1. The demographic and clinicopathological features
This study retrospectively analyzed 547 patients who were first treated with ICIs for malignant solid tumors in the first affiliated hospital of Xi'an jiaotong university from January 1 2019 to January 1 2021. Patients were screened according to strict inclusion and exclusion criteria. Ultimately, 31 patients were excluded due to lack of baseline imaging features, 19 due to ECOG score > 2, and 12 due to lack of records of distant metastatic sites. In addition, 11 patients were excluded for lack of a record of hypertension or diabetes, and 7 patients were excluded for primary tumors that were not in a single site. Five patients were ruled out because they were diagnosed with hematological tumors. Finally, a total of 489 patients were included, covering 13 types of malignant tumors including lung cancer (82.21%), liver cancer (6.74%), gastroesophageal cancer (4.91%), colorectal cancer (2.25%) and renal cancer (1.63%). Following the principle of randomness, patients were divided into training cohort and validation cohort in a ratio of approximately 2:1. A total of 327 patients were eventually included as the training cohort and 162 patients as the validation cohort. The demographic and clinicopathological features of the training cohorts and validation cohort were shown in Table 1.
Among the patients finally analyzed, the median age was 61 years (24-86 years), and a total of 91 patients developed CIP, with an overall incidence of 18.61%. Most of the CIP patients presented with grade 1-2, of which, 32 patients (6.54%) presented with grade 1, 39 patients (7.98%) with grade 2, and 20 patients (4.09%) with grade 3 or higher (Figure S1A). The average time to onset of CIP was 94 days after taking immunotherapy, meaning that most CIP occurred around 3 months after initiation of immunotherapy. Moreover, more than 80% of CIP patients occurred within 4 cycles of immunotherapy, especially within the first or second cycle. However, grade 3 and higher CIP occurs more often after the third to fourth cycle of immunotherapy, which seems to occur later than lower grade CIP (Figure S1B-C).
2. Baseline characteristics of the training cohort
We analyzed the baseline characteristics of the training cohort in three cohorts (Table 2). In terms of patient characteristics, gender, age and smoking history had significant statistical differences in any grade CIP cohort, and male, elderly, and smokers were more likely to develop CIP (P=0.020, 0.006, 0.046, respectively). In tumor characteristics, compared with patients without CIP, patients with less than two metastatic sites and with bone metastases were more likely to develop CIP, and the difference was statistically significant (P=0.033 and 0.002, respectively). From the point of comorbidities, patients with ILD and emphysema at baseline had a higher risk of CIP, with statistically significant differences (P<0.001 for both).
In the grade ≥2 CIP cohort, male and older patients were more inclined to suffer from CIP, with statistically significant differences between the two groups (P=0.043 and 0.001, respectively). For comorbidities, patients with ILD and emphysema at baseline were more likely to develop grade 2 or higher CIP, with statistically significant differences (P< 0.001 for both). The remaining factors were not statistically significant in the grade ≥2 CIP cohort.
The presence of ILD and emphysema at baseline were statistically different in the grade ≥3 CIP cohort (P<0.001 for both), while no statistically significant differences were found for other factors.
3. Correlation between peripheral blood indicator and CIP
ROC curves were plotted for 11 peripheral blood indicators, including CRP, NEUT, ALC, AEC, white blood cell (WBC), PLT, NLR, PLR, procalcalonin (PCT), SII, and PNI. Then the cut-off values of laboratory indicators with AUC greater than 0.5 were obtained and the AUC of each indicator is shown in supplemental table 1. Based on this, the included patients were divided into high group and low group. The results showed that patients with any grade CIP had higher baseline CRP (cut-off value=16.25g/L), AEC (cut-off value=0.215×109 cells/L), WBC (cut-off value=10.7×109 cells/L), PLT (cut-off value=265.5×109 cells/L), and SII index (cut-off value=1592.97), compared with patients without CIP. The differences were statistically significant (P=0.0002, 0.0041, 0.0173, 0.0061, 0.0364, respectively) (Figure S2A-E, respectively). In the grade ≥2 CIP cohort, the results suggested that patients with grade ≥2 CIP had higher baseline CRP (cut-off value=12.65g/L), AEC (cut-off value=0.22×109 cells/L), WBC (cut-off value=5.765×109 cells/L), PLT (cut-off value=176.5×109 cells/L), SII index (cut-off value=411.14), with statistically significant differences (P=0.025, 0.001, 0.011, 0.021, 0.033, respectively) (Figure S3A-E, respectively). The same method was used to analyze the differences of laboratory indicators in the ≥3 grade CIP cohort, suggesting that patients with grade 3 and higher CIP had a high level of CRP (cut-off value=12.65g/L, P=0.025), AEC (cut-off value=0.22×109 cells/L, P=0.001), NEUT (cut-off value=4.0×109 cells/L, P=0.025), and WBC (cut-off value=6.19×109 cells/L, P=0.023) (Figure S4A-D, respectively).
