Classifications of perinatal deaths are needed for health care policy, surveillance, international comparisons, clinical services, and research. There is a wide variety of these classifications in the literature, reflecting differences in criteria and available information for recording stillbirths and in existing health information systems over time and between countries 1 2 . Some of them include categories best suited for epidemiology and health care planning purposes, including risk factors such as small for gestational age (SGA) or twin pregnancy, while others aspire to provide information on the cause of death, focusing on specific clinical groups relevant to biomedical research questions 3 .

After a substantial decrease of the stillbirth rate, by two-thirds from 1950 to 1975, related to prevention and treatment of infection and improved obstetric care, this decline has slowed or halted in high-income countries during the last few decades 4 . Authors of the Lancet’s Stillbirths Series in 2011 suggested that classification should be the first research priority in epidemiological measurement, and underline the need for “the optimum investigation protocol for stillbirth to identify causes and relevant conditions in terms of yield, utility and costs” in high-income countries. Most classifications consistently report up to two-thirds of fetal deaths as being unexplained or unknown 1 . Several factors contribute to increasing the number of unexplained or unknown cases, such as the design of the system itself or the lack of postmortem investigation.

The classification called RECODE (RElevant COndition at DEath) is intended to be used in a strictly hierarchical manner and designed to organize information on the clinical conditions associated with the death rather than why the death occurred 5 . This makes it possible to avoid a case-by-case analysis of the circumstances leading to the death and to apply the classification retrospectively to existing databases. Other strengths of this classification are that is has a clear hierarchical structure, is based on ICD codes, and enables 85% of stillbirth cases to be assigned a relevant condition. In 2009, RECODE was ranked third in the International Stillbirth Alliance out of six contemporary systems designed specifically for stillbirths: Amended Aberdeeen, Extended Wigglesworth, PSANZ-PDC, CODAC and Tulip 3 . They concluded that the best classifications collect all relevant information, use a hierarchical approach as a guide, but rely on expert opinions in order to preserve the relative importance of the narrative 6 7 8 .

The RECODE classification grants significant importance to fetal growth restriction (FGR) relative to other clinical conditions. This is concordant with previous analyses of the pathophysiology of conditions underlying stillbirths 2 . This choice is also supported by the potential preventability of stillbirths associated with FGR 9 . However, the placement of FGR in the RECODE classification may override important information on other related conditions. For instance, when autopsy and placental examinations exist they provide information on placental pathology, which is a frequent antecedent of both FGR and stillbirth 10 . These anomalies are also part of a large group of clinical scenarios associated with maternal vascular disease and FGR 11 12 .

The aim of this study was to test how the hierarchical ranking of FGR affected the classification of stillbirths in a large population-based registry in the Rhône-Alpes region from 2000 to 2010. We compared the RECODE classification with an alternative hierarchy, labelled RECODE-R in which FGR was only retained in the absence of other clinical conditions.

During the study period, 1030 stillbirths were recorded, and 61 (5.9%) were excluded due to missing data on gestational age (n = 1), birth weight (n = 42), sex (n = 24), gestational age below 22 weeks (n = 1) or gender ambiguity (n = 5). They were more often preterm fetal deaths (88.3%, p = 0.001) and multiple pregnancies (26.3%, p = 0.001). The final sample contained 969 stillbirths.

Table 1 presents the main characteristics of the sample. Antepartum and intrapartum deaths represented respectively, 81.6 and 15.0% of the cases, and 26.8% of the cohort were full-term stillbirths. Maternal age was below 25 and above 35 years old in 17.8 and 24.7%, respectively. Twelve per cent were twin pregnancies. The rate of SGA stillbirths was 49.0%, and most of them had a birth weight below the 3^rd centile (39.2%).

Table 2 shows the distribution of RECODE and RECODE-R categories and subcategories for all stillbirths and for all SGA stillbirths (the group non-SGA stillbirths only is not displayed in the table). Category A was composed of lethal congenital anomalies (A1), infection (A2), non-immune hydrops (A3), iso-immunization (A4), feto-maternal haemorrhage (A5), twin–twin transfusion (A6) and FGR (A7), and accounted for 58.7% of conditions retained in the total sample with RECODE. Its largest subcategory was A7 FGR (38.2%). The next three main categories were umbilical cord (B), placenta (C) and amniotic fluid (D), accounting for 6.7, 12.3, and 5.2%, respectively. Each of the other categories (uterus E, mother F, intrapartum G, trauma H) did not exceed 1.3%.

† Subcategories with results equal to zero were not mentioned.

