Introduction
Sintilimab is a human anti-programmed cell death protein 1 (PD-1) IgG4-k monoclonal antibody approved in China to treat classical Hodgkin’s lymphoma.1 Monoclonal antibodies, besides improving the antitumor response, can cause various autoimmune-related adverse reactions via a different mechanism than other systemic therapies, often referred to as immune-related adverse events (irAEs)2 that can affect multiple organs, including the skin, gastrointestinal tract, liver, and lungs.3 The incidence of immune-related hepatotoxicity caused by immune checkpoint inhibitors ranges within 0.7–16%, depending on the dose, duration, and combination regimen.4–6 Immune-related hepatotoxicity is asymptomatic, but some serious complications have been reported in clinical practice, including immune-mediated liver injury caused by checkpoint inhibitors (ILICI).7 The clinical determinants of immune-related hepatotoxicity remain unclear, which complicates risk factor identification and hepatotoxicity timing prediction, and consequently delays the optimal timing of immune-mediated hepatotoxicity treatment. This study aimed to analyze the risk factors for immune-related hepatotoxicity in patients treated with sintilimab, access hepatotoxicity timing and treatment outcome, and construct a risk-scoring model for ILICI. Taken together, the findings of this study are expected to enable clinicians to better predict the risk of hepatotoxicity associated with immune therapy, begin precision treatment as soon as possible, and improve cancer patient outcomes.
Methods
Data collection
Patients treated with sintilimab at the Fujian Medical University Union Hospital (one of the largest emergency hospitals in China) between January 2019 and May 2022 were retrospectively assessed. The Medical Records Inpatient System was used to extract patient health data, including historical epidemiology (sex, age, alcohol consumption, smoking, and family history), concomitant diseases (hypertension, diabetes, hepatitis, coronary heart disease, liver cyst, cholecystitis, and gallbladder or bile duct stones), drug combinations (nonsteroidal anti-inflammatory drugs, proton pump inhibitors, hepatoprotectors, antiemetics, and dose and frequency of sintilimab injections), and baseline laboratory indicators (liver and kidney function, thyroid function indicators, autoimmune antibodies, programmed cell death ligand 1 [PD-L1] expression).
The exclusion criteria were: (1) <18 years of age, (2) abnormal liver function including alanine transaminase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), and total bilirubin 1 month before the first administration of sintilimab, (3) combined therapy with other immune checkpoint inhibitors, (4) insufficient data hindering subsequent analysis such as patients who were not routinely tested for liver enzymes and autoimmune antibodies, and (5) other reasons for liver injury including viral or drug-induced by alternative drugs. The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Fujian Medical Union Hospital. Owing to the retrospective nature of the study, the requirement of written informed consent was waived.
Definition of immune-related hepatotoxicity
According to CTCAE 5.0, ILICI was defined as the occurrence of grade ≥2 liver injury after sintilimab infusion and a Roussel Uclaf Causality Assessment Method score ≥6 (range, −9 to +14 and classified as ≤0, excluded; 1–2, unlikely; 3–5, possible; 6–8, probable; and ≥9, highly probable).8
Statistical analysis
Stata 14 (Stata Corporation, College Station, TX, USA), SPSS (version 25.0; IBM Corp., Armonk, NY, USA), R software version 4.0.5 (R Foundation for Statistical Computing, Vienna, Austria), and Prism 8.0 (GraphPad Software, Boston, MA, USA) were used for data analysis. For continuous variables, normally distributed data were reported as means ± standard deviation, non-normally distributed data were reported as medians and range, and categorical variables reported as percentages. Student’s t, chi-square, or Mann–Whitney U tests were performed for between groups comparisons, and Fisher’s exact test was performed if necessary. Correlation analysis was performed to assess the relationship between two variables. Pearson’s correlation analysis was performed for data with normal or near normal distributions, and Spearman’s correlation analysis were performed for data with non-normal distributions. Stepwise iterative logistic method was adopted based on the Bayesian information criterion to further eliminate redundant variables and the Akaike information criterion model was followed. If the latter were similar, the model with fewer variables was adopted by comparing the differences between the two models.9 Subsequently, independent risk factors related to the incidence of ILICI were determined. Statistical significance was set at p<0.05 (double-tailed). Harrell’s concordance index (C-index) was used to evaluate the effectiveness of prediction and discrimination. Receiver operating characteristic (ROC) curve analysis showed the predictive ability of each risk factor, and combined with the nomogram, revealed the area under the curve (AUC). The R RMS package was used to generate a correction curve reflecting the relationship between the predicted and real incidences. The abscissa depicted the predicted probability, and the ordinate showed the actual probability of the patient (real incidence). Decision curve analysis further showed the net benefit to the observer and was used to evaluate the clinical value of the constructed nomogram.
