Assessing the accuracy gap in early postoperative complication surveillance: ICD-10 codes versus manual curation-clinical and economic implications.

Abstract

Assessing surgical outcomes, notably postoperative complications (PCs), is crucial for healthcare systems. However, reliance on International Classification of Diseases, 10th revision (ICD-10) codes, may be suboptimal. This study aims to compare the accuracy of ICD-10 codes against manual curation of electronic healthcare records (EHRs) for identifying 13 individual PCs and evaluate associated resource utilization. EHR data from 11,827 surgical cases across 18 Danish hospitals in November 2021 were analyzed. PCs were identified and extracted through both manual curation and ICD-10 codes. Outcomes such as readmission, admission days, intensive care unit (ICU) stays, reoperations, and radiology procedures were assessed as proxies for resource consumption. Statistical and economic analyses quantified resource utilization and associated costs. In total, 1047 PCs were found through manual curation and 439 PCs were found through ICD-10 codings. Only 218 of the PCs found through ICD-10 codes were retrieved during manual curation-corresponding to a correct ICD-10 coding of 20.8% of PCs. Patients with PCs experienced significantly higher resource utilization, including a 6.6 times higher readmission rate, 6 additional admission days, 2 extra ICU days, 7.7 times more reoperations. PCs incurred substantial economic costs, with additional admission days alone accounting for €25.5 million annually, over four times higher than estimates from ICD-10 codes. ICD-10 codes inadequately capture early PCs highlighting the need for improved detection strategies. The actual costs associated with PCs far exceed current estimates, emphasizing the necessity for enhanced monitoring for informed decision-making. In the Danish healthcare system, ICD-10 codes only capture approximately 21% of PCs, making it inadequate for surgical quality monitoring. The actual costs related to PCs, based on study assumptions, are more than four times higher than estimated from current standard. This calls for novel strategies for PC detection to improve healthcare as well as political and financial decision-making.