The Predictive Model Construction for Immune-Related Adverse Events in Non-Small Cell Lung Cancer Patients Receiving Immunotherapy

Abstract

Introduction

It is possible to predict immune-related adverse events (irAEs) in the treatment of immune checkpoint inhibitors (ICIs) based on clinical and hematological parameters. Nevertheless, the specific parameters which can predict irAEs are still in the exploration. The purpose of this retrospective study was to develop a predictive model for irAEs in non-small cell lung cancer (NSCLC) patients in the treatment of ICIs.

Methods

Researchers enrolled NSCLC patients treated with at least 1 type of ICIs at Harbin Medical University Cancer Hospital between January 30, 2019 and December 31, 2021. Baseline parameters including demographic, clinicopathology, treatment information, and peripheral blood markers were selected retrospectively. Type, onset time, grade, and treatment of irAEs were also assessed. By analyzing the risk factors for irAEs, an irAEs prediction model was established using univariate and multivariate logistic regression.

Results

In a total of 484 patients, 81 patients experienced 112 irAEs in which thyroid dysfunction was the most common irAE (n = 38, 33.9%) and ICI pneumonitis was the most serious irAE (n = 6, 33.3%). Finally, a prediction model based on lines and combination therapy of ICIs, ECOG performance status, neutrophils/lymphocytes ratio (NLR), platelet (PLT), and lymphocyte (LYM) was established. Multivariate logistic regression analysis showed that 2 or ≥3 lines of immunotherapy, ICIs combination therapy, and ECOG PS 1–2 were independent risk factors for irAEs. Baseline LYM was positively associated with irAEs (OR = 2.599, P = 0.048) while baseline NLR and PLT were negatively associated with irAEs (OR = 0.392, P = 0.047; OR = 0.992, P = 0.035, respectively). The model showed great prediction performance with the AUC value of 0.851 and 0.779 in the training cohort and validation cohort, respectively.

Conclusion

Our study identified the risk factors related to irAEs occurrence and constructed and assessed the predictive model of irAEs in patients with NSCLC treated by ICIs using clinical and hematological parameters, thus guiding clinicians to select precisely the population receiving immunotherapy and develop individualized treatment therapy.

Keywords

non-small cell lung cancer immune-related adverse events predictive model immunotherapy

Introduction

Lung cancer has been characterized by the highest incidence and mortality worldwide, with an estimated 2.2 million newly diagnosed cases and 1.8 million deaths worldwide.¹ As the most frequent type of lung cancer, non-small cell lung cancer (NSCLC) can be treated with a series of therapeutic modalities such as radiation therapy, chemotherapy, targeted therapy, and surgery.^2,3 Nevertheless, the probability of treatment failure is still high.⁴ Recently, immune checkpoint inhibitors (ICIs) have greatly increased therapeutic efficacy and changed treatment paradigms in the management of NSCLC.⁵

Over the past few years, a series of phase III clinical trials have observed the durable therapy responses and great improvement in OS of patients with ICIs or combined platinum-based chemotherapy compared with chemotherapy alone, thus ICIs have become the first-line treatment options for patients with advanced NSCLC without sensitizing mutation site.^6–8 The ICIs in clinical application are mainly antibodies that block the inhibitory signaling molecules expressed on immune cells or tumor cells such as PD-1, CTLA-4, or PD-L1.⁹ Currently, approved ICIs in clinical application mainly include PD-1 inhibitors such as pembrolizumab and nivolumab, PD-L1 inhibitors such as atezolizumab and durvalumab, and the CTLA-4 inhibitor such as ipilimumab.^6–8,10

The use of ICIs may cause hyperactivation of the immune system against normal tissues and immune-mediated toxicity, which are also referred as immune-related adverse events (irAEs).¹¹ Specifically, these agents can enhance T-cell activation and proliferation, weaken Treg cell functions, and increase autoimmune response, thus resulting in multiple irAEs.¹² Based on the diverse mechanisms from above, the affected organs, onset time, and incidence of irAEs are highly heterogeneous. The organs including the skin, lung, colon, endocrine system, and liver are the most frequently affected by irAEs.¹¹ In addition, majority of irAEs initiate from weeks to years after ICIs administration, while others may occur within several years even after the whole treatment.¹³ The incidence of irAEs depends on the ICIs type and combination therapy, with a lower incidence treated with anti-PD-1/PD-L1 antibodies compared to anti-CTLA-4 antibodies and the highest incidence treated with anti-PD-1 + anti-CTLA-4 treatment.¹⁴ Studies have suggested that many factors can be regarded as biomarkers to assess the risk of irAEs, such as host parameters, ICIs type and dose, tumor immune environment, and gut microbiome.¹⁵ Since conducting a lesion biopsy to acquire immunomic-related data is a great challenge, most studies use more accessible data including clinical and hematological factors to predict the occurrence of irAEs. For example, studies have shown that NSCLC patients with neutrophils/lymphocytes ratio (NLR) < 3 and platelets/lymphocytes ratio (PLR) < 180 at the baseline of immunotherapy were more likely to develop irAEs.¹⁶ Another study by Okada et al¹⁷ also found that ECOG PS ≥2 could be used as an independent predictive factor for interstitial lung disease, a serious type of irAEs.

