EE: Healthy Adults: Individual Errors (2005)
What Are Individual Errors and Why Are They Important?
Most of the time, when researchers test to see how accurate a particular equation is they use a group mean error statistic. For example, researchers study groups of people, find the arithmetic mean or average, and then use statistical methods to figure out how wide the group variation is around that mean.
Is this Helpful when Applied to Individuals?
Let's examine how three individual RMR's are reflected in a group mean. For simplicity, let's say Joe, John and Judy all have an actual RMR of 1800 kcals/day. We select a predictive equation (EQ) and discover that in:
Joe, EQ overestimated RMR by 10% = -180 kcals/day error
John, EQ underestimated by 10% = +180 kcal/day error
Judy, EQ underestimated by 2% = +36 kcals/day error
The GROUP MEAN AVERAGE of all 3 is:
EQ estimates within 1% = 12 kcals/day error
If the individual named Joe was overweight, would a 180 kcal/day RMR error significantly change his dietary intake or impact his weight outcome? Also note, the practitioner wouldn’t know if the selected equation over- or underestimated RMR for Joe, John or Judy. Therefore, the practitioner needs to know, not how large or small group mean errors are, but what individual errors are likely and if there is an error, how large of an error will occur. Therefore, the practitioner needs to know
How often will this equation accurately estimate RMR in individuals? (For example, if 80 out of 100 individuals are accurately predicted within 10% of measured RMR, this would be 80%).
If the equation is wrong, how far off (high or low) might the RMR estimates be for this specific individual? (For example, after all the studies were evaluated, the worst underestimation error of an individual may have been 35% below the measured RMR and the worst overestimation error (possibly from another study) might have been 43% over measured RMR).
What Does the Research Tell Us About Individual Errors?
Figure 1 presents the likelihood that the different equations will predict RMR within ±10% of measured RMR and the range of underestimation and overestimation errors in healthy, non-obese adults. After the evidence was summarized, one study reported the Mifflin-St. Jeor equation was accurate (i.e.,within ±10% of measured RMR) about 82% of the time whereas multiple studies reported accuracy of Harris-Benedict equation varied and was accurate 45-81% of the time. The Mifflin-St. Jeor equation stated the worst underestimated error was 18% of RMR whereas the worst underestimated error using the Owen et al equation was 24% of measured RMR.
Figure 2 presents the individual error ranges that the different equations will predict RMR within 10% of measured RMR and the range of underestiamtion and overestimation errors in obese adults. Note, using adjusted body weight and the Harris-Benedict equation was the least accurate way to predict RMR.