MAI change t1- t0 | Model 1 | Model 2 |
Total sample n = 136 | b (SE) | β (p) | b (SE) | β (p) |
Constant | −1.87 (1.46) | | 13.23 (10.2) | |
Intervention | −1.18 (1.99) | −0.05 (0.55) | −3.53 (2.0) | − 0.15 (0.08) |
Baseline MAI | – | – | −0.35 (0.06) | − 0.50 (0.00) |
Age | – | – | −0.07 (0.12) | − 0.05 (0.53) |
Female | – | – | 0.26 (2.0) | 0.01 (0.90) |
CCI | – | – | −0.03 (0.42) | 0.01 (0.94) |
Katz ADL | – | – | 0.42 (0.46) | 0.08 (0.36) |
DSS | – | – | 0.12 (0.23) | 0.05 (0.59) |
R2 | 0.00 | | 0.20a | |
| | Model 3 | Model 4 |
Subgroup wMAIt0 ≥ 23 n = 73 | b (SE) | β (p) | b (SE) | β (p) |
Constant | −4.07 (1.87) | | 9.06 (16.25) | |
Intervention | −5.99 (2.82) | −0.24 (0.04) | −6.31 (3.02) | −0.26 (0.04) |
Baseline MAI | – | – | −0.36 (0.13) | −0.38 (0.01) |
Age | – | – | −0.13 (0.18) | −0.09 (0.50) |
Female | – | – | 3.85 (3.20) | 0.15 (0.23) |
CCI | – | – | 0.33 (0.66) | 0.07 (0.61) |
Katz ADL | – | – | 0.94 (0.69) | 0.18 (0.17) |
DSS | – | – | 0.36 (0.36) | 0.13 (0.32) |
R2 | 0.06 | | 0.12a | |
- Notes: b = unstandardized regression coefficient; SE = standard error; β = standardized regression coefficient. wMAI = Weighted MAI Sum score for long-term medication; CCI = Charlson Comorbidity Index; KATZ ADL = Katz Index of Independence in Activities of Daily Living; DSS = Dementia Screening Score; Subgroup wMAIt0 ≥ 23 = subgroup of NHRs with a wMAI score ≥ 23 at baseline; Bolded βs are statistically significant (p < 0.05)
- a adjusted R2
- Model diagnostics (outliers and influential cases, multicollinearity, homoscedasticity, normal distribution of errors) were performed. 1 extreme outlier (part of CG, change in MAI t1- t0 = 53) was excluded. Standardized residuals for this case were 4.39 (model 1), 4.62 (model 2), 4.06 (model 3) and 3.75 (model 4). Including the case risks overestimation of the intervention effect (if included, model 1 shows a correlation (β - 0.11 (0.19)) and β in model 2 is statistically significant; model 3 and 4 show only marginal increases in both strength of correlation and significance). Additionally, inclusion leads to non-normal distribution of errors in model 2 and slightly worse overall fit of the model
