ISSN 0017-8748 doi: 10.1111/head.12716 Published by Wiley Periodicals, Inc.
Headache C 2015 American Headache Society V
Research Submission Development and Validation of a Four-Item Migraine Screening Algorithm Among a Nonclinical Sample: The Migraine-4 A. Brooke Walters, PhD; Todd A. Smitherman, PhD, FAHS
Objectives.—The present study sought to develop and validate a screening algorithm from the ICHD-3 beta diagnostic criteria for migraine utilizing a nonclinical sample. The goal was to determine the most sensitive and specific symptoms for differentiating migraine from other headache disorders and to validate the derived symptom algorithm as a screening measure. Background.—Despite its prevalence and impact, migraine remains under-recognized and under-treated. The U.S. Headache Consortium recommended development and dissemination of validated screening measures as a means to improve diagnosis. Methods.—Participants were 1829 young adults (71.5% female; 74.4% white; mean age 5 19.09 years [SD 5 2.05]) who reported headache via computerized diagnostic interview. From this group, 158 were found to have ICHD episodic or chronic migraine and were randomly split into experimental and holdout validation samples. Within the experimental sample, receiver operating characteristic (ROC) curve data were obtained for each candidate symptom (item); backward stepwise logistic regression analysis was performed among the items with the most predictive likelihood ratios to determine the optimal model for differentiating migraine from non-migrainous headache. The retained four-symptom algorithm was then validated among the holdout sample, in which various cutoff points were compared to gold standard diagnosis via ROC curve estimations to determine the optimal operating point (OOP) of the algorithm as a screening measure. Results.—Attack duration of 4-72 hours (100% [95% CI 5 95-100%]), severity 5 (91% [82-97%]), photophobia (90% [80-96%]), and phonophobia (90% [80-96%]) showed the highest sensitivity, while vomiting (98% [96-99%]), duration of 4-72 hours (92% [90-94%]), nausea (89% [86-91%]), and headache-related disability (88% [85-91%]) showed the highest specificity. The optimal retained model (Migraine-4) included: duration of 4-72 hours, nausea, photophobia, and phonophobia. Among the holdout validation sample, the OOP was positive endorsement of three out of four symptoms, which had a sensitivity of 94% (95% CI 5 87-98%), a specificity of 92% (90-94%), and an area under the curve of 93% (90-96%; 1LR 5 12.37, 2LR 5 0.06, PPV 5 67%, NPV 5 99%). Conclusions.—The optimal model shares some similarities with previous models but performed better than prior screeners at differentiating migraine from other headache presentations. The Migraine-4 has utility in identifying migraine among nonclinical and young adult samples. Further research with this measure is warranted to determine its utility with treatment-seeking patients and validity in direct comparison to established screening instruments. Key words: migraine, headache diagnosis, screening, assessment, sensitivity and specificity (Headache 2016;56:86-94)
Address all correspondence to T.A. Smitherman, Department of Psychology, University of Mississippi, 207 Peabody, Oxford, MS 38677, USA, email: [email protected]
INTRODUCTION The diagnostic criteria for migraine are outlined in the 2013 International Classification of Headache Disorders, Third Edition (beta version; ICHD-3 beta).1 Migraine is defined as a recurrent headache
Accepted for publication September 10, 2015.
Conflict of Interest: None.
From the Department of Psychology, University of Mississippi, Oxford, MS 38677, USA.
86
Headache disorder characterized by a history of at least five headache attacks of 4-72 hours duration. The headache attacks must have at least two of the following four features: (1) unilateral location; (2) pulsating (throbbing) quality; (3) moderate or severe pain; and (4) causing aggravation or avoidance or routine activity AND either nausea, vomiting, or both photophobia and phonophobia. Migraine can occur with or without aura symptoms, which are temporary sensory distortions, typically visual in nature, that precede headache onset by an hour or less and then resolve. Migraine is further dichotomized based on its frequency of occurrence: attacks with a frequency of less than 15 days per month are classified as episodic migraine (EM), and the much rarer occurrence of attacks on 15 or more days per month is termed “chronic” migraine (CM). Migraine affects 12-17% of the population annu2–5 ally, and the lifetime incidence is 43% for women and 18% for men.6 Data from the World Health Organization indicate that migraine is the third most common medical condition and the eighth leading cause of disability worldwide.5 Because of its high prevalence and negative impact, migraine is a leading cause of neurologist visits7 and the fourth most common reason for emergency department visits.3 Despite its prevalence and impact, migraine remains under-recognized, underdiagnosed, and undertreated. Roughly half of those with migraine never receive a diagnosis from a physician,8,9 and only 26% receive minimally appropriate care.10 Significant barriers to appropriate care include low rates of physician consultation among males and those with limited insurance/economic resources,10,11 misdiagnosis,12 overuse of neuroimaging,13,14 and inadequate or improper treatment.10 One means of improving the diagnosis and treatment of migraine is developing and validating cost-efficient screening measures, as recommended by the U.S. Headache Consortium.15 Prior efforts in this domain have included statistical algorithms of those symptoms most predictive of migraine in a given sample,16–18 as well as the validation of screening questionnaires.19–21 Few studies have attempted to both determine an algorithm for migraine diagnosis and subsequently validate it as a screening measure,22–24 and these efforts were prior to publication
87 of the ICHD-3 beta. Previous studies also have been quite variable in terms of reference groups, candidate items/symptoms, and findings, with only nausea consistently discriminating migraine from other headache types.16–19,22,23,25,26 In light of these limitations and recognition of a need for cost-efficient migraine screeners among individuals who do not regularly seek treatment (eg, underrepresented groups, in population studies, and for research purposes), the current study endeavored to develop a migraine screener comporting with ICHD-3 beta for use with nonclinical samples. Young adults represented the sample of interest because of the need to identify migraine early in the lifespan to foster early recognition and treatment in an effort to prevent headache progression or chronification, and in light of data that migraine prevalence increases precipitously during young adulthood3 and confers substantial disability.27–29 The present study thus was designed to (1) determine statistically the most sensitive and specific ICHD-3 beta symptoms differentiating migraine from other common headache types among a nonclinical population (ie, algorithm development) and then to (2) validate the resulting symptom algorithm among a holdout sample from the larger overall sample (ie, screener validation) and identify an optimal cutoff point for use. In light of previous findings, nausea was hypothesized to be the strongest discriminator of migraine, but no other a priori hypotheses were made about other candidate symptoms due to inconsistency in previous findings.
METHODS Participants and Procedure.—This study received ethical approval from the Institutional Review Board at the University of Mississippi, and all participants provided informed consent via electronic signature prior to providing data. Participants were young adults at a southeastern university who participated voluntarily in exchange for modest course credit. As part of an ongoing data collection effort,30–34 from August 2012 to December 2013 they completed an online battery of measures, including a computer-administered structured diagnostic interview for headache. Those meeting ICHD-3 beta criteria for EM (with or without
88 aura) or chronic migraine comprised the migraine group. Those with episodic, chronic, or probable tension-type headache (TTH), cluster headache, posttraumatic headache, medication overuse headache (MOH), or those who complained of headache but could not be classified comprised the nonmigraine group. Thus, participants with any headache presentation other than migraine or probable migraine (PM; ie, meeting all but one criterion for migraine) comprised the non-migraine comparison group. For secondary analyses, probable migraineurs were included as part of the migraine group because younger adults often have otherwise prototypical migraine attacks of duration less than 4 hours (ie, missing only Criterion B).35,36 Measures.—Structured Diagnostic Interview for Headache-Revised.—The well-validated computeradministered Structured Diagnostic Interview for Headache37 was modified to comport with ICHD-3 beta criteria and be administered online. Modifications were slight and included minor rewording of some items for clarity, addition of forced-choice format for querying headache duration, and removal of appendix items that afforded detailed assessment of cluster and secondary headache disorders. The Structured Diagnostic Interview for Headache-Revised (SDIH-R) includes items pertaining to each ICHD diagnostic symptom of migraine (including aura) and TTH, as well as additional items pertaining to headache frequency, disability, and screening for other primary and secondary headache disorders. Precise diagnostic criteria for cluster headache, posttraumatic headache, and MOH were not assessed in the computerized SDIH-R; rather, they were screened for by inquiring about a past diagnosis of cluster headache (cluster), if headache began after head injury (posttraumatic), or prolonged use of medications to treat headache on 3 days/week coinciding with increased headache frequency (MOH). Statistical Analyses.—The current study occurred in two phases: the dataset was randomly split by approximately 50% using SPSS random sampling (without variable matching), with SDIH-R responses from the first half of participants (ie, experimental sample) used for algorithm development and responses from the second half (ie, vali-
January 2016 dation sample) used for validating the resulting symptom algorithm as a screening instrument. Algorithm Development in the Experimental Sample.—Chi-square analyses and independent samples ttests were used to assess whether the experimental and validation samples differed on demographic or headache variables. For algorithm development, twoby-two tables were constructed using each of the ICHD-3 beta migraine symptoms (present vs absent) and headache diagnoses (migraine vs non-migraine); sensitivity, specificity, predictive values, and likelihood ratios were calculated for each symptom. Receiver operator characteristic (ROC) curves were then plotted for each of the one-variable symptom models to determine which were most predictive of a migraine diagnosis, as indexed by the area under the curve (AUC). Next, using the optimal model guidelines in Martin et al (2005),16 symptoms were eliminated if they did not have 1LRs > 4.5 or 2LRs < 0.25. A backward stepwise logistic regression analysis was then performed using the remaining symptoms to determine the optimal model for differentiating migraine from non-migraine headache. This resulting model was used as the screening measure (Appendix) validated in the second phase of the study. Screener Validation in the Validation Sample.—For screener validation, two-by-two tables were constructed using each possible cutoff point of the screening measure and headache diagnosis, from which sensitivity, specificity, LRs, and PVs were calculated among the holdout validation sample. An ROC curve then was plotted to compare scores on the screening measure with the gold standard (ie, EM or CM diagnosis as per full ICHD-3 beta criteria) and the screener’s utility in differentiating migraine from other headache types. ROC curve analysis was used to assess all possible cutoff points of the new screener and to determine the optimal operating point (OOP)38 of the screener. Finally, the performance of the screener was examined as a function of gender and race, as well as among probable migraineurs, to supplement the main analyses.
