Sage Journals: Discover world-class research

Abstract

The aim was to estimate lifetime sex and age-specific incidence of migraine. Data are from the American Migraine Prevalence and Prevention study, a mailed survey sent to 120 000 US households. Age-specific incidence was estimated using self-reported data relevant to identification of migraine cases, age of onset of migraine and age at interview. Migraine incidence peaked between the ages of 20 and 24 years in women (18.2/1000 person-years) and the ages of 15 and 19 years in men (6.2/1000 person-years). Cumulative incidence was 43± in women and 18± in men. Median age of onset was 25 years among women and 24 years among men. Onset in 50± of cases occurred before age 25 and in 75± before age 35 years. Four of every 10 women and two of every 10 men will contract migraine in their lifetime, most before age 35 years. The incidence estimates from this analysis are consistent with those reported in previous longitudinal studies.

Keywords

Migraine incidence lifetime risk

Evidence on the age-specific incidence of disease offers important clues to aetiology. Although migraine prevalence has been extensively described (1), relatively little is known about its incidence. Prospective studies of migraine incidence are uncommon and confined to selected age groups (2–8). Moreover, estimates from the relatively few incidence studies are highly variable and based on relatively few new-onset cases. As an alternative to longitudinal studies, cross-sectional data can be used to estimate age-specific incidence (9, 10). Although secular trends of increasing or decreasing migraine incidence can influence estimates derived from cross-sectional studies, evidence suggests that migraine prevalence has not increased in a 10-year period (11, 12).

In this study, we used data on current age and age of migraine onset from a large national survey to estimate age-specific incidence. Our findings are consistent with previously reported incidence estimates based on small samples. An advantage of the results from this study in contrast to longitudinal studies is the large sample and broad age coverage. Our analysis indicates that the cumulative lifetime risk of migraine is 2.5–3 times higher than prevalence. We believe these findings have implications for clinical care and for aetiological and health services research.

Methods

In order, we describe the source of the population, data collection method and data collected, and the statistical methods for estimating incidence.

Population survey

Data for estimating incidence rates were obtained from a US nationwide mailed survey of 163 186 individuals ≥ 12 years of age participating in the American Migraine Prevalence and Prevention (AMPP) study. The AMPP was modelled on the methods of the American Migraine Studies 1 and 2, described in detail elsewhere (11–13). A self-administered headache questionnaire was mailed to a stratified random sample of 120 000 US households, drawn from a 600 000-household nationwide panel maintained by the National Family Opinion, Inc. (NFO). All NFO participants first complete a NFO baseline questionnaire used to obtain data on the household, including details on the head of the household (i.e. the self-defined NFO point of contact), the total number of household members and demographic details (e.g. age, sex, education level) of household members. The AMPP questionnaire was sent to the designated head of household, who was instructed to identify all household members suffering from at least occasional self-defined severe headache. Each household member with severe headaches was asked to complete a symptom screening questionnaire, consisting of 21 questions assessing all headache features important in the detection of migraine. The screening questionnaire has been validated [i.e. taking the second edition of the International Classification of Headache Disorders (ICHD) as a reference (14)] in a population sample of migraineurs and of controls with other types of headache. The sensitivity of the survey for this sample was 100%; the specificity was 82.3% (15).

Migraine case definition

Identification of active migraine cases (i.e. reporting at least one attack in the previous year) was based on established ICHD-II criteria (14) and assigned if a respondent reported at least one severe headache in the previous 12 months, with unilateral or pulsatile pain, and either nausea, vomiting, phonophobia with photophobia, or visual or sensory aura before the headache. Individuals who reported having headaches every day, even though they met migraine symptom criteria, were excluded.

Estimating methods

In a typical longitudinal study, the age-specific incidence rate, I_j , for the jth age interval is n_j /t_j , where n_j is the number of new-onset cases occurring in the jth age interval and t_j is the person-time under observation in that interval. We can estimate n_j and t_j if survey data are available on date of interview, date of birth and date of migraine onset (i.e. or equivalent data such as age of onset). We applied three different methods to AMPP data to estimate n_j and t_j. All three methods assume that there are no secular changes in the sex- and age-specific incidence of migraine, an assumption supported by serial prevalence studies conducted 10 years apart using exactly the same survey methods (11, 12).

