Introduction

Hemophagocytic lymphohistiocytosis (HLH) is a severe disease, accompanied with persistent fever, high levels of ferritin, splenomegaly, cytopenia, cytokine storm, and multi-organ dysfunction [1]. Cytokine storm coming from activated natural killer (NK) and/or cytotoxic T cells was the direct cause of HLH [2]. HLH could be developed at any age [3]. In Japan, from 2001 to 2005, adult patients account for 40% of the total HLH [4]. In adults patients, infection and malignancy are the most commonly reported underlying diseases [5]. In addition, when HLH developed, one or more organ functions would be affected (such as liver, kidney, and coagulation disorders). Thus, the prognosis varies widely [6, 7].

Elevated post-ferritin, interleukin-10, Epstein-Barr virus (EBV) infection, fibrinogen (FIB), platelet (PLT) and lactate dehydrogenase (LDH), blood routine parameters, and their ratios are risk markers and models that included some of these indicators have excellent predictive ability to forecast the outcomes of patients with HLH [3, 8,9,10]. To malignancy-induced HLH patients, the prognosis is also poor [11, 12]. However, prognostic efficiency and efficacy would be low as these indexes may be expensive and/or have low sensitivity/specificity. Thus, feasible, non-invasive, and reasonably predictive models are needed.

In clinical practice, routine blood test, biochemical test, and coagulation test are the most common, readily available, and measured tests. In models, combinations of multi-factors possess markedly better presentation than a single marker [13]. Thus, the purpose of this investigation was to build models to predict 30-day death in total and subgroup HLH patients based on available and cheap factors.

Methods

Patients and Data

The research contained 431 adults with HLH (meeting 5 or more of the 8 HLH-2004 criteria) [14] from January 2015 to September 2021 in the hospital. Inclusion criteria: (1) age ≥ 18 years; (2) newly diagnosed HLH; (3) complete clinical results. Exclusion criteria: patients with unknown outcome. Data such as age, gender, clinical signs, admission laboratory parameters, etiology, medicine, and outcomes.

The primary outcome was defined as death within 30 days when HLH was initially confirmed. The research was consistent with the Declaration of Helsinki and permitted by the hospital (Nanjing, China) (2019-SR-066).

Statistical Methods

Differences in categorical biomarkers (frequencies), continuous parameters (means ± SD), and non-parametric indexes (medians (ranges)) were evaluated by the Chi square test, ANOVA, T-test, or Mann-Whitney U test as appropriate. Logistic regression was employed to seek out biomarkers of 30-day mortality. Parameters (p < 0.10) on univariate analysis were included in the multivariable analysis. Receiver operating characteristic (ROC) was utilized to judge the performance of models in predicting adverse events and selected the ideal points for each of the models and parameters. p-value <0.05 was considered statistically significant. Data analysis was performed by IBM SPSS 21.0 statistical software (IBM SPSS Version 21.0. Armonk, NY).

Results

Characteristics

In total, 431 subjects with HLH (252 males, 179 females) were recruited in the retrospective research. The overall mean age was 52.2 ± 16.5 years. Almost all patients were accompanied with fever (> 90%); EBV infection was remarked in 42.92% of the patients. The overall mortality was 29.7%. Other clinical and laboratory parameters covered splenomegaly (208, 48.26%), lymph node enlargement (200, 46.40%), hemophagocytosis (139, 32.25%), and hyperferritin (2358.9 (15.2, 15,000)). Underlying diseases of HLH were listed as follows: infectious diseases (named infection-associated HLH (I-HLH)) (150, 34.56%); malignancies (named malignancy-associated HLH (M-HLH)) (80, 18.56%); mixed connective tissue disease (named MCTD-HLH) (6, 1.39%); unidentified disease (named Unclear-HLH) (33, 7.66%); and two or more different etiologies-associated HLH which could not be classified into any of the above groups (named Mix-HLH) (162, 37.59%). Other features of the total and subgroups of HLH patients are presented in Table 1. In addition, we also listed and compared the types of infections and malignancies in I-HLH, M-HLH, and Mix-HLH (Supplementary table 1).

