Whenever you're ready, walk me through how you'd approach this.
How would you write a function to detect and treat outliers in a highly skewed clinical dataset for Optum?
Assume the input is a pandas DataFrame and the function must process one specified numeric clinical variable. Explain how you would handle missing values, skewness, clinically implausible values, detection, treatment, and auditability.
Use this interface:
process_outliers(df: pd.DataFrame, column: str, method: str = "winsorize") -> tuple[pd.DataFrame, pd.DataFrame]
Return the processed DataFrame and an audit DataFrame containing the original value, transformed value, outlier flag, lower threshold, and upper threshold for each row.