COVID-19 Death Determination Methods, Minnesota, USA, 2020–2022

Our analysis included 14,004 deaths among Minnesota residents: 13,591 confirmed COVID-19 deaths (13,108 confirmed COVID-19 deaths with death certificate and 483 confirmed COVID-19 deaths without death certificate) and 413 ruled-out COVID-19 deaths. We excluded 59 ruled-out deaths because of language that indicated that COVID-19 did not contribute to death. Confirmed COVID-19 deaths most often occurred in persons who were male (54%) and >80 years of age (53%) (Table 1). Most confirmed COVID-19 decedents were White non-Hispanic (87%), were from the Minneapolis/Saint Paul metropolitan area (52%), and lived in a private residence (51%). About 64% persons with confirmed death had a known COVID-19–associated hospitalization before death, and death occurred during hospitalization for 50%. Only 3% of confirmed COVID-19 decedents underwent an autopsy; 95% had >1 documented underlying condition. Many (49%) of the confirmed COVID-19 deaths occurred before Alpha variant predominance in Minnesota. Median time between symptom onset and death was 14.0 days (interquartile range [IQR] 8.0–25.0 days).

We found that death certificates are accurate and timely sources of COVID-19 mortality surveillance data in Minnesota. However, agreement between death certificates and the Minnesota definition of COVID-19 mortality, as well as timeliness of death certificate filing, varied by certain demographics. In addition, access to provisional death certificates is a consistently expeditious method for obtaining reports of potential COVID-19–associated deaths.

In our analysis, median days between DOD and death certificate filing differed by multiple demographic and disease history variables. However, median time was within 8 days for most groups except those who underwent an autopsy (22 days) and pediatric decedents (40 days), enabling timely access to mortality data for public health insight. In addition, autopsies are rare, and delay is expected when they are performed. Median time from DOD to death certificate filing was longer for younger decedents (19–22).

Decedents with agreement between COVID-19 mortality definition and the death certificate were more likely to have a known hospitalization history or completed autopsy. Hospitalization and autopsy provide valuable information and context for both the health department and the certifier, probably contributing to greater agreement. That finding is consistent with a prior study in Olmsted County, Minnesota, that found higher rates of death certificate accuracy for coronary artery disease and an autopsy rate that was twice the national average (23).

Disagreement between the death certificate and the Minnesota mortality definition was more common among decedents 15,16). The program performs postmortem testing for an assortment of infectious diseases on a wide range of decedents, including medical examiner–investigated deaths that may have resulted from nonnatural causes. UNEX/MED-X identifies many young decedents, as well as many persons with incidental COVID-19 infection or nonnatural alternative causes of death (e.g., drug overdose), who may not have otherwise been identified by routine surveillance systems. Inclusion of COVID-19 as a contributing cause of death on the death certificate for young decedents with a nonnatural alternative cause of death probably explains the higher rate of disagreement among pediatric decedents. In contrast, BIPOC persons may be less likely to have a laboratory confirmed SARS-CoV-2 test result or to access healthcare for COVID-19 illnesses (20,22,24,25) and may therefore be more likely to be underreported in case surveillance, resulting in higher rates of death certificate agreement. BIPOC status has been associated with hospitalization risk (26–31), and non–English-speaking status and Black race have been associated with lower rates of SARS-CoV-2 testing (32,33).

Long-term care decedents were more likely to be considered both a confirmed COVID-19 death without death certificate and a ruled-out death when compared with private residents, resulting in a lower rate of agreement (Appendix Table 2). Those findings may result from multiple factors, including detection of mild or asymptomatic SARS-CoV-2 during facilitywide testing and complicated medical histories. Detection of mild or asymptomatic SARS-CoV-2 may increase the likelihood of COVID-19 being indicated on the death certificate in addition to an alternative cause of death. In addition, underlying health conditions (e.g., chronic obstructive pulmonary disease) are more prevalent in older and institutionalized populations (34), and such conditions can complicate cause-of-death determination. Residents may also be less likely to communicate subjective symptoms associated with COVID-19 infection or may exhibit signs not always associated with infection (e.g., falling more frequently). In addition, previous literature has found that death certificates overestimate death caused by coronary artery disease in certain demographic groups (35), particularly out-of-hospital deaths, which are common among long-term care residents (36). Other reports have found underestimates of Legionnaires’ disease by death certificates and misattribution of death to underlying conditions or other illnesses because of nonspecific manifestations (37). It is reasonable to assume a similar phenomenon may occur among COVID-19 case-patients with complex medical histories, particularly if symptoms could be confused for those of an existing chronic condition (e.g., shortness of breath) or are unobservable (e.g., loss of sense of taste).