The first method, labelled the ‘naive’ approach (9), uses all available data, assumes that active migraine cases identified at cross-section include all individuals who have ever had migraine (i.e. no remission). For this approach, n_j is the count of active migraine cases who reported onset in age interval j; t_j is the sum of person-time for the jth age interval contributed by those greater than or equal to age j at interview and who had not reported migraine onset before the beginning of the jth age interval. Individuals reporting migraine onset during the jth age interval contribute person-time for only part of the age interval.

The second method (9) uses only data from migraine cases who reported their onset to have occurred in the recent past. For the purposes of this analysis, recent onset was defined if an active case reported that their migraine first started within 3 years of the interview. In this approach, n_j is confined to counts of active migraine cases who reported onset in the past 36 months and who reported it occurring in the jth age interval; t_j is defined as it is for the naive method, except that person-time is confined to individuals whose age at interview was within 3 years of the upper bound of the jth age interval and who were migraine free at any time during this age interval. There are two advantages to this method. First, errors in reporting age of onset are likely to be minimal. Second, complete remission of migraine among recent-onset cases is unlikely. This method uses data from individuals represented by the diagonal in Table 1. We label this the ‘diagonal’ method (9). A disadvantage of this method is that the number of new-onset cases is limited and cannot be estimated for those ≤ 9 years old.

Table 1

Distribution of active migraine cases and those currently without migraine by age at interview and age at onset (migraine cases only) for 163 186 respondents

Age at interview	Men—age at onset
Age at interview	0–4	5–9	10–14	15–19	20–24	25–29	30–34	35–39	40–44	45–49	50–54	55–59	60–64	65–69	70–74	75–79	80+	n_j
10–14	20	75	79															3 555
15–19	24	62	144	72														6 093
20–24	10	37	89	100	24													4 654
25–29	6	18	68	95	89	38												4 001
30–34	11	18	93	105	97	105	49											5 007
35–39	9	32	57	94	92	78	91	40										5 765
40–44	5	32	62	81	87	75	77	61	29									6 599
45–49	14	25	55	82	74	76	73	56	76	33								6 915
50–54	7	22	49	60	68	63	50	24	61	45	33							7 138
55–59	7	16	33	42	40	47	34	22	39	24	25	16						6 788
60–64	5	8	13	24	20	11	41	5	22	14	11	16	8					5 491
65–69	0	7	12	8	5	8	21	9	13	6	11	3	8	2				4 552
70–74	1	2	6	6	5	2	9	6	5	2	5	3	7	7	1			4 026
75–79	1	0	2	0	3	2	3	5	1	1	2	1	1	3	2	2		3 453
80+	1	3	5	6	2	0	0	4	3	1	2	0	1	2	0	1	5	3 342
Total	121	357	767	775	606	505	448	232	249	126	89	39	25	14	3	3	5	77 379
t_j	386.6	385.4	377.4	347.0	317.3	295.1	272.8	246.3	216.9	184.8	151.4	118.0	88.8	64.3	43.4	24.7	14.7	3 535.1

Age at interview	Women—age at onset
Age at interview	0–4	5–9	10–14	15–19	20–24	25–29	30–34	35–39	40–44	45–49	50–54	55–59	60–64	65–69	70–74	75–79	80+	n_j
10–14	4	46	132															3 431
15–19	13	51	294	162														5 672
20–24	14	57	246	366	158													4 801
25–29	17	78	290	451	392	165												4 840
30–34	25	108	318	377	320	280	168											5 464
35–39	17	87	304	363	342	249	269	120										6 451
40–44	24	76	306	347	346	201	274	239	118									7 288
45–49	19	91	304	331	376	226	222	167	233	100								8 433
50–54	25	74	267	255	226	240	162	94	137	128	48							8 210
55–59	18	50	160	169	140	127	115	54	111	74	77	34						7 143
60–64	9	26	120	91	80	63	85	40	65	24	34	29	23					6 133
65–69	12	22	52	47	39	27	65	32	24	15	29	11	24	10				5 051
70–74	4	5	25	30	21	10	25	12	17	6	10	6	10	8	12			4 387
75–79	1	5	21	10	12	8	19	13	9	5	5	0	3	4	6	5		3 958
80+	4	3	10	13	9	6	8	13	18	6	5	2	1	2	2	3	6	4 545
Total	206	779	2849	3012	2461	1602	1412	784	732	358	208	82	61	24	20	8	6	85 807
t_j	428.6	426.2	412.5	372.4	334.5	306.9	281.0	255.0	225.6	194.6	159.9	128.1	99.3	73.8	51.9	31.6	21.6	3 803.4

