Psychosocial headache measures in Caucasian and African American headache patients: Psychometric attributes and measurement invariance

Abstract

Introduction: As research increasingly examines differences in headache characteristics between Caucasian and African American headache patients, it is requisite to determine the equivalence of existing headache measures for both racial groups.

Methods: This study examined the psychometric properties and measurement invariance of four widely used psychosocial headache measures. Caucasians (N = 173) and African Americans (N = 114) receiving treatment in four headache subspecialty clinics throughout Ohio completed the Headache Disability Inventory (HDI), the Migraine-Specific Quality of Life (MSQL) measure, the Headache Management Self-Efficacy (HMSE) scale, and the Headache-Specific Locus of Control (HSLC) scale prior to initiating new preventive therapies.

Conclusion: All measures demonstrated good internal consistency and measurement invariance. Despite adequate measurement invariance, the recommended factor structure of the HMSE scale yielded a poor fit for both African American and Caucasian patients. With perhaps the exception of the HMSE scale, the HDI, MSQL scale and HSLC scale are psychometrically sound and can be used with both Caucasian and African American headache patients.

Keywords

Psychometric testing ethnic validity measurement invariance headache race migraine

Introduction

Approximately 3% of African Americans in the United States experience headaches on 180 or more days per year (1). A growing amount of research has compared African Americans and Caucasians with headache disorders. Although African Americans are less likely than Caucasians to be diagnosed with primary headache disorders, such as tension-type headache (TTH) and migraines, a higher prevalence rate of probable migraine (PM), a subtype of migraine that satisfies all but one criterion of a migraine diagnosis, has been found in African Americans compared to Caucasians (2). Among persons diagnosed with headache disorders, African Americans often report more frequent and severe headaches than Caucasians (3,4). Moreover, African Americans receiving care in headache subspecialty treatment clinics are twice as likely as Caucasians to terminate their clinic-based headache treatment before treatments are administered in their entirety (5).

Identifying and understanding race-related differences in headache patients is complicated by the fact that: (i) very few headache studies focus exclusively on ethnic minority groups; and (ii) the psychometric properties of multi-item headache measures are rarely evaluated by racial/ethnic group. Many psychosocial measures currently used in headache research, such as the Migraine-Specific Quality of Life (MSQL) scale (6), the Headache Disability Inventory (HDI) (7), and the Headache Management Self-Efficacy (HMSE) scale (8) were constructed 10 to 20 years ago using samples comprised almost exclusively of Caucasians. The psychometric properties and racial equivalence of commonly used headache measures have, for all intents and purposes, never been evaluated by patient race/ethnicity.

To date, efforts to develop equivalent headache measures have been limited to translating English-based measures for use in non–English speaking countries. This research has translated existing headache measures from English to French (9), English to Japanese (10), English to Danish (11) and English to German (12). However, while language translation is necessary when developing an equivalent measure for use with a different linguistic group, different cultures may interpret key constructs of a measure in conceptually different manners. Furthermore, little—if any—research has examined the validity and equivalence of headache measures between racial/ethnic groups within the same country (e.g. African Americans and Caucasians in the United States). This is problematic because terms that are used commonly in health-related research (e.g. self-efficacy, quality of life, locus of control) may have many different meanings between different racial/ethnic groups, even within the same country (13–15).

Responses to behavioral measures in health-related research often differ by race. For example, when completing self-report measures that assess behavior frequency or intensity, many African Americans engage in “telescoping” (i.e. believing that recent behaviors occurred more frequently than they actually did) (16); underreport health-injurious behaviors (e.g. high-risk sexual and substance use behaviors) (17); and overreport health-promoting behaviors (e.g. engaging in cancer screenings) (18). While other research finds that many behavioral measures apply equally well to African Americans and Caucasians (19), race-related biases and errors in self-reports that have been observed may threaten the psychometric integrity of many self-report behavioral measures and lead to erroneous conclusions regarding race differences (or the lack thereof) in important health behaviors (20).

The extent to which race-related reporting biases and errors that threaten the psychometric integrity of behavioral measures generalize to psychosocial measures (e.g. quality of life, locus of control) is unclear. Research has typically found relatively few differences between African Americans and Caucasians with headache disorders (21). It is unclear, however, if Caucasians and African Americans with headaches are truly comparable in headache characteristics or if psychometric limitations of these measures prevent the identification of race-related differences in this population.

This study examined the psychometric attributes and measurement invariance (MI) of the HDI (7), the MSQL scale (6), the HMSE scale (22) and the HSLC scale (23). MI is the extent to which scales have the same conceptual meaning across groups, a prerequisite to conducting between-group comparisons. Using data provided by patients in headache subspecialty treatment clinics, this study examined the psychometric adequacy of these measures by examining and comparing between races each measure’s internal consistency and, for two of the four measures, convergent validity. Multi-group confirmatory factor analyses assessed the scales’ MI in Caucasian and African American headache patients. Given the inconsistent findings from research examining the racial validity of measures in general, and the complete lack of research on this topic in headache patients, this study did not test any formal hypotheses. Instead, this research was exploratory in nature and tested whether widely used psychosocial headache measures applied equally well to African American and Caucasian headache patients.

Methods

Study procedures

Between July 2004 and June 2008, 311 patients were recruited in four headache subspecialty treatment centers in Columbus (N = 120), Toledo (N = 108), Cleveland (N = 68), and Cincinnati (N = 15), all cities in the US state of Ohio. Study inclusion criteria were: (i) 18 years of age or older; (ii) satisfying International Headache Society (IHS) criteria for episodic migraine, chronic migraine, episodic TTH, chronic TTH or medication overuse headache; (iii) a determination by the physician that the patient should begin new preventive medication therapy; and (iv) provision of informed consent. This study focused only on the 287 participants who self-identified as African American (N = 114) or Caucasian (N = 173). The primary objective of this project’s parent study was to determine how patient race was related to headache characteristics, adherence to headache treatment protocols and treatment outcomes. This psychometric analysis of widely used psychosocial headache measures is a secondary study of the larger parent project.

