Embracing Advanced Methodology to Improve Population Health

AJPH has long endorsed the application of novel and powerful state-of-the-art methodological approaches to advance population health. In 2004, we asked David Murray^1,2 and Allan Donner,³ two pioneers in the development and application of group or cluster randomized trials, to provide extensive primers on their design, application, and analysis, and invited them to help keep these tools in our analytic toolbox by providing methodological updates in 2017.^4,5

In 2018, we embraced the notion of returning “cause” to our public health vocabulary, and employing causal inference methods to our data when appropriate. Miguel Hernan⁶ and others⁷ presented a compelling case that while, for many good historical reasons, we kept our expectation of finding and stating causes to a minimum and similarly kept our language away from the term, that it was time to acknowledge that we are indeed after causes. And while we may not be in a design or analytic space to suggest cause, we should not shy away from the application of methods for causal inference when appropriate and should strive to push our research as far down the causal pathway as possible. In 2023, AJPH is again promoting the adoption of advanced methodologies to improve population health with the presentation of three methodologies and approaches.

METHODOLOGICAL ADVANCES

First, Wang and Chakraborty (p. 49) present an outstanding review of a method that overcomes some of the downsides of traditional randomized controlled trials, which often tend to compare one novel treatment to standard or usual care, where participants remain in their assigned treatment groups over the life of the trial. Such trials can often last months or years and can be prohibitively expensive. The sequential multiple assignment randomized trial (SMART) and other such adaptive or dynamic treatment strategies allow investigators to reassign patients to additional intervention arms depending upon response to the previous intervention. The review paper offers an overview of methods and applications and provides links to R code and methods for sample size calculations for SMARTs as well as for the statistical analysis of such trials. A supportive commentary by Seewald in this issue also extends the conversation and provides further insights into the application of SMART designs (Seewald p. 37).

Second, Liu et al. (p. 60) expertly present the design and analysis considerations for a whole class of interventions that recognize that spatially and temporally static interventions are not always optimal. These “just-in-time adaptive interventions” (JITAIs) seek to provide the most effective intervention type at just the right time and place; in a smoking cessation program, a JITAI approach might send reminders to a participant to check their patch or chew replacement therapy gum during periods of high craving, while sending alerts to practice meditation, to exercise, or participate in healthy and alternative behaviors during other times, and perhaps deliver no intervention when all internal and external conditions are optimal for cessation so as not to overburden the patient. Liu et al. describe a design approach, microrandomized trials (MRTs), that are employed to help determine the type and duration of those spatially and temporally optimal interventions. Liu et al. use two published examples of the implementation and analysis of MRTs and provide tables with comprehensive links to published methods and R, SAS, or Stata code for both sample size calculations for, and statistical analysis of, MRTs, whose results can be used to populate the content of JITAIs.

Third, Bauer et al. (p. 40) describe a Bayesian estimation method to help identify testing disparities and inform the magnitude of testing deficiency in the context of small area estimation, using severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) as an example. Bauer et al. present a two-step sampling method that has a number of advantages over traditional approaches. First, just as “all politics are local,” all public health is local as well, and the method is designed to be implemented in smaller, local environments where testing data and information may be sparse.

Second, it can help, in real time, predict the number of tests that will be needed to better inform the supply chain needs and can incorporate the often-dynamic parameters of a pandemic rather than relying on fixed and perhaps outdated inputs. In a public health catch-22, areas with fewer available tests may report lower rates, thereby reinforcing the false need for fewer tests. As in the article by Hernan,⁷ in which he observes that the terror evoked by the term “causal inference” may keep many away from implementing causal methods, uttering the word “Bayesian” may similarly keep investigators at bay. Fortunately, Bauer et al. (p.40) provide links to the necessary R code on GitHub along with a practice data set for easy implementation.

PUSH–PULL

Methodologies and population health often operate in a symbiotic push‒pull relationship, in which stubborn health issues push the development of novel design and analytic methodologies, and new methodologies allow for interrogations to get pulled into new areas and offer potential solutions. We hope that by presenting these methods in user-friendly formats that it makes it easier to enter the population health push‒pull conversation and continue to advance health.