Impact of a simulation-based induction programme in gastroscopy on trainee outcomes and learning curves

Table 4 Generalised estimating equation models of procedure outcomes in cases and controls

	Cases		Controls		P value (Case vs Control)
	Odds ratio (95%CI)	P value	Odds ratio (95%CI)	P value
Unassisted D2 intubation rates
Intercept	0.51 (0.13 – 1.98)	-	-	-	0.332
Gradient (per Doubling of OGD count)	1.99 (1.69 – 2.34)	< 0.001	1.74 (1.53 – 1.98)	< 0.001	0.205
Moderate-severe discomfort
Intercept	0.42 (0.15 - 1.15)	-	-	-	0.09
Gradient (per 10 procedures)	0.97 (0.88 - 1.07)	0.526	0.92 (0.85 - 1.00)	0.044	0.421
Unsedated procedures
Intercept	1.63 (1.09 – 2.46)	-	-	-	0.018
Gradient (per 10 procedures)	0.99 (0.97 – 1.01)	0.28	1.00 (0.98 – 1.02)	0.973	0.445

Results are from generalised estimating equation models of the 200 procedures after the date of the course for each trainee. Each model contained the trainee group (case-control) and the procedure number as covariates, along with an interaction term. As such, the intercept represents the baseline difference between the case and control groups. The gradient represents the change in the outcome rate with increasing experience, with separate gradients reported for the case and control groups. For unassisted D2 rates, the procedure number was log-transformed in the model, hence the resulting coefficients were anti-logged, and gradients were reported per two-fold increase in procedure count. For the discomfort and unsedated procedures outcomes, gradients are reported per 10 additional procedures. The final column compares the gradients between groups, using the P value from the interaction term in the model. Bold P values are significant at P < 0.05.