The IL‐10/IL‐6 Ratio and the Risk Score: Two Cytokines‐Based Predictors for Malignancy‐Associated Hemophagocytic Lymphohistiocytosis in Adults (M‐HLHa)

ABSTRACT

The predictive value of cytokines (CK) for malignancy‐associated adult hemophagocytic lymphohistiocytosis (M‐HLHa) remains uncertain. We evaluated a cytokine‐based Risk Score (RS) and the IL‐10/IL‐6 Ratio to predict M‐HLHa. Adult patients (n = 112) from the French HLH cohort (NCT02113917) with complete data for nine key HLH related CK measured by Luminex were first analyzed. Logistic regression was performed, and a RS was subsequently derived and internally validated. In a post hoc analysis, the IL‐10/IL‐6 ratio obtained using the ELLA cytokine assay was evaluated in 75 patients, 54 of whom also had Luminex testing, plus 25 additional patients from the cohort. The RS and IL‐10/IL‐6 ratio were then jointly assessed in the 54 patients with results from both platforms. Among the 112 patients, 45 had M‐HLHa and 67 non M‐HLHa; median age was 48 years and 64 (57%) were male. Eight variables were associated with M‐HLHa, of which four were retained in the logistic model: age > 48 years (2.9[1.12–7.33], p = 0.03), TNF‐α ≤ 43 pg/mL (3.2[1.16–8.61], p = 0.02), IL‐18 > 574 pg/mL (4.9[1.89–12.75], p = 0.001), and IL‐10/IL‐6 ≥ 1.5 (2.7[1.01–7.31], p = 0.04), with respective weights of 11, 12, 16, and 10. A RS ≥ 20 increased the odds of M‐HLHa 17‐fold (17.2 [5.4–54.5], p < 0.0001). The IL‐10/IL‐6 ratio alone (n = 75) showed good performance (AUC 0.83). In the 54 patients with both assays, the RS and IL‐10/IL‐6 ratio preserved their diagnostic performance (AUC 0.87 and 0.79, respectively). These data support the RS and IL‐10/IL‐6 Ratio as useful tools to improve the identification of M‐HLHa in adults.

Trial Registration: clinicaltrials.gov NCT02113917

Abbreviations

AD‐HLH

HLHa associated diseases

HLHa

hemophagocytic lymphohistiocytosis in adults

HS

hemophagocytic syndrome

MAS

macrophage activation syndrome

M‐HLHa

malignancy associated HLHa

M‐Risk Score

Risk Score associated to malignancy HLH

1. Introduction

Hemophagocytic lymphohistiocytosis (HLH) is a severe condition characterized by dysregulated activation of CD8 T cells, NK cells, and macrophages, resulting in excessive cytokine production and a cytokine storm syndrome. The HLH syndrome [1, 2] is clinically characterized by fever, hepatosplenomegaly, hyperferritinemia, hypofibrinogenemia, and pancytopenia. The key features of the HLH syndrome may result from the effects of excessive CD8 T cells and macrophage cell proliferation, in addition to damage inflicted by HLH‐related cytokines (CK) on tissues and biological processes [3, 4].

HLH is usually divided into primary such as familial (F‐HLH) cases or secondary/acquired cases. F‐HLH is a severe disease, occurring in very young children, in which genetic defects affecting various immune functions, particularly the perforin‐dependent cytotoxic pathway [5, 6], have been identified. Secondary or acquired HLH occurs mainly in adults (HLHa) [7]. It is associated either with malignancies (M‐HLHa), such as T‐ or B‐cell lymphoma accounting for 40%–70% of cases [8], or with autoinflammatory/immune diseases in which HLH is designated as macrophage activation syndrome (MAS) [9, 10]. When no associated disease (AD) is identified, HLH is considered idiopathic. Despite severe forms of HLHa having been linked to minor defects in genes involved in F‐HLH [11], HLHa is still considered sporadic. Regardless of the AD, HLH onset is often triggered by infectious events such as herpesviruses (EBV, CMV) or intracellular bacteria [11].

In secondary forms, HLH is often the presenting feature and thus diagnosing the AD, especially M‐HLHa, is crucial for adult management as it directly influences therapeutic decisions aimed at reducing mortality. M‐HLHa has the poorest prognosis, with a 5‐year overall survival rate ranging from 5% to 30% [12, 13]. However, the diagnosis of HLH and its ADs is often confounded by the nonspecific and overlapping symptoms. Moreover, biopsy procedures must be decided with caution due to the frequent coagulopathies observed in severe HLH cases. Therefore, there is a strong need for rapid and noninvasive blood tests to help clinicians assess the likelihood of malignancy‐associated HLH (M‐HLHa) in patients presenting HLH symptoms.

