Study power
We calculated that, for a two groups survival analysis comparison on a
total time of 3 years and 400 patients per group, we able to detect as
significant an HR of 0.8 or lower with a type 1 error alpha = 0.05 and a
power of 80%.
Statistical analysis
Median and interquartile range (Q1 – Q3) was used to describe age
distribution due to its non-normality. As well, the non-parametric
Wilcoxon test was used to assess the difference of age between the two
groups of patients exposed or non-exposed to montelukast. Pearson’s
Chi-square test for association was used to test the association of
montelukast exposure with the qualitative variables. All the available
variables were, potentially, considered as predictors of a CV event.
However, because some drugs were used by less than 10 patients, those
were not considered in the subsequent analysis. Cox (proportional
hazards) regression analysis was performed pooling IS and MI as target
events. All the CV events occurred among patients taking
anti-hypertensive, anti-platelet, or anti-cholesterol drugs, i.e. among
patients at increased ‘CV risk’. For this reason, those variables could
not be included in a classical Cox regression models, but rather we used
both a propensity score (PS) matching using different calipers (0.4-01)
and a Cox model adjusted for PS. Since our study is a retrospective one,
PS method was used for reducing the effects of confounders. PS has been
calculated using a logistic regression on the treatment variable (i.e.
exposure to montelukast), considering age, gender, residence, education,
income, as well as anti-hypertensive, anti-platelet, diuretic,
hypoglycemic, anti-hypercholesterolemic and anti-androgen drugs as
potential predictors of taking treatment. Concerning qualitative ordinal
variables with more than two levels, analyses were performed both using
all levels and aggregating levels in order to obtain binary variables.
IS or MI were considered as events for survival analysis. For every
patient the observation started January 1st, 2012. All
the patients have been observed up to three years (no drop out) or until
the first event. Event-free Kaplan-Meier survival curves were drawn for
major CV events, and the difference between two curves was tested using
the log-rank statistics. For all the analyses, p-values lower than 0.05
were considered statistically significant. All statistical analyses were
performed using SAS 9.4 (SAS Institute, Cary, NC, USA)