The main changes from RECODE to RECODE-R in the overall sample are also represented in Figure 1. According to the frequencies in the category A subcategories, we distinguished lethal congenital anomalies (A1) from “fetus-others” corresponding to A2–A6, and FGR (A7). Inversely, categories E–H were combined. From RECODE to RECODE-R, category A decreased substantially from 58.7% to 34.6%, its largest subcategory being now lethal congenital anomalies (14.7%) just before FGR (14.0%). This change increased the numbers of cases in the umbilical cord, placenta and amniotic fluid categories, which nearly doubled to 12.0, 24.8 and 10.3%, respectively. For the categories assigned to uterus, mother, intrapartum event, and trauma, only a slight increase (+1.2%) was observed.

Considering the hierarchical rule of RECODE, all SGA stillbirths were classified in category A, and FGR was retained in 77.9% (Table 2). The distribution of death conditions was radically different among non-SGA stillbirths: category A accounted for only 19% (n = 94), including 11.9% (n = 59) lethal congenital anomalies, and the main categories B–H were more frequently assigned. According to RECODE, unclassified deaths (n = 135, 13.9% of the whole sample) come exclusively from non-SGA stillbirths, and accounted for nearly one-third of them (27.3%).

Moving FGR down in the RECODE-R hierarchy had no impact on SGA births initially assigned to the subcategories A1–A6 (n = 105, 22.1%) (Table 2). By the design of RECODE-R, only stillbirths affected by growth failure and other diseases were redistributed. These 234 cases accounted for 24.1% of the whole sample and 49.3% of all SGA stillbirths. Only 136 SGA births (28.6%) remained classified as FGR. The new related conditions assigned to SGA stillbirths were placental insufficiency (13.3%), constricting loop or knot (9.0%) and placenta abruptio (7.4%).

Table 3 compares the characteristics of reclassified (n = 234) and non-reclassified (n = 136) stillbirths among the 370 stillbirths initially classified as FGR according to RECODE-R. The changes were independent of gestational age, sex, birth weight ratio, maternal age and time of death. There was a non-significant tendency for full-term babies (p = 0.06), stillbirths to older women (p = 0.16), singletons (p = 0.07) and moderate SGA babies (3^rd–10^th centile) (p = 0.11) to be reclassified.

NS Not Significant (p > 0.05).

We tested how the RECODE stillbirth classification performed in a retrospective analysis of a large population-based database of stillbirths. By moving FGR down in the RECODE hierarchy, so that low birthweight for gestational age was retained only in the absence of other conditions, the proportion of stillbirths assigned to the FGR category decreased from 38.2 to 14.0%. Related conditions of the umbilical cord, placenta and amniotic fluid were highlighted and selected in nearly half of the cases. In particular, with RECODE-R one stillbirth in four is assigned to the category of placental conditions. For SGA babies without congenital malformations or fetal abnormalities, these outcomes seemed to fit the mechanisms of death more closely and illustrate their diversity.

Surveillance of stillbirths in a population is an important epidemiological aim of a registry. There is a need for standardised classifications to improve our understanding of these events and how they evolve. For each death, a number of conditions are often observed that may have contributed to the death and the synthesis and organization of this information presents a challenge. We took the pragmatic point of view adopted by Gardosi et al. and demonstrated the feasibility of a RECODE hierarchical computer-based programme. Froen et al. distinguishes cause of death and associated conditions of death which only “contribute in explaining the circumstances of death in a significant proportion of deaths” 7 . From a clinical point of view, this approach may be frustrating compared to a case-by-case perinatal audit 7 8 19 . But this strategy is less time-consuming, retrospectively usable, suitable in an exhaustive and long-standing data collection, and avoids inconsistent identification of cause of death between investigators, countries or study periods. Its main drawback, however, is that it follows a pre-established hierarchy, regardless of whether another condition was evidently a more significant contributor.

The ICD was developed to allow the systematic coding, analysis, interpretation and comparison of morbidity and mortality, and worldwide estimates of stillbirths rate are often provided by these routinely collected data 20 . Recent classifications developed in high-income countries give priority to exhaustive individual analysis, even though some of them ensure compatibility with ICD 7 . The NICE and RECODE classifications are probably unique in using a strictly hierarchical and computerized method applied to ICD codes 21 22 . This approach is consistent with recent recommendations of the authors of The Lancet’s Stillbirths series, who advocate a consensus “on a limited number of programmatically relevant, comparable causal categories,… that can be linked to complex classification systems and ICD codes” 23 . This linkage may be improved if mapping could be extended to all ICD codes through a multi-disciplinary action in order to insure consensus on subcategory definitions. Indeed, not all our ICD codes were included in the initial West Midlands algorithm, suggesting that the choice of ICD codes for maternal and fetal conditions may differ by setting. Furthermore, better classifications could be developed if some of the limitations inherent to using ICD codes for the classification of stillbirths are modified in the revision of ICD-11 3 22 .

