Sudden infant death syndrome: a re-examination of temporal trends

Although Sudden Infant Death Syndrome (SIDS) is a leading cause of post-neonatal death in industrialized countries, its etiology is largely unknown [1]. While the reduction in prone sleeping following the back-to-sleep campaign has led to a decline in SIDS in many countries [2–6], there are several puzzling aspects related to this intervention and the epidemiology of SIDS. For instance, the onset of the decline in SIDS preceded the initiation of the back-to-sleep campaign [2–7]. The reduction in SIDS in the United States began in 1989, while the back-to-sleep campaign was initiated in 1994 [6]. Similarly, SIDS rates in the United Kingdom decreased continuously from 1988 onwards, while the back-to-sleep campaign only began in 1991 [7].

Other unexplained epidemiologic features of the temporal reduction in SIDS include the relatively greater reduction in SIDS among term infants, as compared with infants born at preterm gestation. Data from Avon county in England show that term live births among SIDS cases decreased from 88% in 1984–88 to 63% in 1994–98, the period when SIDS rates declined most rapidly. The proportion of term infants among SIDS cases remained stable thereafter (66% in 1999–2003) and SIDS rates did not change dramatically during this period [7]. Also, a larger decline in SIDS was observed among twins as compared with singletons. In England, SIDS among twin live births declined by 71% from 1.4 per 1000 live births in 1993 to 0.4 per 1000 live births in 2003, whereas among singletons, SIDS rates decreased by 50% from 0.6 to 0.3 per 1000 live births during the same period [8, 9]. It is notable that births at term and post-term gestation and twin births (subpopulations which experienced relatively larger reductions in SIDS) also experienced the largest increases in early delivery (i.e., increased obstetric intervention through labour induction and cesarean delivery).

Perhaps the most intriguing finding related to SIDS is the paradoxical association between plurality and birth weight-specific SIDS rates [8, 9]. SIDS rates are higher among twins as compared with singletons among normal birth weight infants (>3,000 g), whereas at lower birth weights the opposite is true. This phenomenon, sometimes referred to as the paradox of intersecting mortality curves, has also been observed when birth weight- and gestational age-specific stillbirth or infant mortality rates are contrasted across plurality, parity, race and other factors [10].

Various explanations [11] have been proposed to explain the paradox of intersecting mortality curves including the fetuses-at-risk approach [12, 13]. This model assumes an intrauterine etiology for the outcomes of interest and gestational age-specific mortality rates calculated using the fetuses-at-risk approach do not exhibit the crossover paradox [14]. Since numerous studies have shown that unexplained antepartum fetal death and SIDS have common features (including similar pathologic characteristics at autopsy [15–17], common risk factors [15, 18, 19] and congruent temporal trends [1–7, 20]), there is good justification for using the fetuses-at-risk approach for examining gestational age-specific SIDS rates. Finally, the contrast between the gestational age-specific patterns of SIDS and diseases of prematurity (e.g., retinopathy of prematurity) suggest that SIDS is a late gestation disease whose incidence may have been affected by temporal changes in the gestational age distribution [14].

In this study we explored the extent to which changes in gestational age distribution and changes in gestational age-specific SIDS rates contributed to the temporal decline in SIDS among singletons and twins.

We used population-based data on singleton and twin births in the United States from 1995–96 to 2004–05 from the National Centre for Health Statistics (NCHS). Information in the NCHS birth/death and fetal death files was abstracted from birth certificates and is publicly available [21], with the birth-infant death linkage carried out by the NCHS (period linked birth-infant death file). We included all infants born at ≥22 week gestation, based on the clinical estimate of gestation at birth [22–24]. States that did not report the clinical estimate of gestational age were not included in the analysis (13.5% of all births). We excluded infants weighing less than 500 grams in order to avoid potential bias due to variable birth registration at the borderline of viability [25–27]. International Classification of Diseases (ICD) codes 798.0 and R95 (ICD 9^th version, and ICD 10^th version, respectively) were used to identify cases of SIDS. Information about maternal and infant risk factors associated with SIDS, including maternal age, education, race, parity, and marital status, was also obtained from the NCHS files. Temporal changes in the frequency of these risk factors were evaluated by contrasting their population prevalence between 1995–96 and 2004–05. Temporal changes in SIDS rates across categories of each risk factor were evaluated using rate ratios (2004–05 vs. 1995–96) and 95% confidence intervals.

