“Infectious mononucleosis” flag, high-fluorescence lymphocyte percentage, and platelet-to-lymphocyte ratio as diagnostic and prognostic biomarkers for infectious mononucleosis in Chinese children

Abstract

Background

Infectious mononucleosis (IM) may lead to severe complications and present diagnostic challenges in certain clinical settings. This study aimed to preliminarily evaluate the clinical utility of novel CBC-derived graphical and numerical indicators as potential tools for rapid, accurate early diagnosis and monitoring of IM in children.

Methods

A total of 204 pediatric patients with IM, exhibiting a favorable prognosis, 109 pediatric patients diagnosed with other infectious diseases, and 86 healthy controls were enrolled from the Third Affiliated Hospital of Zhengzhou University. Multiple complete blood count (CBC)-derived indicators—including the machine learning-based “IM” flag, high-fluorescence lymphocyte percentage (HFLC%), and platelet-to-lymphocyte ratio (PLR)—were analyzed at initial diagnosis and on days 7, 14, and 21.

Results

The “IM” flag, HFLC%, and PLR were independent predictors of IM (all P < 0.01). The “IM” flag and PLR demonstrated high diagnostic efficacy across all pediatric age groups, while HFLC% showed significant diagnostic utility specifically in children over 72 months (all P < 0.001). Optimal diagnostic cutoff values were 1.95 for HFLC% and 46.35 for PLR. During follow-up, the “IM” flag gradually turned negative within 7 days (P < 0.017), HFLC% decreased significantly (all P < 0.01), whereas PLR levels showed a progressively increasing trend over 14 days (all P < 0.001).

Conclusions

The “IM” flag, HFLC%, and PLR demonstrate significant diagnostic and prognostic value in pediatric IM, supporting their potential for clinical application.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13052-025-02182-6.

Introduction

Infectious mononucleosis (IM) is an acute infectious disease caused by primary Epstein–Barr virus (EBV) infection. The incidence of IM in European and American countries predominantly affects individuals aged 10 to 30 years [1], while the disease most commonly occurs in children aged 4 to 6 years in China [2]. Although IM is typically a self-limited condition, a small proportion of cases may progress to chronic active Epstein–Barr virus (EBV) infection or EBV-associated haemophagocytic lymphohistiocytosis, or may experience delayed resolution leading to the development of chronic EBV infection, all of which are associated with an unfavorable prognosis [3, 4]. Furthermore, the diagnosis of IM can pose significant challenges in certain clinical settings. The primary diagnostic difficulties involve distinguishing EBV-induced IM from IM-like syndromes that exhibit similar clinical presentations, identifying atypical forms of IM characterized by unusual clinical features, and differentiating it from lymphoma with overlapping clinical manifestations [5, 6]. Therefore, timely and accurate diagnosis and monitoring of IM are critical for preventing the progression to severe disease and reducing misdiagnosis rates.

The complete blood count (CBC) is a simple, cost-effective, and time-efficient laboratory test that provides comprehensive information about peripheral blood cells. With ongoing advancements in instrumentation and reagent development, novel CBC-derived indicators have become increasingly available, thereby broadening the range of tools used for disease diagnosis and clinical management [7]. The novel CBC indicators consist of two distinct categories: graphical and numerical components. The updated graphical indicators are specifically designed to accurately delineate the regions occupied by distinct cell populations in scatter plots. For example, the “infectious mononucleosis” flag (“IM” flag), identified in this study, represents the first machine learning-based indicator proposed for the detection of IM. The new numerical indicators include actual measured parameters, such as the high-fluorescence lymphocyte percentage (HFLC%), and derived ratios based on mathematical formulas, including the neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), platelet-to-lymphocyte ratio (PLR), systemic immune–inflammation index (SII), systemic inflammation response index (SIRI), and monocyte-to-platelet ratio (MPR) [8, 9]. Previous studies on graphical indicators of IM have primarily focused on identifying apoptotic cell clusters in scatter plots, with limited application of artificial intelligence. The algorithms used for detecting specific cell population segmentations require further refinement and optimization [10]. Previous studies have primarily focused on initial assessments of the diagnostic performance of HFLC% and newly developed computational indicators of IM [11]. However, comprehensive evaluations of diagnostic performance across different age groups, as well as investigations into their dynamic variations, remain limited in the literature. The elevated HFLC% generated by automated instruments is primarily attributable to an increased proportion of activated lymphocytes (Als) in patients with IM. However, few studies have investigated whether the percentage of ALs, quantified by microscopic examination, can be reliably substituted by HFLC% obtained from automated hematology analyzers in pediatric patients.

Our study focuses on novel graphical indicators derived from machine learning models (the “IM” flag), HFLC%, and newly calculated CBC, aiming to evaluate their diagnostic and prognostic value in IM, explore potential associations between inflammatory levels and IM, and preliminarily identify early and effective biomarkers for clinical application.

