Related Experiment Videos
Survival analysis, more than meets the eye
Marko Lucijanic1, Marko Skelin2, Tomo Lucijanic3
1Hematology Department, University Hospital Dubrava, Zagreb, Croatia.
Biochemia Medica
|April 11, 2017
Summary
Different versions of the log-rank test yield varying P values in oncology trials. This discrepancy can impact the interpretation of borderline results, potentially affecting drug efficacy evaluation.
Area of Science:
- Statistics
- Oncology
- Clinical Trials
Background:
- The log-rank test is crucial for phase III oncology clinical trials.
- Multiple mathematical variations of the log-rank test exist, leading to different P values.
- This ambiguity can affect the interpretation of borderline statistical significance.
Purpose of the Study:
- To highlight the discrepancies arising from different log-rank test implementations.
- To address concerns regarding potential data manipulation in drug regulatory contexts.
- To propose a standardized approach for reporting P values in clinical trials.
Main Methods:
- Comparison of three distinct mathematical procedures for the log-rank test.
- Analysis of P value variations across different methods.
- Utilizing an interactive MS Excel spreadsheet for demonstration.
- Case study: Myelofibrosis with high-grade bone marrow fibrosis.
Main Results:
- Different log-rank test variants produce non-identical P values.
- Borderline significant results can be classified as significant or non-significant depending on the method used.
- The choice of log-rank test variant can influence the perceived efficacy of a drug.
Conclusions:
- Standardized reporting of P values from all log-rank test variants is recommended for borderline results.
- This standardization is essential for accurate drug efficacy evaluation in regulatory settings.
- Ensuring transparency and preventing potential data manipulation is critical.