Cluster Analysis of Migraine-Associated Symptoms (CAMS) across the lifespan: a cross-sectional study

Abstract

Background/purpose

Headache-associated symptoms (e.g., sensory sensitivity, nausea) are used to classify primary headache disorders and define treatment outcomes. We developed Cluster Analysis of Migraine-Associated Symptoms (CAMS) to provide an aggregate measure that captures relationships between several headache-associated symptoms. CAMS reveals that a greater number of symptoms, particularly if they include symptoms of autonomic and vestibular dysfunction, are associated with older age and greater headache burden in children and adolescents. Here, we aim to determine if CAMS is generalizable to adults.

Methods

The Penn Online Evaluation of Migraine (POEM) survey collected headache characteristics on participants (18 years) who reported headache. CAMS models were generated based on the presence or absence of 12 headache-associated symptoms. Three models were compared: the original CAMS model developed for children and adolescents, a model generated with adult data, and a comparison pediatric model for data collected across the same timeframe as the adult data. Changes in CAMS as a function of age and headache burden were measured.

Results

CAMS generated from the adult (n = 350) and pediatric comparison (n = 471) datasets were consistent with the original CAMS model (n = 6,662). Across models, the first dimension (CAMS 1; 25.4–31.0% explained variance) represented the absence or abundance of symptoms; the second dimension (CAMS 2; 10.8–11.6% explained variance) separated the typical symptoms of migraine (photophobia, phonophobia, nausea/vomiting) and osmophobia, from vestibular symptoms (vertigo, imbalance, lightheadedness, tinnitus, diplopia). CAMS symptom number increased throughout childhood and adolescence, reaching a plateau in early adulthood. A greater number of symptoms and more autonomic and vestibular symptoms were associated with greater headache burden and allodynia.

Conclusion

CAMS is generalizable across the lifespan, providing a reliable outcome measure that aggregates information about many headache-associated symptoms. Greater symptom number that includes vestibular symptoms is associated with increased headache burden and allodynia.

Key findings

• Migraine-associated symptoms as captured by CAMS are largely consistent between child, adolescent, and adult populations, indicating this model is generalizable across the lifespan.

• Migraine-associated symptoms as captured by CAMS change as a function of age of migraine onset.

• CAMS aggregates information on multiple headache-associated symptoms that correlate with other markers of disease burden, including allodynia, which is a clinical surrogate for central sensitization.

Introduction

Symptoms associated with headache play a critical role in differentiating primary headache disorders. Photophobia and phonophobia, and nausea/vomiting are cardinal symptoms used to differentiate migraine and tension-type headache (TTH) in the International Classification of Headache Disorders (ICHD-3) [1]. It has long been recognized that many other symptoms are associated with migraine including lightheadedness, vertigo [2–6], and brain fog [2, 7]. This highlights that migraine is a heterogenous condition and the constellation of symptoms associated with headache can vary widely from one individual to another.

We developed Cluster Analysis of Migraine-Associated Symptoms (CAMS), a model based on multiple correspondence analysis (MCA) to aggregate information on associated symptoms [8–10]. CAMS provides a data-driven approach to capturing the relationships among headache-associated symptoms in children and adolescents. By including all youth with headache, it removes the boundaries imposed by current diagnostic criteria. The model consists of two numeric dimensions that capture over one-third of the variance in the presence or absence of headache-associated symptoms. The first CAMS dimension (CAMS 1) demonstrates that those who report any headache-related symptom are likely to report multiple symptoms. The second CAMS dimension (CAMS 2) reveals symptom clustering into stereotypical migraine symptoms (nausea, vomiting, photophobia, phonophobia) and symptoms associated with autonomic and vestibular dysfunction (lightheadedness, vertigo, imbalance, and tinnitus). Brain fog tended to cluster with autonomic and vestibular symptoms. The typical symptoms of migraine tend to present in younger children, while adolescents were more likely to report a greater number of associated symptoms, including autonomic and vestibular symptoms. This increase in symptom number and shift in symptom type was also associated with greater headache burden, including increased headache frequency and headache-related disability [8–10]. Differences across age and sex were also observed: adolescents and females tended to report more associated symptoms, particularly vestibular symptoms, than younger children and males.

In children and adolescents, CAMS captures a novel axis of headache burden that has great potential to be used as a quantifiable outcome measure in clinical trials. Specifically, CAMS aggregates information about headache-associated symptoms that provides a richer picture of associated-symptom burden that moves beyond “most bothersome symptom,” which has been considered for use in migraine clinical trials [11]. However, it is unknown if CAMS is generalizable across the lifespan, including adults. To address this open question, we applied CAMS to a cross-sectional sample of adults who filled out the Penn Online Evaluation of Migraine (POEM) survey, which has been validated for ICHD-3 migraine diagnosis [12]. We hypothesized that CAMS would generalize across the lifespan. Furthermore, we hypothesized that headache-associated symptoms would change with age and headache burden, with older age and greater headache burden being associated with a greater number of symptoms and more autonomic and vestibular symptoms.

