Study survey and population
The AMPP study is a two-phase, longitudinal, population-based survey of headache epidemiology. Sampling methods and design have been described elsewhere . Briefly, in phase 1, a self-administered headache screening survey was mailed in June 2004 to a stratified random sample of 120,000 US households. The 2004 sample yielded data for 162,756 household members aged 12 and older with gender-equivalent response rates. A random sample of 24,000 adults (age 18 and older) was selected from the 30,721 respondents reporting at least one severe headache in the past year in the 2004 screener. This random sample was enrolled in phase 2: an ongoing longitudinal follow-up study. To be included in the current study, respondents had to meet criteria for EM in 2005 and return valid questionnaires in 2005, 2006, and 2007. The study was approved by the Albert Einstein College of Medicine, Institutional Review Board.
Headache diagnosis and case definitions
The headache module of the AMPP study includes items that assess migraine features according to ICHD-2 criteria . The module has a sensitivity of 100 % and specificity of 82 % for migraine diagnosis , and sensitivity of 93 % and specificity of 85 % for CM diagnosis . EM was defined as headache occurring <15 days/month on average over the preceding 3 months and fulfilling ICHD-2 criteria for migraine. CM was defined as having headache on ≥15 days/month averaged over the preceding 3 months and meeting ICHD-2 criteria for migraine. The criteria used are a variation from the ICHD-2R definition of CM . The new onset of CM (i.e., transformation) was defined as: (1) having EM in 2005 and developing CM in 2006 and/or (2) having EM in 2006 and developing CM in 2007. The reference groups for the two transformation events were those who did not develop CM within the time interval of interest and who never met criteria for CM in the years 2005–2007.
Assessment of independent variables
Depression was measured in two ways: using a validated questionnaire (the Patient Health Questionnaire-depression module (PHQ-9) ) and based on self-reported physician diagnosis (SRPD-Depression). The PHQ-9 provides a validated measure of current Major Depressive Disorder based on DSM-IV criteria . The PHQ-9 assesses symptoms and functional impairment over the preceding 2 weeks and contains nine items with four frequency response options (scored as 0, 1, 2, and 3). A sum score is used to categorize participants into four depression categories: none/minimal (0–4), mild (5–9), moderate (10–14), moderately severe (15–19), and severe (20–27). A cut score of 15 was used to define the dichotomous depression variable. In models assessing the dose response for depression, we combined none/minimal and mild into the reference group against which moderate, moderately severe, and severe were separately contrasted. The SRPD-depression item assessed rates of ever being diagnosed with major depression by a physician.
A variety of covariate adjustments were included to assess the robustness of the depression effect and unique contribution of the adjusting effects. Covariates included were selected based on theoretical relevance and evidence in the literature linking them to the emergence of CM. Variables employed for adjustment included sociodemographic features, headache features (including attack frequency, pain intensity, symptom severity and allodynia), comorbid health conditions such as anxiety as well as medication use.
Sociodemographic data such as age, gender, weight, insurance, and income were obtained via self-report. Body mass index (BMI) was calculated using the standard formula based on self-reported height and weight.
Headache features included a measure of average pain intensity on a scale of 0–10 with 0 indicating no pain and 10 indicating the most severe pain possible . Average pain responses were dichotomized at scores ≥4, defining moderate to severe pain. In addition, monthly headache frequency in years preceding transformation to CM was assessed by self-report. Monthly headache frequency estimates were obtained by averaging self-reported 3 month frequency values.
Migraine symptom severity was obtained from the sum of the seven ICHD-2 migraine-defining features for migraine without aura (unilateral pain, pulsatile pain quality, pain intensity and pain increased by routine physical activity as well as nausea, photophobia, and phonophobia) plus an item assessing visual aura. The primary symptom items are coded to have the following response options: never/rarely (0), less than half the time (1), and half the time or more (2). The visual aura item was coded as no (0) or yes (2). The sum of these items produces the migraine symptom severity score, with values ranging from 0 to 16.