In this study, we hypothesize that a specific cytokine profile can help distinguish M‐HLHa from non‐M‐HLHa (autoimmune/inflammatory or idiopathic diseases cases). We propose a cytokine‐based Risk Score (RS) and Il‐10/IL‐6 ratio to support clinicians in the diagnosis of HLHa AD and thereby facilitate efficient management of malignancy in HLH adult patients.

2. Material and Methods

2.1. Study Design and Population

Between 2012 and 2019, 195 patients (178 adults, 6 adolescents, and 11 children ages 3–11 years old) with secondary HLH were prospectively recruited from 62 medical centers in a national study entitled “the French HLH‐GENE Cohort” (Funded by the French National Program of Clinical Research, 2010). The inclusion criteria were adapted from the HLH‐2004 diagnostic criteria [14] and with the expertise of the French HLH Study Group, as recommended by La Rosée et al. [15]. Figure S1. Informed consent was obtained from all study participants in accordance with the Declaration of Helsinki. The regional Ethics Committee of Paris (n° ID‐RCB: 2009‐AO00301‐56) approved the study.

2.2. Data Collection

The data collected includes the clinical features, infectious trigger events, treatment and outcomes (refractoriness to treatment or relapse, intensive care unit (ICU) transfer and death) of HLH and the ADs at disease onset and during a 1‐year follow‐up period. In addition, serum samples were obtained from each patient prior to HLH treatment. Selected associated diseases included neoplastic diseases (hematologic malignancies, solid tumors) and autoimmune/inflammatory‐associated diseases (AIDs). The disease was classified as idiopathic if there was no history or occurrence of neoplasia and/or AIDs during the one‐year follow‐up period. The diagnosis of AIDs was based on the American College of Rheumatology (ACR) criteria for systemic lupus erythematosus [16] (SLE) or on Yamaguchi's criteria for Still's disease [17]. Bacterial, viral, parasitic, or miscellaneous infectious events were considered potential triggers of HLH onset and were documented for each AD. Treatment of HLH was recorded, particularly the use of corticosteroids and etoposide.

2.3. Patients Selection

For 128 patients prospectively included in the French HLH cohort, serum levels of 32 CK were analyzed by Luminex CK panel (Bio‐plex human Cytokines Group I 27‐plex Panel and Pro‐Hu Cytokine Group II 3‐plex: IFN‐α2, IL‐18, and Trail). Patients (n = 112) having criteria of adult HLH (aged ≥ 18 years with an “HScore” ≥ 169) [18] were retained for the RS analysis, Figure 1. Among the 32 CK, nine key mediators driving F‐HLH or MAS [19, 20], including interleukin 1‐β (IL1‐β), interleukin 8 (IL‐8), interleukin 18 (IL‐18), interleukin 10 (IL‐10), interleukin 6 (IL‐6), interferon‐γ (IFN‐γ), tumor necrosis factor α (TNF‐α), interferon γ‐induced protein 10 (IP‐10), and soluble interleukin‐2 receptor‐α (sIL2‐R‐α), were selected for the study.

FIGURE 1.

Flow‐chart of the adult HLH‐related cytokines cohort. A total of 195 patients were included in the HLH‐Genes cohort, of whom 128 had serum cytokine levels measured by Luminex method. After applying the selection criteria of age ≥ 18 years and H‐Score ≥ 169, 112 patients were retained for the Risk Score analysis. In a post Hoc analysis, the IL‐10/IL‐6 ratio study was conducted in 75 patients using the ELLA cytokine test. [Color figure can be viewed at wileyonlinelibrary.com]

During the study period, the IL‐10/IL‐6 ratio began to be routinely measured in clinical practice using a faster assay method (Ella cytokine test) as a diagnostic indicator of vitreoretinal lymphoma and primary central nervous system large B‐cell lymphoma [21]. We chose to focus on the IL‐10/IL‐6 ratio rather than on IL‐10 and IL‐6 levels independently and hypothesized the IL‐10/IL‐6 ratio as being another parameter to be significantly different between M‐HLH and non M‐HLH cases. Therefore, we conducted a post hoc analysis to the RS analysis to assess the performance of this IL‐10/IL‐6 ratio evaluated by this second method of cytokine quantification. Sera from 75 patients were assessed by Ella cytokine test, 54 of which were from the original cohort of 112 patients and that had previously been tested using the Luminex method, and 21 sera from new patients fulfilling inclusion criteria for the RS study (Figure 1).

2.4. ELLA Cytokine Test

The ELLA platform is a rapid cytokine quantification system based on three single plex microfluidics ELISA assays run in triplicate within cartridges following the manufacturer's instructions (Bio‐Techne, San Jose, CA, USA). Analytical validation was performed using both reference cytokine controls and biological replicates across different lots of cartridges. The assay showed over 92% reproducibility, with an intra‐assay and inter‐assay coefficient of variation of below 10% for both IL‐6 and IL‐10. The limits of quantifications were 0.6 pg/mL for the IL‐6 and 0.8 pg/mL for the IL‐10.

