To construct a retrospective cohort, we retrieved medical records and follow-up documents of adult patients who were hospitalized at Qingchun, Zhijiang, Yuhang, and Chengzhan divisions of FAHZJU with discharge diagnosis of dermatomyositis (DM), polymyositis (PM), and amyopathic dermatomyositis (ADM) from January 1, 2017 to December 31, 2020. The inclusion criteria of this study were as follows: (1) age over 18 years old; (2) the definite/probable diagnosis of DM, PM, or ADM satisfied the 2017 ACR/EULAR classification criteria, as certified by two experienced rheumatologists (Heng Cao and Bei Xu) (19); and (3) PET/CT scan performed during hospitalization. Exclusion criteria were as follows: (1) confirmed overlap syndromes with other connective tissue diseases (CTDs); (2) myopathy related to strenuous exercise, thyroid dysfunction, drug-induced myositis (i.e., statins, lamivudine, and Chinese herbal medicine), inherited metabolic disorders, etc.; (3) newly identified or unresolved malignancies; (4) hospitalization for reasons unrelated to myositis and its complications such as fracture, pregnancy, acquired immunodeficiency syndrome, and cataract, due to insufficient medical records for this study; and (5) loss to follow-up without death from any cause within 6 months after hospitalization. The research protocol was approved by the Institutional Review Board (IRB) of FAHZJU (Reference Number: 2021-194) and was implemented in accordance with the Declaration of Helsinki. Written informed consent to utilize and publish clinical data was acquired from all the patients involved at hospitalization.

Clinical records of all the enrolled patients were retrospectively screened and collected utilizing the electronic medical record (EMR) system of FAHZJU. Data including demographic information, disease activity assessment, complications, immunosuppressive regimens, and radiological/laboratory findings were acquired and analyzed. Survival data were extracted from the follow-up records. To be exact, IIM patients were followed from the date of hospitalization until the end of follow-up. For patients who died during hospitalization, their dates of death were accurately documented in the EMR system. For patients who were discharged, a routine return visit was arranged 2 weeks after discharge. In addition to the regular inpatient or outpatient visits, a concise telephone interview was carried out 3 months after discharge, and at an annual frequency afterwards. The end of follow-up could be owing to death from any cause, loss to follow-up, or closure of follow-up for the purpose of this study (June 30, 2021).

Baseline disease activity assessment, lung function testing, and laboratory and radiological detections were performed in the first week of hospitalization. On-admission IIM disease activity was routinely measured using the Myositis Disease Activity Assessment Visual Analogue Scales (MYOACT) (13). ILD and its rapid progression were checked and confirmed by an experienced radiologist and a respiratory specialist (YL and BY) using lung HRCTs. A subset of RP-ILD patients was defined as those presenting with progressive dyspnea and progressive hypoxemia, acute worsening of interstitial alteration on the lung radiograph within 1 month after hospitalization, or onset of respiratory symptoms (20–22). Diagnosis of sHLH was made according to the criteria proposed by the Histiocyte Society in 2004 (HLH-2004, Supplementary Table 1 ) (13, 23) and was confirmed by a hematologist (BW). The included IIM patients were hereby divided into an HLH group and a non-HLH group (control group). Since pulmonary infection was easily confused with interstitial lung disease, identification of bacterial, fungal, or tuberculosis infection was a careful and comprehensive decision based on the essential microbiological findings in sputum or blood, clinical manifestations, and radiographic and laboratory abnormalities. Besides, diagnosis of Epstein–Barr virus (EBV) and cytomegalovirus (CMV) infection relied on the detection of serum antibody and DNA. All of the included patients received potent immunosuppressive therapies: (1) systemic prednisolone (PSL) or methylprednisolone (mPSL) with a maximum dosage ≥ 1 mg/kg/day (calculated by prednisolone); (2) combined therapy of PSL/mPSL, disease-modifying anti-rheumatic drugs (DMARDs), or Janus kinase (JAK) inhibitors, with or without intravenous immunoglobulin (IVIG). The DMARDs used in these patients encompassed Mycophenolate, Tacrolimus, Cyclosporine, Methotrexate, Cyclophosphamide, Thalidomide, and Hydroxychloroquine. Meanwhile the JAK inhibitors used in these patients included Tofacitinib and Baricitinib.

