Multinomial model | Transient | Slowly resolving | Permanent |
|---|
Inibitor titre § | 1 | 3.63 | 62.0 |
SE | - | 1.78 | 34.06 |
Pseudo R-Square | 0.24 | | |
Chi2 | 94.26 | | |
P | <0.001 | | |
Inibitor titre § | 1 | 1.76 | 32.83 |
SE | - | 1.03 | 19.11 |
Inversions # | 1 | 1.57 | 1.85 |
SE | - | 0.94 | 1.40 |
Nullmutations # | 1 | 0.60 | 1.43 |
SE | - | 0.41 | 1.15 |
Pseudo R-Square | 0.24 | | |
Chi2 | 66.65 | | |
P | <0.001 | | |
- Model A – Inhibitor titre (180 cases).
- Model B Inhibitor titre and gene mutation (131 cases).
- §Regression coefficient for high titre versus low titre.
- #Regression coefficient versus not null mutations.
- We fitted two separate multinomial logistic regression models: one, in the entire population, assessing the probability of developing a transient, slowly resolving or permanent inhibitor associated with the initial inhibitor titre (model A), and another, in those with the mutation available, assessing the risk associated with both the initial inhibitor titre and the gene mutation (model B), even if the model with the gene mutation alone was not significant (probably due to insufficient sample size). We assumed as baseline risk the patient with low titre inhibitor in model A and the patient with low titre inhibitor and not null mutation in model B.
- In model A, the relative risk ratio for a patient with a high responding over one with low responding inhibitor to have a slowly resolving inhibitor is 3.63 (95% CI 1.39 – 9.47) and to have a permanent inhibitor is 62 (95% CI 21.12 – 182.00). The relative risk ratio for a permanent over a slow resolving was 17.1 (95% CI 9.02 – 25.11). The predicted probabilities estimated by the model and used to calculate the relative risk ratio are reported in table.
- In model B, after adding the gene mutation, the relative risk ratio for a patient with a high responding over one with low responding inhibitor to have a permanent inhibitor over a transient is 32.8 (95% CI 21.5 – 44.1) and over a slowly resolving is 18.6 (95% CI 10.1 – 27.1), with no significant effect associated to the gene mutation. The predicted probabilities estimated by the model and used to calculate the relative risk ratio are reported in table. The full parametric model with the interactions between inhibitor titre and gene mutation (Log likelihood −102.122, LR chi2(10) = 76.87, Prob > chi2 = 0.0000, Pseudo R2 = 0.2734) showed that the effect of the titre was not statistically significant when considered alone; on the contrary, in patients with inversions compared to those with not null mutations a statistically significant and clinically relevant increase in the risk for permanent inhibitors in patients with high titre inhibitors (21.7, 95% CI 1.80 – 263) was found, together with a statistically significant but not clinically relevant increase in patients with low titre inhibitors (0.03,95% CI 0.001–0.73).
