Abstract
In the 1980s as health care costs were burgeoning at remarkable speed, Dr David Eddy, an MD, PhD, bioengineer, and health care economist, was the first to apply Marcov models to clinical medicine, as well as the first to use and popularize the term “evidence-based medicine.” He wrote monthly articles in The Journal of the American Medical Association (JAMA) (1) addressing inconsistency in medical care and outlining the beginnings of evidence-based medicine. He was engaged by the Council of Medical Specialty Societies (CMSS) with funding by large United States (US) corporations to develop a methodology for guideline development that would hopefully lead to best practices and introduce more consistency in health care and reduce health care costs by eliminating unnecessary testing, surgery, and therapy.
In 1989 the American Academy of Neurology (AAN) developed the Quality Standards Subcommittee with a goal of developing clinical guidelines, which were called “practice parameters.” The Academy, as members of CMSS, was invited to select two representatives to attend two week-long summer sessions in Jackson Hole, Wyoming, with Dr Eddy and representatives of other specialty societies. So in the summer of 1991, Dr Jay Rosenberg and I became practice guideline developers utilizing the David Eddy method of meta-analysis using evidence-based analysis and highly rigorous methods for analyzing data in the literature. Each participant had to select a potential guideline. I chose to look at “The Utility of Neuroimaging in Chronic Headache.” This led to the first disease-specific practice parameter published by the AAN in Neurology in 1994 (2). Subsequently under the auspices of the American Headache Society and led by Dr Stephen Silberstein, the US Headache Consortium was founded with representatives from the American Academy of Family Physicians, AAN, American College of Emergency Physicians, American College of Physicians-American Society of Internal Medicine, American Osteopathic Association, and the National Headache Foundation with technical support by the Duke University Center for Health Policy Research and Education. This led to the development of four practice guidelines including neuroimaging, acute treatment of migraine, chronic treatment of migraine, and non-pharmacologic methods for migraine treatment using the rigorous evidence-based medicine approach as developed by Dr Eddy. The 2000 neuroimaging guidelines as well as the other guidelines were published online (3), with a summary of all four imaging and treatment guidelines published by Dr Silberstein in 2000 (4).
The guidelines suggested that for headaches that are “consistent with migraine” the incidence of clinically meaningful pathology was the same as in the general population and that neuroimaging is not necessary. Several caveats were included such as clinical red flags, an abnormal neurologic examination, onset over age 50, seizures, or significant change in headache pattern. The major problem with the guidelines was they were based on a meta-analysis that used poorly conducted retrospective studies with poor clinical diagnostic criteria for headache type. Despite that, the guidelines were widely used, cited, and became the basis for further guidelines. Since 2000, there have been multiple studies looking at the incidence of significant pathology in patients with headache receiving neuroimaging, and they are consistent in showing no real increase in pathology in patients with headaches that meet criteria for migraine (5,6). In addition, guidelines have been published by the European Federation of Neurologic Societies, and the Ontario Health Technology Assessment that agree with the AAN guidelines (7,8).
In this issue of Cephalalgia, Callaghan et al. have used the massive database of the National Ambulatory Medical Care Survey to extract information from more than 50 million headache visits to review the use of neuroimaging in a variety of headache disorders in order to determine the frequency of imaged patients who according to the guidelines do not need neuroimaging. They refer to this as “guideline discordant testing.” This was performed by analyzing a three-year database of outpatient visits. Sensitivity was increased by using all visits with a headache diagnosis made by the physician. Specificity was increased by using visits that included a primary migraine diagnosis when the patient indicated that headache was the primary problem. They then cross-referenced the incidence of brain computed tomography (CT) and/or magnetic resonance imaging (MRI) that were performed after the office visit. They also defined chronic headaches as any patient who indicated chronic on the survey that by definition was of duration >3 months. Using logistical regression techniques, they then estimated the likelihood of neuroimaging in various populations, and in various clinical settings, including after seeing a primary care provider vs. seeing a neurologist. They also looked at new complaint of migraine versus a flare-up of chronic headaches, without defining what chronic headache means other than present for greater than three months and not coded as migraine. They used a statistical method termed a Monte Carlo simulation, with use of a simple Markov model in order to predict the ordering of tests over a period of time of up to five years based on previous behavior.
What they found was not surprising. Headache accounted for 51.1 million visits in this database, with 25.4 million visits carrying a diagnosis of migraine according to International Classification of Diseases, ninth revision (ICD-9) coding. A total of 16.1 million had migraine as the primary diagnosis, and in 7.7 million of those, the patient also indicated headache as the primary problem. It was from this enriched sample that much of the analysis was performed. Of all patients with any headache diagnosis, 12.4% had neuroimaging, and 9.8% of those with a diagnosis of migraine had neuroimaging. This increased to 15.9% and 11.7%, respectively, if the primary diagnosis was migraine vs. migraine listed as the secondary or tertiary diagnosis. After removing patients with identified “red flags,” the neuroimaging frequency was reduced to 8.3% in migraine patients.
