Article Text

Download PDFPDF

Atraumatic headache in US emergency departments: recent trends in CT/MRI utilisation and factors associated with severe intracranial pathology
  1. John W Gilbert1,
  2. Kevin M Johnson2,
  3. Gregory L Larkin3,
  4. Christopher L Moore3
  1. 1Yale School of Medicine, New Haven, Connecticut, USA
  2. 2Department of Diagnostic Radiology, Yale School of Medicine, New Haven, Connecticut, USA
  3. 3Department of Emergency Medicine, Yale School of Medicine, New Haven, Connecticut, USA
  1. Correspondence to John W Gilbert, Yale School of Medicine, Attn: Hilmer Ayuso, 464 Congress Avenue Suite 260, New Haven, CT 06519, USA; john.gilbert{at}


Objectives To estimate recent trends in CT/MRI utilisation among patients seeking emergency care for atraumatic headache in the USA and to identify factors associated with a diagnosis of significant intracranial pathology (ICP) in these patients.

Design/setting/participants Data were obtained from the USA National Hospital Ambulatory Medical Care Survey of emergency department (ED) visits between 1998 and 2008. A cohort of atraumatic headache-related visits were identified using preassigned ‘reason-for-visit’ codes. Sample visits were weighted to provide national estimates.

Results Between 1998 and 2008 the percentage of patients presenting to the ED with atraumatic headache who underwent imaging increased from 12.5% to 31.0% (p<0.01) while the prevalence of ICP among those visits decreased from 10.1% to 3.5% (p<0.05). The length of stay in the ED was 4.6 h (95% CI 4.4 to 4.8) for patients with headache who received imaging compared with 2.7 (95% CI 2.6 to 2.9) for those who did not. Of 18 factors evaluated in patients with headache, 10 were associated with a significantly increased odds of an ICP diagnosis: age ≥50 years, arrival by ambulance, triage immediacy <15 min, systolic blood pressure ≥160 mm Hg or diastolic blood pressure ≥100 mm Hg and disturbance in sensation, vision, speech or motor function including neurological weakness.

Conclusions The use of CT/MRI for evaluation of atraumatic headache increased dramatically in EDs in the USA between 1998 and 2008. The prevalence of ICP among patients who received CT/MRI declined concurrently, suggesting a role for clinical decision support to guide more judicious use of imaging.

  • Headache
  • imaging
  • CT/MRI
  • guidelines

Statistics from

Request Permissions

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.


Headache is one of the most common complaints in the emergency department (ED), accounting for approximately 1–3% of all visits.1 ,2 While the majority of headaches are benign and self-limited, a subset are associated with potentially life-threatening intracranial pathology (ICP). These patients may present with unremarkable symptoms and physical findings, impeding timely diagnosis.3 ,4 Owing to lack of high-quality data, clinical guidelines for imaging in atraumatic headache remain unclear; for example, current recommendations of the American College of Emergency Physicians (ACEP) and American College of Radiology (ACR) are ambiguous for patients with a non-focal neurological examination without red flags such as history of trauma or ‘thunderclap’ headache.5 ,6

Multiple studies have shown that neuroimaging for ED patients who lack any red flags is unlikely to lead to discovery of significant ICP.7 ,8 An ACR expert panel review of CT use in patients with atraumatic headache but normal neurological examination reported a diagnostic yield of 0.4% in 897 studies of patients with migraine and 2.4% in 1825 patients with unspecified headache.9 Despite the low yield of imaging, nearly half of the respondents in an international survey of 2100 emergency physicians stated that every patient who presents to the ED with acute headache should categorically receive CT.10 Several studies indicate that the overall use of CT/MRI in EDs is increasing, including neuroimaging.11 ,12 Increased use of CT/MRI is associated with both monetary and non-monetary costs including increased exposure to ionising radiation.

A recent study evaluating national trends for the use of CT/MRI in EDs for injury-related conditions found that use of these procedures is increasing without a comparable increase in the prevalence of life-threatening diagnoses.13 No study to date has analysed comparable trends over the past decade among ED visits for atraumatic headache. The goals of this investigation are to describe national trends in utilisation of CT/MRI among ED patients with atraumatic headache between 1998 and 2008. We hypothesised that the use of CT/MRI has increased without a commensurate increase in the frequency of diagnosis of ICP. We also aimed to identify independent factors associated with an ICP diagnosis, underscoring the need for future prospective development of a clinical decision aid to guide more judicious use of imaging.


