Proportionality Assumption. The assumption of proportionality, for all time to event analyses, will be assessed firstly by examining plots of complementary log-log (event times) versus log (time) and, if this raises concerns, subsequently fitting a time dependent covariate (adding a treatment-by-time or treatment-by-ln(time) interaction term) to the model to assess the extent to which this represents random variation. If a lack of proportionality is evident then the HR can still be meaningfully interpreted as an average HR over time unless there is extensive crossing of thetime to event curves. Sensitivity analysis The following sensitivity analyses will be performed on PFS data using the ITT. Evaluation-time bias assessment For PFS, a supportive analysis will be performed to assess time assessment bias via an Evaluation-time bias approach (Xxxxxx et al. 1983, Xxx et al. 2010). This will be performed using a validated SAS program for generation of the p-value. The most pragmatic and preferred approach to assessing evaluation-time bias is to analyse, using standard log-rank , the midpoint between the time of progression and the previous visit, this approach has been shown to be robust to even highly asymmetric assessment schedules. To support this analysis, the mean of patient-level average inter-assessment times will be tabulated for each treatment. Attrition bias assessment A further analysis of the primary endpoint will be conducted, whereby patients who received subsequent anti-cancer therapies prior to progression (or death in the absence of progression) will be censored at the time patients received subsequent therapies, while patients who progress or die following two or more missed visits will not be censored. The analysis will use the same approach as the primary analysis model (outlined within Section 4.2.2). A list of relevant anti-cancer therapies will be provided by the study team prior to database lock (DBL). In support of this analysis, a Xxxxxx- Xxxxx plot where the roles of events and censored observations are reversed will be presented to assess the impact of rate and nature of censoring on final results. Additional supportive summaries/graphs A summary of the number of patients who have incomplete PFS follow-up, i.e. were alive and progression free at the date of data cut-off and had not had a RECIST assessment within 12 weeks of the data cut-off will be provided by treatment group. Additionally, summary statistics will be given for the number of days from censoring...
Proportionality Assumption. For a Xxx proportional-hazards model to be appropriate, the assumption of proportionality must be satisfied. This means that the relative hazard of failure across groups must be constant across time (Xxxxxxx, 2003 p. 46-7). In other words, if one firm has twice the risk of failing as another on day one, this relative failure risk must remain constant over all points in time. To test whether the proportionality assumption was satisfied among covariates in my data set, I ran a regression of the primary covariates in my study and saved the scaled Xxxxxxxxxx residuals. Then, I tested whether any covariate significantly deviated from proportionality. Two of the original covariates, “prepaid services” and “gender diversity” failed to satisfy the proportionality assumption. Analyses showed these measures to contribute little explanatory power to the model, even when their effects were examined over the shorter discrete time periods in which the proportionality assumption was satisfied; they were thus dropped from the reported analyses. For all other covariates, there was insignificant evidence to reject the proportionality assumption. Details of these tests are reported in Table 2. In addition, plots of the residuals for each covariate are included in Appendix B. Table 2: Test of Proportional-Hazards Assumption Time: Log(t) Plot Covariate rho chi2 df Prob>chi2 Pass? 2 Industry annual CLEC capital expenditures (ln) –0.11994 2.43 1 0.1189 √ 3 Firm business model (resale or facilities) 0.03025 0.13 1 0.7170 √ 4 Firm sub–model (prepaid services) (Removed) 0.25443 10.57 1 0.0012 No 5 Firm heritage (parent company firm) –0.01506 0.03 1 0.8534 √ 6 Firm start–up capital: Total assets (ln) 0.04717 0.29 1 0.5924 √ 7 Firm venture capital (VC) funding –0.05262 0.44 1 0.5050 √ 8 Firm geographic scope 0.02627 0.09 1 0.7619 √ 9 Team gender diversity (male) (Removed) –0.17848 5.81 1 0.0160 No 10 Team size –0.15425 3.26 1 0.0711 √ 11 Team related “Bell” telecom experience 0.06795 0.63 1 0.4289 √ 12 Team functional diversity 0.13444 2.94 1 0.0862 √ 13 Team functional dominance 0.04053 0.30 1 0.5863 √ 14 Team operations & engineering depth –0.08151 1.27 1 0.2596 √ 15 Team sales & marketing depth 0.08157 1.42 1 0.2333 √ 16 Team entrepreneur experience –0.13096 2.90 1 0.0885 √ 17 Team executive experience (broadly defined) 0.01513 0.04 1 0.8484 √ 18 Team executive tenure (total years) 0.07521 0.66 1 0.4148 √ 20 Team executive experience (high rank) 0.00277 0.00 1 0.9727 √ 22 Team “Be...