Recruitment materials (i.e. brochures and posters) that described the study were distributed to participating clinics for display and distribution to patients. While patients were recruited by site physicians and nurses during the patient’s initial clinic visit, only the physician determined patient eligibility. Because persons of color were relatively underrepresented in all four treatment centers, all racial minority patients who presented for treatment were approached for study enrollment. The larger pool of Caucasian patients enabled the project to use a recruitment strategy in which only every fifth Caucasian patient was approached for study enrollment. Eligible patients who volunteered for the study provided written informed consent in the treatment clinic after being treated by the physician. Patients were initially assessed at pre-treatment and re-assessed at three follow-ups (see Figure 1). The parent study’s methodology is described elsewhere in greater detail (24,25). The study was approved by the university’s Institutional Review Board, and no adverse events were reported during the project.

Figure 1.

Study assessment and treatment schedule. PT, preventive therapy.

Assessment methodologies

This study collected data using four different methodologies: (i) face-to-face headache diagnoses; (ii) telephone-based assessments of psychiatric comorbidity; (iii) 30-day daily diaries (for headache frequency, severity and disability); and (iv) self-report surveys completed by patients during each of their four clinic visits. Each methodology is described below in greater detail. The psychometric properties of study measures were evaluated using data provided by patients during their baseline (M₀) visit.

Headache diagnosis. During the initial patient-physician interaction, the physician diagnosed each patient’s current headache disorder(s) using IHS criteria (26). The physician also documented each patient’s past and current pharmacological headache treatments.

Psychiatric disorder diagnosis. Psychiatric disorders were diagnosed using the Primary Care Evaluation for Mental Disorders (PRIME-MD) (27). The PRIME-MD was administered to all patients enrolled in the study by trained research staff during a telephone interview conducted within two days after the patient completed his or her pre-treatment visit. The PRIME-MD was used because it was designed specifically for use in primary care settings and yields a subset of diagnoses included in the Diagnostic and Statistical Manual of Mental Disorders, including mood and anxiety disorders, alcohol and substance (abuse/dependence), eating disorders and somatoform disorders.

30-Day daily diary. During the interval between the pre-treatment and one-month follow-up visit, patients completed self-administered, paper-and-pencil daily diaries in which they recorded the frequency, severity and disability of headaches experienced during the 30 day assessment period. Headache frequency was the “number of days over the 30-day period during which patients experienced a ‘mild,’ ‘moderate’ or ‘severe’ headache.” Headache severity was assessed for each headache using a 4-point scale (0 = no pain, 1 = mild pain, 2 = moderate pain, 3 = no disability, 1 = mild disability, 2 = moderate disability, 3 = severe disability).

Self-administered surveys. Patients also completed self-administered surveys at the headache treatment clinics during their initial treatment visit. Measures assessed through self-administered surveys are described below:

Headache Disability Inventory (7). The 25-item HDI assessed the perceived impact of headaches on one’s emotional and functional well-being. The HDI comprised two subscales: (i) Emotional-Related Disability (13 items: e.g. “My headaches make me angry”); and (ii) Functional-Related Disability (12 items: e.g. “I do not enjoy social activities because of my headaches”). Higher scores indicated greater overall, emotional and functional disability due to one’s headaches.

Headache-Specific Locus of Control Scale (23). The 33-item HSLC scale measures patients’ beliefs regarding factors that control their headaches. The HSLC scale contained three subscales that assessed patients’ beliefs that their headaches were controlled by their own efforts (Internal subscale, 11 items), chance circumstances (Chance subscale, 11 items) or health care professionals (Health Care Professional subscale, 11 items). Respondents rated the extent to which they agreed with each item using a 5-point scale (1 = “strongly disagree” to 5 = “strongly agree”).

Headache Management Self-Efficacy Scale (8). The 25-item HMSE scale measured patients’ perceived confidence that they could engage in various behavioral strategies to prevent headache episodes or manage headache-related pain and disability. Sample items included: “There are things I can do to prevent headaches,” “Once I have a headache, there is nothing I can do to control it” and “I can prevent headaches by changing how I respond to stress.” Patients rated the extent to which they agreed with each item using a seven-point scale (1 = “strongly disagree” to 7 = “strongly agree”).

Migraine Specific Quality of Life Scale (6). A 14-item adapted version of the MSQL assessed the impact of headache disorders on patients’ quality of life. Because the original MSQL items referred only to migraines, the term “migraine” was replaced with the more general term “headache” when relevant. For example, the original MSQL item “In the past 4 weeks, how often have migraines interfered with how well you dealt with family, friends and others close to you” was changed to “In the past 4 weeks, how often have your headaches interfered with how well you dealt with family, friends and others close to you” (emphasis added). The MSQL scale comprised three subscales: (i) Role Restriction (7 items: e.g. “How often have headaches left you too tired to do work or daily activities?”); (ii) Role Prevention (4 items: e.g. “How often were you not able to go to social activities such as parties and dinner with friends because of your headaches?”); and (iii) Emotional Functioning (3 items: e.g. “How often have you felt fed up or frustrated because of your headaches?”). The overall scale and the three subscales were scored so that lower scores indicated better quality of life.

Demographics. As part of the 30-day diary, patients indicated the racial group with which they most identified, their age, gender, number of years of education, current employment (full-time, part-time, unemployed), type of health insurance coverage (e.g. health maintenance organization, preferred provider, private, government-sponsored) and annual income.

