Epidemiology and determinants of chronic migraine: A real-world cohort study,with nested case-control analysis,in primary care in Italy

Data sources

We used the Health Search Patients Database (HSD), an Italian general practice database that includes patients’ records of a group of over 1000 GPs homogenously distributed across Italy. GPs voluntarily agreed to attend training courses for data entry on patients’ information.

Patients’ demographic details are linked with an encrypted code to clinical records (diagnoses, referrals, tests prescriptions and results), drug prescriptions (drug name, date of filled prescription, and number of days’ supply), lifestyle-related records (i.e. Body Mass Index (BMI), smoking and alcohol), hospital admissions, and date of death. In the HSD, medical diagnoses and prescribed medications are coded using the 9th version of the International Classification of Disease, Clinical Modification (ICD-9-CM) and the Anatomical Therapeutic Chemical (ATC) classification system, respectively. GPs are instructed to couple diagnoses with every prescription (e.g. indication of drug use).

To be considered for participation in epidemiological studies, GPs are required to meet up-to-standard quality criteria concerning the levels of coding, prevalence of well-known diseases, mortality rates, and years of recording (17). When this study was initiated, 700 GPs homogeneously distributed across all areas of Italy, and covering a patient population of 1,163,855 individuals, met the standard quality criteria and were included in the study. The HSD has been used in a number of epidemiological studies (18–21).

With regard to the classification and implementation of observational drug-related research, as issued by the Italian National Drug Agency (an entity belonging to the Italian Ministry of Health), the present study does not require approval by an Ethics Committee in Italy (Italian Drug Agency note dated 3 August 2007). No patients or carers were directly involved in the conducting of this study.

Study population

We selected a cohort of patients active in the HSD aged ≥18 years during the period between 1 January 2002 and December 2016. The date of the first GP’s visit within the eligibility period was the study entry date. From this population, we excluded patients with less than 1 year’s medical history prior to the entry date. These patients were followed up until the occurrence of these events, whichever comes first: diagnosis of CM (see paragraph on “Event definition”), death, end of data registration in the GP’s database, or the end of the study period (31 December 2016). To calculate the prevalence rate, patients with less than 3 months preceding the index date did not enter the cohort. In order to calculate the incidence rate, patients with less than 8 years of clinical history preceding the index date did not enter in the cohort.

Event definition

According to the International Classification of Headache Disorders, CM is specifically defined by the presence of migraine attacks for at least 15 days per month for 3 consecutive months (3). Therefore, we hierarchically elaborated two groups of case definition algorithms for CM including three and two different algorithms, respectively. The first group was based on at least one migraine-related contact coded via ICD-9-CM being registered by month for 3 consecutive months (Algorithm 1A).

Thus, the others case definition algorithms were based on Algorithm 1A or when there were two contacts for 2 consecutive months and/or medications used for migraine attack (Algorithm 1B) and/or prophylaxis (Algorithm 1C) covering 15 days of therapy for 3 consecutive months. For the evaluation of duration of migraine attacks, we selected patients identified by two migraine-related contacts for two consecutive months (coded via ICD-9-CM) who were prescribed with medications used for migraine attack and/or prophylaxis. Using these criteria, we were able to identify patients with migraine episodes based on the duration of each prescription. To do so, for each medication class we adopted the prescribed daily dose (PDD). As such, the overall prescribed amount (e.g. in mg) was divided by the related PDD to obtain the duration for each prescription. The second group of case definition algorithms was based on the same aforementioned definition but keeping the three migraine related contacts for 3 consecutive months coupled with medications use for migraine attack (Algorithm 2A) and/or prophylaxis (Algorithm 2B) (Supplemental Figure 1). This hierarchical approach attempted to define CM diagnoses with an acceptable level of “completeness” to allow the identification of the highest possible number of events and “correctness” to reduce misclassification of cases.

