|LETTER TO EDITOR
|Year : 2011 | Volume
| Issue : 5 | Page : 550
Harihar V Hegde, Rajashekar R Mudaraddi, Vijay G Yaliwal, P Raghavendra Rao
Department of Anaesthesiology, SDM College of Medical Sciences and Hospital, Sattur, Dharwad, Karnataka, India
|Date of Web Publication||14-Nov-2011|
Harihar V Hegde
Department of Anaesthesiology, SDM College of Medical Sciences and Hospital, Dharwad, Karnataka
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
Hegde HV, Mudaraddi RR, Yaliwal VG, Rao P R. Authors' reply. Indian J Anaesth 2011;55:550
The aim of the study  was to compare the accuracy of blood pressure measurements obtained by Datex-Ohmeda GE S/5 and Criticare 8100E nGenuity. In the current study, the same group underwent repeated measurements four times, once each by the two observers by manual checking and once each by the two machines. Measurements were taken within 10 min to imply that there was no other influence than the methods of measurements alone. We initially applied paired t-test and identical results were also observed by doing one-way analysis of variance (ANOVA)-single factor as well as ANOVA-two way; not writing this in the publication explicitly is regretted as we thought repeated measures ANOVA is an extension of the paired t-test only. It was found that the two observers and Datex-Ohmeda GE S/5 agree and only the Criticare 8100E nGenuity is the source of difference observed which recorded statistically significant higher readings. These values may or may not be medically significant and neither have we made any claim of medical significance.
For the paired t-test, the data are dependent, i.e. there is one-to-one correspondence between the values in the two samples. Paired data can occur for two measurements on the same person, e.g. before and after treatment, or the same subject measured at different times. It is incorrect to analyse paired data ignoring the pairing in such circumstances as important information is lost. Some factors that are not controlled in the experiment will affect the before and after measurements equally, so they will not affect the difference between before and after. By looking at only the differences, a paired test corrects for these factors.
The two-sample paired t-test usually tests the null hypothesis that the population mean of the paired differences of the two samples is zero. We assume that the paired differences are independent. To perform the paired t-test, we calculate the difference between each set of pairs and then perform a one-sample t-test on the differences with the null hypothesis that the population mean of the differences is zero. 
Repeated measures ANOVA is also useful in non-interventional single sitting comparison of two methods as in our study. Repeated measures ANOVA-one way as well as the paired t-test do use the same 95% confidence limits and essentially look for the same limits of agreement, whether it is significant.
Bland and Altman analysis is useful when a new different technique is being tested to compare its reliability with an already established method by quantifying the agreement between the two methods through the mean difference (i.e. the estimate of the systematic bias of one method relative to the other) and the standard deviation of the differences between measurements taken by the methods on the same subjects (i.e. an indication of the variability of these differences across subjects).  This method is used when neither method is an established gold standard, and the new method needs to be compared with an old method to merely decide if the two methods agree sufficiently with each other.
| References|| |
|1.||Hegde HV, Mudaraddi RR, Yaliwal VG, Rao PR. Comparison of oscillometric blood pressure measurement by two clinical monitors: Datex Ohmeda GE S/5 and Criticare 8100E nGenuity. Indian J Anaesth 2011;55:199-200. |
|2.||Everitt BS, Palmer CR, editors. The Encyclopaedic Companion to Medical Statistics. London: Hodder Arnold; 2005. p. 339. |
|3.||Everitt BS, Palmer CR, editors. The Encyclopaedic Companion to Medical Statistics. London: Hodder Arnold; 2005. p. 193-4. |