RESULTS Sample Comparison.—The total sample consisted of 1966 participants who responded positively to
Headache
89 Table 1.——Demographics and Headache Diagnoses in Experimental and Validation Samples Experimental Sample (n 5 887)
Age M (SD) Gender Female Male Race White Black Other
19.05 (1.96)
Validation Sample (n 5 942)
P5
19.12 (2.13)
628 (70.8%) 259 (29.2%)
680 (72.2%) 262 (27.8%)
661 (74.5%) 147 (16.6%) 79 (8.9%)
699 (74.2%) 184 (19.5%) 59 (6.3%)
.48 .51
Entire Sample (N 5 1829)
19.09 (2.05) 1308 (71.5%) 521 (28.5%)
.88 1360 (74.4%) 331 (18.1%) 138 (7.5%)
Headache Type
CM EM w/out aura EM w/aura All migraine †PM CTTH ETTH PTTH Cluster Posttraumatic MOH No diagnosis All non-migraine
23 29 17 69 266 12 184 219 18 40 1 78 552
(2.6%) (3.3%) (1.9%) (7.8%) (30.0%) (1.4%) (20.7%) (24.7%) (2.0%) (4.5%) (0.1%) (8.8%) (62.2%)
18 45 26 89 301 16 177 203 18 52 1 85 552
(1.9%) (4.8%) (2.8%) (9.4%) (32.0%) (1.7%) (18.8%) (21.5%) (1.9%) (5.5%) (0.1%) (9.0%) (58.6%)
.32 .10 .23 .14 .36 .55 .29 .11 .85 .32 .97 .86 .14
41 74 43 158 567 28 361 422 36 92 2 163 1,104
(2.2%) (4.0%) (2.4%) (8.6%) (31.0%) (1.5%) (19.7%) (23.1%) (2.0%) (5.0%) (0.1%) (8.9%) (60.4%)
Mean Headache Days Per Month (0-30)
Mean Headache Severity (0-10)
18.54 (SD 5 4.28) 6.86 (4.37) 6.77 (3.27)
6.41 (SD 5 1.22) 6.03 (1.79) 6.33 (1.34)
7.33 17.96 4.88 5.09 9.22 7.87 21 2.57
5.19 4.82 3.95 3.67 5.39 5.11 7 2.82
(5.72) (4.77) (3.17) (4.48) (5.08) (6.03) (5.66) (2.45)
(1.63) (1.28) (1.36) (1.48) (1.89) (1.85) (2.83) (1.55)
†Probable migraineurs were excluded from the primary analyses, but were included as part of the migraine group in secondary analyses. CM 5 chronic migraine; CTTH 5 chronic tension-type headache; EM 5 episodic migraine; ETTH 5 episodic tension-type headache; MOH 5 medication overuse headache; PM 5 probable migraine; PTTH 5 probable tension-type headache.
the initial question “Do you ever get headaches?” Excluding 137 with missing data, 1829 were retained and classified into 10 headache categories (see Table 1) based on their SDIH-R responses. The sample of 1829 was split randomly by approximately 50% to ensure equivalence between the experimental (n 5 887) and validation (n 5 942) samples. Nonsignificant chi-square tests (for gender, race [white vs non-white], and headache diagnosis) and independent samples t-tests (for age, headache frequency, and headache severity) confirmed that the two samples did not differ significantly on any of these variables (see Table 1). Algorithm Development.—Within the experimental sample, the performance of each migraine symptom was compared to the gold standard diagnosis of migraine as acquired via the SDIH-R. Two-by-two tables were created showing the number of migrai-
neurs and non-migraineurs endorsing each symptom (N 5 621), from which sensitivity, specificity, positive and negative likelihood ratios, and positive and negative predictive values were calculated (see Table 2). Headache duration of 4-72 hours showed the highest sensitivity (100%), followed by severity 5 (91%), photophobia (90%), and phonophobia (90%). In contrast, vomiting showed the highest specificity (98%), followed by duration of 4-72 hours (92%), nausea (89%), and disability (defined as headaches interfering with work, school, or personal life; 88%). Duration of 4-72 hours had the largest AUC (96%), followed by photophobia (78%), severity 5 (78%), and nausea (77%). Unilateral location, pulsing quality, and worsening by physical activity were ruled out as part of the optimal model as they did not have an 1LR > 4.5 or 2LR < 0.25. By way of backward stepwise logistic regression analysis (performed on the
(99-100%) (87-93%) (92-97%) (96-99%) (94-98%) (93-97%) (88-93%) (96-99%) (95-99%) (93-97%) 100% 91% 95% 98% 96% 95% 91% 98% 98% 95% (52-71%) (10-19%) (13-21%) (19-30%) (18-29%) (33-52%) (33-74%) (20-32%) (16-25%) (31-50%) Disease prevalence 5 11% (excluding probable migraineurs).