2.5. Statistical Analysis

The data collected were described using the number and the percentage (%) for the qualitative variables. In other cases, the median and interquartile range (IQR) were used for the quantitative variables. Univariate tests (Fisher's exact test or χ ² test for qualitative variables and t‐test or Mann–Whitney U test for quantitative variables) were used to compare the characteristics between M‐HLHa and other forms (autoimmune or idiopathic) of HLH.

Logistic regression was used to develop and validate the RS predicting M‐HLHa. The selection of the variables to include was made by the stepwise method using the p‐value. The threshold p value for enter and leave from the model were set at ≤ 0.30 and ≥ 0.05, respectively. Quantitative covariates were dichotomized at their median values. This approach was chosen because several variables violated the linearity assumption in the logistic regression model, and threshold values suggested by generalized additive model (GAM) plots led to imbalanced group sizes. Using the median as a uniform cut‐off helped maintain balanced groups.

Model calibration and discrimination were each assessed using the Hosmer‐Lemeshow and receiver operating characteristic (ROC) curves, respectively. A 10‐fold cross‐validation was carried out to consider overfitting bias. A predictive RS was then derived from the final logistic model by assigning weights to selected covariates based on their regression coefficients. The optimal cut‐off for the score was determined by maximizing the Youden index (sensitivity + specificity−1), allowing classification of patients into low‐ and high‐risk categories for M‐HLHa (score ≥ or < threshold).

The diagnostic performance of the IL‐10/IL‐6 ratio was evaluated through ROC curve analysis, with sensitivity, specificity, and AUC reported. Additionally, in a subset of 54 patients with cytokine measurements available from both the Luminex and ELLA platforms, we separately assessed and compared the diagnostic accuracy of the predictive RS and the IL‐10/IL‐6 ratio.

Multiple imputation chain equation (MICE) was performed to account for the missingness in covariates. The number of multiple imputations was set to 12 with 10 iterations. Twenty‐three covariates were used, namely age, gender, lymphocyte, neutropenia, ferritin, ASAT, ALAT, hemoglobin, platelet, fibrinogen, CD8/CD25, interleukin 1‐β (IL1‐β), interleukin 8 (IL‐8), interleukin 18 (IL‐18), interleukin 10/interleukin 6 (IL‐10/IL‐6), interferon‐γ (IFN‐γ), tumor necrosis factor α (TNF‐α), interferon γ‐induced protein 10 (IP‐10), soluble interleukin‐2 receptor‐α (sIL2‐R‐α), severity, infectious diseases, CD4/CD8 and M‐HLHa or non M‐HLH (autoimmune/inflammatory diseases or idiopathic form). All following analyses were carried out based on the imputed datasets by combining estimates from each completed dataset using Rubin's rules [22]. p value < 0.05 was considered statistically significant. Statistical analyses were performed using R Project for Statistical Computing, Version 3.6.1 software (The R Foundation for Statistical Computing, Vienna, Austria, http://www.r‐project.org/).

3. Results

3.1. Study Population Characteristics

The RS analysis included 112 adult patients selected on H‐Score criteria (“HScore” ≥ 169). None had a familial history of HLH. The median [IQR] age of patients at diagnosis was 48 years [36–61], and the sex ratio (M/F) was 1.3 (64/48). The majority of patients (65/112, 58%) had no previous disease at HLH onset. In 45 patients (40%), AD was malignancy (M‐HLHa) distributed as B cell‐Non Hodgkin Lymphoma (B‐NHL) (n = 22, EBV+, n = 9), T cell‐NHL (n = 12, EBV+, n = 7), Hodgkin lymphoma (HL) (n = 3), or other (n = 8). In the remaining patients, HLH AD included autoimmune/inflammatory (n = 29) or idiopathic forms (n = 38). Infectious events occurred in 70 patients, with predominance for virus/intracellular bacteria occurring in both M‐HLHa and non M‐HLHa (autoimmune/inflammatory), Table 1.

TABLE 1.

Cohort characteristics.