Materials and methods

Study subjects

This cross-sectional study was conducted at the Third Affiliated Hospital of Zhengzhou University from January 2022 to December 2024. The experimental group consisted of 204 children diagnosed with IM, and the control group included 195 children, 109 of whom had been diagnosed with other infectious diseases and 86 of whom were healthy individuals who underwent routine physical examinations. The inclusion criteria for the IM group were as follows: (1) met the diagnostic criteria for IM [12], (2) did not receive any out-of-hospital treatment before the initial diagnosis, (3) did not have any complications and/or a poor prognosis related to IM following standard treatment, (4) were aged ≤ 14 years, and (5) had relatively comprehensive clinical data. Patients with other concurrent types of infections, malignant tumors, or congenital anomalies were excluded.

The complete blood count (CBC)

All venous blood samples were collected from peripheral blood in ethylenediaminetetraacetic acid (EDTA) anticoagulant tubes. CBC was conducted within 4 h of sample arrival at the laboratory by the Mindray 7500[NR]CRP hematology analyzer (Mindray, Shenzhen, China). The CBC comprises a range of parameters, including white blood cell (WBC) count, neutrophil count, lymphocyte count, monocyte count, platelet count, and other relevant hematological components. This instrument performs white blood cell (WBC) differential counts by analyzing side and forward laser scatter (side scatter [SSC] and forward scatter [FSC], respectively) as well as side fluorescence (SFL) in the DIFF channel.

Analysis of a novel CBC indicator, HFLC%

During the CBC, the Mindray BC-7500CRP hematology analyzer (Mindray, Shenzhen, China) measured the HFLC% parameter, which reflects the proportion of lymphocytes with elevated nucleic acid fluorescence intensity as detected by the side fluorescence (SFL) channel.

Analysis of a novel CBC indicator, the “IM” flag

During the CBC, the Mindray BC-7500CRP hematology analyzer (Mindray, Shenzhen, China) generated an “IM” flag by applying machine learning algorithms. The “IM” flag validated in this study was generated by a machine learning model that analyzes 41 hematological parameters. This model was co-developed by the Children’s Hospital of Zhejiang University School of Medicine and Mindray. It operated based on detecting the distinctive “rocket-shaped” distribution pattern of ALs in the DIFF scattergram observed in patients with IM, in conjunction with additional characteristics of relevant cell populations (see Fig. 1). The development of this model followed the rigorous TRIPOD guidelines for machine learning. A large-scale retrospective dataset was used and randomly divided into a training set (2,205 patients), a validation set (2,204 patients), and a test set (1,102 patients) in a 4:4:2 ratio. The most discriminative combination of features was identified through recursive feature elimination with cross-validation. After a systematic comparison of multiple algorithms, the random forest classifier was selected as the optimal algorithm. To ensure robustness and generalizability, several measures were implemented during training—specifically, addressing class imbalance, performing hyperparameter tuning via 5-fold cross-validation, and conducting final evaluation on an independent test set—to mitigate the risk of overfitting. The detailed construction and preliminary validation of this model have been elaborated in a separate article [13].

Fig. 1.

Lymphocyte cluster morphologies in the DIFF three-dimensional scatter plot. Cellular clusters identification: neutrophils (azure); lymphocytes (green); monocytes (violet); eosinophils (red); noise (blue). a. IM group, which shows a “rocket-shaped” distribution pattern of lymphocyte clusters in the DIFF three-dimensional scatter plot; b. other infectious diseases group; c. healthy control group

Computation of inflammatory indices derived from CBC

The values of these inflammation-related indices were calculated using the standardized formulas outlined below, based on CBC data:

NLR = neutrophil count/lymphocyte count;

PLR = platelet count/lymphocyte count;

MLR = monocyte count/lymphocyte count;

SII = platelet count × neutrophil count / lymphocyte count;

SIRI = neutrophil count × monocyte count / lymphocyte count;

MPR = mean platelet volume/platelet count.

Quantification of the proportion of ALs

The percentage of ALs in peripheral blood smears stained with Wright-Giemsa stain (BaSo, Zhuhai, China) was determined by optical microscopy (Olympus, Tokyo, Japan).

Statistical analysis

All the data were analyzed using SPSS 26.0 (IBM Corp., Armonk, NY, USA), GraphPad Prism 9.0.0 (GraphPad Software, San Diego, CA, USA), and MedCalc 11.4.2.0 (MedCalc Software Ltd., Ostend, Belgium). Normally distributed continuous variables are expressed as the means ± standard deviations (SDs), whereas nonnormally distributed variables are presented as medians with interquartile ranges [Q25, Q75]. Independent-sample t tests and Mann–Whitney U tests were used to compare two independent groups, whereas one-way analysis of variance (ANOVA) and the Kruskal–Wallis test were employed for comparisons among multiple groups. Categorical variables are presented as n (%), and intergroup differences were evaluated using the chi-square test. Binary logistic regression analysis was performed to evaluate the independent predictive value. A receiver operating characteristic (ROC) curve was constructed by plotting sensitivity against 1-specificity, and the area under the ROC curve (AUC) was calculated. Optimal cut-off values were determined using Youden’s J statistic. Comparison of AUCs was performed using ROC curve analysis, with the significance level for multiple pairwise comparisons was adjusted via the Bonferroni method. The consistency of the continuous variables was assessed using linear regression, Passing–Bablok regression, and Bland–Altman difference plots. Statistical significance was defined as a P value < 0.05.

