Executive Summary

Drawn entirely from the chapters that follow; every figure below is sourced and tagged in the body text.

Cluster headache (CH) is defined by ICHD-3 as attacks of severe, strictly unilateral orbital, supraorbital and/or temporal pain lasting 15–180 minutes untreated, occurring between once every other day and eight times a day, accompanied by ipsilateral autonomic signs (tearing, nasal congestion, drooping eyelid) and/or restlessness. It comes in two forms: episodic (bouts separated by remissions of three months or more) and chronic (a year or more without such remission). Chronic CH is roughly 10–15% of Western cases but markedly rarer (about 2–7.5%) in East Asian cohorts, a difference nobody has explained. The best quantitative pain study available (n=1,604) puts mean attack pain at 9.7 ± 0.6 out of 10 — 72.1% of respondents rated it a full 10 — significantly above childbirth, pancreatitis and kidney stones. The nickname “suicide headache” is old and its etymology unverified, but the burden behind it is documented: suicidality is strongly state-dependent (passive suicidal ideation 64.2% during attacks, near zero in remission), while whether CH carries a true population-level excess suicide risk is genuinely contested between a 2025 CH-specific meta-analysis and a large Danish registry study.

Historically, the condition was described piecemeal for three centuries — priority is contested among Tulp (1641), Willis (1672), van Swieten (1745) and others — recognised as its own entity by Wilfred Harris in 1926, named “cluster headache” by Kunkle in 1952, and made an independent diagnosis only in 1988. Mechanistically, the 1998 Lancet PET finding of posterior hypothalamic activation during attacks moved the field from vascular theories to a central, brain-origin model; the current view treats the hypothalamus as a “crossroads” in a wider network rather than a sole generator. The trigeminal-autonomic reflex explains why pain and same-side tearing arrive together. Much of the rest is contested: CGRP was classically elevated in CH but a well-powered 2024 Danish study found it lower than controls; orexin findings are contradictory; hypothalamic volume findings do not replicate. Genetics is firmer ground — the 2023 GWAS found eight loci, SNP heritability of 14.5%, and Mendelian-randomisation evidence that smoking is causal for developing CH — which sits paradoxically alongside consistent clinical evidence that quitting smoking essentially never improves established disease. The disease’s clockwork is its most distinctive feature: about 70% of patients show circadian attack timing with a 2 a.m. peak, and bouts cluster around the equinoxes, tracking the speed of daylight change rather than the solstices.

Epidemiologically CH affects roughly 1 in 1,000 people (pooled lifetime prevalence 124/100,000), with a pooled male:female ratio of 4.3:1 — whether that ratio has genuinely narrowed over recent decades is contested, with the most population-representative source unable to confirm the narrowing that clinic cohorts report. Onset peaks around age 30. Diagnostic delay pooled across studies is 10.43 years, but it is shrinking fast (Danish data: 39 years for 1950s onset, 0.9 years for 2010s onset); migraine, trigeminal neuralgia, sinusitis and dental disease are the standard misdiagnoses, and a meaningful minority of patients undergo unnecessary sinus or dental surgery first. No population-based prevalence study exists anywhere in the Southern Hemisphere, including Australia.

In treatment, one pattern dominates everything: almost every therapy performs worse in chronic than episodic CH, and the headline figure quoted is almost always the episodic one. The acute mainstays are high-flow oxygen (78% relief at 15 minutes vs 20% on air in the pivotal RCT; no daily ceiling, which makes it the structural workhorse for frequent attacks) and subcutaneous sumatriptan 6 mg (75% relief at 15 minutes, NNT 2.4, but capped at two injections per day — formally under-treating anyone with more than two attacks daily). Citizen-science data find the two comparably effective at adequate oxygen flow. For bridging, suboccipital/GON steroid injection has two positive RCTs and was the only Level A preventive in the AHS 2016 guideline. Verapamil remains first-line prevention on a strikingly thin, 26-year-old randomised base, with real arrhythmia risk and documented monitoring failure (41% of one 217-patient series had never had an ECG). CGRP monoclonal antibodies produced one marginal win (galcanezumab, episodic only) and failed every chronic-CH trial across three drugs. In neuromodulation, the best-evidenced device (SPG stimulation) is commercially unavailable after its manufacturer collapsed, stranding 700+ implanted patients; ONS is obtainable but two controlled designs failed to show the electricity itself is the active ingredient; DBS carries a documented death. Access, not efficacy, is often the binding constraint: Germany and Japan formally license and reimburse CH oxygen, while Australia — on the verified comparison in Part V — has the worst access position of the four jurisdictions reviewed: no PBS pathway for oxygen, no PBS listing for the injectable sumatriptan that is TGA-labelled specifically for CH, and no nasal triptan on the market at all.

The patient community has repeatedly outrun formal medicine. The clearest case is psychedelics: from a 1998 message-board report through Clusterbusters (founded 2002) to a 2006 Harvard survey and controlled Yale trials from 2022, with the community’s claim that sub-hallucinogenic doses work now substantially validated. The high-dose vitamin D3 “Batch protocol” is widely used but its one formal RCT was terminated in 2024 on recruitment failure; energy-drink aborts, reported constantly, have never been formally tested. In citizen-science surveys, serotonergic psychedelics rank as the most effective self-reported prophylactic (~75%), above verapamil and corticosteroids — the largest community-vs-trial gap identified anywhere in these chapters.

At the frontier, PACAP is arguably the best-evidenced untried drug target in headache medicine (plasma PACAP-38 elevated 49.8% in chronic CH) with precisely zero CH trials anywhere; orexin has never been tested in CH at all. The live pipeline includes LSD minidosing (recruiting), Ceruvia’s non-hallucinogenic BOL-148 (Phase 1/2 planned), candesartan Phase 3s, sodium oxybate for nocturnal attacks, and the first circadian light-therapy trial. Chronic CH increasingly looks biologically distinct — pro-inflammatory where episodic-in-bout looks anti-inflammatory — and has been argued for rare-disease designation. Structurally, the field is starved: the UK’s three major funders have given CH £0 against roughly £227 million for comparably prevalent multiple sclerosis, the NIH runs two CH-specific grants, and patient organisations — on budgets as small as OUCH(UK)‘s £29,444 a year — are doing the registry, survey and advocacy work that public funders are not. The Watchlist at the end of this document is the live monitoring table for what happens next.