Statistical analysis

Baseline characteristics were summarized according to LVEF trajectory class using proportions for categorical variables while mean (standard deviation (SD)) and median (interquartile range (IQR)) were utilized for normally and non-normally distributed continuous variables, respectively. The purpose of this table is primarily descriptive; no hypothesis testing was performed to determine the differences across the classes.

Details on the procedure implemented to define the LVEF trajectory classes have been described previously, and these are additionally presented in the supplementary file (12). Univariable and multivariable multinomial regression analyses were performed to evaluate associations between baseline variables and LVEF trajectory class membership. A minimal set of prespecified variables deemed to be clinically and/or pathophysiologically relevant (i.e., age, radiation therapy, cancer therapy regimen, baseline LVEF, hypertension and cardiac medication use (angiotensin converting enzyme inhibitor (ACEI), angiotensin receptor blocker (ARB) or beta-blocker)) were included in multivariable analysis.

To characterize the changes in markers of cardiac dysfunction over time in the different LVEF trajectory classes, mean changes in longitudinal strain, NT-proBNP and hs-cTnT from baseline were estimated at 6 months, 1, 2, and 3.5 years using repeated-measures linear regression estimated via generalized estimating equations (GEE). Each of the models were adjusted for the baseline value of the marker under consideration and time since baseline assessment (modeled using a cubic spline with 3 degrees of freedom). A robust variance estimator was used to account for within subject clustering. Separate models were fit in each of the identified LVEF trajectory classes. A similar procedure was implemented to evaluate changes in the MDASI-HF symptom score over time. The Kaplan-Meier estimator was used to estimate the proportion of patients who experienced all-cause mortality during follow-up in each trajectory class. Difference across classes was tested with the log-rank test.

A two-sided alpha level of 0.05 was used to assess statistical significance. All analyses were performed using R 3.4.0 (R Foundation for Statistical Computing, Vienna, Austria).