The study was approved by the ethics committee of the Faculty of Medicine and Dentistry at the University of Technical Sciences and Technologies, Bamako, under the number 2020/209/CE/FMOS/FAPH. It was conducted in accordance with the Declaration of Helsinki [13]. All participants gave verbal consent to inclusion.

Data were gathered anonymously, and managed in accordance with data protection legislation.

The study was a cross-sectional survey among adults in the general population of Mali, adopting the standardized sampling methodology of the Global Campaign against Headache [14]. Trained interviewers, visiting randomly selected households unannounced, employed a structured questionnaire during face-to-face interviews.

The study was conducted from January to October 2021.

Through multistage cluster sampling with random selection, we aimed to generate a representative sample of the adult general population (aged 18–65 years).

Firstly, we selected seven of the country’s eleven regions to reflect its ethnic and cultural diversities: (1) Kayes and (2) Koulikoro in the west, (3) Bamako in the southwest, (4) Sikasso in the south, (5) Mopti and (6) Tombouctou (Timbuktu) in the central region, and (7) Gao in the east.

Secondly, from these regions, we randomly selected health districts: from Bamako (Mali’s most populous city and the country’s Capital), four urban districts, two from each side of the Niger river, which divides Bamako into left and right banks; from Sikasso (the second most populous region), the urban district of Koutiala and rural district of Sélingué; from Mopti (the third most populous city), the urban district of Djenné and rural district of Badiangara; and from each of the less populated regions, a single district: Koniacary from Kayes and Maracacoungo from Koulikoro (both rural), the urban district of Tombouctou, and, for security reasons (rural areas being unsafe), the urban district of Gao.

Thirdly, from health-district official lists, interviewers randomly selected four villages or city areas in each district, then one or more blocks or circumscribed areas within each village or city area. Within each block, they systematically visited consecutive dwellings (omitting empty properties and commercial premises), unannounced in the first instance (“cold-calling”).

Fourthly, at each selected dwelling, the interviewer first identified the number of families living there (a family was defined as a group of people living together and sharing a kitchen). The head of each biologically unrelated family was asked to list all adult members (aged 18–65 years) living within that household. From this list, one person (the selected participant) was randomly selected for interview by the lottery method. Those refusing were counted, but not replaced from that household, in accordance with published guidelines [14]. If the selected participant was not present, another time was arranged for interview. If this and one further appointment were not kept, he or she was considered to be withholding consent and counted as a non-participant.

When the door to a selected dwelling was not answered at first visit, the dwelling was replaced by the next. This continued until the required number of participants was achieved in each selected village or city, and in the study overall.

We aimed for a minimum sample size of N = 2,000, again in accordance with guidelines [14].

The eight interviewers were physicians or final-year medical students trained for the purpose, selected for their knowledge of the principal local languages of Mali.

The study employed modules from the Global Campaign’s Headache-Attributed Restriction, Disability, Social Handicap, and Impaired Participation (HARDSHIP) questionnaire [15], translated into Central African French in the version previously used in Cameroon [2] and Benin [1]. The interviewers first gathered demographic information, then enquired into headache with neutral screening questions (“Have you ever had a headache?” and “Have you had a headache during the last year?“). Participants who answered positively to both were asked further questions enquiring into frequency of headache and of acute medication use, and into attributable burden using selected modules from HARDSHIP [15]. The last included impaired participation (lost time from paid and household work and from social or leisure activities, utilizing the Headache-Attributed Lost Time [HALT] questionnaire [16]), willingness to pay [WTP] for effective headache treatment, and quality of life (QoL) using the WHOQoL-8 questionnaire [17]. These enquiries were supplemented by questions about headache on the previous day (“headache yesterday” [HY]), its characteristics (duration and intensity) and its impact on activities. Participants reporting no headache in the preceding year were asked only about their QoL, to provide normative data.