Variable Median [IQR]	n	ALL (n = 112)	Non M‐HLH (n = 67)	M‐HLH (n = 45)	p
Age, year	112	48 [36; 61]	41 [30; 55]	56 [44; 68]	< 0.001
Male, N (%)	112	64 (57%)	34 (51%)	30 (67%)	0.14
Infectious diseases, N (%)	111				0.001
None		42 (38%)	20 (30%)	22 (50%)
Herpes virus		39 (35%)	21 (31%)	18 (41%)
Intra cellulaire bacteria		16 (14%)	16 (24%)	0 (0%)
Others virus and others infectious agents		14 (13%)	11 (15%)	4 (9%)
Fibrinogen, g/L	83	3 [2; 5]	3 [2; 4]	4 [2; 5]	0.65
Total lymphocytes, /mm3	71	730 [396; 1388]	1068 [572; 1744]	600 [390; 890]	0.04
ASAT, UI/l	104	138 [58; 296]	158 [62; 332]	83 [53; 223]	0.04
ALAT, UI/l	105	81 [42; 211]	109 [54; 280]	57 [34; 127]	0.02
Hemoglobin, g/L,	111	9 [8; 11]	9 [8; 11]	8 [8; 10]	0.02
Platelets, G/L	109	82 [41; 125]	98 [64; 142]	66 [33; 104]	0.002
Neutrophile, /mm3	96	2.130 [1.201; 4.622]	2.700 [1.500; 4.900]	1.620 [903; 4.210]	0.05
Ferritin, μg/L	105	7.000 [2.144; 14.760]	6.288 [2.143; 10.578]	8482 [2.300; 18.014]	0.15
CD8/CD25+, /mm3	13	73 [58; 161]	65 [52; 161]	80 [62; 119]	0.66
IL‐1β, pg/mL	73	6 [2; 15]	7 [3; 21]	3 [1; 9]	0.20
IL‐8, pg/mL	107	169 [47; 673]	192 [61; 1063]	161 [42; 434]	0.52
IFN‐γ, pg/mL	94	160 [73; 513]	136 [73; 516]	202 [75; 445]	> 0.9
TNF‐α, pg/mL	103	43 [25; 73]	55 [36; 82]	35 [15; 44]	0.001
IL‐18, pg/mL	112	574 [123; 1859]	256 [74; 946]	1.045 [403; 2.494]	< 0.001
IP‐10, pg/mL	111	10.004 [20.54; 35.450]	6786 [1.499; 20.975]	21.514 [3.880; 64.888]	0.008
sIL2‐R‐α, pg/mL	107	36.700 [56.25; 92.100]	13.000 [3.375; 53.098]	83.400 [35.350; 189.000]	< 0.001
IL‐10, pg/μL	107	42 [16; 173]	26 [10; 48]	167 [51; 997]	< 0.001
IL‐6, pg/μL	106	46 [15; 129]	33 [10; 128]	75 [17; 133]	0.04
IL‐10/IL‐6	103	1.4 [0.6;3.9]	1.1 [0.3;2.0]	3.5 [1.1;17.9]	0.001
IL‐10/IL‐6 ratio, N (%)	103				0.002
< 1.5		54 (52.4%)	41 (65.1%)	13 (32.5%)
≥ 1.5		49 (47.6%)	22 (34.9%)	27 (67.5%)
CD4/CD8 ratio, N (%)	81				0.02
≤ 1		44 (54%)	20 (43%)	24 (71%)
> 1		37 (46%)	27 (57%)	10 (29%)

Note: Clinical and biological characteristics of the 112 patients prospectively included in the HLH‐cytokines study, whose cytokine levels were analyzed using the Luminex method.

Abbreviations: IFN‐γ, interferon‐γ; IL, interleukin; IP‐10, interferon γ‐induced protein 10; IQR, interquartile range; sIL2‐R‐α, soluble interleukin‐2 receptor‐α; TNF‐α, tumor necrosis factor α.

Among the 45 patients with malignancy, 31 (71%) patients had lymphoma revealed by HLH. Among the nine key HLH mediators, the CK IL‐1β, IL‐8, and TNF‐α were generally higher in non M‐HLHa. The median value [IQR] of IL‐10 was markedly elevated in M‐HLHa compared to non M‐HLHa (167 [51; 997] /26 [10; 48]) (p < 0.001). Moreover, the median [IQR] value of the IL‐10/IL‐6 ratio was 1.1 [0.3; 2.0] in non M‐HLHa and 3.5 [1.1; 17.9] in M‐HLHa (p = 0.001). IFN‐signature markers such as IP‐10 and markers of CD8/NK lymphocyte activity such as sIL2R or CD4/CD8 ratio < 1 or IL‐18 were elevated in M‐HLHa and, to a lesser extent, in non M‐HLHa.

3.2. RS Development

A total of 20 potential predictors (including nine key HLH‐related CK) were selected, Table S1. From the predictors selected, eight were significantly associated with the development of M‐HLHa in univariate analysis. The factors included demographics (age > 48y; p = 0.004), clinical (hemoglobin ≤ 9 g/L (p = 0.03), platelets ≤ 82 g/L (p = 0.03), lymphocytes CD4/CD8 ≤ 1 (p = 0.03)) and cytokine levels (IL‐10/IL‐6 ≥ 1.5 (p = 0.002), TNF‐α ≤ 43 pg/mL (p = 0.001), IL‐18 > 574 pg/mL (p = 0.0002), sIL2‐R‐α > 36 700 pg/mL (p = 0.0004)).