Table 1 The demographic, clinical, and laboratorial characteristics of the patients included in the study
Full size table

Independent Parameters for Total or Subgroup HLH

Based on the follow-up results, patients were assigned to survivors and nonsurvivors. Compared with survivors, there was a distinctly difference in nonsurvivors in the following laboratory data: age, ferritin, EBV infection, lymphocyte (LY), neutrophil, hemoglobin, PLT, prothrombin time (PT), international normalized ratio (INR), activated partial thromboplastin time (APTT), FIB, thrombin time (TT), d-dimer, alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin (TB), direct bilirubin (DB), indirect bilirubin (IB), triglycerides (TG), albumin, globulin, urea nitrogen (UREA), creatinine (CREA), uric acid (UA), chloride for total HLH (all p < 0.10); age, ferritin, EBV infection, LY, PLT, APTT, FIB, TT, d-dimer, ALT, AST, TB, DB, TG, UREA, chloride for I-HLH patients (all p < 0.10); ferritin, EBV infection, LY, neutrophil, PLT, PT, INR, APTT, FIB, TT, ALT, AST, TB, DB, IB, TG, globulin, UREA, CREA for M-HLH patients (all p < 0.10); age, ferritin, PLT, PT, INR, APTT, TT, d-dimer, AST, TB, DB, IB, TG, UREA, CREA, UA for Mix-HLH patients (all p < 0.10) (Table 2).

Table 2 The laboratorial characteristics of the patients included in the study
Full size table

To select the significant independent markers, ROC was employed to discover the optimal value for each of the continuous factors and models. Then, we reset these parameters in Table 2 with p < 0.10 (except for EBV infection, as not all patients were tested for EBV) into categorical parameters.

On univariate and multivariable logistics regression analyses for HLH patients: age (hazard ratio (HR): 2.990, 95% confidence interval (CI): 1.657–5.397, p < 0.001), ferritin (HR: 4.264, 95% CI: 2.513–7.238, p < 0.001), LY (HR: 0.391, 95% CI: 0.232–0.659, p < 0.001), INR (HR: 2.513, 95% CI: 1.382–4.568, p = 0.003), TT (HR: 2.077, 95% CI: 1.193–3.616, p = 0.010), globulin (HR: 0.560, 95% CI: 0.301–0.848, p = 0.010), UA (HR: 2.637, 95% CI: 1.394–4.988, p = 0.003), and chloride (HR: 0.479, 95% CI: 0.285–0.806, p = 0.006) remained as independent factors of 30-day death. Model was set up depending on these markers (Table 3) (Table 4).

Table 3 Logistic regression analysis of independent factors and 30-day mortality
Full size table
Table 4. Diagnostic statistics for clinical predication models
Full size table
$$Clinical\;model:Logit P=1.095\times age+1.450\times ferritin-0.938\times LY+0.921\times INR+0.731\times TT-0.682\times globulin+0.969\times UA-0.736\times chloride-4.355$$

After adjusted confounding factors for I-HLH patients: age (HR: 8.813, 95% CI: 2.602–29.848, p < 0.001), LY (HR: 0.235, 95% CI: 0.078–0.710, p = 0.010), APTT (HR: 5.028, 95% CI: 1.709–14.799, p = 0.003), TT (HR: 3.736, 95% CI: 1.233–11.321, p = 0.020), AST (HR: 3.656, 95% CI: 1.219–10.965, p = 0.021), TG (HR: 4.016, 95% CI: 1.356–11.894, p = 0.012), and chloride (HR: 0.214, 95% CI: 0.074–0.615, p = 0.004) were still in the multivariate model. A model was developed (Table 3).

$$Clinical\;model:Logit P=2.176\times age-1.499\times LY+1.615\times APTT+1.318\times TT+1.296\times AST+1.390\times TG-1.543\times chloride-7.484$$

On univariate and multivariable exploration for M-HLH patients: ferritin (HR: 53.839, 95% CI: 7.772–372.961, p < 0.001), LY (HR: 0.086, 95% CI: 0.016–0.467, p = 0.011), and TB (HR: 12.326, 95% CI: 1.765–86.063, p = 0.005) were confirmed as independent indexes. A multivariate model was formed (Table 3).

$$Clinical\;model:Logit P=3.986\times ferritin-2.458\times LY+2.512\times TB-6.274$$

On univariate and multivariable logistics regression analyses for Mix-HLH patients: ferritin (HR: 2.215, 95% CI: 1.009–4.864, p = 0.048), LY (HR: 0.387, 95% CI: 0.175–0.859, p = 0.020), APTT (HR: 4.075, 95% CI: 1.470–11.294, p = 0.007), IB (HR: 2.516, 95% CI: 1.129–5.609, p = 0.024), TG (HR: 3.120, 95% CI: 1.037–9.386, p = 0.043), and UA (HR: 4.680, 95% CI: 1.918–11.418, p = 0.001) persisted as significant factors of 30 days mortality. A multivariate model was created (Table 3).

$$Clinical\mathit\;model:\;LogitP=0.759\times ferritin-0.948\times LY+1.405\times APTT+0.923\times IB+1.138\times TG+1.543\times UA-6.652$$

We further explored the distribution of mortality. Figure 1 showed the percentage of mortality by the cumulative number of independent biomarkers. The more risk factors, the higher the mortality.

