Predicting Treatment Response with Clinical Trial Data
A central challenge in medicine is determining when to prescribe treatment for a patient when it’s unclear whether the benefits outweigh the associated side effects. This challenge is often compounded by a lack of scientific knowledge about how a patient’s individual characteristics – their age, gender, or co-morbidities, for example – might affect the patient’s response to the treatment. An editorial published in JAMA highlights a clinical trial that tackled this problem head on by creating a risk score, based on trial data, that predicts which patients are at increased risk for adverse events with treatment.
Researchers conducting the Dual Antiplatelet Therapy (DAPT) trial set out to calculate a risk score that could predict whether extended therapy after heart surgery was more likely to result in positive outcomes (reduced risk of ischemic events such as heart attacks) or a negative one (increased internal bleeding). Taking into account patient factors that could affect treatment outcome, such as history of heart disease and smoking status, the study team was able to produce a risk score able to distinguish when the benefits of extended therapy outweighed the risks. Patients with a score predictive of a negative outcome, for example, were more likely to experience bleeding with treatment than those who took placebo.
This efficient use of clinical trial data is promising for its potential to aid clinicians and patients in making more informed, individualized treatment decisions. It can also serve as a model for the design of future clinical trials concerned with heterogeneity in patient response to treatment. As described in the editorial, however, there is room for improvement. Specifically, the researchers’ risk score weighed the benefits and harms of the treatment’s outcome without taking into account each patient’s assessments of those factors. For instance, one patient might be more concerned about the possibility of internal bleeding than another, and thus would want the potential for bleeding to be given higher weight in the risk prediction model. Adding this level of personalization would complicate risk model development, but could prove invaluable for patients and their clinicians.