Epstein-Barr virus (EBV) establishes persistent infection in almost 90% of the people worldwide, mostly asymptomatic (1, 2). However, once developed into EBV-related hemophagocytic lymphohistiocytosis (EBV-HLH), it becomes life-threatening. HLH comprises a heterogeneous class of disorders characterized by extreme immune activation (3). HLH can be classified as primary (genetic) or secondary (acquired), and EBV infection was found to be implicated in the onset of both primary and secondary HLH (4). EBV-HLH is one of the most common type of HLH in childhood, especially in Asia area (4, 5). One multi-state clinical trial included 369 pediatric patients with HLH, and found that EBV infection was associated with 20% (n=74) of the patients (6). One study from United States found that approximately 1/3 of the HLH patients were positive for EBV (7). In one Chinese series, EBV infection was found in 94.7% (90/95) of the pediatric HLH patients (8). One study from Vietnam found that 47% (40/85) of the pediatric HLH patients were EBV positive (9). Like all types of HLH, EBV-HLH can rapidly deteriorate and lead to multiple organ failure or even death. Timely diagnosis and early administration of effective therapy are essential for survival (10–12).

The consequence of EBV-HLH could be life-threatening, whilst the diagnose of EBV-HLH is challenging. The diagnose of HLH can be established if a molecular (genetic) diagnosis consistent with HLH or five out of eight criteria fulfilled, which include fever, splenomegaly, cytopenias affecting 2 or 3 lineages, hypertriglyceridemia and/or hypofibrinogenemia, hemophagocytosis, low or absent NK-cell activity, hyperferritinemia, and elevated soluble CD25 (3). Some of these criteria, like genetic test, hemophagocytosis, ferritin, NK-cell activity, and soluble CD25, were unlikely to be tested unless HLH or other relevant diseases were suspected. Our previous study showed that the performance of adequate diagnostic tests for HLH is the most important contribute factor for early diagnosis of HLH (13). However, deciding when to suspect HLH and perform diagnostic tests in patients with EBV is challenging, given the high prevalence of EBV infection, the relatively low incidence of EBV-HLH, and the wide spectrum of disease presentations. It has been suggested that HLH should be suspected when any of the diagnostic, typical, and unique features of HLH are present (5, 12, 14). For example, cytopenias of more than 2 lineages may raise a suspicion for HLH. However, since the diagnostic and typical features of HLH may nonspecific and could been seen in patients with other disorders, the efficacy of suspecting HLH when any of the HLH feature is presented needs further evaluation. Moreover, in order to improve the diagnostic process of EBV-HLH, it is imperative to develop evidence based methods for the identification of high risk patients who would be benefit from the HLH-diagnostic test. In this study, we proposed that developing a screening tool for EBV-HLH would help clinical decision on when to suspect HLH among children with acute EBV infection.

The aim of this study was to develop an EBV-HLH screening model for pediatric patients diagnosed with acute infection of EBV. The strategy of our model development had two considerations: first, to be easily used in clinical practice, the model parameters should be selected from the most prevalently checked laboratory tests; second, for the ease of use, the model should be transformed into a scoring criteria. We extracted clinical data of pediatric patients diagnosed with acute infection of EBV from Hunan Children’s Hospital; the outcome of interest was the diagnose of HLH during hospitalization; laboratory tests that had been check on more than 90% patients with acute infection of EBV were selected as candidate model parameters; a logistic model were developed based on the method of Akaike Information Criterion (AIC)-optimal selection through cross-validation (15), and then this primary model was transformed into a scoring criteria, with a suggested cut-off point. The model was developed in a training set (70% of study population), and was validated in a validation set (30% of the study population). The performance of the screening score model was compared with the performance of the screening method based on the number of cytopenias lineages. The EBV-HLH scoring criteria could serve as a simple tool for the screening of HLH among pediatric patients with acute infection of EBV.

During the study period, 3523 pediatric inpatients were tested positive for active EBV infection. Among these, 8 patients were excluded because they were clinically suspected for EBV-HLH but the diagnoses were not confirmed; 332 patients were excluded because missing essential laboratory data. A total of 3183 pediatric EBV-positive pediatric patients were included ( Supplementary Figure S1 ). Table 1 summarized the general characteristics of the study population. The incidence of EBV-HLH in the inpatient EBV cohort were 4.62%. The distributions of age, sex, and incidence of EBV-HLH in the training and validation cohort were not statistically significant from each other (P>0.05, Table 1 ).

Among the 37 investigated laboratory parameters, 32 showed to be significantly different between EBV-HLH and EBV-nonHLH patients (P <0.05, Supplementary Table S1 ). The PCA and PLS-DA plots were presented in Figure 1 . Total bilirubin, direct bilirubin, and indirect bilirubin were the top three defining parameters in the PCA; and LDH, hematocrit, and hemoglobin were the top three defining parameters in the PLS-DA. Both PCA and PLS-DA showed limited ability to separating the EBV-HLH patients from EBV patients ( Figure 1 ).