With the extensive use of ICIs in NSCLC therapy, it is crucial to establish biomarkers of immunotherapy response to optimize patient benefit and minimize toxicities risk. As the biomarker with the greatest focus, high PD-L1 expression on tumor cells tends to indicate better clinical benefit, but PD-L1 testing alone is not sufficient. Recently, emerging studies have demonstrated the utility of tumor mutational burden (TMB) in predicting ICI benefits, regardless of PD-L1 level.¹⁸ Immune gene signatures, T-cell receptor clonality, oncogenic driver mutation, and tumor-infiltrating lymphocytes (TILs) within the tumor microenvironment are also being investigated as they are more likely to affect immunotherapy response.^19,20 However, the application of these biomarkers is still limited by many problems, such as the difficulty in obtaining optimal cutoff value, high technical difficulty, and the complexity of the tumor immune microenvironment, which needs to be deeply investigated and validated. A healthy positive correlation between onset of irAEs and response to immunotherapy has been identified in multiple of studies. In a single-center retrospective study of 155 patients receiving monotherapy with a PD-1 inhibitor for advanced NSCIC, dermal toxicity was associated with higher objective response rate (ORR) and disease control rate (DCR; ORR:57% vs 19%, P < 0. 001; DCR: 94% vs 55%, P < 0.001).²¹ Similarly, a retrospective study analyzing data from 48 patients with advanced NSCLC in KEYNOTE-001 found that thyroid function abnormality was significantly associated with the prolongation of OS (40 months vs 14 months, HR = 0.29, 95%CI: 0.09-0.94, P = 0.04).²² Moreover, there is better tolerance with PD-1 and PD-L1 inhibitors than CTLA-4 inhibitors, and CTLA-4 inhibitors were more likely to cause grade 3 or 4 irAEs reported by a systematic review.^23,24 Therefore, it is urgent for us to develop a better understanding of irAE onset and explore potential biomarkers of irAE development.

In our retrospective study, we developed and validated successfully a predictive model to evaluate the incidence of irAEs collecting clinical and hematological parameters from real-world evidence and further explored the type, onset time, grade, and treatment of irAEs. Besides, we also investigated the correlation of irAEs with efficacy in ICI treatment.

Materials and Methods

Patients Enrollment

A total of 522 NSCLC patients were enrolled in our study. The inclusion criteria for NSCLC patients were as follows: pathologically confirmed NSCLC between January 30, 2019 and December 31, 2021, age ≥18 years, treated with anti-PD-1 antibody (nivolumab, pembrolizumab, camrelizumab, sintilimab, tislelizumab, toripalimab, etc) or anti-PD-L1 antibody (atezolizumab, durvalumab) combined or not with anti-CTLA-4 antibody (ipilbuzumab) at least 1 cycle until disease progression, toxicity intolerance, withdrawal, or death from the registry data in Department of Medical Oncology, Harbin Medical University Cancer Hospital. The exclusion criteria included the following: incomplete baseline demographic, clinicopathology and therapeutic information, autoimmune disease history, liver and kidney dysfunction, and acute infections. Written informed consent from patients that medical data during the treatment process could be used for scientific research was confirmed before ICI treatment. The flow diagram of the patients’ enrollment is presented in Figure 1. The protocol in this study was approved by the ethics committee of Harbin Medical University Cancer Hospital on October 28, 2020 with the approval number 2020-140-IIT. The reporting of this study conforms to the TRIPOD guidelines.²⁵

Figure 1.

Flow diagram of patients’ enrollment.

Data Collection

We collected retrospectively baseline demographic (sex, age, BMI, smoking history), clinicopathology and treatment (stage, performance status, single or combination agent therapy, type and line of ICI treatment) information in our research. Baseline peripheral blood biomarkers comprised white blood cell (WC), lymphocytes (LYM), eosinophil (EOS), monocytes (MONO), neutrophils (NEU), basophils (BASO), platelet (PLT) counts, serum albumin level (ALB), and lactate dehydrogenase (LDH) as well as NLR, PLR, monocytes-to-lymphocyte ratio (MLR). All baseline parameters for enrolled NSCLC patients were selected within 1 week before ICI therapy. We also assessed optimal efficacy including complete response (CR), partial response (PR), stable diseases (SD), and progressive diseases (PD) in patients during ICI cycle. In addition, the overall survival of patients receiving ICIs was identified by the researchers. Besides, we used the median to perform simple interpolation for missing values.

Outcome Assessment

IrAEs were defined as adverse events causing immunological dysregulation in patients treated with ICIs such as thyroid dysfunction, ICI pneumonitis, rash, and myocarditis. Researchers who did not know about the relevant information in advance evaluated irAEs using the Common Toxicity Criteria for Adverse Events (CTCAE 4.0) of the National Cancer Institute during or after the treatment of ICIs. Data about irAEs including type, onset time, grade, and treatment of irAEs were obtained.

Statistical Analyses

Based on random seed, the NSCLC immunotherapy data was divided into the training cohort and validation cohort randomly to establish and verify the predictive model. We applied receiver operating characteristic curve (ROC) to identify the optimal cutoff value of NLR, PLR, and MLR and classified patients into 2 cohorts based on the cutoff value (NLR <3.2 and ≥ 3.2, respectively; PLR <143 and ≥ 143, respectively; MLR < 0.3 and ≥ 0.3, respectively). We also performed student's t-test and chi-square test to compare baseline characteristics between patients with irAEs and those without irAEs. Univariate and multivariate logistic regression analyses were used to confirm factors tightly linked to the occurrence of irAEs and included in the establishment of predictive model using nomogram through “rms” and “Dynnom” R package. ROC, calibration curves, and decision curve analyses (DCA) were applied to verify the predictive performance of model through “survivalROC” and “ggplot2” package. The Kaplan-Meier survival curve was performed to compare the OS between the irAEs and non-irAEs group using R packages “survival” and “survminer.” All statistical tests were disposed by R software (version 4.0.2), and P < 0.05 was considered statistically significant.

