Despite improvements to antepartum care, stillbirth, commonly defined as fetal death at 20 weeks of gestation or greater, remains an important problem in obstetrics. In the United States, stillbirth affects approximately 1 in 200 pregnancies and is emotionally devastating for patients, families, and clinicians.1
Determining the cause of stillbirth is beneficial for bereavement, counseling families regarding the recurrence risk, and, in some cases, improving outcome in subsequent pregnancies. In addition, it will ultimately facilitate research aimed toward prevention of this devastating complication. Unfortunately, determination of cause remains difficult and unexplained stillbirth persists as the largest single contributor to perinatal mortality among nonanomalous fetuses in the Western world over the past 3 decades.2 Elucidating the cause of stillbirth is complicated by a number of factors including incomplete postmortem evaluation, multifactorial etiologies, a lack of a universally accepted classification scheme, and inaccurate reporting. Numerous classification systems have been developed for the cataloging of stillbirths, each with its own merits and drawbacks. There is no agreement on the ideal system.3,4 The Wigglesworth classification system was introduced in 1980. It includes nine categories and for several decades was one of the most commonly used classification systems. This system catalogs neonatal death as well as stillbirths.5 Gardosi and colleagues introduced the Relevant Condition at Death (ReCoDe) classification system in 2005.6 This system excludes neonatal deaths and is a hierarchical system starting with conditions directly affecting the fetus, moving outward in anatomic groups.
Most population-based studies examining conditions associated with or contributing to stillbirth use vital statistics records, in the form of fetal death certificates, as a primary source of data. Estimating the accuracy of these records is therefore crucial in performing and evaluating studies aimed toward investigating stillbirth etiology and contributing factors. However, vital records are not always accurate. Indeed, even adult death certificates are prone to numerous inaccuracies. Adult death certificates are not always properly completed and discrepancies often exist between data obtained from death certificates and those derived from autopsies and medical records.7,8 We hypothesized that similar problems exist in fetal death certificates. Our primary aim for this study was to compare and validate stillbirth data as recorded in vital statistics with those recorded on or deduced from medical records. Secondarily, we aimed to investigate differences in diagnostic evaluation between tertiary care and community hospitals.
MATERIALS AND METHODS
In this retrospective cross-sectional study, fetal death certificate data from the Utah Department of Health were used to identify individuals with stillbirths delivering in eight Salt Lake City hospitals from 1998 to 2002. All cases for which a fetal death certificate was on file and a medical record available for review were included. The study was approved by the Institutional Review Board at the University of Utah Health Sciences Center.
State vital statistics records based on fetal death certificates were evaluated for stated cause of death, maternal demographics, and obstetric details. Inpatient medical records were abstracted by a single physician investigator (J.H.) using a standardized data collection form. Data were recorded about demographics, medical and obstetric history, features of the pregnancy, diagnostic tests performed in an attempt to identify a cause of stillbirth, and reported etiologies of stillbirth. Study investigators then came to a consensus regarding the cause as determined from medical records.
Data were compared between death certificates and the medical record by calculation of κ coefficient9 for categorical variables or Kendall's τ-b coefficients10 based on the number of concordant and discordant pairs of observations for continuous variables. We chose to interpret the coefficient values as proposed by Landis and Koch in the following way: 0 showing no agreement, 0–0.2 showing slight agreement, 0.21–0.4 showing fair agreement, 0.41–0.6 showing moderate agreement, 0.61–0.8 showing substantial agreement, and 0.81–1 showing almost complete agreement.9 We also calculated a “percent concordance” for each category. This value was obtained by dividing the number of concordant pairs by the total number of concordant and discordant pairs. The proportion of cases undergoing specific tests as part of an evaluation for potential causes of stillbirth were compared between community and tertiary care hospitals with χ2 or Fisher exact test as appropriate.
A total of 556 stillbirths were identified from vital statistics during the study interval. Of these, 461 (83%) had charts available for review. The stillbirths occurred in eight hospitals as shown in Table 1. Fifty-six percent were in teaching hospitals, which were also tertiary care centers, and 75% were in community-based hospitals as compared with 25% at a university-based hospital.
Table 2 compares characteristics of study population derived from fetal death certificates and medical records. Correlation was reasonably good, but some differences existed for maternal race or ethnicity, gestational age, and birth weight. The gestational age was equal in 58% of cases, but differed by 1–7 days in 40%, by 8–14 days in 1.76%, and by more than 15 days in 0.66%. Birth weight was identical in both sources in 79% of cases and differed by 0–100 g in 89% of cases, by 101–200 g in 1%, by 201–300g in 0.5%, and by more than 300 g in 9.5%. There was a mean difference of 146.17 g for birth weight between death certificates and the medical record.