Results

Clinical presentation and epidemiological insights

In this study, 204 paediatric patients with IM were evaluated. The detailed clinical characteristics are summarized in Table 1. The majority of patients presented with fever (89.22%), angina (87.25%), and cervical lymphadenopathy (92.16%). Approximately half of the cohort exhibited hepatomegaly (50.98%) and splenomegaly (55.88%), suggesting significant systemic involvement. However, only a minority of patients had a rash (6.86%).

Table 1.

Clinical manifestations of pediatric patients with IM

Clinical presentation	Positive rate
Fever	89.22%(182/204)
Angina	87.25%(178/204)
Purulent discharge on the tonsils	50.98%(104/204)
Cervical Lymphadenopathy	92.16%(188/204)
Hepatomegaly	50.98%(104/204)
Splenomegaly	55.88%(114/204)
Eyelid Edema	41.18%(84/204)
Rash	6.86%(14/204)
Nasal obstruction	22.54%(46/204)

Characteristics of novel CBC graphical and numerical indicators from the study subjects

There were no statistically significant differences in age or sex between the two groups (P > 0.05). Patients in the IM group showed a significantly higher positive rate of the “IM” flag, as well as increased levels of HFLC% and MPR, compared to the control group. Conversely, the IM group showed significantly lower levels of NLR, MLR, SII, SIRI, and PLR compared to those in the control group (P < 0.001; see Table 2).

Table 2.

The characteristics of novel CBC graphical and numerical indicators from the study subjects

Characteristics	IM patients (n = 204)	Control (n = 195)	p-value
Gender
Male: female	104: 100	99:96	0.957
Age (months)
Median age (P25, P75)	46(34,69)	48(27,78)	0.656
Novel CBC Indicators
The positive rate of the “IM” flag	94.60% (192/204)	11.46% (22/195)	<0.001
HFLC%	9.00(6.00,13.10)	0.70(0.30,1.90)	<0.001
NLR	0.32(0.19,0.51)	0.88(0.48,1.64)	<0.001
MLR	0.05(0.03,0.08)	0.14(0.10,0.23)	<0.001
SII	63.51(35.99,99.96)	253.95(129.89,448.50)	<0.001
SIRI	0.14(0.06,0.24)	0.39(0.18,0.81)	<0.001
PLR	21.41 (15.69, 29.12)	84.07(63.94, 114.63)	<0.001
MPR	0.05(0.04,0.06)	0.03(0.02,0.04)	<0.001

The “IM” flag, the “Infectious Mononucleosis” flag; HFLC%, high-fluorescence lymphocyte percentage; NLR, neutrophil-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; SII, systemic immune inflammation index; SIRI, systemic inflammation response index; PLR, platelet-to-lymphocyte ratio; MPR, mean platelet volume-to-platelet ratio

Binary logistic regression analysis of novel CBC graphical and numerical indicators

A binary logistic regression analysis was performed with the patient’s disease status as the dependent variable (IM patients were assigned a value of 1; the control group was assigned a value of 0) and novel CBC indicators as independent variables. The collinearity analysis conducted on the novel CBC indicators demonstrated statistically significant differences between the IM group and the control group. The results revealed that the variance inflation factors (VIFs) of the SII and SIRI were 20.513 and 15.315, respectively, whereas their corresponding tolerances were 0.075 and 0.092, both of which were close to zero (see Supplementary Fig. S1). Given the potential multicollinearity caused by the repeated use of components such as neutrophils, lymphocytes, and monocytes in the calculation formulas of SII and SIRI, these two indicators were excluded from the set of independent variables to ensure model stability. Following their removal, the VIFs of the remaining variables were all below 10, indicating acceptable multicollinearity levels. These indicators included the “IM” flag, HFLC%, NLR, MLR, PLR, and MPR. Compared with the control population, the “IM” flag (OR = 18.218, 95% CI: 3.712–89.411, P < 0.001), HFLC% (OR = 4.508, 95% CI: 1.812–11.216, P = 0.001), and PLR (OR = 0.926, 95% CI: 0.892–0.961, P < 0.001) were identified as independent risk factors for IM (see Table 3; Fig. 2).

Table 3.