Independent risk factors for M‐HLHa are shown in Table 2, and include four factors: Patients over 48 years of age (OR: 2.86 [1.12–7.33], p = 0.03), IL‐10/IL‐6 ≥ 1.5 (OR: 2.71 [1.01–7.31], p = 0.04), IL‐18 > 574 pg/mL (OR: 4.90 [1.89–12.75], p = 0.001), and TNF‐α ≤ 43 pg/mL (OR: 3.16 [1.16–8.61], p = 0.02). All four factors exhibit an elevated risk of M‐HLHa.

TABLE 2.

Multivariate logistic model analysis selected four associated predictor factors.

Variable	OR [95% CI]	p‐value	Scoring a
Age > 48 years	2.86 [1.12–7.33]	0.03	11
TNF‐α ≤ 43 pg/mL	3.16 [1.16–8.61]	0.02	12
IL‐18 > 574 pg/mL	4.90 [1.89–12.75]	0.001	16
IL‐10/IL‐6 ratio ≥ 1.5	2.71 [1.01–7.31]	0.04	10

Note: Four factors were selected as predictors for the multivariate analysis of the risk of M‐HLHa. Age > 48 years, TNF‐α ≤ 43 pg/mL, IL‐18 > 574 pg/mL, IL‐10/IL‐6 ratio ≥ 1.5.

Abbreviations: 95% CI, 95% confidence interval; IL, interleukin; OR, odds ratio; TNF‐α, tumor necrosis factor‐α.

^a The scoring points in the final model were assigned based on the regression coefficients β multiplied by 10 and rounded to the nearest integer.

Based on the β coefficients, which were multiplied by 10 for each independent risk factor, a total point score was assigned to each risk factor. Eleven points were assigned to age > 48 years, 12 points to TNF‐α ≤ 43 pg/mL, 16 points to IL‐18 > 574 pg/mL and 10 points to IL‐10/IL‐6 ratio ≥ 1.5. No points are attributed to patients with following parameters: age ≤ 48 years, TNF‐α > 43 pg/mL, IL‐18 ≤ 574 pg/mL or IL‐10/IL‐6 ratio < 1.5. The formula for the RS model was as follows: RS = (11 × 1, if age was higher 48y) + (12 × 1, if TNF‐α was less than or equal 43 pg/mL) + (16 × 1, if IL‐18 was higher 574 pg/mL) + (10 × 1, if IL‐10/IL‐6 ratio was higher than or equal to 1.5). The total RS ranged from 0 to 49 points.

3.3. Model Performance and Validation

The ROC curve is shown in Figure 2a. The ability to discriminate between M‐HLHa and other (AID/idiopathic) was quantified by the AUC; the mean AUC in the cohort was 0.83 (95% CI, 0.76; 0.89), and the AUC in the cross‐validation (internal validation) was 0.80 (95% CI, 0.76–0.90). The goodness‐of‐fit test p‐value was 0.71, suggesting a good agreement between observation and prediction.

FIGURE 2.

Receiver operating characteristic (ROC) curve—Risk Score. (a) ROC curve for the predictive score in the cohort of 112 patients; (b) ROC curve for the IL‐10/IL‐6 ratio in the subset of 75 patients tested with the ELLA platform; (c) ROC curve for the predictive score in the subset of 54 patients with both Luminex and ELLA measurements; (d) ROC curve for the IL‐10/IL‐6 ratio in the same subset of 54 patients.

3.4. Risk Categorization

A cut‐off value of 20 points was determined as a clinically relevant threshold for M‐HLHa risk, with a sensitivity [95% CI] of 0.91 [0.79; 0.98] and specificity [95% CI] of 0.63 [0.50; 0.74] (positive predictive value (PPV) [95% CI] = 0.62 [0.49; 0.74] and negative predictive value (NPV) [95% CI] = 0.91 [0.79; 0.98]). As a result, we categorized the score into two risk categories as low risk (0–19 points) and high risk (≥ 20 points).

The logistic model demonstrated that high‐risk patients for M‐HLHa had a 17‐fold increased risk of presenting an underlying neoplasia compared to low‐risk patients (OR [95% CI] = 17.2 [5.4;54.5], p < 0.0001). The positive and negative likelihood ratios were 2.44 [95% CI, 1.77; 3.37] and 0.14 [95% CI, 0.05;0.37], respectively, Table 3.

TABLE 3.

Risk Score categorization.

Variable	Categories	Score (points)	OR	IC95%	p
Risk for M‐HLHa	Low risk	0–19	1	—	—
	High risk	20–49	17.2	[5.4–54.5]	< 0.0001

Note: A cutoff value of 20 points was determined as a relevant threshold for M‐HLH. The total score was categorized in low risk (range 0–19) and high risk (20–49).