Rates of testing as compared between tertiary care centers and community hospitals are shown in Table 3. An alarming proportion of stillbirth cases did not undergo recommended tests as part of a “workup” for possible etiologies of stillbirth. The American College of Obstetricians and Gynecologists recommends autopsy if the parents consent, karyotype, and placental evaluation in all cases. Other tests may include complete blood count, testing for maternal-fetal hemorrhage, blood type and screen, testing for antiphospholipid antibodies or thrombophilias, toxicology screening, testing for infections, and testing for endocrine disorders.1 Autopsy was performed in only 25% of cases overall and in only 13% and 34% of cases in community and tertiary care hospitals, respectively (P<.01). Tertiary care hospitals performed autopsy; karyotype; and toxoplasmosis, other viruses, rubella, cytomegalovirus, herpes simplex viruses titers; Kleihauer-Betke stain; complete blood count; maternal toxicology screen; and evaluation for thrombophilias more often than community hospitals.
The cause of death based on vital statistics and medical record abstraction are shown in the Wigglesworth classification system in Table 4. There are important differences notable in the cause of death as assigned on the fetal death certificate and as deduced from the medical record abstraction. Using the medical record as the referent, the “cause of death” was correct on the stillbirth certificate only 54.9% of the time when using this classification system. Agreement between the two sources varied by category. The κ value of 0.4 in the “congenital defect” category supports only fair-to-moderate agreement, but the other categories show even less correlation. Notably, the percentage of unknown or unexplained antepartum death substantially decreased after medical record abstraction from 74% to 45%. More congenital defects were noted after review of the medical records.
Table 5 depicts cause of death from both sources as fitted into the ReCoDe classification system. As with the Wigglesworth classification, there are important differences in the cause of death as assigned on the fetal death certificate and as deduced from evaluation of the medical records. Using the medical record as the referent, the “cause of death” was correct on the stillbirth certificate only 52.2% of the time when using this classification system. Similarly, there was a substantial decrease in “unclassified” stillbirths when chart abstraction was used to assign cause of death. Anomalies, umbilical cord knots, twin–twin transfusion syndrome, and abruption were more frequently identified as causes of death after chart abstraction. In cases in which information in the medical record led to a change in the suspected cause of death, 75% involved changes from “unclassified” to a specific cause of death. Fifteen percent of cases involved changing from an initial specific diagnosis to an “unclassified” cause of death. The most common death certificate diagnoses that were changed to “unclassified” on review of the medical record included cord knots or constrictions, vasa previa, intrapartum asphyxia, and abruption. The remaining 10% involved changing from one specific diagnosis to another. Many of these were changes that would alter patient counseling regarding recurrence risk or management of future pregnancies.
When community and tertiary care centers were evaluated separately, the rate of death certificate and medical record concordance was 57% and 53% (P=.25), respectively, for the Wigglesworth classification and 54% and 49% (P=.39), respectively, for ReCoDe. Using the ReCoDe classification, 51% of stillbirths remained unexplained after review of the medical record in community hospitals, as compared with 42% in tertiary care hospitals (P=.37).
These data confirm our hypothesis that important differences exist between fetal death certificates and medical records with regard to demographic variables, contributing factors, and possible etiologies of stillbirth. Correlation was “almost perfect” for fields such as maternal age, birth weight, and gestational age. However, that even objective variables do not show 100% correlation illustrates that even they are subject to incorrect reporting. In both classification schemes, correlation coefficients were all less than 0.5 for recorded cause of stillbirth. Furthermore, in both classification systems, the number of unexplained stillbirths decreased markedly after a complete review of the medical record was undertaken. This illustrates that the proportion of stillbirths for which a cause can be assigned increases substantially with complete medical record abstraction. Our findings also highlight the differences in stillbirth evaluation between tertiary care and community hospitals, with tertiary care hospitals completing a more thorough evaluation. Albeit statistically not significant, this translated into a lower incidence of “unknown” cause in tertiary care hospitals, emphasizing the benefits of completing such an evaluation.
A number of studies have noted discrepancies between adult death certificates and findings on autopsy. However, only a handful have examined these discrepancies in perinatal deaths and stillbirths. Headley and colleagues recently published a review of 86 stillbirths assessing the incorporation of standard stillbirth evaluation guidelines on cause of death classification in Australia.11 Similar to our study, they found that many components of the evaluation were completed in only 53% of cases and that a complete evaluation frequently changed the cause of death from that originally documented on the death certificate. The number of stillbirths that were “unexplained” decreased after complete investigation and medical record abstraction. In a review of 125 stillbirths, Duke et al noted that actively ascertained data from a population-based birth defect surveillance system yielded fewer unexplained stillbirths than when death certificates alone were examined.12 Another study examined the agreement between death certificates and results of autopsy investigations in Western Australia and found that for identifying stillbirths as explained or unexplained, the correlation was “fair” with a κ score of 0.23.13
A comparison of fetal death certificates with data derived from the Wisconsin Stillbirth Service Project noted significant differences in stillbirth cause. The authors speculated that this may have resulted from death certificates being filed by nonmedical personnel.14 The death certificates in our study were signed exclusively by physicians. Such practice might be anticipated to increase the accuracy of the death certificate. Nonetheless, we also had major discrepancies between death certificates and the actual data in medical records.