Gestational age-specific SIDS rates were compared using two different approaches: A) the traditional method which expressed gestational age-specific SIDS rates as the number of SIDS cases at any gestation divided by the number of live births at that gestation; and B) the fetuses-at-risk approach. Under the fetuses-at-risk approach, gestational age-specific SIDS rates were calculated as the number of SIDS cases among infants born at any gestation divided by the number of fetuses in-utero who were at risk of birth (live birth or stillbirth) at that gestation [12, 13]. This latter model assumes an intrauterine etiology for SIDS and has been used previously for estimating gestational age-specific rates of stillbirth, neonatal death and cerebral palsy [12, 13]. Both approaches were used to examine the temporal trends in SIDS, because they embody different perspectives; the traditional approach models the gestational age-specific risk of SIDS after birth assuming that live births are the appropriate candidates for SIDS, whereas the fetuses-at-risk approach models an in-utero etiology and assumes that fetuses are the appropriate candidates for SIDS.

where N1 and N2 denote SIDS rates in 2004–05 and 1995–96, respectively, R₁ and R₂ refer to gestational age-specific SIDS rates in 2004–05 and 1995–96, F₁ and F₂ represent proportions of live births in gestational age category i for each respective time period, and i denotes gestational age category (in weeks). The first part of the equation represents the relative contribution of changes in the gestational age distribution to the overall difference in SIDS rates, and the latter part of the equation represents the relative contribution of changes in gestational age-specific SIDS rates. For the decomposition using the traditional approach, the gestational age distribution at gestational week i was defined as the number of live births at that gestation expressed as a proportion of all live births; for the decomposition using the fetuses-at-risk approach, the gestational age distribution at gestational week i was expressed as the number of live births at that gestation expressed as a fraction of all fetuses in-utero at that gestation.

The Kitagawa decomposition was carried out separately for singletons and twins born at term vs pre-term gestation (≥37 weeks vs 22–36 weeks).

New birth certificates were introduced in the United States in 2003 and led to some increases in missing values for a few variables of interest (e.g., educational status, congenital malformations). Sensitivity analyses were carried out to assess how these changes affected results by restricting temporal trends to a period before the introduction of the new birth certificate i.e., between 1995–96 and 2001–02. All analyses were carried out using SAS version 9.2. Data used in this study were publicly accessible from the National Centre for Health Statistics [21].

The rate of SIDS declined from 8.3 to 5.6 per 10,000 live births from 1995–96 to 2004–05 among singletons (rate difference −2.7, 95% CI: -2.4 and −3.0), and from 14.2 to 10.6 per 10,000 live births among twins (rate difference −3.6, 95% CI: -1.4 and −5.9). On a relative scale SIDS rates declined by 33% (rate ratio 0.67, 95% CI: 0.67-0.67) among singletons and by 25% among twins (rate ratio 0.75, 95% CI: 0.74-0.75).

(Figure 1a shows the gestational age-specific rates of SIDS among singleton and twin live births between 28 and 40 weeks gestation as calculated under the traditional perinatal model. Rates of SIDS declined with increasing gestational age among both singletons and twins. Rates of SIDS were lower among twins at preterm gestation compared with singleton live births at preterm gestation, but the opposite was true at later gestational ages (paradox of intersecting perinatal mortality curves). (Figure 1b shows gestational age-specific rates of SIDS among singletons and twins under the fetuses-at-risk model. Rates of SIDS increased with increasing gestation among both singletons and twins and SIDS rates were higher among twins at all gestational ages.

Substantial changes occurred in the gestational age distribution of singleton live births between 1995–96 and 2004–05 ((Figure 2a). The proportion of singleton live births at gestational ages up to 39 weeks increased, while the proportion after 39 weeks declined. Under the traditional model, gestational age-specific SIDS rates showed a temporal decline at all gestational ages (Figure 2b), while under the fetuses-at-risk approach, gestational age-specific SIDS rates showed a temporal decline at 39 weeks and later (Figure 2c). (Figure 3a shows changes in the gestational age distribution among twins between 1995-96 and 2004–05; the proportion of live births up to 37 weeks increased and there was a decline in the proportion of births at 38 weeks and later. Gestational age-specific SIDS rates among twins showed a temporal decline at all gestations under the traditional model ((Figure 3b), while under the fetuses-at-risk model, no decline in rates was evident except at 40 weeks of gestation (Figure 3c).