3.5. Post Hoc Analysis of IL‐10/IL‐6 Ratio as a Biomarker

A post hoc analysis was performed on a randomly selected subset of 75 patients, Table S2A, to assess the IL‐10/IL‐6 ratio measured using the ELLA cytokine assay. In this subset, the median IL‐10/IL‐6 ratio was 0.81 (IQR: 0.17–4.83). The median [IQR] IL‐10/IL‐6 ratio was 4.83 [0.82–27.70] in patients with M‐HLHa (n = 34), significantly higher than in those with non M‐HLHa (0.21 [0.09–0.96] (n = 41), p < 0.001).

The AUC of the IL‐10/IL‐6 Ratio was 0.83 [95% CI: 0.72–0.91], indicating good performance of the model, Figure 2b. The optimal threshold maximizing the Youden index was 1.7, with a sensitivity of 0.65 [95% CI: 0.46–0.80] and specificity of 0.88 [95% CI: 0.74–0.96]. The PPV was 0.81 [95% CI: 0.62–0.94] and the NPV was 0.75 [95% CI: 0.60–0.86].

3.6. Post Hoc Analysis of the 54 Patients With Complete Cytokine Data

Among the 54 patients with both cytokine panels available, Table S2B, the RS calculated in the 54‐patient subset reached an AUC of 0.87 [95% CI: 0.77–0.95], Figure 2c, with a sensitivity of 0.96 [95% CI: 0.79–1.00] and a specificity of 0.60 [95% CI: 0.41–0.77] using the threshold of 20. These results are in line with the initial analysis on the full cohort of 112 patients. Similarly, the IL‐10/IL‐6 Ratio yielded an AUC of 0.79 [95% CI: 0.68–0.90], Figure 2d, with a sensitivity of 0.62 [95% CI: 0.41–0.81] and a specificity of 0.83 [95% CI: 0.65–0.94] at the predefined threshold of 1.7. These results are consistent with the initial analysis performed on the 75 patients for whom the IL‐10/IL‐6 ratio was available.

4. Discussion

In this study, we evaluated clinical and biological markers including cytokine levels of 112 adults diagnosed with HLHa to assess their association with M‐HLHa. Four parameters were significantly associated with M‐HLHa in the multivariate logistic model. These variables were incorporated into a predictive score ranging from 0 to 49 points, with a threshold of 20 points to provide optimal discrimination, with placing above this threshold corresponding to a 17‐fold increased risk ratio for developing malignancy. The score demonstrated robust diagnostic performance, with an AUC of 0.83 and high sensitivity and NPV, supporting its utility as a sensitive tool to screen for occult malignancy in HLH patients.

A post hoc analysis confirmed that use of the IL‐10/IL‐6 ratio as a routine, rapid screening tool was effective in identifying the presence of underlying hematologic malignancies. This ratio alone yielded a comparable AUC to the multivariable RS, but with better specificity. It is notable that the level of both CK is higher in the M‐HLHa, and in particular there is no overlap in the biological values for IL‐10 between the two groups. Nevertheless, the IL‐10/IL‐6 ratio remains a reliable and rapid first‐line marker to guide clinical suspicion toward an underlying hematologic malignancy. Importantly, the threshold of 1.7 optimized diagnostic accuracy and may help prioritize patients for further diagnostic procedures, such as bone marrow biopsy.

It is important to note that the IL‐10/IL‐6 ratio showed lower sensitivity than the composite RS, highlighting a risk of false negatives—particularly in patients with low cytokine levels (IL‐10), such as those previously treated with immunosuppressive therapy or chemotherapy (unpublished data). In such cases, the full composite RS, integrating demographic and cytokine parameters, may provide a more sensitive second‐line tool. To further assess the clinical applicability of our findings, an analysis was conducted on a subset of 54 patients who underwent both cytokine assays. This subset analysis confirmed the strong association between the IL‐10/IL‐6 ratio and M‐HLHa, with diagnostic performance metrics similar to those observed in the main cohorts. Altogether, these findings support two diagnostic strategies: the IL‐10/IL‐6 ratio could serve as a highly specific and rapid screening test at HLH onset. However, the diagnosis may remain uncertain despite aggressive and repeated investigations. In such cases, applying the multivariate score could provide additional diagnostic clarity. This dual strategy could enhance early identification of M‐HLH and facilitate timely diagnostic work‐up and therapeutic decisions in a population where delays in diagnosis affect prognosis.

