Study patients
A total of 293 consecutive patients with a primary diagnosis of adult secondary hemophagocytic lymphohistiocytosis (sHLH) between January 1, 2014, and January 1, 2020, were enrolled. The HLH-2004 criteria developed by Henter and colleagues [11] and the HScore by Fardet L [12] were routinely used to help with the diagnosis of HLH. The inclusion of patients was based on 1) age ≥18 years and 2) fulfillment of at least 5 of the 8 criteria proposed by the Histiocyte Society in 2004. The exclusion criteria included 1) Patients less than 18 years old (n=5); 2) patients with a history of anticoagulant therapy (n=6); and 3) patients with a history of cirrhosis or other serious liver diseases (n=4). Figure 1 shows the flow chart of the patient inclusion and exclusion process in this study. Our study was approved by the ethics committee of the First Affiliated Hospital of Nanjing Medical University (Clinical Trial: ChiCTR2000032421). The study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained to review patient medical records.
Clinical data
Patients’ histories and clinical characteristics including age; gender; fever; complete blood cell counts (CBC); hepatosplenomegaly; and blood biochemical tests (including aspartate aminotransferase (AST), alanine aminotransferase (ALT), triglycerides (TG), lactate dehydrogenase (LDH), fibrinogen (FIB), ferritin, serum soluble interleukin-2 receptor (sIL-2R, sCD25), and β2-microglobulin (β2-MG) tests) were reviewed from their medical records on admission. Epstein-Barr virus (EBV) was evaluated by both serology and EBV DNA real-time quantitative polymerase chain reaction (RQ-PCR) analysis. The NK-cell cytotoxicity assay is not available at our facility. Bone marrow aspiration and biopsy samples were reviewed at the first diagnosis. Tumor or lymph node biopsy and PET/CT confirmed malignancy-associated hemophagocytic lymphohistiocytosis (MHLH).
The FIB levels were measured in the clinical laboratory of our hospital following regular procedures and assayed by the Clauss method with an automatic coagulometer (Sysmex CS5100, Japan). The FIB levels were categorized in tertiles according to the distributions of the study population: tertile 1 (T1) ≤ 1.20, 1.21 ≤ tertile 2 (T2) ≤ 1.97 and tertile 3 (T3) > 1.97.
Follow-up and endpoints
The primary outcome of the current study was overall survival (OS), which was calculated as the time in days from sHLH diagnosis to the date of death from any cause or the last follow-up. The survival status of all participants was confirmed with death records or a telephone call to the patient’s relatives or to the patient themselves and ascertained by checking government records of death.
Treatment according to causes of sHLH
Two hundred and seventy-eight patients (95.2%) received a specific treatment for sHLH. In our 169 MHLH patients, 120 patients had received systemic combination chemotherapy as the first-line therapy for sHLH; 33 patients were treated with HLH-94 as the initial therapy; 10 patients with progressive multiple organ dysfunction were treated with only intravenous immunoglobulins (IVIg) and glucocorticoid (GC). In our 124 non-malignancy-associated hemophagocytic lymphohistiocytosis (non-MHLH) patients, first-line treatment using HLH-94 was administered in 14 patients; GC was administered in 11 patients; GC + IVIg was administered in 46 patients; GC + etoposide was administered in 26 patients; GC + IVIg + cyclosporine was administered in 15 patients; and GS + IVIg + cyclophosphamide was administered in 4 patients. No difference in treatment regimens occurred across the FIB groups.
Statistical analysis
Continuous variables were presented as the means with standard deviations (SDs) for normally distributed variables or the median with the interquartile range (IQR) for nonnormally distributed variables, and categorical variables were described as numbers and proportions (percentages). Differences across FIB tertiles were analyzed using one-way ANOVA and the Kruskal-Wallis and chi-squared tests where appropriate. We used univariate and multivariable Cox proportional hazards models to estimate hazard ratios (HRs) and 95% CIs to examine the relationship between fibrinogen concentrations and survival (variables with P <0.05 in the univariate Cox regression were subsequently entered into the multivariable model). The fibrinogen of T3 was used as a reference to estimate HR and 95% CI. Kaplan-Meier curves for tertiles of fibrinogen were plotted to show mortality rates. Survival curves were adjusted for covariates derived from the final model according to multivariable Cox proportional hazards models.
A restricted cubic spline (with 3 knots located at the 10th, 50th, and 90th percentiles; the number of knots was selected according to the Akaike information criterion) was conducted to address the nonlinear relationship between fibrinogen and the risk of mortality after adjusting for confounding factors. If nonlinearity was detected, we calculated the inflection point using a recursive algorithm and then constructed a two-piecewise Cox proportional hazards model on both sides of the inflection point. The threshold level was calculated according to the inflection point with the maximum model likelihood using the trial and error method. A log likelihood ratio test comparing the one-line Cox proportional hazards model with a two-piecewise Cox proportional hazards model was performed to examine the statistical significance.
Stratified analyses were conducted according to gender (male or female), age (≤ 60 years or > 60 years), ferritin (≤ 10,000 µg/L or > 10,000 µg/L), sCD25 (≤ 20,000 ng/L or > 20,000 ng/L), EBV status (no EBV infection or EBV infection), and etiologies (MHLH or non-MHLH). We tested for potential effect modification by these stratification variables by including interaction terms between fibrinogen and a potential effect modifier (overall survival) in the multivariate adjusted model and by conducting a likelihood ratio test (LRT) comparing the models with and without interaction terms. Tests for trends were calculated by including the median value for each corresponding tertile as a continuous variable in the models [13]. All statistical analyses were conducted with R software (version 3.6.0; The R Foundation for Statistical Computing) and STATA statistical software (version 14.0; StataCorp, TX, USA). A two-sided P < 0.05 was considered statistically significant for all analyses.