The decline in SIDS among twins followed a similar pattern. Under the traditional Kitagawa decomposition, the entire temporal change in SIDS rates (−3.6 cases per 10,000 live births) was due to changes in gestational age-specific mortality (Table 3). Under the fetuses-at-risk approach, however, the temporal shift in the gestational age distribution was responsible for 63% of the decline in SIDS (−2.3 cases per 10,000 fetuses-at-risk), while change in gestational age-specific SIDS rates were responsible for 37% of the observed SIDS reduction (−1.3 SIDS cases per 10,000 fetuses-at-risk, Table 3).

The decomposition of the SIDS decline yielded different results at preterm vs term gestation. Under the traditional model, the change in the gestational age distribution had a relatively larger adverse effect at preterm gestation among singletons, whereas among twins, the change in the gestational age distribution adversely affected SIDS rates at preterm gestation only. Changes in gestational age-specific SIDS rates were responsible for a larger proportion of the SIDS decline at preterm gestation compared to term gestation among both singletons and twins (Table 3). Under the fetuses-at-risk approach, gestational age distribution changes contributed substantially to the SIDS decline at term gestation among both singletons and twins. On the other hand, changes in gestational age-specific SIDS rates contributed more to the SIDS decline at preterm gestation compared to term gestation (Table 3).

Results from sensitivity analyses restricted to years prior to the introduction of new birth certificates in 2003 showed that between 1995–96 and 2001–02, the decline in SIDS rates was similar among singletons and twins. SIDS rates declined from 8.3 to 5.9 per 10,000 live births among singletons and from 14.2 to 10.5 per 10,000 live births among twins (rate ratio for singletons 0.72, 95% CI 0.69-0.75 and for twins 0.74, 95% CI 0.62-0.89).

Our study confirms the paradoxical relationship between plurality and SIDS under the traditional perinatal model; twins had lower rates of SIDS at preterm gestation, while singletons had lower rates at later gestation. The fetuses-at-risk approach eliminated the paradoxical crossover in SIDS rates by plurality and showed that, in fact, twins had higher rates of SIDS at all gestational ages. Analyses using the fetuses-at-risk approach also showed that temporal changes in the gestational age distribution of live births were responsible for 45% of the temporal decline in SIDS rates among singletons and for 64% of the decline among twins. The remainder of the decline in SIDS (55% among singletons and 37% among twins) was attributed to reductions in gestational age-specific rates of SIDS (e.g., due to the back to sleep campaign, etc.). This effect of a temporal shift in gestational age distribution was evident predominantly among term singletons and term twins suggesting that the temporal reduction in SIDS occurred, in part, due to a shift in the gestational age distribution among term infants.

While the exact cause of SIDS is unknown, the ‘triple risk hypothesis’ implicates three causes in the etiology of SIDS, namely, genetic factors, a critical developmental period, and environmental factors [29]. These three risk factors may affect the fetus/infant sequentially at different stages of development, and progressively increase the risk of SIDS. Genetic predisposition is related to male gender and race [29, 30] and several gene polymorphisms involved in autonomic function, neurotransmission, energy metabolism, and response to infection have been implicated [31, 32]. However, temporal changes in genetic factors are unlikely to explain the temporal decline in SIDS.

Critical developmental periods may occur at various time-windows during fetal or infant development. Suboptimal conditions in-utero, due to hypoxia, have been implicated in the origins of SIDS [15]. Studies have shown that infants who died of SIDS had a higher incidence of subcortical leukomalacia, brainstem gliosis, and other changes in central nervous system, when compared with infants not affected by SIDS [33]. These lesions have been found to originate, in part, in the antenatal period, suggesting that an antepartum hypoxic insult may constitute a predisposing risk factor for SIDS [16, 33]. Due to a resulting dysfunction of the autonomic nervous system, SIDS victims have a diminished capacity to respond to physiological challenges during a vulnerable developmental period between 2 to 4 months after birth, when the majority of SIDS occurs [15, 30, 31]. Similar pathologic characteristics at autopsy have also been found among stillborn fetuses, suggesting that SIDS and unexplained stillbirth represent the same phenomenon [15–17]. SIDS and unexplained stillbirth share similar risk factors including male gender, seasonality, maternal smoking, a parity of 3 or more, race, extremes of maternal age, low education, single marital status and low socio-economic status [15, 18, 19]. Antepartum hypoxia is believed to be responsible for the majority of unexplained antepartum stillbirths [34]. Many complications of pregnancy that necessitate iatrogenic delivery involve fetal hypoxia [35]. The recent temporal increase in medically indicated deliveries has been shown to coincide with a reduction in stillbirth rates [36] and the decline in SIDS follows a similar temporal pattern.

