Quality of Information Registered on Fetal Deaths Certifi Cates in São Paulo, Southeastern Brazil

OBJECTIVE: To evaluate the quality of information registered on fetal death certifi cates. METHODS: Records were reviewed from 710 fetal deaths registered in the consolidated database of deaths from the State System for Data Analysis and the São Paulo State Secretary of Health, for deaths in São Paulo municipality (Southeastern Brazil) during the fi rst semester of 2008. Completeness was analyzed for variables on fetal death certifi cates issued by hospitals and autopsy service. The death certifi cates from a sub-sample of 212 fetal deaths in hospitals of the National Unifi ed Health System (public) were compared to medical records and to the records from Coroner's Offi ce. RESULTS: Among death certifi cates, 75% were issues by Coroner's Offi ce, with Coroner's greater frequency in public hospitals (78%). Completeness of variables on death certifi cates issued by hospitals was higher among non-public hospitals. There was greater completeness, agreement and sensitivity in death certifi cates issued by hospitals. There was low agreement and high specifi city for variables related to maternal characteristics. Increased reporting of gender, birth weight and gestational age was observed in certifi cates issued by Coroner's Offi ce. Autopsies did not result in improved ascertainment of cause of death, with 65.7% identifi ed as unspecifi ed fetal death as 24.3% as intrauterine hypoxia, while death certifi cates by hospitals reported 18.1% as unspecifi ed and 41.7% as intrauterine hypoxia. CONCLUSIONS: Completeness and the ascertainment of cause of fetal death need to be improved. The high proportion of autopsies did not improve information and ascertainment of cause of death. The quality of information generated by autopsies depends on access to hospital records.


INTRODUCTION
Although fetal deaths account for the majority of perinatal deaths, they are not well understood and studied, 7,8,16,17 and remain almost invisible throughout the world.Systematic and reliable records of these events are essential to understand the problem.Nonetheless, quantifi cation of fetal deaths remains insuffi cient; detailed information is necessary to support interventions to prevent fetal mortality. 8Information from mortality information systems are an important a Fundação Seade.Nascidos vivos, nascidos mortos, óbitos neonatais precoces e perinatais e taxas de mortalidade segundo distritos do Município de São Paulo.São Paulo; 2008 [cited 2011 Jun 12].Available from: http://www.seade.gov.br/produtos/mortinf/tabelas/2008/pdf/tabela06_2008_distritos.pdf tool to identify potential risk factors.Based on mortality data, studies have identifi ed the role of twins, 8 maternal age 5,13,17 and gestational age 5,13 on fetal mortality.
The Euro Perinatal Statistics project of the European Union defi ned specifi c indicators for monitoring perinatal mortality, like birth weight, gestational age, time of death (antenatal/intrapartum), maternal age and parity. 25The indicators are obtained from vital statistics databases.
Causes of death are also important for understanding and preventing fetal deaths.Performance of autopsies can improve data quality.Some countries such as the United Kingdom propose that 100% of fetal deaths should be autopsied, although a study in Whales found an autopsy rate of 60.5% for fetal deaths between 1994 and 2003. 1 Comparison of vital statistics with hospital records is a method to evaluate information quality and should be used in quality improvement. 10,12,20,21ere have been few studies about fetal mortality in Brazil, 3,7,18,22 and this indicator is not included among the basic Brazilian health indicators.Eight Brazilian federative units have perinatal mortality data considered as good quality.Under-notifi cation of fetal deaths and low completeness of information on death certifi cates contribute to this problem. 2,4Nonetheless, perinatal and fetal mortality merits increased attention in national health research, given their growing importance.
In São Paulo, São Paulo state, 99.4% of deliveries occur in hospital wards, mainly in high complexity hospitals. 23Fetal mortality was 7.2 per one thousand births and accounted for 56% of perinatal deaths in 2008.It has been the principal component of perinatal mortality since 1996.a Fetal deaths were 14.6 times more frequent than maternal deaths and constituted 60% of infant deaths in 2008.a Vital statistics in São Paulo have good coverage of events.Data from 2006 a show that the majority of fetal death certifi cates (70%) were autopsied, a similar proportion as in developed countries, which may suggest good quality information.
The study objective was to evaluate the quality of information from fetal death certifi cates in São Paulo municipality, in southeast Brazil.

