2.3 Statistical Analysis
Patient-level characteristics for the hip and knee osteoarthritis patient cohorts were compared via analysis of variance (ANOVA) for continuous variables and chi square tests for categorical variables where appropriate. These analyses accounted for patient clustering within HVHC systems through the use of robust standard errors. Multivariate logistic regression examined the association between patient-level characteristics and alignment between patient-expressed post-decision aid treatment choices and treatments received among patients with hip and knee osteoarthritis who were exposed to decision aids across HVHC systems between 2012-2015. The outcome of interest – alignment between post-decision aid treatment choice and treatment received – is a dichotomous variable where alignment = 1 if patients chose surgery and received surgery or chose non-surgical treatment and received non-surgical treatment, and alignment = 0 if patients’ treatment choices did not match received treatments (e.g., chose non-surgical treatment and received surgery). Another multivariate logistic regression was run to explore the association between patient-level characteristics and choice-treatment alignment for hip or knee patients who specifically chose surgery since the majority of both patient cohorts preferred surgery after exposure to decision aids. Patient-level characteristics included in all final models are age, sex, race, marital status, health insurance type, educational attainment, Charlson comorbidity score, HOOS or KOOS pain score and decision-making stage.
All regression models utilized organization-level fixed effects to account for clustering of patients within systems. The output of these models is reported as odds ratios (OR) with corresponding 95% confidence intervals (CI). Any missingness across variables of interest was assumed to be random and these observations were excluded from final analyses. Analyses were conducted using SAS version 9.4.