A Study of Clinical Coding Accuracy in Surgery: Implications for the Use of Administrative Big Data for Outcomes Management.

Background: Clinical coding is the translation of clinical activity into a coded language. Coded data drive hospital reimbursement and are used for audit and research, and benchmarking and outcomes management purposes.

Methods: We undertook a 2-center audit of coding accuracy across surgery. Clinician-auditor multidisciplinary teams reviewed the coding of 30,127 patients and assessed accuracy at primary and secondary diagnosis and procedure levels, morbidity level, complications assignment, and financial variance. Postaudit data of a randomly selected sample of 400 cases were reaudited by an independent team.

Results: At least 1 coding change occurred in 15,402 patients (51%). There were 3911 (13%) and 3620 (12%) changes to primary diagnoses and procedures, respectively. In 5183 (17%) patients, the Health Resource Grouping changed, resulting in income variance of £3,974,544 (+6.2%). The morbidity level changed in 2116 (7%) patients (P < 0.001). The number of assigned complications rose from 2597 (8.6%) to 2979 (9.9%) (P < 0.001). Reaudit resulted in further primary diagnosis and procedure changes in 8.7% and 4.8% of patients, respectively.

Conclusions: The coded data are a key engine for knowledge-driven health care provision. They are used, increasingly at individual surgeon level, to benchmark performance. Surgical clinical coding is prone to subjectivity, variability, and error (SVE). Having a specialty-by-specialty understanding of the nature and clinical significance of informatics variability and adopting strategies to reduce it, are necessary to allow accurate assumptions and informed decisions to be made concerning the scope and clinical applicability of administrative data in surgical outcomes improvement.