Changes in environmental factors, such as the infant’s sleeping position, safe sleep environments and second hand smoke have changed over time, and likely contributed to a substantial fraction of the decline in gestational age-specific rates of SIDS. More recently, side-sleeping and bed-sharing have been identified as risk factors for SIDS [37]. Other explanations for the temporal reduction in SIDS rates have been proposed including temporal changes in SIDS case ascertainment and death certificate coding practices. Studies have shown that the reduction in SIDS in the United States paralleled a temporal increase in deaths caused by ‘accidental suffocation and strangulation’ [38]. However, the decline in SIDS has been observed in many countries and a similar shift in coding practices worldwide is unlikely [1, 20, 39]. For instance, a significant temporal reduction in SIDS was observed in the United Kingdom, where SIDS cases were carefully evaluated by the Confidential Enquiry into Stillbirths and Death in Infancy team. This included a thorough clinical and criminal investigation and a post-mortem examination by a pediatric pathologist [7].

Our study adds to the literature suggesting that unexplained fetal death and SIDS may have common causal pathways. The fetuses-at-risk approach shows that gestational age-specific rates of stillbirth and SIDS both increase with advancing gestation. Further, the fetuses-at-risk approach resolves the paradoxical relationship between gestational age-specific rates of SIDS and plurality and shows that twins are at higher risk of SIDS at all gestational ages. This adds to a growing body of evidence that suggests an intrauterine etiology for SIDS [15–17]. Our finding that changes in the gestational age distribution contributed significantly to the temporal reduction in SIDS rates also supports a common causal pathway between stillbirth and SIDS. The iatrogenic left shift in the gestational age distribution at term and post-term gestation that has occurred in recent decades has had a positive impact on both stillbirth and SIDS.

Our study has several limitations. We used death certificates to identify SIDS and this information is subject to potential misclassification as death certification processes are limited in their ability to incorporate all information about the circumstances leading to death [40, 41]. A systematic misclassification, such as an increasing preference for other types of diagnoses than SIDS may have contributed to the temporal decline in gestational age-specific rates of SIDS in our study. The diagnostic code for SIDS changed from ICD-9 (code 798.0) to ICD-10 (code R95) in 2000, and SIDS cases were more likely to be reported under the new ICD-10 coding rules [42]. This change was estimated to artificially increase the SIDS rates by about 3% [42]. However, any effect of changes in coding practices would likely be uniform across all gestational ages. The relative reduction in SIDS in our study was larger among singletons compared with twins between 1995–96 vs 2004–05, whereas a similar relative reduction was observed for singletons and twins between 1995–96 and 2001–02. This is in contrast to data from England, where the relative SIDS reduction was larger in twins as compared with singletons [9]. Potential explanations for this discordance include differences in the time period when the largest changes in the gestational age distribution and in SIDS rates occurred in the 2 countries. Data from the United States show that the SIDS decline was larger among twins as compared with singletons from 1990 to 2005 (rate ratio 0.35, 95% CI: 0.28-0.43 among twins vs rate ratio 0.43, 95% CI: 0.41-0.45 among singletons). However, clinical estimates of gestational age were not available prior to 1995 and this restricted the time period of our study.

In conclusion, our study indicates that in addition to the back-to-sleep campaign, temporal changes in the gestational age distribution may have contributed to the overall reduction in SIDS. This effect of a temporal shift towards earlier gestation at delivery has been observed predominantly at term and post-term gestation, and to a larger extent among twins. These findings support the hypothesis that antenatal factors contribute to the origins of SIDS, and endorse the concept of similar causal pathways between unexplained fetal death and SIDS.

None of the authors have a personal financial relationship relevant to this article.

The authors declare that they have no competing interests.