(219) Identification of migraine in electronic healthcare data sources: a headache for pharmacoepidemiologists – a systematic literature review

Background: Ascertaining migraine in electronic healthcare data is challenging because of likely diagnosis underrecording and treatment with over-the-counter analgesics, which cannot be used as disease proxies. Migraine prevalence will depend on the characteristics of the migraine-identifying algorithm (e.g., diagnosis codes, treatments or combinations of both) and healthcare data source used.

Objectives: To describe migraine-identifying algorithms implemented in observational studies that use healthcare administrative claims, electronic medical records, or other data sources and any related validation efforts; and to summarize the prevalence of migraine observed in those studies.

Methods: This systematic literature review searched PubMed in November 2023 for peer-reviewed, English-language, original research articles published since 2013 that identified migraine in adults using electronic algorithms in electronic healthcare data. Two reviewers independently screened articles for inclusion; data were extracted by one researcher and checked by the other. The study was registered with PROSPERO (CRD42023491279).

Results: The literature search identified 630 unique articles. After reviewing titles and abstracts, 50 (8%) were selected for full-text review; of those, 41 (82%) were included.
Among the 41 articles, 18 (44%) identified migraine using only diagnosis codes, 3 (7%) used only migraine treatments, 11 (27%) used treatments or diagnosis codes, and 9 (22%) used combinations of diagnosis codes and treatments in different settings (e.g., primary care, hospital discharge, specialist). Sixteen studies were conducted in Europe, 13 in the United States or Canada, and 12 in Asia. The most common diagnosis codes to ascertain migraine were ICD-9 346.xx and ICD-10 G43.xx. Triptans were the most frequently drugs used to identify migraine (ATC: N02CC).
Where reported, the prevalence of migraine in the general population ranged between 4% and 17%. Algorithms combining diagnosis codes, treatments, and additional conditions including setting identified specific types of migraine or cases of treatment-resistant migraine.
Only 2 studies reported validation results: one identified prevention-eligible patients with migraine (PPV = 97%), and one identified migraine based on a calculated probability algorithm with PPVs between 68% and 92% depending on the diagnostic threshold and the gold standard used.

Conclusions: Most studies included in this systematic review identified migraine using diagnosis codes. Only studies that included a combination of diagnosis codes and treatments in different settings identified specific migraine types and reported results on validation. Prevalence varied by type of data source.