Analysis of covariance, Part 2

In the previous article I used an example comparing residual tooth irregularity after 6 months in 2 patient groups treated with different nickel-titanium wires. I discussed the adjustment of the analysis for the baseline value of crowding, indicating that differences between the initial amount of crowding between treatment groups could confound the results.

The unadjusted and adjusted models are shown in Table I . In this article, I will further discuss the baseline adjustment approach.

Table I
Comparison of unadjusted and adjusted models
Tooth irregularity at 6 months
Unadjusted model Adjusted model
β coefficient 95% CI β coefficient 95% CI
Group A vs B 0.21 −0.18, 0.61 0.28 −0.06, 0.61
Crowding before treatment 0.19 0.12, 0.27
R 2 0.019 0.314

As previously discussed, a common approach is to use the final value, such as the residual crowding at the end of 6 months, without accounting for the initial crowding ( t test or unadjusted linear regression can be used for analysis purposes). This method is sufficient if the correlation between the pretreatment and posttreatment values is less than 0.5 and is preferred over change from baseline (discussed below), since it will be more precise under the less-than-0.5-correlation scenario. Another approach is to use the change from baseline as shown in Table II . The interpretation of the results is the same: the use of wire A compared with wire B results in lowering the change from baseline by 0.54, indicating more residual crowding for A; this agrees with the previous models.

Apr 4, 2017 | Posted by in Orthodontics | Comments Off on Analysis of covariance, Part 2
Premium Wordpress Themes by UFO Themes