Duration (4-72 hours) Unilateral Pulsing Severity 5 Worsened by activity Nausea Vomiting Photophobia Phonophobia Disability
100% 49% 80% 91% 78% 65% 19% 90% 90% 64%
(95-100%) (37-62%) (68-88%) (82-97%) (66-87%) (53-76%) (11-30%) (80-96%) (80-96%) (51-75%)
92% 62% 49% 65% 67% 89% 98% 67% 55% 88%
(90-94%) (57-66%) (45-54%) (61-69%) (63-71%) (86-91%) (96-99%) (63-71%) (51-60%) (85-91%)
13.14 1.28 1.57 2.60 2.39 5.90 9.42 2.71 2.02 5.37
(9.83-17.58) (.99-1.67) (1.36-1.82) (2.27-2.97) (2.01-2.85) (4.40-7.91) (4.39-20.20) (2.35-3.13) (1.78-2.28) (4.02-7.17)
0.82 0.41 0.13 0.33 0.39 0.83 0.15 0.18 0.41
– (0.65-1.05) (0.26-0.66) (0.06-0.29) (0.21-0.51) (0.28-0.54) (0.74-0.93) (0.08-0.31) (0.09-0.37) (0.30-0.56)
62% 14% 16% 25% 23% 42% 54% 26% 20% 40%
PPV 2LR 1LR Specificity Sensitivity
Table 2.——Experimental Sample: Migraine Symptom Performance (95% Confidence Intervals)
NPV
96% 56% 65% 78% 72% 77% 58% 78% 72% 76%
(95-98%) (49-63%) (58-71%) (73-83%) (66-79%) (70-84%) (50-66%) (73-83%) (67-78%) (69-83%)
January 2016
AUC
90
remaining seven symptoms), the retained optimal model consisted of duration of 4-72 hours (B 5 21.08), nausea (B 5 1.18), photophobia (B 5 1.87), and phonophobia (B 5 1.54). When including those with PM in the migraine group (N 5 887), photophobia showed the highest sensitivity (90%), followed by phonophobia (89%), pulsing quality (77%), and severity 5 (73%). Photophobia showed the highest AUC (78%), followed by phonophobia (72%), nausea (70%), and severity 5 (69%). Optimal likelihood ratios when including those with PM were found for nausea (B 5 1.77) and photophobia (B 5 2.73). Because this was a supplementary aim of the present study, validation was not attempted on this two-symptom model. Instead, the validity of the original foursymptom screener was explored within probable migraineurs in subsequent validation analyses. Screener Validation.—To validate the prior fourfactor model as a screening instrument in the holdout validation sample, two-by-two tables were created for each possible cutoff point of the screener, from which sensitivity, specificity, positive and negative likelihood ratios, and positive and negative predictive values were calculated (see Table 3). The OOP for the screener was positively endorsing three out of four items, which had a sensitivity of 94% and a specificity of 92% (1LR 5 12.37, 2LR 5 0.06, PPV 67%, NPV 99%). ROC curve analyses were used to determine the accuracy of each cutoff point as compared to the gold standard diagnosis of migraine (see Table 3). The AUC of positively endorsing one or more items was 69% [95% CI: 65-74%], of two or more items was 87% [85-90%], of three or more items was 93% [90-96%], and of all four items was 78% [72-85%]. When including probable migraineurs, the four-item screener showed an OOP of responding positively to two or more items (sensitivity 5 98% [96-99%], specificity 5 75% [71-78%], PPV 5 74% [70-77%], NPV 5 99% [97-99%], AUC 5 87% [84-89%]). All items in the screening algorithm were significantly intercorrelated (rs from 0.18 to 0.51, Ps .01) and were highly correlated with the OOP of the screener (rs from 0.42 to 0.71, Ps .01). Group Differences in Screener Performance.— The screening measure performed better statistically
91
69% 87% 93% 78% 100% (98-100%) 100% (99-100%) 99% (98-100%) 94% (91-95%)
(65-74%) (85-90%) (90-96%) (72-85%)
among men than in women at all four cutoff points (Ps .001; X2 at OOP 5 26.896, P < .001); however, this statistically significant difference is almost certainly a function of the large sample size and not clinically meaningful, as the AUC at the OOP of the screener was very high among both genders: 94% among men vs 93% among women. The screener performed equally well among white and non-white individuals.
Disease prevalence 5 13% (excluding probable migraineurs).
One or more Two or more Three or more All four
100% 100% 94% 57%
(96-100%) (96-100%) (87-98%) (46-68%)
39% (35-43%) 75% (71-78%) 92% (90-94%) 99% (98-100%)
1.64 3.96 12.37 61.84
(1.53-1.75) (3.42-4.57) (9.17-16.67) (25.37-150.73)
– – 0.06 (0.03-0.15) 0.43 (0.34-0.55)
21% 39% 67% 91%
(17-25%) (33-46%) (58-75%) (80-97%)
AUC NPV PPV –LR 1LR Specificity Sensitivity
Table 3.——Validation Sample: Screener Performance at Different Cutoff Points (95% Confidence Intervals)
Headache
DISCUSSION The current study sought to identify an algorithm for identifying ICHD-3 beta migraine and to validate this algorithm as a screening instrument in a nontreatment-seeking sample of young adults. The four characteristics of duration of 4-72 hours, nausea, photophobia, and phonophobia comprised the optimal model for differentiating migraine from non-migraine headache and showed high sensitivity, specificity, and accuracy at the OOP of three out of four positive endorsements. The screening measure had very strong clinical utility among men and women and among white and non-white individuals at its OOP. This is the first study to find this unique combination of these four symptoms as an effective screener for migraine. Several prior studies found that nausea and photophobia were among the most sensitive and specific symptoms for distinguishing migraine from other headache types,16–19,22,23,25,26 and these variables emerged as distinctive within the present study. Five of these studies also found phonophobia to be one of the most sensitive and specific symptoms for identifying migraine,17–19,22,25,26 but it was included as part of the optimal model in only one study.22 A notable difference between this study and others is the lack of predictive importance of disability within the current study. Studies that found disability to be an important predictor of migraine18,19,23 varied considerably in how disability was operationalized, and therefore, the utility of this variable cannot be directly compared. For example, Lipton et al (2003) defined disability as functional limitation reported on any one day in the past 3 months, while Wang et al (2008) operationalized disability as headache that limited the patient from studying, working, housekeeping, family, social life, or leisure activities. In addition, these
92 studies were conducted with treatment-seeking samples that are overall more likely to endorse a greater level of disability, while the current study was conducted using a comparison group of individuals having numerous forms of primary and secondary headaches and thus varying levels of disability. Another difference in findings between the current study and previous literature is the importance of duration in establishing migraine.17,19,26 This is likely because the current study did not include PMs within the migraine group as did other studies,16,18,25 and because other studies of this type have often not included duration as a candidate symptom, despite its requirement per ICHD-3. Individuals with PM often endorse migraine attacks of shorter than 4-hour duration, particularly young adults who have otherwise prototypical migraine attacks.35,36 This was the case in the present study, wherein 90.6% of those with PM met all diagnostic criteria for migraine except duration of 4-72 hours. The screener performed well among probable migraineurs, however, at a different OOP that did not include duration (positively endorsing at least two out of four items), showing a sensitivity of 98% and specificity of 75%. Whether duration is of similar importance in older and treatmentseeking samples is an empirical question. The OOP of the current four-question screener was a positive endorsement on three or more items, showing 94% sensitivity and 92% specificity. These sensitivity and specificity values exceed those obtained in validation studies of the ID MigraineTM 23 (81% sensitivity, 75% specificity), the three-item screener developed by Cady et al. (2004) (77% sensitivity), the screener recommended by PrysePhillips et al. (2002) (86% sensitivity and 73% specificity), and even a longer measure, the MS-Q22 (92% sensitivity and 81% specificity). Although this screener showed high sensitivity, specificity, and accuracy at identifying migraine, as with any screening instrument the Migraine-4 should not be used in isolation; a positive screen should be followed by diagnostic confirmation of migraine. Some strengths of the current study include the use of recent ICHD-3 beta criteria, the large sample size of migraineurs and individuals with nonmigrainous headache, the strong statistical approach,
January 2016 and the inclusion of young adult headache sufferers that are less likely to be complicated by chronification or medication overuse. However, some limitations exist. One main limitation is that the authors did not readminister independently to the holdout sample the four items that comprised the screening measure. Although there is not yet any evidence to suggest that symptom endorsement would differ if the screening measure were administered in isolation (vs with other items), this possibility exists and awaits empirical confirmation. In addition, headache sufferers in the current study were diagnosed using a wellvalidated computer-administered headache interview, which does not allow for clarification of responses, as does an in-person interview. Last, the observed prevalence of migraine was less than that typically reported in population studies, which likely reflects the large number of young adults with otherwise prototypical migraine but who fell short of the required 4-72 hour attack duration and thus were instead classified as “probable” migraineurs. In light of these strengths and limitations, future research should attempt to replicate these results via in-person diagnostic headache interviews and within a treatment-seeking sample of older adults. If the Migraine-4 performs similarly well, direct comparisons with other validated screeners are warranted. Given its strong performance among a nonclinical sample, the screener may be particularly useful in research efforts to estimate the prevalence of migraine in a given population or to prescreen migraineurs for subsequent experimental studies. Because physicians often do not have time to assess all diagnostic criteria or conduct a lengthy structured interview,16 inclusion of this four-item measure in clinical settings could be used to identify migraineurs that so often go undiagnosed,8,9 so long as it performs well in future studies with clinical samples. Collectively, this study and future validation efforts will clarify the potential utility of this promising screener.
STATEMENT OF AUTHORSHIP Category 1 (a) Conception and Design Brooke Walters and Todd Smitherman
Headache (b) Acquisition of Data Todd Smitherman (c) Analysis and Interpretation of Data Brooke Walters and Todd Smitherman
93
8.
Category 2 (a) Drafting the Manuscript Brooke Walters and Todd Smitherman (b) Revising It for Intellectual Content Brooke Walters and Todd Smitherman Category 3
9.
10.
