Median percent change: a robust alternative for assessing temporal trends

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Temporal trend analysis commonly uses average percent change (APC), often estimated via generalized linear models assuming log-linear relationships.
• Least-squares estimators for APC are susceptible to outliers, potentially compromising trend accuracy.
• Violations of standard model assumptions can further challenge reliable estimation of temporal trends.

Purpose of the Study:

• To propose a robust and easily computable measure for temporal trends: the median percent change (MPC).
• To evaluate the performance of various trend estimation models under violated assumptions.
• To offer practical guidance for estimating temporal trends across different models and scenarios.

Main Methods:

• Development of the median percent change (MPC) measure, utilizing the median of rates instead of the mean.
• Conducting a simulation study to assess the robustness and performance of the MPC against outliers.
• Investigating mathematical properties of the median for handling zero counts in log-scale trend estimation.
• Analysis of an English cancer registration dataset to demonstrate the MPC method.

Main Results:

• The simulation study confirmed that MPC significantly reduces estimation bias and variance in the presence of outliers.
• The MPC method effectively addresses challenges posed by zero counts in log-scale trend analysis.
• The proposed method provides a more robust estimation of temporal trends compared to traditional APC.

Conclusions:

• The median percent change (MPC) offers a robust alternative to the average percent change (APC) for temporal trend analysis.
• Presenting both APC and MPC is recommended when data sensitivity to outliers is a concern.
• The MPC method enhances the reliability of trend estimation, particularly in epidemiological and biostatistical applications.