METHODS
The study analyzed 710 fetal deaths among resident mothers delivering in hospitals of São Paulo municipality, during the fi rst semester of 2008.Data were obtained from the unifi ed database of deaths from the Fundação Sistema Estadual de Análise de Dados (SEADE, State System Foundation of Data Analysis) and the São Paulo State Health Secretary.Deaths that occurred at home were excluded.Deaths from deliveries in hospitals with less than 100/births per year were also excluded due to infrequent fulfi lling of fetal death certifi cates.
Identifi cation of the institution responsible for fulfi lling of the death certifi cate was performed using fi eld 52 of the death certifi cate (Did signing physician attended the death?1-Yes; 2-Substitute; 3-Medical Legal Institute; 4-Coroner's Offi ce [CO] and 5-Others).Responses 1, 2 and 5 were considered as death certifi cates by hospitals.Events identifi ed as Medical Legal Institute and with those not fi lled in were excluded.
The hospital where death occurred was identifi ed on the death certifi cate and subsequently classifi ed as part of the Sistema Único de Saúde (SUS, National Unifi ed Health System) or non-SUS, according to information from the Cadastro Nacional de Estabelecimentos de Saúde (CNES, National Registration of Health Establishments).Death certifi cates without hospital name and code were excluded.
Completeness of data on certifi cates was classifi ed as excellent (above 95%); good (from 90% to 95%); moderate (from 70% to 90%); poor (from 50% to 70%); and very poor (less than 50%). 6The variables analyzed were maternal characteristics (age; education; number of previous children alive and deceased); pregnancy/delivery (gestational age, type of pregnancy and delivery mode) and fetal (sex and birth weight).
In a second step of the study, fetal deaths from a sample of SUS hospitals in São Paulo municipality were analyzed.The 11 SUS hospitals studied (ten high complexity and one medium complexity) 23 are representative of the SUS network.Fetal deaths occurring in four municipal owned hospitals, two state hospitals, three university hospitals (two public and one philanthropic) and two from the Social Organizations of Health were studied.Education (medical residency) take place in the university hospitals and in three other public institutions.
Data completeness (percentage filled in) and the primary cause of death on death certifi cates fulfi lled by hospitals and CO were compared.Causes of death were classifi ed based on the International Statistical Classifi cation of Diseases and Related Health Problems (ICD-10).The cause of death coding was performed by SEADE Foundation.A specifi c protocol was used to extract information from the obstetric medical records at hospitals.A specifi c protocol was applied to obtain data from the records of the CO.
Agreement was calculated for each variable based on data registered in hospital records and death certifi cates.Sensitivity and specifi city were estimated for the information recorded, with the gold standard being hospital records. 15Sensitivity was defi ned as the proportion of information present in medical records and death certifi cates in relation to the total hospital records with information.Specifi city was defi ned as the proportion of information missing in the two data sources, among events that were not included in hospital records.
The project was approved by the Research Ethics Committee of the Faculdade de Saúde Pública da Universidade de São Paulo (Process nº 049/07) on 20 of April of 2007.