Results

Patient Characteristics

A total of 484 patients who were treated with at least 1 cycle ICIs at anonymized issue Cancer Hospital have been included. Most of the patients (n = 441, 91.12%) received anti-PD-1 antibody, 27 patients were administered anti-PD-L1 antibody, 5 patients with anti-PD-L1 + anti-CTLA-4 antibody, and 11 patients with anti-PD-1 + anti-CTLA-4 antibody. In total, 53.3% of patients were treated with ICIs monotherapy whereas other patients received ICIs combined therapy including chemotherapy, antiangiogenic therapy, and radiotherapy. Majority of the patients were given the first-line treatment of ICIs (62.4%) and had 0 to 1 ECOG performance status (92.15%). Patients with IV stage (75.41%), male (71.69%), age <65 (64.88%), BMI ≤23.9 (59.92%), and smoking history (50.41%) also had a larger proportion of the total population. In addition, the NSCLC immunotherapy data was divided into the training cohort (n = 339) and validation cohort (n = 145) randomly to establish and verify the predictive model according to random seed. The baseline characteristics of enrolled patients were compared between the training and validation cohort, and no statistical significance was found (P > 0.05; Table 1).

Table 1.

Baseline Characteristics of Enrolled Patients in Overall, Training, and Validation Cohort.

Baseline characteristics	Overall cohort (n = 484)	Training cohort (n = 339)	Validation cohort (n = 145)	P
Stage				0.0705
III	119 (24.59)	75 (22.12)	44 (30.34)
IV	365 (75.41)	264 (77.88)	101 (69.66)
Sex				0.2294
Male	347 (71.69)	249 (73.45)	98 (67.59)
Female	137 (28.31)	90 (26.55)	47 (32.41)
Age				0.5931
<65	314 (64.88)	223 (65.78)	91 (62.76)
≥65	170 (35.12)	116 (34.22)	54 (37.24)
BMI				0.9668
≤23.9	290 (59.92)	202 (59.59)	88 (60.69)
24–27.9	161 (33.26)	114 (33.63)	47 (32.41)
>27.9	33 (6.82)	23 (6.78)	10 (6.90)
Smoking				0.475
No	240 (49.59)	164 (48.38)	76 (52.41)
Yes	244 (50.41)	175 (51.62)	69 (47.59)
ECOG PS				0.0559
0	159 (32.85)	119 (35.10)	40 (27.59)
1	287 (59.30)	199 (58.70)	88 (60.69)
2	38 (7.85)	21 (6.19)	17 (11.72)
Line of immunotherapy				0.6343
1	302 (62.40)	216 (63.72)	86 (59.31)
2	153 (31.61)	104 (30.68)	49 (33.79)
≥3	29 (5.99)	19 (5.60)	10 (6.90)
Type of immunotherapy				0.4802
Anti-PD-1	441 (91.12)	310 (91.45)	131 (90.34)
Anti-PD-L1	27 (5.58)	16 (4.72)	11 (7.59)
Anti-PD-L1 + anti-CTLA-4	5 (1.03)	4 (1.18)	1 (0.69)
Anti-PD-1 + anti-CTLA-4	11 (2.27)	9 (2.65)	2 (1.38)
Treatment				0.4238
I	258 (53.31)	188 (55.46)	70 (48.28)
I + A	34 (7.02)	22 (6.49)	12 (8.28)
I + C	183 (37.81)	124 (36.58)	59 (40.69)
Others	9 (1.86)	5 (1.47)	4 (2.76)
irAEs
Yes	81 (16.74)	60 (17.70)	21 (14.48)	0.4621
No	403 (83.26)	279 (82.30)	124 (85.52)

I: immunotherapy; I+A: immunotherapy+antiangiogenic therapy; I+C: immunotherapy+chemotherapy; neutrophil-tolymphocyte ratio (NLR) platelet-to-lymphocyte ratio (PLR) monocytes-to-lymphocyte ratio (MLR) white blood cell (WC) lymphocytes (LYM) eosinophil (EOS) monocytes (MONO) neutrophils (NEU) serum albumin level (ALB) platelet (PLT) basophils (BASO)lactate dehydrogenase (LDH).

Immune-Related Adverse Events

As shown in Table 2, a total of 81 patients experienced 112 irAEs of any grade and 18 irAEs of grade 3 or higher. Thyroid dysfunction was the most common irAEs with the incidence rate of 33.9%, followed by ICI pneumonitis (21.4%), rash (17.9%), and myocarditis (16.7%; Figure 2). Notably, ICI pneumonitis was the most serious adverse event with the proportion of 33% in all grade ≥3 irAEs. A total of 34.8% of patients who developed irAEs received corticosteroid treatment, while the remaining patients were treated with symptomatic treatment and immunosuppressants as recommended by CSCO, ASCO, and NCCN guidelines for irAEs management. Most patients with irAEs showed symptom relief after irAEs treatment, and 61.6% (n = 69) of them underwent immunotherapy retreatment (Table 2). We also investigated the onset time of various irAEs and found that the average time of irAEs occurrence was 0.3 to 122 weeks, which showed high variability among them (Figure 2). Furthermore, the relationship of irAEs with efficacy was explored. In total, 59.3% and 52.1% of irAEs and non-irAEs patients showed CR or PR, respectively, although there was no significance found between the 2 groups (P > 0.05; Figure 3). Kaplan-Meier survival curve has revealed that compared with the non-irAE group, the overall survival of patients in the irAE group was longer (P < 0.05), tending to have a better prognosis (Figure 3).