4. Exploration of independent risk factors for CIP in training cohort
To investigate independent risk factors associated with CIP, multivariate logistic regression analysis was performed for 25 factors, including patients’ baseline characteristics, tumor characteristics, treatment characteristics, comorbidity characteristics and peripheral blood indicators. In any grade CIP cohort, multivariate logistic regression analysis showed that several variables were independent risk factors for CIP, including ILD at baseline (P<0.001, OR=5.404, 95%CI=2.581-11.317), emphysema at baseline (P=0.006, OR=3.383, 95%CI=1.420-8.062), higher baseline CRP level (P=0.004, OR=4.967, 95%CI=1.650-14.953), and higher baseline AEC level (P=0.022, OR=2.553, 95%CI=1.146-5.687) (Table 3).
Multivariate logistic regression analysis was also performed in the grade ≥2 CIP cohort. The result revealed lung metastasis (P=0.012, OR=24.319, 95%CI=2.045-289.234), less than two sites of metastasis (P=0.031, OR=5.764, 95%CI=1.375-24.674), ILD at baseline (P=0.026, OR=8.949, 95%CI=1.300-61.614), emphysema at baseline (P=0.022, OR=9.252, 95%CI=1.373-62.329), higher baseline CRP level (P=0.045, OR=5.527, 95%CI=1.041-29.344), higher baseline AEC level (P=0.017, OR=10.442, 95%CI=1.509-72.258), and higher baseline SII level (P=0.028, OR=61.901, 95%CI=1.567-2444.729) were independent risk factors for grade 2 and higher CIP (Table 3). Similarly, multivariate logistic regression analysis was performed for the ≥3 CIP cohort, and no independent risk factors associated with CIP were found (Supplemental table 2).
5. Construction and validation of the nomogram model
Based on the independent risk factors, the predictive nomogram model was established derived from multivariate regression analysis. We used ILD and emphysema at baseline and baseline CRP and AEC levels as key factors to construct a nomogram model for predicting any grade CIP (Figure 2). The C index of the model in the training cohort and validation cohort were 0.814 (95%CI=0.760-0.870) and 0.919 (95%CI=0.842-0.941), respectively. And the calibration curves and ROC curves of the training cohort and validation cohort after internal verification were generated and illustrated in Figure 3A-D, respectively. The calibration curves suggested that the predicted and actual incidence rates are almost identical, which indicates that the nomogram performs well in predicting any grade CIP. Besides, the DCA curve showed that the nomogram model has better clinical predictive power than any single predictor, suggesting that the nomogram could serve as an effective diagnostic tool for CIP (Figure 3E-F, respectively). ROC curve was used to compare the value of combined nomogram model and single predictor in predicting the occurrence of CIP and the results indicate that the combined nomogram model has the highest AUC for both the training cohort and the validation cohort, which again verifies the predictive power of the nomogram model (Figure 3G-H, respectively).
Meanwhile, based on the independent risk factors for grade 2 and higher CIP, we construct another nomogram to predict the occurrence of grade 2 or higher CIP (Figure 4). After internal verification, the calibration curve of training cohort and validation cohort showed a small difference between the predicted incidence and the actual incidence, manifesting the nomogram was quite consistent with the actual observations (Figure 5A-B, respectively). The C index of training cohort and validation cohort were 0.879 (95%CI=0.81-0.92) and 0.936 (95%CI=0.812-0.982), respectively, which indicated the nomogram could perform well in effectively predicting the grade 2 or higher CIP (Figure 5C-D, respectively). DCA curve also manifested that the nomogram model had a stronger clinical prediction power than the single factor (Figure 5E-F, respectively). Moreover, the AUC of the nomogram model has the highest compared to every single indicator, which also supports the high predictive power for predicting grade 2 or higher CIP (Figure 5G-H, respectively).
6. The predictive power of the nomogram model scoring system
In order to further evaluate the predictive power of the nomogram prediction system, we divided the patients into high risk group and low risk group according to the cut-off value of the nomogram prediction model score. For any grade CIP cohort, the cut-off value of the total score of nomogram prediction model is 145 points, and patients will be divided into high risk group (total points ≥145 points) and low risk group (total points <145 points). After the cut-off value was applied to the training cohort, univariate analysis showed that there was a statistically significant difference in the occurrence of any grade CIP between the high risk group and the low risk group (P<0.001, OR=8.831, 95%CI=4.899-15.921), which was consistent with the results of the validation cohort (P<0.001, OR=6.457, 95%CI=2.339-17.824) (Table 4).
Similarly, in the grade ≥2 CIP cohort, patients were divided into high-risk group (total score ≥175) and low-risk group (total score < 175) with 175 points as the cut-off value. Univariate logistic regression analysis both manifested statistically significant differences in training cohort and validation cohort (training cohort: P<0.001, OR= 8.16, 95%CI=3.725-17.873; validation cohort: P<0.001, OR=35.100, 95%CI=4.447-277.065, respectively) (Table 4).