Adverse event development in clinical oncology trials

Gita Thanarajasingam and colleagues' Article1 in The Lancet Oncology reports on a novel longitudinal approach for adverse event analysis and reporting. Comprehensive adverse event reporting in clinical oncology trials is essential to monitor tolerability of new cancer treatments. In view of the shift towards a more personalised care approach, including treatments that are given for a prolonged period of time, adverse event reporting needs to be reformed. A refocus from reporting only the maximum adverse event grade towards individual development of adverse event progression over time is desirable. Therefore, Thanarajasingam and colleagues' Article 1 fulfills an unmet need within oncology research.

The proposed ToxT package improves graphical representation of adverse event data for individual patients. Beyond the scope of drug trials, this new method is crucial for radiation oncology to distinguish between acute and late occurring adverse events. Moreover, the fact that the ToxT package can be applied to both the Common Terminology Criteria for Adverse Events and the increasingly popular Patient Reported Outcomes version2 is very convenient. However, the analysis techniques incorporated within the package might be improved when studying treatment effects. For example, the authors used linear mixed modelling to calculate one-size-fits-all average trajectories of adverse events. This statistical method does not allow revelation of subgroups with distinct developmental trajectories of adverse events. Although the method potentially leads to enhanced understanding of adverse event development, the assumption that one average trajectory represents the whole treatment group might be too simplistic. A more innovative approach to analysing such data is latent class growth mixture modelling.3 This type of modelling allows identification of otherwise unobserved subgroups with distinct developmental trajectories of adverse events.3 The identification of such treatment subgroups, including their determinants and consequences, might ultimately lead to an improved understanding of which adverse event is likely to occur in whom and when, by categorising patients on the basis of heterogeneity within the data and not on subgroups determined a-priori from existing knowledge.4 This way of analysing longitudinal data can be a next step towards personalised medicine and moves beyond graphical data presentation of average treatment effects.

Another limitation, also acknowledged by the authors, is that information about the type and timing of supportive care measures and other potential influencing factors that are registered and might vary over time are not taken into account by the applied methods within the ToxT package. Characterisation of within-patient patterns or the longitudinal association of these dynamic variables with adverse event development over time would lead to an improved understanding of how and when to prevent and treat patients at risk for adverse events that reduce quality of life. Although complex, more dynamic joint models might be introduced to allow adjustment for time-dependent confounders and to study the dependency and association between repeated measurements.5

To conclude, the ToxT analysis package for longitudinal adverse event data is promising and fills an important gap in adverse event reporting and graphical representation. However, in view of the ongoing tendency towards personalised and more dynamic care models, the authors are encouraged to add more options to the analysis package.

We declare no competing interests.