To acquire the profiles of myositis-specific antibodies (MSAs) and myositis-associated antibodies (MAAs) as well as peripheral lymphocyte subsets and cytokines, serum samples were routinely acquired and detected within the first week of hospitalization. The 12 MSAs (anti-MDA5, anti-TIF1γ, anti-Jo-1, anti-EJ, anti-OJ, anti-PL-7, anti-PL-12, anti-Mi-2α, anti-Mi-2β, anti-SAE1, anti-SRP, and anti-NXP2) and 4 MAAs (anti-Ro-52, anti-PM-Scl75, anti-PM-Scl100, anti-Ku) were assessed by an immunoblotting assay utilizing the EUROLINE Autoimmune Inflammatory Myopathies 16 Ag (IgG) commercial line blot assay (Euroimmun, Lübeck, Germany) including a membrane strip with the 16 autoantigens as per the manufacturer’s instructions. Peripheral lymphocyte subsets and cytokine profiles were detected by the Department of Clinical Laboratory (FAHZJU) and uploaded onto the EMR system (detected values). Specifically, peripheral lymphocyte subsets were determined as percentages of CD3⁺CD4⁺, CD3⁺CD8⁺, CD3^-CD16⁺CD56⁺, and CD3^- CD19⁺ cells using CD45-PE, CD3-PC5, CD4-FITC, CD8-PE, CD3-FITC-CD(16 + 56)-PE, and CD19-FITC mouse anti-human fluorescence monoclonal antibodies (BD Bioscience) and the BD FACScanto™ II flow cytometer (Becton Dickinson, San Jose, CA, USA). Meanwhile serum levels of IL-2, IL-4, IL-6, IL-10, IL-17A, tumor necrosis factor (TNF)-α, and IFN-γ were quantified utilizing the cytometric bead array (CBA) kit BD™ CBA Human Th1/Th2/TH17 Cytokine Kit (BD Biosciences, San Jose, CA, USA) and the flow cytometer (described above). Measurement of cytokine was implemented as per the manufacturer’s instructions. The lower and upper limits of cytokine detection were 0.10 pg/ml and 5,000.00 pg/ml, respectively. The data were engendered in graphical and tabular format utilizing FCAP Array™ software (BD Biosciences, San Jose, CA, USA).

Whole-body detection of CT and PET, which was performed with a combined PET/CT scanner (Biograph, Sensation 16, Siemens systems), covered a wide range from the meatus of the ear to the mid-thigh. Patients fasted overnight or for at least 6 h prior to the PET/CT detection. Blood glucose levels were confirmed to be within normal limits before the injection of 4.0 MBq/kg of [18F] FDG. Patients rested for 30 min so that non-specific FDG uptake in muscles was minimized. Imaging acquisition was systematically performed at 60 min after injection. SUV (standard uptake value) was calculated by the following formula: SUV (g/ml) = regional radioactivity concentration (Bq/ml)/[injected dose (Bq)/body weight (g)]. Region of interest (ROI, 20 mm diameter) was manually placed by a single trained radiologist (YL) at the region with the highest FDG uptake in the following organs: liver, spleen, bone marrow (thoracic, T10–T12, lumbar, L2–L4) (24), bilateral lung, esophagus, stomach, small intestine, colon/rectum, bilateral cerebellum, and bilateral proximal muscles (namely, trapezius, deltoid, biceps, iliopsoas, gluteus medius, gluteus maximus, and quadriceps) (25), excluding the region markedly influenced by FDG uptake in adjacent anatomical structures. In order to avoid noise and acquire the value representing a certain volume of targeted organs, SUV was calculated as the mean value of ROI (SUVmean) rather than the maximum value at a single pixel (25). For bilaterally distributed organs, the SUVmean was counted as the maximum SUVmean value of the symmetrical sides. The radiologist was blinded to the diagnostic subtypes, complications, and outcome of the included patients when assessing the SUVmean value of each and every targeted organ.

