Sample size for post‐marketing safety studies based on historical controls