Results
Patient characteristics
Between January 2019 and May 2022, 772 patients received sintilimab treatment. A total of 187 patients were excluded from the study, among whom 173 had abnormal ALT, AST, ALP, GGT, or total bilirubin levels in the previous cycle, eight had received clinical trial placebos, and six had insufficient data for analysis. The remaining 585 patients were included in the analysis, among whom 71 were included in the ILICI group, representing an ILICI incidence of 12.1% (Fig. 1). Thirty-seven patients (52.1%) had grade 2 hepatotoxicity, 29 (40.8%) had grade 3 hepatotoxicity, and five (7.0%) had grade 4 hepatotoxicity. Among 71 ILICI cases, 63 (88.7%) were probable and 8 (11.3%) were highly probable based on the updated RUCAM, and the average RUCAM score of the ILICI group was 7.06 ± 1.22. Seventeen cases (23.9%) were classified as hepatocellular type, 32 (45.1%) as cholestatic type, and 22 (31.0%) as mixed type hepatotoxicity.
The median age, sex, and incidence of tumor liver metastasis was similar between the ILICI group and the 514 patients without liver injury (Table 1). The median sintilimab use cycle in the ILICI group was two (range: 1–16), and the cumulative dose was 400 mg, which differed significantly from the non-ILICI group (p<0.0001). Among baseline laboratory indicators, red blood cell, platelet, eosinophils, and basophils counts, as well as hemoglobin, liver function indicators (ALT, AST, ALP, GGT, and total bilirubin), albumin were significantly different (p<0.05).
Table 1Participant baseline epdemographic and disease characteristics
Variable | All cases, n=585 | ILICI, n=71 | Non-ILICI, n=514 | p-valuea |
---|
Age in years, mean (range) | 60.0 (12–86) | 55.4 (14–75) | 58.1 (12–86) | 0.708 |
Sex as male/female, n | 450/135 | 57/14 | 393/121 | 0.474 |
Primary cancer type, n | | | | 0.031 |
NSCLC | 198 | 10 | 185 | |
esophageal cancer | 159 | 23 | 134 | |
gastric cancer | 119 | 22 | 97 | |
lymphoma | 56 | 11 | 45 | |
other cancers | 26 | 2 | 23 | |
intestinal cancer | 18 | 3 | 14 | |
liver cancer | 9 | 0 | 9 | |
Liver metastasis as yes/no, n | 37/548 | 6/65 | 31/478 | 0.547 |
Cycle of liver damage | 2.0 (1–34) | 2.5 (1–16) | / | / |
Laboratory index, mean (range) | | | | |
White blood cells as ×109/L | 5.76 (0.01–92.55) | 7.12 (0.01–63.52) | 6.93 (0.11–65.33) | 0.152 |
Red blood cells as × 109/L | 3.83 (0.02–54.1) | 3.95 (0.03–44.1) | 4.44 (0.02–54.1) | <0.001 |
Hemoglobin in g/L | 116 (1–205) | 105.45 (1–166) | 116.62 (1–205) | <0.001 |
Platelets as × 109/L | 200 (0–1,373) | 204.98 (0–769) | 221.59 (0–1,373) | <0.001 |
Monocyte count, % | 6.9 (0–70.8) | 7.63 (0–70.8) | 7.54 (0–56) | 0.423 |
Lymphocyte count, % | 22.2 (0–100) | 22.73 (0–100) | 24.66 (0–97.8) | <0.001 |
Eosinophil count, % | 1.5 (0–58.8) | 2.60 (0–42.7) | 2.21 (0–58.8) | <0.001 |
Basophil count, % | 0.4 (0–5.6) | 0.49 (0–5.6) | 0.51 (0–4.3) | 0.270 |
ALT in IU/L | 18 (1–1,375) | 43.33 (4–1,375) | 22.29 (1–790) | <0.001 |
AST in IU/L | 21 (2–2,064) | 45.96 (2–5,215) | 25.05 (2–948) | <0.001 |
ALP in IU/L | 76 (21–725) | 104.08 (33–1,328) | 84.01 (18–1,959) | <0.001 |
GGT in IU/L | 25 (5–908) | 78.92 (7–1,728) | 36.95 (4–1,533) | <0.001 |
Total bilirubin in IU/L | 9 (0.7–9.0) | 11.95 (0.8–9.0) | 10.52 (0.7–9.0) | <0.001 |
Direct bilirubin in IU/L | 2.6 (0–116) | 4.30 (0–112.8) | 3.13 (0–125) | <0.001 |
