elimination for the two interval cut-points defined above, namely, the prehospital phase (time from headache onset to hospital registration) and the hospital phase (time from hospital registration to CT).
We also performed subgroup analyses using only patients with no missing information. This complete case analysis assessed each variable in the univariable analysis and retained them if they were significant at the 0.25 level, and sequentially deleted the variable from the multivariable logistic model if not significant at the 0.05 level by the Wald test. We confirmed significance using the likelihood ratio test. We performed all analyses using SAS software V.9.3 (SAS Institute, Cary, North Carolina, USA).
From the 4130 patients originally enrolled in the two study phases, 759 were excluded for lack of CT and another 56 were excluded due to delay to presentation of more than 14 days, CT being performed prior to registration or inaccessible records (figure 1). Of the remaining 3315 eligible patients, 903 patients had some missing data, leaving 2412 with complete information (table 1). No individual variable was missing in more than 10% of cases, and all but three (pain score at peak, CTAS score and neck stiffness on exam) were missing in no more than 3% of the cases. Therefore, the total number of subjects used was 3315 for the imputation analysis and a total of 2412 for the remaining analyses.
Of the 2412 patients, 194 (8.0%) had a final diagnosis of SAH (table 1), 178 of whom (92%) were identified on unenhanced CT. Subjects with SAH were more likely to be older (52.7 years vs 44.2 years, p<0.001), and to have arrived by ambulance (56.2% vs 21.7%, p<0.001), vomited (65.5% vs 26.8%, p<0.001) and experienced a witnessed loss of consciousness (7.7% vs 3.2%, p<0.001).
The median time from headache onset to CT was 12 hours (IQR 5.3–48.0). This time interval was substantially shorter in patients with SAH (median 6.4 hours, IQR 3.5–27.1) compared with those without SAH (12.6 hours, IQR 5.5–48.0, p<0.001) (table 2). Most of this difference was due to disease positive patients presenting to hospital several hours earlier (4.5 hours, IQR 1.7–22.7) than disease negative patients (9.6 hours, IQR 2.8–46.0, p<0.001) (figure 2). The in-hospital interval from registration to imaging was generally short, and statistically shorter in disease positive patients, but the difference averaged under 1 hour (1.9 hours, IQR 1.2–2.8 vs 2.5 hours, IQR 1.5–3.9, p<0.001) (table 2).
The diagnostic yield of CT imaging (proportion of studies positive for subarachnoid blood) was much higher when performed within 6 hours of headache onset with over 1 in 10 studies being positive. Just over half of the patients with SAH (91) had imaging studies obtained within 6 hours of headache onset and two-thirds of the patients with SAH had imaging within 12 hours (figure 3). Thus, a total of 87 patients had CT imaging positive for SAH obtained more than 6 hours after the headache onset. Of note, among the 1685 imaging studies completed more than 6 hours after the headache onset, 16 were falsely negative (ie, SAH identified by CSF analysis). Thus, the negative predictive value for SAH was 99% when CT was performed more than 6 hours after the headache onset.
In the final stepwise logistic regression model incorporating multivariable imputation for missing information, the odds of having CT within 6 hours of headache onset were almost two-thirds greater in subjects with SAH, and more than three times greater in those with higher triage acuity and arrival by ambulance (table 3). The model for CT within 2 hours of hospital registration was very similar to the model assessing CT within 6 hours. Not surprisingly, patients transferred from another facility (presumably for imaging) were less likely to be imaged within 6 hours of headache onset, but more likely to be imaged within 2 hours of arrival at the referral hospital.
The complete case analysis, which included only cases with no missing data, agreed very closely with those of main multiple imputation analysis (see online supplementary appendices 1 and 2). Being SAH positive, triage acuity, arrival by ambulance and transfer from another hospital were consistently identified as independently associated with earlier neuroimaging, with similar ORs. Depending on the model, other factors that also appeared included onset during exertion, neck pain, higher systolic BP, vomiting, older age, worst headache ever, being male, shorter time to hospital presentation and time to peak headache.
