# Table 2 Mean-based comparisons of HCE between compliers to recommended secondary prevention (column headings) and non-compliers

4-class combination (main analysis) 3- & 4-class combination (sensitivity analysis)
Full population (n = 1840)
Compliers
Median [IQR]
12,504 [11,061; 13,948]
(n = 595)
13,942 [12,720; 15,164]
(n = 1081)
Non-compliers
Median [IQR]
17,764 [16,081; 19,447]
(n = 1245)
19,084 [16,660; 21,508]
(n = 759)
Crude difference, mean [95%CI] −5260 [− 7890; − 2630] − 5142 [− 7640; − 2644]
TwoPM, unweighted
Predicted difference [95%CI]
−2144 [− 4956; 668] − 2737 [− 6081; 606]
TwoPM, IPTW
Predicted difference [95%CI]
−4865 [−8027; −1703] − 3837 [− 8703; 1030]
No prior exposure (n = 542)
Compliers
Median [IQR]
8134 [6380; 9887]
(n = 236)
8360 [7082; 9638]
(n = 397)
Non-compliers
Median [IQR]
11,069 [7319; 14,818]
(n = 306)
13,708 [6037; 21,379]
(n = 145)
Crude difference, mean [95%CI] −2935 [− 7466; − 1596] − 5347 [−10,410; − 285]
TwoPM, unweighted
Predicted difference [95%CI]
−1708 [− 4688; 1273] − 4389 [− 10,158; 1380]
TwoPM, IPTW
Predicted difference [95%CI]
−4048 [− 8727; 632] − 6974 [− 17,959; 4011]
1. HCE amounts represent Swiss francs (CHF).Two-part models were estimated for compliance status to a 4- or 3- and 4- class combination treatment (main variable of interest shown in the table) and adjusted for age, sex, having a high deductible, participating in a managed care model, having at least one supplementary insurance, living in a French-speaking or Italian-speaking canton, degree of urbanity of place of living, having had high medication expenditures of at least CHF 5′000 within 360 days before the index date, having had an inpatient hospital stay within 360 days before the index date (other than the index hospitalization for myocardial infarction), as well as the presence of pharmaceutical cost groups (co-morbidites) as confounders (coefficients not shown)
2. In addition, inverse probability for compliance to a specific drug combination (IPTW) was applied to further adjust for the “healthy adherer bias”. Inverse probability weights were estimated by means of a multivariable logistic regression with compliance with a specific drug (as indicated by the column heading) as outcome variable and the same variables as for the two-part model potential predictors
3. Abbreviations: HCE health care expenditures, TwoPM Two-part model, IPTW Inverse probability of treatment weights, IQR Interquartile Range, 95% CI 95% Confidence Intervals. Estimates with CI not including 0 are in bold