Fig. 1
figure 1

Distribution of mortality by cumulative number of risk factors. A Distribution of mortality for HLH patients. B Distribution of mortality for I-HLH patients. C Distribution of mortality for M-HLH patients. D Distribution of mortality for Mix-HLH patients

Full size image

Performance of Models

ROC was hired to estimate the performance of these clinical models. AUC of these models was listed as follow: HLH patients: 0.842 (95% CI 0.804–0.875, p < 0.001); I-HLH patients: 0.913 (95% CI 0.856–0.953, p < 0.001); M-HLH: 0.921 (95% CI 0.838–0.969, p < 0.001), and 0.809 (95% CI 0.740–0.866, p < 0.001) for Mix-HLH patients (Fig. 2).

Fig. 2
figure 2

Performance of the models. A AUCHLH for model. B AUCI-HLH for model. C AUCM-HLH for model. D AUCMix-HLH for model. E Calibration curves for HLH model. F Calibration curves for I-HLH model. G Calibration curves for M-HLH model. H Calibration curves for Mix-HLH model.

Full size image

Discussion

In this study, models for total or subgroup HLH patients were established based on admission laboratory results. LY was an independent marker for total or subgroup HLH patients. For HLH patients, age, ferritin, LY, INR, TT, globulin, UA, and chloride were the most stable factors and the combination of them is the most predictive. For I-HLH patients, age, LY, APTT, TT, AST, TG, and chloride were the most stable factors and the combination of them is the most predictive. For M-HLH patients, ferritin, LY, and TB were the most stable factors and the combination of them is the most predictive. For Mix-HLH patients, ferritin, LY, APTT, UB, TG, and UA were the most stable factors and the combination of them is the most predictive.

Laboratory markers could be useful in monitoring the prognosis of HLH patients. It is reported that old age, low LY, elevated ferritin, high post-treatment ferritin, TB, triglycerides, and AST were positively related with adverse events such as early death, long prognosis in adult HLH patients, or malignancy-associated HLH [3, 8, 15,16,17,18,19]. Similar results were also concluded in our research. Furthermore, to pediatric HLH patients, INR > 1.5, APTT, TB > 19 μmol/L, and globulin < 20 g/L were risk factors for 7 days and within 30 days prognosis [20]. In this investigation, similar results were gained in adult HLH patients.

Serum bilirubin from the metabolism of heme may be effective in protecting the body from oxidant inflammation and cancers in colorectal cancer [21, 22]. Previous research has described that elevated IB was a risk factor in mortality in the acute phase of ischemic stroke patients [23]. UA, an end product of purine, plays a key role in antioxidants [24,25,26]. Chloride ion, a plentiful anion in the extracellular fluid, is an essential part of many functions to work with the body [27]. A high level of serum chloride (≥ 105.4 mmol/L) is an adverse prognosis marker for IgAN patients [28]. Our study also provides clinical data on the prognosis role of the above parameters in adult HLH patients. In addition, the thrombin time (TT), a common coagulation index, has rarely been studied in many diseases. In this research, we firstly found that high TT was an independent and risk factor for HLH patients in 30-day death.

In this research, we built four models based on admission laboratory results. Firstly, we divided the total patients into I-HLH, M-HLH, and Mix-HLH groups according to the underlying diseases. Then, univariate and multivariable logistics regression analyses were applied to select the independent factors to predict 30-day mortality in total and subgroups. We found that independent factors were different in different groups with different values. Therefore, the choice of which laboratory marker should be the primary outcome may depend on the underlying disease. For example, LY is an independent factor to total and subgroup HLH. Ferritin is an independent factor in total, M-HLH, and Mix-HLH. Age, TT, and chloride are independent factors to total and I-HLH.

Strengths of the investigation included that data were from a single center with large samples. Patients were evaluated as total and subgroups with biomarkers that are available and affordable in clinical practice. UA, TT, and chloride were firstly confirmed as independent factors in adult HLH.

As a retrospective research, selection bias is unavoidable. Internal and external validation is lacking. In addition, less information on treatment and genetic testing is included. We encourage these findings to be tested in other studies with multi-centers.

In conclusion, we built four models with biomarkers that are available and affordable in clinical practice. With these models, high-risk patients with different underlying disease could be easily identified.