t_j , person-years (in 1000s of years) during the age interval, taking into account the number of cases who reported onset of migraine during or before the age interval; n_j , total number of persons interviewed within each age.

The third method combines advantages of the naive (i.e. uses all data) and the diagonal (i.e. minimizes bias) methods. Like the naive approach, data from all individuals were used to improve precision of estimates. In contrast to the naive approach, regression methods were used to adjust for bias. Two sources of bias were of interest. Under-ascertainment occurs when a respondent forgets to report a history of severe headache or a cardinal migraine-defining symptom. In prevalence surveys of active migraine cases, under-ascertainment also occurs implicitly, as individuals with migraines that have remitted are not ascertained. Exclusion of inactive cases is likely to be the dominant source of bias and more common with increasing age. Temporal reporting bias, another error, occurs when case status is accurately reported (i.e. no ascertainment bias), but age of onset is over- or underestimated (16, 17). A more complete discussion of these errors has been reported elsewhere (9, 10). The third method assumes that errors are related to two temporal variables: age at interview and how long before the interview migraine first occurred. LAG was defined as the difference between the age at interview and the reported age of onset of migraine. It is important to recognize that LAG is only defined for active cases. LAG predicts variation in age-specific incidence. Variation occurs largely because inactive cases are not reported in the numerator and because the number of inactive cases naturally increases as LAG increases. We modelled data assuming that the number of active cases interviewed at age interval k with age of onset in interval j, n_jk , is approximated by a negative binomial with mean:

μ_{j k} = N_{i j} \times exp (α_{j} + {γX}_{j k}) j =, \dots, k,

where N_jk is the number of person-years at risk for that age-onset age group, X_jk is a vector of covariates and γ is the corresponding vector of parameters. If there were no covariates, then exp(α_j) would be the event rate at age j. The naive method essentially uses all of the data to estimate α_j. However, it yields a biased estimate, because remitted cases are not represented in the estimate. The diagonal approach uses only recent cases to estimate α_j. The third approach uses all of the data, but adjusts the incidence estimate for recall and ascertainment errors (9). For example, consider the simple model exp(α_j + γ₁LAG_jk). Think of exp(α_j) as the true incidence, and exp(γ₁LAG_jk) as the adjustment factor. As LAG increases (i.e. the years since onset), the incidence rate estimate would decrease (due to inactive cases), which implies that γ₁ < 0. The event rate for someone with LAG = 10 would be the true incidence rate multiplied by exp(γ₁ × 10). For example, if exp(γ₁ × 10) = 0.5, this would suggest that half of the cases remitted (or were inactive) within 10 years. One advantage of this method over the diagonal method is that all data are used to estimate α_j and γ. If the model is correctly specified, the incidence rate exp(α_j) will be less biased than the naive approach and more efficient than the diagonal approach. The model described above was just for illustration. More realistic models would include non-linear terms for AGE and LAG, as well as interactions between them. Interaction terms would be necessary if recall errors or inactive rates varied with age. The estimated true incidence curve is obtained by setting LAG to 0 (i.e. the expectation is there are no inactive cases) and using the predicted age-specific values from the model solution. Use of this method should minimize effects of underreporting of inactive cases and recall errors, while retaining all of the data and allowing us to extrapolate to ages not available with the diagonal method.

