- Email: [email protected]
A competing risk model of sudden infant death syndrome incidence in two US birth cohorts
A competing risk model of sudden infant death syndrome incidence in two US birth cohorts Harold A. Pollack, MPP, PhD, and John G. Frohna, MD, MPH Objectives: To compare changing incidence and changing risk factors associated with sudden infant death syndrome (SIDS) in the 1989 and 1996 US birth cohorts. Study design: All available singleton births over 500 g from the 1989 linked birth–infant death file and the 1996 and 1997 Perinatal Mortality files were examined. A log-logistic survival model was used to explicitly account for declining competing risks among low birth weight infants. Results: Controlling for maternal prenatal smoking and other confounders, SIDS incidence declined by >33% between the 2 survey years (adjusted odds ratio = 0.628 with 95% CI [0.598, 0.660]). Self-reported declines in maternal prenatal smoking were also associated with significant declines in SIDS incidence. African American infants and infants born weighing <1000 g experienced increased relative risk compared with non-Hispanic white infants born weighing >2500 g. Hispanic/Latino infants had significantly lower SIDS risk than non-Hispanic white infants in both years. Accounting for declining competing risks and other factors, relative SIDS risks among infants born between 500 and 1000 g increased over the study period. Conclusions: SIDS incidence sharply declined between 1989 and 1996. High incidence of SIDS in African Americans and increased relative SIDS risk for infants born weighing <1000 g require increased attention from clinicians and public health policy makers. (J Pediatr 2001;138:661-7)
Sudden infant death syndrome is the most common cause of post-neonatal infant death in the United States. In 1992, the American Academy of Pediatrics recommended that “healthy term
infants” be placed on their backs to sleep.1 The decline in the prevalence of prone infant sleep was associated with declining SIDS incidence.2 The prevalence of prone sleeping has since de-
From the University of Michigan School of Public Health, Ann Arbor.
Submitted for publication May 17, 2000; revisions received Aug 17, 2000, and Sept 29, 2000; accepted Oct 11, 2000. Supported in part (data collection) by grant No. 031605 from the Robert Wood Johnson Foundation Substance Abuse Policy Research Program. Any opinions expressed or policy recommendations are those of authors and do not reflect those of the Robert Wood Johnson Foundation. Reprint requests: Harold A. Pollack, MPP, PhD, Assistant Professor of Health Management and Policy, University of Michigan School of Public Health, 109 Observatory, Ann Arbor, MI 48109-2029. Copyright © 2001 by Mosby, Inc. 0022-3476/2001/$35.00 + 0 9/21/112248 doi:10.1067/mpd.2001.112248
clined from 70% in 1992 to 24% in 1996.3,4 In subsequent analyses of 1994-98 trends, Willinger et al5,6 documented that the prevalence of prone sleeping declined from 44% to 17% among whites and from 53% to 32% among African Americans. Sleep position for premature and low birth weight infants has been a matter of greater uncertainty. Partly out of concern that premature infants might incur special risks from supine sleeping, the initial AAP recommendation did not apply to “premature infants experiencing respiratory distress.”1 However, in 1996, the Academy removed the specific exception for premature infants, stating that the benefits of supine sleeping outweigh the potential risks, except in cases of relatively rare specific diagnoses.7 AAP AOR ELBW LBW SIDS VLBW
American Academy of Pediatrics Adjusted odds ratio Extremely low birth weight Low birth weight Sudden infant death syndrome Very low birth weight
Low birth weight infants have a significantly higher risk of SIDS, a risk tightly correlated with prematurity. The period of greatest risk for SIDS occurs 4 to 6 weeks later in premature infants than in term infants.8 The Nordic Epidemiological SIDS study suggested important interaction effects between low birth weight and both prone and side sleeping. Compared with normal birth weight infants sleeping in the supine position, premature infants sleeping prone had an adjusted odds ratio for SIDS of 48.8 (95% CI 19-128).9 High SIDS incidence among low birth weight, very low birth weight, 661
POLLACK AND FROHNA
and extremely low birth weight infants persists despite increased overall survival within these infant populations.10 The decrease in SIDS incidence for VLBW infants lags behind that for other birth weight groups.11,12 Although SIDS incidence has decreased among VLBW infants, SIDS incidence within this population remains much higher than for infants of normal birth weight.13 The greater risk of SIDS for preterm infants was noted in the recent AAP statement, and the supine sleep position is now explicitly recommended for these infants.14 Given the increased survival of VLBW and ELBW infants, we hypothesize that the relative risk of SIDS for these infants may be even higher now than it was before the “Back to Sleep” campaign. We have used an explicit survival model to test the alternative possibility that higher relative incidence of SIDS is merely a statistical artifact that reflects declining VLBW and ELBW infant mortality from other causes. We examined the impact of birth weight and other SIDS risk factors for nearly 8 million singleton infants born within the 1989 and 1996 birth cohorts, using a competing risk analysis to account for increased survival among low birth weight infants.