One explanation is that although death certificates in Utah may have ultimately been signed by physicians, they were often filled out by less experienced clinicians. Perhaps even more importantly, the death certificate is filled out at the time of delivery, before results of most testing for possible causes of stillbirth is available. In many cases, autopsy results are unavailable for up to 6 weeks after delivery.
The study had several weaknesses including the retrospective nature and relatively small sample size. Furthermore, 17% of stillbirths located in vital statistics were unable to be found in medical records. These records could not be located by the medical records personnel at their hospitals. It is unknown whether this was due to lost or misplaced records, incorrect information regarding vital statistics, or both. It was likely some combination of these factors. This is a possible source of bias although systematic bias is unlikely. However, the study also has several strengths. First, we had a population from a single geographic region that included tertiary care and community hospitals. Second, medical records were meticulously reviewed for information pertinent to potential causes of death by a physician investigator. Finally, we focused on the cause of death and contributing factors using popular classification schemes.
The implications of these findings are relevant for both clinicians and investigators. In cases in which patients have suffered a previous stillbirth, the clinician should make every effort to obtain medical records because elucidating the cause of the initial loss may change management in the subsequent pregnancy. Furthermore, clinicians caring for patients suffering a fetal loss should discuss with patients that a thorough workup, including autopsy, gives important information that may have future reproductive implications. Also, it may facilitate counseling and bereavement. Ideally, clinicians should review results from postmortem evaluation and other workup with families at the postpartum examination.
Vital statistics provide a unique and generally population-based resource for answering clinical questions, although bias can still exist. Also, the large number of individuals in most data sets provides adequate power to study rare events. Several measures can be undertaken to improve the accuracy of fetal death certificates. First, forms that capture more appropriate data can be developed and implemented. Second, more experienced clinicians could fill out the forms as accurately as possible. Finally, a mechanism should be developed to allow for the inclusion of test results that are available only at a later date such as perinatal autopsy in vital statistics records of fetal deaths.
In conclusion, these data underscore the need for modifications regarding the type, quality, and timing of data collected in fetal death certificates. Future research into the causes of stillbirth should attempt to use complete medical record data whenever possible. Complete evaluation, review of the medical record, and efforts to increase accurate reporting may decrease the number of unexplained stillbirths.
1. Management of stillbirth. ACOG Practice Bulletin No. 102. American College of Obstetricians and Gynecologists. Obstet Gynecol 2009;113:748–61.
2. Frøen JF, Arnestad M, Frey K, Vege A, Saugstad OD, Stray-Pedersen B. Risk Factors for sudden intrauterine unexplained death: epidemiologic characteristics of singleton cases in Oslo, Norway, 1986–1995. Am J Obstet Gynecol 2001;184:694–702.
3. Reddy UM, Goldenberg R, Silver R, Smith GC, Pauli RM, Wapner RJ, et al. Stillbirth classification–developing an international consensus for research: executive summary of a National Institute of Child Health and Human Development workshop. Obstet Gynecol 2009;114:901–14.
4. Silver RM, Varner MW, Reddy U, Goldenberg R, Pinar H, Conway D, et al. Work-up of stillbirth: a review of the evidence. Am J Obstet Gynecol 2007;196:433–44.
5. Wigglesworth JS. Classification of perinatal deaths. Soz Praventivmed 1994;39:11–4.
6. Gardosi J, Kady SM, McGeown P, Francis A, Tonks A. Classification of stillbirth by relevant condition at death (ReCoDe): population based cohort study. BMJ 2005;331:1113–7.
7. Moriyama IM. Problems in measurement of accuracy of cause-of-death statistics. Am J Public Health 1989;79:1349–50.
8. Sington JD, Cottrell BJ. Analysis of the sensitivity of death certificates in 440 hospital deaths: a comparison with necropsy findings. J Clin Pathol 2002;55:499–502.
9. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33:159–74.
10. Kendall M. A new measure of rank correlation. Biometrika 1938;30:81–9.
11. Headley E, Gordon A, Jeffery H. Reclassification of unexplained stillbirths using clinical practice guidelines. Aust NZ J Obstet Gynaecol 2009;49:285–9.
12. Duke W, Williams L, Correa A. Using active birth defects surveillance programs to supplement data on fetal death reports: Improving surveillance data on stillbirths. Birth Defects Res A Clin Mol Teratol 2008;82:799–804.
13. Measey MA, Charles A, d'Espaignet ET, Harrison C, Deklerk N, Douglass C. Aetiology of stillbirth: unexplored is not unexplained. Aust NZ J Public Health 2007;31(5):444–9.
14. Greb AE, Pauli RM, Kirby RS. Accuracy of fetal death reports: comparison with data from an independent stillbirth assessment program. Am J Public Health 1987;77:1202–6.