Methods

This is a cross-sectional survey-based study.

Ethics and consent

The study was conducted in accordance with the Declaration of Helsinki. Adult data collection via an anonymized online survey was approved by the University of Pennsylvania IRB and deemed the study exempt from consent collection due to minimal risk. For the pediatric dataset, The Children’s Hospital of Philadelphia (CHOP) IRB approved extracting data from the electronic health record and entering it into a research registry, with a waiver of consent and assent to maximize generalizability.

Participants

Adult participants were recruited through a Facebook advertisement that circulated in the United States from February to April 2024. Adults who completed the questionnaire and reported headache were included; all were at least 18 years old based on self-report. Diagnosis was based on ICHD-3 migraine criteria applied to questionnaire answers [12]. Migraine, probable migraine, with and without aura, and non-migraine headache that occurred outside of illness were included.

We compared the CAMS model derived from adult participants’ data to CAMS models derived from two groups collected at CHOP. Participants were included if they were 6 to 17 years old, with a diagnosis of migraine, probable migraine, with and without aura, and non-migraine headache:

1. Participants who completed headache intake questionnaires between June 2017 and February 2022, which were used to produce the original CAMS model [8].

2. Participants who completed headache intake questionnaires over the same time window as the adult data, between February and April 2024, to provide a temporal comparison to the adult data and demonstrate the variability of the CAMS model within children and adolescents.

Patient survey data from children and adolescents were managed using REDCap electronic data capture tools hosted at CHOP [13, 14]. ICHD-3 diagnosis was made using a diagnostic algorithm applied to the patient survey data, as described in our prior work from this database [15].

Data collection

Participants completed the POEM survey built with the Qualtrics platform, that assessed the quality, duration, location, exacerbating factors, frequency, severity, and associated symptoms of headache needed to evaluate ICHD-3 diagnosis of migraine. The diagnostic algorithm is explained in greater detail in our prior work [12]. The presence or absence of 12 symptoms associated with headache were assessed for CAMS: photophobia, phonophobia, osmophobia, nausea/vomiting, lightheadedness, brain fog, blurry vision, vertigo, imbalance, neck pain, tinnitus, and diplopia. Data were only stored on completed questionnaires. Headache burden metrics were also collected. Headache severity was measured on a 0–10 Likert-type scale. Frequency of any headache and migraine headache was defined as the number of days with any headache in the following categories: “0 days (never in the last 3 months),” “1–3 days (once a month or less),” “4–11 days (between 1–4 days per month),” “12–24 days (1–2 days per week, 4–8 days per month),” “25–48 days (2–4 days per week, 8–16 days per month),” “49–72 days (4–6 days per week, 16–24 days per month),” “73–90 days (every day or nearly every day).” Frequency of bad headache was captured in the following categories: “0–4 days per month,” “5–9 days per month,” “10–15 days per month,” “16–20 days per month,” “>20 days per month.” Age at headache onset was defined as “less than 12 years (elementary school or before,” “12–18 years old (in middle or high school),” “18–30 years old,” “30–40 years old,” or “older than 40.”

For the pediatric dataset, participants filled out a headache questionnaire prior to or at their first visit in general neurology or headache clinic who presented with a chief complaint of headache. Data were collected on the quality, duration, location, exacerbating factors, frequency, severity, and associated symptoms of headache needed to evaluate ICHD-3 diagnosis of headache disorders, including migraine. The questionnaire and diagnostic algorithm has been explained in more detail in our prior work [15].

Validated measurements

All questionnaires and measurements have been previously used and validated. The POEM questionnaire has been validated to diagnose migraine in adults [12], and the CHOP headache questionnaire has been previously validated to diagnose headache disorders in children with migraine [15]. Headache-related disability as determined by MIDAS score, which is a validated questionnaire for measuring headache-related disability with a possible range of 0 to 270 [16]. Little to no disability was defined as 0–5 days, mild was 6–10 days, moderate was 11–20 days, and severe was > 20 days over a 3-month period. Allodynia was captured with the Allodynia Symptom Checklist (ASC-12), which has been previously published and validated [17]. A score of 0 indicated no allodynia symptoms, and a maximum score of 24 indicated substantial symptoms of allodynia. The CAMS model has been previously developed and validated [8–10].