The model build from the method of AIC-optimal selection through cross-validation included three variables, including platelet, albumin, and LDH. After adding hemoglobin and neutrophil to the model, the AIC was improved from 445.6 to 392.5, and the AUC was improved from 93.1% to 95.2%. Therefore a preliminary model (Model 1) with five parameters, namely hemoglobin, platelet, neutrophil, albumin, and LDH, were determined. The model parameters were presented in Supplementary Table S2 . Figures 2A–E showed the distribution of the selected parameters in EBV patients with and without HLH.

Following the method described above, Model 2 was developed using categorized variables and Model 3 was developed using a score as the predictive variable. The AUCs of three models were similar between the training and validation sets (range from 94.9% to 95.3%) ( Figure 3 ). The median screening score in EBV-nonHLH patients was 0 (Q1, Q3: 0, 22), and the median score in EBV-HLH patients was 62.5 (Q1, Q3: 41, 77; P<0.0001) ( Figure 2F ). Using a cut-of value of 29, the scoring system had a sensitivity of 91.7% and a specificity of 90.0% in the training set, and a sensitivity of 89.2% and a specificity of 89.5% in the validation set. The FNR, FPR, PPV, and NPV in the validation set was 10.8%, 10.5%, 26.8%, and 99.5%, respectively, similar to that of the training set ( Table 2 ). The screening methods using the number of cytopenias lineages (≥ 2 or =3) showed high specificities (>97%), however, with low sensitivities (57.8% for cytopenias of ≥2 lineages and 25.7% for cytopenias of 3 lineages, training set), demonstrating the superiority of the newly developed scoring model. The scoring criteria of the screening score was presented in Table 3 .

Among the 3183 patients included in this study, 2989 had available lab data to calculate the screening score. Among these patients, 2028 were diagnosed with acute infection of EBV according to EBV-DNA copy numbers, and 961 were diagnosed by serological profiles. The performance of the scoring model in the sub-populations divided by two types of EBV tests were presented in Supplementary Table S3 . Although the model accuracy and specificity among patients diagnosed by serology tests (accuracy=94.2%, specificity=94.2%) were higher than that among patients diagnosed by EBV DNA copy numbers (accuracy=87.9%, specificity=87.6%), the model showed better sensitivity in the EBV DNA-diagnosed subpopulation (91.4%) than that in the serology-diagnosed subpopulation (85.7%).

In this study, we developed a score for the screening of HLH among pediatric patients with acute EBV infection. The score model has five parameters, including hemoglobin, platelet, neutrophil, albumin, and LDH. The sensitivity and specificity of the score model in the validation set were 89.2% and 89.5%, respectively. The performance of the screening score was stable in the training and validation set, as well as in the two sub-populations that using different EBV diagnostic tests.

Patients with acute infection of EBV may develop some of the features of HLH, like fever, splenomegaly, and cytopenias (5). Therefore, HLH may be difficult to distinguish from EBV. In clinical practice, patients who presents more sever manifestations of disease may undergo further evaluation for suspected HLH. However, the diagnosis of HLH could be delayed because the diagnosis of HLH requires multiple tests and there are no clinical thresholds that guide the routine practice for the identification of patients who need further evaluations for possible HLH. To improve the diagnostic procedures of HLH, researchers had been working on the development of new diagnostic tools. Smits et al. (2021) identified a minimal parameter set for the prediction of HLH (19); the minimal parameter set consisting phagocytosis, splenomegaly, cytopenias, increased ferritin, and increased triglycerides/low fibrinogen, with a sensitivity of 95% and a specificity of 94%. One of the strengths of the minimal parameter set is it excludes the checking of NK activity and soluble CD25, which cannot be measured in routine practice in many settings and can be time consuming. The HScore was developed for the diagnosis of reactive HLH in adult (18), and has been validated in several adult and pediatric populations (20, 21). HScore has 9 variables, including known underlying immunosuppression, high temperature, organomegaly, triglyceride, ferritin, serum glutamic oxaloacetic transaminase, fibrinogen levels, cytopenia, hemophagocytosis features on bone marrow aspirate. HScore also excludes the use of NK activity and soluble CD25. Although the utility and generalizability of the minimal parameter set and the HScore need further investigation and validation, the promising results suggested that alternative diagnostic tools for HLH may be developed in the near future. However, these two criteria still consist a variety of parameters, including clinical, biologic, and cytologic variables. The question of when to suspect HLH and take the diagnostic tests remained.