RESULTS
Among the 710 deaths, 75.1% of death certifi cates were issued by the CO and 24.9% by hospitals that provided delivery care (Table 1).Death certifi cates were issued by the CO 19% more frequently for deaths in SUS hospitals compared to non-SUS hospitals (p < 0.001; 95%CI 1.06;1.34).
Completeness for fetal sex and birth weight was excellent for fetal death certifi cates issued by CO and were respectively excellent and good in death certifi cates emitted by hospitals.The variables related to maternal characteristics (age, education and previous children) had poor completeness, and hospitals performed better than the CO, especially non-SUS hospitals which had completeness between moderate and good.Completeness was good and excellent for variables related to pregnancy and delivery in non-SUS hospitals and moderate in the SUS network.Death certifi cates issued by CO showed large variability in data completeness: good (gestational age and birth weight), moderate (type of pregnancy) and very poor (delivery mode).Death certifi cates from hospitals, especially non-SUS hospitals, presented greater completeness than those fi lled in by the CO (Table 2).
Fetal deaths in the 11 SUS hospitals represented 39% of the total deaths in SUS hospitals during the study period.Of the sample, 66% of death certifi cates were emitted by CO and 34% by hospitals (Table 3).
For referral of fetal deaths to CO, a police report must be made, authorizing removal of the corpse from the In regards to the fetal characteristics, completeness was high in medical records and death certifi cates, especially in the later (Table 3).
Among death certificates issued by hospitals, all variables with the exception of maternal age present greater completeness in death certifi cates than medical records, even though completeness is unsatisfactory in both sources.The variables related to gestation/birth, type of pregnancy and type of delivery presented higher completeness in medical records.Sex and birth weight had excellent completeness and were slightly better in death certifi cates (Table 3).
The agreement between medical records and death certifi cates issued by the CO was obtained.Data from medical records were considered the gold standard, since they are a primary information source and easily recovered (Table 3).In death certifi cates emitted by CO, there was high agreement for the variables birth weight and sex and low agreement for the other variables.Birth weight and sex presented high sensitivity, and the other variables had low sensitivity.
In death certifi cated fulfi lled by hospitals, there was good agreement between the variables sex, birth weight, type of pregnancy and maternal age.High sensitivity (above 80%) was found for birth weight, sex and type of pregnancy, indicating good recording of these variables in the two data sources.Nonetheless, the high agreement for maternal age was due to the absence of information in the two sources.
Fetal deaths sent to the CO were predominantly preterm (96.4%), with high concentration during the period of 32 to 36 weeks (35.7%).There were differences in the cause of death between the hospital death certifi cates and those from CO (Table 4).Unspecifi ed fetal death was the principal cause of death on death certifi cates issues by CO, and more frequent than those fulfi lled by hospitals (OR=8.SUS and non-SUS hospitals presented differences in completeness for variables based on information routinely collected in hospital care, demonstrating greater care in the transcription of data from hospital records to death certifi cates.
Performance of fetal autopsies is supposed to produce more detailed and higher quality information.Two-thirds of fetal deaths were referred to the CO for autopsy, a level that approximated developed countries. 1,14Nonetheless, recording of data on death certificates from CO was less consistent than in hospitals.
In death certifi cates from CO, the variables obtained either directly or indirectly by autopsy (sex, birth weight and gestational age) presented similar, and sometimes higher, completeness than death certifi cates emitted by hospitals.CO do not have access to hospital records and instead only have access to data registered on the police reports emitted by the police precinct.The use of the Cadaver Referral Guide b was unable to resolve the High completeness for birth weight was observed for death certifi cates emitted by CO, although the measurement was performed at autopsy and did not represent true birth weight.Gestational age, which had good completeness overall, is provided by family members and is written in months in CO records.The calculation for weeks is performed when death certifi cates are fi lled out, which can involve measurement errors in data transcription.
The high proportion of fetal death certifi cates emitted by CO can be related to the combination of four factors: a) performance of fetal autopsies is thought to improve the identifi cation of cause of death by obstetricians; b) the high proportion of antepartum fetal deaths (approximately 90%) 3 combined with the lack of integration between prenatal care and delivery, results in the woman's health records not being sent to the delivery hospital; c) the relative proximity of CO in the capital; d) the reluctance of health professionals to report unfavorable outcomes. 8These factors can incentivize obstetricians to judge antepartum fetal deaths as undefi ned deaths and to refer them to autopsy.The situation was more frequent in SUS hospitals, probably due to the need of family authorization.
Recording of data in hospital records was more complete than information in death certifi cates, except for the variables birth weight, sex and gestational age.For almost all the other variables there was a loss of data during transcription from medical records to death certifi cates, especially those emitted by the CO.This results in a low degree of agreement for the majority of variables.
The type of pregnancy had lower agreement for death certifi cates emitted by SUS hospitals and CO, when compared to a study in Washington in the USA. 15The variable for previous live births had even lower agreement than in the Washington study. 15These results suggest a lower preoccupation in the recording of information on death certifi cates emitted in São Paulo.
The study in the United States compared the values recorded in death certifi cates and medical records, and in the current study, agreement refers to the presence of information and not their values in death certifi cates and medical records.
The presence of birth weight and gestational age had a high sensitivity in the death certifi cates emitted by hospitals, indicating simultaneous recording of both documents.With the exception of age, there was low recording of maternal characteristics in medical records and therefore low completeness of death certifi cates.
Although the objective of performing autopsies is to improve information on cause of death, the majority of death certifi cates emitted by CO attributed the primary cause of death as unspecifi ed.The most frequent cause of death indicated by CO was unspecifi ed fetal death, followed by intra-uterine hypoxia.These positions were reversed in death certifi cates emitted by hospitals.
A similar result was obtained in Massachusetts in the USA, 13 where a comparison of the cause of death on death certifi cates with autopsy results demonstrated that a portion of deaths indicated as asphyxia on death certifi cates, were subsequently considered as unknown following autopsy.Perinatal autopsies performed in a referral hospital in Curitiba 19 found 15.7% of fetal deaths as unspecifi ed.
The variety in the classifi cation of fetal deaths 9 makes the comparison of results complex.The proportion of deaths due to unknown or unexplained causes depends on the classifi cation used by various studies.A study performed in West Midlands 9 in England found 66% of deaths were considered as unexplained fetal deaths using the Wigglesworth classification, 24 and when employing the Classifi cation System according to the Relevant Mortality Code, 9 15.1% were classifi ed as unknown.A study involving the classifi cation system adopted by Australia and New Zealand found that the cause of death reported by hospitals was undetermined in 26.7% of antepartum fetal deaths, although this proportion increased to 38.4% when reviewed by the Perinatal Mortality Committee. 11 death certifi cates emitted by hospitals, besides the aforementioned causes of death, there was a high proportion of deaths due to gestational complications, as well as placenta and umbilical cord complications, which were absent from death certifi cates emitted by CO.This fi nding may be due to the absence of clinical information concerning the pregnancy and delivery, during autopsy.In addition, the placentas do not accompany the fetuses for autopsy which complicates the identifi cation of cause of death.
Some authors indicate that there are problems in the use of ICD-10 for the study of fetal deaths, 8 specifying that the codes referring to placental conditions are incomplete and include few pathologies.There is also overlap between obstetric and perinatal conditions which could contribute to the high proportion of undefi ned or unexplained deaths, demonstrating the need to introduce modifi cations in the next ICD revision. 8nformation about fetal deaths can be improved with the introduction of a specifi c fetal death certifi cate, which could include more detailed information about gestational complications. 14 2005, fetal deaths of 2,500 or more grams were included in investigations that should be performed by the Infant and Fetal Death Surveillance Committees c to improve information and obtain a better understanding for prevention programs.It is possible that compulsory surveillance of infant and fetal deaths by the SUS health services (public and private) d will improve information quality.
The removal of the fetuses form the hospital to the CO occurs after a police report is submitted by the police station responsible for the area where the hospital is located.In the sample of SUS hospitals, the police reports for almost 90% of deaths referred to CO were requested by family members.