(a) Final Approval of the Completed Manuscript Brooke Walters and Todd Smitherman
REFERENCES 1. Headache Classification Committee of the International Headache Society. The International Classification of headache disorders. Cephalalgia. 2013; 33:629-808. 2. Burch RC, Loder S, Loder E, Smitherman TA. The prevalence and burden of migraine and severe headache in the United States: Updated statistics from government health surveillance studies. Headache. 2015;55:21-34. doi: 10.1111/head.12482 3. Lipton RB, Bigal ME, Diamond M, Freitag F, Reed ML, Stewart WF. Migraine prevalence, disease burden, and the need for preventive therapy. Neurology. 2007;68:343-349. 4. Smitherman TA, Burch R, Sheikh H, Loder E. The prevalence, impact, and treatment of migraine and severe headaches in the United States: A review of statistics from national surveillance studies. Headache. 2013;53:427-436. doi: 10.1111/head.12074 5. Vos T, Flaxman AD, Naghavi M, et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990–2010: A systematic analysis for the global burden of disease study 2010. The Lancet. 2013;380:2163-2196. doi: 10.1016/ S0140-6736(12)61729-2 6. Stewart WF, Wood C, Reed ML, Roy J, Lipton RB. Cumulative lifetime migraine incidence in women and men. Cephalalgia. 2008;28:1170-1178. 7. Carson AJ, Ringbauer B, MacKenzie L, Warlow C, Sharpe M. Neurological disease, emotional disorder, and disability, they are related: A study of 300 consecutive new referrals to a neurology out-
11.
12.
13.
14.
15.
16.
17.
18.
19.
patient department. J Neurol Neurosurg Psychiatry. 2002;68:202-206. Diamond S, Bigal ME, Silberstein S, Loder E, Reed M, Lipton RB. Patterns of diagnosis and acute and preventive treatment for migraine in the United States: Results from the American Migraine Prevalence and Prevention study. Headache. 2007;47:355-363. Lipton RB, Diamond S, Reed M, Diamond ML, Stewart WF. Migraine diagnosis and treatment: Results from the American Migraine Study II. Headache. 2001;41:638-645. Lipton RB, Serrano D, Holland S, Fanning KM, Reed ML, Buse DC. Barriers to the diagnosis and treatment of migraine: Effects of sex, income, and headache features. Headache. 2013;53:81-92. doi: 10.1111/j.1526-4610.2012.02265.x Lipton RB, Stewart WF, Simon D. Medical consultation for migraine: Results from the American Migraine Study. Headache. 1998;38:87-96. Schreiber CP, Hutchinson S, Webster CJ, Ames M, Richardson MS, Powers C. Prevalence of migraine in patients with a history of self-reported or physician-diagnosed “sinus” headache. Arch Intern Med. 2004;164:1769-1772. Mafi JN, Edwards ST, Pedersen NP, Davis RB, McCarthy EP, Landon BE. Trends in the ambulatory management of headache: Analysis of NAMCS and NHAMCS data 1999-2010. J Gen Intern Med. 2015;30:548-555. doi: 10.1007/s11606-014-3107-3 Sempere AP, Porta-Etessam J, Medrano V, et al. Neuroimaging in the evaluation of patients with non-acute headache. Cephalalgia. 2005;25:30-35. Dowson AJ, Lipscombe S, Sender J, Rees T, Watson D. New guidelines in the management of migraine in primary care. Curr Med Res Opin. 2002;18:414-439. Martin VT, Penzein DB, Houle TT, Andrew ME, Lofland KR. The predictive value of abbreviated migraine diagnostic criteria. Headache. 2005;45:1102-1112. Michel P, Henry P, Letenneur L, Jogeix M, Corson A, Dartigues JF. Diagnostic screen for assessment of the IHS criteria for migraine by general practitioners. Cephalalgia. 1993;13:54-59. Wang SJ, Fuh JL, Huang SY, et al. Diagnosis and development of screening items for migraine in neurological practice in Taiwan. J Formos Med Assoc. 2008;107:485-494. Cady RK, Borchert LD, Spalding W, Hart CC, Sheftell FD. Simple and efficient recognition of
94
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
January 2016 migraine with 3-question headache screen. Headache. 2004;44:323-327. Gervil M, Ulrich V, Olesen J, Russell MB. Screening for migraine in the general population: Validation of a simple questionnaire. Cephalalgia. 1998; 18:342-348. Maizels M, Burchette R. Rapid and sensitive paradigm for screening patients with headache in primary care settings. Headache. 2003;43:441-450. Lainez MJA, Dominguez M, Rejas J, et al. Development and validation of the Migraine Screen Questionnaire (MS-Q). Headache. 2005;45:1328-1338. Lipton RB, Dodick D, Sadovsky R, et al. A selfadministered screener for migraine in primary care: The ID migraine validation study. Neurology. 2003;61:375-382. Pryse-Phillips W, Aube M, Gawel M, Nelson R, Purdy A, Wilson K. A headache diagnosis project. Headache. 2002;42:728-737. Smetana GW. The diagnostic value of historical features in primary headache syndromes. Arch Intern Med. 2000;160:2729-2737. Detsky ME, McDonald DR, Baerlocher MO, Tomlinson GA, McCrory DC, Booth CM. Does this patient with headache have migraine or need neuroimaging? JAMA. 2006;296:1274-1283. Bigal ME, Bigal JM, Betti M, Bordini CA, Speciali JG. Evaluation of the impact of migraine and episodic tension-type headache on the quality of life and performance of university student population. Headache. 2001;41:710-719. Smitherman TA, McDermott MJ, Buchanan EM. Negative impact of migraine on a university population: Quality of life, functional impairment, and comorbid psychiatric symptoms. Headache. 2011;51:581-589. Walters AB, Hamer JD, Smitherman TA. Sleep disturbance and affective comorbidity among episodic migraineurs. Headache. 2014;54:116-124. doi: 10.1111/head.12168 Peck KR, Smitherman TA. Mediator variables in headache research: Methodological critique and exemplar using self-efficacy as a mediator of the relationship between headache severity and disability. Headache. 2015;55:1102-1111. Turner DP, Smitherman TA, Black AK, et al. Are migraine and tension headache diagnostic types or points on a severity continuum? An exploration of the latent taxometric structure of headache. Pain. 2015;156:1200-1207.
32. Smitherman TA, Davis RE, Walters AB, Young J, Houle TT. Anxiety sensitivity and headache: Diagnostic differences, impact, and relations with headache triggers. Cephalalgia. 2015;35:710-721. doi: 10.1177/0333102414557840 33. Black AK, Fulwiler JC, Smitherman TA. The role of fear of pain in headache. Headache. 2015;55: 669-679. doi: 10.1111/head.12561 34. Smitherman TA, Kolivas ED. Trauma exposure versus posttraumatic stress disorder: Relative associations with migraine. Headache. 2013;53:775-786. doi: 10.1111/head.12063 35. Rains JC, Penzien DB, Lipchik GL, Ramadan NM. Diagnosis of migraine: Empirical analysis of a large clinical sample of atypical migraine (IHS 1.7) patients and proposed revision of the IHS criteria. Cephalalgia. 2001;21:584-595. 36. Rasmussen B, Jensen R, Olesen J. A populationbased analysis of the diagnostic criteria of the International Headache Society. Cephalalgia. 1991; 11:129-134. 37. Andrew ME, Penzien DB, Rains JC, Knowlton GE, McAnulty RD. Development of a computer application for headache diagnosis: The headache diagnostic system. Int J Biomed Comput. 1992;31:17-24. 38. Halpern EJ, Albert M, Krieger AM, Metz CE, Maidment AD. Comparison of receiver operating characteristic curves on the basis of optimal operating points. Acad Radiol. 1996;3:245-253.
APPENDIX : FOUR-ITEM SCREENING MEASURE Migraine-4 1. If left untreated or unsuccessfully treated, about long would these headaches usually last? a. Less than 30 minutes b. At least 30 minutes but less than 2 hours c. At least 2 hours but less than 4 hours d. Between 4 hours and 3 days e. Between 3 days and 7 days f. Longer than 7 straight days 2. Do you often feel nauseous or sick to your stomach during your headaches? a. Yes b. No 3. Are you often sensitive to light during your headaches? a. Yes b. No 4. Are you often sensitive to sound during your headaches? a. Yes b. No *These four items followed a preliminary question of “Do you ever get headaches?”; An answer of “d” on question 1 and “Yes” to questions 2-4 are affirmative responses for migraine.
