In 2015, Ohio had the 12th highest state infant mortality rate in the nation (US DHHS n.d.). Infant mortality is an important maternal and child health indicator of overall population health (Reidpath and Allotey 2003). Accurate infant cause of death reporting provides crucial data for focused strategy development to reduce infant mortality. In Ohio, two infant cause of death reporting systems exist—death certificates and child fatality review (CFR).

Death certificates are a primary source of infant mortality statistics and are used to rank leading causes of infant death nationally using the 71 rankable causes of infant death (Heron 2016). Since 1908, Ohio death certificates have consistently tracked infant mortality changes for public health purposes. Mortality statistics rely on the underlying cause of death, defined by the World Health Organization (WHO) as, “the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury” (WHO 2011). Death certificates collect basic demographics but not detailed circumstances of the death.

In contrast, CFR was developed to document and understand circumstances related to fatal child abuse deaths (Durfee et al. 2009), and different CFR versions have developed nationally and internationally. Today, CFR scope has broadened to include other injury and medical causes of death, but the purpose remains to identify preventable causes of death. In 2000, Ohio mandated a CFR modeled after the National Center for Fatality Review and Prevention (NCFRP) process (Covington et al. 2005) to more fully understand and prevent child deaths. All deaths for persons aged < 18 years are reviewed, including all infant deaths. Local and state public health officials use CFR findings to develop infant mortality prevention initiatives that address safe sleep, child abuse, and infant home visiting programs (ODH n.d.).

In 2013, Ohio Department of Health (ODH) began publishing an annual infant mortality report to raise awareness of slowing improvements in the state’s infant mortality compared with national trends (ODH n.d.). Differences in leading causes of death between the two classification systems used by ODH at the time—71 rankable causes from death certificates and cause of death categories reported through CFR—were identified while preparing the report. Previous studies have noted limitations of death certificates in accurately reporting cause of death; however, few studies have examined infant deaths specifically and no previous studies have directly compared death certificates with CFR (Percy et al. 1981; Pritt et al. 2005; Seske et al. 2017). We investigated to compare the two systems directly and delineate the utility of each system. The primary objective was to examine the frequency, characteristics, similarities, and differences in cause of death classification between death certificates and CFR to enable selection of the most appropriate cause of death ranking system for Ohio.

During 2016 we conducted a cross-sectional, secondary data analysis. We reviewed all deaths among live-born Ohio residents aged < 1 year during 2009–2013. Fetal deaths and deaths outside Ohio were excluded.

For infants who died during 2009–2013, death certificates totaled 5192, CFR reviewed 5198 deaths, and 5030 records (96.9% death certificates; 96.8% CFR) were linked. More CFR reviews were conducted than death certificates because some infant deaths were reviewed by more than one board. Unlinked records occurred when CFR boards submitted insufficient personal identifiers. Population demographics are described in Table 1. Most infants died during the neonatal period (0– < 28 days) (68%), were boys (56%), were white, non-Hispanic (60%), resided in an urban county (82%), and were born early preterm (< 34 weeks) (59%).

This is the first study to systematically compare infant cause of death between death certificates and CFR. We found notable differences in leading causes of death when each system was ranked separately, and we used linked data to explore the most common areas of discordance. Prematurity was more often selected by CFR than death certificates. Secondly, injury categories had higher concordance than medical categories. Lastly, half of SIDS deaths on death certificates (230 of 456) were categorized as missing/unknown/undetermined by CFR. We explore possible explanations as follows.

Prematurity ranks higher when using CFR rankings than death certificate because death certificates and CFR have different approaches to defining prematurity. For death certificates, ICD-10 coding instructions consider prematurity a mode of death rather than a cause of death and recommend that prematurity not be coded unless it was the only condition known (WHO 2004). National leading causes of infant death are based on 71 rankable causes (Heron 2016), with prematurity defined narrowly as, “disorders related to short gestation and low birth weight, not elsewhere classified (P07).” Other prematurity-related causes (e.g., neonatal hemorrhage, respiratory distress syndrome, or necrotizing enterocolitis) are ranked separately. Using this narrow definition for prematurity likely contributed to the low concordance for prematurity between death certificates and CFR. One solution is to classify prematurity-related causes together, as done by either NCHS (Mathews et al. 2015) or the modified Dollfus classification scheme (Nakamura et al. 2015). The Dollfus prematurity category groups prematurity-related conditions together with short gestation and low birth weight (P07). By using a broader prematurity category, modified Dollfus more inclusively estimates prematurity burden on infant mortality (Nakamura et al. 2015).

