If there is one topic that is top of mind for anyone in any service industry, it is a “personalized” approach. We live in a world where we want customized experiences. We have become used to companies like Amazon making “recommendations” for things we might like to buy based on past purchases. As part of that process, we have sacrificed data and privacy, but the convenience we get seems to be a price we are willing to pay.
Personalization in healthcare is also gaining ground with new healthcare systems touting care models based on your genetic data in addition to crucial markers of health via blood work. A new paper by Noa Dagan and colleagues from the Clalit Health Services in Tel Aviv published in Nature Digital Medicine examines the use of clinical trial data on the path to making personalized recommendations.
The authors chose Intensive Blood Pressure Treatment (IBPT) as a clinical example, specifically the data from the Systolic Blood Pressure Intervention Trial (SPRINT). Every day doctors use RCTs data to inform clinical decisions; they also take subjective elements into account when making treatment decisions. This study divided the SPRINT dataset into three phases; the first phase used prediction models to examine the pros and cons of IBPT. The second looked at predicted risk reductions using the benefit/harm ratio of a yes/no type of recommendation to treat IBPT. Lastly, the third phase involved the development of a decision support tool based on the analysis outputs.
Findings suggest that 62% of the original SPRINT trial participants would have benefited from a “yes” recommendation looking at the benefit/risk ratios. The value of this approach is building on prior methodologies by using eight separate outcomes that were ranked by severity versus a compositive severity score. Building out these eight outcomes allows for a deeper shared-decision making conversation between doctor and their patient regarding the pros and cons of intensive treatment.
Moving from population to individual-level guidance has been fraught with difficulty given the difference in the data signal. This study provides new options to consider when translating RCTs to the individual patient level.
One limitation of this study is that the trial used to develop the predictive models had only three-years worth of follow up data. Epidemiology trials show us that cardiovascular disease and its associated risks build up over decades- a three year view might be too short to be robust. The application of this methodology to RCT data with longer follow-up timelines will become necessary. The authors have, however, handed us a vital blueprint- an enhancement of what has gone before that can be built upon by future work.
I look forward to seeing more applications of these methods in other studies in other clinical areas.
Thanks for reading – Trina
(Opinions are my own)
Translating clinical trial results into personalized recommendations by considering multiple outcomes and subjective views.