Operationally, migraine-related contacts were identified using the following ICD-9-CM codes: 346.0*, 346.1*, 346.2*, 307.81, 784.0, all coupled with the term “migraine” as commented by GP in the diagnosis descriptor. We excluded the codes: 346.00, 346.10, 346.20, 346.80, 346.2 coupled with “abdominal”, “ciliary” or “neuralgic migraine” in the coding field. Furthermore, among patients with the aforementioned ICD-9-CM diagnosis, we identified prescriptions of medication for acute migraine attacks, including: i) triptans (ATC: N02CA*); ii) ergot derivatives (ATC: N02CC*); iii) NSAIDs (ATC: M01A*); iv) paracetamol (ATC: N02BE01, N02BE51, N02BE71); v) paracetamol-codeine combinations (ATC: N02AA59 besides the previous codes coupled with “codeine” in drug description). Then, we considered prescriptions for migraine prophylaxis, including: i) antidepressants (tricyclic (ATC: N06AA*) and SSRI (ATC: N06AB*)); ii) antiepileptic drugs (ATC: N03A*); iii) beta blockers (ATC: C07*); iv) dopaminergic drugs (ATC: N05A*); v) calcium antagonists (ATC: C08*); vi) pizotiphen (ATC: N02CX01); vi) antihistamines (ATC: N07CA*, R06A*).

Nested case-control analysis

The date of CM definition was the study index date, which was defined as the 31st day of the third month of the related case definition algorithm. For each case, 20 controls were randomly selected from the risk set of the respective case using the incidence density sampling method (i.e. at the time of each case diagnosis) (22). Controls (i.e. person-times) were therefore matched with cases by month and year of cohort entry, age (±5 years), and duration of follow-up. The index date of the controls was the date of the respective cases.

We used a nested case-control analysis because of the time-varying nature of exposure, the size of the cohort, and the duration of follow-up. When compared with a time-dependent survival analysis, a nested case-control analysis is computationally more efficient, while producing odds ratios that are unbiased estimators of incidence rate ratios (RR), with little or no loss in precision (22). Thus, while a cohort study using a survival analysis uses the full risk set (i.e. all patients still at risk of the event at the time of the case’s event), the nested case-control analysis uses a random sample of patients (i.e. controls) from that same risk set. In other words, the nested case-control analysis should provide nearly identical results as that of the cohort approach with the advantage of being computationally more efficient and easier to implement.

Candidate determinants

We considered all potential determinants being registered in the HSD which, along with sex and age, were known to affect the occurrence of incident CM (5–7,23). The following variables were measured ever before or on the entry date, unless otherwise specified. Namely, obesity (ICD-9-CM: 278.0* or BMI ≥30 kg/m²), alcohol abuse and/or alcohol-related disease, smoking (current, former, non-smokers; last available measurement); depression (in the overall period preceding or on the entry date); NSAID overuse (30 days of cumulative use during 90 consecutive days) was defined in the period preceding the index date by adding a latency of 90 days (i.e. this period was considered as free of use for NSAIDs) to minimize the reverse causation (3,24,25).

Statistical analysis

We reported descriptive statistics for continuous (mean (standard deviation-SD)) and categorical values (N and percentage (%))

Prevalence and incidence rate of CM were calculated for each case definition algorithm defining the event. To calculate the prevalence rates, we identified both “old” and “new” cases of CM being captured in 2016 as the numerator. The number of patients being active in the database and with at least 3 months of follow-up being registered at 31 December 2016 was the denominator. Given that up to 26% of cases of CM go into remission within 2 years (26), this calculation considered as prevalent cases of CM (i.e. increasing the numerator) those identified up to 8 years earlier (up to 2008 included) as well, using the same selection criteria adopted for 2016. The incidence rate was calculated in the study cohort, dividing the number of new cases identified in 2016, by excluding those with the same diagnosis in the previous 8 years (numerator) by the total number of person-years cumulated during follow-up (denominator). All patients with at least 3 months of follow-up therefore contributed with their individual person-time.

A conditional logistic regression was used to calculated odds ratios (ORs) and related 95% confidence intervals (CI) for each potential determinant of CM as defined by every case definition algorithm. In particular, the logistic regression was conditioned on calendar period (i.e. month and year of cohort entry), age (i.e. ±5 years) and duration of follow-up (i.e. matching criteria). Those determinants were statistically significantly associated with CM and were linearly combined to retrieve a single estimate with related 95% CI.

We also conducted a sensitivity analysis to test the potential misclassification of old cases of CM being counted in 2016 as the numerator. In essence, patients diagnosed with CM before 2016 might be healed in 2016 with a rate of almost 25% every two years (27). The prevalence rate of CM was therefore re-calculated by limiting the cases of CM to those with old CM (from 2009 to 2015) and at least one current (in 2016) contact of migraine coded via ICD-9-CM.

We used STATA software, version 11 (Stata Statistical Software: Release 11. College Station, TX: StataCorp 2009 LP.).