Methods

Study subjects

All patients who had received trastuzumab at Osaka Medical and Pharmaceutical University Hospital from January 1, 2017 to December 31, 2020 were identified and the following exclusion criteria were applied: gastric cancer, missing information on HER2 status, missing eosinophil information, or use of other treatments. This study was conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of Osaka Medical and Pharmaceutical University (Approval ID: 2020-175). Since this is a retrospective observational study without intervention or invasion, the requirement for informed consent was waived. This study was conducted according to the STROBE statement.45

Outcome variable

The severity of IRRs was assessed using the Common Terminology Criteria for Adverse Events (CTCAE) version 5.0, and the outcome measure was the occurrence of an IRR of grade 1 or higher after starting trastuzumab administration. For each patient, all electronic medical record data at the time of trastuzumab administration were retrospectively reviewed for potential IRR cases by a clinical pharmacist with specific training in screening and treatment of adverse effects in cancer chemotherapy.

Explanatory variables

Patients received repeated doses of trastuzumab. The data analyzed in this study were obtained from multiple doses administered to the same individual; that is, it had a two-level hierarchical structure consisting of a macro-level (patient level, level 2) and micro-level (infusion level, level 1). With reference to previous studies,39,42 the following factors were examined in this study. The macro-level variables were height, HER2 status, estrogen receptor (ER) status, progesterone receptor (PR) status, allergy history, and baseline eosinophil levels; the micro-level variables were age, stage, metastasis, concomitant medications, number of courses, weight, BMI, status (preoperative, postoperative, and/or recurrent progression), trastuzumab dose, and dexamethasone dose. Eosinophils were not measured for each dose of trastuzumab; therefore, baseline values were used as macro-level variables.

Data processing

Centering at the pooled mean was performed for the following macro-level variables: height, HER2 status, ER status, PR status, allergy history, and baseline eosinophil levels. Although age, stage, metastasis, and concomitant medications were micro-level variables, centering at the pooled mean was applied because there was little variation within individual patients. On the other hand, centering within the cluster was applied to the following micro-level variables: the number of courses, weight, BMI, status (preoperative, postoperative, and/or recurrent progression), trastuzumab dose, and dexamethasone dose.46

Statistical analysis

Descriptive statistics at the macro- and micro-levels were calculated. Fisher’s exact test was used for nominal variables, Wilcoxon rank-sum test was used for continuous variables, and Cochran-Armitage trend test was used for ordered variables (stage, trastuzumab dose, and dexamethasone dose) to examine the trends of IRR occurrence rate. Owing to the hierarchical structure of the data, the independence of the observed values was evaluated using the intraclass correlation coefficient (ICC).47 Subsequently, four models, including a null model, were tested using multilevel logistic regression analysis to identify the preventive effects of dexamethasone premedication on IRR occurrence as well as the relationships between the macro- and micro-level independent risk factors and IRR development. Model 0, the null model, is a model with no objective or explanatory variables and was used to determine ICC. ICC is a measure to assess similarity within a group,47 and in this study, the patients represented the group and one infusion of trastuzumab represented the individual. Model 1 incorporated micro-level variables, Model 2 incorporated macro-level variables, and Model 3 incorporated both micro- and macro-level variables. For the selection of candidate explanatory variables, p -values from univariate analysis were considered. Akaike’s information criterion (AIC), Bayesian information criterion (BIC), and likelihood ratio tests were used to compare model goodness-of-fit. Variance inflation factors (VIFs) of ≥10 were considered evidence of multicollinearity. All p -values were reported using two-tailed tests, and the significance level was set at 5%. Analyses were performed using R version 4.0.2 (R Development Core Team, Vienna, Austria).