Data analytic procedures

Coefficient alpha was calculated for each measure by race to assess its internal consistency. The convergent validity of the HDI and MSQL scale was assessed by calculating the Pearson Product Moment correlation coefficient between these two scales and measures that were assessed through different methodologies but that, theoretically, should have correlated with the multi-item measures. Finally, robust maximum-likelihood estimation multigroup confirmatory factor analyses (CFA) were used to test the equivalence of the MSQL scale, the HDI, the HMSE scale and the HSLC scale in African American and Caucasian samples. Multigroup CFAs were conducted using LISREL 8.54 (28). Because the amount of missing data was minimal (<5%), missing values were not imputed; analyses were conducted using only those cases that contained complete data. Test-retest reliability coefficients were not calculated for study measures because the changes and/or adjustments in medications administered at the M₀ baseline visit (e.g. changes in acute therapies and/or previously prescribed preventive therapies) may have altered patients’ headache activity between the M₀ and M₁ treatment visits and reduced each measure’s test-retest reliability.

Results

Participant characteristics and headache treatments at baseline

The average patient was 36.5 years of age, female (88%) and had completed 14.0 years of education. The modal headache diagnosis was episodic migraine (with or without aura). The average patient experienced 17.3 headache days per month and had been living with headaches for 7.9 years (standard deviation [SD] = 9.9 years, median = 4.0 years). Acute medications being taken by patients at their M₀ baseline visit (and prior to the initiation of new preventive headache therapies) included triptans (34%), ergotamine/antiemetic (21%) and simple/compound analgesics (15%). Preventive therapies being taken by patients at their M₀ baseline visit were vitamins (16%), anticonvulsants (9%), other antidepressants (7.4%) tricyclic antidepressants (6%), and beta blockers (4%).

Baseline differences between Caucasians and African Americans

One-way ANOVA and chi-square tests of association identified psychosocial differences between the 173 Caucasian and 114 African American patients prior to initiating treatment (see Table 1). Caucasians (mean = 35.6 years) were significantly younger than African Americans (mean = 38.4 years), F (1,245) = 4.2, p < .05, d = .25. Caucasians (mean = 14.3 years) had completed more years of education than African Americans (mean = 13.5 years), F (1,253) = 7.9, p < .01, d = .34. African Americans (92%) were slightly more likely than Caucasians (85%) to be female, χ² (1) = 2.8, p < .10, d = .37. African Americans were significantly more likely to be diagnosed with major depressive disorder (Caucasians = 24.1%, African Americans = 40.4%, χ²{1} = 8.3, p < .004, d = .46), dysthymia (Caucasians = 15.3%, African Americans = 30.3%; χ²{1} = 12.1, p < .001, d = .51), and slightly more likely to receive a diagnosis of anxiety/not otherwise specified (Caucasians = 23.5%, African Americans = 30.3%; χ²{1} = 3.0, p < .010, d = .21). African Americans experienced more headache days during the past month (mean = 18.9 days) than Caucasians (mean = 16.3), F (1,190) = 4.9, p < .03, d = .27. Finally, African Americans reported more severe headaches (mean = 1.77) than Caucasians (mean = 1.62), F(1, 190) = 7.2, p < .01, d = .32. African Americans and Caucasians did not differ in employment, annual income, health insurance coverage, number of years with headache disorder(s) or headache diagnoses (all p _s > .10).

Table 1.

Sociodemographic and clinical characteristics of Caucasian and African American patients; mean (SD)

Characteristic	Overall Sample (N = 230)	Caucasians (N = 156)	African Americans (N = 92)	χ² or t	p <
Age	36.5 (10.4)	35.6 (10.2)	38.4 (10.5)	2.1	.04
Being female	87.8%	85.2%	92.1%	2.8	.10
Years of education	14.0 (2.3)	14.3 (2.1)	13.5 (2.2)	2.8	.005
Annual income < USD 40,000	66.4%	65.4%	70.6%	6.7	.01
Diagnosed with major depressive disorder	30.5%	24.1%	40.4%	8.3	.004
Diagnosed with dysthymia	22.2%	15.3%	33.0%	12.1	.001
Diagnosed with anxiety disorder (not otherwise specified)	27.2%	23.5%	33.0%	3.0	.09
Diagnosed with chronic migraine (with or without aura)	24.4%	24.0%	25.0%	0.04	0.85
Diagnosed with episodic migraine (with or without aura)	45.2%	52.0%	34.8%	8.1	.005
Diagnosed with chronic tension-type headache	17.3%	15.2%	20.5%	1.3	.25
Diagnosed with episodic tension-type headache	2.1%	0.00%	5.4%	9.4	.002
Years with headache disorder(s)	7.9 (9.9)	7.6 (8.8)	8.5 (11.7)	.70	.50
Headache days/month*	17.3 (7.8)	16.3 (7.9)	18.9 (7.4)	2.2	.03
Headache severity (past month)*	1.68 (0.39)	1.62 (0.32)	1.77 (0.47)	2.7	.008

SD, standard deviation; USD, US dollars. *Based on data obtained through daily diaries completed after patient’s initial appointment.

Coefficient alpha of headache measures by race

As shown in Table 2, coefficient alpha for each multi-item measure was generally very good for both Caucasian and African American patients. Among Caucasian patients, coefficient alpha ranged from 0.82 to 0.95 (mean α = 0.87). Among African American patients, coefficient alpha ranged from 0.81 to 0.96 (mean α = 0.90).

Table 2.

Psychometric attributes of headache-specific measures by race

Scale	Caucasians (N = 173)	African Americans (N = 114)
Internal consistency (α)
Headache Disability Inventory (total)	0.94	0.95
Emotional subscale	0.89	0.92
Functional subscale	0.88	0.91
Headache-Specific Locus of Control
Health Care Provider subscale	0.82	0.84
Internal Locus of Control subscale	0.84	0.86
Chance Locus of Control subscale	0.82	0.81
Migraine-Specific Quality of Life (total)	0.95	0.96
Role Restriction subscale	0.91	0.95
Role Prevention subscale	0.90	0.91
Emotional Functioning subscale	0.83	0.86
Headache Management Self-Efficacy	0.85	0.88
Convergent validity (Pearson r)
Correlation between HDI (self-report) and headache disability assessed through 30-day daily diaries	0.38***(N = 122)	0.48***(N = 58)
Correlation between Headache-Specific Locus of Control^# (self-report) and headache severity assessed through daily diaries	0.28**(N = 124)	0.27*(N = 63)

HDI, headache disability inventory. ***p < .001. **p < .01. *p < .05. ^#Higher scores indicate poorer quality of life.

