Concordance of Sepsis Diagnosis in Decedents Between Inpatient Claims Data and Death Certificates

Introduction: Death certificates are the primary source for estimating the population-level burden of sepsis mortality, while ICD codes are often used to identify sepsis deaths in hospital claims data. Based on ICD codes, sepsis is present in up to 1 in 2 hospital decedents in the United States. A prior study comparing the number of sepsis decedents in hospital claims data and death certificates showed markedly lower figures among the latter, but the findings are dated and different ICD coding approaches were used for claims and death certificate data. We evaluated the contemporary concordance of similarly coded sepsis diagnosis between population-level hospital claims data and death certificates.


Methods: We used a statewide dataset to identify all in-hospital decedents aged ≥18 years in Texas and the corresponding aggregate inpatient death certificate data from CDC WONDER during 2020-2023. An identical, previously reported ICD-10 code-based taxonomy was used to identify sepsis in claims data and death certificates. Normal approximation was used to estimate the rate difference [RD] and rate ratio [RR] in the frequency of sepsis diagnosis overall and in demographic and COVID-19 diagnosis strata, and for their comparisons across categories within examined strata.


Results: There were 282,958 decedents in claims data and 321,660 in death certificates, with 136,886 (48.4%) and 67,212 (20.9%) reporting sepsis, respectively (RD 27.5% [95% CI 27.2-27.8]; RR 2.32 [95% CI 2.30-2.33]). RD in the frequency of sepsis diagnosis between claims data and death certificates ranged from 23.0% (non-COVID-19) to 41.6% (COVID-19), with corresponding RR 2.04 and 3.49, respectively. RD in the frequency of sepsis diagnosis between claims data and death certificates and their RR varied across the examined strata categories, being highest among decedents aged 18-44 years (vs ≥65 years), males (vs females), Hispanics (vs Whites), and those with COVID-19 (vs non-COVID-19).


Conclusions: Death certificates substantially underestimated the burden of sepsis-associated mortality among hospital decedents, when compared to hospital claims data in Texas, with the use of an identical ICD code-based algorithm. The gaps in the frequency of sepsis diagnosis were greater among younger decedents, males, Hispanics, and those with COVID-19.