Serum resistin, adiposity, and insulin resistance in Saudi women: conflicting data or statistical flaws?

To the Editor: I have read with great interest the recently published article by Al-Harithy and Al-Ghamdi¹ in the Annals of Saudi Medicine and I appreciated the authors’ efforts and work. However, I would like to make few comments on it because of the apparent conflicting results of different studies investigating the differences in resistin levels between healthy subjects and type 2 diabetics, which the author admitted in their discussion. The authors mentioned that “these conflicting data may reflect the lack of adjustment for potential confounding factors, i.e. age and body fat distribution”.

I agree with the authors, partially, in their explanation. However, conflicting studies could also be attributed to non-fulfillment of certain prerequisites or assumptions before running a statistical model, a common problem in health research in which the authors themselves have fallen. The authors mentioned that “resistin concentrations were not correlated with BMI in lean subjects whereas there was a highly significant positive correlation between resistin and BMI in overweight/obese (OW/OB) nondiabetic and diabetic women”. That means the authors have calculated a Pearson correlation coefficient between BMI and resistin level for each of the three studied groups separately. Albeit that the first and second group constituted 21 and 24 participants, respectively, the authors did not mention in their methodology that they tested the three groups for linearity before running Pearson’s correlation for each. Linearity would be suspected given the paucity of participant numbers, especially for the first two groups. Pearson’s correlation coefficient is a measure of linear association. Two variables can be perfectly related, but if the relationship is not linear, Pearson’s correlation coefficient is not an appropriate statistic for measuring their association. If linearity is not met, we can use a transformation or one of the rank correlation methods.² I think it would also be better if the authors would pool the data of the three groups together (summing to 89 participants) and run a partial correlation between BMI and resistin adjusted for the type of group. Moreover, the results of any study should be consistent irrespective of the statistical tool used, especially if it is on the bivariate level. BMI was found to be significantly associated with resistin level by the ANOVA test (in Table 1). Accordingly, results of simple correlation should be in the same vein.

Moreover, when the authors mentioned in the results section that “HOMA-R were similar in lean and OW/OB subjects, but significantly higher in diabetic compared with non-diabetic women”, they were semantically incorrect. The right description of what their data revealed is that there was no significant difference in HOMA-R between the lean and OW/OB groups, whereas the difference was significant between the lean and diabetic group as well as the OW/OB and diabetic group.

As regards the adjustment of confounding variables mentioned above, the authors, in the methods section, mentioned that “the findings from the bivariate correlation analysis were further explored using stepwise multiple linear regression analysis with resistin concentration as the dependent variables.” I think it was better to enumerate the predictors or the independent variables in the regression model. If we look in Table 1 we could easily detect that many variables with strong co-linearity were significantly associated with resistin in bivariate analysis. The co-linearity of waist circumference with waist-hip ratio, and weight with BMI is manifest. Selecting one of any strongly co-linear variables should be done before running regression models.

Finally, I wish to close with what Altman et al³ concluded in their study: “Statistical input to medical research is widely recommended but inconsistently obtained. Individuals providing such expertise are often not involved until the analysis of data and many go unrecognized by either authorship or acknowledgment.”