Figure 2.

(A) Distribution characteristics of irAEs. (B) Median time for the development of different irAEs.

Figure 3.

(A) Association of irAEs with efficacy of immunotherapy. (B) Kaplan-Meier survival curve of NSCLC with the irAE and non-irAE group.

Table 2.

Immune Checkpoint Inhibitor-Related Adverse Effects.

irAEs events	Any grade	Grade ≥3	Corticosteroid treatment	Retreatment with ICIs
Any irAEs	112 (100.0)	18 (100.0)	39 (34.8)	69 (61.6)
Thyroid dysfunction	38 (33.9)	0 (0.0)	0 (0.0)	28 (73.7)
ICI pneumonitis	24 (21.4)	6 (33.3)	20 (83.3)	10 (41.7)
Rash	20 (17.9)	2 (11.1)	4 (20.0)	16 (80.0)
Myocarditis	7 (6.3)	3 (16.7)	7 (100.0)	2 (28.6)
Hepatitis	7 (6.3)	2 (11.1)	2 (28.6)	4 (57.1)
Colitis	5 (4.5)	2 (11.1)	1 (20.0)	3 (60.0)
Fatigue	3 (2.7)	0 (0.0)	0 (0.0)	3 (100.0)
Nephritis	3 (2.7)	1 (5.6)	2 (66.7)	1 (33.3)
Pain	1 (0.9)	0 (0.0)	0 (0.0)	1 (100.0)
Encephalitis	1 (0.9)	1 (5.6)	1 (100.0)	0 (0.0)
Myositis	1 (0.9)	1 (5.6)	1 (100.0)	0 (0.0)
Hypophysitis	1 (0.9)	0 (0.0)	1 (100.0)	NA
Allergy	1 (0.9)	0 (0.0)	0 (0.0)	1 (100.0)

Construction of a Predictive Model for IrAEs

The risk factors for irAEs were assessed by univariate binary logistic regression analysis in the training cohort. Then, variables including BMI, ECOG PS, line of immunotherapy, type of immunotherapy, treatment, NLR, PLR, MLR, LYM, and PLT were selected for inclusion in the multivariate logistic regression analysis based on statistical significance in univariate analysis (P < 0.05; Table 3). A final irAEs prediction model was constructed using the significant variables (P < 0.05) including ECOG PS, line of immunotherapy, treatment, NLR, LYM, and PLT in multivariate logistic regression analysis (Table 4). We can conclude from multivariate logistic regression analysis that patients with ECOG PS 1 to 2, 2 or ≥3 line of immunotherapy, and ICIs combined therapy are independent risk factors for irAEs. Baseline LYM are positively associated with irAEs (OR = 2.599, P = 0.048) while baseline NLR and PLT are negatively associated with irAEs (OR = 0.392, P = 0.047; OR = 0.992, P = 0.035, respectively; Table 4). A nomogram was established as a quantitative tool to predict the incidence of irAEs in NSCLC patients more precisely (Figure 4). Furthermore, the web-based nomogram (https://weitonggao.shinyapps.io/DynNomapp/) was also established by which you can select the corresponding variable options, and then click Predict on the page to get the probability of irAEs in NSCLC patients and its 95% confidence interval.

Figure 4.

Nomogram of predictive model in the training cohort.

Table 3.

Univariate Binary Logistic Regression Analysis.

Baseline characteristics	OR	95%CI	P
Stage
III	reference
IV	0.919	0.474–1.783	0.804
Sex
Male	reference
Female	0.907	0.477–1.722	0.765
Age
<65	reference
≥65	1.139	0.637–2.037	0.66
BMI
≤23.9	reference
24–27.9	1.246	0.678–2.29	0.48
>27.9	2.942	1.15–7.528	0.024
Smoking
No	reference
Yes	1.39	0.791–2.444	0.253
ECOG PS
0	reference
1	4.172	1.894–9.188	0
2	5.55	1.692–18.206	0.005
Line of immunotherapy
1	reference
2	2.679	1.463–4.906	0.001
≥3	5.556	2.041–15.128	0.001
Type of immunotherapy
Anti-PD-1	reference
Anti-PD-L1	1.145	0.315–4.161	0.837
Anti-PD-L1 + anti-CTLA-4	1.654	0.169–16.213	0.666
Anti-PD-1 + anti-CTLA-4	3.969	1.031–15.283	0.045
Treatment
I	reference
I + A	1.767	0.546–5.72	0.342
I + C	2.766	1.508–5.072	0.001
Others	11.929	1.883–75.551	0.008
NLR
<3.2	reference
≥3.2	0.341	0.176–0.658	0.001
MLR
<0.3	reference
≥0.3	0.557	0.318–0.977	0.041
PLR
<143	reference
≥143	0.223	0.123–0.403	0
WC	1.023	0.931–1.122	0.64
LYM	3.122	1.855–5.255	0
EOS	1.275	0.516–3.152	0.598
MONO	0.96	0.755–1.22	0.739
NEU	0.976	0.865–1.101	0.69
ALB	1.013	0.945–1.085	0.726
PLT	0.994	0.99–0.998	0.003
BASO	0.089	0–1237.236	0.619
LDH	1	0.998–1.003	0.774

Table 4.