Convergent validity of HDI and MSQL by race

Table 2 shows the results of analyses assessing the convergent validity of the HDI and the MSQL scale by race. The 25-item HDI completed using the study’s self-report survey correlated significantly with headache disability scores assessed via 30-day self-report daily diaries in both Caucasians, r(121) = .38, p < .001 and African Americans, r(57) = .48, p < .001. These correlations did not differ significantly (z = 0.75, p = 0.46). The correlation between the MSQL scale completed using a self-report survey and headache severity assessed through 30-day daily diaries in Caucasian patients, r(123) = .28 was essentially identical to the correlation between the two measures in African American patients, r(62) = .27 (z = 0.07, p = 0.95).

Analyses of measurement invariance

Robust maximum-likelihood estimation multi-group confirmatory factor analyses tested the equivalence of the MSQL scale, the HDI, the HMSE scale and the HSLC scale between African American and Caucasian samples. This method computed a Satorra-Bentler (scaled chi square) (SB χ²) and robust standard errors that adjusted for multivariate kurtosis (29). The absolute fit of the measurement model was evaluated using the SB χ², with a preference for models with a SB χ² to degrees of freedom ratio of ≤2 (30). Because the traditional chi-square statistic as a test of absolute fit is sensitive to sample size (31), the Comparative Fit Index (CFI) was used. CFI values range from 0 to 1, with values ≥ .90 and ≥ .95 representing adequate and excellent model fit, respectively (32,33). The Root Mean Square Error of Approximation (RMSEA) (34) statistic was supplied as an indication of the population error variance (35). Interpretation of RMSEA values in terms of fit were as follows: good (<0.05), acceptable (0.05–0.08), marginal (0.08–0.10), and poor (>0.10) (35–37). For differences between nested models, the SB χ² difference test was used (38).

MI analyses determined if the same factor structures provided an adequate fit for both African American and Caucasian samples by testing models that had increasingly restrictive and hierarchical constraints. More specifically: Model 1, Configural Invariance, tested the measurement model in which the factor pattern (pattern of factor loadings) was equal between groups; Model 2, Weak Factorial Invariance, added equal factor loadings to the constraint of equal factor pattern; Model 3, Strong Factorial Invariance, added equal intercepts; Model 4, Strict Factorial Invariance, tested equal factors loadings, intercepts, and error variances; and Model 5, Strict Factorial Invariance, added equal factor variances. If the constraints placed on the factor structures resulted in a worse model fit (as evidenced by a significant increase in the SB χ² difference test) (38), the model was not considered to meet that particular level of invariance.

Factor structure and measurement invariance of the MSQL scale. The MSQL scale was constructed to include three subscales: Role Restriction, Role Prevention and Emotional Functioning. The multi-group CFA (Table 3) revealed an excellent model fit that met criteria for Strict factorial invariance. Factor loadings of the MSQL scale appear in Table 4.

Table 3.

MSQL scale goodness of fit model statistics for multi-group confirmatory factor analyses

Model/description	χ²	SB χ² /df	p value	RMSEA	CFA	ΔSB χ²
Model 1: configural invariance (same pattern of loadings)	403.48	265.84/148 = 1.80	.00	0.075	0.98
Model 2: weak factorial invariance (equal factor loadings)	371.86	282.88/159 = 1.78	.00	0.067	0.98	Model 1 to 2, p = 0.02
Model 3: strong factorial invariance (equal factor loadings and intercepts)	371.86	287.33/170 = 1.69	.00	0.063	0.98	Model 2 to 3, p = 0.99
Model 4: strict factorial invariance (equal factor loadings, intercepts and error variances)	371.86	284.38/184 = 1.55	.00	0.056	0.99	Model 3 to 4, p = 0.99
Model 5: strict factorial invariance (equal factor loadings, intercepts, error and factor variances)	371.86	284.03/190 = 1.49	.00	0.054	0.99	Model 4 to 5, p = 0.99

MSQL, migraine-specific quality of life scale; SB, Satorra-Bentler (scaled chi square); df, differential; RMSEA, root mean square error of approximation; CFA, confirmatory factor analysis.

Table 4.

Factor structure of the MSQL for Caucasians and African Americans

	Factor loadings
Scale item	Role restriction	Role prevention	Emotional functioning
How often have you had difficulty in performing work or daily activities because of headache symptoms?	0.89	—	—
How often did headaches limit your ability to concentrate on work or daily activities?	0.88	—	—
How often did headaches keep you from getting as much done at work or at home at work or at home?	0.87	—	—
How often have headaches interfered with your leisure time activities, such as reading or exercise?	0.84	—	—
How often have headaches left you too tired to do work or daily activities?	0.82	—	—
How often have headaches interfered with how you dealt with family, friends and others who are close to you?	0.78	—	—
How often have headaches limited the number of days you have felt energetic?	0.76	—	—
How often have you had to cancel work or daily activities because you had a headache?	—	0.86	—
How often did you have to stop work or daily activities to deal with headache symptoms?	—	0.85	—
How often were you not able to go to social activities such as parties, dinner with friends, because you had a headache?	—	0.83	—
How often did you need help in handling routine tasks such as every day household chores, doing necessary business, shopping or caring for others, when you had a headache?	—	0.79	—
How often have you been afraid of lettings others down because of your headaches?	—	—	0.84
How often have you felt like you were a burden on others because of your headaches?	—	—	0.86
How often have you felt fed up or frustrated because of your headache?	—	—	0.70

MSQL, migraine-specific quality of life scale.

