Sas 9

Analysis was performed using SAS 9.3 and STATA 11.2. The analysis plan, prepared before the database was locked, included predetermined cut-offs for outcome and predictor variables with a primary focus on perfect (100%) versus incomplete adherence. These predetermined cutoffs were used to summarize the prevalence of adherence across the different measures.
To identify individual modifiable risk factors and programme characteristics associated with incomplete ART adherence, a multiple mixed effects logistic regression model was used for the adherence measure that correlated best with the HIV RNA measurement. The model contained fixed effects for all factors of interest and a random intercept term effect for the mean adherence at each site. Cut-offs for the adherence measures were selected on the basis of receiver operating characteristic (ROC) analysis with HIV RNA at least 1000 copies/ml as the reference standard. The model was constructed using a hierarchical stepwise procedure, with individual-level factors associated at the 0.10 level added first, followed by the programme characteristics significant at the 0.20 level. The model was simplified using step-wise deletion retaining only significant factors and interactions at 0.05. The estimates were corrected for predictor data missing using multiple imputations (see supplemental Table 1, http://links.lww.com/QAD/A616) [36 ].

Statistical analyses were performed using SAS 9.3 and Stata 14.2. Statistical significance was defined as P-value <0.05. This study was considered to be non-human subjects research due to the de-identified nature of the data. The study was approved by the Human Research Protection Office at Washington University in St Louis. All analysis was performed in compliance with the MCBS data use agreement.

The home safety observation was conducted for 97%, 65% and 77% of participants at baseline, 6-month and 12-month follow-ups, respectively (Wang et al., 2018 (link)). There were no significant differences on baseline socio-demographic characteristics or home safety score by loss to follow-up, with the exception that lost dyads had marginally lower high school diploma/equivalent completion (40.2% vs 54.7%, p=0.055). Therefore, maternal education was included as a covariate. Under Missing at Random assumption (MAR), Maximum Likelihood Estimation (MLE) accounted for missingness in the continuous outcome variables to provide unbiased parameter estimates (Allison, 2012 ). SAS 9.3 and STATA 14.0 were used, with p<0.05 indicating significance and 0.05<=p<0.10 indicating marginal significance (Pritschet, Powell, & Horne, 2016 (link)).

We used logistic regression analysis to compare treatment in a VAMC-year classified in the lowest hospice exposure quin-tile (HEQ) with treatment in the next 4 quintiles, for the odds of experiencing each outcome. We used the same approach to estimate the association of hospice exposure with the odds of veterans’ surviving 180 days after diagnosis. All models were adjusted for all covariates, year of diagnosis, and facility fixed effects.
Our cost analyses regressed total daily costs on VAMC-level exposure, while controlling for the same covariates. We compared generalized linear models, ordinary least squares, and semi-log regression models on model fit using Hosmer and Lemeshow deciles.^{21 (link)} A modified Park test was used to guide our selection of distribution and link functions in the generalized linear models.^{22 (link),23 (link)} We corrected the standard errors for repeated sampling within person. Since costs in the first week post diagnosis were highly heterogeneous, we analyzed costs in days 8 through 180, and in these analyses, ordinary least squares provided the best fitting model. Statistical analyses were performed using SAS 9.3 and Stata statistical software.

We assessed patients’ demographic characteristics and described trends in the use of the branded or biosimilar infliximab, adalimumab and etanercept. A segmented linear regression model was used to examine utilization patterns of infliximab (the branded and biosimilar) and other TNF inhibitors (adalimumab and etanercept) before and after the introduction of the biosimilar infliximab. This model included the number of claims each month as the outcome variable, intercept and two slope terms that described the trend in use of TNF inhibitors per month before and after introduction of the biosimilar infliximab in November 2012. We did not include a term for a step change on introduction of the biosimilar infliximab, as we did not expect an immediate impact of the product. Newey-West standard errors were used to allow for autocorrelation up to the second order.(8 ) All analyses were performed using SAS 9.3 and STATA 13.
The study protocol was approved by the Institutional Review Boards of the Brigham and Women’s Hospital and Seoul National University Hospital. Patient informed consent was not required, as the dataset was de-identified.