During a trial, it is important to measure baseline variables such as demographics (age, sex) and clinical characteristics (Angle classification, oral hygiene status, initial crowding). We usually measure baseline characteristics to describe participants, assess the similarity of treatment groups, and decide what will be included in the statistical analysis.
In this article, I will discuss the use of baseline values in statistical analysis that apply in dental trials that are usually small. The following example will be used to better explain this topic.
Research question: In a trial, 2 types of nickel-titanium wires were used to assess the effect of wire type in the resolution of crowding after 6 months of treatment. The variables used are wire, irregularity index before treatment (irprtx), and irregularity index at the end of the 6-month treatment period (irat6mo). The variables are summarized in Table I by wires A and B.
One option in analyzing this data set is to compare the mean irregularity index at the end of therapy using an independent t test as shown in Table II .
|n||Mean||SE||95% CI||P value|
We can see from the t test that there is no difference on average between wires A and B in terms of the amount of crowding at the end of the 6-month therapy.
Difference = 0.21 mm (95% CI, −0.18, 0.61; P = 0.29)
Exactly the same approach of the independent t test can be implemented using linear regression. Table III predicts the amount of crowding at the end of the 6-month therapy based on the wire type. Table III shows that the wire coefficient is 0.21, and this indicates that the amount of crowding at the end of the 6-month therapy is greater by 0.21 mm (95% CI, −0.18, 0.61; P = 0.29) with wire A compared with wire B. The results are exactly the same with the independent t test. To further clarify the interpretation, we can say that if we go from wire A to wire B (or the opposite, depending on which wire was used as the reference), we have a difference in the amount of residual crowding at 6 months of 0.21 mm between wires A and B. The range of this difference implied by the 95% confidence interval is –0.18 to +0.61 mm, and it is a statistically nonsignificant finding ( P = 0.29).