Stillbirth is an important public health concern. Despite improvements in prenatal and perinatal care in recent decades, stillbirth occurs in approximately 1 out of every 200 pregnancies and has a tremendous emotional and psychological impact upon families. ^{1, 2}

Although reporting requirements vary, stillbirth is a reportable event in all 50 states and U.S. territories. Data on stillbirths are regularly collected, analyzed and reported by the National Center for Health Statistics (NCHS) through collaborative agreements with states as part of the National Vital Statistics System.³ Based on these data provided to NCHS, in 2005 the prevalence of stillbirths in the United States was 6.22 per 1,000 live births plus stillbirths.⁴ The use of vital records for surveillance purposes, however, has been problematic.^5–8 The American College of Obstetricians and Gynecologists (ACOG) has published recommended guidelines for conducting postmortem stillbirth evaluations.⁹ However, several studies have shown that data on fetal death certificates (FDC) not only yield inaccurate and incomplete information with respect to certain variables such as maternal health conditions, presence of a birth defect, and causes of death, but they also potentially underestimate the true prevalence of this event. ^10–15 Without reliable population-based data, the conduct of epidemiologic studies of risk factors and causes of stillbirth are challenging.

In 2005, the Centers for Disease Control and Prevention (CDC) funded two pilot projects – one in Iowa and one in metropolitan Atlanta, to assess the feasibility of expanding existing population-based birth defects surveillance programs to include surveillance of stillbirths with or without birth defects. The hypothesis was that using the infrastructure of established birth defects surveillance programs employing active case finding methods to collect, analyze and report data on stillbirths could enhance existing surveillance information on stillbirths. These enhancements would need to demonstrate improvements not only in the quantity and quality of information collected, but also completeness of case ascertainment. In 2008, Duke and colleagues evaluated a revised data collection tool for use in the surveillance of stillbirths as part of the Metropolitan Atlanta Congenital Defects Program (MACDP).¹⁰ After linking MACDP-identified stillbirths with FDC, the analysis demonstrated that overall there was less missing information for critical variables, such as birth weight and fetal sex, compared with corresponding information on FDC. Also, the amount and quality of clinical and pathological information abstracted from the medical record through MACDP surveillance was improved and could potentially allow for a better understanding of the contributing factors associated with the fetal death. The current paper reports the results of a follow-up study; the objectives were to evaluate the completeness of case ascertainment and compare prevalence of stillbirths identified through MACDP, FDC, and both data sources combined for the years 2006 and 2008.

In 2006 and 2008 there were 2,252 stillbirths reported in the state with just under half of these occurring among mothers residing in the 5-county metropolitan Atlanta area (Table 1). Because Georgia law requires that all fetal deaths be reported regardless of gestational age if brought to the attention of a healthcare provider, the majority of fetal deaths in Georgia are losses before 20 weeks of gestation (Table 1). The year 2008 had substantially more fetal deaths with a missing gestational age than the year 2006 because of differences in data sources from the state. The data from 2006 from the Office of Health Indicators for Planning had missing clinical estimates of gestational age recoded as the gestational age based on the last menstrual period, if available. The data from 2008 were the raw vital statistics data that did not undergo this assignment process.

MACDP captured 863 stillbirths and FDC captured 862. Of these, 665 stillbirths were independently captured by both sources (Table 2). MACDP captured an additional 198 stillbirths for which no FDC could be found. Similarly, a total of 197 stillbirths were identified based soley on the FDC for case finding (Table 2). These cases would have been missed if FDC had not been available for retrospective case finding and abstraction. Using capture-recapture methods, 58 stillbirths were estimated as missed by both sources [(198 x 197)/665 = 58], resulting in 1118 total stillbirths (95% confidence interval [CI]: 1052, 1183) in metropolitan Atlanta during 2006 and 2008. The estimated sensitivities for capturing a stillbirth were 77.1% (95% CI: 74.7%, 79.7%) for FDC alone, 77.1% (95% CI: 74.7%, 79.7%) for MACDP alone and 94.8% (95% CI: 93.5%, 96.1%) for both sources combined. Of the 197 stillbirths identified solely through FDC, medical records were sought but not found for 30 of them, and 26 additional stillbirths occurred among mothers who resided within the catchment area, but delivered in a facility outside of it. The medical records for these 26 stillbirths were not sought by MACDP.