Multivariate Binary Logistic Regression Analysis.

Baseline characteristics	OR	95%CI	P
BMI
≤23.9	reference
24–27.9	1.095	0.511–2.347	0.816
>27.9	3.2	0.973–10.527	0.056
ECOG PS
0	reference
1	5.272	2.106–13.2	0
2	5.413	1.13–25.922	0.035
Line of immunotherapy
1	reference
2	4.023	1.879–8.613	0
≥3	8.477	2.25–31.93	0.002
Type of immunotherapy
Anti-PD-1	reference
Anti-PD-L1	1.113	0.181–6.857	0.908
Anti-PD-L1 + anti-CTLA-4	8.354	0.645–108.145	0.104
Anti-PD-1 + anti-CTLA-4	5.095	0.959–27.077	0.056
Treatment
I	reference
I + A	2.504	0.611–10.26	0.202
I + C	2.785	1.324–5.857	0.007
Others	46.811	3.706–591.334	0.003
NLR
<3.2	reference
≥3.2	0.392	0.155–0.988	0.047
PLR
<143	reference
≥143	0.69	0.229–2.078	0.51
MLR
<0.3	reference		0.525
≥0.3	1.301	0.578–2.93
LYM	2.599	1.009–6.689	0.048
PLT	0.992	0.985–0.999	0.035

Evaluation and Validation of a Predictive Model for irAEs

Figure 5A exhibits the ROC curve, and the area under the curve (AUC) was 0.851 (95% confidence interval: 0.804-0.899). The corresponding sensitivity and specificity were 86.7% and 69.2%, respectively. As depicted in the calibration plots, there was excellent agreement between the observed and predicted probabilities, demonstrating the superiority of the nomograms (Figure 6A). In decision curve analysis (DCA), the net benefit of model was significantly higher than all positive and negative curves, showing high reliability (Figure 7A). In order to validate the prediction model, the above parameters were enrolled into the validation cohort, and ROC curves were drawn with the AUC of 0.779 (95% CI: 0.690-0.867). The corresponding sensitivity and specificity were 95.2% and 54.8%, respectively (Figure 5B). Furthermore, the calibration and DCA plots also exhibited good predictive efficiency in the validation cohort, which was consistent with that in the training cohort (Figures 6B and 7B).

Figure 5.

ROC curve analysis of the predictive model. (A) Training cohort: AUC = 0.851, sensitivity 86.7%, specificity 69.2%. (B) Validation cohort: AUC = 0.779, sensitivity 95.2%, specificity 54.8%.

Figure 6.

Calibration plots showed the agreement between predicted and observed probabilities of the predictive model in the training cohort (A) and validation cohort (B).

Figure 7.

DCA curve showed the net benefit of model was significantly higher than all positive and negative curves in training cohort (A) and validation cohort (B).

Discussion

Over the past few decades, immunotherapy dominated by ICIs including anti-PD-1, anti-PD-L1, and CTLA-4 antibody has dramatically altered the landscape of NSCLC treatment.²⁶ However, with increased application of ICIs, many irAEs have emerged that impose adverse effects on patients’ quality of life and long-term outcomes.²⁷ Thus, it is important for clinicians to early identify patients with high risk of irAEs to reduce or prevent the occurrence of severe irAEs. Emerging studies confirmed irAEs-related biomarkers in multiomics data, such as genomics, epigenomics, and transcriptomics, from tumor samples that was not easily accessible.^28,29 In contrast, baseline demographic, clinicopathology, and treatment information in combination with laboratory test results are the most convenient and available predictive factors. Recently, Zhao et al³⁰ established successfully a nomogram including interleukin (IL)-2, IL-8, hepatic metastases, and duration of time to predict the occurrence of irAEs in cancer patients receiving ICIs. As a result, based on clinical and hematological parameters from real-world evidence, a predictive model was established to early identify and reduce the onset of irAEs in our study.

In our research, 81 patients experienced 112 irAEs treated with ICIs in which thyroid dysfunction was the most common irAEs (n = 38, 33.9%), followed by ICI pneumonitis (n = 24, 21.4%), and rash (n = 20, 17.9%). Similar with above, data from a systematic review showed that pneumonitis, hypothyroidism, arthralgia, and vitiligo were more common with patients treated by PD-1 inhibitors.²⁴ One meta-analysis also found that arthralgia and pneumonitis are more likely to develop with pembrolizumab, endocrine toxicities with nivolumab, and hypothyroidism with atezolizumab.³¹ Besides, as the most commonly fatal irAE treated with anti-PD-1/PD-L1 antibodies,³² 33% of ICI pneumonitis was graded ≥3 by our researchers. It has been demonstrated that most of irAEs can occur within 2 to 16 weeks after the initial ICI therapy.³³ Nevertheless, some irAEs are likely to occur within a few days at the beginning of ICIs, and some even occur up to 1 year after the end of therapy.^34,35 Similarly, the onset time of ICIs in our study was highly heterogeneous with the range of 0.3 to 122 weeks. As glucocorticoids are regarded as the mainstay of ICIs treatment with the exception of endocrine irAEs,³⁶ 39 out of 112 (34.8%) patients received glucocorticoid therapy in our study.

