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


Survey Instrument

We developed a 60-item survey following the Theoretical Domains Framework (TDF).[24] The TDF was specifically designed for the purpose of improving clinician behavior in implementation research, has been rigorously validated,[25] has been used to understand clinician behavior in a variety of contexts,[26,27] and successfully applied in prior implementation initiatives.[28–30] Survey items were developed through a consensus process with the study authors, along with qualitative feedback provided by two neurologists and two primary care clinicians on survey clarity, simplicity, and alignment with key questions. The survey consisted of Likert Scale questions assessing clinician opinions on the harms, benefits, confidence, barriers, and attitudes toward MRI utilization for patients with headaches. The survey also posed three different headache neuroimaging scenarios to determine how often clinicians believed they would order an MRI, find abnormalities, find tumors, and cause harm or benefit to the patient. The risk scenarios describe hypothetical patients with low risk of management changing lesions (migraine with normal neurologic examination) and high risk (lung cancer, subacute onset, and incoordination). There is no consensus definition of red flags for headache neuroimaging, however, to understand the effect that potential red flags with limited evidence to support their use have on neuroimaging decisions, we created another low-risk scenario with three such potential red flags (tingling in both hands, worse with exertion, and new onset in a 65-year-old patient) that have been suggested in previous studies. The complete survey instrument is included as Supporting File S1.

Sampling Design

Eligible clinicians were identified using the VA Corporate Data Warehouse. Specifically, we identified all nonresident clinicians in primary care and neurology that saw at least three patients with headaches in the preceding 6 months. This included physicians (authors M.D. and D.O.) and APC. The APCs included those that identified themselves as a nurse practitioner (NP) or physician assistant (PA). Our a priori power calculations determined that by targeting 1500 clinicians, at a 50% response rate, with a type 1 error rate of 0.05, for a given response on a five-response Likert scale survey item, if the "true" proportions varied from 10% to 50%, we could estimate parameters within the following 95% confidence intervals (CIs): 10%, 7.9%–12.4%, 20%, 17.1%–23.1%, 30%, 26.7%–33.5%, 40%, 36.4%–43.7%, and 50%, 46.3%–53.7%. On September 11, 2018, a prenotification email was sent to the survey population describing the survey and informing them a link will be sent the following week. On September 19, 2018, each member of the survey population received an email containing a link to an anonymous, online, closed survey using Research Electronic Data Capture (REDCap). From September 27, 2018, to October 18, 2018, nonresponders received four email reminders. The REDCap survey did not include a completeness check or review step. On October 29, 2018, and November 30, 2018, nonresponders were sent a follow-up paper survey via mail. Respondents were entered into a lottery for a chance to win one of fifteen $100 gift cards. The analysis included all survey responses, including those that were not fully completed.

Statistical Analysis

Descriptive statistics were used to summarize survey responses, stratified by clinician type (physicians or APCs, NPs, and PAs) and specialty (physicians: neurology or primary care). Specifically, for categorical survey responses, we reported the frequency and percentage of each survey item response. To summarize ordinal survey items, we also calculated the median, 25th, and 75th percentile of responses. For the continuous responses corresponding to the three headache neuroimaging scenarios, we have reported the mean and SD (mean (SD)), percentage of time clinicians reported that they would order an MRI, find abnormalities, find tumors, and cause harm or benefit to the hypothetical patient.

Wilcoxon rank sum tests were used to compare responses to ordinal survey questions between clinician types (APC vs. physicians) and specialties (physicians: neurology vs. primary care). One-way analysis of variance (ANOVA) was used to assess differences in responses to the three headache neuroimaging scenarios across clinician groups (APC, neurologists, and primary care). Paired t-tests were used to assess within-clinician changes in the percentage of time clinicians expected to order an MRI, find abnormalities, find tumors, and cause harm or benefit across the three headache neuroimaging scenarios. Assumptions of normality for the one-way ANOVA and paired t-tests were assessed using histograms. In the case of non-normality, Wilcoxon signed rank tests and Kruskal-Wallis tests were used instead of paired t-tests and one-way ANOVA, respectively.

Available case analysis was used to compare survey responses between clinician types (APC vs. physicians) and specialties (physicians: neurology vs. primary care). In addition, two-tailed p values were calculated and statistical significance was determined using a p value threshold of 0.05.

The primary analyses were to describe clinician responses to the three neuroimaging scenarios. All other analyses were secondary. All analyses were specified a priori, and these data are not in publication or reported elsewhere.

All analyses were completed using R version 3.6.1.

Ethics Approval and Participant Consent

This study was approved by the VA Institutional Review Board. Given that this study posed no more than minimal risk to clinicians, the Institutional Review Board deemed that it was not necessary to obtain written informed consent.