Headache Neuroimaging

A Survey of Current Practice, Barriers, and Facilitators to Optimal Use

Evan L. Reynolds PhD; James F. Burke MD, MS; Lacey Evans MPH; Faiz I. Syed MD, MS; Eric Liao MD; Remy Lobo MD; Wade Cooper DO; Larry Charleston IV MD, MSc; Brian C. Callaghan MD, MS


Headache. 2022;62(1):36-56. 

In This Article


Neuroimaging is overused in patients with headaches, however, factors that lead to suboptimal utilization are unknown. We administered a survey to 431 VA headache clinicians and found that a majority of clinicians believed that neuroimaging is overused for patients with headaches and generally believed that much neuroimaging is low value, yet many clinicians still indicated they would order neuroimaging in low-risk scenarios where the probability of changing patient management is small. Conflicting guidelines that lack details regarding neuroimaging utilization in the presence of red flags, asymmetric valuation of harms via omission than commission, and unbalanced beliefs in neuroimaging harms and benefits likely contribute to overutilization. Importantly, potential red flags without strong evidence led to much higher neuroimaging; therefore, future studies are needed to further define the role of currently used red flags. We also found that APCs indicate use of headache neuroimaging more often than physicians and are more amenable to possible interventions, making them an ideal target for future optimization efforts.

In the hypothetical scenarios for patients with headaches, the presence of potential red flags resulted in increased neuroimaging utilization. We found that clinicians typically were able to correctly identify patients at high risk of finding management changing lesions and subsequently decide whether to perform neuroimaging. However, all clinician types indicated that they would order neuroimaging over 25.0% of the time in patients with migraine headaches and no red flags despite guidelines that explicitly recommend against this practice. Furthermore, there was a substantial jump in utilization for patients with potential red flags. Specifically, we found that clinicians indicated they would order neuroimaging 67.1% of the time in scenarios with red flags that have limited evidence to support their use. The use of "red flags" to identify underlying medical conditions originated in back pain,[31,32] but has been extended to a number of settings,[33] including other neurologic conditions, such as Parkinson's disease[34] and cauda equina syndrome.[35] In conditions other than headache, systematic reviews typically find that red flags result in low sensitivity and a high false positive rate.[31,34,35] On the other hand, the use of red flags are particularly important to primary care physicians who need to surveil a wide range of potential conditions.[36] Current guidelines recommend against neuroimaging for patients with headaches, except in the presence of red flags; however, red flags are inconsistently described across guidelines and are based on limited or non-existing evidence,[37] which likely results in overutilization of neuroimaging and subsequent false positive findings. One retrospective study found that 77% of patients with headaches had at least one red flag that justified neuroimaging, but only 3.4% of those patients had an abnormal MRI finding,[38] suggesting that relying on red flags may lead to more harms than benefits. Another study assessed the diagnostic accuracy of different red flags in patients with headaches,[39] finding limited sensitivity and specificity even when considering the best combination of red flags. More studies with larger sample sizes are needed to validate these results and ultimately update clinical guidelines. Our study provides additional evidence that headache clinicians rely on red flags to make neuroimaging decisions. Therefore, more evidence is needed to determine which red flags are consistently associated with benefits following neuroimaging that exceeds any harms associated with these tests.

The majority of clinicians believed that neuroimaging was overused in patients with headaches, but also believed that the likelihood of benefits was greater than the likelihood of harms in all three different clinical scenarios. Although clinicians may believe neuroimaging is broadly overused for patients with headaches, ordering neuroimaging is the sensible clinical decision when the perceived likelihood of benefits outweighs harms. These contradictory beliefs imply disagreements with headache neuroimaging guidelines, which is a primary reason clinicians make guideline-discordant decisions according to a systematic review.[14] Therefore, interventions should attempt to align clinicians' perceptions of neuroimaging harms and benefits more closely with the best available data, which may ultimately improve decision making. In addition to improving the guidelines themselves, embedding decision-making rules through a CDSS in the EHR could improve neuroimaging utilization.[17,40] Specifically, including an algorithm that uses patient characteristics to predict the probability of harm, benefit and potential for change in management could correct clinicians' perceptions of neuroimaging harms and benefits and ultimately increase guideline-concordant neuroimaging decisions.[41] Our survey also indicated that many clinicians were open to interventions that provide reminders of optimal use within the EHR.

