Trials for Precision Medicine

As noted above, despite the value of RCTs in demonstrating treatment efficacy, in real life these treatments do not work for all similarly. The demand for more individualized interventions has caused researchers to question whether traditional RCTs are the best method to evaluate personalized medicine. One RCT design is the crossover trial design where, rather than randomization to a treatment arm, patients are randomized to a sequence of treatments. This is most commonly an AB/BA trial design whereby patients assigned to the AB group receive treatment A followed by treatment B, and vice versa for the BA group. The patient takes one treatment for a prespecified period, followed by the second treatment for the same amount of time. Appropriate measurable outcomes are recorded, allowing the treatment effects for the two interventions to be compared. Since each patient receives both interventions, each patient acts as their own control, which avoids the problems that arise from the analysis of groups of patients. Comparing treatment effects within each individual also means there is no opportunity for confounding due to patient differences. To ensure that the effects of one drug does not “carry over” to the following treatment period, which would confound treatment effects, there is often a “wash-out” period between treatments. This is a period in which the patient is not exposed to any intervention, to allow the effects of the first drug to wear off. Crossover trials use the comparison of two interventions within individuals to establish the effects of both interventions in the absence of confounding.

An extension of the crossover trial design is the “N-of-1” trial design, which incorporates the basic elements of a crossover study. Where crossover studies are used to assess the average treatment effect of a drug in a certain population, N-of-1 trials are used to assess the individual treatment effect in a single patient. N-of-1 trials are randomized, multiple crossover trials which are conducted in a single patient, and are used to determine the most effective treatment in an individual (Guyatt et al., 1986). N-of-1 studies are a promising way to advance individualized medicine and a method for gaining insights into comparative treatment effectiveness among a wide variety of patients. Although N-of-1 trials have generated a lot of interest among both physicians and researchers over the last 30 years, it is well recognized that they are more time-consuming and costly than standard care. With ubiquitous mobile digital devices and advances in Al and machine learning, the potential of N-of-1 trials is now realizable for personalized medicine. N-of-1 trials are associated with more intensive data collection, and a large number of observations collected on a patient require time-series analysis and accommodating serial correlation between measures and possible carryover effects. Machine learning methods are essential for combining and evaluating multiple N-of-1 trials to make population-level estimates (Zucker, Ruthazer, & Schmid, 2010), identification of common characteristics among patients who are ultimately found to respond best to a particular intervention, treatment repositioning and integration of wireless data with electronic medical records.

Conclusion

The application of Al and machine learning can improve clinical trial subject identification, recruitment and retention. Furthermore, they can enable enhanced trial designs for evaluating precision medicine therapies, uncover hidden structure in trial data and lead to novel therapeutic discoveries and treatment repositioning. Generation of synthetic data under differential privacy with deep neural networks offers a technical solution for data sharing and the use of controls without endangering patient privacy.