Expert Consensus and Artificial Intelligence in Medical Decision Making in Patients with Malignant Brain Tumors

1\) Statistical analysis plan: 1. this document provides the details of statistical analyses planned for the EC-AIM Brain study. 2. Objective of this study: i. The primary objective is to determine whether patients whose treatment aligns with expert consensus have a superior outcome to those who treatment does not align with expert consensus. 1. This will be determined with a primary endpoint of progression free survival (alive and without tumor growth at last follow up). 2. This will be tested with the log-rank test for equality of survival curves (EC/aligned vs EC/unaligned). 2\) Abbreviations: 1. EC - expert consensus 2. HR - hazard ratio 3. OS - overall survival 4. PFS - progression free survival 5. KPS - Karnofsky Performance Score 3) Target population: a. Inclusion criteria: i. Consecutive patients treated by neurosurgeons at a single center between April 2018 and July 2023 for malignant brain tumors, including glioma, metastasis, and lymphoma. b. Exclusion criteria: 1. Patients without available MRI dicom images 2. Patients with other CNS malignancies 3. Patients with multiply recurrent gliomas undergoing treatment for primarily palliative purposes 4. Patients younger than 16 years old 4) Sampling: Consecutive 5) Sample size assessment: a. It is challenging to determine an estimated effect size that EC would provide. From previous research on expert panels in lumbar spondylolisthesis, a sample size of approximately 200 patients was noted to be sufficient for initial assessment. As such, the study will include 225 subjects as our cohort size in order to account for \~10% patient loss to follow up. 6\) Data Acquisition Methods: 1. Standardized clinical vignettes with de-identified clinical data and standardized radiographic images and video will be uploaded to Acesis HPOP platform. i. Uploaded files will include representative multiplanar magnetic resonance or computed tomography images b. An email with a link to the survey will be distributed to the brain tumor expert panel c. Expert respondents will have 72 hours to respond to a standardized survey d. Expert respondents may answer questions regarding suggested treatment as well as any adjunctive technologies needed 7) Blinding: The chart review portion will be performed by a separate study investigator without knowledge of cohort assignment; cohort allocation (based on survey results and actual treatment) will be performed by a separate study investigator. 8\) Missing data: a. All subjects will be included in primary outcome analysis. They will be censored at the last follow up assessment. All missing data will be assumed to be missing at random. 9\) Quality assurance plan: 10) Outcome Group Allocation: a. Treatment will be defined into the following four categories: i. Observation/no treatment ii. Biopsy only iii. Resection iv. Radiation treatment b. Expert consensus will be defined as at least 80% agreement among at least 10 survey respondents c. Aligned will be defined as the recommended treatment category above matching the actual treatment provided 11) Outcome Analysis: a. Primary Outcome Variable Analysis: i. The primary outcome measure is progression free survival (alive and without evidence of tumor growth at follow up) ii. Hypothesis: 1. H0: hEC\_aligned ≤ hEC\_unaligned 2. HA: hEC\_aligned \> hEC\_unaligned iii. This will be tested with the log-rank test for equality of survival curves (EC/aligned vs EC/unaligned). iv. Test will be one-sided (alpha=0.025) b. Additional analyses of the primary outcome: i. Adjustment for known confounders including diagnosis, age, KPS, extent of resection ii. Exploratory data analysis using descriptive statistics and logistic regression/linear regression will be used to evaluate for interactions and confounding co-variates c. Secondary Outcome Variable Analysis: i. Overall survival ii. Data analysis will be exploratory and supportive, and, thus, the statistical plan will not account for multiplicity iii. Secondary analyses will be performed for OS and PFS for two- and three- cohort groups 1. EC/aligned vs EC/unaligned 2. EC vs no EC 3. EC/aligned vs EC/unaligned vs no EC 12) Planned analysis for predictors of EC 1. Data analysis will be exploratory and supportive, and, thus, the study will not account for multiplicity 2. Descriptive statistics will be used to evaluate differences between EC and no EC groups 3. Descriptive statistics will be used to evaluate differences between EC/aligned and EC/unaligned groups 4. Fisher exact tests and Chi-squared tests will be used for categorical variables depending on test assumptions 5. Student's t test and Mann-Whitney U tests will be used for normally distributed and non-normally distributed continuous variables, respectively 6. Logistic regression models will be used to evaluate for predictors of EC 7. Concordance between treatment performed and expert consensus will be evaluated using Spearman coefficients 8. All statistical tests used for descriptive statistics will be two-tailed (alpha=0.05) 13) Planned interim analysis: None 14) Adverse events: None expected as this is a retrospective chart review and survey based study.