Regression to the Mean: A Statistical Phenomenon of Worthy Consideration in Anemia Research

Abstract

Background Regression to the mean (RTM) is a statistical phenomenon where second measurements are more likely to be closer to the mean. This is particularly observed in those with baseline values further from the mean. Anemic individuals (hemoglobin < 120g/L) are often recruited when evaluating iron supplementation programs, as they are more likely to elicit a greater hemoglobin response; however, they are also at greater risk for RTM as their baseline values are lower than the overall population mean. Objective To calculate and apply RTM to a previously conducted iron supplementation trial of women in Cambodia at increasingly severe baseline anemia cut-offs (hemoglobin < 120g/L, <115g/L and <110g/L). Methods Women received either 60mg/day iron (n = 191) or placebo (n = 185) for 12 weeks. Hemoglobin was measured at baseline and 12 weeks (endline), and change in hemoglobin was calculated in each group, for each cut-off. RTM was calculated in the placebo group at each cut-off and applied to the change observed at each cut-off in the iron group to obtain the RTM-free effect. Results In the placebo group, mean change in hemoglobin increased as cut-offs became more extreme (0.9g/L to 1.9g/L in those with baseline hemoglobin < 120g/L and <110g/L, respectively). RTM estimates similarly increased: 1.0g/L (<120g/L), 1.3g/L (<115g/L), and 1.8g/L (<110g/L). When applying RTM to the iron group, we found that ∼10% of the “treatment effect” could be attributable to RTM at each cut-off. However, iron supplementation was still effective in increasing hemoglobin, with an increased effect in those with lower baseline values, as proven by the RTM-free effect at each cut-off: 8.7g/L (<120g/L), 10.9g/L (<115g/L), and 13.6g/L (<110g/L). Conclusion RTM may have accounted for ∼10% of the observed change in hemoglobin following iron supplementation; however, appropriate use of a placebo group in the statistical analyses of the trial controls for this potential RTM effect.

JPEN Journal Club 74. Regression to the mean

“…11 Methods have been developed to quantify regression to the mean, although they are complicated and are not calculable by the average reader. 10,12 You may have other thoughts about this paper. If so, I would encourage you to discuss them with your colleagues or even send an email to me (rkoretz@msn.com).…”

“…We have contrasted these types of comparisons previously (differences in nominal significance) 11 . Methods have been developed to quantify regression to the mean, although they are complicated and are not calculable by the average reader 10,12 …”