To better predict the immunotherapy response of ICIs, we investigated the correlation of irAEs occurrence with therapeutic efficacy. We found that compared with the non-irAEs patients, a larger percentage of the irAEs patients showed a CR or PR. It was also revealed that the overall survival of patients in the irAEs group was significantly longer than that of the non-irAEs patients, tending to have a better prognosis. This may owe to the mechanism of irAEs onset that ICIs can activate several key cellular events, such as activation and proliferation of T cells, inhibitory function of Treg cells, the induction of cross-reactivity between antigens on normal tissues and antitumor T cells, the production of cytokines as well as antibody.¹² Therefore, it is reasonable for us to infer that once the antitumor immune response was stimulated by receiving ICIs, the activation of the immune system could also lead to the development of irAEs. In the future, we will deeply explore the influence of irAE occurrence time and type as well as induced by different types of ICIs on treatment efficacy.

Next, we evaluated the demographics, clinical characteristics, and peripheral blood biomarkers in NSCLC patients treated with ICIs and identified the risk factors related to irAEs. Ultimately, 6 parameters including lines and combined therapy of ICIs, ECOG PS, baseline NLR, PLT, and LYM were integrated into the construction of predictive model for irAEs using univariate and multivariate logistic regression analysis. The AUC of ROC curve in the training cohort was 0.851 (95% confidence interval: 0.804-0.899). The corresponding sensitivity and specificity were 86.7% and 69.2%, respectively. In the validation cohort, ROC curve was drawn with the AUC of 0.779 (95% CI: 0.690-0.867). The corresponding sensitivity and specificity were 95.2% and 54.8%, respectively, showing a good predictive ability.

An increasing use of combination therapeutic regimens (ICIs combined with traditional cancer therapy, such as chemotherapy, antiangiogenic therapy, and radiotherapy) might improve the effectiveness of the treatment but amplify irAEs, which was consistent with our finding.³⁷ Several preclinical studies have demonstrated that most chemotherapeutic drugs stimulate the immune system by reducing the density of immunosuppressive cells and increasing the infiltration of T cells, thus enhancing immunogenicity.³⁸ For example, neoadjuvant FOLFIRINOX can increase CD8 + T-cells amount and decrease M2 macrophage and regulatory T-cells amount to reactivate the immune system in pancreatic ductal adenocarcinoma.³⁹ Similarly, antiangiogenic agents and immunotherapy exhibit synergistic effects in preclinical models that may be explained by the immune status regulation.⁴⁰ Radiotherapy can also induce immunostimulatory mechanisms that trigger the immune response especially when combined with immunotherapy.⁴¹ In our study, ICIs combined with chemotherapy and antiangiogenic agents (I + C + A) or ICIs combined with radiotherapy (I + R) were both included in other groups, which predicted high incidence of irAEs. Moreover, our multivariate regression analysis found that the incidence of irAEs was higher in ≥3 lines of therapy, reflecting high tumor burden.⁴² A retrospective study conducted by Sakata et al⁴³ showed that severe irAEs were correlated with a high tumor load among advanced NSCLC patients treated with ICIs. It could be explained by the mechanisms that tumor size can influence the intensity of the drug-induced T-cell response.⁴⁴ Besides, in agreement with our findings, a study by Okada et al¹⁷ identified the relationship between ECOG PS and lung irAEs in lung cancer patients. However, patients with multiple lines of treatments tend to have poor physical activity, showing a potential interaction between therapy lines and ECOG PS.

We also observed that baseline NLR (OR = 0.392, P = 0.047), LYM (OR = 2.599, P = 0.048), and PLT (OR = 0.992, P = 0.035) were associated with the risk of irAEs. The NLR and LYM are common biomarkers of systemic inflammation, which can be readily detected by peripheral blood analysis.⁴⁵ Inflammation can trigger immunosuppression and compromise anticancer immune response, thus reflecting a dysfunctional host immune status.⁴⁶ Liu et al⁴⁷ also found an increased risk of severe irAEs (grade 3/4) was linked to decreased baseline NLR and PLR among NSCLC patients receiving anti-PD-1 treatment. LYM was regarded as a risk factor for the occurrence of irAEs in patients with NSCLC treated by PD-1 inhibitors in a study by Isono et al.⁴⁸ As another component of peripheral blood, PLT have been recognized as immune cell to facilitate inflammation and immune response.⁴⁹ A recent study demonstrated that PLT may have immune suppressive role in tumor microenvironment by inhibiting the tumor lysis mediated by natural killing cells.⁵⁰ Consistent with our results, a decreased amount of PLT was found significantly in patients with thyroid dysfunctions under the immunotherapy.⁵¹

However, there are still some limitations in our study. Firstly, this was a retrospective study in which potential confounders could not be excluded. Secondly, emerging irAEs-related biomarkers such as chemokines and cytokines were not included in the establishment of our prediction model, which needs to be further explored.^30,52 Finally, the internal validation we applied in our study may not be sufficient compared to the external validation.

Conclusion

In conclusion, our study identified the risk factors related to irAEs occurrence and constructed and assessed successfully the predictive model of irAEs in patients with NSCLC treated by ICIs using clinical and hematological parameters from real-world evidence. Although it needs further investigation, the predictive model may be possibly incorporated into clinician decision-making and minimize adverse effects while optimizing the quality of life of the cancer patients receiving immunotherapy.