Other interesting tidbits gleaned from this study include no difference in imaging of migraine with aura and migraine without aura, increased imaging when a nonspecific headache diagnosis was used as compared with the patients who carried a diagnosis of migraine, increased imaging when the patient is new to a practice, and increased neuroimaging by neurologists as compared with primary care providers. Using a statistical predictive model, they also estimated that a patient with new migraine has a 39% chance of receiving neuroimaging in the next five years, while those with a flare-up of chronic headache have a 51% chance of receiving neuroimaging in the same time frame.
They conclude that neuroimaging in patients with migraine and chronic headache without red flags is over-used, especially if the predicted five-year imaging is accurate. They suggest that referral to specialists will not reduce unnecessary testing, but that enhancing continuity of care to avoid more first visits to new health care providers will likely lead to less unnecessary neuroimaging.
This is an interesting paper that provides evidence of neuroimaging when the widely available guidelines suggest imaging is not necessary. I found the statistical methods to be very complicated, especially the predictive value using the Monte Carlo simulation. The accuracy of the study is affected by the appropriate use of diagnostic coding. Many physicians, especially when they are performing their own coding, do not accurately code, use very generalized diagnoses, or use the diagnoses easily obtained on their billing sheet or in their electronic medical record. Studies like this depend on accurate coding. We also do not have data concerning what other factors may have been involved in the decision to image a headache patient, such as a progressively worsening headache, a change in headache pattern, or atypical features and comorbidity.
While the previous guidelines are fairly clear on the lack of utility in imaging migraine, the data concerning imaging in chronic headache are not so clear. In the 1994 Utility of Neuroimaging paper (2), it was clear that a diagnosis of “chronic headache” was a heterogeneous group of headaches, and in that group, a higher incidence of pathology was noted. The authors reference a guideline published by the American College of Radiology (ACR) in 2000 and updated online in 2013 (9) in reference to imaging in chronic headache. However, that guideline states that in patients with chronic headache with a normal neurologic examination and no change in pattern, MRI may be appropriate. They go on to say that when headache is in the orbit, periorbital region or skull base, then CT or MRI is usually appropriate. The ACR also had a Choosing Wisely guideline on neuroimaging in headache with a normal neurologic examination. The ACR Choosing Wisely Guideline (10) is not specific, but rather just states neuroimaging is not indicated in most stable headaches. The authors do not reference any other papers or guidelines that indicate neuroimaging in chronic headache is not necessary. I take issue with the assumption of the authors that neuroimaging is inappropriate in chronic headache as this is not part of any guideline.
So what is the bottom line? Most headache specialists agree that neuroimaging of patients with migraine is not necessary, but still many continue to use neuroimaging even when we know the study will be normal, or findings, if any, are unlikely to be related to migraine. I believe several forces drive this. The fear of litigation remains one of the most powerful motivators when one looks at physician testing behavior. We are all fearful of missing the incidental tumor or other lesion. Several years ago a group of headache physicians were informally surveyed and asked “if the evidence was clearly convincing that there is no increased risk of intracranial pathology in migraine patients, how many of you will still check a study in your new headache patients?” Greater than 50% indicated that they would still check a neuroimaging study.
The other major factor is expectation, both by the patient and the referring physician. Patients these days have very high expectations, and nearly all have visited the Internet, so they come to us with a list of concerns and requests. CT these days never seems to be adequate, so they want an MRI as well, and how about a magnetic resonance angiography (MRA)? I have a useful approach I have used in the clinic. When patients with migraine request a scan, I indicate that while it is not medically necessary, if they would like an MRI to ease their worry, I am happy to order it, but insurance will not cover it because it is not medically necessary. This puts the ball in their court, and rarely have I had a patient take me up on my offer. In these days of patient satisfaction surveys and physician ratings based on satisfaction that may be tied to compensation, we all want happy patients, and saying “No” does not go over well. There is also the expectation of the referring physician. They may be sending you a patient just to get imaging, as their utilization review department will not allow imaging in headache without it being ordered by a neurologist or a headache specialist. I find this less intimidating than the demanding patient, as the referring physician has covered himself or herself by sending the patient to the specialist, and as long as the consult discusses the reasons for not needing neuroimaging, they are usually satisfied.
Currently there are several guidelines concerning neuroimaging in migraine, including the recently published Choosing Wisely campaign, which was sponsored by The American Board of Internal Medicine Foundation (11) as well as the practice parameter of the AAN, both of which recommend against neuroimaging in migraine with a normal neurologic examination without red flags, seizures, or change in headache pattern. I would ask all of the readers to be circumspect in ordering neuroimaging and try to do the “right thing.”
Footnotes
Conflict of interest
None declared.