There was no diagnostic enquiry beyond establishing frequency and, in those reporting H15+, identifying probable medication-overuse headache (pMOH: H15 + associated with acute medication use on ≥ 15 days/month).

All data were entered using an electronic platform for data capture (ONA, datafax or Redcap). At the end of each day the team coordinator assessed the day’s data for completeness, inconsistencies and wrong or missed entries. Following this review, the data were downloaded and kept secure at the University of Technical Sciences and Technologies, Bamako.

Gender was recorded as male or female. Age was recorded as a continuous variable, then categorized for further analyses as 18–25, 26–35, 36–45, 46–55 or 56–65 years. The distributions of these variables were compared with those of the national population aged 18–65 years.

Marital status was recorded as single, married, widowed, separated or divorced, the last three analysed as a single category. Educational level was recorded and analysed as none, primary school, secondary school or college+. Household income was recorded in West African francs (XOF) in four categories (< 10,000; 10,000–20,000; 20,001–50,000; >50,000).

Participants were classified as having no headache (no headache in the last year), episodic headache (reported frequency < 15 days/month), pMOH or other H15+.

Headache-attributed burden was analysed overall and for each of these three types. Symptom burden was estimated in participants with HY from the symptoms associated with HY. Headache intensity was reported on a 3-point scale (1 = mild, 2 = moderate, 3 = severe), with means calculated as though these were continuous data. Time spent in the ictal state (TIS) was calculated as a product of duration of HY (assuming this to be typical for the participant) and headache frequency (in days/month), and reported as a proportion of total time (pTIS). These estimates, adjusted for frequency, were extrapolated to the whole sample.

Headache-attributed impaired participation recalled by participants over the preceding 3 months (HALT-90) was analysed according to established procedure: “nothing achieved” and “less than half achieved” were counted as entire days lost; to counterbalance, “more than half achieved” was reckoned as no loss, along with “everything achieved” [16]. There were separate enquiries for income-generating work (“worktime”), household chores and leisure/social activities. For those with HY, impaired participation yesterday was analysed in similar manner by counting “less than half achieved” as “nothing achieved” and “more than half achieved” as “everything achieved”. WTP was recorded in XOF/month (at the time of the study, USD 1.00 ~ XOF 590). QoL scores (in the range 8–40) were derived by summation of responses to the eight items (each on a scale of 1–5), higher scores signifying better QoL.

An assessment of headache-care need was carried out using criteria for expectation of benefit from care: (1) having pMOH or other H15+; (2) having episodic headache and either or both of (a) pTIS > 3.3% and/or (b) losing ≥ 3 work and/or household days over the preceding 3 months.

In descriptive analyses, we used means with standard deviations (SDs) or standard errors of means (SEMs) and medians as appropriate.

We estimated 1-year prevalences of any headache, episodic headache, pMOH and other H15 + as percentages (%) with 95% confidence intervals (CIs). We adjusted observed values for age and gender according to their distributions in the national population aged 18–65 years [9]. Point prevalence of headache was calculated from reported HY, and predicted point prevalence from observed 1-year prevalence and mean reported headache frequency in days/month.

In the association analyses, demographic and social status variables were considered as independent variables and headache type as dependent. Unadjusted odds ratios (ORs) were calculated in bivariate analyses, and adjusted ORs (aORs) in multivariate analyses, each with 95% CIs. We evaluated associations between gender and symptom burden (headache frequency and duration of HY) and lost time using ANOVA. Intensity and impaired participation with HY were compared between genders using chi-squared tests. WTP and WHOQoL by headache type were analysed using ANOVA, and WHOQoL data were also displayed graphically.

Population-level estimates of pTIS, and of impaired participation in the preceding 3 months and yesterday, were derived by factoring in age- and gender-adjusted 1-year or 1-day prevalences as appropriate.

Significance was set at p < 0.05. We used Microsoft Excel to calculate adjusted prevalences and SPSS version 28 for all other analyses.