Factor structure and measurement invariance of the HDI. The HDI was purported to comprise two subscales: Emotional Disability and Functional Disability. The multi-group CFA (Table 5) revealed an excellent model fit that met criteria for Weak Factorial Invariance. Given that subsequent constraints (Models 3–5) suggested only unequal intercepts, a modified invariant model (Model 6) was tested in which mean structures were excluded. This model found that, aside from unequal intercepts, the HDI had invariant factor loadings, error variances, and factor variances. HDI factor loadings based on Model 6 appear in Table 6.

Table 5.

HDI goodness of fit model statistics for multi-group confirmatory factor analyses

Model/description	χ²	SB χ² /df	p value	RMSEA	CFA	ΔSB χ²
Model 1: configural invariance (same pattern of loadings)	1260.50	842.96/548 = 1.54	.00	0.062	0.95
Model 2: weak factorial invariance (equal factor loadings)	1269.91	872.74/571 = 1.53	.00	0.061	0.95	Model 1 to 2, p = .70
Model 3: strong factorial invariance (equal factor loadings and intercepts)	1309.47	923.12/594 = 1.55	.00	0.063	0.94	Model 2 to 3, p = < .001
Model 4: strict factorial invariance (equal factor loadings, intercepts and error variances)	1331.42	1047.82/619 = 1.69	.00	0.070	0.94	Model 3 to 4, p = .99
Model 5: strict factorial invariance (equal factor loadings, intercepts, error and factor variances)	1332.52	1051.74/622 = 1.69	.00	0.070	0.94	Model 4 to 5, p = .54
Model 6: modified factorial invariance excluding mean structure (equal factor loadings, error and factor variances)	1297.30	1004.89/599 = 1.68	.00	0.069	0.94	Model 2 to 6, p = .99

HDI, headache disability inventory; SB, Satorra-Bentler (scaled chi square); df, differential; RMSEA, root mean square error of approximation; CFA, confirmatory factor analysis.

Table 6.

Factor structure of the HDI for Caucasians and African Americans

Scale item	Factor loadings
Emotional	Functional
I feel desperate because of my headaches.	0.81	—
Sometimes I feel that I am going to lose control because of my headaches.	0.74	—
My headaches are so bad that I feel I am going to go insane.	0.73	—
My outlook on the world is affected by my headaches.	0.72	—
My headaches place stress on my relationships with family or friends.	0.66	—
My headaches make me feel frustrated.	0.64	—
My headaches make me feel confused.	0.63	—
I feel irritable because of my headaches.	0.62	—
I am afraid to go outside when I feel that a headache is starting.	0.61	—
Because of my headaches, I feel handicapped.	0.60	—
My headaches make me angry.	0.58	—
No one understands the effects that my headaches have on my life.	0.52	—
My spouse (significant other), or family and friends, have no idea what I am going through because of my headaches.	0.39	—
I believe my headaches are making it difficult for me to achieve my goals in life.	—	0.73
I do not enjoy social gatherings because of my headaches.	—	0.71
I avoid traveling because of my headaches.	—	0.72
I am unable to think clearly because of my headaches.	—	0.70
Because of my headaches I am less likely to socialize.	—	0.69
I find it difficult to focus my attention away from my headaches and on other things.	—	0.65
Because of my headaches I feel restricted in performing my routine daily activities.	—	0.65
I am concerned that I am paying penalties at work or at home because of my headaches.	—	0.65
I restrict my recreational activities (e.g. sports, hobbies) because of my headaches.	—	0.64
I avoid being around people when I have a headache.	—	0.58
I find it difficult to read because of my headaches.	—	0.57
I get tense (e.g. muscle tension) because of my headaches.	—	0.47

HDI, headache disability inventory.

Factor structure and measurement invariance of the HMSE scale. The HMSE scale was modeled to have one total score (8). The multi-group CFA (Table 7) revealed a model that provided a poor fit to the data but that also met criteria for Weak Factorial Invariance. HMSE scale factor loadings based on Model 2 appear in Table 8.

Table 7.

HMSE scale goodness of fit model statistics for multi-group confirmatory factor analyses

Model/description	χ²	SB χ² /df	p value	RMSEA	CFA	ΔSB χ²
Model 1: configural invariance (same pattern of loadings)	1693.84	1302.86/551 = 2.36	.00	0.098	0.80
Model 2: weak factorial invariance (equal factor loadings)	1687.46	1336.26/574 = 2.33	.00	0.097	0.80	Model 1 to 2, p = .80
Model 3: strong factorial invariance (equal factor loadings and intercepts)	1865.28	1527.93/598 = 2.55	.00	0.105	0.78	Model 2 to 3, p = < .001
Model 4: strict factorial invariance (equal factor loadings, intercepts, and error variances)	1894.32	1594.07/623 = 2.56	.00	0.105	0.78	Model 3 to 4, p = < .001
Model 5: strict factorial invariance (equal factor loadings, intercepts, error and factor variances)	1894.83	1597.88/624 = 2.56	.00	0.105	0.78	Model 4 to 5, p = .25
Model 6: modified factorial invariance excluding mean structure (equal factor loadings, error and factor variances)	1711.54	1400.55/600 = 2.33	.00	0.097	0.80	Model 2 to 6, p = < .001
Model 7: (equal factors loadings, error &factor variances)	1688.04	1339.26/575 = 2.33	.00	0.097	0.80	Model 2 to 7, p = .02
Model 8: (equal factors loadings, error &factor variances)	1711.19	1397.10/599 = 2.33	.00	0.097	0.80	Model 2 to 8, p = < .001

HMSE, headache management self-efficacy scale; SB, Satorra-Bentler (scaled chi square); df, differential; RMSEA, root mean square error of approximation; CFA, confirmatory factor analysis.