Headache C 2015 American Headache Society V
Research Submission Development and Validation of a Four-Item Migraine Screening Algorithm Among a Nonclinical Sample: The Migraine-4 A. Brooke Walters, PhD; Todd A. Smitherman, PhD, FAHS
Objectives.—The present study sought to develop and validate a screening algorithm from the ICHD-3 beta diagnostic criteria for migraine utilizing a nonclinical sample. The goal was to determine the most sensitive and specific symptoms for differentiating migraine from other headache disorders and to validate the derived symptom algorithm as a screening measure. Background.—Despite its prevalence and impact, migraine remains under-recognized and under-treated. The U.S. Headache Consortium recommended development and dissemination of validated screening measures as a means to improve diagnosis. Methods.—Participants were 1829 young adults (71.5% female; 74.4% white; mean age 5 19.09 years [SD 5 2.05]) who reported headache via computerized diagnostic interview. From this group, 158 were found to have ICHD episodic or chronic migraine and were randomly split into experimental and holdout validation samples. Within the experimental sample, receiver operating characteristic (ROC) curve data were obtained for each candidate symptom (item); backward stepwise logistic regression analysis was performed among the items with the most predictive likelihood ratios to determine the optimal model for differentiating migraine from non-migrainous headache. The retained four-symptom algorithm was then validated among the holdout sample, in which various cutoff points were compared to gold standard diagnosis via ROC curve estimations to determine the optimal operating point (OOP) of the algorithm as a screening measure. Results.—Attack duration of 4-72 hours (100% [95% CI 5 95-100%]), severity 5 (91% [82-97%]), photophobia (90% [80-96%]), and phonophobia (90% [80-96%]) showed the highest sensitivity, while vomiting (98% [96-99%]), duration of 4-72 hours (92% [90-94%]), nausea (89% [86-91%]), and headache-related disability (88% [85-91%]) showed the highest specificity. The optimal retained model (Migraine-4) included: duration of 4-72 hours, nausea, photophobia, and phonophobia. Among the holdout validation sample, the OOP was positive endorsement of three out of four symptoms, which had a sensitivity of 94% (95% CI 5 87-98%), a specificity of 92% (90-94%), and an area under the curve of 93% (90-96%; 1LR 5 12.37, 2LR 5 0.06, PPV 5 67%, NPV 5 99%). Conclusions.—The optimal model shares some similarities with previous models but performed better than prior screeners at differentiating migraine from other headache presentations. The Migraine-4 has utility in identifying migraine among nonclinical and young adult samples. Further research with this measure is warranted to determine its utility with treatment-seeking patients and validity in direct comparison to established screening instruments. Key words: migraine, headache diagnosis, screening, assessment, sensitivity and specificity (Headache 2016;56:86-94)
Address all correspondence to T.A. Smitherman, Department of Psychology, University of Mississippi, 207 Peabody, Oxford, MS 38677, USA, email: [email protected]
INTRODUCTION The diagnostic criteria for migraine are outlined in the 2013 International Classification of Headache Disorders, Third Edition (beta version; ICHD-3 beta).1 Migraine is defined as a recurrent headache
Accepted for publication September 10, 2015.
Conflict of Interest: None.
From the Department of Psychology, University of Mississippi, Oxford, MS 38677, USA.
86
Headache disorder characterized by a history of at least five headache attacks of 4-72 hours duration. The headache attacks must have at least two of the following four features: (1) unilateral location; (2) pulsating (throbbing) quality; (3) moderate or severe pain; and (4) causing aggravation or avoidance or routine activity AND either nausea, vomiting, or both photophobia and phonophobia. Migraine can occur with or without aura symptoms, which are temporary sensory distortions, typically visual in nature, that precede headache onset by an hour or less and then resolve. Migraine is further dichotomized based on its frequency of occurrence: attacks with a frequency of less than 15 days per month are classified as episodic migraine (EM), and the much rarer occurrence of attacks on 15 or more days per month is termed “chronic” migraine (CM). Migraine affects 12-17% of the population annu2–5 ally, and the lifetime incidence is 43% for women and 18% for men.6 Data from the World Health Organization indicate that migraine is the third most common medical condition and the eighth leading cause of disability worldwide.5 Because of its high prevalence and negative impact, migraine is a leading cause of neurologist visits7 and the fourth most common reason for emergency department visits.3 Despite its prevalence and impact, migraine remains under-recognized, underdiagnosed, and undertreated. Roughly half of those with migraine never receive a diagnosis from a physician,8,9 and only 26% receive minimally appropriate care.10 Significant barriers to appropriate care include low rates of physician consultation among males and those with limited insurance/economic resources,10,11 misdiagnosis,12 overuse of neuroimaging,13,14 and inadequate or improper treatment.10 One means of improving the diagnosis and treatment of migraine is developing and validating cost-efficient screening measures, as recommended by the U.S. Headache Consortium.15 Prior efforts in this domain have included statistical algorithms of those symptoms most predictive of migraine in a given sample,16–18 as well as the validation of screening questionnaires.19–21 Few studies have attempted to both determine an algorithm for migraine diagnosis and subsequently validate it as a screening measure,22–24 and these efforts were prior to publication
87 of the ICHD-3 beta. Previous studies also have been quite variable in terms of reference groups, candidate items/symptoms, and findings, with only nausea consistently discriminating migraine from other headache types.16–19,22,23,25,26 In light of these limitations and recognition of a need for cost-efficient migraine screeners among individuals who do not regularly seek treatment (eg, underrepresented groups, in population studies, and for research purposes), the current study endeavored to develop a migraine screener comporting with ICHD-3 beta for use with nonclinical samples. Young adults represented the sample of interest because of the need to identify migraine early in the lifespan to foster early recognition and treatment in an effort to prevent headache progression or chronification, and in light of data that migraine prevalence increases precipitously during young adulthood3 and confers substantial disability.27–29 The present study thus was designed to (1) determine statistically the most sensitive and specific ICHD-3 beta symptoms differentiating migraine from other common headache types among a nonclinical population (ie, algorithm development) and then to (2) validate the resulting symptom algorithm among a holdout sample from the larger overall sample (ie, screener validation) and identify an optimal cutoff point for use. In light of previous findings, nausea was hypothesized to be the strongest discriminator of migraine, but no other a priori hypotheses were made about other candidate symptoms due to inconsistency in previous findings.
METHODS Participants and Procedure.—This study received ethical approval from the Institutional Review Board at the University of Mississippi, and all participants provided informed consent via electronic signature prior to providing data. Participants were young adults at a southeastern university who participated voluntarily in exchange for modest course credit. As part of an ongoing data collection effort,30–34 from August 2012 to December 2013 they completed an online battery of measures, including a computer-administered structured diagnostic interview for headache. Those meeting ICHD-3 beta criteria for EM (with or without
88 aura) or chronic migraine comprised the migraine group. Those with episodic, chronic, or probable tension-type headache (TTH), cluster headache, posttraumatic headache, medication overuse headache (MOH), or those who complained of headache but could not be classified comprised the nonmigraine group. Thus, participants with any headache presentation other than migraine or probable migraine (PM; ie, meeting all but one criterion for migraine) comprised the non-migraine comparison group. For secondary analyses, probable migraineurs were included as part of the migraine group because younger adults often have otherwise prototypical migraine attacks of duration less than 4 hours (ie, missing only Criterion B).35,36 Measures.—Structured Diagnostic Interview for Headache-Revised.—The well-validated computeradministered Structured Diagnostic Interview for Headache37 was modified to comport with ICHD-3 beta criteria and be administered online. Modifications were slight and included minor rewording of some items for clarity, addition of forced-choice format for querying headache duration, and removal of appendix items that afforded detailed assessment of cluster and secondary headache disorders. The Structured Diagnostic Interview for Headache-Revised (SDIH-R) includes items pertaining to each ICHD diagnostic symptom of migraine (including aura) and TTH, as well as additional items pertaining to headache frequency, disability, and screening for other primary and secondary headache disorders. Precise diagnostic criteria for cluster headache, posttraumatic headache, and MOH were not assessed in the computerized SDIH-R; rather, they were screened for by inquiring about a past diagnosis of cluster headache (cluster), if headache began after head injury (posttraumatic), or prolonged use of medications to treat headache on 3 days/week coinciding with increased headache frequency (MOH). Statistical Analyses.—The current study occurred in two phases: the dataset was randomly split by approximately 50% using SPSS random sampling (without variable matching), with SDIH-R responses from the first half of participants (ie, experimental sample) used for algorithm development and responses from the second half (ie, vali-
January 2016 dation sample) used for validating the resulting symptom algorithm as a screening instrument. Algorithm Development in the Experimental Sample.—Chi-square analyses and independent samples ttests were used to assess whether the experimental and validation samples differed on demographic or headache variables. For algorithm development, twoby-two tables were constructed using each of the ICHD-3 beta migraine symptoms (present vs absent) and headache diagnoses (migraine vs non-migraine); sensitivity, specificity, predictive values, and likelihood ratios were calculated for each symptom. Receiver operator characteristic (ROC) curves were then plotted for each of the one-variable symptom models to determine which were most predictive of a migraine diagnosis, as indexed by the area under the curve (AUC). Next, using the optimal model guidelines in Martin et al (2005),16 symptoms were eliminated if they did not have 1LRs > 4.5 or 2LRs < 0.25. A backward stepwise logistic regression analysis was then performed using the remaining symptoms to determine the optimal model for differentiating migraine from non-migraine headache. This resulting model was used as the screening measure (Appendix) validated in the second phase of the study. Screener Validation in the Validation Sample.—For screener validation, two-by-two tables were constructed using each possible cutoff point of the screening measure and headache diagnosis, from which sensitivity, specificity, LRs, and PVs were calculated among the holdout validation sample. An ROC curve then was plotted to compare scores on the screening measure with the gold standard (ie, EM or CM diagnosis as per full ICHD-3 beta criteria) and the screener’s utility in differentiating migraine from other headache types. ROC curve analysis was used to assess all possible cutoff points of the new screener and to determine the optimal operating point (OOP)38 of the screener. Finally, the performance of the screener was examined as a function of gender and race, as well as among probable migraineurs, to supplement the main analyses.