Despite following headache neuroimaging guidelines, more clinicians were bothered by harms stemming from the omission of neuroimaging (88.5% of respondents) compared to commission (77.6% of respondents). This was surprising as examples of "omission bias"[42] are common in healthcare[43,44] and result in clinician inaction when caring for patients. Interestingly, a multicenter study at VA health care systems, found that harms via omission were more common and typically more serious compared with harms via commission.[45] The unbalanced burden of omission and commission likely results in overutilization of neuroimaging and may be explained by the fact that 29.9% of clinicians believed underlying headache-causing diagnoses are missed during the initial evaluation and 80.4% believed it is their responsibility to never miss a headache caused by a brain tumor. In addition, this perception of clinician responsibility directly contradicts sentiments that headache neuroimaging is overutilized. Unfortunately, the frequency and severity of harms stemming from headache neuroimaging decisions are not well known. Future studies that determine the downstream harms of guideline-discordant neuroimaging may decrease the current imbalance of perceived harms via commission and omission. Given that VA clinicians have immunity from malpractice liability as part of the United States Federal Tort Claims Act, it was surprising to find that 26.9% of clinicians often/almost always consider malpractice concerns when making headache neuroimaging decisions. One possible explanation may be that clinicians had dual appointments at the VA and outside institutions, without the protections of the Federal Tort Claims Act. Nonetheless, this surprising result in this population highlights the importance that malpractice concerns have in headache neuroimaging decisions, which may further contribute to the differences in burden from acts of omission versus commission, especially in healthcare systems other than the VA.

Compared with physicians, we found that APCs overutilized neuroimaging more often in low-risk scenarios. A similar conclusion was made in a study of Medicare clinicians, where APCs were more likely to order diagnostic imaging compared with primary care clinicians.[46] Overutilization by APCs may have been driven by strong beliefs in neuroimaging benefits, smaller beliefs in harms stemming from a false positive MRI finding, less confidence in the ability to make decisions to obtain an MRI, and less awareness of neuroimaging overutilization. In addition, the fact that APCs believed they would identify a brain tumor in 21.6% of patients with a normal neurologic examination and no other red flags, suggests that these clinicians lack accurate knowledge regarding the likelihood of finding abnormalities and brain tumors in patients with headaches, which may ultimately drive overutilization. Although guideline-discordant neuroimaging decisions were common amongst APCs, these clinicians were consistently more receptive to implementation interventions compared to physicians, especially those involving improved guidelines. Specifically, only 30.1% of APCs reported having good knowledge of headache neuroimaging guidelines, but 89.3% believed updated guidelines would improve optimal utilization. Therefore, APCs are an ideal target population to improve neuroimaging utilization and should be the focus of future implementation initiatives.

In three separate headache neuroimaging scenarios, we found that neurologists would utilize neuroimaging at similar rates to primary care clinicians. This was particularly surprising in the low-risk scenario with minor red flags, given that neurologists reported having significantly better knowledge of headache neuroimaging guidelines. This apparent contradiction may be explained by findings from a systematic review, which found that specialists were more likely to follow clinical guidelines, but were also more likely to perform diagnostic testing compared with general practitioners.[47] For neurologists, the decreased testing associated with guideline adherence is likely offset by the increased diagnostic testing that is typically performed by specialists. Although neurologists stated that they had a better knowledge of clinical guidelines, their belief that neuroimaging resulted in a higher likelihood of changed management and smaller likelihood of harms, false positive findings, unnecessary procedures, and unnecessary consultations may have increased their guideline-discordant utilization. Neurologists also had increased confidence to order, interpret, and act on MRI findings, and ordered neuroimaging upon patient request more often than primary care clinicians, which may also have increased utilization beyond guideline recommendations. Another possibility is that neurologists overestimated their actual knowledge of headache neuroimaging guidelines, which may explain the apparent contradiction. Importantly, primary care clinicians were consistently more accepting of the potential interventions to improve neuroimaging decisions compared with neurologists. Therefore, separate strategies are necessary to improve patterns of headache neuroimaging utilization for neurologists and primary care clinicians. For neurologists, improving the clinical guidelines themselves is essential, especially with increased consistency regarding red flags and increased details of the potential harms and the likelihood of changes in patient management. For primary care clinicians, they are more likely to respond to a variety of implementation strategies to reduce unnecessary neuroimaging for patients with headaches.

Study limitations include the potential for nonresponse bias. However, our response rate was comparable to other physician surveys. Other limitations include the lack of generalizability to clinicians outside of the VA. Specifically, it is unknown whether malpractice concerns for VA clinicians are generalizable to other healthcare systems. Furthermore, we were able to identify all VA neurologists and a representative sample of VA primary care and APC clinicians, which makes generalizability within the VA robust. Our survey only details what clinicians indicate they would do, but this does not necessarily reflect their actual practice. Moreover, the limited clinical information in our survey's scenarios may not have adequately accounted for the diverse presentations that occur in routine care. Although survey items were developed by following the well-validated TDF, the reliability and validity of the individual survey items are unknown. In addition, we were unable to account for potential reporting bias, especially in regard to the importance of financial incentives in neuroimaging decision making. In addition, the online survey did not include a review step or completeness check, which is a limitation of the survey design. We also investigated many comparisons without statistical corrections; however, these results are meant to be hypothesis generating and not definitive.