An improved index for diagnosis and mortality prediction in malignancy-associated hemophagocytic lymphohistiocytosis

Abstract

Hemophagocytic lymphohistiocytosis (HLH) is a life-threatening inflammatory syndrome that may complicate hematologic malignancies (HMs). The appropriateness of current criteria for diagnosing HLH in the context of HMs is unknown because they were developed for children with familial HLH (HLH-2004) or derived from adult patient cohorts in which HMs were underrepresented (HScore). Moreover, many features of these criteria may directly reflect the underlying HM rather than an abnormal inflammatory state. To improve and potentially simplify HLH diagnosis in patients with HMs, we studied an international cohort of 225 adult patients with various HMs both with and without HLH and for whom HLH-2004 criteria were available. Classification and regression tree and receiver-operating curve analyses were used to identify the most useful diagnostic and prognostic parameters and to optimize laboratory cutoff values. Combined elevation of soluble CD25 (>3900 U/mL) and ferritin (>1000 ng/mL) best identified HLH-2004-defining features (sensitivity, 84%; specificity, 81%). Moreover, this combination, which we term the optimized HLH inflammatory (OHI) index, was highly predictive of mortality (hazard ratio, 4.3; 95% confidence interval, 3.0-6.2) across diverse HMs. Furthermore, the OHI index identified a large group of patients with high mortality risk who were not defined as having HLH according to HLH-2004/HScore. Finally, the OHI index shows diagnostic and prognostic value when used for routine surveillance of patients with newly diagnosed HMs as well as those with clinically suspected HLH. Thus, we conclude that the OHI index identifies patients with HM and an inflammatory state associated with a high mortality risk and warrants further prospective validation.

© 2022 by The American Society of Hematology.

Figures

Graphical abstract

Figure 1.

The distribution of HLH-defining features reveals significant overlap between patients with uncomplicated HMs and patients with malignancies complicated by HLH, regardless of clinical suspicion for HLH. Normal range and the diagnostic threshold level from the HLH-2004 study (used for FHL) are shown in patients examined as routine at HM diagnosis (left bars) and patients tested due to HLH suspicion (right bars). These values are the preliminary measurements at the initial presentation of HLH/malignancy. The distribution of maximally pathologic values during the index encounter is shown in supplemental Figure 2. (A) Distribution of the inflammatory markers. The values of sCD25 between HM-HLH with suspected HLH and those evaluated as routine were not significantly different (P = .09), whereas the ferritin values were (P = .016). (B) Distribution of the blood lineages. (C) Distribution of other markers. (D) Categorical parameters. The statistics of the quantitative parameters were analyzed by using the Mann-Whitney U test. The statistics of the categorical parameters were analyzed with Fisher’s exact test. Values shown were available for >90% of the patients. NK activity was available for only 4% of the patients and therefore is not shown. HM = patients with uncomplicated HMs; HM-HLH = patients fulfilling 5 of 8 HLH-2004 diagnostic criteria.*P < .05, **P < .01, ***P < .001 ****P < .0001. ANC, absolute neutrophil count; Hb, hemoglobin; N/A, not available; ns, not significant; PLT, platelets; Tg, triglycerides.

Figure 2.

Inflammatory markers best discriminate between patients with uncomplicated HMs and patients with HLH in the context of HMs. (A) CART, including initial assessment of all markers, is shown. CART analysis determines the relative importance of different variables for identifying homogeneous groups within a data set. CART’s growth limits were defined with the Gini method, limiting 4 cases in the child node and 5 cases in the parent node. The minimum change in improvement was defined as 0.0001 and pruning with the minimum difference in risk of 1 standard error. (B) ROC analysis for individual HLH-defining markers at presentation (initial measurement) is shown. The distribution of maximally pathologic values during the index encounter is shown in supplemental Figure 4. The dotted line indicates the pertinent point on the curve identified as the best balance between sensitivity and specificity (with the highest Youden index). (C) The table summarizes the performance of each marker and thresholds defined by the HLH-2004 and ROC analyses. AUC, 95% CIs, P values, sensitivity (Sens), and specificity (Spec) for the HLH-2004 thresholds and the optimized ones at initial presentation are presented. ANC, absolute neutrophil count; Tg, triglycerides.

Figure 3.