RESULTS Sample Comparison.—The total sample consisted of 1966 participants who responded positively to
Headache
89 Table 1.——Demographics and Headache Diagnoses in Experimental and Validation Samples Experimental Sample (n 5 887)
Age M (SD) Gender Female Male Race White Black Other
19.05 (1.96)
Validation Sample (n 5 942)
P5
19.12 (2.13)
628 (70.8%) 259 (29.2%)
680 (72.2%) 262 (27.8%)
661 (74.5%) 147 (16.6%) 79 (8.9%)
699 (74.2%) 184 (19.5%) 59 (6.3%)
.48 .51
Entire Sample (N 5 1829)
19.09 (2.05) 1308 (71.5%) 521 (28.5%)
.88 1360 (74.4%) 331 (18.1%) 138 (7.5%)
Headache Type
CM EM w/out aura EM w/aura All migraine †PM CTTH ETTH PTTH Cluster Posttraumatic MOH No diagnosis All non-migraine
23 29 17 69 266 12 184 219 18 40 1 78 552
(2.6%) (3.3%) (1.9%) (7.8%) (30.0%) (1.4%) (20.7%) (24.7%) (2.0%) (4.5%) (0.1%) (8.8%) (62.2%)
18 45 26 89 301 16 177 203 18 52 1 85 552
(1.9%) (4.8%) (2.8%) (9.4%) (32.0%) (1.7%) (18.8%) (21.5%) (1.9%) (5.5%) (0.1%) (9.0%) (58.6%)
.32 .10 .23 .14 .36 .55 .29 .11 .85 .32 .97 .86 .14
41 74 43 158 567 28 361 422 36 92 2 163 1,104
(2.2%) (4.0%) (2.4%) (8.6%) (31.0%) (1.5%) (19.7%) (23.1%) (2.0%) (5.0%) (0.1%) (8.9%) (60.4%)
Mean Headache Days Per Month (0-30)
Mean Headache Severity (0-10)
18.54 (SD 5 4.28) 6.86 (4.37) 6.77 (3.27)
6.41 (SD 5 1.22) 6.03 (1.79) 6.33 (1.34)
7.33 17.96 4.88 5.09 9.22 7.87 21 2.57
5.19 4.82 3.95 3.67 5.39 5.11 7 2.82
(5.72) (4.77) (3.17) (4.48) (5.08) (6.03) (5.66) (2.45)
(1.63) (1.28) (1.36) (1.48) (1.89) (1.85) (2.83) (1.55)
†Probable migraineurs were excluded from the primary analyses, but were included as part of the migraine group in secondary analyses. CM 5 chronic migraine; CTTH 5 chronic tension-type headache; EM 5 episodic migraine; ETTH 5 episodic tension-type headache; MOH 5 medication overuse headache; PM 5 probable migraine; PTTH 5 probable tension-type headache.
the initial question “Do you ever get headaches?” Excluding 137 with missing data, 1829 were retained and classified into 10 headache categories (see Table 1) based on their SDIH-R responses. The sample of 1829 was split randomly by approximately 50% to ensure equivalence between the experimental (n 5 887) and validation (n 5 942) samples. Nonsignificant chi-square tests (for gender, race [white vs non-white], and headache diagnosis) and independent samples t-tests (for age, headache frequency, and headache severity) confirmed that the two samples did not differ significantly on any of these variables (see Table 1). Algorithm Development.—Within the experimental sample, the performance of each migraine symptom was compared to the gold standard diagnosis of migraine as acquired via the SDIH-R. Two-by-two tables were created showing the number of migrai-
neurs and non-migraineurs endorsing each symptom (N 5 621), from which sensitivity, specificity, positive and negative likelihood ratios, and positive and negative predictive values were calculated (see Table 2). Headache duration of 4-72 hours showed the highest sensitivity (100%), followed by severity 5 (91%), photophobia (90%), and phonophobia (90%). In contrast, vomiting showed the highest specificity (98%), followed by duration of 4-72 hours (92%), nausea (89%), and disability (defined as headaches interfering with work, school, or personal life; 88%). Duration of 4-72 hours had the largest AUC (96%), followed by photophobia (78%), severity 5 (78%), and nausea (77%). Unilateral location, pulsing quality, and worsening by physical activity were ruled out as part of the optimal model as they did not have an 1LR > 4.5 or 2LR < 0.25. By way of backward stepwise logistic regression analysis (performed on the
(99-100%) (87-93%) (92-97%) (96-99%) (94-98%) (93-97%) (88-93%) (96-99%) (95-99%) (93-97%) 100% 91% 95% 98% 96% 95% 91% 98% 98% 95% (52-71%) (10-19%) (13-21%) (19-30%) (18-29%) (33-52%) (33-74%) (20-32%) (16-25%) (31-50%) Disease prevalence 5 11% (excluding probable migraineurs).
