A validation study of the identification of haemophagocytic lymphohistiocytosis in England using population-based health data.

Bishton, Mark J; Stilwell, Peter; Card, Tim R; Lanyon, Peter; Ban, Lu; Elliss-Brookes, Lucy; Manson, Jessica; Nanduri, Vasanta; Earp, Kate; Flower, Luke; Amarnani, Raj; Rankin, Judith; Sen, Ethan S; Tattersall, Rachel S; Crooks, Colin J; Aston, Jeanette; Siskova, Veronika; West, Joe; Bythell, Mary

We assessed the validity of coded healthcare data to identify cases of haemophagocytic lymphohistiocytosis (HLH). Hospital Episode Statistics (HES) identified 127 cases within five hospital Trusts 2013-2018 using ICD-10 codes D76.1, D76.2 and D76.3. Hospital records were reviewed to validate diagnoses. Out of 74 patients, 73 were coded D76.1 or D76.2 (positive predictive value 89·0% [95% Confidence Interval {CI} 80·2-94·9%]) with confirmed/probable HLH. For cases considered not HLH, 44/53 were coded D76.3 (negative predictive value 97·8% [95% CI 88·2-99·9%]). D76.1 or D76.2 had 68% sensitivity in detecting HLH compared to an established active case-finding HLH register in Sheffield. Office for National Statistics (ONS) mortality data (2003-2018) identified 698 patients coded D76.1, D76.2 and D76.3 on death certificates. Five hundred and forty-one were coded D76.1 or D76.2 of whom 524 (96·9%) had HLH in the free-text cause of death. Of 157 coded D76.3, 66 (42·0%) had HLH in free text. D76.1 and D76.2 codes reliably identify HLH cases, and provide a lower bound on incidence. Non-concordance between D76.3 and HLH excludes D76.3 as an ascertainment source from HES. Our results suggest electronic healthcare data in England can enable population-wide registration and analysis of HLH for future research.