Malignancy‐associated HLH (M‐HLHa) presents in two forms: first, in patients already diagnosed with malignancy for whom the Optimized HLH Inflammatory (OHI) Index [23] or IL‐10 [24] have already been proposed to diagnose HLH and second, as a presenting feature of an underlying malignancy at initial diagnosis or at relapse. Our biomarkers were designed for the latter form, aiming to support clinicians in developing a management strategy that rapidly leads to the diagnosis of malignancy while minimizing the risk of organ failure and death by initiating chemotherapy early. Interestingly, Xu et al. [25] reported that IL‐10 is elevated in lymphomas, whereas IL‐6 is not, suggesting that the IL‐10/IL‐6 ratio is also increased. Consistent with our data, they further report that IFN‐γ and TNF‐α are normal or only moderately elevated in lymphomas.

Multiplex immunoassay platforms, such as Luminex (magnetic bead‐based) and ELLA (Simple Plex, Bio‐Techne), are widely used for the simultaneous quantification of CK and other inflammatory proteins. In this study, we initially performed cytokine quantification using the Luminex platform, followed by validation and comparison with the ELLA system, to evaluate their respective strengths and ensure robust data interpretation. The ELLA platform became routinely used in our hospital for clinical cytokine and chemokine quantification using Simple Plex cartridges (ProteinSimple) in the context of renal transplantation immunomonitoring, occular lymphoma and detection of cytokine storm. The ELLA platform utilizes automated microfluidic technology, providing enhanced sensitivity and reproducibility, especially for low‐abundance CK. In contrast, Luminex technology, based on encoded magnetic beads, offers higher multiplexing capacity but is more susceptible to interference from heterophilic antibodies, soluble proteins, and matrix effects, particularly in complex samples such as serum or plasma [26]. Given the differences in calibration standards and reference curves between the two platforms, significant variations in absolute reported values can occur. Therefore, we performed cross‐platform comparisons to ensure the reliability and clinical relevance of our findings [27].

Multivariate analysis identified four parameters—age, TNF‐α, IL‐18, and the IL‐10/IL‐6 ratio—as significantly associated with M‐HLHa. It is well established that patients with malignancy‐associated HLH are generally older than those with HLH linked to other conditions. In a cancer‐focused review, the authors recommended thoroughly investigating the possibility of an underlying malignancy in older patients presenting with HLH [15]. IL‐18 induces IFN‐γ production which is one of the major HLH CK. In a previous study, IL‐18 was found at elevated levels in HLH patients with a range from 600 to 3000 pg/mL [28]. Excessive production of IL‐18 (around 10 000 pg/mL) was observed in genetic autoinflammatory disorders predisposed to HLH such as those associated with NLRP4 or XIAP alterations. In 70 EBV lymphoproliferative diseases associated with HLH, co‐elevation of IL‐10 and IL‐18 has been reported as being at higher levels than 20 other TH1 CK. Moreover, receiver operating curve analysis further revealed high values of area under curves for IL‐18 and IL‐10 [29].

Other criteria identified through uni‐ and multivariate analyses may reflect the biological mechanisms underlying M‐HLHa. In this study, excess CD8+ T lymphocytes, soluble IL‐2 receptor alpha (sIL‐2Rα), and IFN‐γ signatures emerged as potential hallmarks of M‐HLHa. These findings are consistent with previous studies suggesting that these markers are associated with lymphoma‐related HLH [30, 31, 32]. These biomarkers reflect CD8+ T‐cell proliferation and activation, which lead to IFN‐γ synthesis—a characteristic feature of familial HLH pathophysiology [33]. In familial HLH (F‐HLH), CD8+ T‐cell activation typically results from a genetic defect affecting their cytotoxic function [34]. The phenotypic similarities between M‐HLHa and F‐HLH could be related to the presence of variants in HLH‐related genes, as we have previously demonstrated [11]. Alternatively, this resemblance may result from a “tumor antigen” effect that overstimulates the immune system, particularly NK and CD8+ T cells. In this context, the cytotoxic activity of these cells may be impaired due to mechanisms of tumor immune evasion.

A key finding of our study is the strong association between the IL‐10/IL‐6 ratio and M‐HLH. IL‐10 is a powerful anti‐inflammatory cytokine, and its role in HLH is complex [35]. In experiments with a secondary inflammatory HLH mouse model, mice continuously injected with 5′‐Cytosine‐phosphate‐Guanine‐3′ (CpG) exhibited a severe HLH‐like syndrome with elevated serum levels of IFN‐γ, IL‐6, IL‐10, and IL‐18 [36]. Abrogation of the IL‐10 signal led to fulminant HLH with hemophagocytosis, suggesting that IL‐10 is a key regulatory cytokine in this secondary HLH mouse model. In this study, the median ratio close to 1 found in the non‐M‐HLHa patients suggests the presence of a potential homeostatic balance between IL‐10 and IL‐6 in inflammatory HLH. IL‐10's biological function in this context may act as a “counter regulation cytokine” in charge of countering the deleterious effect of pro‐inflammatory CK.

