BMR estimation equations can miss your individual metabolism

BMR equations can be useful without being exact for you. Research shows why average bias, individual errors, and the population tested all matter.

日本語で読む（Japanese version）

You enter your weight in the morning and a screen displays your basal metabolic rate. Another calculator takes the same height and age but returns a slightly different answer. Which number should guide lunch? A result with neat decimal places can feel as though someone has looked inside your body, rather than performed a calculation using a few details about it.

Calculated metabolism and measured metabolism are different things. A systematic review covering adults with and without obesity found Mifflin-St Jeor relatively reliable among commonly used equations, while still identifying substantial errors when predictions were applied to individuals.[15] The problem is bigger than a typing mistake. Even with correct inputs, you may be someone for whom an otherwise useful equation performs poorly.

The related feature on underreporting in food records takes up the question of how intake is recorded. Here, the focus is the origin of the expenditure number: what it represents, who it has been tested in, and how closely predictions matched measurements. The aim is to give a calculation an appropriate place in your decisions about food, rather than ask it to settle questions its validation studies did not answer.

Is the metabolism number measured or estimated?

The research uses both BMR, meaning basal metabolic rate, and RMR, meaning resting metabolic rate. These names identify what the researchers measured. This feature keeps each paper’s terminology rather than automatically treating the terms as interchangeable. Studies testing equations for basal metabolism and studies testing equations for resting metabolism both use indirect calorimetry as the comparison measurement.[2][15] Before comparing results, establish which outcome the equation predicts.

Indirect calorimetry is the method used to measure metabolism in these papers. Its role differs from that of an equation predicting metabolism from height, weight, and other inputs. The review of general adult populations explains that valid measurement can avoid prediction errors, while also emphasizing the need for a suitable protocol to minimize measurement error.[15] A measurement has conditions attached to it; the word “measured” does not remove the need to examine how it was obtained.

Equations are convenient. The study of people living with overweight or obesity explains that indirect calorimetry requires time and money, so predictive equations are often used instead.[2] A calculator can provide an estimate where measurement equipment is unavailable. That is a practical reason to use it, but also a reason to remember its function. The calculation draws on information entered into a model, and does not reveal everything about the particular person entering it.

Resting metabolism also differs from total energy requirements. A study developing equations for older adults describes RMR as accounting for 50–70% of total energy needs.[6] That statement belongs to the older-adult context of the paper. It should not become a universal percentage for converting everyone’s resting estimate into a daily food allowance. The useful distinction is between energy at rest and the requirements of life as a whole.

Imagine checking the number before dinner. If the screen says only “metabolism,” it may leave unanswered whether the figure is a prediction or a measurement, which equation produced it, and whether it refers to basal or resting metabolism. Those questions come before deciding whether the number looks high or low. The adult review’s concern is precisely that prediction errors can matter when estimates are used in an individual’s nutrition care.[15]

A display can be easy to read without being easy to interpret. Height, weight, and age are familiar, so a result based on them may feel personal. Yet entering personal details into a population-derived equation is different from having your metabolism measured. Keeping the equation’s name alongside the result makes its origin visible, and gives you something concrete to compare with the research evaluating it.[15]

A small average error can conceal large individual errors

“Bias” describes the average direction and size of the difference between predicted and measured values. Individual accuracy asks how close each person’s prediction came to their measurement. The systematic review of athletes explicitly separates closeness of average predicted values from the proportion of people whose estimates were within ±10% of their measured RMR.[4] Both questions matter, but they describe different aspects of performance.

An average difference near zero can arise when overestimates and underestimates cancel. It therefore describes the center of a group’s differences without directly showing their spread. A calculator used for dinner planning confronts the error for the person using it, not just the average error across the research sample. Studies in Chinese adults and older adults illustrate why both the center and the spread need attention.[1][6]

A study developing a new BMR equation in normal-weight Chinese adults used a development sample of 516 people. The new equation’s average bias was only 0.2 kcal/day. Its limits of agreement, however, extended from −514.3 to 513.9 kcal/day.[1] Reporting the tiny average alone would give a very different impression from reporting the average together with the range of agreement. The same equation can look reassuring at group level while leaving considerable uncertainty for individuals.

