Balance of Covariates Sample Clauses

Balance of Covariates. To test the balance of covariates using propensity score methods, the associ- ation of patient characteristics with the indicator of treatment was examined for change after conditioning on the proposed propensity scores. Table 2.3 presents the p-values of these associations when examined marginally, condi- tioned on the standard propensity score as calculated from a logistic regression of the treatment indicator, and conditioned on the generalized propensity score as calculated from a Xxx proportional hazards regression of the time to treat- ment. Overall, the use of propensity scores does balance the covariates across treatment groups. Most patient characteristics have a significant marginal association with treatment. After conditioning on the standard propensity score, all associations of patient characteristics with treatment become non- significant, indicating that balance of covariates is successfully achieved. Con- ditioning on the generalized propensity score is also fairly successful in balanc- ing covariates, with most patient characteristic associations with treatment Table 2.2: Comparison of PEG treated and untreated populations 1 year post- baseline (N=580) Treated Untreated (N=200) (N=380) Mean/P SD or n Mean/P SD or n p-value BMI at to 23.87 5.52 24.89 5.63 0.05 Baseline (BL) BMI 24.92 5.89 25.62 5.68 0.19 ∆BMIto , -1.13 2.12 -0.65 2.92 0.04 FVC at to 46.84 21.44 63.49 25.11 < 0.01 Baseline (BL) FVC 65.24 25.58 73.98 25.84 < 0.01 ∆BMIto -18.86 22.09 -10.42 16.96 < 0.01 Age at Diagnosis 64.91 10.15 61.91 12.17 < 0.01 ∆TDX > 30 days 0.21 41 0.19 71 0.68 Proportion Surviving 0.42 83 0.54 206 < 0.01 Prop. of Females 0.52 103 0.41 156 0.02 Prop. of Spinal Onset 0.43 86 0.80 304 < 0.01 becoming non-significant. Though two patient characteristics (baseline FVC and number of visits in the first year of follow-up post-baseline) remain as- sociated with treatment after conditioning on the generalized propensity score.