Data analysis

Matlab (Mathworks, Natick, MA) was used to produce custom-written code for all analyses.

CAMS

We applied the CAMS modeling approach to each dataset to determine relationships between migraine-associated symptoms [8–10]. CAMS is based on multiple correspondence analysis (MCA) using an indicator matrix adapted from a publicly available Matlab function [18] that reduces the dimensionality of binary data (i.e., the presence or absence of migraine-associated symptoms) in Euclidean space [19]. Both the symptoms themselves and the symptoms of individual participants can be represented in this space. The closer two symptoms are to each other, the more likely they are to co-occur. The original CAMS model included youth with headache, regardless of headache diagnosis, to explore the relationship between associated symptoms and headache in a broader context, removing the bounds imposed by current diagnostic criteria. Because nausea and vomiting were combined in the adult questionnaire, they were also combined in the CHOP datasets and models were re-run to allow for direct comparison of CAMS (see Fig. 1). This did not substantively change the structure of the model. CAMS 1 represents the overall number of symptoms, with negative values indicating no or few symptoms and positive values indicating many symptoms. CAMS 2 distinguished between symptom types: those that are typical features of migraine (negative values) and those consistent with autonomic and vestibular dysfunction (positive values). While the multicenter validated CAMS model has been presented across 3 dimensions [9], we focused on the first two dimensions for analysis because they are well correlated with other migraine burden metrics. When CAMS is represented graphically, we do not include the numeric values as they have meaning only within the CAMS space. Instead, for symptom number (CAMS 1), we list the areas of the graph that are associated with few versus many symptoms, and for symptom type (CAMS 2), we label areas associated with typical migraine features (e.g., photophobia, phonophobia, nausea) versus vestibular symptoms (e.g., vertigo, double vision, tinnitus) to provide a more intuitive understanding of the model.

Fig. 1.

CAMS models. CAMS derived from (a) the original dataset, (b) the adult dataset, and (c) the CHOP chronological comparison dataset are shown. Bubbles represent the presence of symptoms, with bubble size indicating the proportion of respondents that reported that symptom. The closer symptoms are in CAMS space, the more likely they are to co-occur. The x-axis shows the absence of symptoms (left), a few symptoms (middle), and many symptoms (right). Symptoms that are in the middle of the graph (e.g. photophobia) may be reported with only one or two other symptoms, while symptoms on the right (e.g. diplopia) are only present with many other symptoms. The y-axis separates symptom type. Those on the bottom are typically associated with migraine including the cardinal symptoms of photophobia, phonophobia, nausea/vomiting, as well as osmophobia. Those on the top are associated with autonomic dysfunction (lightheadedness, blurry vision) and vestibular dysfunction (vertigo, imbalance, tinnitus, diplopia)

Given the similarity in the structure of the CAMS model across our three datasets, we applied the CAMS model developed from the original dataset (June 2017 – February 2022) to all participants (the chronological dataset from CHOP, and adult dataset from POEM) collected between February and April 2024. This was done because the original CAMS model was developed across a much larger dataset, but the results did not change substantively regardless of the model used. These scores were used to compare CAMS across the lifespan and determine the relationship between headache burden and CAMS in adults.

Statistical analysis

No a priori sample size calculations were performed. Sample size was based on available data as the goal was to explore this developed and validated model in the adult population. Furthermore, CAMS model has been confirmed in the pediatric population based on over 10,000 respondents, and has been consistent across different datasets [8–10], and shows the same general structure when hundreds of pediatric responses were used (see Fig. 1). Categorical variables were reported as percentages. CAMS scores as a function of age and headache onset were graphically represented as mean with 95% confidence intervals by bootstrap analysis. Age was treated as a continuous variable in statistical analysis. Original data points across age are shown in figures. Additionally, the mean with 95% confidence intervals were graphically represented binning the data into < 12 years old, 12–17 years, 18–29 years, 30–39 years, 40–64 years, and 65 years and older. As in our prior work, CAMS 2 analysis only included participants who reported at least 1 associated symptom as CAMS 2 scores associated with symptom type are only meaningful for those who report at least 1 associated symptom [8–10]. Chi-squared was used to compare differences between the pediatric and adult samples across the categorical variables, sex assigned at birth and headache diagnosis. Spearman’s Rho was used to determine correlation between CAMS and headache severity, frequency, and headache-related disability to account for non-normal distribution of headache characteristics. Small effect size was defined as ρ ≥ 0.1, medium effect size was defined as ρ ≥ 0.3, and large effect size was defined as ρ ≥ 0.5, consistent with our prior work [9]. To address the high proportion of females in the POEM dataset relative to the CHOP datasets, sensitivity analysis was run on females across age. For all statistical modelling of CAMS, significance was defined by p < 0.05. Hypothesis testing was two-tailed. We did not correct for multiple comparisons because we chose complementary outcomes and provide effect size data. Thus, we presented raw p-values to allow readers to interpret the full results. Missing data were < 5% in the CHOP dataset, and there were no missing data in the POEM dataset because incomplete questionnaires were not saved. This limitation is noted in the discussion.