The negative binomial model, a generalized version of Poisson regression, was selected because of over-dispersion in the Poisson model (18). Cumulative incidence was estimated using the product-limit formula (19, 20) and median age at onset was defined as age at which cumulative incidence was 50%.

Separate estimates were derived for men and women, since the incidence of migraine is known to be different by sex. For this analysis, I_j was estimated for 5-year intervals. Data displayed in Table 1 for women and men, respectively, were used to derive age-specific incidence estimates for each of the three models. Model parameters were estimated using the GENMOD procedure in SAS version 8.1 (SAS Inc., Cary, NC, USA). In the model, we included age at interview, reported age of onset (ONST) and LAG as input variables. We considered higher ordered terms (i.e. to account for non-linear effects of incidence in relation to ONST and LAG) for each of these variables, as well as interactions between ONST and LAG. For regression modelling, ONST and LAG were centred. Model fit was assessed using a likelihood ratio test (LRT) to determine if higher order models significantly improved fit to the data compared with simpler models. A higher order model was deemed to improve fit to the data if the LRT statistic was equal to or greater than the critical value (i.e. P < 0.05) for the χ² statistic with degrees of freedom equal to the number of additional parameters in the higher order model. The optimal model for both men and women was based on the following parameters: ONST + ONST² + ONST³ + LAG + LAG² + ONST × LAG.

Results

Survey participants were between 12 and 100 years of age; 55% were female. A total of 4364 (6%) of the 77 379 men and 14 604 (17%) of the 85 807 women met criteria for migraine. Person-years of observation, t_j (i.e. must be multiplied by 1000), is displayed in the last row by age interval (Table 1).

Applying the naive method to data on men (Table 2, Fig. 1), migraine incidence peaks between 15 and 19 years of age at 2.2 cases per 1000 person-years. Among women, migraine incidence peaks during the same age interval at 8.09 cases per 1000 person-years. The naive median age of onset in men is 25.5 years and in women it is 23.2 years (Table 2, Fig. 2), and the cumulative incidence in men is 7.5% and in women 21% (Fig. 2).

Figure 1

(a) Male 5-year age-specific migraine incidence for the naive, diagonal and model-based estimates. (b) Female 5-year age-specific migraine incidence for the naive, diagonal and model-based estimates.

Figure 2

(a) Male 5-year age-specific cumulative incidence rates of current migraine headache comparing the model method and the naive method. (b) Female 5-year age-specific cumulative incidence of current migraine headache comparing the model method and the naive method.

Table 2

Sex and 5-year age-specific incidence rates (per 1000) of migraine headache for the naive, diagonal and model-based estimates

Age of onset	Men			Women
Age of onset	Naive	Diagonal	Model	Naive	Diagonal	Model
0–4	0.3	NA	0.8	0.5	NA	0.5
5–9	0.9	NA	2.7	1.8	NA	3.1
10–14	2.0	6.4	5.0	6.9	13.4	9.2
15–19	2.2	5.0	6.2	8.1	13.0	15.6
20–24	1.9	2.4	6.1	7.4	12.3	18.2
25–29	1.7	3.5	5.3	5.2	18.2	16.9
30–34	1.6	2.7	4.2	5.0	14.2	13.7
35–39	0.9	2.4	3.3	3.1	10.1	10.4
40–44	1.2	1.8	2.5	3.3	7.8	7.5
45–49	0.7	1.4	1.8	1.8	6.6	5.3
50–54	0.6	1.9	1.3	1.3	3.0	3.6
55–59	0.3	1.0	0.9	0.6	1.7	2.4
60–64	0.3	0.2	0.6	0.6	1.2	1.5
65–69	0.2	0.1	0.4	0.3	1.0	0.9
70–74	0.1	0.2	0.3	0.4	1.1	0.6
75–79	0.1	0.2	0.2	0.3	0.5	0.4
80+	0.3	0.0	0.2	0.3	0.0	0.3
Median age of onset	25.5	21.2	24.1	23.2	26.1	25.2

The diagonal method uses data on only 431 male cases compared with 4364 male cases used in the other two methods. Diagonal estimates (Table 2, Fig. 1) cannot be derived for ages ≤ 9 years and estimates below age 20 are unstable (i.e. small n). Nevertheless, diagonal estimates are consistently higher than the comparable naive estimate, particularly for women (i.e. age-specific estimates are two to three times higher than the comparable naive estimate). Cumulative incidence was not calculated because incidence could not be estimated at or below age 9 years.

