Asheley Cockrell Skinner1,2, TaShauna U Goldsby3,4, David B Allison3,4,5. 1. 1 Department of Pediatrics, Division of General Pediatrics and Adolescent Medicine, The University of North Carolina at Chapel Hill , Chapel Hill, NC. 2. 2 Department of Health Policy and Management, The University of North Carolina at Chapel Hill , Chapel Hill, NC. 3. 3 Nutrition Obesity Research Center, University of Alabama at Birmingham , Birmingham, AL. 4. 4 Office of Energetics, University of Alabama at Birmingham , Birmingham, AL. 5. 5 Department of Biostatistics, University of Alabama at Birmingham , Birmingham, AL.
Abstract
OBJECTIVE: In this paper we discuss what regression to the mean (RTM) is, the magnitude of RTM in realistic situations, interpretation of RTM, and recommendations for how to address RTM in study design. METHODS: Public health research faces many challenges in conducting gold standard randomized, controlled trials (RCT). Although there are many threats to validity in uncontrolled trials, RTM is often overlooked or not adequately considered. RTM is a statistical phenomenon that occurs with any pair of variables that have a correlation not equal to |1.0|. With RTM, subjects' average values on an outcome variable (e.g., BMI) change in a systematic direction over time despite there being no treatment effect. Without a proper control group, changes thought to be associated with an intervention may be due entirely to RTM. Investigators may draw erroneous conclusions based on results showing greater declines in a variable among participants with higher baseline of that variable compared to those with lower baseline of that variable, and label this evidence for differential treatment efficacy. CONCLUSIONS: Ignoring RTM can lead to unsubstantiated conclusions about the effects of treatments. These conclusions can lead to the waste of time, money, and other resources, which distract from finding appropriate interventions. When a true RCT design is not feasible, reasonable design alternatives involving nonrandomized control groups should be implemented.
OBJECTIVE: In this paper we discuss what regression to the mean (RTM) is, the magnitude of RTM in realistic situations, interpretation of RTM, and recommendations for how to address RTM in study design. METHODS: Public health research faces many challenges in conducting gold standard randomized, controlled trials (RCT). Although there are many threats to validity in uncontrolled trials, RTM is often overlooked or not adequately considered. RTM is a statistical phenomenon that occurs with any pair of variables that have a correlation not equal to |1.0|. With RTM, subjects' average values on an outcome variable (e.g., BMI) change in a systematic direction over time despite there being no treatment effect. Without a proper control group, changes thought to be associated with an intervention may be due entirely to RTM. Investigators may draw erroneous conclusions based on results showing greater declines in a variable among participants with higher baseline of that variable compared to those with lower baseline of that variable, and label this evidence for differential treatment efficacy. CONCLUSIONS: Ignoring RTM can lead to unsubstantiated conclusions about the effects of treatments. These conclusions can lead to the waste of time, money, and other resources, which distract from finding appropriate interventions. When a true RCT design is not feasible, reasonable design alternatives involving nonrandomized control groups should be implemented.
Authors: Diana M Thomas; Nicholas Clark; Dusty Turner; Cynthia Siu; Tanya M Halliday; Bridget A Hannon; Chanaka N Kahathuduwa; Cynthia M Kroeger; Roger Zoh; David B Allison Journal: Am J Clin Nutr Date: 2020-02-01 Impact factor: 7.045
Authors: Kelsey M Cochrane; Brock A Williams; Jordie A J Fischer; Kaitlyn L I Samson; Lulu X Pei; Crystal D Karakochuk Journal: Curr Dev Nutr Date: 2020-09-24
Authors: Andrew W Brown; Douglas G Altman; Tom Baranowski; J Martin Bland; John A Dawson; Nikhil V Dhurandhar; Shima Dowla; Kevin R Fontaine; Andrew Gelman; Steven B Heymsfield; Wasantha Jayawardene; Scott W Keith; Theodore K Kyle; Eric Loken; J Michael Oakes; June Stevens; Diana M Thomas; David B Allison Journal: Obes Rev Date: 2019-08-19 Impact factor: 9.213