A binary logistic regression model for novel CBC-derived graphical and numerical indicators

Novel CBC graphical and numerical Indicators	Regression Coefficient	Standard Error	Wald χ2 Value	P-value	OR	95% CI
						Lower limit	Upper limit
the “IM” flag	2.902	0.812	12.787	<0.001	18.218	3.712	89.411
HFLC%	1.506	0.465	10.487	0.001	4.508	1.812	11.216
NLR	-0.010	0.057	0.029	0.864	0.990	0.886	1.107
MLR	-6.683	6.560	1.038	0.308	0.001	0.000	479.859
PLR	-0.077	0.019	16.213	<0.001	0.926	0.892	0.961
MPR	-3.135	8.360	0.141	0.708	0.044	0.000	568023.604
Constant	1.088	1.236	0.775	0.379	2.970

The “IM” flag, the “infectious mononucleosis” flag; HFLC%, high-fluorescence lymphocyte percentage; NLR, neutrophil-to-lymphocyte ratio; MLR, monocyte-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; MPR, mean platelet volume-to-platelet ratio

Fig. 2.

Tree diagram: correlations between the “IM” flag, HFLC%, PLR, and IM. The “IM” flag, the “infectious mononucleosis” flag; HFLC%, high-fluorescence lymphocyte percentage; PLR, platelet-to-lymphocyte ratio

Diagnostic performance of novel CBC graphical and numerical indicators in children with IM

To evaluate the utility of the novel CBC indicators in classifying patients with IM, ROC curve analysis was performed. The AUC provides an estimation of the probability of correctly classifying a randomly selected subject. An AUC value between 0.7 and 0.9 indicates moderate diagnostic accuracy, while a value above 0.9 reflects high diagnostic accuracy. Therefore, the diagnostic performance of SII and SIRI was at best moderate. Given that the primary aim of this study was to identify novel biomarkers with high diagnostic utility for pediatric IM, excluding these two indices before conducting multivariate logistic regression analysis was unlikely to have a substantial impact on the overall findings (see Supplementary Table S1). Both the “IM” flag and the PLR demonstrated high diagnostic accuracy [14]. The “IM” flag showed a sensitivity of 94.60% and a specificity of 89.40% (P < 0.001), whereas for PLR, at an optimal cutoff value of 46.35, sensitivity reached 91.20% and specificity reached 89.60% (P < 0.001). The HFLC% showed moderate diagnostic performance, with a sensitivity and specificity of 87.20% and 83.10%, respectively, at a cut-off value of 3.86 (P < 0.001; see Table 4).

Table 4.

Diagnostic performance of novel CBC indicators and conventional diagnostic indicators in children with IM

Indicators	AUC	95%CI	Sensitivity	Specificity	Cut off value	P-value
The “IM” flag	0.920	0.883 ~ 0.957	0.946	0.894	-	<0.001
HFLC%	0.888	0.845 ~ 0.930	0.872	0.831	3.86	<0.001
PLR	0.944	0.914 ~ 0.974	0.912	0.896	46.35	<0.001
ALs%	0.993	0.987 ~ 1.000	0.732	0.992	9.50	<0.001
lymphocyte percentage(%)	0.848	0.803 ~ 0.893	0.951	0.564	50.15	<0.001
lymphocyte count(109/L)	0.923	0.864 ~ 0.941	0.853	0.828	5.065	<0.001
EBV-DNA (copies/mL)	0.953	0.935 ~ 0.978	0.977	0.922	2500	<0.001

The “IM” flag, the “infectious mononucleosis” flag; HFLC%, high-fluorescence lymphocyte percentage; PLR, platelet-to-lymphocyte ratio; ALs: activated lymphocytes -: no result

Comparison of the diagnostic performance between novel CBC indicators and conventional diagnostic indicators in children with IM

The traditional quantitative diagnostic indicators for IM primarily include AL%, lymphocyte percentage, lymphocyte count, and EBV-DNA load. Among these, AL%, lymphocyte count, and EBV-DNA load demonstrated high diagnostic performance, whereas lymphocyte percentage exhibited only moderate diagnostic utility. To assess their relative discriminatory accuracy, pairwise comparisons were conducted between each of these three established markers and two novel candidate indicators—The “IM” flag and PLR—using a significance threshold of P < 0.0083. The results revealed that AL% achieved a significantly higher AUC than both the “IM” flag and PLR (Z = -3.751, P < 0.001; Z = -3.132, P = 0.002), with higher specificity, although its sensitivity was suboptimal at 73.2%. For lymphocyte count, no statistically significant difference in AUC was observed compared to either the “IM” flag or PLR (Z = 0.054, P = 0.957; Z = 0.841, P = 0.406), and both its sensitivity and specificity were lower than those of the two new indicators. In contrast, although the AUC of EBV-DNA load did not differ significantly from that of the “IM” flag or PLR (Z = 1.51, P = 0.132; Z = 0.482, P = 0.631), it demonstrated significantly higher sensitivity and specificity (P < 0.001, see Table 4).