Using the parameters described in the methods, estimates were derived using the negative binomial model. Once fitted, estimated incidence curves (for men and women) were derived by obtaining predicted values for each value of age from the model with LAG equal to ‘0’. The model-based median age of onset for men is 24.1 years and for women it is 25.2 years (Table 2, Fig. 2), and the cumulative incidence is 18% for men and 43% for women. Among men, > 75% of new-onset cases occur after age 14 years and > 85% of new-onset female cases occur after age 14 years. For both men and women, approximately 75% of new-onset cases occur before age 35 years. The model-based 5-year age-specific incidence rates displayed in Table 2 and in Fig. 1 are substantially higher than the comparable naive estimates. In contrast, the model-based estimates for women, in particular, are comparable to the corresponding diagonal estimate.

Including the 273 daily headache cases has a negligible effect on incidence estimates. For example, the peak incidence in women (i.e. the greatest difference) increased from 18.2 to 18.4 per 1000 person-years.

Discussion

The model-based estimates suggest that the cumulative lifetime risk of migraine is very high and that the median age of onset for migraine occurs after puberty. At first glance, the assumption of no remission for the naive estimate is invalid, as evidence clearly indicates that migraine does remit over time (2–4). The naive method therefore underestimates incidence and cumulative risk. In contrast, the diagonal method relies on self-reports from active cases of migraine reporting onset in the past 3 years. While age of onset is subject to reporting errors, the diagonal method approximates a longitudinal study with relatively complete representation of all cases whose onset occurred in the recent past. Cases would be excluded in the diagonal method if migraine first occurred in the past 3 years and remitted at least 1 year before the interview, or the respondent simply forgot to report relevant symptoms or the occurrence of severe headache. The striking contrast between the naive and diagonal age-specific estimates is likely to be largely explained by remission and indicates that active prevalent cases represent only a share of the total number of historical cases.

The model-based approach overcomes limitations (e.g. unstable estimates, lower age limit to estimates) to the diagonal method. Our estimate of cumulative incidence is consistent with prevalence estimates. Previous studies have consistently reported a peak male prevalence of 8% and a peak female prevalence of 25–27% occurring around age 38 years (1, 11). If migraine remits, as indicated by the rapidly declining prevalence after age 40, the peak prevalence indicates that the cumulative incidence of migraine must be > 8% for men and > 25% for women.

Our model-based estimates are largely consistent with a Danish longitudinal study, but substantially lower than estimates from two other studies (Fig. 3a,b) (2–4). In a Danish 12-year follow-up study of 453 migraine-free adults (mean age 42 years at baseline) (3) we derived age-specific estimates from Fig. 2 and accompanying tables. The 95% confidence intervals for seven of the eight estimates (see Fig. 3a,b) overlap with the model-based estimates. Over a 2-year period, Breslau et al. (2) estimated an annual incidence of 19.5/1000 person-years among a predominantly female (79%) cohort with a mean age of 42 years. This estimate, which is a weighted composite of the rate for those with and without depression, is more than twice (i.e. statistically significant) as high as the female model-based estimate. Annual incidence of migraine was estimated in an occupational cohort of men (n = 180) and women (n = 60) with a median age of 46 years at baseline and an average age of 51 years between the two surveys. Annual incidence estimates, based on two surveys 10 years apart (4), were approximately 8.8/1000 person-years for women and 3.8/1000 person-years for men. Both of these estimates are substantially higher than the corresponding model-based estimates. Although there are differences between several of the longitudinal estimates and the corresponding models, in total these estimates show a declining incidence after age 30 years, a pattern consistent with our model.

