Methodological critique reveals critical statistical and imaging limitations in glioblastoma survival modeling, highlighting the need for standard analytical rigor.
Key Points
To evaluate the methodological validity, imaging protocols, data integrity, and statistical modeling in a published study using DTI and MRI to predict glioblastoma survival.
Critically appraised the imaging acquisition protocol, accounting of established clinical prognostic covariates, and volumetric analysis methods.
Evaluated the statistical integrity of reported correlation analyses, follow-up classifications, and multivariable Cox proportional hazards regression models.
Identified critical imaging limitations, including using only 6 diffusion-encoding directions for fractional anisotropy rather than the recommended minimum of 25–30 directions, alongside an omission of essential prognostic covariates such as age, KPS, extent of resection, and MGMT methylation status.
Demonstrated severe statistical model overfitting with 30 survival events evaluated across >10 predictors, producing unstable estimates including a hazard ratio of 45.9 (95% CI: 3.01–703.23) for microhemorrhage exceeding 25%.
Identified data recording anomalies reporting biologically implausible tumor volumes approaching 80% to 100% of intracranial volume, alongside improper application of Pearson's correlation to non-normally distributed data.