Diagnostic performance of novel CBC graphical and numerical indicators by age in children with IM

All IM patients were classified into two groups according to age: those ≤ 72 months and those > 72 months. The sensitivity, specificity, cut-off value, and AUC comparisons are summarized in Table 5. The AUCs of the “IM” flag, which demonstrated high diagnostic value, were significantly different between the two groups (P < 0.001). In the > 72 months group, this indicator showed superior diagnostic performance compared to the ≤ 72 months group, with both very high sensitivity and specificity. Furthermore, the AUCs of PLR, which also had high diagnostic value, were not significantly different between the two groups (P > 0.05). The AUC for HFLC% reached 0.971 in the > 72 months group, with both the sensitivity and specificity exceeding 90.00% at a cut-off value of 1.95 (P < 0.001).

Table 5.

Diagnostic performance of novel CBC graphical and numerical indicators in children with IM, stratified by age

Novel CBC graphical and numerical indicators	AUC	95%CI	Sensitivity	Specificity	Cut off value	P-value of the ROC curve	AUC difference
							Z Value	P-value
The “IM” flag
≤ 72 months	0.919	0.877 ~ 0.961	0.962	0.876	-	<0.001	-6.191	<0.001
>72 months	0.955	0.905 ~ 1.000	0.939	0.972	-	<0.001
HFLC%
≤ 72 months	0.855	0.793 ~ 0.917	0.925	0.806	3.90	<0.001	-3.912	<0.001
>72 months	0.971	0.935 ~ 0.994	0.981	0.912	1.95	<0.001
PLR
≤ 72 months	0.953	0.912 ~ 0.994	0.962	0.910	38.17	<0.001	-1.855	0.064
>72 months	0.981	0.955 ~ 1.000	0.959	0.943	67.62	<0.001

The “IM” flag, the “infectious mononucleosis” flag; HFLC%, high-fluorescence lymphocyte percentage; PLR, platelet-to-lymphocyte ratio; -: no result

Consistency test between AL% and HFLC%

The HFLC% is generated based on ALs in peripheral blood. Therefore, it is crucial to assess the consistency between the HFLC% obtained from a CBC analyzer and the AL% determined by microscopic examination. Cusum tests for linearity (P = 0.99) indicated that Passing–Bablok regression analysis was suitable for assessing the agreement between the two parameters. In the Passing–Bablok regression analysis, the intercept was − 3.34, with a 95% confidence interval ranging from − 0.92 to 0.08, which included 0. The slope was 1.14, with a 95% confidence interval ranging from 0.83 to 1.65, which included 1.00. These results suggest that there was no systematic or proportional bias between the two methods. However, linear regression analysis revealed an R² value of only 0.065 (P = 0.002), substantially below the acceptable threshold of 0.95 [15], indicating a weak correlation between HFLC% and AL%. Additionally, Bland–Altman analysis showed a mean difference of -2.56 between the two methods, with a 95% confidence interval ranging from − 3.68 to -1.43, which does not include zero and is statistically significant (P < 0.001). The combination of a weak correlation and a mean difference significantly different from zero in the Bland–Altman analysis demonstrates a lack of agreement between HFLC% and AL% (see Fig. 3).

Fig. 3.

Correlation between HFLC% and AL%. a. Passing–Bablok linear regression; b. Bland–Altman difference plot. HFLC%, high-fluorescence lymphocyte percentage; ALs: activated lymphocytes

Dynamic changes in novel CBC graphical and numerical indicators throughout treatment in children with IM

Patients in the IM group were categorized into four distinct subgroups based on the course of the disease: 0-day, 7-day, 14-day, and 21-day groups (see Fig. 4). The positive rate of the “IM” flag was significantly lower in the 7-day group compared to that in the 0-day group (P = 0.002). In contrast, no statistically significant differences were found between the 7-day and 14-day groups or between the 14-day and 21-day groups (P > 0.017). From Day 0 to Day 14, the HFLC% progressively decreased (all P < 0.01), whereas the PLR gradually increased (all P < 0.001). In contrast, no statistically significant differences were found between the 14-day and 21-day groups (P > 0.017).

Fig. 4.

Bar chart illustrating novel CBC indicators across different stages of the treatment course. a. the positive rate of the “IM” flag; b. HFLC%, high-fluorescence lymphocyte percentage; c. PLR, platelet-to-lymphocyte ratio

Comparison of novel CBC graphical and numerical indicators among children with different hospitalization durations

The median hospitalization duration of 204 children with IM was 24 days. Based on this median value, patients were stratified into two groups: those with a hospitalization duration of less than 24 days and those with a duration of 24 days or longer. A statistically significant difference in PLR was observed between the two groups (P < 0.001), whereas no significant differences were observed in the positive rate of the “IM” flag or HFLC% (P > 0.05; see Fig. 5).

Fig. 5.

