Improved interpretation of studies comparing methods of dietary assessment: combining equivalence testing with the limits of agreement

Abstract

The aim of this study was to demonstrate the use of testing for equivalence in combination with the Bland and Altman method when assessing agreement between two dietary methods. A sample data set, with eighty subjects simulated from previously published studies, was used to compare a FFQ with three 24 h recalls (24HR) for assessing dietary I intake. The mean I intake using the FFQ was 126·51 (sd 54·06) µg and using the three 24HR was 124·23 (sd 48·62) µg. The bias was -2·28 (sd 43·93) µg with a 90% CI 10·46, 5·89 µg. The limits of agreement (LOA) were -88·38, 83·82 µg. Four equivalence regions were compared. Using the conventional 10 % equivalence range, the methods are shown to be equivalent both by using the CI (-12·4, 12·4 µg) and the two one-sided tests approach (lower t=-2·99 (79 df), P=0·002; upper t=2·06 (79 df), P=0·021). However, we make a case that clinical decision making should be used to set the equivalence limits, and for nutrients where there are potential issues with deficiency or toxicity stricter criteria may be needed. If the equivalence region is lowered to ±5 µg, or ± 10 µg, these methods are no longer equivalent, and if a wider limit of ±15 µg is accepted they are again equivalent. Using equivalence testing, acceptable agreement must be assessed a priori and justified; this makes the process of defining agreement more transparent and results easier to interpret than relying on the LOA alone.

Keywords: 24HR 24 h recall; Agreement; BA Bland and Altman; Bland and Altman method; Dietary assessment; Equivalence; LOA limits of agreement.