Indirect bilirubin in IU/L | 7.1 (0.1–9.0) | 9.29 (0.1–8.0) | 9.63 (0.5–9.0) | 0.172 |
Albumin in IU/L | 39.2 (17.2–56.6) | 37.81 (15.3–54.6) | 38.42 (17.2–56.6) | <0.001 |
Creatinine in µmol/L | 65 (19–561) | 69.89 (22–813) | 70.55 (12–2,454) | 0.663 |
Lifestyle | | | | |
Smoking, n (%) | 334 (57.1) | 35 (49.3) | 297 (58.6) | 0.445 |
Drinking, n (%) | 271 (57.1) | 31 (43.7) | 240 (46.7) | 0.984 |
Underlying disease, n (%) | | | | |
Hypertension | 90 (15.4) | 7 (9.9) | 83 (16.1) | 0.078 |
Diabetes mellitus | 32 (5.5) | 4 (5.6) | 28 (5.4) | 0.687 |
Coronary heart disease | 9 (1.5) | 2 (2.8) | 7 (1.4) | 0.381 |
Hepatitis | 35 (6.0) | 4 (5.6) | 31 (6.0) | 0.321 |
Stroke | 10 (1.7) | 4 (5.6) | 6 (1.2) | 0.885 |
Hepatic cyst | 45 (7.7) | 7 (9.9) | 38 (7.4) | 0.346 |
Abnormal lipids metabolism | 28 (4.8) | 4 (5.6) | 24 (4.7) | 0.063 |
NAFLD | 14 (2.4) | 1 (1.4) | 13 (2.5) | 0.308 |
Choledocholithiasis | 16 (2.7) | 5 (7.0) | 11 (2.1) | 0.221 |
Cholecystitis | 18 (3.1) | 6 (8.5) | 17 (3.3) | 0.200 |
Other characteristics | | | | |
Number of hospital visitations, mean (range) | 52 (1–541) | 69 (3–233) | 31 (1–541) | 0.052 |
Number of hospitalizations, mean (range) | 22 (1–77) | 29 (2–65) | 12 (1–77) | 0.085 |
Tumor surgical history, n (%) | 145 (24.8) | 19 (26.8) | 126 (24.5) | 0.640 |
History of blood disease, n (%) | 4 (0.7) | 1 (1.4) | 3 (0.6) | 0.430 |
Family medical history (grade 1/2), n | 64/2 | 8/0 | 55/2 | 0.923 |
Relapse or not, n (%) | 18 (3.1) | 3 (4.2) | 15 (2.9) | 0.673 |
PD-1 cycle, mean (range) | 2 (1–34) | 2 (1–16) | 2 (1–34) | <0.0001 |
Accumulated dose (mg), mean (range) | 600 (0–12,800) | 400 (200–5,000) | 600 (0–12,800) | <0.0001 |
Independent risk factors associated with the development of hepatotoxicity in the training cohort
The samples were randomly divided into training and verification sets in a 6:4 ratio. Correlation analysis revealed no relationship between the variables of the training set (Fig. 2). Stepwise iterative logistic regression identified three independent risk factors associated with ILICI, namely dyslipidemia (odds ratio [OR]=2.546, 95% confidence interval [CI]=1.246–5.202, p<0.001), thyroid peroxidase (TPO) antibodies (OR=22.992, 95% CI=1.749–302.272, p=0.018), and hypoproteinemia (OR=4.111, 95% CI=1.733–9.755, p=0.001). Based on these risk factors, a nomogram model for predicting ILICI risk (ILICI-nomogram) was established for patients using sintilimab (Fig. 3A). Scores corresponding to the prognostic index were added to obtain the total score, and the ILICI risk probability was obtained from the total score. After 1,000 internal validations of bootstrap self-sampling, the model showed a C-index value of 0.713 (95% CI=0.632–0.794), indicating that the predictions obtained from this nomogram were consistent with the real observations and that the nomogram sensitivity met the standard. The calibration chart was in good agreement between the observed and predicted probability of ILICI risk. Decision curve analysis further showed that the nomogram had the largest net benefit over any single factor, indicating the best clinical diagnostic value for a comprehensive nomogram (Fig. 3B). The p-value of the goodness-of-fit was 0.9659, which suggested that the model was reasonable. We used the verification set and the full sample to verify the model, and the AUCs were 0.563 and 0.652, respectively (Supplementary Table 1 and Supplementary Fig. 1).