Statistical analysis was performed using SPSS 22.0 (Chicago, IL, USA), R 3.6.1, and GraphPad Prism 8.0. In comparison between HLH and non-HLH patients, independent sample t-test was utilized to compare normally distributed continuous variables; meanwhile Mann–Whitney U test was used to compare skewed continuous variables or ordinal categorical variables. Chi-square test and Fisher’s exact test were utilized to compare unordered categorical variables. p-values in comparisons and univariate analyses were adjusted by false discovery rate (FDR) correction, using p.adjust function in R.3.6.1, to acquire adjusted p-values and minimize type I error. The univariate and the following multivariate logistic regression analyses were used to identify significant factors associated with development of sHLH in IIM patients. Factors with p < 0.05 in univariate analyses or comparisons were entered into the multivariate analysis. Binary logistic regression was also used to evaluate the contributary roles of statistically significant factors to development of sHLH. Receiver operating characteristic (ROC) curve analysis was applied to quantify the predictive value of continuous variables. The correlation between two continuous variables was measured by the Pearson linear analysis. Survival in different groups was evaluated by the Kaplan–Meier method with log-rank test. All tests were two-sided, and p <  0.05 was considered statistically significant.

A total of 352 adult IIM patients, including 11 patients who developed sHLH (3.1%), were hospitalized at the Qingchun, Zhijiang, Yuhang, and Chengzhan divisions of FAHZJU from January 1, 2017 to December 31, 2020. Among them, 69 patients who satisfied the inclusion/exclusion criteria were finally included into the study ( Supplementary Figure 1 ), encompassing 45 with DM, 12 with PM, and 12 with ADM. Twenty-eight (40.6%) were males, and the mean age of all the patients included was 56.52 ± 11.71 years old. Thirty-one patients (44.9%) died in follow-up and the medium follow-up time was 13.33 (4.70, 26.45) months. Among the 69 patients, 10 patients developed sHLH ( Figure 1 and Supplementary Tables 2 and 3 ), and the 59 patients without sHLH constituted the control group. Eight of the 10 sHLH patients were anti-MDA5 antibody positive, while the remaining two patients were both complicated with bacterial infection beforehand and were anti-PL-7 antibody positive and anti-SRP antibody positive, respectively. Besides, all of the sHLH events were identified after performance of PET/CT scan. IIM patients complicated with sHLH were found to suffer from worse survival (p = 0.005, Figure 2 ) with seven patients (70.0%) perished within 6 months. In patients who developed sHLH, infection (50.0%) was the most common contributor to death. Meanwhile exacerbation of ILD (including RP-ILD, 39.1%) was identified as the most frequent cause for death in the non-HLH group, followed by carcinoma (21.7%) ( Supplementary Table 4 ). Three of the IIM patients with sHLH received etoposide therapy; however, none of them survived beyond 3 months ( Supplementary Table 5 ). Apart from the conventional therapy of steroid, DMARDs, and IVIG, four patients received combined therapy of steroid and JAK inhibitors (two with Tofacitinib and two with Baricitinib). None of the four patients under the medication of steroid and JAK inhibitors developed sHLH.

To acquire an initial understanding on clinical factors related to sHLH in IIM patients, a comparison between IIM patients with or without sHLH was carried out and identified that patients who developed sHLH had a female dominance (p = 0.041); more complication of pulmonary bacterial infection (p = 0.011); higher disease activity (p = 0.016); lower percentage of peripheral CD3^-CD16⁺CD56⁺ lymphocytes (p = 0.024); higher serum levels of IL-4 (p = 0.004, Figure 3A ), IL-6 (p = 0.004, Figure 3B ), IL-10 (p < 0.001, Figure 3C ), TNF-α (p = 0.002, Figure 3D ), and IFN-γ (p < 0.001, Figure 3E ); higher serum levels of alanine transaminase (ALT, p = 0.023) and aspartate transaminase (AST, p = 0.010); higher levels of spleen SUVmean (p < 0.001, Figure 4A ), bilateral lung SUVmean (p = 0.034, Figure 4B ), colon and rectum SUVmean (p = 0.013, Figure 4C ); as well as positivity of anti-MDA5 antibody (p = 0.005). After FDR correction, nevertheless, only a few remained statistically significant, including higher spleen SUVmean (p < 0.001) and positivity of anti-MDA5 antibody (p = 0.049), as well as higher serum levels of IL-4 (p = 0.045), IL-6 (p = 0.045), IL-10 (p < 0.001), TNF-α (p = 0.034), and IFN-γ (p < 0.001) ( Table 1 ).