This study used a data-driven approach to develop the screen model for EBV-HLH. Candidate model parameters were selected from the most frequently checked laboratory parameters. The purpose of this approach was that the screening results could be obtained during routine lab tests, which would assist in the first step of identifying suspected HLH among pediatric patients with EBV. The screening score included five parameters, including three lineages of cytopenias, albumin, and LDH. Cytopenias in HLH can be explained by direct hemophagocytosis, active disseminated intravascular coagulation, and high concentrations of tumor necrosis factor-α and interferon–γ, which stimulated the process of consumptive micropinocytosis (22, 23). Decreased albumin and elevated LDH can attributed to the organ dysfunction in HLH. Cytopenias were also parameters from the HLH-2004 diagnostic criteria, and this feature was developed in up to 100% of patients with HLH (13). Since cytopenias were typical features of HLH, we investigated the performance of ≥2 or 3 lineages of cytopenias as screening tests for EBV-HLH. Our results showed that these screening methods had high specificities (up to 99.4%), however, with low sensitivities (range from 25.7% to 59.5%), suggesting cytopenias along were not satisfactory screening tools for HLH among patients with EBV. Decreased albumin and elevated LDH had been observed in both pediatric and adult HLH (24, 25), and was found to be associated with poor prognosis (25–27). In a pediatric HLH case series, 113 of 116 patients (97.4%) had decreased albumin, and 110 of 111 patients (99.1%) had elevated LDH (26). In our study, the identified cut-off value for albumin was 38.5.g/L; and 23.7% of the patients with EBV-nonHLH and 86.3% of the patients with EBV-HLH had albumin < 38.5.g/L in their first albumin test (P<0.05). The cut-off value for albumin was similar to the normal reference range in the local clinical laboratory (<35 g/L). For LDH, the cut-off value in the screening score (>637 IU/L) was higher than the normal reference range in the local clinical laboratory (>450 IU/L). In the study population, 44.43% of patients with acute EBV infection had LDH > 450; 15.09% of patients with EBV-nonHLH had LDH >637 IU/L, and 76.03% of patients with EBV-HLH had LDH >637 IU/L (P<0.05). In other words, elevated LDH was developed in nearly half of the EBV patients, but those with HLH had more sever manifestation, which leads to the higher thresholds of screening score than the local lab reference range.

Strengths of this study included it used a data-driven approach to select model parameters, which could minimize subjective bias. Dozens of variables had been found to be associated with HLH, many of them were proposed to be used in the diagnosis of HLH; to build a robust model and avoid overfitting, we used the method of AIC-optimal selection through cross-validation, which was developed to overcome the drawbacks of stepwise selection and selection over a single sample (15). The second strength was the simplicity of the screening score. With only five parameters which are frequently checked in routine practice, the score is easy to use; it could be embedded into the electronic medical record system and provide early warning for EBV-HLH.

This study has several limitations. First, it was a single center study, which could subject to selection bias. The screening score need to be validated in other populations, and the cut-off point might need to be adjusted to different populations. Second, this study only included inpatients, which also could subject to selection bias. HLH is a severe syndrome which in most cases requires hospitalization, therefore, we only included inpatients. Besides, patients who diagnosed with acute infection with EBV in the outpatient department were not followed up for the development of HLH. Since the diagnostic outcome could not be confirmed in outpatients, data from this population were excluded from this study. Therefore, according to the design of this study, the EBV-HLH score was supposed to be used in pediatric inpatients. Further studies are needed to investigate whether the score could be used to identify patients with early EBV-HLH before the need of hospitalization. Third, this study excluded patients with suspected but unconfirmed HLH and patients whose infection status of EBV could not be determined. Although the excluded population was relatively small and should had limited influence on the performance of the score, further validation studies are essential for the evaluation of the performance and generalizability of the EBV-HLH screening score. Fourth, the EBV-HLH screening score was supposed to be used in patients diagnosed with acute infection of EBV, even though the diagnose of EBV infection could be delayed. Although suspecting for EBV-HLH after the diagnosis of EBV infection was reasonable in clinical practice, it could delay the diagnosis of HLH. To promote early identification of EBV-HLH, further studies are needed to optimize the screening procedure and targeted screening population.

In conclusion, we developed a screening score for the identification of HLH among pediatric patients with acute infection of EBV. With five common laboratory parameters, the EBV-HLH score provides a simple tool to assist the identification of EBV patients who require further evaluation of HLH. Further studies are needed to evaluate the generalizability of the score and optimize the diagnose process for EBV-HLH.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by The Medical Ethics Committee of the Hunan Children’s Hospital. Written informed consent from the participants’ legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

XLi, HY, and ZX contributed to the study’s conception, analysed and interpreted the data, and wrote the manuscript. TL, LX, YY, LG, ZT, JH, XZ, and MZ contributed to the study design, interpreted the data, and revised the manuscript. ZY, PZ, and DZ performed chart reviews, interpreted the data, and revised the manuscript. XLu designed the study, interpreted the data, and revised the manuscript. All authors read and approved the final manuscript. The corresponding author (XLu) had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors approved the final version of the manuscript.

This study was supported by the Hunan Provincial Science and Technology Department Project (2020SK1014-3 and 2020SK2114, grant to XLu; 2018SK2135, grant to ZX), the Hunan Provincial Key Laboratory of Emergency Medicine for Children (No. 2018TP1028, grant to ZX), the National Natural Science Foundation of China (Young Scientists Fund, No. 82102285, grant to XLi), the Natural Science Foundation of Hunan Province (No. 2021JJ40270, grant to XLi), and the Scientific Research Project of Hunan Provincial Health Commission (No. 202112050360, grant to XLi). The study sponsors have no role in the study design, data collection, data analysis, data interpretation, or writing of the report.

The authors would like to thank Dan Yang and Yu Tian for their assistance in data management.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.