Results

A total of 385 adult participants at least 18 years old (age range 18–86 years) completed the POEM questionnaire with broad distribution across the United States, but with highest concentrations coming from the Northeast. Thirty-five participants were excluded who did not experience headache outside of illness, leaving 350 participants who were included in CAMS. In the chronological CHOP sample, 471 respondents 6–17 years started the CHOP headache intake questionnaire between February and April 2024. All with a chief complaint of headache who filled out the headache intake questionnaire were included in the analysis. Participant characteristics of the adult POEM and CHOP samples are reported. Detailed demographics are reported (Table 1). The pediatric and adult groups differed in that there was a greater proportion of females in the adult group ( = 94.9, p < 0.001), and a greater proportion in the adult group met full migraine diagnosis ( = 98.9, p < 0.001).

Table 1.

Respondent demographics. *Gender identity information not available for the CHOP sample. – indicates the category was not asked. To ensure anonymity, any group where there were < 3 participants, are reported as < 3 participants

	CHOP (6–17 yrs)	POEM (18 yrs)
Number	471	350
Median Age [IQR]	13 [10, 15]	48 [33, 61]
Sex assigned at birth (%)
Female	303 (64.3)	324 (92.6)
Male	168 (35.7)	24 (6.9)
Prefer not to answer	--	< 3 (< 0.9)
Gender Identity (%)*
Cisgender Man	--	22 (6.3)
Cisgender Woman	--	309 (88.3)
Non-Binary	--	12 (3.4)
Transgender Man	--	4 (1.1)
Prefer not to answer	--	3 (0.9)
Transgender Woman	--	< 3 (< 0.9)
Race
American Indian/Alaska Native	< 3 (< 0.6)	< 3 (< 0.9)
Asian	20 (4.2)	11 (3.1)
Black	79 (16.8)	10 (2.9)
Middle Eastern/North African	--	< 3 (< 0.9)
More than One Race	--	14 (4.0)
Other/prefer not to answer	89 (18.9)	10 (2.9)
Pacific Islander	< 3 (< 0.6)	< 3 (< 0.9)
White	272 (57.7)	303 (86.6)
Ethnicity
Hispanic	50 (10.6)	14 (4.0)
Not Hispanic	411 (87.3)	336 (96.0)
Headache Phenotype (%)
Non-migraine	83 (17.6)	25 (7.1)
Probable migraine	138 (29.3)	38 (10.9)
Migraine	250 (53.1)	287 (82.0)
Missing data (%)	23 (4.9)	--

Comparison of CAMS models across age

We compared the original CAMS model constructed from CHOP headache registry data [8] to the CAMS model derived from adult POEM data and their CHOP comparator group (Fig. 1). CAMS scores represent two-dimensions that captures the most variability in the presence or absence of 12 headache-associated symptoms. Bubbles represent the presence of a certain associated symptom, and bubble size represents the proportion of participants who reported that symptom. In this space, symptoms that appear closer together are more likely to co-occur. For example, the typical symptoms of migraine are tightly clustered together indicating that when one of these symptoms are present, the others are also likely to be present. All three CAMS models showed a similar structure: CAMS 1 (x-axis), which accounted for the greatest amount of variability in the data (25.4–31.0%), captured symptom number, ranging from no symptoms (left) to all symptoms (right). When a symptom is further to the left, it indicates it is more likely to be the only associated symptom (e.g. photophobia), while symptoms further to the right are likely to be present with many other symptoms (e.g. diplopia). CAMS 2 (y-axis), which accounted for the second-largest amount of variability (10.8–11.6%), defined symptom type. The lower area of the plot contained the typical symptoms of migraine (photophobia, phonophobia, nausea/vomiting) and osmophobia. The upper area of the plot contained symptoms of autonomic and vestibular dysfunction (lightheadedness, blurry vision, vertigo, imbalance, tinnitus, diplopia). There were also differences between models. A greater proportion of adult respondents reported each symptom compared to child and adolescent respondents. Indeed, osmophobia, in addition to brain fog and neck pain that shifted into the typical migraine cluster in the adult model, was notably more common in the adult sample.