Furthermore, MACDP captured 61 stillbirths for which induction of labor was performed due to the fetus being affected by a birth defect. Of these, 31 linked to FDC and 30 could not be linked. MACDP ascertained another 49 stillbirths for which induction of labor was performed secondary to a pregnancy complication such as pre-eclampsia, premature rupture of membranes, or chorioamnionitis. Thirty-seven of these were issued a FDC and 12 did not link (Table 3). The 30 cases for which the medical record could not be found and the 26 cases that were delivered outside the catchment area are not included in the assessment of ascertainment by pregnancy outcome reported in Table 3.

Lastly, there were an additional 114 stillbirths identified through FDC with a gestational age of 20 or more weeks that were subsequently excluded after reviewing the medical record. The reasons for excluding these cases are listed in Table 4. Forty-five cases were excluded after review of the medical record clearly indicated that the death occurred before 20 weeks of gestation. Thirty-four cases were singleton stillbirths for which 2 identical FDC were generated. Twenty cases had medical record documentation that the fetus was born alive and expired shortly after birth. Another 13 cases were excluded because the mother did not reside in the surveillance catchment area and 2 cases had the wrong year of birth on the FDC.

There were 55,707 and 54,581 live births and stillbirths (the stillbirths in the denominator were based on the number ascertained by FDC) delivered in the metropolitan Atlanta surveillance area in 2006 and 2008 respectively. Using only those stillbirths identified from FDC, the prevalence of stillbirth was 8.2 per 1,000 live births plus stillbirths in 2006 (95% CI: 7.4, 8.9) and 7.4 per 1,000 live births plus stillbirths in 2008 (95% CI: 6.7, 8.2). Using only ascertainment by MACDP, the estimates were 8.0 and 7.6 per 1,000 live births plus stillbirths (95% CIs: 7.3, 8.7 and 6.9, 8.4), respectively. Using both sources for ascertainment yielded estimates of 9.9 and 9.3 per 1,000 live births plus stillbirths (95% CIs: 8.1, 10.7 and 8.5, 10.1), respectively (Figure 2). Prevalence estimates were compared including and excluding stillbirths occurring after induction of labor as a medical intervention (Figure 3). Excluding stillbirths occurring after induction of labor naturally reduced the prevalence; we observed a greater reduction for MACDP identified stillbirths than FDC identified stillbirths.

The use of FDC alone for population-based surveillance of stillbirths is limited and uncertainty about the utility of FDC data for risk factor analysis has been previously noted. ^{7, 10–12} Complete and reliable surveillance data is needed if hypothesis-driven epidemiologic studies are to be conducted. The current study demonstrated that expanding the capabilities of an existing birth defect surveillance system to include active ascertainment of stillbirths and combining that information with what is gathered from FDC, is feasible and results in the ascertainment of cases that would have otherwise been missed by either system alone. Expanding MACDP to include surveillance of stillbirths required a few modifications to the birth defects surveillance protocol, such as accessing emergency department records (to capture cases arriving as an emergency and potentially getting discharged without hospital admission), and autopsy and placental histopathology information. There was no additional staff required to implement stillbirth surveillance as sources for case finding overlapped with sources already used for birth defects surveillance and were already being reviewed by clinical abstractors.

Each data source – FDC and MACDP – has limitations. The current analysis, as well as previous studies, indicate that FDC underreport stillbirths, as well as often misreport the pregnancy outcome and contain large amounts of missing information for critical variables such as gestational age, birth weight, and the cause of death. ^{10, 18} With respect to cause of death, this may in part be explained by the fact that the majority of FDC are completed before all post-mortem evaluation information is available. ²¹ For MACDP, our analyses suggest that a large number of stillbirths would have been missed if not for the availability of FDC as a source for case finding. Routine precedures for MACDP normally involve obtaining FDC on a monthly basis, from which stillbirths can be identified on an ongoing basis. This was not the case for our study years, allowing for the application of capture-recapture methods to estimate the number of stillbirths occurring in the surveillance population. Active ascertainment by MACDP was only able to collect what was available from the medical record. However, when both MACDP and FDC were used together, they ascertained more stillbirths than either system captured independently. The factors influencing case ascertainment within each data source are not clear. The underascertainment of stillbirths is not likely a random event; it may be associated with factors such as maternal race/ethnicity, gestational age, delivery facility or the cause of death, or perhaps factors that are not even recorded. More analyses need to be undertaken to better understand the role that these factors may play in the ascertainment of stillbirths. This could provide potentially valuable information to inform training needs and strategies to improve the reporting process.