Table 8.

Factor structure of the HMSE scale for Caucasians and African Americans

Scale item	Factor loading
There are things I can do to prevent headaches.	0.66
I can do things to control how much my headaches interfere with my life.	0.61
When I sense a headache is coming, there is nothing I can do to stop it.	0.59
I can keep a mild headache from disrupting my day by changing the way I respond to the pain.	0.59
I can prevent headaches by recognizing headache triggers.	0.55
I can do things that make a headache seem not so bad.	0.55
I can prevent headaches by changing how I respond to stress.	0.54
I can do things to cope with my headaches.	0.50
If I do certain things every day, I can reduce the number of headaches I will have.	0.49
If I can catch a headache before it begins, I can stop it.	0.49
I can control the intensity of headache pain.	0.49
I can do things that will control how long a headache lasts.	0.47
I can reduce the intensity of a headache by relaxing.	0.46
There are things I can do to reduce headache pain.	0.46
Nothing I do will keep a mild headache from turning into a bad headache.	0.46
When I'm not under a lot of stress, I can prevent many headaches.	0.42
Once I have a headache, there is nothing I can do to control it.	0.41
When I'm tense, I can prevent headaches by controlling the tension.	0.40
Nothing I do reduces the pain of a headache.	0.40
I cannot control the tension that causes my headaches.	0.36
If I am under a lot of stress, there is nothing I can do to prevent headaches.	0.30
When I'm in some situations, nothing I do will prevent headaches.	0.27
If I am upset, there is nothing I can do to control the pain of a headache.	0.25
I can keep even a bad headache from disrupting my day by changing the way I respond to the pain.	0.24
Nothing I do will keep a bad headache from disrupting my day.	0.24

HMSE, headache management self-efficacy scale.

Factor structure and measurement invariance of the HSLC scale. The HSLC scale was modeled to have three subscales: Internal Locus of Control, Chance Locus of Control and Health Care Provider Locus of Control. Results of the multi-group CFA (Table 9) were marginal overall, given the mixed fit indices. All indices suggested a good to acceptable fit except for the CFI, which neared acceptable. In terms of invariance, the HSLC met the highest standard, Strict Invariance. HSLC factor loadings based on Model 5 appear in Table 10.

Table 9.

HSLC goodness of fit model statistics for multi-group confirmatory factor analyses

Model/description	χ²	SB χ² /df	p value	RMSEA	CFA	ΔSB χ²
Model 1: configural invariance (same pattern of loadings)	2214.54	1930.12/987 = 2.00	.00	0.075	0.84
Model 2: weak factorial invariance (equal factor loadings)	2079.89	1821.37/1014 = 1.80	.00	0.068	0.87	Model 1 to 2, p = > .99
Model 3: strong factorial invariance (equal factor loadings and intercepts)	2079.89	1827.90/1044 = 1.75	.00	0.066	0.87	Model 2 to 3, p = .94
Model 4: strict factorial invariance (equal factor loadings, intercepts and error variances)	2079.89	1835.66/1077 = 1.70	.00	0.064	0.87	Model 3 to 4, p = >.99
Model 5: strict factorial invariance (equal factor loadings, intercepts, error and factor variances)	2079.89	1836.58/1083 = 1.70	.00	0.064	0.88	Model 4 to 5, p = .99

HSLC, headache-specific locus of control scale; SB, Satorra-Bentler (scaled chi square); df, differential; RMSEA, root mean square error of approximation; CFA, confirmatory factor analysis.

Table 10.

Factor structure of the HSLC scale for Caucasians and African Americans

Scale item	Factor loadings
Internal	Chance	Health care
By not becoming agitated or overactive, I can prevent many headaches.	0.75	—	—
When I drive myself too hard, I get headaches.	0.68	—	—
My headaches are worse when I am coping with stress.	0.66	—	—
I can prevent some of my headaches by not getting emotionally upset.	0.65	—	—
When I worry or ruminate about things, I am more likely to get headaches.	0.64	—	—
I can prevent some of my headaches by avoiding certain stressful situations.	0.60	—	—
If I remember to relax I can avoid some of my headaches.	0.53	—	—
My actions influence whether I get headaches.	0.53	—	—
When I have not been taking proper care of myself, I am likely to experience headaches.	0.53	—	—
I am directly responsible for getting some of my headaches.	0.51	—	—
My headaches are sometimes worse because I am overactive.	0.22	—	—
I am likely to get headaches no matter what I do.	—	0.73	—
Often I feel that no matter what I do, I will still have a headache.	—	0.67	—
No matter what I do, if I am going to get a headache, I will get a headache.	—	0.62	—
My headaches are beyond my control.	—	0.57	—
My not getting headache is largely a matter of good fortune.	—	0.54	—
I am completely at the mercy of my headaches.	—	0.53	—
When I get headaches, I just have to let nature run its course.	—	0.50	—
It’s a matter of fate whether I have a headache.	—	0.49	—
When I have a headache, there is nothing I can do to affect its course.	—	0.48	—
Luck plays a big part in determining how soon I will recover from a headache.	—	0.45	—
I’m just plain lucky for a month when I don’t get headaches.	—	0.42	—
Having regular contact with my physician is the best way for me to control my headaches.	—	—	0.79
My headaches can be less severe if medical professionals take proper care of me.	—	—	0.68
When I have headaches, I should consult a medically trained professional.	—	—	0.66
Health professional keep me from getting headaches.	—	—	0.61
Following the doctor’s medication regimen is the best way for me to not be laid-up with a headache.	—	—	0.54
If I don’t have the right medication, my headaches will be a problem.	—	—	0.54
Only my doctor can give me ways to prevent my headaches.	—	—	0.49
I usually recover from headaches when I get proper medical help.	—	—	0.48
Just seeing my doctor helps my headaches.	—	—	0.46
When my doctor makes a mistake, I am the one to suffer with headaches.	—	—	0.38
My doctor’s treatments can help my headaches.	—	—	0.32

HSLC, headache-specific locus of control scale.