The National Hospital Ambulatory Medical Care Survey (NHAMCS) is conducted annually by the Division of Health Care Statistics of the National Center for Health Statistics (NCHS), Center for Disease Control (CDC). The survey represents a four-stage probability sample of visits to randomly selected non-institutional general hospitals and short-stay (<30 days) hospitals, with the exception of federal, military and VA hospitals, in the USA.14 The sampling design includes primary selection of units corresponding to geographical regions (eg, counties, townships, metropolitan statistical areas), hospitals within those regions, EDs and outpatient units within those hospitals and patient visits within those EDs and outpatient units.15

Hospital staff trained by a field interviewer from the Census Bureau are responsible for collecting data from a random sample of patients during annual recording periods of 4 weeks' duration using standardised patient record forms. Visits are excluded if no health services are rendered. Further information regarding survey methodology is available online.16 Given that NHAMCS data is publicly accessible and de-identified, this study is non-human subjects research exempt from Institutional Review Board review.

We identified patient cases from the 1998–2008 NHAMCS surveys using the following NCHS-assigned patient ‘reason-for-visit’ classification codes which reflect the patient's description of his or her problem upon presentation to the ED: headache (1210.0), sinus headache (1410.0), migraine (2365.0) or facial pain (1055.4). Patients with any of the above codes in any of the three reason-for-visit fields were included.

To ensure an accurate representative sample, visits related to trauma or injury were excluded in several ways: by excluding all patient cases for which the survey item asking if the visit was related to injury/poisoning/adverse event was marked ‘yes’; by excluding all patients with injury or poisoning-related International Classification of Diseases (ICD-9-CM) diagnosis codes specified (800–999); and by excluding all patients with causes of injury/poisoning/adverse event coded via the Supplementary Classification of External Causes of Injury and Poisoning (E-codes) within the survey. Patient records with absent reason for visit and/or diagnosis codes for all available fields were also excluded. The NHAMCS dataset was used in its entirety as a reference to compare our sample of headache-related ED visits with a representative sample of all ED visits over the same time period.

Absolute numbers of ED visits were approximated using NCHS-assigned sample weights tabulated for each patient visit record, taking into account the four-stage design of the survey, thus providing an unbiased estimate of the annual number of visits.14 ,15 For each case we recorded whether diagnostic CT or MRI was performed. The number of cases in which MRI was ordered was too low to evaluate separately and was also treated as a combined variable in some NHAMCS survey years; consequently, we created a combined CT/MRI variable for all years examined, as has been done in recent studies (eg, by referring to CT/MRI as ‘advanced radiology’).13 Comparing this variable against the study population, ED utilisation of imaging for atraumatic headache was approximated by year. Calculation of the average length of visit is based on data from 2001 onwards because the NHAMCS did not include this variable as a survey item prior to 2001.

We next recorded the frequency with which diagnosis of ICP was associated with use of CT/MRI. ICP was defined from a list of available ICD-9 codes that signify potentially life- or limb-threatening diagnoses made on the basis of neuroimaging. This included subarachnoid haemorrhage, benign or malignant brain neoplasm, stroke (ischaemic, haemorrhagic, embolic, thrombotic), intracranial haemorrhage, cerebrovascular anomaly, thrombosis, cerebral aneurysm, intracranial abscess and vertebral dissection. Trends over time were calculated using the same method as imaging utilisation.

Imaging studies in patients with CNS infection (encephalitis/meningitis) may be normal and the results are frequently non-specific, neither excluding nor definitively establishing a diagnosis. The diagnosis of meningitis is made on the basis of clinical assessment and CSF analysis. The most important role of CT is to identify contraindications to lumbar puncture—that is, occult causes of increased intracranial pressure such as cerebral abscess. For these reasons, the meningitides/encephalitides were excluded from our definition of ICP diagnosed on the basis of imaging.