In contrast to death certificates, CFR boards were more likely to record prematurity rather than a perinatal condition for infant decedents born prematurely. Perinatal conditions that contributed to or resulted from prematurity are recorded in a different section of the CFR data collection form. However, in Ohio these data are often not recorded, either because boards do not or cannot obtain detailed infant or maternal medical records. Because CFR focuses on preventability, some boards do not review medical deaths as closely as injury deaths. For example, one Ohio county partially reviewed two-thirds of all infant deaths because they were due to medical causes or prematurity. Consequently, the opportunity to collect information on perinatal and maternal conditions through CFR might be underutilized in Ohio. NCFRP has recognized the limitation of reviewing medical and premature deaths. National data quality initiatives, including online webinar training, have been developed to raise awareness that many medical infant deaths can be prevented, to make reviews more effective, and to improve recommendations for prevention activities (NCFRP 2016).

Our second finding of higher concordance for injury deaths than medical deaths might be attributed to injury deaths often occurring from a single event, whereas medical deaths are often multifactorial (Jacob et al. 2015). For example, bacterial meningitis (G00) can be categorized as a neurological condition or an infection. When NCHS selects the underlying cause from the death certificate and assigns an ICD-10 code, it follows a nosological system standardized nationally. In contrast, each CFR board in Ohio has independent methods for selecting cause of death. The high concordance for injury categories reveals that CFR collects reliable cause of death data for injury-related deaths, which are typically considered preventable. Other studies have reported that CFR is more effective than death certificates in identifying certain injury categories, especially child maltreatment (Schnitzer et al. 2008; Crume et al. 2002).

Finally, we revealed that most SIDS deaths on death certificates were not categorized as SIDS by CFR. In CFR, SIDS is subcategorized under medical causes, and most SIDS deaths were reported as undetermined if injury or medical cause. This is understandable as CDC defines SIDS as, “the sudden death of an infant under one year of age that cannot be explained after a thorough case investigation, including a complete autopsy, examination of the death scene, and review of the clinical history” (CDC n.d.). The difference in SIDS reporting between death certificates and CFR might reflect differences in interpretation of SIDS criteria. CFR has an additional section on deaths related to sleeping or the sleep environment, and Ohio avoids this confusion by reporting sleep-related deaths rather than SIDS. Before using this CFR measure, other states might consider evaluating the data quality of the SIDS cause of death variable.

In summary, both systems revealed limitations for tracking infant cause of death, particularly prematurity-related deaths. Death certificate underlying cause codes underestimated the broader contribution of prematurity when ranked using the 71 rankable causes, while CFR missed opportunities to record perinatal and maternal medical conditions. Data completeness on death certificates and the standardized methodology used by NCHS are advantages to using death certificates for cause of death rankings. In response, Ohio now uses death certificate causes of death for annual infant mortality reporting (ODH n.d.). To more inclusively capture prematurity-related causes of death, Ohio groups causes using the modified Dollfus classification.

This analysis had several limitations. Each county conducts an independent review process, so data collection and determination of cause of death might vary by county. Additionally, no validated method of assigning ICD-10 codes to CFR cause of death categories exists. ICD-10 assignments to CFR injury categories are published in Ohio’s annual CFR report (ODH n.d.), and assignments to CFR medical categories were determined by the authors. This analysis was restricted to underlying cause codes on death certificates rather than multiple cause codes because national rankings are based on the underlying cause (Heron 2016). Further-more, kappa statistic does not have universally accepted cut-offs for high or low concordance. Knowing this limitation, we used kappa statistic only to show relative levels of concordance among categories. Despite these limitations, Ohio provided a unique opportunity to conduct this assessment because Ohio CFR reviews all infant deaths, and by linking these datasets, we could quantitatively compare two widely used infant death reporting systems.

Differences in infant causes of death between death certificates and CFR were not attributable to inaccuracies or reporting errors but to an artifact of using different classification schemes. Each reporting system serves distinct, valuable purposes. Strengths of death certificates are in capturing higher degree of detail and in using a classification system consistent across counties, states, and years. Strengths of CFR are in identifying preventable causes of death and in examining circumstances surrounding a death. If CFR findings are used to rank causes of death, it should be stated that the purpose of CFR is to identify preventable causes of death, otherwise rankings could be misleading. Because the two systems complement each other, states could consider using linked death certificate-CFR records to better understand infant mortality.