Factors influencing the robustness of P-value measurements in CT texture prognosis studies
Physics in Medicine and Biology
Aim: Several studies have recently reported on the value of CT texture analysis in predicting survival, although the topic remains controversial, with further validation needed in order to consolidate the evidence base. The aim of this study was to investigate the effect of varying the input parameters in Kaplan-Meier analysis, to determine whether the resulting P-value can be considered to be a robust indicator of the parameter's prognostic potential. Methods: A retrospective analysis of the
... e analysis of the CT-based normalised entropy of 51 patients with lung cancer was performed and overall survival data for these patients were collected. A normalised entropy cut-off was chosen to split the patient cohort into two groups and log-rank testing was performed to assess the survival difference of the two groups. This was repeated for varying normalised entropy cut-offs and varying follow-up periods. Our findings were also compared with previously published results to assess robustness of this parameter in a multi-centre patient cohort. Results: The P-value was found to be highly sensitive to the choice of cut-off value, with small changes in cut-off producing substantial changes in P. The Pvalue was also sensitive to follow-up period, with particularly noisy results at short follow-up periods. Using matched conditions to previously published results, a P-value of 0.162 was obtained. Conclusions: Survival analysis results can be highly sensitive to the choice in texture cut-off value in dichotomising patients, which should be taken into account when performing such studies to avoid reporting false positive results. Short follow-up periods also produce unstable results and should therefore be avoided to ensure the results produced are reproducible. Previous published findings that indicated the prognostic value of normalised entropy were not replicated here, but further studies with larger patient numbers would be required to determine the cause of the different outcomes.