External adjustment for unmeasured confounders improved drug-outcome association estimates based on health care utilization data