Photo credit: www.sciencedaily.com
The restricted mean survival time (RMST) analysis method has made significant strides since its introduction to healthcare research approximately 25 years ago. Its application has broadened to fields such as economics, engineering, and business.
In clinical environments, RMST serves as a valuable tool for assessing the average survival duration—essentially the time patients live after receiving a diagnosis or treatment, alongside the factors influencing that duration—within a defined time frame.
Unlike Cox regression models and other prevalent methods, RMST estimations and comparisons do not depend on the proportional hazard assumption, which posits that the risk of an event remains constant over time.
However, challenges remain. “RMST can assess differences in treatment effects between groups from a baseline to a designated time—known as the threshold—but pinpointing the optimal threshold in clinical and epidemiological research can be complex,” explained Gang Han, PhD, a biostatistics professor at Texas A&M University School of Public Health. “This complexity can result in less statistically robust outcomes than desired.”
To tackle this issue, Han, along with academic and industry colleagues, has pioneered a new approach utilizing a pre-existing mathematical tool—the reduced piecewise exponential model—to identify the ideal threshold time in RMST analysis for comparing two groups.
“This advancement is particularly significant in medical research because the likelihood of certain events can vary during different treatment phases,” noted Matthew Lee Smith, PhD, a health behavior professor at Texas A&M School of Public Health, who contributed to this study.
The research team determined the optimal threshold by identifying significant changepoint(s) in hazard rates and comparing these findings with the maximum possible threshold time.
Published in the American Journal of Epidemiology, their findings demonstrated the effectiveness of the proposed method across various simulation studies and two practical examples—a clinical trial and an epidemiological study.
In their simulations, the researchers evaluated Type 1 error rates and statistical power, contrasting a group with a constant hazard rate against another with a varying rate. They compared results obtained from the standard logrank test with those from the new model.
“Our model outperformed the others,” stated Marcia G. Ory, PhD, Regents and Distinguished Professor in the School of Public Health, who studies evidence-based prevention techniques. “We observed the same superior performance when applying the model to two real-world cases.”
In both practical examples, traditional statistical methods failed to show significant differences between the two treatments. However, when the new method was applied, it indicated a clear superior treatment for each scenario.
The first scenario assessed two therapies over seven months among patients with non-small-cell lung cancer who exhibited lower levels of a crucial biomarker. The second investigated the time until a decline in cognitive function among individuals with mild dementia, comparing those living with caregivers to those without.
“These preliminary results are encouraging, though further research is needed to explore comparisons involving more than two groups and to incorporate multiple covariates such as age, ethnicity, and socioeconomic status,” Han emphasized. “Nevertheless, our early findings suggest that this method has the potential to outperform existing techniques for two-group comparisons in time-to-event outcome analyses.”
Other contributors to the study included Laura Hopkins, a doctoral student in the Department of Epidemiology and Biostatistics, Raymond Carroll, PhD, a Distinguished Professor in the Texas A&M Department of Statistics, as well as external collaborators from Eli Lilly and Company and the H. Lee Moffitt Cancer Center & Research Institute.
Source
www.sciencedaily.com