Overall, death certificates were generally in high agreement with case-based investigation using the Minnesota definition of COVID-19 mortality. Undercounting (missing confirmed COVID-19 deaths) was observed more often than overcounting (i.e., erroneously identifying a decedent as a COVID-19 death). Those results are consistent with the existing literature (3–7). In our analysis, agreement ranged from 74% among pediatric decedents to 98% among those who underwent an autopsy; agreement was >90% for all groups except pediatric decedents. The small sample size for pediatric deaths complicates efforts to understand differences in characteristics. As more data become available, mortality surveillance systems should adapt as needed to better serve and identify special populations. Adaptions may include additional investigations into potential COVID-19 pediatric deaths (e.g., medical record review) to describe those deaths as accurately as possible.

Death certificates can be combined with case surveillance data (e.g., SARS-CoV-2 testing data) to form a robust COVID-19 mortality surveillance system. Misclassification of COVID-19 deaths is inevitable. Our analysis suggests that misclassification may skew toward underreporting and that some groups are more likely than others to be misclassified. Factors to consider when evaluating misclassification or cases missed by a surveillance system are racial and ethnic disparities in healthcare access, healthcare-seeking behavior, and infection and severe illness risk. For example, pediatric decedents may be overcounted because of intense scrutiny and comprehensive testing, and BIPOC persons may be undercounted because of lack of access to testing. Reliance on death certificates as a primary source of COVID-19–associated death reporting may serve as an accurate surveillance system, especially with limited public health resources. However, when available, public health entities should consider investing resources in more investigations for populations who may be more likely to be misclassified by death certificates, such as pediatric decedents and long-term care residents.

Among the limitations of our analysis, high reliance on the death certificates in the MDH COVID-19 death identification and determination process may skew results toward greater congruence between the death certificate and COVID-19 mortality definition. Although death certificates were an integral part of the Minnesota definition of COVID-19 mortality, we conducted further investigation for specified COVID-19 language, for deaths that occurred >1 year after a positive test result and for deaths that occurred within 30 days of a positive test result and did not have COVID-19 indicated on the death certificate. Therefore, our analysis compares a resource-intensive approach, including review of all available sources, with reliance on death certificate reporting alone. In addition, our analysis identified populations for which an intensive investigation was less likely to agree with the death certificate, which could have implications for future mortality surveillance and analysis.

The Minnesota definition of COVID-19 mortality underwent several revisions during the study period; specifically, a prior definition categorized deaths as probable if the decedent had undergone only antigen testing for SARS-CoV-2. However, probable and confirmed deaths were both still considered COVID-19 deaths and therefore did not affect the overall mortality reporting used for our analysis. In addition, resources to investigate deaths fluctuated throughout the pandemic, so it is possible that not all deaths underwent the same level of investigation. Because intensive investigations were used to rule out and rule in COVID-19 deaths, we expect that the net effect of resource differences over time would minimally affect our findings.

The Minnesota case definition was in place for several years from the beginning of the pandemic, resulting in consistent detailed data collection beyond the initial phase of intensive COVID-19 contact tracing and case investigations. Confirmed Minnesota COVID-19 deaths in our analysis are laboratory-confirmed COVID-19 cases. Documented inequities in access to COVID-19 testing (32,33,38) may have biased mortality and case surveillance. Access to (and use of) at-home tests, which are not reported to MDH or counted as laboratory-confirmed cases, has increased, particularly during the latter stages of the pandemic, and are probably used at different rates by different populations. The population of our analysis constitutes mostly White non-Hispanic persons in an upper Midwestern state with robust public health resources, including the UNEX/MED-X program, and thus, results may not be generalizable to other states. Variables assessed in our analysis may be correlated (e.g., long-term care residency and race/ethnicity), potentially affecting analysis outcomes and generalizability to states with different demographics. Future research may use regression models to explore the relationship between demographic and disease history variables.

Trained public health professionals in Minnesota evaluated death certificates for any mention of COVID-19 rather than searching for specific ICD codes or underlying causes of death, which enabled more timely death reporting and inclusion of language from the entire death certificate. That approach also avoided errors associated with incorrect identification of underlying cause of death on the death certificate, which previous research suggests may be common (39,40). Last, our analysis adds to the body of COVID-19 knowledge by evaluating a COVID-19 mortality surveillance approach for accuracy and timeliness, providing context for future mortality surveillance development and evaluation of emerging diseases in Minnesota and beyond.

In conclusion, we analyzed a large sample of possible COVID-19 deaths that were reviewed by using a standard case definition throughout several years of the COVID-19 pandemic. Comparisons of the Minnesota definition of COVID-19 mortality and inclusion of COVID-19 on the death certificate indicated that death certificates are an efficient and timely source of COVID-19 mortality data when paired with SARS-CoV-2 testing data and should be an integral part of COVID-19 mortality surveillance. Supplemental investigations may be warranted for key groups, such as pediatric decedents, as resources allow. Mortality surveillance is a vital aspect of disease surveillance, particularly during emergence of a new disease, and surveillance evaluations are needed to improve existing systems and prepare for the next public health challenge.