Duration (4-72 hours) Unilateral Pulsing Severity 5 Worsened by activity Nausea Vomiting Photophobia Phonophobia Disability
100% 49% 80% 91% 78% 65% 19% 90% 90% 64%
(95-100%) (37-62%) (68-88%) (82-97%) (66-87%) (53-76%) (11-30%) (80-96%) (80-96%) (51-75%)
92% 62% 49% 65% 67% 89% 98% 67% 55% 88%
(90-94%) (57-66%) (45-54%) (61-69%) (63-71%) (86-91%) (96-99%) (63-71%) (51-60%) (85-91%)
13.14 1.28 1.57 2.60 2.39 5.90 9.42 2.71 2.02 5.37
(9.83-17.58) (.99-1.67) (1.36-1.82) (2.27-2.97) (2.01-2.85) (4.40-7.91) (4.39-20.20) (2.35-3.13) (1.78-2.28) (4.02-7.17)
0.82 0.41 0.13 0.33 0.39 0.83 0.15 0.18 0.41
– (0.65-1.05) (0.26-0.66) (0.06-0.29) (0.21-0.51) (0.28-0.54) (0.74-0.93) (0.08-0.31) (0.09-0.37) (0.30-0.56)
62% 14% 16% 25% 23% 42% 54% 26% 20% 40%
PPV 2LR 1LR Specificity Sensitivity
Table 2.——Experimental Sample: Migraine Symptom Performance (95% Confidence Intervals)
NPV
96% 56% 65% 78% 72% 77% 58% 78% 72% 76%
(95-98%) (49-63%) (58-71%) (73-83%) (66-79%) (70-84%) (50-66%) (73-83%) (67-78%) (69-83%)
January 2016
AUC
90
remaining seven symptoms), the retained optimal model consisted of duration of 4-72 hours (B 5 21.08), nausea (B 5 1.18), photophobia (B 5 1.87), and phonophobia (B 5 1.54). When including those with PM in the migraine group (N 5 887), photophobia showed the highest sensitivity (90%), followed by phonophobia (89%), pulsing quality (77%), and severity 5 (73%). Photophobia showed the highest AUC (78%), followed by phonophobia (72%), nausea (70%), and severity 5 (69%). Optimal likelihood ratios when including those with PM were found for nausea (B 5 1.77) and photophobia (B 5 2.73). Because this was a supplementary aim of the present study, validation was not attempted on this two-symptom model. Instead, the validity of the original foursymptom screener was explored within probable migraineurs in subsequent validation analyses. Screener Validation.—To validate the prior fourfactor model as a screening instrument in the holdout validation sample, two-by-two tables were created for each possible cutoff point of the screener, from which sensitivity, specificity, positive and negative likelihood ratios, and positive and negative predictive values were calculated (see Table 3). The OOP for the screener was positively endorsing three out of four items, which had a sensitivity of 94% and a specificity of 92% (1LR 5 12.37, 2LR 5 0.06, PPV 67%, NPV 99%). ROC curve analyses were used to determine the accuracy of each cutoff point as compared to the gold standard diagnosis of migraine (see Table 3). The AUC of positively endorsing one or more items was 69% [95% CI: 65-74%], of two or more items was 87% [85-90%], of three or more items was 93% [90-96%], and of all four items was 78% [72-85%]. When including probable migraineurs, the four-item screener showed an OOP of responding positively to two or more items (sensitivity 5 98% [96-99%], specificity 5 75% [71-78%], PPV 5 74% [70-77%], NPV 5 99% [97-99%], AUC 5 87% [84-89%]). All items in the screening algorithm were significantly intercorrelated (rs from 0.18 to 0.51, Ps .01) and were highly correlated with the OOP of the screener (rs from 0.42 to 0.71, Ps .01). Group Differences in Screener Performance.— The screening measure performed better statistically
91
69% 87% 93% 78% 100% (98-100%) 100% (99-100%) 99% (98-100%) 94% (91-95%)
(65-74%) (85-90%) (90-96%) (72-85%)
among men than in women at all four cutoff points (Ps .001; X2 at OOP 5 26.896, P < .001); however, this statistically significant difference is almost certainly a function of the large sample size and not clinically meaningful, as the AUC at the OOP of the screener was very high among both genders: 94% among men vs 93% among women. The screener performed equally well among white and non-white individuals.
Disease prevalence 5 13% (excluding probable migraineurs).
One or more Two or more Three or more All four
100% 100% 94% 57%
(96-100%) (96-100%) (87-98%) (46-68%)
39% (35-43%) 75% (71-78%) 92% (90-94%) 99% (98-100%)
1.64 3.96 12.37 61.84
(1.53-1.75) (3.42-4.57) (9.17-16.67) (25.37-150.73)
– – 0.06 (0.03-0.15) 0.43 (0.34-0.55)
21% 39% 67% 91%
(17-25%) (33-46%) (58-75%) (80-97%)
AUC NPV PPV –LR 1LR Specificity Sensitivity
Table 3.——Validation Sample: Screener Performance at Different Cutoff Points (95% Confidence Intervals)
Headache
DISCUSSION The current study sought to identify an algorithm for identifying ICHD-3 beta migraine and to validate this algorithm as a screening instrument in a nontreatment-seeking sample of young adults. The four characteristics of duration of 4-72 hours, nausea, photophobia, and phonophobia comprised the optimal model for differentiating migraine from non-migraine headache and showed high sensitivity, specificity, and accuracy at the OOP of three out of four positive endorsements. The screening measure had very strong clinical utility among men and women and among white and non-white individuals at its OOP. This is the first study to find this unique combination of these four symptoms as an effective screener for migraine. Several prior studies found that nausea and photophobia were among the most sensitive and specific symptoms for distinguishing migraine from other headache types,16–19,22,23,25,26 and these variables emerged as distinctive within the present study. Five of these studies also found phonophobia to be one of the most sensitive and specific symptoms for identifying migraine,17–19,22,25,26 but it was included as part of the optimal model in only one study.22 A notable difference between this study and others is the lack of predictive importance of disability within the current study. Studies that found disability to be an important predictor of migraine18,19,23 varied considerably in how disability was operationalized, and therefore, the utility of this variable cannot be directly compared. For example, Lipton et al (2003) defined disability as functional limitation reported on any one day in the past 3 months, while Wang et al (2008) operationalized disability as headache that limited the patient from studying, working, housekeeping, family, social life, or leisure activities. In addition, these
92 studies were conducted with treatment-seeking samples that are overall more likely to endorse a greater level of disability, while the current study was conducted using a comparison group of individuals having numerous forms of primary and secondary headaches and thus varying levels of disability. Another difference in findings between the current study and previous literature is the importance of duration in establishing migraine.17,19,26 This is likely because the current study did not include PMs within the migraine group as did other studies,16,18,25 and because other studies of this type have often not included duration as a candidate symptom, despite its requirement per ICHD-3. Individuals with PM often endorse migraine attacks of shorter than 4-hour duration, particularly young adults who have otherwise prototypical migraine attacks.35,36 This was the case in the present study, wherein 90.6% of those with PM met all diagnostic criteria for migraine except duration of 4-72 hours. The screener performed well among probable migraineurs, however, at a different OOP that did not include duration (positively endorsing at least two out of four items), showing a sensitivity of 98% and specificity of 75%. Whether duration is of similar importance in older and treatmentseeking samples is an empirical question. The OOP of the current four-question screener was a positive endorsement on three or more items, showing 94% sensitivity and 92% specificity. These sensitivity and specificity values exceed those obtained in validation studies of the ID MigraineTM 23 (81% sensitivity, 75% specificity), the three-item screener developed by Cady et al. (2004) (77% sensitivity), the screener recommended by PrysePhillips et al. (2002) (86% sensitivity and 73% specificity), and even a longer measure, the MS-Q22 (92% sensitivity and 81% specificity). Although this screener showed high sensitivity, specificity, and accuracy at identifying migraine, as with any screening instrument the Migraine-4 should not be used in isolation; a positive screen should be followed by diagnostic confirmation of migraine. Some strengths of the current study include the use of recent ICHD-3 beta criteria, the large sample size of migraineurs and individuals with nonmigrainous headache, the strong statistical approach,
January 2016 and the inclusion of young adult headache sufferers that are less likely to be complicated by chronification or medication overuse. However, some limitations exist. One main limitation is that the authors did not readminister independently to the holdout sample the four items that comprised the screening measure. Although there is not yet any evidence to suggest that symptom endorsement would differ if the screening measure were administered in isolation (vs with other items), this possibility exists and awaits empirical confirmation. In addition, headache sufferers in the current study were diagnosed using a wellvalidated computer-administered headache interview, which does not allow for clarification of responses, as does an in-person interview. Last, the observed prevalence of migraine was less than that typically reported in population studies, which likely reflects the large number of young adults with otherwise prototypical migraine but who fell short of the required 4-72 hour attack duration and thus were instead classified as “probable” migraineurs. In light of these strengths and limitations, future research should attempt to replicate these results via in-person diagnostic headache interviews and within a treatment-seeking sample of older adults. If the Migraine-4 performs similarly well, direct comparisons with other validated screeners are warranted. Given its strong performance among a nonclinical sample, the screener may be particularly useful in research efforts to estimate the prevalence of migraine in a given population or to prescreen migraineurs for subsequent experimental studies. Because physicians often do not have time to assess all diagnostic criteria or conduct a lengthy structured interview,16 inclusion of this four-item measure in clinical settings could be used to identify migraineurs that so often go undiagnosed,8,9 so long as it performs well in future studies with clinical samples. Collectively, this study and future validation efforts will clarify the potential utility of this promising screener.
STATEMENT OF AUTHORSHIP Category 1 (a) Conception and Design Brooke Walters and Todd Smitherman
Headache (b) Acquisition of Data Todd Smitherman (c) Analysis and Interpretation of Data Brooke Walters and Todd Smitherman
93
8.
Category 2 (a) Drafting the Manuscript Brooke Walters and Todd Smitherman (b) Revising It for Intellectual Content Brooke Walters and Todd Smitherman Category 3
9.
10.
