Models*
n controls=1319
n exposed=1729
ATT95% CIControls matched (unweighted), nExposed dropped, nPairs matched (weighted), nUnique controls, n
Model C*6.31(1.46 to 11.16)8895701659955
Model D*‡7.06(2.57 to 11.56)92721717121001
  • The matching used was many-to-one with replacement. Some exposed patients were not similar enough to any control patients according to the calliper threshold and these individuals were dropped from the analysis (exposed dropped). Some controls were not similar enough to any exposed patients and were thus not used as potential matches and dropped from the analysis. The remaining controls (unique controls) were then ‘copied’ a number of times to be used as potential matches (controls matched unweighted). Each control was not matched individually, but rather weighted to form one matched comparator for each treatment patient. These matched comparator patients were matched to the treatment patients to form matched pairs (pairs of controls and treated cases matched). The number of pairs may thus be higher than the total initial sample size as some controls were used more than once and some were not used at all.

  • *Models C and D omit variables with >25% missing data.

  • †Model A includes linear and quadratic forms of continuous covariates and omits variables with >50% missing data to estimate the propensity score. Variables included in the final propensity score are those listed in bold in the caption to figure 1.

  • ‡Models B and D omit quadratic forms of continuous covariates.

  • ATT, average effect of treatment on the treated.