Validation of the HScore and the HLH‐2004 diagnostic criteria for the diagnosis of hemophagocytic lymphohistiocytosis in a multicenter cohort

Abstract

Timely diagnosis of hemophagocytic lymphohistiocytosis (HLH) is critical and relies on clinical judgment. The HLH‐2004 criteria are commonly used diagnostic criteria, whereas HScore was recently developed for reactive HLH. Objective In this external validation study, we sought to compare the diagnostic accuracy of the HLH‐2004 criteria and HScore and identify optimal cutoffs stratified by underlying etiology. Methods In this retrospective cohort of all hospitalized adults in Alberta, Canada, (1999–2019) who had ferritin >500 ng/ml and underwent either biopsies or soluble CD25 testing, we calculated the diagnostic accuracy of HLH‐2004 and HScore for the overall population and different etiologies. Results Of 916 patients, 98 (11%) had HLH. HLH‐2004 criteria ≥5 predicted HLH with a sensitivity of 91%, specificity of 93%, positive predictive value of 90%, and negative predictive value of 94% (c‐statistic 92%). HScore ≥169 predicted HLH with better sensitivity (96%) but reduced specificity (71%), whereas the optimal cutoff ≥200 performed comparably to HLH‐2004. HLH‐2004 criteria outperformed HScore in most etiologies, whereas HScore improved sensitivity in inflammatory/autoimmune‐HLH. The optimal cutoff of HScore was higher in hematopoietic cell transplant due to higher prevalence of fevers and cytopenias. Conclusion HLH‐2004 criteria and HScore demonstrated excellent discriminatory power in identifying HLH. HScore may improve diagnostic accuracy in autoimmune‐HLH.

Diagnosing Hemophagocytic Lymphohistiocytosis with Machine Learning: A Proof of Concept

“…The two main available scores for HLH diagnosis are currently the HLH-2004 criteria and the Hscore [ 6 , 15 ]. External validation studies revealed that those scores had high sensitivity and specificity, which could vary depending on the scored population [ 7 , 17 ]. HLH-2004 criteria and Hscore perform better in critically ill patients.…”

“…HLH-2004 criteria and Hscore perform better in critically ill patients. Hscore performs better in non-IMID-related HLH patients since the original study included only 3% of patients with IMID-related HLH and thus, this linear model is probably unsuitable for populations with a high proportion of IMID-related HLH [ 7 , 16 ]. These performances are to be modulated depending on the cutoff score that is chosen to define HLH when using Hscore.…”

“…This choice may lead to information loss, especially for extreme values (i.e., in the case of extreme hyperferritinemia), but was essential for the deployment of the tool in order to be as simple as possible without significant performance loss. Although in external validation studies, the cutoff that was proposed by the authors failed to be sufficiently specific, yielding to high sensitivity, the logistic regression design allows to adapt the score cutoff to fit a given population [ 7 , 16 , 17 ]. Finally, the parameters that were selected to build this tool were chosen from a web-based Delphi study [ 31 ].…”

“…HLH diagnosis in adults relies mainly on the clinical expertise of physicians, although the classification criteria proposed by The Histiocyte Society (HLH-2004) are regularly used as a reference in studies [ 6 ]. However, these criteria were developed for children with primary HLH and have not been formally validated for sHLH or adults [ 7 ]. In addition, these criteria may hamper early diagnosis as some criteria (e.g., cytopenias and hemophagocytosis) may occur late and some tests (e.g., NK cell activity and sCD25 levels) are not routinely available in hospital laboratories [ 1 ].…”

“…Several subsequent studies aimed at testing the validity of the Hscore on cohorts of patients with more varied triggers have all found poorer diagnostic performance. In particular, specificity was variable (specificity between 71–93%, depending on the study vs. 86% in the original population) [ 7 , 15 , 16 , 17 ]. In total, none of these scores are completely satisfactory for the diagnosis of sHLH and none have achieved complete consensus.…”