Bar chart illustrating novel CBC indicators across different hospitalization durations. a. the positive rate of the “IM” flag; b. HFLC%, high-fluorescence lymphocyte percentage; c. PLR, platelet-to-lymphocyte ratio

Discussion

Most individuals remain asymptomatic carriers of EBV after contracting it during childhood. When various external factors disrupt the balance between EBV and the host, the virus can exert its pathogenic potential, leading to the development of clinical manifestations of IM, such as fever, hepatosplenomegaly, and lymphadenopathy [16]. Following primary infection, EBV establishes a productive infection within B cells; subsequent B lymphocyte activation triggers the host immune system to produce EBV-specific cytotoxic T lymphocytes, which are critical for viral clearance [17]. During this immune response, cytotoxic T lymphocytes undergo morphological and functional transformation into ALs, characterized by increased nucleic acid synthesis and blast-like morphological changes that indicate cellular activation [18]. This transformation ultimately enhances the host’s capacity to combat viral infection effectively [19]. Based on the immune mechanisms and the principles of the blood cell analyzer, this study explored the diagnostic and prognostic significance of novel CBC graphical and numerical indicators in children across different age groups with IM.

HFLCs represent lymphocytes exhibiting elevated nucleic acid fluorescence intensity, as detected by the SFL channel in the CBC scatter plot, without the need for additional reagents or costs [20]. In this study, HFLC% demonstrated strong diagnostic accuracy for IM, with enhanced specificity and sensitivity observed in children older than 72 months, a finding consistent with that reported in adults with IM [21]. In contrast, the diagnostic performance of HFLC% was only moderate in children under 72 months, showing reduced specificity. This discrepancy may be attributed to age-related physiological variations in immune response mechanisms [22]. Therefore, HFLC% may serve as a more reliable biomarker for diagnosing IM in children older than 72 months. IM is an immunopathological disease in which the clinical manifestations are predominantly mediated by an excessive immune response to EBV infection [23]. Following standardized therapeutic interventions, the immune response gradually stabilizes, resulting in the alleviation of clinical symptoms in most pediatric patients. Concurrently, a decrease in the percentage of ALs has been documented [24]. Given these findings, it is clinically significant to explore whether HFLC% shows a corresponding reduction. Richa Juneja et al. reported that a dynamic decrease in HFLC% during the first 10 days following dengue infection onset may have potential clinical utility in monitoring platelet count recovery [25]. However, this finding was based on a small sample size of only 15 patients. In contrast, our study analyzed a larger cohort of 204 pediatric patients with IM and demonstrated that HFLC% progressively declined during the first 14 days after disease onset, suggesting its potential as a dynamic prognostic biomarker. Notably, a reduced HFLC% was found to correlate with the approaching clinical recovery from IM.

Microscopic classification of blood cells is a time-consuming and labor-intensive process that is inherently prone to subjective interpretation. Previous studies on IM have primarily focused on evaluating the concordance between hematology analyzers and microscopic methods for detecting apoptotic cells [26]. However, the presence of apoptotic cells is not specific to IM, as similar phenomena are also frequently observed in other pathological conditions, including hepatitis [27], AIDS [28], and chronic lymphocytic leukemia [29]. Can the blood cell analyzer-based method, specifically HFLC%, serve as a reliable and effective alternative to the microscopic method for determining AL% in pediatric IM patients? The results of this study demonstrated a weak correlation between the two methods, and Bland–Altman analysis revealed a mean difference significantly different from zero, indicating a lack of agreement between them. This finding is consistent with previous reports [30]. In this study, nine patients exhibited an AL% exceeding 10%, while the corresponding HFLC% levels were either abnormally low or undetectable. This discrepancy may be attributed to two potential factors: first, the DIFF scatter plot revealed significant abnormalities that impaired the instrument’s capacity to quantify HFLC%; second, the ALs region partially overlapped with the monocyte region, which interfered with the precise measurement of HFLC% [31]. Additionally, two patients with IM exhibited elevated HFLC%, whereas the AL% remained abnormally low. This inconsistency may result from the inherent subjectivity associated with manual enumeration of ALs [32]. The replacement of microscopic analysis with automated instrumental techniques for quantifying ALs still requires further refinement and validation. Consequently, machine learning models have been introduced to enhance the accuracy of AL region identification, thereby supporting the diagnostic process of IM.

Machine learning models have been widely applied in the analysis of various diseases, including acute promyelocytic leukemia [33], red cell lysis [34], and genetic hemoglobinopathy [35]. Previous studies on IM have predominantly focused on identifying specific cell clusters in scatter plots, typically using specialized gating software, rather than leveraging artificial intelligence-based approaches. The machine learning-based indicator innovatively explored in this study is the “IM” flag, which is designed to detect AL clusters and enable the automated identification of IM-related CBC scatter plots. In this study, we present the first preliminary evidence that the “IM” flag demonstrates high diagnostic accuracy and may serve as a reliable diagnostic indicator for pediatric IM. The diagnostic performance of the “IM” flag surpasses that of HFLC% in children under 72 months of age with IM, highlighting its superior utility in this paediatric population. In older children, the sensitivity of the “IM” flag remains high, while specificity increases significantly, further supporting its robustness across age groups.