An optimized combination of sCD25 and ferritin with the OHI index improves mortality prediction in patients with HMs. (A) Binary linear regression for the OHI index (combination of sCD25 and ferritin) was performed to calculate the predicted probabilities of diagnosing HLH (per HLH-2004) and prediction of mortality at 500 days. ROC for the predicted probabilities to predict HLH-2004 is presented on the left, and ROC for the predicted mortality probabilities at day 500 is presented on the right. (B) ROC analysis for sCD25 and ferritin identifying mortality by day 500 is shown. The blue arrow indicates the optimal point identified in the prior ROC analysis for identifying patients meeting the HLH-2004 criteria (Figure 2B). The red arrow indicates the pertinent point on the curve identified as the best balance between sensitivity and specificity (with the highest Youden index) for identifying mortality. These values were determined as the OHI index values. The table shows the sensitivity (Sens) and specificity (Spec) of the OHI for identifying ≥5 HLH-2004 parameters. The sensitivity and specificity were calculated by using a contingency table. (C) Kaplan-Meier curves of patients classified at the initial presentation (utilizing initial measurement) by HLH-2004 (blue), OHI (sCD25 ≥3900 U/mL and ferritin ≥1000 ng/mL, red), and “optimized” HLH-2004, O-HLH2004 (the same framework as HLH-2004 but with sCD25 ≥3900 U/mL and ferritin ≥1000 ng/mL, gray). The number at risk is presented for each group. Statistics were calculated with the log-rank (Mantel-Cox) test. *P < .05. All diagnostic indices were highly distinct between patients who were positive and negative (P < .0001).

Figure 4.

The OHI index is highly predictive of mortality in patients with lymphoid or myeloid malignancies. ROC analyses for sCD25 and ferritin for predicting HLH-2004 diagnosis (≥5 diagnostic criteria) and mortality prediction in patients with lymphoma (A) and patients with myeloid malignancies (B). The blue arrow indicates the pertinent point on the curve identified as the best balance between sensitivity and specificity (with the highest Youden index). The red arrow indicates the point closest to the OHI values (sCD25 >3900 U/mL and ferritin >1000 ng/mL). (C) Survival of patients with lymphoma (B-cell lymphoma, T-cell lymphoma, and Hodgkin lymphoma; n = 167), classified per the indicated HLH indices. (D) Survival of patients with myeloid malignancies (acute myeloid leukemia, myelodysplastic syndrome, and myeloproliferative neoplasms; n = 48) as in panel C. Classification for this Kaplan-Meier analysis is based on the peak/nadir laboratory values obtained during the initial presentation of HLH/malignancy. The number at risk is presented for each group. Statistics were calculated with the log-rank (Mantel-Cox) test. **P < .001, ***P = .001, ****P < .0001.

Figure 5.

The OHI index recognizes patients at high mortality risk not identified by current diagnostic indices. The patients were classified based on the HLH-2004 (≥5 diagnostic criteria), optimized HLH-2004 (O-HLH2004, the HLH-2004 framework with sCD25 >3900 U/mL and ferritin >1000 ng/mL), the OHI index (sCD25 >3900 U/mL and ferritin >1000 ng/mL), or HScore ≥ 169 based on the peak/nadir of relevant laboratory values obtained during the index hospitalization. (A) Kaplan-Meier curve for patients classified concordantly or discordantly by using the HLH-2004 criteria and the OHI index. (B) Kaplan-Meier curve for patients classified concordantly or discordantly by using the optimized HLH-2004 criteria (O-HLH2004) and the OHI index. (C) Kaplan-Meier curve for patients classified concordantly or discordantly by using the HScore and the OHI index. (D) Kaplan-Meier curves of patients classified by using the OHI index or parameters of the OHI index. The number at risk is presented for each group. Statistics were calculated with the log-rank (Mantel-Cox) test. *P < .05, ***P < .001, ****P < .0001.

Figure 6.

The OHI index performs better when assessed during routine surveillance and in treatment-naive patients. Kaplan-Meier curves of subgroup analyses of patients classified by using the OHI index. (A) Centers evaluating sCD25 and ferritin upon HLH suspicion only vs centers that have routine surveillance of patients with newly diagnosed lymphoma (Toyama Hospital in Japan and MD Anderson Cancer Center in the United States). (B) Treatment-naive patients vs pretreated patients. Although patients who received treatments that will undoubtedly influence their laboratory results (eg, cytoreductive therapy/high-dose steroids) were excluded from the study, patients who received minimal pretreatment (eg, lower dose steroids for only a few days) before assessment of relevant studies were included and defined in the pre-treatment group. The number at risk is presented for each group. Statistics were calculated with the log-rank (Mantel-Cox) test. *P < .05, **P < .01, ****P < .0001.