Figure 3

(a) Model-based male 5-year age-specific migraine incidence compared with estimates from longitudinal studies. (b) Model-based female 5-year age-specific migraine incidence compared with estimates from longitudinal studies.

We completed sensitivity analysis to determine how age-specific incidence estimates varied in relation to the upper age limit of the study population (Table 3, Fig. 4). The model-based estimates did not vary substantially. In contrast, the naive age-specific and cumulative incidence estimates declined as the upper age limit increased. When the naive method is applied to a younger age population (i.e. restricted to those < 30 years old), age-specific and cumulative incidence estimates (i.e. < 30) are similar to the model-based estimates (Table 3). In contrast, as the upper age limit is increased, the naive age-specific and cumulative incidence estimates are lower than the comparable model-based estimates. For example, when the upper age limit is 50 years, the cumulative incidence up to age 30 years for women is 22% for the naive method compared with 28% for the model-based method. Similarly, the cumulative incidence for all age groups is 21% for the naive method vs. 43% for the model-based method. The striking difference between the naive and model-based cumulative incidence estimates suggests that a majority of those who report migraine remit or experience a substantial change in symptoms associated with headache. This finding is consistent with longitudinal studies of adults (4) and adolescents (8). For example, in a workplace cohort, migraine changed to a non-migraine headache or remitted 10 years later in 63% of active migraine cases 35–50 years old at baseline (8).

Figure 4

Model-based female 5-year age-specific migraine incidence compared with naive estimates derived for the total population and when the upper age was truncated below 30 and 50 years.

Table 3

Age-specific (per 1000 person-years) and cumulative incidence (%) of migraine for women when naive and model-based estimates are applied to subpopulations truncated at an older age of 30, 40, and no limit

	Naive estimate age range used for sensitivity analysis						Model-based estimate age range used for sensitivity analysis
	12–30		12–50		12+		12–30		12–50		12+
	I_j	C_j	I_j	C_j	I_j	C_j	I_j	C_j	I_j	C_j	I_j	C_j
0–4	0.5	0%	0.6	0%	0.5	0%	1.0	1%	0.2	0%	0.4	0%
5–9	2.5	1%	2.6	2%	1.8	1%	3.1	2%	2.0	1%	3.1	2%
10–14	11.3	7%	10.1	6%	6.9	5%	10.7	7%	8.6	5%	9.2	6%
15–19	17.6	15%	13.2	13%	8.1	8%	14.6	14%	17.2	13%	15.6	14%
20–24	19.3	23%	13.2	18%	7.4	12%	16.0	21%	19.4	22%	18.2	21%
25–29	18.5	30%	9.2	22%	5.2	14%	18.5	28%	16.3	28%	16.9	28%
30–34			9.5	26%	5.0	16%			13.1	33%	13.7	33%
35–39			7.1	28%	3.1	17%			11.6	37%	10.4	36%
40–44			7.6	31%	3.2	19%			10.2	40%	7.5	39%
45–49			6.0	33%	1.8	20%			5.7	42%	5.3	40%
50–54					1.3	20%					3.6	41%
55–59					0.6	20%					2.4	42%
60–64					0.6	21%					1.5	42%
65–69					0.3	21%					0.9	43%
70–74					0.4	21%					0.6	43%
75–79					0.3	21%					0.4	43%
80–84					0.5	21%					0.3	43%

I_j , the age-specific incidence for the jth age category; C_j , the cumulative incidence up through the jth age category.

Our estimates of cumulative incidence (Fig. 2) suggest a very high liability for migraine in the general population. The median age of onset for migraine between 24 and 25 years for men and women would suggest that induction (onset) is largely attributable to factors that occur well past puberty.

Aetiological factors for migraine induction, persistence and remission are likely to be the product of genetic and environmental factors. Duration of illness reflects determinants of disease persistence and remission. Age of onset largely reflects determinants of disease induction. For example, the known relation of higher migraine prevalence with lower income could reflect the influence of environmental factors on the rate of induction or on disease persistence. In contrast, the finding that younger age of migraine onset in probands predicts familial aggregation for migraine (21, 22) suggests a possible genetic influence on early-life induction or persistence of migraine.

