Validation of ICD-10 Consensus Code Set for Cirrhosis Detection Using Electronic Health Records in an Asian Population

Abstract

Background

Systematic identification of patients with cirrhosis through electronic healthcare records (EHRs) using ICD-10 codes is essential for epidemiological research but is prone to discrepancies. We aim to validate and improve a recent consensus code set of nine ICD-10 codes to identify cirrhosis in a multi-ethnic Asian population.

Methods

We applied an initial broad algorithm of 25 ICD-10 codes related to cirrhosis and its complications to identify patients potentially with cirrhosis admitted to Singapore General Hospital in 2018 and confirmed true cirrhosis cases via manual EHR review. We evaluated the consensus code set's sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) in identifying cirrhosis cases. We examined alternative code sets to improve cirrhosis identification and validated them in another local hospital.

Results

One thousand, seven hundred thirty-three patients potentially with cirrhosis were identified, with 937 (54.1%) confirmed. The median age at diagnosis was 71 years (IQR: 64–78), with 65.6% males, 75.2%/8.8%/9.3%/6.7% Chinese/Indians/Malays/Others, and 56.7% Child-Pugh A. The main etiologies were chronic hepatitis B (29.5%) and metabolic dysfunction–associated steatotic liver disease (25.5%). The consensus code set demonstrated sensitivity/specificity/PPV/NPV of 76.1%/82.0%/83.3%/74.5%, respectively. We identified a set of 10 ICD-10 codes (SingHealth Chronic Liver Disease Registry [SoLiDaRity]-10) with sensitivity/specificity/PPV/NPV of 76.5%/84.8%/85.6%/75.4%, respectively, demonstrating an improved specificity versus the consensus code set (p = 0.001). External validation in another local hospital with 578 patients potentially with cirrhosis demonstrated improved sensitivity of the SoLiDaRity-10 code set versus the consensus code set (p = 0.033) (sensitivity/specificity/PPV/NPV: 78.0%/93.6%/94.1%/76.4% vs. 76.2%/93.6%/94.0%/75.0%, respectively).

Conclusions

While the consensus code set performs well in identifying patients with cirrhosis in a multi-ethnic Asian population, we propose the improved SoLiDaRity-10 code set.