To identify factors associated with an ED diagnosis of ICP, we examined visits by several demographic variables including age, sex, race and geographical location. We also examined variables related to the patient's visit such as mode of arrival and triage urgency. Some variables (eg, mode of arrival) were not included in the survey data for certain years prior to 2003, so we used years 2003–8 for analysis. Finally, we examined vital signs and several features of presentation by assessing reason-for-visit codes. A total of 18 factors were examined, many of which comprised multiple related reasons for visit in the NHAMCS survey. Adjusted ORs were calculated to identify a subset of variables significantly associated with receiving CT/MRI, defined as an adjusted OR >1.00 with a 95% CI not including 1.00.

Statistical analyses were performed using Statistical Analysis Software (SAS) V.9.1.3 (SAS Institute). Unless specifically noted otherwise, ED visit sample estimates and associated CIs were calculated using relative standard errors (dividing the SE by the estimate itself) of <30% and with at least 30 raw data items in accordance with NCHS recommendations. Adjusted ORs for factors associated with ICP were calculated using logistic regression analysis. The least squares method of linear regression was used for analysis of trends, with p<0.05 deemed statistically significant.


A total of 15 062 patient records of atraumatic headache-related ED visits were identified for our sample. This represents an estimated 54 million total ED visits (95% CI 49.8 to 58.1 million) by patients with atraumatic headache in the USA between 1998 and 2008, accounting for 4.5% (95% CI 4.3% to 4.6%) of all ED visits during this time period. Baseline demographic characteristics of this sample are summarised in table 1. Relative to the non-headache ED population, visits for headache were overwhelmingly for patients aged 18–49 years (65.1% vs 47.0%, p<0.01). There is a greater representation of women overall in the atraumatic headache versus non-headache ED population but no significant differences on the basis of race or ethnicity (68.8% vs 53.0%, p<0.01).

Table 1

Demographics and characteristics of US emergency department (ED) visits for headache-related conditions, 1998–2008

Of the headache-related visits sampled, CT or MRI was ordered in approximately 11 million (95% CI 9.9 to 12.2 million) patients, representing 20.4% (95% CI 19.0% to 21.8%) of all ED visits for headache between 1998 and 2008. The mean age of patients who received imaging for atraumatic headache was 41.5 years (95% CI 40.6 to 42.5). Trends in CT/MRI utilisation and prevalence of ICP for ED visits with atraumatic headache over the past decade are shown in figure 1. Notably, the percentage of visits in which CT/MRI was ordered increased from 12.5% to 31% between 1998 and 2008 (p<0.01). This corresponded to a similar increase in CT/MRI utilisation across the same period among all ED visits in the survey, from 4.2% in 1998 to 15.0% in 2008 (p<0.01). Length of ED visit was 4.6 h (95% CI 4.4 to 4.8) among those who received CT/MRI compared with 2.7 h (95% CI 2.6 to 2.9) for those who did not (p<0.01).

Figure 1

US trends in CT/MRI utilisation and yield of CT/MRI to diagnose severe intracranial pathology (ICP) in the emergency department (ED) evaluation of atraumatic headache, 1998–2008.

The observed trends in imaging utilisation remained consistent across several demographic variables including age, gender and race or ethnicity. The percentage of visits by men receiving CT/MRI increased from 13.9% to 35.6% between 1998 and 2008, while the percentage of visits by women increased from 11.8% to 29% (p<0.01 for both). A comparable and significant increase was seen among non-Hispanic white (11.8% to 30.5%), black (9.8% to 34.8%) and Hispanic (13.1% to 41.5%) individuals (p<0.01 for all). A significant increase in the use of imaging among adult patients from 18.9% to 44.9% in those aged ≥50 years and from 11.1% to 28.3% in those aged 18–49 years was also seen (p<0.01 for both).

