Surveyreg

The unadjusted relationships between depression treatment categories and baseline characteristics were tested with Chisquare statistics. The unadjusted relationship between depression treatment categories and the HRQoL measures (PCS and MCS) were tested using F tests. The adjusted associations between depression treatment categories and the HRQOL measures (PCS and MCS) were examined in separate ordinary least squares (OLS) regression models. A series of four OLS regression models were fit. Model 1 included depression treatment categories, predisposing, enabling, and need factors; model 2 additionally included the personal health practices; model 3 included all the factors in model 2 plus the external environment characteristics; and the final model 4 included predisposing, enabling, need factors, personal health practices, the external environment characteristics, and the baseline HRQoL measures. The regression models were compared using F statistics and adjusted R². For simplicity, we only present the results from models 3 and 4 in the paper. All tests were two-sided, and the significant levels were set at 0.05. We used the survey procedures (SURVEYMEANS and SURVEYREG) available in SAS 9.4 (SAS Institute Inc. 2010) to account for the complex survey design of the MEPS and to obtain weighted parameter estimates and standard errors [26 (link)].

The primary outcomes were in-hospital mortality and LOS. Generalized linear regression methods for complex survey data were used. Because the KID uses a complex sampling design, sampling weights supplied in the KID were utilized to obtain valid standard errors. More specifically, multivariable logistic regression models with sample weights, stratification and clustering were used to assess the effects of race and other risk factors on in-hospital mortality. For LOS, multivariable linear regression models with sample weights, stratification and clustering were used to assess the association of race and other risk factors on hospital LOS. LOS was analyzed after the logarithmic transformation, log (LOS+1), to correct skewness in the distribution of LOS. For ease of interpretation, we also provide results for LOS in days in its original scale, without logarithmic transformation. SAS version 9.4 procedures SURVEYLOGISTIC and SURVEYREG (SAS Institute, Cary, NC) were used for analysis. Two-sided tests with p < 0.05 were considered statistically significant.