Statistical analysis
An overall model to assess differences in force according to the method
of teaching was used. The dependent variable was the “post-teaching
measurements”, categorical variables indicated patient status
(“awake” and “anaesthetised”) and teaching method (“biofeedback”,
“nose” and “syringe”), and their interaction, and “baseline
measurements” were considered as a continuous covariate. This model is
conventionally formulated as an analysis of covariance (ANCOVA) and in
Stata can be estimated using the “anova” (analysis of variance),
“regress” (linear regression) or “mixed” (linear mixed [random
effects] model, LMM)[21]. For the LMM, a random coefficient model
was formulated with participants considered as random intercepts and
baseline measurements as random effect slopes (for each participant).
The incorporation of both “anaesthetised” and “awake” tests into the
overall model allowed the parts of the model to borrow strength from
each other. Model predictions were calculated using the “margins”
command of Stata™[22]. Model adequacy for the ANCOVA, linear
regression and LMM models was adjudged according to the information
criteria; lower scalar values preferred (AIC, assessing best model
prediction), Bayesian information criterion (BIC, best overall model)
and the likelihood ratio test[23]. Model estimates for each
“method” and “status” are displayed in graph format using the Stata™
“marginsplot” command.
Exact P-values are reported for parameter estimates; P-values
<0.05 were deemed to be significant. Statistical analysis was
performed using Stata™ (V 15.1) software.