There are a few examples of the RECODE classification system in population-based samples. Our rate of unexplained cases was close to the West Midlands cohort of 2625 stillbirths, the Dutch sample of 485 antepartum singleton stillbirths, or the Italian sample of 154 stillbirths (16.0, 14.2 and 14.3%, respectively) 5 12 19 . Like Gardosi et al., we reported 15% lethal congenital anomalies, but our stillbirths classified as FGR (A7) was slightly lower (38.2% versus 43.0%). In the two other case series, the authors found only 30.3% and 16.9% FGR 12 19 . These differences could be due to population selection and most probably to different definitions of SGA births. In particular we were unable to use customized norms which require data on maternal height and weight. This adjustment strengthens the association between SGA and maternal and fetal complications, and the rate of SGA stillbirths was probably slightly underestimated in our study 24 .

In our alternative RECODE-R hierarchy of classification, we considered SGA as a common modifier of other underlying maternal and fetal conditions, but not as a specific condition in itself, unless SGA was isolated. In the six classification systems for stillbirth analyzed by Flenady et al., RECODE is the only one with FGR classified as a specific condition 3 . Four of them do not mention FGR, and isolated FGR is put with unexplained cases at the bottom of the list 6 7 25 26 27 . The PSANZ (Perinatal Society of Australia and New Zealand) classification ranks FGR 8^th of 11 categories and placental histology defines the subcategories, resulting in a FGR rate of 3.2% in a recent analysis in New South Wales from 2002 to 2004 6 28 .

The impact of RECODE-R concerns SGA stillbirths associated with various conditions except fetal conditions. Nevertheless, the main characteristics of reclassified SGA stillbirths did not differ from those of non-reclassified SGA stillbirths. We only found a tendency for full-term, singletons, moderate SGA stillbirths, and stillbirths to mothers aged 35 years or more to be more often reclassified. Several explanations are plausible. Due to specific fetal anomalies, multiple pregnancies are more likely to stay in one of the group A subcategories. The reason that stillbirths to older mothers presented placental, umbilical or maternal conditions more often, and consequently were reclassified more often, may be related to a higher frequency of maternal complications with advanced maternal age. The mechanisms for full-term stillbirths is less clear especially as late stillbirths are those that are more likely to remain unexplained 29 30 . On the other hand, post-mortem investigations could be performed more often for full-term stillbirths, so that this information is highlighted. The mild severity of growth failure among full-term versus preterm stillbirths had already been described 31 . Finally the fact that severe compared to moderate SGA stillbirths stay preferentially in the FGR category might be a reasonable argument for using RECODE-R. The impact of RECODE-R on the classification of SGA stillbirths according to their characteristics, and the hypothesized mechanisms should be confirmed in larger studies.

Monitoring stillbirth rates and capturing the reality of primary clinical conditions associated with fetal death remains an ambitious challenge. Using a hierarchical system within a classification requires defining priorities among the circumstances of death; this strategy is a complementary approach to the perinatal audit designed to identify cause of death. RECODE underlines the frequency of growth failure among stillbirths and the importance of improving prenatal detection of FGR. In contrast, RECODE-R may be closer to etiological mechanisms leading to death and supports the use of post-mortem investigations. Given that the selection of a classification leads to important differences in the clinical conditions which are underscored; these choices should be made explicit and justified with respect to the objective of the analyses.

SGA: Small for gestational age; FGR: Fetal growth restriction.

The authors declare that they have no competing interests.

AE and JZ have been involved in conception and design, analysis and interpretation of data, and draft of the manuscript. PB has been involved in analysing the data and drafting the manuscript. CC, PSJ and TD revised it for important intellectual content and gave final approval of the version to be published. SC, AF and MBM contributed to acquisition of data. All authors read and approved the final manuscript.

This work was supported by the Institut National de la Santé et de la Recherche Médicale (INSERM), the Institut National de Veille Sanitaire (INVS), and the Conseils Généraux of the districts of Isère, Savoie and Haute-Savoie. We particularly appreciated the experience and the help of Prof. Jason Gardosi for encoding the different RECODE conditions. We are also grateful to all the RHEOP members, especially Claire Ambrico, Magalie Piret and Catherine Tronc, for their contribution in data collection, completeness and quality control.