Person:

Moscoe, Ellen

When patients receive an intervention based on whether they score below or above some threshold value on a continuously measured random variable, the intervention will be randomly assigned for patients close to the threshold. The regression discontinuity design exploits this fact to estimate causal treatment effects. In spite of its recent proliferation in economics, the regression discontinuity design has not been widely adopted in epidemiology. We describe regression discontinuity, its implementation, and the assumptions required for causal inference. We show that regression discontinuity is generalizable to the survival and nonlinear models that are mainstays of epidemiologic analysis. We then present an application of regression discontinuity to the much-debated epidemiologic question of when to start HIV patients on antiretroviral therapy. Using data from a large South African cohort (2007–2011), we estimate the causal effect of early versus deferred treatment eligibility on mortality. Patients whose first CD4 count was just below the 200 cells/μL CD4 count threshold had a 35% lower hazard of death (hazard ratio = 0.65 [95% confidence interval = 0.45–0.94]) than patients presenting with CD4 counts just above the threshold. We close by discussing the strengths and limitations of regression discontinuity designs for epidemiology.

Regression discontinuity analyses can generate estimates of the causal effects of an exposure when a continuously measured variable is used to assign the exposure to individuals based on a threshold rule. Individuals just above the threshold are expected to be similar in their distribution of measured and unmeasured baseline covariates to individuals just below the threshold, resulting in exchangeability. At the threshold exchangeability is guaranteed if there is random variation in the continuous assignment variable, e.g., due to random measurement error. Under exchangeability, causal effects can be identified at the threshold. The regression discontinuity intention-to-treat (RD-ITT) effect on an outcome can be estimated as the difference in the outcome between individuals just above (or below) versus just below (or above) the threshold. This effect is analogous to the ITT effect in a randomized controlled trial. Instrumental variable methods can be used to estimate the effect of exposure itself utilizing the threshold as the instrument. We review the recent epidemiologic literature reporting regression discontinuity studies and find that while regression discontinuity designs are beginning to be utilized in a variety of applications in epidemiology, they are still relatively rare, and analytic and reporting practices vary. Regression discontinuity has the potential to greatly contribute to the evidence base in epidemiology, in particular on the real-life and long-term effects and side-effects of medical treatments that are provided based on threshold rules – such as treatments for low birth weight, hypertension or diabetes.

Background The pathway from evidence generation to consumption contains many steps which can lead to overstatement or misinformation. The proliferation of internet-based health news may encourage selection of media and academic research articles that overstate strength of causal inference. We investigated the state of causal inference in health research as it appears at the end of the pathway, at the point of social media consumption.

Methods We screened the NewsWhip Insights database for the most shared media articles on Facebook and Twitter reporting about peer-reviewed academic studies associating an exposure with a health outcome in 2015, extracting the 50 most-shared academic articles and media articles covering them. We designed and utilized a review tool to systematically assess and summarize studies’ strength of causal inference, including generalizability, potential confounders, and methods used. These were then compared with the strength of causal language used to describe results in both academic and media articles. Two randomly assigned independent reviewers and one arbitrating reviewer from a pool of 21 reviewers assessed each article.

Results We accepted the most shared 64 media articles pertaining to 50 academic articles for review, representing 68% of Facebook and 45% of Twitter shares in 2015. Thirty-four percent of academic studies and 48% of media articles used language that reviewers considered too strong for their strength of causal inference. Seventy percent of academic studies were considered low or very low strength of inference, with only 6% considered high or very high strength of causal inference. The most severe issues with academic studies’ causal inference were reported to be omitted confounding variables and generalizability. Fifty-eight percent of media articles were found to have inaccurately reported the question, results, intervention, or population of the academic study.

Conclusions We find a large disparity between the strength of language as presented to the research consumer and the underlying strength of causal inference among the studies most widely shared on social media. However, because this sample was designed to be representative of the articles selected and shared on social media, it is unlikely to be representative of all academic and media work. More research is needed to determine how academic institutions, media organizations, and social network sharing patterns impact causal inference and language as received by the research consumer.

Although the challenges of population health differ widely between rich and poor countries, fundamental features of health behavior shed light on how individuals make choices about their health. These insights that can cut across countries and cultures. In this thesis, I apply concepts from behavioral economics to provide insights into how cognitive biases and social influences guide health behavior.

Paper 1 addresses inter-household spillovers and knowledge of HIV status. Using regression discontinuity design and a population-based dataset from South Africa, I estimate how a person's ART eligibility affects their household member’s HIV status knowledge. ART led to a large increase in HIV status knowledge among the patient's male household members. Although prior studies have noted a correlation between ART expansion and testing rates, this study is among the first to causally link ART initiation to increased awareness of HIV status among household members.

Paper 2 assesses the role of present bias and salience in malaria prevention behavior and risk perception in northern Ghana. Using lab-in-the-field measurement and high-frequency surveys of market vendors in Tamale, Ghana, I find that time preferences do not predict spending on malaria prevention or bednet utilization, but recent illnesses are associated with malaria prevention spending. I investigate the role of beliefs about malaria risk and find that respondents whose children had been ill in the past two weeks report higher subjective expectations of malaria risk, suggesting that recent episodes of illness may increase an individual's perception of risk and lead to increase spending on malaria prevention.

Paper 3 uses a behavioral field experiment to evaluate whether personal, goal-oriented reminders are an effective means to increase exercise frequency. I ran a 12-month randomized controlled trial on members of a chain of gyms in Montreal, Quebec. The trial compared generic SMS reminders with personalized reminders that recalled members' own exercise goals, which were elicited via a questionnaire at the time of study enrollment. I find that individuals who received personalized reminders did not exercise more frequently than the general reminder group and present suggestive evidence that recalling their goals generated a discouragement effect.