Accurate predictors of immune checkpoint inhibitors in patients with gallbladder cancer

Abstract

Immune checkpoint inhibitors (ICIs) are effective for biliary tract cancers, but data on gallbladder cancer (GBC) are limited. In a recent issue of Cancer Science, Cheng et al.¹ aimed to assess the efficacy of ICIs in GBC and explore the clinicopathologic and molecular markers associated with ICI benefits. Based on logistic regression analysis, they found that alcohol intake history, a carcinoembryonic antigen (CEA) level ≥ 100 U/mL, and cutaneous immune-related adverse events (irAEs) were independent prognostic factors for these patients. High carcinoembryonic antigen (CEA) levels, cutaneous irAEs, high CD8⁺ T-cell infiltration, and an immune inflamed phenotype could be useful for predicting the efficacy of ICIs in GBC patients. However, in this letter, we raise concerns about the statistical method used in this study, while the prognostic factors or predictors for GBC patients may be different.

For a prognostic factors or predictors logistic regression analysis, the basic statistical rule demands 1 covariate per 10 outcome events.^2-4 However, the univariable and multivariable Cox proportional hazards regression model in Table 2 of Cheng's study breaks this basic statistical rule. We could observe that there were 21 variables in Table 2 of Cheng's paper that were generated from only 43 PD patients (outcome) who developed tumor recurrence or progressed to death. In other words, analysis of these 21 variables needs at least 210 PD outcome patients, not the 43 PD patients reported in this study. Thus, these overfitted univariable and multivariable logistic models could not produce reliable results, therefore the results in Cheng's study may not be accurate predictors for these patients in the clinic.

To reduce the variables in the predictor logistic regression analysis, the author could compare the outcome group and the non-outcome group. Finding the significant variables between these two groups and using the reduced variables to carry out the predictor logistic regression analysis would lead to more reliable statistical results. Additionally, to finally corroborate Cheng's conclusion, other large sample size results or a validation cohort is needed to validate the predictor results reported in this study.

Last, we congratulate Cheng et al. for their outstanding work despite these comments.

Naimei Li: Writing – original draft. Shuang Deng: Conceptualization; writing – original draft; writing – review and editing.

None.

The authors declare no conflict of interest.

Approval of the research protocol by an Institutional Reviewer Board: N/A.

Informed consent: N/A.

Registry and the Registration No. of the study: N/A.

Animal Studies: N/A.