However, this balance appears to be disrupted in M‐HLHa, as the median ratio is around 3, attesting to a deregulated production of IL‐10 and favoring an immunosuppressive IL‐10‐dominated profile that likely reflects an underlying immune dysregulation. Previous studies have established IL‐10 and the IL‐10/IL‐6 ratio as an diagnostic indicator of vitreoretinal lymphoma, a primary nervous system B Large Cell lymphoma [37, 38]. One plausible explanation is the secretion of IL‐10 by lymphoma cells or neighboring cells, which may alter the expression profile of immunomodulatory genes, inhibit the release of pro‐inflammatory mediators, decrease antigen presentation and phagocytosis, and concomitantly enhance the inhibitory and tolerance functions of immune cells especially CD8 or NK cells against tumors. This, in turn, may worsen the HLH syndrome, with the anti‐inflammatory effect of IL‐10 secreted by lymphoma cells paradoxically acting as a “booster” of HLH pathogenesis [39, 40, 41].

Our study presents several strengths and limitations. Notably, our RS is the first to use clinical and cytokine data to predict M‐HLHa patients. Furthermore, the RS and IL‐10/IL‐6 ratio were developed from a prospective study population, thus avoiding potential selection biases and obtaining reliable data with non M‐HLHa patients as the control population (instead of a non‐HLH lymphoma cohort). One limitation of our study is the relatively small sample size. Given that HLH is a rare disease, the number of patients included is more than acceptable; however, this constrains for the development of a predictive score, thereby limiting the number of parameters that can be included in the model. Reliable and valid statistical methods were used, as well as multiple imputation methods to avoid any data loss. Good performances were observed in both our development and validation populations. However, we chose to dichotomize the parameters to facilitate the score's usability for clinicians, which could be a limitation of our study resulting in a loss of information. Although two different cytokine assay methods were used, the comparison of the IL‐10/IL‐6 ratio medians showed no significant difference between the methods, and the composite score demonstrated similar diagnostic performance (identical AUCs), supporting the robustness of our findings.

Despite these limitations, our findings pave the way for improving the early detection and risk stratification of M‐HLHa. The integration of novel cytokine biomarkers, such as the IL‐10/IL‐6 ratio, offers a promising step toward improved management for this complex and often rapidly progressing syndrome. Additionally, the development of a predictive score combining clinical and cytokine data provides an essential tool for clinicians to refine diagnostic accuracy and guide their treatment decisions.

5. Conclusion

We have devised a RS for M‐HLHa that incorporates both clinical factors and cytokine levels. Our study has revealed significant pathophysiological differences between patients with M‐HLHa and those without. This scoring system shows potential for enhancing diagnostic precision and guiding timely interventions for patients at elevated risk of M‐HLHa. This approach aims to improve the management and outcomes of this complex clinical condition. In addition, the use of a more current cytokine quantification method for determining the IL‐10/IL‐6 ratio reinforces the potential of this biomarker as a complementary and accessible tool for rapid risk stratification in clinical practice.

Author Contributions

C.B. is the Principal Investigator of the research, and supervised the research and biostatistics, and wrote the paper. O.H. is the Scientific Director and supervised the research and the biostatistics, and wrote the paper. G.S.B. are co‐Scientific Directors and reviewed the paper. Y.L. and R.L. reviewed the paper. N.S. led the biostatistics analysis and contributed to performing biostatistics. S.C. performed the Ella Cytokines Test (IL‐10/IL‐6) on the 75 patients and reviewed the paper. M.G. is a Laboratory Technician and performed the Luminex cytokines tests on the 112 patients. O.L., D.L., C.L., E.L., F.L., M.M., J.‐M.M., P.M., L.T., J.‐F.V., G.U., F.S., M.C., included patients in the study. All authors read, edited, and approved the manuscript.

Funding

This work was supported by Clinical Research Direction of Assistance Publique‐Hopitaux de Paris, AOM 10219. Funded by Programme Hospitalier de la Recherche Clinique 2010 of the French Ministry of Social Affairs and Health.

Ethics Statement

The regional Ethics Committee of Paris (n° ID‐RCB: 2009‐AO00301‐56) approved the study.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Univariate logistic model analysis.

Table S2A‐B: Characteristics of patients with cytokine measurements on the ELLA platform only and on both Luminex and ELLA platforms.

Table S2A: Patients with cytokine measurements on both Luminex and ELLA platforms.

Table S2B: Patients with cytokine measurements on the ELLA platform only.

Figure S1: Inclusion Criteria in the HLH—GENES cohort

Bloch C., Chhun S., Sritharan N., et al., “The IL‐10/IL‐6 Ratio and the Risk Score: Two Cytokines‐Based Predictors for Malignancy‐Associated Hemophagocytic Lymphohistiocytosis in Adults (M‐HLHa),” American Journal of Hematology 101, no. 9 (2026): 2214–2224, 10.1002/ajh.70410.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.