CAMS symptoms across the lifespan

Given the similarity across models, we fit adult and pediatric temporal comparison associated symptom responses to the original CAMS model, derived from a large (6,662-person) dataset [8]. We have previously found that the number of symptoms increased, and vestibular symptoms became more common in adolescents compared to younger children [8–10]. To determine how CAMS changes across the lifespan, we evaluated CAMS scores as a function of participants’ age at the time they completed the survey (Fig. 2a). There was an increase in the number of symptoms reported with increasing age (CAMS 1), reaching a plateau in early adulthood. There was less variation in symptom type: CAMS 2 scores were relatively flat across age, though a small peak in autonomic and vestibular symptoms was seen in the 3rd decade of life. We ran sensitivity analysis on females across age since sex differences in CAMS scores have been observed and the adult POEM dataset was over 90% female. This sensitivity analysis yielded a similar result.

Fig. 2.

CAMS score as a function of current age and age of headache onset. A CAMS symptom number (top) and symptom type (bottom) as a function of current age. CHOP respondents (light gray) and POEM respondents (dark gray) with mean and 95% CI are shown. For symptom type, those with no symptoms are shown in light gray and were not included in the mean and 95% CI calculations because CAMS symptom type is not meaningful for those with no symptoms. B Mean with 95% CI CAMS symptom number (top) and symptom type (middle) are shown as a function of headache onset for adult POEM data only. Number of participants in each category of age of headache onset is shown (numbers in parentheses). HA = headache, yrs = years, no. = number

To determine if the number and type of associated symptoms varied depending on when someone first developed headache, CAMS was also evaluated as a function of age of headache onset in the POEM dataset only (Fig. 2b). The pediatric dataset was excluded from this analysis because it would have skewed the data with many more individuals experiencing childhood or adolescent-onset headache compared to adulthood-onset. The number or type of symptoms reported in adulthood did not significantly change based on age of headache onset, though overall symptom number trended towards decreasing, and vestibular symptoms trended towards increasing with increased age of headache onset.

CAMS correlates with headache burden in adults

Using CAMS, we have previously shown that increased symptom number with more autonomic and vestibular symptoms is associated with increased headache burden in children and adolescents [8, 9]. Thus, we next aimed to determine if this was also seen in adults (Fig. 3). The following analysis included only the POEM dataset. Specifically, we evaluated how CAMS 1 (symptom number) and CAMS 2 (symptom type) scores changed as a function of frequency of any headache, frequency of migraine headache, and headache-related disability. Across all adults with headache (n = 350), the median frequency category for any headache was 1–2 days per week [IQR: 1 − 4 days per month, 4–6 days per week]. The median frequency category of migraine headache was 1–2 days per week [IQR 1–4 days per month, 2–4 days per week]. Median usual headache severity was 7/10 [IQR 5, 8]. Overall, headache-related disability was high: 10.1% had little to no disability, 9.3% mild, 12.5% moderate, and 68.1% severe, based on the MIDAS. A greater number of symptoms (CAMS 1; Fig. 3a) correlated with increased frequency of any headache (ρ = 0.38, p < 0.001) and bad headache (ρ = 0.33, p < 0.001), and increased headache-related disability (ρ = 0.40, p < 0.001). There was not a significant correlation between symptom number and headache severity (Rho = 0.10, p = 0.073). Symptom type (Fig. 3b; CAMS 2; n = 345) was not significantly correlated with any headache burden metrics (ρ 0.07, p 0.140).

Fig. 3.

Relationship between CAMS and headache burden in adults. a CAMS symptom number, and (b) CAMS symptom type are shown as a function of headache burden metrics. The mean with 95% confidence intervals of CAMS scores is shown for all respondents with headache (black dots) and sensitivity analysis of only those meeting criteria for migraine, including probable migraine (white dots) for frequency of any headache (left), bad headache (middle), and headache-related disability (right). HA = headache, no. = number, d = days, wk = week, mo. = month, mod. = moderate

We considered the possibility that CAMS would hold more meaning for those with a migraine diagnosis, as associated symptoms were chosen to represent those seen in migraine. Therefore, we performed a sensitivity analysis assessing only those meeting criteria for migraine or probable migraine (n = 325; Fig. 3, white circles). Consistent with the full dataset, a greater number of symptoms (CAMS 1) correlated with increased frequency of any headache (ρ = 0.30, p < 0.001) and bad headache days (ρ = 0.29, p < 0.001), and headache-related disability (ρ = 0.33, p < 0.001). However, unlike the larger dataset, more autonomic and vestibular symptoms (CAMS 2) correlated with increased frequency of any headache (ρ = 0.16, p = 0.003), migraine headache frequency (ρ = 0.14, p = 0.012), and headache-related disability (ρ = 0.21, p < 0.001). Neither CAMS 1 nor CAMS 2 correlated with headache severity (ρ 0.05, p 0.326).

