Sample size and power considerations for ordinary least squares interrupted time series analysis : a simulation study