Measuring re-identification risk using a synthetic estimator to enable data sharing