In parallel with the observed increase in CT/MRI utilisation, the proportion of visits in which ICP was diagnosed by CT/MRI (ie, the yield to detect ICP) decreased from 10.1% to 3.5% between 1998 and 2008 (p<0.05, figure 1), with an average yield across the study period of 4.9% (95% CI 3.9% to 5.8%). Notably, the average yield to detect ICP was 10.4% (95% CI 7.8% to 13.1%) for patients aged ≥50 years but only 2.3% (95% CI 1.5% to 3.1%) for those aged <50 years in the study population (p<0.0001). No significant differences in yield were observed on the basis of race or ethnicity; however, a difference was noted based on the type of presenting headache complaint to the ED. The proportion of visits in which ICP was diagnosed among those with a non-migraine headache complaint who received imaging was 5.2% (95% CI 4.1% to 6.2%), while only three patient records in the survey sample across the study period with a reason for visit of migraine were associated with ICP upon imaging, representing a yield of 1.0% (95% CI 0% to 2.6%, N <30). The proportion of visits in which ICP was diagnosed among patients with private insurance who received imaging was 3.8% (95% CI 2.6% to 5.0%) compared with 6.6% (95% CI 4.5% to 8.7%) in those with Medicare, Medicaid or Worker's Compensation (p<0.01). Correspondingly, private insurance was associated with an increased likelihood of receiving CT/MRI (OR 1.25, 95% CI 1.1 to 1.4). Finally, 10 of the 18 variables examined were found to be associated with a significantly increased odds of ICP among the ED population with autraumatic headache (table 2).

Table 2

Adjusted odds ratios of selected variables for a diagnosis of ICP among patient visits to the emergency department (ED) for atraumatic headache, 2003–8


This study found that the use of CT/MRI in the evaluation of atraumatic headache in the ED increased dramatically in the USA between 1998 and 2008. The proportion of visits in which imaging was ordered and ICP was diagnosed (diagnostic yield to detect ICP) decreased concurrently. The use of CT/MRI in this setting is associated with an increased length of stay in the ED, healthcare burden and exposure to ionising radiation.

Our results are in line with several previous investigations demonstrating that the overall use of diagnostic imaging in the ED is increasing.11 ,12 Compared with a previous demographic study using older NHAMCS data from 1992 to 2001 and reporting an overall imaging rate of 14% among patients with atraumatic headache, the cumulative imaging rate in our study for the years 2003–8 represents an increase of 76%.17 Our reported trends raise the concern that imaging resources may be overutilised in the emergency evaluation of patients with headache, as argued in a recent article on indications for headache imaging by the ACR utilisation rounds of the Massachusetts General Hospital.18 A similar concern was echoed in a recent investigation of the diagnostic yield of CT in the management of patients in the ED with headache and a normal neurological examination which found a low percentage of positive findings and limited cost efficacy.19

Several factors help to explain the observed trends. In a prospective review study of CT utilisation for headache evaluation in the ambulatory setting, clinicians reported ordering 17% of scans due to patient demand and/or medicolegal concerns.20 In addition to medicolegal fears, clinical guidelines remain ambiguous; for example, the ACEP Subcommittee on Acute Headache did not issue any recommendations in its clinical policy statement representing ‘generally accepted principles for patient management that reflect a high degree of certainty’ based on currently available evidence.6 Given the potentially devastating consequences of a missed diagnosis, many emergency physicians may feel uncomfortable trusting their own judgement in the absence of clear evidence-based guidelines. This may explain why nearly 50% of ED physicians in a recent survey reported that all patients presenting to an ED with headache should categorically receive imaging.10 Finally, the proliferation of faster scanners and their increased availability and proximity to ED treatment areas may promote increased imaging utilisation more generally.

The particularly low prevalence of ICP among patients with a reason for visit of migraine who received imaging is probably related to the fact that a small number of ED patients with severe headache may comprise a disproportionately large share of headache-related ED visits.21 These patients often have pre-existing diagnoses such as migraine and presumably use the ED more for purposes of chronic disease management than treatment of underlying life-threatening pathology. Our finding that private insurance is associated with an increased likelihood of receiving CT or MRI can also be interpreted in several ways. A previous study reported a similar influence of insurance status on the CT scan rate using NHAMCS data and suggested that this may be due to a lack of access and underutilisation of imaging in socioeconomically disadvantaged patients.17 Alternatively, patients with a lack of access could be more likely to use the ED for chronic disease management of conditions (such as abortive therapy for migraine), with the result that these patients have a decreased likelihood of receiving imaging merely as a consequence of the manner in which the ED is being used. Further studies are needed to investigate this potential disparity.

