Page 72 - 2018_10-Haematologica-web

E. Gars et al.
marrow report of all 40 patients with HLH, compared to 12 of 38 (32%) in the non-HLH group (P<0.01). Although evi- dence of hemophagocytosis may be reassuring for clini- cians in making the diagnosis of HLH, hemophagocytosis is not required. In our cohort, the majority of patients ulti- mately diagnosed with HLH (73%) fulfilled 5 or more crite- ria independently of evidence of hemophagocytosis in the bone marrow aspirate (Table 1). However, 9 patients (23%) in the HLH group and 8 patients (21%) in the non-HLH group met 4 criteria excluding hemophagocytosis, indicat- ing that morphological assessment of the bone marrow aspirate was critical to the diagnosis in 22% of patients pre- senting with clinical features of concern for HLH.
Two patients in the HLH group satisfied fewer than 5 of the HLH-2004 criteria; however, both of these patients were surgically asplenic and the remainder of their clinical and laboratory findings, including the presence of hemo- phagocytosis on bone marrow aspirate, were compatible with the diagnosis. Four patients in the non-HLH group satisfied 5 criteria but were not diagnosed with HLH. The clinical notes of the treating physicians indicate that the chronicity of symptoms was considered to be inconsistent with HLH. None of these patients had hemophagocytosis identified in their bone marrow.
Figure 2. Flow chart for classification of patients. *Patients were excluded from the analysis if hemophagocytosis was incidentally noted independent of clinical concern for hemophagocytic lymphohistiocytosis (HLH), slides were not avail- able for review, HLH was considered but the diagnosis was equivocal after workup, or a documented history of HLH-directed treatment was noted prior to biopsy.
Developing predictive models for a morphological diagnosis of hemophagocytic lymphohistiocytosis
Patients diagnosed with HLH (n=40) displayed signifi- cantly higher values of total hemophagocytes, and hemo- phagocytes with any of the cell lineages (RBCs, nRBCs, granulocytes, and lymphocytes) compared to non-HLH patients (n=38) (P<0.001 for all lineages) (Figures 1 and 3). Correlation analysis demonstrated that each of the clinical and laboratory criteria included in the HLH-2004 criteria (excluding NK function) has a significant positive correla- tion with total number of hemophagocytes and hemo- phagocytes with each of the individual cell lineages, indi- cating that the degree of hemophagocytosis correlates with the diagnosis of HLH (Online Supplementary Figure S1).
Independently of one another, each lineage distin- guished between HLH and non-HLH fairly well. Ingested nRBCs and ingested granulocytes had the highest area under the curve (AUC), indicating a high degree of correct classification of HLH and non-HLH subjects (AUC: 0.92), with threshold values of 2 ingested cells and 1 ingested cell, respectively. The sum of all four lineages also per- formed well at distinguishing between HLH and non- HLH with a threshold value of 6 (AUC: 0.92, 95%CI: 0.85, 0.98).
Dichotomizing each lineage based on the threshold val- ues described above and including all four in a decision tree, hemophagocytes ingesting granulocytes was chosen as the most important predictor of HLH, followed by nRBCs and lymphocytes (Figure 4A). Patients with an absence of hemophagocytes ingesting granulocytes had a 3% chance of having an HLH diagnosis. Patients with a presence of at least one hemophagocyte with an ingested granulocyte, two or more hemophagocytes with ingested nRBCs, and one hemophagocyte with ingested lympho- cytes together were guaranteed to have an HLH diagnosis (100% chance). Cross-validated (CV) AUC of this CART was 0.90; 95%CI: 0.83-0.97.
Identifying lymphocytes within HPCs is rare, even in the presence of florid HLH. The highest number of HPCs containing lymphocytes was identified in a patient with
Figure 3. Values of hematopoietic progenitor cell (HPC) lineages and sum by hemophagocytic lymphohistiocytosis (HLH) diagnosis. Patients ultimately diag- nosed as HLH had significantly higher values of all variables compared to non- HLH patients (Kruskal-Wallis rank sum test, P<0.001).
1638
haematologica | 2018; 103(10)