Next, we assessed the relationship between CAMS 1 (symptom number) and CAMS 2 (symptom type) scores and the ASC-12 score, which is a validated score that measures the severity of allodynia, with larger scores indicating greater allodynia (Fig. 4). This analysis was performed on the adult POEM dataset only since the ASC-12 was not collected in the pediatric CHOP dataset. The median allodynia score was 4 [IQR 1, 8], but ranged from 0 to 24. The allodynia score demonstrated a positive correlation with symptom number (ρ = 0.55, p < 0.001; large effect size; Fig. 4a) and symptom type (ρ = 0.15, p = 0.007; small effect size; Fig. 4b). These relationships were the same when only those who met criteria for migraine or probable migraine were included in the analysis (CAMS 1 ρ = 0.53, p < 0.001, large effect size; CAMS 2, ρ = 0.22, p < 0.002, small effect size).

Fig. 4.

The relationship between CAMS and allodynia. a CAMS symptom number, and (b) CAMS symptom type shown as a function of Allodynia Symptom Checklist (ASC-12) scores. An ASC-12 score of 0 indicates no allodynia, a score of 24 indicates substantial symptoms of allodynia. Individual respondents (dark gray dot, dots are translucent to show overlapping datapoints) and mean with 95% confidence intervals are shown

Discussion

We demonstrate that CAMS captures relationships among headache symptoms that are consistent across child, adolescent, and adult populations. Indeed, we find that a greater number of associated symptoms, including those beyond typical migraine symptoms, are associated with more frequent and disabling headache in adults with migraine, consistent with our prior work in children and adolescents [8, 9]. Our work demonstrates that the CAMS model is generalizable across the lifespan and captures a novel axis of headache burden.

CAMS generally supports the classification system for primary headache disorders. CAMS 1, which captures the most variance in associated symptom behavior, distinguishes individuals with low headache burden who report no or only a few symptoms from those with high headache burden and many symptoms. This is consistent with defining TTH as a milder headache phenotype with no associated symptoms, and migraine as a more severe headache phenotype with abundant associated symptoms [1]. However, we previously observed a subset of youth meeting criteria for TTH, or probable migraine reported headache-associated symptoms not included in the cardinal symptoms that distinguish TTH from migraine, indicating that blurry lines separate current diagnostic categories [10]. This could be viewed as support for the “continuum hypothesis,” which posits that TTH and migraine lie on a continuum of headache severity [20, 21]. However, we cannot rule out the possibility that TTH is physiologically distinct from migraine [22]. It may be that considering a greater number of headache-associated symptoms is needed to separate TTH from migraine cleanly. CAMS has the potential to advance this separation by providing a means of quantifying and consolidating information on many headache-associated symptoms.

CAMS 2 separates symptoms typical of migraine from those that are associated with vestibular migraine, which is included in the appendix of the ICHD-3 [1]. Importantly, many symptoms beyond the cardinal features of migraine, as well as individuals who did not meet an ICHD-3 diagnosis of migraine, were included in generating the CAMS model. Thus, CAMS is not dependent on ICHD-3 criteria. Indeed, CAMS is agnostic to the ICHD-3, meaning it does not assume, for instance, that light sensitivity, sound sensitivity, and nausea/vomiting (the cardinal headache-associated symptoms needed to make a migraine diagnosis) must occur together. The finding that they do, indicates that current diagnostic criteria do a reasonably good job of distinguishing naturally occurring headache-associated symptom clusters in migraine.

The diversity of symptoms likely reflects the complexity of migraine pathophysiology, which involves alterations in multiple brain regions involved in sensory and pain processing within the cortex, subcortex, brain stem, and cerebellum [23, 24]. Central sensitization of the trigeminal nociceptive system is a key pathophysiologic step in migraine and chronic pain, and may help tie together the range of observed headache-associated symptoms [25]. As we have argued previously, these symptom clusters may reflect involvement of separate but overlapping neural networks [8, 9]. Sensory hypersensitivity (photo-, phono-, and osmophobia) tended to co-occur, consistent with other findings [26], that may result from enhanced responses in the corresponding sensory system [27]. Interestingly, while osmophobia is not included in the diagnostic criteria, it has high specificity for migraine which has led to the proposal that it should be included in the ICHD [28]. Nausea/vomiting frequently co-occurred with sensory hypersensitivities to complete the standard migraine cluster, which suggests involvement of the rostral dorsal medulla and periaqueductal gray area of the brain stem [29]. Symptoms of autonomic dysfunction (lightheadedness, blurry vision) cluster together, sharing many similarities with hemodynamic orthostatic dizziness [30], and may indicate involvement of the trigemino-autonomic reflex. It has been proposed that vertigo, imbalance, tinnitus, and diplopia localize to vestibular regions and the brainstem [1], though alternatively these symptoms could originate in the parieto-temporal cortex [31]. Interestingly, brain fog and neck pain were the two symptoms that did not consistently occur in the same cluster across all datasets used to generate CAMS. This may suggest that the brain fog and neck pain are intermediate symptoms that can be present across a range of situations. Notably, the frequency of neck pain and brain fog substantially increased with age (see Fig. 1).

