Identifying Sepsis Cases Using ICD Codes

Sepsis admissions were identified using hospital discharge diagnoses in the NDCMS database, based on the official 10-digit Chinese version of International Classification of Diseases, Tenth Revision (ICD-10) diagnosis codes (an expansion from the 4-digit WHO version [21 ]) and ICD-9 procedure codes [2 (link), 22 (link)–27 (link)]. Explicit sepsis cases were those with an ICD-10 code referencing sepsis explicitly (Additional file 1: Table S1). For example, the category (3-digit) code O08 pertains to the condition “complications following ectopic and molar pregnancy”, its secondary category (4-digit) code O08.2 pertains to the specific condition “embolism following ectopic and molar pregnancy”, while its tertiary category (10-digit) code O08.200 × 006 pertains to the specific condition “embolism following abortion and ectopic and molar pregnancy (septicopyaemic)”. Therefore, O08.200 × 006 was labeled as an explicit sepsis code.
Implicit sepsis cases were identified based on the commonly used algorithm that required at least one acute infection code (Additional file 1: Table S2) and an organ dysfunction code, defined by ICD-10 diagnosis codes (Additional file 1: Table S3) or ICD-9 procedure codes (Additional file 1: Table S4). The implicit-coded algorithm was validated based on a prospective, single-center, cohort study which was designed to assess the diagnostic value of qSOFA for sepsis in general ward and ICU [28 (link)]. Among all implicit sepsis, duplicated admissions, and admissions with unreasonable age (age < 0 day or age > 120 years) or hospital length of stay (LOS, LOS < 0 day or > 365 days) information, missing sex information was excluded. Because 26–60% of children younger than 5 years old in China were not registered to resident identity [29 (link)], we used the UGID as the UID for child ≤ 6 years old with missing UID information. Thereafter, patients with missing UID or fake UID (one UID corresponds to more than 4 UGIDs per year, which means one patient has more than 4 sepsis-related admissions each year) were further excluded (Additional file 1: Fig. S2). The implicit sepsis cases included explicit sepsis codes [3 (link)].