METHODS Population Data are from the National Center for Health Statistics’ public-use birth certificate files. Data for the 1989 birth cohort are from the 1989 Linked Birth Infant Death File. This file contains matched individual-level data regarding almost all US births and, when applicable, infant death records. A linked birth—infant death file was unavailable for the 1996 US birth cohort. However, analogous data were contained in the 1996 and 1997 National Center for Health Statisitics Perinatal Mortality files.15 The 1996 file contained data on known births from 1996 and on infant deaths during the same 662
THE JOURNAL OF PEDIATRICS MAY 2001 year. The 1997 Perinatal Mortality data are needed for infants born during 1996 who died during 1997. Because infants who died during 1997 appear as live births in the 1996 data set and again as 1997 infant deaths, we followed the algorithm of Pollack, Lantz, and Frohna16 to find and remove redundant data entries. We matched infants across the data sets based on state of residence, birth weight in grams, gestation, maternal race, and maternal age at delivery to eliminate double-counting. Such matching is especially important for ELBW infants, whose high mortality rate would otherwise produce frequent double-counting. In particular, 3300 infants born during calendar year 1996 were recorded as 1997 deaths. We could match 2514 of these infants to a unique 1996 birth record. The remaining 786 were matched to multiple 1996 births. In such cases, the first matching live birth record was retained for the analysis. We exclude twins and higher plurality births because these infants raise issues outside the current analysis. We also exclude infants weighing <500 g at birth because the non-SIDS infant mortality rate is virtually 100% in this group and because infant and fetal deaths may be difficult to distinguish at such low birth weights. The resulting files include individuallevel birth certificate data for virtually all US births. These files provide many data items pertinent to SIDS risk. Maternal characteristics such as age, marital status, race/ethnicity, and years of completed education are available in these birth certificate data. Infant characteristics such as birth weight and gestational age are also provided. We chose to focus on birth weight, because this appears to be more reliably measured than gestational age in national birth certificate data.17 Data regarding maternal smoking during pregnancy—a key risk factor for SIDS—are also available for most US births. Unfortunately, several states
including New York and California do not provide smoking data in public-use data. We have included a dummy variable to mark births for which maternal smoking data are unavailable. No data are available to measure paternal smoking during pregnancy or postnatal tobacco exposure. In cases of infant death, these data files provide linked birth certificate and death certificate information for most infants. Supporting documentation indicates that more than 97% of infant deaths were successfully linked with accompanying birth certificate data.15,18 To address competing risks and changes in the timing of SIDS deaths, we have also analyzed age at infant death, as recorded on the infant death certificate.
Impact of Birth Weight An important feature of this analysis is examination of changing patterns of SIDS incidence among low birth weight infants (birth weight <2501 g), VLBW infants (birth weight <1501 g), and ELBW infants (birth weight <1001 g). The populations of VLBW and ELBW infants, although small, are important because these infants face high mortality risks from SIDS and other causes. Because VLBW and ELBW infants represent a small fraction of all births, large national data sets are required to examine SIDS risks. In the 1996 birth cohort, we identified fewer than 100 SIDS deaths among infants weighing between 500 and 1500 g. Failure to model improved neonatal survival might lead researchers to conclude that SIDS risks had increased for VLBW and ELBW infants, even if this pattern is entirely an artifact of declining competing risks. Over time, VLBW and ELBW infants are increasingly likely to survive into the postneonatal period of peak SIDS risk. Because of this increased exposure, the number of SIDS deaths among VLBW and ELBW infants might have increased between 1989 and 1996, even
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 if the probability of SIDS death per infant per day had actually declined over the same period. We use survival analysis to account for these competing risks.
