search

erikhuizinga / Bland-Altman-Analysis

A MATLAB function for Bland-Altman Analysis of agreement between measurement methods.
5 stars 3 forks source link

Different p-value from MedCalc #21

Open mahjalili opened 5 years ago

mahjalili commented 5 years ago

Dear Sir, The analysis result are different from MedCalc result. Your function result is as below: muD: -1.3491 muDCI: [-10.1650 7.4668] loa: [-965.4156 962.7174] loaCI: [2×2 double] sD: 491.8797 rSMuD: 0.2301 pRSMuD: 1.9598e-143

but the result from MedCalc program is: Arithmetic mean | -1.3491 95% CI | -10.1650 to 7.4668 P (H0: Mean=0) | 0.7642 Standard deviation | 491.8797 Lower limit | -965.4333 95% CI | -980.5004 to -950.3662 Upper limit | 962.7351 95% CI | 947.6680 to 977.8022

difference is in p-value. In fact with muDCI: [-10.1650 7.4668], the p-value cannt be significant. Thanks

erikhuizinga commented 5 years ago

Hi @mahjalili! Thanks for reaching out to me and for your interest in my work.

Are you comparing pRSMuD: 1.9598e-143 from my BA function with P (H0: Mean=0) | 0.7642 from MedCalc? Those are not references to the same p-value! Mine is the Spearman rank correlation's (rSMuD) p-value at the alpha (default = 0.05) significance level.

On a side note, I don't know what use a p-value for a null hypothesis that the mean of the analysis (muD / Arithmetic mean) is zero. It has no value, because:

If measurement methods A vs. B have x arithmetic mean, where x is nonzero, you can just subtract x from the measurements of A to obtain the same mean as measurement done with method B. In other words: if you would subtract that mean from A and then do the BA analysis, you would get an arithmetic mean of exactly zero.

Note that neither A nor B are correct (although one of the two sometimes is considered the gold standard, that needs not be the case). So aiming for zero mean difference is useless, as a correction is easily made if necessary to achieve that.