Application and evaluation of the ILICI-nomogram in grade ≥3 and ≥4 ILICI cohorts
The C-index of the grade 3 and 4 ILICI cohort was 0.752 (95% CI=0.627–0.878). The C-index of the ≥grade 4 ILICI cohort was 0.811 (95% CI=0.554–1.000). The calibration curve showed that the predicted incidence was close to the real incidence, and the nomogram agreed with the real observed outcomes (Fig. 4). ROC analysis further showed that the optimal cutoff value of the model was 10.0%; under this threshold, the decision curve of the model was above the none and all lines, indicating that the model had certain clinical practicability.
Subgroup analysis
An independent assessment of the predictive effect of the ROC model was also performed in some important subgroups of patients (Table 2). The model showed better ILICI predictive potential in males and patients ≥65 years. The C-indices for patients ≥65 years of age and men were 0.746 (95% CI=0.678–0.892) and 0.723 (95% CI=0.639–0.890), respectively (Fig. 5)
Table 2Subgroup receiver operating characteristic analysis
Subgroup | Number of patients | Area under the curve |
---|
Age in years | | |
≥65 | 101 | 0.7458 |
<65 | 250 | 0.6925 |
Sex | | |
Female | 266 | 0.6296 |
Male | 85 | 0.7230 |
Cancer type | | |
Non-small cell lung cancer | 114 | 0.5810 |
Esophageal cancer | 99 | 0.6281 |
Gastric cancer | 72 | 0.6281 |
Lymphoma | 36 | 0.6846 |
Other cancers | 13 | 0.6250 |
Intestinal cancer | 10 | 0.6500 |
Correlation between baseline characteristics and ILICI
The relationship between baseline clinical features, various laboratory indicators, and ILICI were also evaluated (Supplementary Table 2). In the ILICI cohort, patients with positive thyroglobulin (p=0.004) and TPO (p<0.001) antibodies, dyslipidemia (p=0.035), hypoproteinemia (p=0.001), and renal insufficiency (p=0.011) were more likely to develop ILICI. Similarly, ILICI was more likely to occur in the grade ≥3 ILICI cohort with thyroglobulin antibodies (p=0.004), TPO antibodies (p=0.001), hypoproteinemia (p=0.030), and renal insufficiency (p<0.001). Female, liver metastases, thyroglobulin antibodies, antinuclear antibodies, coronary heart disease, hepatitis, dyslipidemia, choledocholithiasis, renal insufficiency, and cholecystitis were factors associated with the likelihood of developing grade ≥4 ILICI (all p<0.05). ROC analysis was performed that included the risk factors (Supplementary Fig. 2).
Clinical features in patients with ILICI
The median time from the start of immunotherapy to immune liver injury onset was 87 days, with large individual differences (1–828 days). We did not observe differences in the time to reach grade 2 or higher liver injury following stratification by the severity of liver injury (p=0.0784; Fig. 6). Of the 71 patients with liver injury, 34 were permanently discontinued, and 12 restarted sintilimab therapy; only six received corticosteroid therapy at a dose of 40 mg/qd, and their liver function indicators eventually returned to normal. Thirty-four patients received hepatoprotectors along with conventional hepatoprotective drugs, usually in combinations of two or three, and the drugs of choice are butyl disulfonic acid, compound glycyrrhizin, glutathione, polyene phosphatidylcholine, and monoammonium cysteine glycyrrhizinate.
Discussion
Risk factors and the incidence of ILICI were evaluated in cohorts of patients with cancer who received sintilimab treatment. In addition, a diagnostic, intuitive, and personalized ILICI-nomogram was developed and verified. Finally, hypoproteinemia, dyslipidemia, and TPO antibodies were identified as independent risk factors for ILICI.
