Original articleRegression to the mean and changes in risk behavior following study enrollment in a cohort of U.S. women at risk for HIV
Introduction
In human immunodeficiency virus (HIV) and sexually transmitted infection (STI) prevention trials reported risk behaviors are often observed to decline over the course of the study, even in the control group [1], [2]. Such declines may be attributed to a number of factors, including counseling messages and services provided by the study, a general increase in feelings of self-worth due to study participation, healthy survivor effects, aging of the participants, and/or the Hawthorne effect [3]. However, an important alternative explanation that should be considered, and can be quantified, is the phenomenon of regression to the mean [4].
Regression to the mean (RTM) is a statistical phenomenon that occurs when study participants are selected for extreme values of characteristics or behaviors that vary over time. RTM predicts that subsequent measures of those characteristics or behaviors will, on average, be closer to the underlying population mean than the original values. For example, if subjects are selected such that 100% of participants have a particular behavior at enrollment, then (assuming the behavior varies over time) RTM predicts that the prevalence of that behavior will decline as the study progresses, even in the absence of any intervention. However, it is not clear how much of a decline might be expected due to RTM, and how rapidly the decline might occur.
RTM has been studied extensively, usually in the context of continuous measures (e.g., blood pressure [5], heights of fathers and sons [6], and medical costs [7]) or count data (sex acts, sex partners [8]). However, little has been written about RTM in the context of binary measures (e.g., unprotected sex in the past 6 months) even though such measures are common in epidemiologic and behavioral studies. In this manuscript, we develop methods for estimating the expected change in the prevalence of a selected binary behavior due to RTM. We apply these methods to data from the HIV Prevention Trials Network (HPTN) Women's HIV SeroIncidence Study HPTN 064 [9] to quantify the amount of change in a key risk behavior—unprotected sex in the past 6 months—that is associated with study participation versus RTM.
Section snippets
Study design
HPTN 064 was a multisite, longitudinal cohort study designed to estimate HIV incidence among women at elevated risk for HIV infection. Details regarding the study design have been previously described [9]. In brief, eligible women were enrolled between May 2009 and July 2010 from 10 urban and periurban communities in six geographic areas of the United States (Atlanta, GA; Baltimore, MD; New York City, NY; Newark, NJ; Raleigh-Durham, NC; Washington, DC) using venue-based recruitment.
Results
A total of 8029 women were screened for HPTN 064 (Fig. 1). Of those, 4126 women satisfied all eligibility criteria, aside from the requirement of unprotected sex in the last 6 months. Thus, these 4126 women represent (a sample from) the unselected population. At screening, a total of 3234 of these women (78.4%) reported unprotected sex with a male during the past 6 months (screened “positive”) and 2099 (64.9%) of these chose to participate in HPTN 064. Details on follow-up and retention are
Discussion
When individuals are selected for high or low values of a time-varying quantity (e.g., presence of a risk behavior or biologic marker) and then measured again on that quantity, regression to the mean is inevitable. This phenomenon has been demonstrated in a wide variety of settings including education (e.g., test scores [14]; medicine, e.g., treatment for hypertension [5]; and sports, e.g., the “sophomore slump” [15]), among others. Although less often discussed in the context of binary
Acknowledgment
The authors thank the study participants, community stakeholders, and staff from each HPTN 064 study sites. In addition, they acknowledge Lynda Emel, Jonathan Lucas, Nirupama Sista, Kathy Hinson, Elizabeth DiNenno, Lisa Diane White, Waheedah Shabaaz-El, LeTanya Johnson-Lewis, Carlos del Rio, Christin Root, Manya Magnus, Christopher Chauncey Watson, Quarraisha Abdool-Karim, and Sten Vermund.
This study was sponsored by the National Institute of Allergy and Infectious Diseases, National Institute
References (16)
- et al.
Blood pressure reductions: correcting for regression to the mean
Prev Med
(1983) - et al.
Feasibility of identifying a cohort of US women at high risk for HIV infection for HIV vaccine efficacy trials: longitudinal results of HVTN 906
J Acquir Immune Defic Syndr
(2013) - et al.
The occurrence of vaginal infections among HIV-infected and high-risk HIV-uninfected women: longitudinal findings of the women's interagency HIV study
J Acquir Immune Defic Syndr
(2006) - et al.
The Hawthorne effect: a randomized controlled trial
BMC Med Res Methodol
(2007) - et al.
Statistic notes: regression towards the mean
BMJ
(1994) Regression towards mediocrity in hereditary stature
The J Anthropological Institute of Great Britain and Ireland
(1886)Assessing regression to the mean effects in health care initiatives
BMC Med Res Methodol
(2013)- et al.
Modeling seroadaptation and sexual behavior among HIV+ study participants with a simultaneously multilevel and multivariate longitudinal count model
Biometrics
(2013)
Cited by (0)
The views expressed herein are solely the responsibility of the authors and do not necessarily represent the official views of the National Institute of Allergy and Infectious Diseases, the National Institute of Mental Health, the National Institutes of Health, the HPTN, or its funders.
This study is registered at Clinicaltrials.gov (NCT00995176, http://clinicaltrials.gov).