Navigating Controversy: Ordinal Outcomes in Clinical Trials
The Historical Roots of Ordinal Outcomes
Clinical trials trace their roots to 1747 when James Lind, a Scottish naval surgeon, sought a cure for scurvy aboard the HMS Salisbury. Despite the rudimentary design, which lacked randomization, his trial marked a pivotal moment in medical research with an organized design to collect—and publish—empirical evidence of the effect of different therapies. Lind's outcomes involved observing ranked categories of patient improvement—an early illustration of what we now refer to as ordinal outcomes. The challenge Lind faced was not just in identifying effective treatments but how to characterize the clinical outcome of those therapies. It took 47 years for the British Navy to accept the role of vitamin C in the treatment of scurvy.
The first randomized clinical trial for humans is attributed to Bradford Hill, which he published in 1948, with a randomized trial of streptomycin for the treatment of pulmonary tuberculosis. The primary endpoint in the trial was a 6-point ordinal scale, ranging from “considerable improvement,” to “death.” Hill analyzed the outcome by looking at multiple cut points on the scale for a difference in effect – which streptomycin showed a huge advantage across the entire scale.
While there is a rich history of using ordinal outcomes, and they are powerful tools for measuring the clinical outcome for patients, we argue about how to analyze these outcomes, usually resorting to very crude analyses of the rich clinical outcomes.
Ordinal Outcomes: Pivotal Yet Contentious
Moving to contemporary clinical trials, ordinal outcomes have become integral yet their analysis contentious. These outcomes, categorized in a rank order, but typically without definitive differences in the various levels. The tuberculosis trial by was a landmark case, using a six-level ordinal scale, which is very similar to numerous trials using the modified Rankin score, a measure of the neurological status of a patient, utilizing a seven-level scale in neurological assessments from “no neurological deficit” to “death.”. The scale is a rich one, measuring across the spectrum the neurological status that makes us human. Statisticians and researchers debate vigorously and contentiously how to analyze this beautiful 7-point scale.
Despite their utility, ordinal outcomes challenge researchers to balance simplification against clinical relevance. Dichotomization—reducing categories into binary options like responder versus non-responder—is prevalent but problematic. It equates disparate conditions, potentially diminishing the granularity crucial for meaningful analysis. Yet, its simplicity makes it tempting for trial designs.
Statistical Approaches: Exercising Caution and Precision
Diverse statistical approaches attempt to address the intricacies of ordinal data, each bringing specific strengths and challenges. Proportional odds models stand out as a popular choice, assuming a consistent effect across outcome levels. This method respects ordinal structure while enabling detailed insights, albeit hinging on statistical assumptions that may falter in varied clinical settings.
A proportional odds model requires a constant odds ratio of a treatment across all outcome levels. If misapplied, this assumption could make interpretation challenging, confounding treatment efficacy conclusions. Unfortunately, reactions to a lack of proportion odds assumption lead people to dichotomize the outcome, obfuscating clinical meaning, and ignoring that the lack of proportionality still holds even if the dichotomization ignores it. A proportional odds model imposes a utility on each of the outcomes that is based on their prevalence, but not necessarily their clinical utility.
Alternative methods like the Wilcoxon test and Cox models also tackle ordinal data, drawing from time-to-event analysis traditions. Each method attempts to honor the ordinal nature while providing interpretable results. However, they too face scrutiny for their assumptions and computational complexity, also imposing unspecified utilities.
Looking to the Future: A Call for Meticulous Innovation
The handling of ordinal outcomes is a pivotal issue with far-reaching implications for clinical advancements. As technology paves the way for more granular health data from wearables and real-time monitoring, the landscape of clinical trials is poised for transformation as are the depth of the outcomes. The challenge remains in integrating this data into a coherent framework that maintains statistical rigor and clinical relevance.
Efforts must focus on developing methods embracing the potential of detailed ordinal data. This means moving beyond conventional dichotomization or simplistic categorizations to embrace the hard, but necessary challenge of explicitly equating utilities to the outcomes. By aligning statistical innovations with clinical objectives, researchers can ensure trials deliver meaningful, actionable insights.
Ultimately, acknowledging and addressing the complexities of ordinal outcomes is both an important technical challenge and a strategic necessity. As clinical trials evolve, embracing nuanced, transparent approaches to analyzing these outcomes will be crucial for advancing medical research and treatment efficacy. By resisting oversimplification and advocating for measuring clinical meaningfulness, we can ensure every trial not only meets scientific standards but also serves the broader quest for improved health outcomes.