The activation of regulatory T cells against cytotoxic T lymphocytes contributes to all irAEs related to immune checkpoint inhibitors; however, the potential signaling pathway underlying hepatocyte damage and ILICI remains unclear. The main possible mechanisms are based on changes in self-tolerance and T-cell-mediated immune system activation.10 Previous studies showed that the PD-1 pathway helps the liver protect itself from immune-mediated destruction and that PD-1 or PD-L1 inhibitors restore the antitumor immunity of cytotoxic T lymphocytes by blocking the interaction between PD-1 and PD-L1. However, activated cytotoxic T lymphocytes respond to the liver.11–13 Therefore, increased expression of PD-L1 may play a role in immune-mediated hepatotoxicity. PD-L1 expression was reported to be significantly up-regulated in patients with high body fat and that there are more T helper 1 and 17 cells in adipose tissue. Furthermore, dyslipidemia can promote the secretion of pro-inflammatory cytokines (such as interleukin-1β, monocyte chemoattractant protein-1, and tumor necrosis factor-α) by immune cells, which promote inflammation and liver tissue damage.14 Therefore, it is reasonable to consider hyperlipidemia as a risk factor for immune-mediated hepatotoxicity.
Sharma et al.15 found that the incidence of drug-induced liver injury in patients with hypoproteinemia before antituberculosis drug treatment was 2.3-fold that in nonhypoproteinemia patients. This study had similar findings, with a higher risk of ILICI, a type of drug-induced liver injury, observed in patients with hypoproteinemia. Furthermore, hypoproteinemia can reflect the degree of inflammation in the body and predict the release of various inflammatory markers. In various solid tumors, systemic inflammatory markers play an important role in the development of irAEs.16 Moreover, when plasma albumin content decreases, the properties of the immune checkpoint inhibitor binding site change, reducing the binding of the drug to plasma proteins, which results in slower drug elimination and longer biological half-life, thus increasing its toxicity to the body.17 Therefore, patients with hypoproteinemia should be closely monitored for ILICI.
Previous studies have shown that patients with thyroglobulin antibodies in the serum are at a considerable risk of developing thyroid dysfunction during immunotherapy. Similarly, patients with autoimmune antibodies and rheumatoid factor who receive immunotherapy are at increased risk of irAEs.18,19 Our study had similar findings. However, it is unclear why these antibodies are associated with irAEs. A possible reason is that PD-1 is regulated by T cell-independent and -dependent mechanisms, and therefore it is expressed in large quantities in activated B cells.20 T cells enhance the therapeutic effect of anti-PD-1 antibodies and B cells are induced to produce autoantibodies, which enhance irAEs incidence.21,22 Therefore, the presence of autoantibodies, such as antinuclear, TPO, and thyroglobulin antibodies, may be related to therapeutic efficacy and irAEs. This suggests that future research should focus on developing biomarkers to predict the onset, treatment outcome, and resolution of liver injury.
To date, risk factors for immune-mediated hepatotoxicity have been rarely studied. Cho et al.23 showed an increased association between 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitor use and ILICI development, with a 4.7-fold increase in risk and a 78.8% attributable risk. Tsung et al.24 reported that liver injury was significantly more common in patients with liver metastasis (53% vs. 21%). However, they focused only on the analysis of influencing factors without building a prediction model. Therefore, in this study, risk factors were reassessed and an ILICI-nomogram that included patient history, clinicopathological features, peripheral blood markers, and combination therapy was created. We incorporated multiple risk factors into an easily used nomogram that conforms to the methodology criteria for prediction models, ultimately incorporating three new variables. Overall, the designed ILICI-nomogram showed a good C-index, which could be robustly applied to patients with grade ≥3 liver injury and different subgroups of patients.
The nomogram is helpful for the early identification of high-risk groups for ILICI among sintilimab-treated patients, providing exciting potential for clinical application. However, the stability of the proposed prediction model is related to the number of covariates and outcome events. Furthermore, model accuracy was determined using an internal verification approach; therefore, the insufficient sample size could lead to unsatisfactory validation set results. We aim to collect multicenter external validation data in future research. Taken together, this study shows that ILICI incidence in patients receiving sintilimab is high, with risk factors including hypoproteinemia, TPO antibodies, and dyslipidemia. The developed ILICI-nomogram herein can assist oncologists to develop better and more personalized treatment strategies for patients receiving immunotherapies.