Table 2.
Estimates of difference in healthcare costs between AF patients and controls: comparison of alternative models
| Dependent variable | Model | ||||
| Two-part GLM | Two-part OLS | Propensity score matching | OLS | ||
| Total costs | OR (logistic part) | 1.50 (1.46 to 1.54) | – | – | |
| Marginal effect/cost estimate (GLM/OLS part) | 6374 (5609 to 7139) | 5743 (5210 to 6277) | – | – | |
| Combined two part/direct estimate | 5588 | 5187 | 5692 | 5124 | |
| Outpatient costs | OR (logistic part) | 1.46 (1.42 to 1.50) | – | – | |
| Marginal effect/cost estimate (GLM/OLS part) | 1299 (1097 to 1501) | 1043 (860 to 1226) | – | – | |
| Combined two part/direct estimate | 1425 | 1246 | 1342 | 1124 | |
| Inpatient costs | OR (logistic part) | 1.13 (1.08 to 1.17) | – | – | |
| Marginal effect/cost estimate (GLM/OLS part) | 35 154 (28 827 to 41 481) | 37 322 (32 916 to 41 728) | – | – | |
| Combined two part/direct estimate | 2779 | 2957 | 4350 | 3999 | |
The two-part models used a logistic regression in the first part, and GLM or OLS, respectively, in the second part. Propensity score matching was done 1:1, and OLS refers to a direct (non-two-part) OLS estimate. The brackets show 95% CIs. An exchange rate of EUR1.0=CHF1.1 can be used to convert the costs into Euros to facilitate comparison with other countries.
AF, atrial fibrillation; GLM, generalised linear model; OLS, ordinary least squares regression.