Table 2

Conditional and marginal model estimates for time-to-management failure

ModelsAverage HR (95% CI)*
ConditionalMarginal
Maineffectsmodel
BMI percentile (for every 10 units)1.06 (1.01 to 1.10)1.05 (1.01 to 1.10)
ICS vs treatment non-compliance0.91 (0.69 to 1.20)0.93 (0.70 to 1.23)
LTRA vs treatment non-compliance1.05 (0.68 to 1.61)1.11 (0.72 to 1.70)
LTRA vs ICS1.15 (0.74 to 1.80)1.52 (0.72 to 3.22)
Model withinteractionterm
BMI percentile (for every 10 units)1.09 (1.03 to 1.16)1.08 (1.02 to 1.15)
ICS vs treatment non-compliance1.45 (0.66 to 3.16)1.37 (0.61 to 3.10)
LTRA vs treatment non-compliance7.94 (2.56 to 24.69)4.62 (1.27 to 16.88)
ICS × BMI percentile0.94 (0.85 to 1.03)0.95 (0.86 to 1.04)
LTRA × BMI percentile0.78 (0.67 to 0.91)0.83 (0.70 to 0.99)
LRT pvalue<0.001<0.001
RERI ICS×BMI percentile−0.06 (−0.16 to 0.04)−0.05 (−0.14 to 0.05)
RERI LTRA×BMI percentile−1.29 (−3.55 to 0.97)−0.52 (−1.76 to 0.71)
  • *After accounting for age, sex, ethnicity, income, user type, global assessment of severity score, number of exacerbations in previous year, exposure to smoke, asthma-related comorbidities, triggers, % predicted FEV1, ICS rescue use and season.

  • †LRT assesses improved goodness of fit when comparing nested models (the model with interaction terms vs the main effects model); a p value <0.05 indicates a statistically significant improved fit, that is, explaining a greater proportion of the variance in the outcome, and the likely presence of effect measure modification on the multiplicative scale.

  • ‡RERI=HRTherapy×BMI – HRTherapy – HRBMI +1; a negative RERI can be interpreted as the hazard reduction due to interaction on the additive scale (subadditivity), adjusted for measured confounders.

  • BMI, body mass index; FEV1, forced expiratory volume in 1 s; ICS, inhaled corticosteroids; LRT, likelihood ratio test; LTRA, leukotriene receptor antagonist; RERI, relative excess risk due to interaction.