Individual covariates [age, sex, race and morphine dose in mg/kg POD1 and 2, preoperative anxiety score (VAS), preoperative pain score, duration of surgery, vertebral levels fused, propofol and remifentanil doses used during surgery (per kg), use of intravenous acetaminophen/ketorolac (Yes/No), diazepam doses (mg/kg), and sequential scores for CASI, PCS-C, PCS-P and PPH] were analysed to identify those associated with AUC using simple linear regression models. Similarly, the same factors as well as AUC were evaluated to identify factors associated with CP and PP using univariate logistic regression models; those associated at a p < 0.10 were entered into multivariable models and stepwise selection used to derive a final model for each outcome where only variables with a p < 0.05 were retained (Bursac et al., 2008 (link)). Correlation between two continuous variables was examined using either Spearman’s or Pearson’s correlation coefficient as appropriate. Linear trajectories of pain scores were estimated for different combinations of PP and CP outcomes for subjects who had both CP and PP outcomes reported. (i.e. CP = No, PP = No; CP = No, PP = Yes; CP = Yes, PP = No; and CP = Yes, PP = Yes).
Power calculation was done using PASS (Power and Sample Size,© 2008, Kaysville, Utah). Assuming an incidence of our main binary outcome (Y = persistent pain) to be 20–30% (based on prior studies which show an incidence of 22% (Page et al. 2013b (link)) to 29.5% (Landman et al., 2011 (link); Cudilo et al., 2014 ), and our pilot data), for a logistic regression of (Y) on a continuous, normally distributed variable (X). With a sample size of 100, we have 80% power to detect an effect size of 2–2.7 at a 0.05 significance level (α = 0.05). This assumes X’s multiple correlations with covariates already in the model is 0.5. With lower multiple correlation between X and other covariates, we can achieve the same power with less samples, or detect a smaller effect size. The sample size required increases to 118 assuming an expected loss to follow-up of 15%.
Power calculation was done using PASS (Power and Sample Size,© 2008, Kaysville, Utah). Assuming an incidence of our main binary outcome (Y = persistent pain) to be 20–30% (based on prior studies which show an incidence of 22% (Page et al. 2013b (link)) to 29.5% (Landman et al., 2011 (link); Cudilo et al., 2014 ), and our pilot data), for a logistic regression of (Y) on a continuous, normally distributed variable (X). With a sample size of 100, we have 80% power to detect an effect size of 2–2.7 at a 0.05 significance level (α = 0.05). This assumes X’s multiple correlations with covariates already in the model is 0.5. With lower multiple correlation between X and other covariates, we can achieve the same power with less samples, or detect a smaller effect size. The sample size required increases to 118 assuming an expected loss to follow-up of 15%.
