Abstract
Background: Non-traumatic headache accounts for a substantial minority of emergency department (ED) attendances. Only a small proportion of these presentations arise from a life-threatening secondary cause, but the consequences of a missed diagnosis are severe. Guidance concentrates heavily on subarachnoid haemorrhage, and it is unclear how the wider differential should be structured at the bedside.
Objective: To synthesise, without meta-analysis, the accuracy and diagnostic yield of clinical features, decision rules, imaging, cerebrospinal fluid analysis, biomarkers and bedside adjunct tests for identifying life-threatening secondary headache in adults presenting to the ED with non-traumatic headache, and to derive an evidence-based diagnostic algorithm.
Methods: PubMed/MEDLINE (two strands), Europe PMC and ClinicalTrials.gov were searched on 30 June 2026 for reports published between 1 January 2000 and 30 June 2026, supplemented by forward and backward citation searching. Eligible studies enrolled adults presenting to an ED or equivalent acute assessment setting with non-traumatic headache and reported diagnostic accuracy, or the frequency of a serious secondary cause within a defined denominator or index-test stratum. One reviewer screened records, extracted data and appraised studies using QUADAS-2. Synthesis was by structured tabulation and narrative comparison grouped by index test, with no statistical pooling. Confidence in the body of evidence was assessed using a structured, explicitly labelled framework.
Results: Of 6,465 unique records screened, 199 studies (210 reports) were included, conducted in 34 countries; 57 reports described prospective designs and 128 retrospective designs. A serious secondary cause was identified in approximately 5% to 13% of ED non-traumatic headache populations, and in 10.9% of thunderclap presentations. A normal neurological examination did not exclude a serious cause (4.8% of such patients). Composite decision rules were highly sensitive but poorly specific: the Ottawa Subarachnoid Haemorrhage Rule reached 100% sensitivity in several validation cohorts with specificity of 13.6% to 37.0%, and applied to fewer than one quarter of ED headache presentations. Non-contrast CT performed within six hours of headache onset reached sensitivities of 97% to 100% with negative predictive values of 99.6% to 99.9%, falling to approximately 90% beyond six hours. After a negative CT, the diagnostic yield of lumbar puncture was 0.4% to 1%, against a complication rate of about 6%, although 7% of confirmed haemorrhages were still identified only by lumbar puncture. Jolt accentuation of headache showed sensitivities from 21% to 75% and cannot exclude meningitis. Selected adjuncts, including D-dimer combined with a clinical probability score for cerebral venous thrombosis and non-mydriatic fundus photography for papilloedema, showed useful rule-out or detection properties. Only 5 of 210 reports were at low risk of bias across all four QUADAS-2 domains; index-test blinding was unclear in 182 reports.
Conclusions: The evidence supports a staged, time-anchored diagnostic pathway in which a validated rule determines who requires investigation, early non-contrast CT does most of the diagnostic work for subarachnoid haemorrhage, and phenotype-directed second-line tests address the conditions a normal CT does not exclude. Confidence in the underlying evidence is low to moderate, dominated by single-centre retrospective studies with incomplete reporting, and the proposed algorithm requires prospective external validation before implementation.