Footnotes

Abbreviations

Authors’ Note

W.G. and Q.L. contributed equally to this work. Y.Y. conceived the manuscript. W.G. wrote the initial draft of the manuscript and processed the data. Q.Q. collected and processed data. Y.Z. and M.Y. participated in the revision of the manuscript. All authors have read and approved the article.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Yan Yu

References

Sung

Ferlay

Siegel

, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209-249.

Herbst

Morgensztern

Boshoff

. The biology and management of non-small cell lung cancer. Nature. 2018;553(7689):446-454.

Arbour

Riely

. Systemic therapy for locally advanced and metastatic non-small cell lung cancer: A review. JAMA. 2019;322(8):764-774.

Marconi

Croce

Chiorino

, et al. A circulating risk score, based on combined expression of Exo-miR-130a-3p and fibrinopeptide A, as predictive biomarker of relapse in resectable non-small cell lung cancer patients. Cancers 2022;14(14):3412

Mok

Kudaba

, et al. Pembrolizumab versus chemotherapy for previously untreated, PD-L1-expressing, locally advanced or metastatic non-small-cell lung cancer (KEYNOTE-042): A randomised, open-label, controlled, phase 3 trial. Lancet. 2019;393(10183):1819-1830.

Reck

Rodríguez-Abreu

Robinson

, et al. Pembrolizumab versus chemotherapy for PD-L1-positive non-small-cell lung cancer. N Engl J Med. 2016;375(19):1823-1833.

Hellmann

Paz-Ares

Bernabe Caro

, et al. Nivolumab plus ipilimumab in advanced non-small-cell lung cancer. N Engl J Med. 2019;381(21):2020-2031.

Herbst

Giaccone

de Marinis

, et al. Atezolizumab for first-line treatment of PD-L1-selected patients with NSCLC. N Engl J Med. 2020;383(14):1328-1339.

Pardoll

. The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer. 2012;12(4):252-264.

10.

Gray

Villegas

Daniel

, et al. Three-year overall survival with durvalumab after chemoradiotherapy in stage III NSCLC-update from PACIFIC. J Thorac Oncol. 2020;15(2):288-293.

11.

Postow

Sidlow

Hellmann

. Immune-related adverse events associated with immune checkpoint blockade. N Engl J Med. 2018;378(2):158-168.

12.

Yang

Yao

Zhou

, et al. Immune-related adverse events of checkpoint inhibitors: Insights into immunological dysregulation. Clin Immunol. 2020;213:108377.

13.

Ghisoni

Wicky

Bouchaab

, et al. Late-onset and long-lasting immune-related adverse events from immune checkpoint-inhibitors: An overlooked aspect in immunotherapy. Eur J Cancer. 2021;149:153-164.

14.

Kennedy

Salama

. A review of cancer immunotherapy toxicity. CA Cancer J Clin. 2020;70(2):86-104.

15.

Jia

Geng

Jiang

, et al. The biomarkers related to immune related adverse events caused by immune checkpoint inhibitors. J Exp Clin Cancer Res. 2020;39(1):284.

16.

Pavan

Calvetti

Dal Maso

, et al. Peripheral blood markers identify risk of immune-related toxicity in advanced non-small cell lung cancer treated with immune-checkpoint inhibitors. Oncologist. 2019;24(8):1128-1136.

17.

Okada

Matsuoka

Sakurada

, et al. Risk factors of immune checkpoint inhibitor-related interstitial lung disease in patients with lung cancer: A single-institution retrospective study. Sci Rep. 2020;10(1):13773.

18.

Sholl

Hirsch

Hwang

, et al. The promises and challenges of tumor mutation burden as an immunotherapy biomarker: A perspective from the international association for the study of lung cancer pathology committee. J Thorac Oncol. 2020;15(9):1409-1424.

19.

Gibney

Weiner

Atkins

. Predictive biomarkers for checkpoint inhibitor-based immunotherapy. Lancet Oncol. 2016;17(12):e542-e551.

20.

Bodor

Boumber

Borghaei

. Biomarkers for immune checkpoint inhibition in non-small cell lung cancer (NSCLC). Cancer. 2020;126(2):260-270.

21.

Aso

Toi

Sugisaka

, et al. Association between skin reaction and clinical benefit in patients treated with anti-programmed cell death 1 monotherapy for advanced non-small cell lung cancer. Oncologist. 2020;25(3):e536-e544.

22.

Kim

Lee

, et al. Development of thyroid dysfunction is associated with clinical response to PD-1 blockade treatment in patients with advanced non-small cell lung cancer. Oncoimmunology. 2018;7(1):e1375642.

23.

Larkin

Chiarion-Sileni

Gonzalez

, et al. Combined nivolumab and ipilimumab or monotherapy in untreated melanoma. N Engl J Med. 2015;373(1):23-34.

24.

Khoja

Day

Wei-Wu Chen

, et al. Tumour- and class-specific patterns of immune-related adverse events of immune checkpoint inhibitors: A systematic review. Ann Oncol. 2017;28(10):2377-2385.

25.

Collins

Reitsma

Altman

, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement. Br Med J. 2015;350:g7594.

26.

Robert

. A decade of immune-checkpoint inhibitors in cancer therapy. Nat Commun. 2020;11(1):3801.

27.