It may be that individual genetic and environmental differences result in varying susceptibilities of different neural networks to central sensitization, with broader neural network disruption yielding more headache-associated symptoms and a more severe headache phenotype. Several observations from our current and past work developing CAMS support this proposal. In both children and adults, we find that headache burden increases with a greater total number of symptoms and more autonomic and vestibular symptoms [8, 9]. In this study, we also found that the CAMS 1 symptom number strongly correlated with the allodynia score. Allodynia is considered a clinical surrogate for central sensitization as it is associated with changes in the brainstem pain modulation in individuals with migraine [32]. Furthermore, as allodynia can predict treatment response in migraine [33–35], CAMS may also hold predictive power. This proposal is consistent with other studies, which have found that the presence of cranial autonomic symptoms is associated with greater central sensitization and increased headache frequency in people with migraine [36, 37]. This may similarly indicate that symptoms beyond those of typical migraine are indicative of central sensitization and migraine chronification. Therefore, CAMS may help measure or even predict treatment response in this condition. Indeed, we have demonstrated that CAMS symptom counts decrease with treatment-responsive youth, though we did not find that CAMS predicted treatment response in this child and adolescent cohort, where overall treatment response was high [38]. Currently, resolution or improvement in “the most bothersome symptom” has been used as an outcome measure for acute and preventive clinical trials in migraine research [11, 39]. CAMS provides a quantitative alternative that summarizes information about multiple associated symptoms in two numeric values.

We find age-related changes in the number of symptoms associated with headache. The number of headache-associated symptoms increases from childhood, hitting a plateau around the fourth decade of life. By comparison, symptom type was relatively stable across the entire lifespan, indicating that the types of symptoms experienced do not differ substantially with age. There was a small peak in vestibular symptoms in the 3rd decade of life, which is consistent with reports from child, adolescent, and adult studies that vestibular migraine tends to present in late adolescence/early adulthood [40, 41]. The increase in CAMS symptom number mirrors the increased incidence of chronic migraine when moving from childhood into the fourth decade of life.⁴⁵ These trends were not seen across age associated with headache onset, indicating that associated symptoms differ across developmental stage rather than a phenotypic difference in age when the headache first occurred. It is important to note that we had few participants who developed a headache after the age of 40 years, and there was a trend to report more vestibular symptoms without reporting a greater number of symptoms. This suggests that headache developing after age 40 years is not generally consistent with typical migraine, but a larger sample size in this age group is needed to clarify this trend.

Limitations that should be considered in the interpretation of the results include possible sources of bias. First, there is selection bias in the adult sample, limited to those who use Facebook and agreed to take the survey. The adult POEM dataset was over 90% female, limiting the generalizability to adult males, which is a challenge facing much of migraine research. The percentage of participants who reported symptoms consistent with migraine was substantially higher than the general population, indicating those with migraine were more likely to take the survey. Furthermore, we do not have data on individuals who started the form and did not complete it. Thus, those who finished the questionnaire may have further biased our sample population, but we are not able to quantify this. In comparing the adult and pediatric samples, there were more males and higher rates of probable migraine in our child and adolescent dataset. This may be due to children and adolescents not meeting criteria because they did not meet associated symptom criterion [10], but we cannot rule out the possibility of a difference in sampling bias across the datasets. It is important to point out that CHOP and POEM datasets had different sources of selection bias, because CHOP recruitment was in a clinic-based setting, while POEM was based on self-identification through an online advertisement. A strength of CAMS is we still observe consistent patterns in CAMS despite different sources of bias across the populations studied. The study was cross-sectional; thus we cannot determine how CAMS changes over time across the lifespan. The study design was also based on participant report opening the possibility of recall bias. The pediatric data likely represented both patient and caregiver symptom reporting. The POEM survey was designed to identify headache phenotype, but we cannot assess additional causes or contributors to headache (e.g. post-traumatic headache, new daily persistent headache), though we have found that features of these types of headaches also exist on the CAMS continuum from studies in children and adolescents [8, 9]. Additionally, our dataset does not allow us to determine how CAMS operates in more rare primary headache disorders like the trigeminal autonomic cephalalgias. Finally, there are multiple symptoms associated with migraine (e.g., the cranial autonomic symptoms) that were not included in CAMS, that may be relevant. This method can be used to include additional migraine-like symptoms in future iterations. Finally, as we chose not to correct for multiple comparisons opting to report effect sizes and choosing complementary outcomes, this does increase the risk of Type I error. Despite these limitations, CAMS demonstrated reproducibility and generalizability.