Limits of agreement describe the spread of differences between predictions and measurements. They are not a promise that your own error must fall inside that range, and they are not a range within which it is appropriate to add or remove food. They characterize the agreement found in the study. The separate validation sample helps show how the equation performed beyond the data used to develop it.[1]

The researchers also tested the new equation in a separate sample of 41 adults, where its reported accuracy was 75.6%. They still called for further validation.[1] That qualification is part of the result. A model can perform well enough to deserve more investigation without already having established how reliably it will work in many other people. A promising result and broad evidence of applicability are different stages of evaluation.

During development, 30 participants were excluded using a statistical criterion.[1] The selection and analysis of participants help define the conditions under which the equation achieved its reported performance. They should remain attached to the favorable numbers. Describing the result as “almost no error” on the strength of average bias alone would overlook both the spread of differences and the way the development data were handled.

Decimal places describe how a result is displayed, not how precisely a model knows your metabolism. When reading a claim about an accurate calculator, look for an individual accuracy rate or limits of agreement as well as average bias. Without them, it is difficult to tell whether the claim concerns the group’s center or the predictions people actually received.[1][6] That distinction is more useful than the apparent precision of the final digit.

The best equation depends on the adults being studied

The 2005 systematic review compared widely used equations in adults with and without obesity, giving particular weight to research reporting individual data. Among the equations examined, Mifflin-St Jeor placed more people within 10% of measured RMR and had the narrowest error range. However, older adults and ethnic minority groups living in the United States were underrepresented in both equation development and validation.[15] Its favorable ranking came with limits on generalization.

A 2013 validation study of 337 ambulatory, community-living adults also supported its usefulness. Mifflin-St Jeor was within 10% of measured RMR in 82% of participants; Livingston achieved 79%. The 95% confidence interval for the Mifflin-St Jeor mean difference was −26 to 8 kcal/day, with no clear group-level bias.[16] This confidence interval concerns the average difference. It does not describe the range of errors that individual participants experienced.

Within that same study, Mifflin-St Jeor accuracy was 87% in adults without obesity and 75% in adults with obesity.[16] The equation’s name stayed the same while its performance changed with the population. An overall accuracy rate can be useful, but a breakdown by relevant participant characteristics may tell you more about the evidence closest to your situation. The headline result does not make the subgroup difference disappear.

A cross-sectional study at a Belgian obesity outpatient clinic examined 731 people. Their mean age was 43 years, 79.5% were women, and mean BMI was 35.6. BMI is an index based on weight and height. Of 14 equations tested, Henry, Ravussin, and Mifflin-St Jeor reached the highest accuracy rate, at 73%. Mifflin-St Jeor and Henry were unbiased on average.[2] An equal accuracy rate did not mean identical average bias across those equations.

The assessment changed again in a 2024 systematic review and meta-analysis of adults with severe obesity. BMI across the included studies ranged from 40.0 to 62.4. The review included 40 studies, with 20 reporting mean differences contributing to the meta-analysis. WHO and Harris-Benedict were judged the most accurate and precise overall, while the other equations tended to underestimate BMR.[3] The ranking for severe obesity differed from the ranking in general adult populations.

For WHO, the mean difference was −12.44 kcal/day, with a 95% confidence interval from −81.4 to 56.5 kcal/day. For Harris-Benedict, it was −18.9 kcal/day, with a 95% confidence interval from −73.2 to 35.2 kcal/day.[3] These are group-level results from the analysis. They do not establish the size of your own error, even if your BMI falls within the ranges studied.

Most studies in that review had a high risk of bias, and additional analyses suggested that equation performance might differ across obesity BMI ranges.[3] The high risk of bias in these studies limits how confidently their combined results can be interpreted. When searching for a recommended equation, ask whether the recommendation concerns general community-living adults, an obesity clinic, or severe obesity. Different rankings can reflect different populations and evaluation measures rather than an inconsistency in arithmetic.[2][3][15]

Numbers: BMR estimation and individual agreement

Community-living adults, 337 participants: Mifflin-St Jeor accuracy was 82%, defined as being within 10% of measured RMR.[16]

Obesity outpatient clinic, 731 participants: the highest equation accuracy was 73%; average bias was assessed separately.[2]

Normal-weight Chinese adults, new equation: average bias was 0.2 kcal/day, while limits of agreement were −514.3 to 513.9 kcal/day.[1]

Older adults, 1,686 participants: population mean bias was about 1%, while individual limits of agreement were approximately ±25%.[6]

Athlete meta-analysis: Ten-Haaf placed 80.2% within ±10% of measured RMR, a result from the athletic populations studied.[4]