Acknowledgement

This study was sponsored by the National Headache Foundation through a grant from Ortho-McNeil Neurologics, Inc.

References

1 Stewart

Simon

Shechter

Lipton

. Population variation in migraine prevalence: a meta-analysis. J Clin Epidemiol 1995; 48:269–80.

2 Breslau

Lipton

Stewart

Schultz

Welch

. Comorbidity of migraine and depression: investigating potential etiology and prognosis. Neurology 2003; 22:1308–12.

3 Lyngberg

Rasmussen

Jorgensen

Jensen

. Incidence of primary headache: a Danish epidemiologic follow-up study. Am J Epidemiol 2005; 1:1066–73.

4 Nachit-Ouinekh

Dartigues

Chrysostome

Henry

Sourgen

El Hasnaoui

. Evolution of migraine after a 10-year follow-up. Headache 2005; 45:1280–7.

Camarda

Monastero

Santangelo

Raimondo

Puma

Pipia

. Migraine headaches in adolescents: a five-year follow-up study. Headache 2002; 42:1000–5.

6 Bille

. A 40-year follow-up of school children with migraine. Cephalalgia 1997; 17:488–91; discussion 487.

7 Bille

. Migraine in childhood and its prognosis. Cephalalgia 1981; 1:71–5.

Wang

Fuh

Juang

Hsu

Chen

. Evolution of migraine diagnoses in adolescents: a 3-year annual survey. Cephalalgia 2005; 25:333–8.

9 Stewart

Brookmyer

Van Natta

. Estimating age incidence from survey data with adjustments for recall errors. J Clin Epidemiol 1989; 42:869–75.

10.

10 Stewart

Linet

Celentano

Van Natta

Ziegler

. Age- and sex-specific incidence rates of migraine with and without visual aura. Am J Epidemiol 1991; 15:1111–20.

11.

11 Stewart

Lipton

Celentano

Reed

. Prevalence of migraine headache in the United States. Relation to age, income, race, and other sociodemographic factors. JAMA 1992; 1:64–9.

12.

12 Lipton

Scher

Kolodner

Liberman

Steiner

Stewart

. Migraine in the United States: epidemiology and patterns of health care use. Neurology 2002; 26:885–94.

13.

Lipton

Bigal

Diamond

Freitag

Reed

Stewart

. Migraine prevalence, disease burden, and the need for preventive therapy. Neurology 2007; 30:343–9.

14.

14 Headache Classification Subcommittee of the International Headache Society. The International Classification of Headache Disorders: 2nd edition. Cephalalgia 2004; 24 (Suppl. 1):9–160.

15.

15 Stewart

Lipton

Liberman

. Variation in migraine prevalence by race. Neurology 1996; 47:52–9.

16.

16 Wagenaar

. My memory: a study of autobiographical memory over six years. Cogn Psychol 1986; 18:225–52.

17.

17 Bradburn

Rips

Shevell

. Answering autobiographical questions: the impact of memory and inference on surveys. Science 1987; 10:157–61.

18.

18 Gardner

Mulvey

Shaw

. Regression analyses of counts and rates: Poisson, overdispersed Poisson, and negative binomial models. Psychol Bull 1995; 118:392–404.

19.

19 Kalbfleisch

Prentice

. The statistical analysis of failure time data. New York: Wiley 1980.

20.

20 Cox

Oakes

. Analysis of survival time data. New York: Chapman & Hall/CRC 1984.

21.

21 Stewart

Staffa

Lipton

Ottman

. Familial risk of migraine: a population-based study. Ann Neurol 1997; 41:166–72.

22.

22 Stewart

Bigal

Kolodner

Dowson

Liberman

Lipton

. Familial risk of migraine: variation by proband age at onset and headache severity. Neurology 2006; 14:344–8.

Cumulative Lifetime Migraine Incidence in Women and Men

Abstract

Keywords

Methods

Population survey

Migraine case definition

Estimating methods

Results

Discussion

Acknowledgement

References