Author: Dr. Doug Popken,
SVP Analytics, NextHealth Technologies
Health care payers (insurers) and providers are seeking innovations that demonstrably improve health and lower the medical costs of the populations they serve by impacting positive behavioral changes. Many of today’s population health interventions are already being implemented – such as cancer screenings, ER diversion, and clinical interventions. But it is difficult to know what will work best for any given population, or even how to measure success. Randomized controlled trials have long been considered the gold standard in medical research to establish the causal impact of health interventions. However, in the health care industry their use has been limited. A more common approach has been to compare the metrics describing a given population before and after an intervention or “nudge”. The problem with this approach is that the population will likely have been influenced by factors other than the intervention, such as systemic changes in economic conditions, pricing, regulation, or seasonal variations. It is then impossible to know the true impact of the nudge versus other influential factors.
The control group tells us what would have happened to individuals if they had not received the nudge. To avoid introducing systemic variations between trials and controls (or between trials experiencing different nudges), random selection is used to determine the group to which an eligible individual belongs. We can then compute causal “lift” by comparing the average value for any given evaluation metric between the trial and control groups. We can then go on to evaluating the relative effectiveness of different nudges by comparing their lift. This basic process is at the core of the NextHealth analytics based platform – NextNudge™.
Since most health metrics are subject to high degrees of random (versus systemic) variation, it is also important to determine whether lift values are statistically significant. Fortunately, standard statistical approaches to comparing population means are available that can provide both confidence intervals and “P-values” that establish significance. For example, if the 95% confidence interval for lift does not contain zero, we can say that the results are significant. Similarly, a P-value less than .05 would also indicate significance. In our experience, it may take several months of data collection to begin seeing statistical significance in health related population metrics. However, once achieved, significance rapidly becomes very strong.
In today’s dynamically changing market, establishing and leveraging causality is critical to optimized resource allocation and effective consumer engagement.
For more information about NextHealth Technologies’ NextNudge™ platform, please contact us at firstname.lastname@example.org or visit http://www.nexthealthtechnologies.com/nextnudge/.