Sas system for windows

Baseline characteristics were described using means ± standard deviations for continuous variables and frequencies for categorical variables. A chi-square test and t-test were used to examine group differences in baseline characteristics by SPPB score. A generalized estimating equation model analysis using the GENMOD procedure in SAS was performed to estimate the odds ratio of having low physical performance (SPPB < 7) over 20-years as a function of cognitive impairment. The models used a logit link binomial distribution and an autoregressive order covariance matrix. All variables including cognitive impairment were analyzed as time varying (with the potential to change as time progressed), except for sex and education. Additional analysis was performed using the general linear mixed models MIXED procedure with an unstructured covariance matrix in SAS to estimate the average population change in SPPB over time as a continuous score as a function of cognitive impairment. These methods were chosen to account for missing or incomplete observations, as well as the modeling of time-dependent change between the variables and the modeling effects of time on the outcome. A p-value of <0.05 was set for statistical significance. SAS System for Windows (version 10; SAS Institute, Cary, NC) was used to perform all analyses.

Data were analyzed using IBM SPSS Statistics for Windows, Version 25 (Armonk, NY: IBM Corp) and SAS software, Version 9.4 of the SAS System for Windows (Cary, NC: SAS Institute Inc.). Data screening procedures indicated that both the tracking and memory data were non-Gaussian in nature. For this reason, generalized estimating equation (GEE) analyses were conducted using an exchangeable correlation matrix and a binary logistic distribution to model the dichotomous outcome of responding correctly or incorrectly to each trial. The main effects of group (CWS vs. CWNS), condition (homogeneous vs. heterogeneous), and the two-way interaction between group and condition were assessed and reported using Wald Chi-Square statistics (with 1 df using the Type III sum of squares approach, two-tailed). Chronological age was also added as a covariate to these analyses as it was moderately to highly correlated with tracking and memory accuracy in both conditions. Planned pairwise comparisons, with Bonferroni corrections applied, were used to further explore differences between means. Finally, correlational analyses between tracking and memory accuracy were conducted using Spearman’s rank partial correlation coefficients, with chronological age added as a covariate.