Grangeon

Tomasini

Chaleat

, et al. Association between immune-related adverse events and efficacy of immune checkpoint inhibitors in non-small-cell lung cancer. Clin Lung Cancer. 2019;20(3):201-207.

28.

Bomze

Hasan Ali

Bate

, et al. Association between immune-related adverse events during anti-PD-1 therapy and tumor mutational burden. JAMA Oncol. 2019;5(11):1633-1635.

29.

Jing

Liu

, et al. Multi-omics prediction of immune-related adverse events during checkpoint immunotherapy. Nat Commun. 2020;11(1):4946.

30.

Zhao

Jiang

Shang

, et al. Construction and evaluation of clinical prediction model for immunotherapy-related adverse events and clinical benefit in cancer patients receiving immune checkpoint inhibitors based on serum cytokine levels. J Immunother. 2023;46(8):310-322.

31.

Chen

, et al. Comparative safety of immune checkpoint inhibitors in cancer: Systematic review and network meta-analysis. BMJ (Clinical Research ed.). 2018;363:k4226.

32.

Wang

Salem

Cohen

, et al. Fatal toxic effects associated with immune checkpoint inhibitors: A systematic review and meta-analysis. JAMA Oncol. 2018;4(12):1721-1728.

33.

Weber

D'Angelo

Minor

, et al. Nivolumab versus chemotherapy in patients with advanced melanoma who progressed after anti-CTLA-4 treatment (CheckMate 037): A randomised, controlled, open-label, phase 3 trial. Lancet Oncol. 2015;16(4):375-384.

34.

Yoest

. Clinical features, predictive correlates, and pathophysiology of immune-related adverse events in immune checkpoint inhibitor treatments in cancer: A short review. Immunotargets Ther. 2017;6:73-82.

35.

Parakh

Cebon

Klein

. Delayed autoimmune toxicity occurring several months after cessation of anti-PD-1 therapy. Oncologist. 2018;23(7):849-851.

36.

Ramos-Casals

Brahmer

Callahan

, et al. Immune-related adverse events of checkpoint inhibitors. Nature Reviews Disease Primers. 2020;6(1):38.

37.

Hoos

. Development of immuno-oncology drugs - from CTLA4 to PD1 to the next generations. Nat Rev Drug Discovery. 2016;15(4):235-247.

38.

Heinhuis

Ros

Kok

, et al. Enhancing antitumor response by combining immune checkpoint inhibitors with chemotherapy in solid tumors. Ann Oncol. 2019;30(2):219-235.

39.

Michelakos

Cai

Villani

, et al. Tumor microenvironment immune response in pancreatic ductal adenocarcinoma patients treated with neoadjuvant therapy. J Natl Cancer Inst. 2021;113(2):182-191.

40.

Tartour

Pere

Maillere

, et al. Angiogenesis and immunity: A bidirectional link potentially relevant for the monitoring of antiangiogenic therapy and the development of novel therapeutic combination with immunotherapy. Cancer Metastasis Rev. 2011;30(1):83-95.

41.

De Martino

Daviaud

Vanpouille-Box

. Radiotherapy: An immune response modifier for immuno-oncology. Semin Immunol. 2021;52:101474.

42.

Feng

Zhang

, et al. Prediction of immune-related adverse events in non-small cell lung cancer patients treated with immune checkpoint inhibitors based on clinical and hematological markers: Real-world evidence. Exp Cell Res. 2022;416(1):113157.

43.

Sakata

Kawamura

Ichikado

, et al. The association between tumor burden and severe immune-related adverse events in non-small cell lung cancer patients responding to immune-checkpoint inhibitor treatment. Lung Cancer. 2019;130:159-161.

44.

Huang

Postow

Orlowski

, et al. T-cell invigoration to tumour burden ratio associated with anti-PD-1 response. Nature. 2017;545(7652):60-65.

45.

Sun

Zhang

. The clinical use of pretreatment NLR, PLR, and LMR in patients with esophageal squamous cell carcinoma: Evidence from a meta-analysis. Cancer Manag Res. 2018;106:167-179.

46.

Taniguchi

Karin

. NF-κB, inflammation, immunity and cancer: Coming of age. Nat Rev Immunol. 2018;18(5):309-324.

47.

Liu

, et al. Peripheral blood markers associated with immune-related adverse effects in patients who had advanced non-small cell lung cancer treated with PD-1 inhibitors. Cancer Manag Res. 2021;137:65-71.

48.

Isono

Kagiyama

Takano

, et al. Outcome and risk factor of immune-related adverse events and pneumonitis in patients with advanced or postoperative recurrent non-small cell lung cancer treated with immune checkpoint inhibitors. Thorac Cancer. 2021;12(2):153-164.

49.

Best

Sol

In ‘t Veld

, et al. Swarm intelligence-enhanced detection of non-small-cell lung cancer using tumor-educated platelets. Cancer Cell. 2017;32(2):238-252.e9.

50.

Nieswandt

Hafner

Echtenacher

, et al. Lysis of tumor cells by natural killer cells in mice is impeded by platelets. Cancer Res. 1999;59(6):1295-1300.

51.

Kurimoto

Inaba

Ariyasu

, et al. Predictive and sensitive biomarkers for thyroid dysfunctions during treatment with immune-checkpoint inhibitors. Cancer Sci. 2020;111(5):1468-1477.

52.

Khan

Luo

, et al. Immune dysregulation in cancer patients developing immune-related adverse events. Br J Cancer. 2019;120(1):63-68.