14 MANOVA for multivariate continuous outcomes

14.1 MANOVA: joint treatment effect across multiple outcomes

MANOVA extends ANOVA to several continuous outcomes analyzed together. Here, it tests whether treatment has an overall effect across SBP change, QoL change, and biomarker change.

man_dat <- dat |> drop_na(sbp_change,qol_change,biomarker_change)

man_fit <- manova(cbind(sbp_change,qol_change,biomarker_change)~treatment_arm,data=man_dat)
summary(man_fit,test="Pillai")
#>                Df Pillai approx F num Df den Df              Pr(>F)    
#> treatment_arm   2  0.514     37.6      6    652 <0.0000000000000002 ***
#> Residuals     327                                                      
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

A significant MANOVA result indicates that treatment groups differ across the outcomes jointly, but it does not identify which outcomes differ. Outcome-specific analyses are therefore used as follow-up tests, with attention to multiple testing.

summary.aov(man_fit)
#>  Response sbp_change :
#>                Df Sum Sq Mean Sq F value              Pr(>F)    
#> treatment_arm   2   4854    2427    38.5 0.00000000000000096 ***
#> Residuals     327  20612      63                                
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  Response qol_change :
#>                Df Sum Sq Mean Sq F value        Pr(>F)    
#> treatment_arm   2   2204    1102    23.7 0.00000000025 ***
#> Residuals     327  15210      47                          
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  Response biomarker_change :
#>                Df Sum Sq Mean Sq F value              Pr(>F)    
#> treatment_arm   2  36245   18122     104 <0.0000000000000002 ***
#> Residuals     327  57214     175                                
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

MANOVA does not replace the need to define primary and secondary outcomes. In confirmatory studies, the outcome hierarchy and multiple-testing strategy should be specified in advance.