The univariate logistic regression analyses for sHLH revealed that pulmonary bacterial infection (p = 0.007), disease activity (p = 0.012), percentage of peripheral CD3^-CD16⁺CD56⁺ lymphocytes (p = 0.040), IL-4 (p = 0.002), IL-6 (p = 0.007), IL-10 (p = 0.049), TNF-α (p = 0.003), IFN-γ (p = 0.001), AST (p = 0.011), spleen SUVmean (p = 0.001), bilateral lung SUVmean (p = 0.045), colon/rectum SUVmean (p = 0.042), anti-MDA5 antibody (p = 0.008), and use of PSL/mPSL+IVIG (p = 0.040) were associated with development of sHLH in IIM patients ( Supplementary Table 6 ). The following multivariate logistic regression analysis identified IFN-γ (p = 0.017), spleen SUVmean (p = 0.035), and anti-MDA5 antibody (p = 0.049) as clinical factors that were significantly correlated with development of sHLH in IIM patients ( Supplementary Table 7 ).

Utilizing ROC curve analyses, the optimal cutoff value of IFN-γ for sHLH was >4.68 pg/ml, with a sensitivity of 80.0%, a specificity of 83.1%, and an AUC of 0.840 ( Figure 4D ). Meanwhile the optimal cutoff value of spleen SUVmean for sHLH was >2.029, with a sensitivity of 100.0%, a specificity of 79.7%, and an AUC of 0.910 ( Figure 4D ). To establish a more efficient and balanced predictor for development of sHLH, a series of models were constructed using logistic regression analysis about IFN-γ, spleen SUVmean, and anti-MDA5 antibody (two or three factors). In the present study, the coefficients were as follows: −10.274 + 0.343 * IFN-γ + 3.226 * spleen SUVmean for IFN-γ and spleen SUVmean (optimal cutoff value of −1.586, sensitivity of 90.0%, specificity of 89.8%, and AUC of 0.937); −5.890 + 0.511 * IFN-γ + 3.222 * anti-MDA5 antibody for IFN-γ and anti-MDA5 antibody (optimal cutoff value of −0.468, sensitivity of 80.0%, specificity of 100.0%, and AUC of 0.899); −11.182 + 3.952 * spleen SUVmean + 1.982 * anti-MDA5 antibody for spleen SUVmean and anti-MDA5 antibody (optimal cutoff value of −1.888, sensitivity of 90.0%, specificity of 81.4%, and AUC of 0.912); −12.453 + 0.492 * IFN-γ + 3.154 * spleen SUVmean and 2.943 * anti-MDA5 antibody for combination of the three factors (optimal cutoff value of 0.047, sensitivity of 90.0%, specificity of 100.0%, and AUC of 0.946) ( Figure 4E ). Through Pearson linear analysis, a moderate correlation was identified between ferritin and spleen SUVmean (p = 0.001, r = 0.380, Figure 4F ) as well as IFN-γand spleen SUVmean (p = 0.001, r = 0.398, Figure 4G ).

To probe into the predictive value of PET/CT scan in survival, the 10 IIM patients with sHLH were further divided into two groups depending on whether he or she died within 3 months (six patients) or survived beyond this threshold (four patients). Only age (p = 0.019), higher level of serum creatine kinase (CK, p = 0.038), and DM subtype (p = 0.005) exhibited statistical significance. However, the significance vanished after FDR correction ( Supplementary Table 8 ). In the six patients who perished within 3 months, all of them were anti-MDA5-antibody-positive DM patients. Due to the limitation of sample size, we failed to identify significant difference in anti-MDA5-antibody-positive/negative IIM patients who developed sHLH (p = 0.149, Figure 5A ). By narrowing down to anti-MDA5-antibody-positive DM patients (vs. other patients), a significant difference was identified in survival (p = 0.004, Figure 5B ).