Cutaneous allodynia was assessed with the 12-item Allodynia Symptom Checklist, which includes questions about the frequency of various allodynia symptoms associated with headaches . Total scores range from 0 to 24 with scores ≥3 defining the presence of allodynia.
Anxiety (SRPD-anxiety), a health condition comorbid with depression, was addressed using self-reported medical diagnosis, employing the same open recall period as that of depression. Weekly alcohol consumption and smoking behavior, two additional health conditions comorbid with depression were assessed by self-report. Because alcohol and smoking measures were only available in the 2006 AMPP battery, they were not included in longitudinal analyses.
Two classes of medication use, use of anti-depressants and medication overuse, were also examined because each may contribute to confounding of the link between depression and transformation to CM. Our measure of anti-depressant use was based on self-reported current use of any anti-depressant compounds (including duloxetine, venlafaxine, amitriptyline, nortriptyline, paroxetine, fluoxetine, sertraline). Medication overuse was assessed using the acute medication module, which evaluated self-reported use of simple analgesic compounds, triptans, ergotamines, opioids, and other compounds. Respondents met criteria for our medication overuse definition if they used triptans, opioids, or ergotamines 10 or more days per month, or used simple analgesics 15 or more days per month.
Analyses were performed using SAS Version 9.2 (SAS Institute Inc., Cary, NC, USA). We used a nominal alpha level of 0.05 for statistical testing. Analysis proceeded in two stages. In the first stage, descriptive analyses were conducted characterizing unadjusted differences between those who developed CM and those who did not from 2005 to 2006 and then for the 2006–2007 interval. Contrasts for normally distributed variables (e.g., age, BMI) were based on t tests. For binary variables (e.g., gender, and allodynia), contrasts were based on Chi-square tests of the logistic regression odds ratio (OR). For ordered categorical variables (e.g., income and PHQ-9 categories), contrasts were based on the Chi-square test associated with the cumulative logistic OR. Because headache frequency and weekly alcohol use were count variables, inference for the mean comparisons was based on Negative Binomial models.
In the second stage of analysis, two-stage logistic transition models were used to model the new onset of CM in 2006 and 2007 as a function of covariates in the preceding year. The GLIMMIX Procedure for generalized linear mixed models was used for estimation. We parameterized the model using a binomial response distribution, logit inverse link function, and constant subject-specific random effect (detailed information on the model, estimation procedure, and code are available upon request). CM development could be reported in 2006 or in 2007. As depression is a risk factor whose status changes over time, we used a lagged predictor approach. Specifically, odds of developing CM in any subsequent year were modeled from predictor values in the preceding year. Sociodemographics, such as gender and income, were fixed within subject, thus lags were not employed. Other effects, including, but not limited to, age, BMI, headache frequency, allodynia, and depression status were lagged.
We examined a series of four models, with the depression effect included in each. Model 1 focused on adjustment for sociodemographic variables including age, gender, income, insurance, and BMI (both a linear and quadratic trend). Model 2 was fully adjusted by the addition of the lagged cutaneous allodynia effect and lagged effects of SRPD-anxiety, average headache pain intensity, headache frequency, and migraine symptom severity, use of anti-depressant medication and medication overuse.
To elucidate the relationship between depression and CM onset, two additional models were fit. Model 3 examined a dose–response effect of depression across categories (moderate, moderately severe, and severe) relative to participants with none/mild levels of depressive symptoms adjusting for sociodemographics, cutaneous allodynia, SRPD-anxiety, and migraine pain intensity. Model 4 included an additional adjustment for monthly headache day frequency and SRPD-depression in the years preceding transformation to CM.
Because analyses were restricted to subjects contributing in each year from 2005 to 2007, missing data were not caused by attrition, but by nonresponse to specific predictors and covariates. PROC GLIMMIX does not permit missing values on predictors, therefore, we implicitly assume that covariate nonresponse was missing completely at random . Predictor-specific missing values were consistent across all adjusted models and therefore a common sample size for events and trials exists for models contained in Tables 2 and 3 as well as all web tables.