Currently, the commonly used laboratory diagnostic markers for IM in clinical practice include the proportion of ALs, lymphocyte percentage and absolute count, serological testing, and quantitative EBV-DNA detection [36]. In this study, the “IM” flag, the proportion of atypical lymphocytes, and the lymphocyte count all demonstrated high diagnostic performance. Among these three indicators, the proportion of ALs exhibited significantly superior diagnostic performance compared to the other two, while no statistically significant difference was observed between the “IM” flag and the lymphocyte count. Although the specificity of the “IM” flag was lower than that of ALs, it exceeded that of the absolute lymphocyte count; notably, its sensitivity was significantly higher than both of these markers. Moreover, compared to lymphocyte percentage, the “IM” flag not only achieved superior overall diagnostic performance but also exhibited substantially greater specificity. Our results were consistent with findings reported in the literature [37]. Serological methods are widely utilized to support IM diagnosis by detecting EBV-specific antibodies, including VCA-IgM, VCA-IgG, EBNA-1-IgG, and VCA-IgG affinity. A study by Ting Shi et al. showed that while the VCA-IgG affinity assay showed high specificity (96.33%) and acceptable sensitivity (84.08%)—values that surpassed the specificity but fell short of the sensitivity of the “IM” flag—its AUC was only 0.80. In contrast, EBV-VCA-IgM had limited sensitivity, EBV-VCA-IgG had suboptimal specificity, and EBNA-1-IgG performed poorly in both sensitivity and specificity [38]. Notably, a subset of patients with confirmed IM remain negative for VCA-IgM during acute infection [39], highlighting the risk of false-negative diagnoses when relying solely on this marker. Furthermore, EBV VCA-IgM testing may yield false-positive results, particularly due to cross-reactivity with other herpesviruses, such as cytomegalovirus [40]. Additionally, the absence of EBNA-1-IgG antibodies did not definitively exclude primary EBV infection, further complicating serological interpretation [41]. Moreover, current serological testing protocols are labor-intensive and time-consuming, which imposes significant constraints on timely clinical decision-making [42]. Therefore, in the early rapid diagnosis of pediatric infectious mononucleosis, the “IM” flag may offer advantages over EBV antibody panel testing. EBV DNA quantitative detection has emerged as a clinically valuable biomarker, with viral load levels demonstrating a strong correlation with disease activity and patient prognosis [43]. The study confirmed that EBV-DNA quantification exhibited high sensitivity and specificity, consistent with findings reported in the current literature [44]. The AUC values for both the “IM” flag and plasma EBV-DNA quantification exceeded 0.90, indicating excellent diagnostic accuracy for both methods, although the “IM” flag showed slightly lower sensitivity and specificity compared to plasma EBV-DNA quantification. Nevertheless, EBV DNA quantification alone is insufficient for reliably distinguishing among different EBV-associated diseases [45]. Moreover, the method entails relatively high costs, and critical aspects—including assay quality control, the definition of clinical intervention thresholds for viral load, standardization of quantitative units, and identification of the optimal specimen type [46]-remain inadequately standardized [47]. Therefore, compared with conventional indicators, the “IM” flag showed similar or higher sensitivity but slightly lower specificity. Given its high sensitivity, the “IM” flag is recommended as a preliminary screening marker for suspected cases of IM in Chinese children and should be incorporated into microscopic re-examination protocols during CBC analysis. When the “IM” flag is triggered, a peripheral blood smear should be performed for quantitative assessment of ALs, and confirmatory testing—including EBV-specific serology and EBV-DNA quantification—should be initiated.

