The Reliability of Big "Patient Satisfaction" Data

Big data in healthcare can bring significant clinical and cost benefits. Of equal but often overlooked importance is the role of patient satisfaction data in improving the quality of healthcare service and treatment, where satisfaction is measured through feedback by patients on their meetings with medical specialists and experts. One of the major problems in analyzing patient feedback data is the nonstandard research designs often used for gathering such data: the designs can be uncrossed, unbalanced, and fully nested. Traditional measures of data reliability are more difficult to calculate for such data. Also, patient data can contain significant proportions of missing values that further complicate the calculation of reliability. This paper describes a reliability approach that is robust in the face of nonstandard research designs and missing values for use with large-scale patient survey data. The dataset contains nearly 85,000 patient responses to over 2,000 healthcare practitioners in five different subtypes over a 15-year period in the United Kingdom. Reliability measures are calculated to provide benchmarks involving minimum numbers of patients and practitioners for deeper drill-down analysis. The paper concludes with a demonstration of how regression models generated from big patient feedback data can be assessed in terms of reliability at the total data level as well as drill-down levels.