Dear Editor,

We read the interesting publication from Hasson et al. [1] comparing blood levels of ionized magnesium (iMg) to total magnesium (tMg) in pediatric patients who had undergone kidney transplantation. Magnesium, similar to calcium, exists in the blood in either ionized (free) form, bound to a protein such as albumin, or in chelated form. The ionized forms of magnesium and calcium represent the biologically active forms. Based on their observations examining both iMg and tMg, the authors concluded that iMg and tMg levels correlated strongly, with strong categorical agreement between these parameters in pediatric patients. After reviewing the evidence, we feel that these conclusions are not supported by the data presented and we wish to clarify this point, as it may impact clinical decision making.

The authors used two statistical tests for clinical category agreement, which is comprehensive. Fisher’s exact test showed high statistical significance (P=0.0003), implying a low probability that the agreement between iMg and tMg occurred by chance. The unweighted kappa statistic was 0.27 (95% CI: 0.056–0.47) and was also statistically significant (P=0.013).

In interpreting the kappa statistic for agreement, the literature indicates that a value of 0.27 is interpreted as fair. For example, Landis and Koch [2] categorized a kappa coefficient between 0.21 and 0.40 as a fair strength of agreement, which is weaker than moderate, substantial, or almost perfect. In other sources [3-5], kappa in the 0.21–0.39 range was considered minimal [4], and <0.4 was categorized as poor [5]. Therefore, a low value, such as 0.27 is generally not accepted to be interpreted as good or strong. Rather, more accurate interpretations would be fair, minimal, or poor agreement between iMg and tMg.

To further evaluate the category agreements, we tabulated their percentages. Although the sample size was small (as the authors mentioned), overall, the results of 24/43 (56%) samples were in clinical agreement. Based on this low percentage of agreement and the interpretation of the kappa statistic, we consider it more accurate to state that the category agreement between iMg and tMg was statistically significant but poor, in contrast to the authors conclusion that “iMg and tMg showed strong category agreement” (stated in the abstract) or that there was “good category agreement” (stated in the last paragraph). In support of these arguments, a recent study by Bargnaux et al. [6] measured the clinical agreement between iMg and tMg in dialysates of patients undergoing dialysis. Bargnaux et al. reported 56.9% agreement and a weighted kappa statistic of 0.31, which were nearly identical to those observed by Hasson et al. [1]. However, Bargnaux et al. correctly interpreted these statistics as “indicating a minimal agreement to assess Mg status” [6].

A more subtle point can be made regarding interpreting the correlation between iMg vs. tMg as strong based on the Pearson’s r coefficient (0.707) and least-squares R² value (0.5003). In this matter, some authors have interpreted r values between 0.7 and 0.8 as indicating good correlations. Certainly, this would be considered a good correlation for two unrelated or distinct parameters. However, iMg and tMg represent two different forms of the same molecule; therefore, it is expected that their levels should generally increase together and that some correlation should exist. When comparing two assays for measuring the same analyte, the stringent requirements of the Food and Drug Administration often require correlation values of 0.9, 0.95, or greater to support the statement or conclusion that two assays are comparable or equivalent. Therefore, the interpretation of a strong correlation may depend on the frame of reference. With a stringent requirement for comparable assays or interchangeable/equivalent assays, the r value found in the publication of interest and in other publications would justify an interpretation that iMg and tMg were not equivalent or interchangeable.

Although these definitions and delineations between strength of correlation and strength of agreement can be subjective, they are important for clinical decision making. Therefore, accurately interpreting category agreements or disagreements between iMg and tMg is important for understanding how measuring them may impact clinical care. Overall, we believe that this well controlled study demonstrated a statistically significant yet poor clinical category agreement between biologically active iMg and tMg in the blood of pediatric patients who underwent kidney transplantation. Based on the correlation data between iMg and tMg, we suggest that these tests were not interchangeable or equivalent for this patient population.