Compared to the independent assessment of neutrophils, lymphocytes, monocytes, and platelets, CBC-derived calculated indices are less susceptible to confounding factors and may provide greater predictive accuracy in the evaluation of inflammatory responses, such as PLR [48, 49]. The decrease in platelet count and the increase in lymphocyte count play significant roles in the inflammatory response to IM. An increase in the number of lymphocyte subsets, combined with enhanced functional activity of these cells, constitutes a central mechanism underlying the immune response in IM [50]. Approximately one-third of patients with IM exhibit thrombocytopenia (platelet count lower than 150 × 10⁹/L) [51]. The underlying mechanism of thrombocytopenia in IM is primarily peripheral and immune-mediated. Antiplatelet antibodies and platelet-bound IgG have been identified in certain cases of thrombocytopenia associated with IM [52]. Splenic sequestration represents another mechanism, as an inverse correlation has been observed between spleen size and platelet counts [53]. In addition, EBV may interact with platelets, and impaired platelet function during the course of IM has been documented [54]. In this study, the PLR in patients with IM was significantly lower than that in the control group. The PLR exhibited high diagnostic performance, characterized by favorable sensitivity and specificity, which exceeded that of PLR in neonatal sepsis [55]. Therefore, the PLR may serve as a reliable inflammatory biomarker for the diagnosis of IM in Chinese pediatric populations. The PLR serves as an indicator of the balance between inflammation and thrombosis, and its elevation or reduction has been associated with the severity of the inflammatory response [56]. Hulya Albayrak et al. reported a decrease in the PLR following treatment in patients with psoriasis [57]. In contrast, this study revealed a progressive increase in the PLR during the initial 14 days of treatment, with elevated PLR levels potentially indicating improvement in patients with IM. The discrepancy in PLR trends may be attributed to differences in the underlying pathogenic mechanisms. Therefore, the PLR may serve as a potential biomarker for monitoring the clinical course of IM. Furthermore, to explore its clinical relevance, we analyzed the association between the PLR levels and hospitalization duration—a well-established surrogate marker for disease severity. We found that shorter hospital stays were significantly associated with higher PLR levels. The association between higher PLR levels and shorter hospitalization duration suggests that this increase may be indicative of a more favorable disease progression.

Despite the valuable insights provided by this study, several limitations must be acknowledged. First, the single-center design, modest sample size, and reliance on institution-specific hematology analyzers inherently limit the generalizability of our findings, suggesting that the performance of the novel CBC-based indicators and the proposed clinical thresholds may not be directly generalizable to other populations or more diverse clinical settings. Second, although these biomarkers are closely associated with lymphocytes, their combined diagnostic value warrants further systematic evaluation. Third, although this study investigated the association between CBC-derived indicators and hospitalization duration and characterized their dynamic trajectories across key treatment time points, the prognostic utility of these indicators in IM requires rigorous validation against well-defined, standardized clinical endpoints. Finally, the finding that the “IM” classification yielded negative results in 12 confirmed cases of IM highlights a critical need for further training and refinement of the underlying artificial intelligence algorithm to enhance its sensitivity and clinical reliability. To address these limitations, our future research will incorporate large-scale, multicenter cohorts across multiple brands of hematology analyzers to enable robust external validation and establish meaningful correlations with hard clinical endpoints, including time to normalization of liver function, duration of clinical symptoms, and complication rates. These efforts will be accompanied by continuous model optimization through iterative learning and expanded training data, ensuring progressive improvement in clinical performance and real-world applicability.

Conclusion

We conducted the first comprehensive assessment of the diagnostic and prognostic value of machine learning-based CBC scatter plots and derived parameters in pediatric patients with IM in China, and evaluated the potential of HFLC% as an alternative to traditional manual diagnostic methods. It is preliminarily recommended that the presence of the “IM” flag and a PLR value below 46.35 be used as diagnostic indicators for IM in pediatric patients, whereas an HFLC% exceeding 1.95 may serve as a diagnostic criterion specifically in children older than 72 months. When these criteria are met, prompt EBV aetiological testing and appropriate treatment for IM should be initiated. Furthermore, a negative conversion of the “IM” flag, a decrease in HFLC%, and an increase in the PLR indicate a favorable prognosis. These findings provide novel insights and practical tools for future research and clinical applications, ultimately contributing to the development of more targeted and efficacious treatment strategies for IM.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

IM

Infectious Mononucleosis

AL

Activated lymphocyte

CBC

Complete blood count

HFLC%

High-fluorescence lymphocyte percentage

EBV

Epstein–Barr virus

NLR

Neutrophil-to-lymphocyte ratio

MLR

Monocyte-to-lymphocyte ratio

PLR

Platelet-to-lymphocyte ratio

SII

Systemic immune–inflammation index

SIRI

Systemic inflammation response index

MPR

Monocyte-to-platelet ratio

PCR

Polymerase chain reaction

Author contributions

Conceptualization, Yuanyu Wei and Peng Wang; methodology, Yuanyu Wei, Kai Zhang; statistical analysis, Yuanyu Wei, Kai Zhang; data curation, Enwu Yuan; writing—original draft preparation, Yuanyu Wei; writing—review and editing, Kai Zhang and Peng Wang; funding acquisition, Enwu Yuan. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the Project Funding Joint Construction Project of Henan Medical Science and Technology Tackling Plan (Grant Number: LHGJ20200458).

Data availability

The data that support the findings of this study are available from the corresponding authors (Yuanyu Wei, weiyuanyu2019@163.com) upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of the Third Affiliated Hospital of Zhengzhou University (ethics batch number: 2022-156-01). Written informed consent was obtained from the parents or legal guardians of all participants by the researchers. All methods were carried out following the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

Informed consent

Informed consent was obtained from all subjects involved in the study.

Additional information

Correspondence and requests for materials should be addressed to Yuanyu Wei.

Competing interests

The authors declare no competing interests.