(a) Final Approval of the Completed Manuscript Brooke Walters and Todd Smitherman
REFERENCES 1. Headache Classification Committee of the International Headache Society. The International Classification of headache disorders. Cephalalgia. 2013; 33:629-808. 2. Burch RC, Loder S, Loder E, Smitherman TA. The prevalence and burden of migraine and severe headache in the United States: Updated statistics from government health surveillance studies. Headache. 2015;55:21-34. doi: 10.1111/head.12482 3. Lipton RB, Bigal ME, Diamond M, Freitag F, Reed ML, Stewart WF. Migraine prevalence, disease burden, and the need for preventive therapy. Neurology. 2007;68:343-349. 4. Smitherman TA, Burch R, Sheikh H, Loder E. The prevalence, impact, and treatment of migraine and severe headaches in the United States: A review of statistics from national surveillance studies. Headache. 2013;53:427-436. doi: 10.1111/head.12074 5. Vos T, Flaxman AD, Naghavi M, et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990–2010: A systematic analysis for the global burden of disease study 2010. The Lancet. 2013;380:2163-2196. doi: 10.1016/ S0140-6736(12)61729-2 6. Stewart WF, Wood C, Reed ML, Roy J, Lipton RB. Cumulative lifetime migraine incidence in women and men. Cephalalgia. 2008;28:1170-1178. 7. Carson AJ, Ringbauer B, MacKenzie L, Warlow C, Sharpe M. Neurological disease, emotional disorder, and disability, they are related: A study of 300 consecutive new referrals to a neurology out-
11.
12.
13.
14.
15.
16.
17.
18.
19.
patient department. J Neurol Neurosurg Psychiatry. 2002;68:202-206. Diamond S, Bigal ME, Silberstein S, Loder E, Reed M, Lipton RB. Patterns of diagnosis and acute and preventive treatment for migraine in the United States: Results from the American Migraine Prevalence and Prevention study. Headache. 2007;47:355-363. Lipton RB, Diamond S, Reed M, Diamond ML, Stewart WF. Migraine diagnosis and treatment: Results from the American Migraine Study II. Headache. 2001;41:638-645. Lipton RB, Serrano D, Holland S, Fanning KM, Reed ML, Buse DC. Barriers to the diagnosis and treatment of migraine: Effects of sex, income, and headache features. Headache. 2013;53:81-92. doi: 10.1111/j.1526-4610.2012.02265.x Lipton RB, Stewart WF, Simon D. Medical consultation for migraine: Results from the American Migraine Study. Headache. 1998;38:87-96. Schreiber CP, Hutchinson S, Webster CJ, Ames M, Richardson MS, Powers C. Prevalence of migraine in patients with a history of self-reported or physician-diagnosed “sinus” headache. Arch Intern Med. 2004;164:1769-1772. Mafi JN, Edwards ST, Pedersen NP, Davis RB, McCarthy EP, Landon BE. Trends in the ambulatory management of headache: Analysis of NAMCS and NHAMCS data 1999-2010. J Gen Intern Med. 2015;30:548-555. doi: 10.1007/s11606-014-3107-3 Sempere AP, Porta-Etessam J, Medrano V, et al. Neuroimaging in the evaluation of patients with non-acute headache. Cephalalgia. 2005;25:30-35. Dowson AJ, Lipscombe S, Sender J, Rees T, Watson D. New guidelines in the management of migraine in primary care. Curr Med Res Opin. 2002;18:414-439. Martin VT, Penzein DB, Houle TT, Andrew ME, Lofland KR. The predictive value of abbreviated migraine diagnostic criteria. Headache. 2005;45:1102-1112. Michel P, Henry P, Letenneur L, Jogeix M, Corson A, Dartigues JF. Diagnostic screen for assessment of the IHS criteria for migraine by general practitioners. Cephalalgia. 1993;13:54-59. Wang SJ, Fuh JL, Huang SY, et al. Diagnosis and development of screening items for migraine in neurological practice in Taiwan. J Formos Med Assoc. 2008;107:485-494. Cady RK, Borchert LD, Spalding W, Hart CC, Sheftell FD. Simple and efficient recognition of
94
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
January 2016 migraine with 3-question headache screen. Headache. 2004;44:323-327. Gervil M, Ulrich V, Olesen J, Russell MB. Screening for migraine in the general population: Validation of a simple questionnaire. Cephalalgia. 1998; 18:342-348. Maizels M, Burchette R. Rapid and sensitive paradigm for screening patients with headache in primary care settings. Headache. 2003;43:441-450. Lainez MJA, Dominguez M, Rejas J, et al. Development and validation of the Migraine Screen Questionnaire (MS-Q). Headache. 2005;45:1328-1338. Lipton RB, Dodick D, Sadovsky R, et al. A selfadministered screener for migraine in primary care: The ID migraine validation study. Neurology. 2003;61:375-382. Pryse-Phillips W, Aube M, Gawel M, Nelson R, Purdy A, Wilson K. A headache diagnosis project. Headache. 2002;42:728-737. Smetana GW. The diagnostic value of historical features in primary headache syndromes. Arch Intern Med. 2000;160:2729-2737. Detsky ME, McDonald DR, Baerlocher MO, Tomlinson GA, McCrory DC, Booth CM. Does this patient with headache have migraine or need neuroimaging? JAMA. 2006;296:1274-1283. Bigal ME, Bigal JM, Betti M, Bordini CA, Speciali JG. Evaluation of the impact of migraine and episodic tension-type headache on the quality of life and performance of university student population. Headache. 2001;41:710-719. Smitherman TA, McDermott MJ, Buchanan EM. Negative impact of migraine on a university population: Quality of life, functional impairment, and comorbid psychiatric symptoms. Headache. 2011;51:581-589. Walters AB, Hamer JD, Smitherman TA. Sleep disturbance and affective comorbidity among episodic migraineurs. Headache. 2014;54:116-124. doi: 10.1111/head.12168 Peck KR, Smitherman TA. Mediator variables in headache research: Methodological critique and exemplar using self-efficacy as a mediator of the relationship between headache severity and disability. Headache. 2015;55:1102-1111. Turner DP, Smitherman TA, Black AK, et al. Are migraine and tension headache diagnostic types or points on a severity continuum? An exploration of the latent taxometric structure of headache. Pain. 2015;156:1200-1207.
32. Smitherman TA, Davis RE, Walters AB, Young J, Houle TT. Anxiety sensitivity and headache: Diagnostic differences, impact, and relations with headache triggers. Cephalalgia. 2015;35:710-721. doi: 10.1177/0333102414557840 33. Black AK, Fulwiler JC, Smitherman TA. The role of fear of pain in headache. Headache. 2015;55: 669-679. doi: 10.1111/head.12561 34. Smitherman TA, Kolivas ED. Trauma exposure versus posttraumatic stress disorder: Relative associations with migraine. Headache. 2013;53:775-786. doi: 10.1111/head.12063 35. Rains JC, Penzien DB, Lipchik GL, Ramadan NM. Diagnosis of migraine: Empirical analysis of a large clinical sample of atypical migraine (IHS 1.7) patients and proposed revision of the IHS criteria. Cephalalgia. 2001;21:584-595. 36. Rasmussen B, Jensen R, Olesen J. A populationbased analysis of the diagnostic criteria of the International Headache Society. Cephalalgia. 1991; 11:129-134. 37. Andrew ME, Penzien DB, Rains JC, Knowlton GE, McAnulty RD. Development of a computer application for headache diagnosis: The headache diagnostic system. Int J Biomed Comput. 1992;31:17-24. 38. Halpern EJ, Albert M, Krieger AM, Metz CE, Maidment AD. Comparison of receiver operating characteristic curves on the basis of optimal operating points. Acad Radiol. 1996;3:245-253.
APPENDIX : FOUR-ITEM SCREENING MEASURE Migraine-4 1. If left untreated or unsuccessfully treated, about long would these headaches usually last? a. Less than 30 minutes b. At least 30 minutes but less than 2 hours c. At least 2 hours but less than 4 hours d. Between 4 hours and 3 days e. Between 3 days and 7 days f. Longer than 7 straight days 2. Do you often feel nauseous or sick to your stomach during your headaches? a. Yes b. No 3. Are you often sensitive to light during your headaches? a. Yes b. No 4. Are you often sensitive to sound during your headaches? a. Yes b. No *These four items followed a preliminary question of “Do you ever get headaches?”; An answer of “d” on question 1 and “Yes” to questions 2-4 are affirmative responses for migraine.