Table 1 .
Institutions responsible for issuance of fetal death certificates, by type of hospital at delivery.São Paulo, Southeastern Brazil, 2008.

Table 2 .
Completeness of variables on fetal death certifi cates, by type of hospital and institution responsible for issuance.São Paulo, Southeastern Brazil, 2008.
SUS: Brazilian Unifi ed Health System CO: Coroner's Offi ce

Table 3 .
Completeness, agreement, sensitivity and specifi city of variables in fetal death certifi cates from deliveries in hospitals of the Brazilian Unifi ed Health System, according to the institution responsible for issuance.São Paulo, Southeastern Brazil, 2008.In 88.6% of deaths, the police report was solicited by family members, and in the remaining cases, by other people, including hospital staff made the request.The Municipal Health Secretary introduced Cadaver Referral Guidelines, in order to improve hospital information sent to the CO; nonetheless, the guidelines were present in only 21.4% of records at the CO.The analysis according to source of death certifi cate (hospital or CO) did not show signifi cant differences in completeness of records.Differences were observed when comparing the proportion of variables recorded on medical records and death certifi cates, although there was no consistent trend.Maternal characteristics had lower completeness in death certifi cates issued by CO compared to hospital records.Recording of maternal education was minimal (3.6%).Gestational age was complete in 97.8% of death certifi cates and 66.4% of medical records, and type of pregnancy was complete in 80.4% and 86.4%, respectively.Type of delivery was practically not registered on death certifi cates (1.4%), although it was present in 98.6% of hospital records.

Table 4 .
Primary cause of fetal death listed on death certifi cates for hospitals of the Brazilian Unifi ed Health System, according to the institution responsible for issuance.São Paulo, Southeastern Brazil, 2008., since it was present in CO records for 21.4% of reviewed deaths (SUS hospitals), which explains the low completeness of variables that depend on information unobtainable by fetal autopsy.
b Secretaria Municipal da Saúde de São Paulo; Sistema Único de Saúde.Guia de encaminhamento de cadáver.São Paulo; 2006 [cited 2011Jun 12].Available from: http://ww2.prefeitura.sp.gov.br/arquivos/secretarias/saude/legislacao/0077/Guia_Encaminhamento_Cadaver.pdfproblem Instead of family members receiving counseling and support from health services, they are referred to police stations to obtain the document.Families in São Paulo state that use SUS and experience a fetal loss, appear to receive a different level of care than offered in the United States, United Kingdom, Canada and Australia, where support and counseling activities are provided to families.eInconclusion, it is necessary to improve completeness and reporting for the cause of death in fetal deaths.The high proportion of autopsies did not improve information quality.Improvement of information quality on CO death certifi cates depends on access to hospital information.Good quality vital statistics are a critical fi rst step for the research, monitoring and prevention of fetal mortality.Sensitization of managers and health professionals is fundamental to improve the information available about fetal deaths.It is possible to develop routines to facilitate the adequate fl ow of information between prenatal care services and hospitals and subsequently to CO.In addition, it is necessary to improve support for families that experience a fetal death.