Discussion

As researchers investigate potential race differences in persons with headache disorders, a critical prerequisite is the establishment of MI of the multi-item headache measures used to detect these differences. Currently, it is assumed that psychosocial headache measures apply equally well to headache patients of different racial groups. This research tested this assumption by assessing the MI of widely used psychosocial headache measures in Caucasian and African American headache patients.

The psychosocial headache measures assessed in this study evinced good internal consistency for both Caucasian and African American headache patients. Coefficient alpha for the psychosocial measures always exceeded 0.80 for both groups, a criterion commonly used to demonstrate adequate internal consistency (39). Two widely used psychosocial headache measures (i.e. the HDI and the MSQL scale) demonstrated good convergent validity. Specifically, the self-report HDI correlated significantly with headache disability assessed through 30-day daily diaries while the self-report MSQL scale correlated significantly with headache severity recorded via 30-day daily diaries in both racial groups. These findings should increase researchers’ confidence in headache patients’ abilities to provide accurate data about their headache disability and quality of life assessed through more global measures that use longer retrospective recall periods.

MI was established for the HDI, the MSQL scale, and the HSLC scale. This suggests that these measures apply equally well to Caucasian and African American headache patients and that race-related differences observed on these measures are, in all likelihood, identifying actual between-group differences that are not the result of measurement artifacts.

While the HMSE scale evinced adequate MI, the measure’s established factor structure provided a poor fit for both Caucasians and African Americans as assessed by several indices (CFA ≤ .80, RMSEA ≥ .10; and SB χ² /df ≥ 2.33). These findings suggest that while Caucasians and African Americans conceptualized self-efficacy similarly, the scale not be a particularly good assessment of headache management self-efficacy for either group. These results were somewhat unexpected given the scale’s high coefficient alpha in Caucasians (α = 85) and African Americans (α = .88). These findings also demonstrate why relying solely on a scale’s coefficient alpha is insufficient to substantiate its psychometric adequacy. Because the scale’s developers (8) speculated that the HMSE scale may contain two subscales (i.e. Headache Prevention and Headache Management subscales), the scale may have provided a better fit to data collected in the current study if two factors had been specified.

Future research should test if the HMSE scale is unidimensional, comprised of two or more subscales, or, alternatively, develop new measures of headache management self-efficacy. The poor fit of the scale in this study may have also been due to one or more characteristics of the sample (i.e. patients in participating subspecialty clinics reported very frequent and disabling headaches), the current study’s use of CFA as opposed to exploratory factor analyses (EFA), or other methodological factors of which we are unaware. A future measure of headache management self-efficacy might also be more psychometrically rigorous, and clinically meaningful, if it assesses self-efficacy in a specific domain (e.g. headache prevention self-efficacy, behavioral headache management self-efficacy).

This study has several limitations. All participating headache treatment centers were located in relatively large cities in Ohio; the extent to which study findings generalize to other geographic areas is unclear. The study’s sample had only adequate numbers of patients to assess the psychometric properties of study measures in Caucasian and African American headache patients. Future research should assess the psychometric attributes of these and other psychosocial headache measures with additional racial groups, such as Latinos/Latinas, Asian Americans/Pacific Islanders, and Native Americans.

Despite these limitations, this study adds to the existing headache literature by establishing MI between Caucasian and African American headache patients on several widely-used psychosocial headache measures. In both racial groups, the HDI, the HSLC scale, and the MSQL scale evidenced acceptable internal consistency, convergent validity, and MI, suggesting that these measures are likely to perform well for Caucasian and African American headache patients in cross-sectional research. Findings demonstrating MI between Caucasians and African Americans indicated that the two racial groups were using the same conceptual framework when completing these measures and that between-group comparisons on these measures are appropriate.

As headache researchers continue to examine potential race-related differences in headache characteristic in headache patients (or between-group differences on other variables such as gender, age and socioeconomic status), it will be increasingly important to consider carefully the psychometric attributes and MI of measures used in research of this type. As increasingly sophisticated data analytic methods become available to headache researchers (e.g. multi-group CFA), it is important that these approaches are integrated into their research to ensure that measures being used are assessing the constructs they purport to assess and to justify the conduct of between-group comparisons that are often the focus of contemporary headache research.

Footnotes

Acknowledgements

This research was supported by Grant K01 NS046582 from the National Institute of Neurological Disorders and Stroke (B. Heckman, P.I.).

References

Scher

Stewart

Liberman

Lipton

. Prevalence of frequent headache in a population sample. Headache 1998; 38: 497–506.

Silberstein

Loder

Diamond

. Probable migraine in the United States: results of the American migraine prevalence and prevention (AMPP) study. Cephalalgia 2007; 27: 220–229.

Stewart

Lipton

Liberman

. Variation in migraine prevalence by race. Neurology 1996; 47: 52–59.

Lipton

Bigal

Diamond

Freitag

Reed

Stewart

. Migraine prevalence, disease burden, and the need for preventive therapy. Neurology 2007; 5: 343–349.

Heckman

Holroyd

O'Donnell

. Race differences in adherence to headache treatment appointments in persons with headache disorders. J Natl Med Assoc 2008; 100: 247–255.

Jhingran

Osterhaus

Miller

Lee

Kirchdoerfer

. Development and validation of the migraine-specific quality of life questionnaire. Headache 1998; 38: 295–302.

Jacobson

Ramadan

Aggarwal

Newman

. The Henry Ford Hospital headache disability inventory (HDI). Neurology 1994; 44: 837–842.