In this large multicentre prospective cohort study conducted over 10 years, we found that the interval from headache onset to CT was moderately associated with the presence of subarachnoid blood on imaging, as well as other potential markers of disease severity, namely arrival by ambulance and higher acuity at triage. We observed these effects despite restricting our population to alert and neurologically intact patients. Moreover, the interval from headache onset to imaging was primarily due to the interval prior to arrival in the ED, and thus, the healthcare-seeking behaviour of the patient rather than delays from hospital arrival to diagnostic imaging. Patients ultimately diagnosed with SAH presented earlier after headache onset, and underwent rapid diagnostic imaging. Very few imaging studies obtained more than 6 hours after headache onset were falsely negative.
We support rapid neuroimaging for the indication of ‘rule out SAH’, but urge caution with any well-intentioned intervention urging the public to present sooner to hospital with headache as the prevalence of disease is very low in late-presenting patients. The majority of patients with this catastrophic diagnosis already seek immediate emergency care, many by ambulance. These findings should be interpreted in the context of Canada's universal health insurance tradition and of limited access block to emergency care.
Retrospective studies from different countries have also reported nearly perfect sensitivity of early CT using current multidetector, high-resolution imaging technology and interpreted by an experienced radiologist.11–14 Two European studies had remarkably high disease prevalence rates of 44%11 to 59%,12 with approximately one in two scans being positive, even though one of the studies11 specifically included only awake and neurologically intact patients. In contrast, our cohort had a disease prevalence of 8% with less than 1 in 10 CT scans being positive. We also prospectively enrolled patients, and retained those in whom LP was not performed by using a proxy outcome based on 14-day follow-up. Most prior studies do not report a measure of disease severity for subjects who were deemed to have SAH. Edlow and Fisher express concern that these very high prevalences may inflate the sensitivity estimates.5 One strength of our study is that we have better defined the test characteristics of CT in the target population of chief interest to emergency physicians, grouped by delay to imaging, in keeping with principles first expressed by Ransohoff and Feinstein many years ago.6 We can also estimate that any spectrum effect is at most modest in the stratum of alert and neurologically intact patients with headache.
Our study has some limitations. First, we did not examine patients with more severe disease including those who present with cranial nerve palsies or other neurological deficits, or who are obtunded or comatose. Our decision rule work has always excluded such patients in whom neuroimaging is necessary. Thus, our subjects would all be classified as Hunt and Hess Grade I at enrolment. We also did not attempt to grade CT findings, but would expect most disease positive patients to be of low grade. The appearance of subarachnoid blood on CT including estimates of haemorrhage volume also evolves over time, precluding an analysis in which the delay to CT and appearance on CT are assumed to be independent.
Second, we relied on a proxy outcome when LP was not performed, as we could not mandate that enrolling physicians perform an LP in all subjects. We have used this proxy outcome since the inception of this decision rule work and have never identified a missed SAH on telephone follow-up or review of medical records. The few patients missed on initial presentation re-presented to their local neurosurgical centre prior to telephone follow-up.
In conclusion, time from headache onset to imaging is strongly associated with SAH, yet this interval is largely composed of the delay to initial hospital presentation. When neuroimaging is available within 2 hours of arrival in otherwise awake and alert patients with headache, about half of the SAH patients will be identified on CT obtained within 6 hours of headache onset. Moreover, most of the remaining disease positive patients will still be diagnosed on initial CT despite longer intervals. As such, altering hospital systems to achieve a 6-hour CT target for patients with acute headache would appear to have limited impact on decreasing the number of LPs performed.
Presented at the European Stroke Conference, Nice, France, May 2014. Presented at the Canadian Association of Emergency Physicians (CAEP) Annual Scientific Meeting, Ottawa, Ontario, Canada, June 2014.
Contributors The author contributions were as follows: JJP and MLAS conceived the idea and prepared the manuscript with MK. JS coordinated the study, collected data and contributed to the writing of the manuscript, and GAW provided considerable statistical assistance and revised the manuscript. MLAS, MJB, ME, AW, CMH, JSL, ME, MP, HL and IGS assisted with study design and revised the manuscript.
Funding Ontario Ministry of Health and Long Term Care, Physician Services Incorporated Foundation, Canadian Institutes of Health Research (grant numbers: MOP-67107, MSH-87715).
Competing interests None declared.
Ethics approval Ottawa Hospital Research Ethics Board.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.