Table 6.
Random parameters logit regression results: disease severity subgroups (N=323)
| Attribute | Preference weight | 95% confidence interval | ||
|---|---|---|---|---|
| Severe disease subgroup | ||||
| Skin sensor | 0.953** | 0.697 | 1.210 | |
| Control of the speed of injection | 0.529** | 0.324 | 0.734 | |
| Step-by-step, on-screen instructions in real time | 0.331** | 0.122 | 0.540 | |
| Injection date reminder | 0.474** | 0.265 | 0.684 | |
| Electronic log | 0.512** | 0.311 | 0.714 | |
| Large grip size | 0.145 | −0.220 | 0.510 | |
| Out-of-pocket cost | £0 | 5.098** | 4.188 | 6.008 |
| £15 | 4.236** | 3.480 | 4.992 | |
| £30 | 3.375** | 2.773 | 3.977 | |
| £50 | 2.226** | 1.829 | 2.623 | |
| £90 | −0.072** | −0.085 | −0.059 | |
| £125 | −2.082** | −2.454 | −1.711 | |
| £150 | −3.518** | −4.146 | −2.891 | |
| £250 | −9.263** | −10.916 | −7.610 | |
| Constant for standard disposable self-injection device | 1.474** | 0.466 | 2.481 | |
| Moderate disease subgroup | ||||
| Skin sensor | 0.566** | 0.341 | 0.790 | |
| Control of the speed of injection | 0.422** | 0.249 | 0.594 | |
| Step-by-step, on-screen instructions in real time | 0.299** | 0.126 | 0.472 | |
| Injection date reminder | 0.610** | 0.411 | 0.810 | |
| Electronic log | 0.491** | 0.313 | 0.669 | |
| Large grip size | −0.458** | −0.787 | −0.128 | |
| Out-of-pocket cost | £0 | 4.286** | 3.488 | 5.084 |
| £15 | 3.562** | 2.899 | 4.225 | |
| £30 | 2.837** | 2.309 | 3.366 | |
| £50 | 1.871** | 1.523 | 2.220 | |
| £90 | −0.060** | −0.072 | −0.049 | |
| £125 | −1.751** | −2.077 | −1.425 | |
| £150 | −2.958** | −3.509 | −2.407 | |
| £250 | −7.788** | −9.238 | −6.338 | |
| Constant for standard disposable self-injection device | −0.821* | −1.574 | −0.067 | |
Notes: All variables, with the exception of out-of-pocket cost, were dichotomous and effects-coded. Cost was included in the model as the interaction of continuous cost and the natural log of income. The preference weights in the table above have been created using the estimated coefficient for cost. A dummy-coded variable for moderate disease was interacted with the main variables to model the subgroups. The parameters were modeled with normally distributed random parameters, assumed to be the same across the disease severity subgroups. Additionally, we included an error component to better accommodate the opt-out bias that could be caused by the standard device alternative. **P≤0.01; * P≤0.05.