There are multiple promising directions to take CAMS for future research. The neural substrate for CAMS remains unclear. Studies that explore brain regions involved in these symptom constellations will help further understand the pathophysiology of primary headache disorders. This may also pave the way to use CAMS to predict response to specific therapies, which provides an avenue of developing precision headache care. Another open question is if phenotypic differences identified by CAMS represent varying levels of genetic risk. Better characterization of headache-associated symptoms may help us understand and identify different genetic clusters within migraine as migraine genetics studies have thus far shown great heterogeneity. One study has already found success with this type of approach, identifying a gene locus specifically associated with sound sensitivity trait [42]. If a relationship were found between CAMS and polygenic risk score for migraine [43], this may create a new way forward to personalize headache care.

Conclusions

CAMS aggregates information about relationships between headache-associated symptoms. The increase in number of headache-associated symptoms (CAMS 1), and clustering of typical migraine symptoms and vestibular migraine (CAMS 2) demonstrate a similar pattern in children and adults, indicating that this model is applicable across the lifespan. CAMS appears to represent a novel axis of migraine chronification with a greater number of symptoms, and the presence of symptoms outside of the cardinal features of migraine corresponding to an increase in other markers of headache burden. Thus, CAMS is a promising candidate to be used as an outcome measure in clinical trials. Specifically, it could be used to quantify changes in the number and type of headache-associated symptoms that provide a richer description than “the most bothersome symptom.” [11, 44] Further work should focus on identifying the neural substrates for CAMS and determining if it can be used to target therapy.

Abbreviations

CAMS

Cluster Analysis of Migraine-associated Symptoms

CHOP

Children’s Hospital of Philadelphia

CI

Confidence Intervals

ICHD

International Classification of Headache Disorders

MCA

Multiple Correspondence Analysis

POEM

Penn Online Evaluation of Migraine

Authors’ contributions

C.P.G., G.K.A., and E.K. conceptualized the study, designed data collection, and collected the data. C.P.G. performed the analysis, visualized the data, and wrote the manuscript. G.K.A., E.K., C.L.S., and A.D.H. provided scientific feedback and editing to the main manuscript. All authors approved the final manuscript.

Funding

This work was supported in part by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health K23NS124986 (to C.P.G).

Data availability

Data collected are not publicly available due to privacy concerns but are available from the corresponding author on reasonable request pending IRB approval and execution of a Data Use Agreement. Code used to analyze these data is publicly available on github.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki. Adult data collection via an anonymized online survey was approved by the University of Pennsylvania IRB and deemed the study exempt from consent collection due to minimal risk. For the pediatric dataset, The Children’s Hospital of Philadelphia (CHOP) IRB approved extracting data from the electronic health record and entering it into a research registry, with a waiver of consent and assent to maximize generalizability.

Consent for publication

The requirement for informed consent was waived by The University of Pennsylvania and CHOP IRBs for adult and pediatric datasets, respectively due to the minimal risk nature of this study and the use of de-identified participant data to ensure no adverse effects on participant rights.

Competing interests

C. Patterson Gentile: Dr. Patterson Gentile is currently or was recently funded by the National Institutes of Health/National Institute of Neurological Disorders and Stroke and the CHOP Foerderer Institutional grant. E. Kaiser: Dr. Kaiser was recently funded by the National Institutes of Health/ National Institute of Neurological Disorders and Stroke. C.L. Szperka.: Dr. Szperka is currently receiving research/grant support from PCORI. Dr. Szperka or her institution have received compensation for her consulting work for Eli Lilly; Teva Pharmaceutical Industries Ltd; Upsher-Smith Laboratories, LLC; and Abbvie. A.D. Hershey: Dr. Hershey or his institution have received compensation for serving as a consultant for AbbVie, Amgen, Biohaven, Eli Lilly, Lundbeck, Supernus, Teva, Theranica and Upsher-Smith. His institution has also received research support from Amgen, Biohaven, Eli Lilly, Theranica, Upsher-Smith, and the NIH NINDS/NICHDS. G.K. Aguirre: Dr. Aguirre receives funding/grant support from the National Institute of Neurological Disorders and Stroke, the National Eye Institute, and the Binational Science Foundation.