French

Holroyd

Pinell

Malinoski

O'Donnell

Hill

. Perceived self-efficacy and headache-related disability. Headache 2000; 40: 647–656.

Gerbaud L Navez

Couratier

. Validation of the combined SF36/MSQOL test of evaluation of quality of life in migraine patients in France. Revue Neurologique 2002; 158: 719–727.

10.

Ohbu

Igarashi

Okayasu

. Development and testing of the Japanese version of the Migraine-Specific Quality of Life Instrument. Qual Life Res 2004; 13: 1489–1493.

11.

Hansen

Bendtsen L Jensen

. Psychometric properties of the Danish versions of headache-specific locus of control scale and headache management self-efficacy scale. J Headache Pain 2009; 10: 341–347.

12.

Fritsche

Hueppe

Kukava

. Validation of a German language questionnaire for screening for Migraine, Tension-Type Headache, and Trigeminal Autonomic Cephalalgias. Headache 2007; 4: 546–551.

13.

Pérez E. Lost in translation? Validity and reliability in bilingual political surveys. In: Proceedings of the Midwestern Political Science Association 2007 Annual Meeting, 12 April, 2007. Ipswich, MA: PUBLISHING ORG, 2007, p.1–39.

14.

Geisinger

. Cross-cultural normative assessment: translation and adaptation issues influencing the normative interpretation of assessment instruments. Psychol Assessment 1994; 6: 304–312.

15.

Sandoval

Duran

Language . In: Jonathan

Sandoval

Craig

Frisby

(eds) Test interpretation and diversity: achieving equity in assessment. Washington, DC: American Psychological Association, 1998, 181–211.

16.

Warnecke

Johnson

Chávez

. Improving question wording in surveys of culturally diverse populations. Ann Epidemiol 1997; 7: 334–342.

17.

Krall

Valadian

Dwyer

Gardner

. Accuracy of recalled smoking data. Am J Public Health 1989; 79: 200–202.

18.

Matthews

Nattinger

Anderson

. Accuracy and certainty of self-report for colorectal cancer testing among ambulatory patients. Psych Health Med 2005; 10: 1–15.

19.

Cherpitel

Bazargan

. Screening for alcohol problems: comparison of the audit, RAPS4 and RAPS4-QF among African American and Hispanic patients in an inner city emergency department. Drug Alcohol Depend 2003; 71: 275–280.

20.

Carle

. Cross-cultural validity of alcohol dependence across Hispanics and non-Hispanic Caucasians. Hispanic J Behav Sci 2008; 30: 106–120.

21.

Nicholson

Rooney

O'Laughlin

Gordon

. Migraine care among different ethnicities: do disparities exist? Headache 2006; 46: 754–765.

22.

Martin

Holroyd

Rokicki

. The headache self-efficacy scale: adaptation to recurrent headaches. Headache 1993; 33: 244–248.

23.

Martin

Holroyd

Penzien

. The headache-specific locus of control scale: adaptation to recurrent headaches. Headache 1990; 30: 729–734.

24.

Heckman

Holroyd

Tietjen

. Whites and African-Americans in headache specialty clinics respond equally well to treatment. Cephalalgia 2009; 29: 650–661.

25.

Heckman

Holroyd

Himawan

. Do psychiatric comorbidities influence headache treatment outcomes? Results of a naturalistic longitudinal treatment study. Pain 2009; 146: 56–64.

26.

Headache Classification Subcommittee of the International headache Society. International classification of headache disorders. 2nd edn. Cephalalgia 2004; 24(Suppl 1): 9–160.

27.

Spitzer

Williams

JBW

Kroenke

. Utility of a new procedure for diagnosing mental disorders in primary care: the PRIME-MD 1000 study. J Am Med Assoc 1994; 272: 1749–1756.

28.

Jöreskog

Sörbom

. LISREL 8.54: user’s reference guide. Chicago: Scientific Software International, 2003.

29.

Satorra

Bentler

. Corrections to test statistics and standard errors in covariance structure analysis. In: von Eye

Clogg

(eds) Latent variable analysis: applications for developmental research. Thousand. Oaks, CA: Sage Publications, Inc., 1994, 399–419.

30.

Ullman

. Structural equation modeling. In: Tabachnick

Fidell

(eds) Using multivariate statistics, 4th edn. New York: Allyn &Bacon, 2001, 653–771.

31.

Bentler

. Evaluating model fit. In: Hoyle

(ed.) Structural equation modeling: concepts, issues, and applications. Thousand Oaks, CA: Sage Publications, Inc., 1995, 76–99.

32.

Bollen

. Structural equations with latent variables. Oxford: John Wiley &Sons, 1989.

33.

Hoyle

Panter

. Writing about structural equation modeling. In: Hoyle

(ed.) Structural equation modeling: concepts, issues, and applications. Thousand Oaks, CA.: Sage Publications, Inc, 1995, 158–176.

34.

MacCallum

Browne

Sugawara

. Power analysis and determination of sample size for covariance structure modeling. Psychol Methods 1996; 1: 130–149.

35.

Browne

Cudeck

. Alternative ways of assessing model fit. In: Bollen

Long

(eds) Testing structural equation models. Newbury Park, CA: Sage Publications, Inc., 1993, 136–161.

36.

Fabrigar

Wegener

MacCallum

Strahan

. Evaluating the use of exploratory factor analysis in psychological research. Psychol Methods 1999; 4: 272–299.

37.

Bentler

. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct Equ Modeling 1999; 6: 1–55.

38.

Satorra

. Scaled and adjusted restricted tests in multi-sample analysis of moment structures. In: Heijmans

RDH

Pollock

DSG

Satorra

(eds) Innovations in multivariate statistical analysis: A Festschrift for Heinz Neudecker. London: Kluwer Academic Publishers, 2000, 233–247.

39.

Nunnally

. Introduction to psychological measurement. New York: McGraw-Hill, 1970.