The protocol was approved by the Local Ethics Committee for Biomedical Research of the University of Parakou (CLERB-UP) under number 0168/CLERB-UP/P/SP/SA of April 10, 2019. The study was conducted in accordance with the Declaration of Helsinki [14].

Necessary authorizations from academic and administrative authorities were obtained. All participants were informed of the nature and purpose of the study and gave oral consent before enrolment. Anonymity and confidentiality of the information collected were respected in accordance with data-protection laws.

This was a cross-sectional study of the adult general population in Benin, using cluster-sampling to select a representative sample. In unannounced door-to-door visits to households, trained interviewers randomly selected, and interviewed, one member of each aged 18–65 years.

Pre-pilot and pilot studies tested the questionnaire and methods prior to the main study.

The pre-pilot study was carried out at the Centre Hospitalier Universitaire et Départemental over a period of 1 month using the first draft of the questionnaire to test its acceptability and comprehensibility. Respondents were 80 adults (aged 18–65 years), of whom 40 were patients presenting with headache and 40 were accompanying persons not complaining of headache. The study provided the basis for finalisation of the questionnaire.

Interviewers used the Headache-Attributed Restriction, Disability, Social Handicap and Impaired Participation (HARDSHIP) structured questionnaire developed by LTB [15], in the Central African French version used in Cameroon [7]. This questionnaire, described in detail previously, consists of several modules, three of which were used here: the demographic and social status module (enquiring into gender, age, habitation [urban or rural], marital status, education level and household income), and the headache diagnostic and description modules [15]. All participants completed the first module and answered the screening questions of the second (“have you ever had headache?” and “have you had headache during the last year?”). Only those responding positively to the latter question continued with the headache modules, describing their most bothersome headache if they had more than one type. Two questions asked whether headache had occurred on the day prior to the interview (“headache yesterday” [HY]) and whether, if so, it was of this type.

To obtain a representative sample of the general population of Benin, we recruited from three geographical regions (departments): Borgou (Parakou [urban] and Pèrèrè [rural]), Atlantique (Torri-Bossito [rural]) and Littoral (Cotonou [urban]) (Fig. 1).

The 12 interviewers were physicians or senior medical or epidemiology students. In a two-day session at Unité de Recherche Clinique et Epidémiologique (URCE), Faculty of Medicine, University of Parakou, they received instruction in headache disorders (basic epidemiological and clinical aspects) and the study (design, purpose and practical aspects). Competence was ensured in supervised interviews.

These interviewers conducted both the pilot and the main studies.

The pilot study field-tested the final version of the questionnaire in communities in Porto-Novo, a city 40 km from Cotonou, and in the surrounding rural areas. Using a mix of convenience and purposive sampling, 160 non-biologically related adults (aged 18–65 years) were interviewed, 40 urban and 120 rural.

Data from the pilot study were not included in the main analysis, but provided an estimate (with confidence interval [CI]) of the expected refusal rate.

This was completed between May 11th and July 4th, 2020.

Map of administrative departments of Benin indicating those surveyed (hatched)

We used simple random sampling to select four towns in the three departments: Cotonou (urban) in Littoral; Tori-Bossito (rural) in Atlantique; and Parakou (urban) and Pèrèrè (rural) in Borgou (Fig. 1). In these we followed a three-stage sampling procedure. At the first level, we randomly selected 28 villages and/or town districts in rural areas and 12 town districts in urban areas. At the second level, we selected 30 dwellings per village or town district in a randomly chosen direction from a central starting point. At the third level, one individual per household was randomly selected.

First visits to dwellings were unannounced (“cold-calling”). At each, the interviewer identified the number of non-biologically related families living there (families connected by a first- or second-degree relative [parent, sibling, child, uncle, aunt, nephew, niece or first cousin] were considered as one). Each family was regarded as a household. The wife or head of household was asked to list all adult family members (aged 18–65 years) living there, from which one person (the participant) was selected by lottery. Only this person was eligible: if he or she was not then present, an appointment was made for interview. Refusals were counted, but not replaced from the household.

The sample size of N = 2,400 was estimated for an expected prevalence of 24.8%, a precision of 1.8% and an anticipated refusal rate of 10%. Methodological guidelines recommend a minimum sample size of N = 2,000 [13].

At interview, demographic enquiry was followed by headache screening questions (“Have you ever had headache?” and “Have you had headache in the last year?”), with the full interview covering headache characteristics (symptoms, including those contributing to diagnosis of headache type, and attributed burden [not reported here]) proceeding only when answers to both were positive.

As data were collected in each region, they were checked for quality by two supervising neurologists. Additionally, the principal investigator reviewed completed questionnaires as the study proceeded.

All data were entered twice into Excel at URCE under the supervision of a senior epidemiologist, and discrepancies resolved. The original questionnaires were retained securely in the unit.

Gender was recorded as a binary variable (male or female). Age was recorded as a continuous variable but later categorized: 18–25, 26–35, 36–45, 46–55 or 56–65 years. Habitation was recorded as urban or rural, marital status as single, married or widowed/separated/divorced, education level as none, primary school, secondary school or college+, and household income in West African francs (XOF) as < 10,000, 10,000–20,000, 20,001–50,000 or > 50,000. In June 2020, USD 1 = XOF 583.

Headache diagnoses were made algorithmically during analysis, according to responses to the HARDSHIP diagnostic module, which applied modified ICHD-3 criteria [16]. Only one diagnosis was allowed in each participant. The algorithm first identified those reporting headache on ≥ 15 days/month (H15+). Participants also reporting acute medication use on ≥ 15 days/month were categorized as having probable MOH (pMOH), on the assumption that only simple analgesics were available to the vast majority in a low-income country. The others were categorized as “other H15+”, with no further attempt at diagnosis. Participants reporting headache on < 15 days/month were categorized, as stipulated by ICHD-based criteria [16], in hierarchical order: definite migraine, definite TTH, probable migraine and probable TTH. Any remaining cases were unclassified.

Mean age, gender and habitation distributions in the sample were compared, using chi-squared tests, with those of Benin’s population aged 18–65 years.

We estimated 1-year prevalences as proportions (%) with 95% confidence intervals (CIs), reporting observed and age-, gender- and habitation-adjusted values for each headache type. We combined definite and probable migraine and definite and probable TTH in these and all further analyses. We estimated point prevalence of headache, as a proportion (%), from reported HY, and calculated the predicted point prevalence from the observed 1-year prevalence and mean headache frequency in days/month.

We used bivariate and multivariate analyses (calculating odds ratios [ORs] and adjusted ORs [aORs] with 95% CIs) to identify associations, if any, between each headache type and the demographic and social status variables. Significance was set at p < 0.05.

We used Microsoft Excel version 16 to calculate adjusted prevalences, and IBM-Statistical Package for Social Sciences statistical software (SPSS) version 28 (SPSS Inc., Chicago, IL) for all other analyses.