Active case finding of stillbirths has several strengths. Trained abstractors visit area hospitals, locate medical records for potential cases, and record the relevant information. The abstracted information for each potential stillbirth is systematically reviewed by one or more MACDP clinicians to ensure that inclusion criteria are met and to designate the appropriate outcome classification. Previous studies by Duke and colleagues, using data from MACDP, have demonstrated that active ascertainment and medical chart review improves upon the quantity and quality of the data collected. ^{10, 22} In addition, the in-depth medical record review resulted in more accurate classification of pregnancy outcomes, which provided insight into the potential misclassification of pregnancy outcomes by FDC. This information is important to better understand stillbirth prevalence estimates that are based on FDC alone. Active ascertainment can allow for the inclusion or exclusion of stillbirths resulting from the medical induction of labor or stillbirths that were actually live births. As shown in Table 3, about 50% of the inductions performed for a fetus affected by a birth defect linked to a FDC, whereas about 75% of those inductions performed for other pregnancy complications were issued a FDC. It is not possible to know if the cases that did not link were issued an Induced Termination of Pregnancy (ITOP) certificate, data for which are deidentified and unlinkable with information from other sources. Anectdotally the majority of inductions performed in the setting of a fetus affected by a birth defect are done subsequent to the administration of intrauterine potassium chloride, and identifying these events through the review of medical records is relatively straightforward. Therefore, from a surveillance perspective, the intent of the procedure is apparent, and ITOP certificate should have been issued. However many of these birth defects can be considered lethal anomalies and should be considered when understanding fetal mortality rates. On the other hand, inductions performed in the context of other clinical scenarios, such as severe chorioamnionitis, are most often conducted in the best medical interest of the mother and the intent may very well have been to produce a live birth. It is likely that these clinical situations explain the differences in whether a FDC was issued or not. Active case finding allows for these events to be captured and documented based on the thorough review of medical records, potentially improving our understanding of the impact of these events on estimates of the prevalence of stillbirths. These distinctions cannot be made when using FDC data alone.

We capitalized on a lapse of availability of FDC to MACDP for case ascertainment. Having two independent data sources for stillbirth ascertainment allowed us to conduct a capture-recapture analysis to estimate the total number of stillbirths occurring within metropolitan Atlanta – and the number potentially missed by the two data sources working independently. This normally cannot be done when FDC are obtained and used on an ongoing basis as a source for case finding.

This analysis is subject to several limitations, however. First, we were limited to only two years of data; a similar analysis is planned for stillbirths occurring in 2009 and later. Second, we did not assess or compare data quality between sources - an important next step to further demonstrate the utility of this approach to stillbirth surveillance. Third, it was not possible or practical to capture every fetal death. Stillbirths that occurred to mothers residing in metropolitan Atlanta but delivering outside the catchment area were missed by MACDP and were therefore not subjected to medical chart review. However, they could be identified if issued a FDC and were included in the analysis as shown in Table 2. Similarly, medical records for terminations and stillbirths delivered at abortion clinics are not accessible by MACDP. Lastly, it is not clear why such a large number of medical records could not be located (n=30). This may reflect inadequate staffing and resources to conduct an exhaustive search for the medical record as many healthcare facilities store medical records off-site after a certain length of time.

Fetal death reporting by states to the National Vital Statistics System is and will remain the core infrastructure for stillbirth surveillance in the United States; however, expanding existing birth defects surveillance programs to include active ascertainment of stillbirth is potentially a valuable approach to help address our current knowledge gaps about the frequency and risk factors for stillbirths. More importantly, improvements to surveillance data on stillbirths will require multidisciplinary efforts to increase and standardize the use of ACOG recommended clinical guidelines for postmortem stillbirth evaluation.