Of the 10 factors found to be significantly associated with a diagnosis of ICP, arrival by ambulance is certainly not surprising. Previous studies have shown the highest concordance in measures of diagnostic accuracy between experts at headache centres and ED physicians in patients arriving by ambulance, and also demonstrated this factor to be a clear marker of disease severity.22 Similarly, triage status has been shown to correlate with short-term survival and is a reasonable (if not always reliable) index of disease severity.23 Disturbances in sensation, vision, speech or motor functions (including neurological weakness) are all correlates of CNS abnormalities that typically warrant further investigation with imaging, as recommended by the ACEP. Age >50 years is one of the few risk factors for ICP that ACEP explicitly recognises in its clinical guidelines, and has been demonstrated as a predictor of ICP in prior studies.5

To our knowledge, only one attempt has been made in the literature to develop a decision aid for patients who present to the ED with acute headache. A prospective cohort study identified a number of high-risk clinical characteristics for subarachnoid haemorrhage (SAH) and proposed three decision rules, all with 100% sensitivity, that incorporated some combination of the following variables: age (>40, >45 or 45–55 years); neck pain/stiffness; vomiting; systolic blood pressure >160 mm Hg or diastolic blood pressure >100 mm Hg; and arrival by ambulance, among others.24 Our study identifies some of the same characteristics as being significantly associated with ICP (age ≥50 years, systolic and diastolic blood pressure, arrival by ambulance), while others failed to reach significance (neck pain/stiffness, vomiting). This is probably due to the different endpoints of the study (SAH vs ICP), as our variables encompass a wider range of pathology with more diverse clinical manifestations. While patient characteristics were similar between these studies (mean age 43.4 years vs 41.5 years in our sample, 68.8% vs 60% female), other features appear more specific to SAH such as neck pain, vomiting and self-reported ‘worst headache of life’ reported by more than three-quarters of patients in the SAH study. These differences highlight the difficulty inherent in developing clinical guidelines for the evaluation of acute headache, given its myriad causes and associated symptoms.

The key strength of our study is the ability to make reliable national estimates based on the NHAMCS probability-based sample. However, studies like this one also have several limitations inherent to their retrospective design. First, the dataset is reliant on information in patient records that may be incorrectly coded or incomplete; however, the highly rigorous survey methodology makes this less of a concern. Second, the dataset does not represent missing diagnoses due to the healthcare provider failing to record accurately all conditions diagnosed during the visit and is vulnerable to other forms of coding bias. Third, there is no way to track diagnoses or further imaging that may have occurred after initial ED evaluation due to lack of patient follow-up data. Particularly in the earlier years of the study when patients were less likely to receive imaging, significant pathology may have been missed if there was initially an incorrect diagnosis given in the ED. Fourth, while the dataset offers a useful broad snapshot of national trends in ED utilisation, it limits our search for factors associated with a diagnosis of ICP to the items that are included in the annual surveys. For example, it is impossible to ascertain from the NHAMCS survey data whether patients had a focal neurological examination, thunderclap headache or other indices of disease severity.

Our study suggests that, while imaging for atraumatic headache is increasing dramatically, our improvement in diagnosing ICP is minimal and that decision support may be helpful in deciding who requires imaging. Analysing trends in CT/MRI utilisation for evaluation of atraumatic headache and identifying factors that may be associated with the need (or absence of a need) for imaging is a critical first step in understanding when imaging should be obtained. The benefits of reducing unnecessary imaging include reduction of healthcare costs, shorter patient length of stay, decreased exposure to ionising radiation and improved ED flow and efficiency. Together with nuclear imaging, CT accounted for over 75% of total radiation delivered to patients despite comprising only 20% of examinations in a review of over 950 000 adults across the USA.25 It has been estimated that up to 2.0% of all cancers in the USA and UK are associated with ionising radiation exposure secondary to CT.26 Future studies should be directed at prospectively developing a clinical decision aid to guide the use of imaging in the ED evaluation of atraumatic headache in an effort to minimise these associated monetary and non-monetary costs.



  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Statistical code is available from the corresponding author at john.gilbert{at} NHAMCS dataset is publicly available online at Consent was not necessary because the presented data are anonymised as part of the NHAMCS survey methodology.

Linked Articles

  • Primary survey
    Simon Carley