Table I. Descriptive statistics
Multivariate Analysis Our main statistical approach is to estimate a survival model of SIDS risk. Within this framework, the dependent variable is an infant’s reported age at death, measured in days. We estimate a log-logistic specification, which provides the simplest and most readily interpreted extension of binary logistic regression to account for competing risk.19 In the absence of competing risk, the log-logistic survival model and the standard binary logistic model yield equivalent results. We also analyze an interaction specification to examine the impact of changing competing risks over the study period. In our competing risk approach, we follow standard methodology in considering non-SIDS mortality to be a form of non-informative right-censoring. Accounting for observed risk factors, an infant death from other causes on day T is considered to reveal no additional information about that infant’s probability of subsequent SIDS death, had he or she survived to day T + 1. It is possible that SIDS-vulnerable infants are also vulnerable to other life-threatening events. In principle, a more elaborate parametric model might explore this possibility.19,20 Unfortunately, current data do not allow reliable investigation of these effects. Within the log-logistic model, SEs accompanying adjusted odds ratios are adjusted by using the delta method to reflect SEs in the estimated scale parameters.
RESULTS Population SIDS incidence declined by almost half between the 2 study years, from 1.35 per 1000 live births in 1989 to 0.76 per 1000 in 1996 (Table I). The infant mortality rate in our study sam-
Race/ethnic group Non-Hispanic white African American Mexican Puerto Rican Other Other Hispanic Mother smoked during pregnancy Mother not married at birth Mother married at birth Mother completed <12 grades of schooling at birth Mother at least high school graduate at birth Mother college graduate at birth Birth weight ≥2501 g 1501-2500 g 1001-1500 g 500-1000 g
1996 Singleton mean (% of sample)
1989 SIDS incidence, singleton sample (per 1000)
1996 SIDS incidence, singleton sample (per 1000)
67.9 14.8 12.8 1.4 5.1 3.7 13.1
1.2 2.3 0.9 0.7 1.5 0.9 3.0
0.7 1.5 0.5 0.9 0.7 0.4 2.2
32.5 67.5 23.7
2.4 1.0 2.1
1.4 0.5 1.3
76.3
1.2
0.6
21.6
0.5
0.2
94.1 5.0 0.5 0.5
1.2 3.6 5.4 2.2
0.7 2.0 2.9 2.1
Percentages do not add to 100.0 because of rounding.
ple is slightly lower than reported rates for the overall population of US infants, reflecting the restricted group of singleton infants with birth weights that exceed 500 g. Overall SIDS incidence declined among African Americans. However, the relative risk slightly increased over the study period when compared with non-Hispanic whites. In the 1996 cohort, African Americans experienced 2.5 times the SIDS incidence of nonHispanic whites. Our point estimates suggest that most, though not all, of this racial disparity is attributable to group differences in observed risk factors such as birth weight, maternal educational attainment, and marital status. SIDS incidence was extremely low among Mexican Americans and Puerto Ricans.
Maternal smoking during pregnancy declined by almost one third between the 2 study years. Because maternal smoking is associated with more than a doubling of SIDS risk, declining maternal smoking is estimated to produce a significant decline in SIDS incidence. Overall infant mortality declined from 7.7 per 1000 to 5.36 per 1000, although the reported incidence of low birth weight, VLBW, and ELBW was virtually unchanged. Non-SIDS neonatal mortality significantly declined within these categories between 1989 and 1996. Within the 1996 cohort, some of these marginal survivors survived into the post-neonatal period when they subsequently died of SIDS. Partly because of such declining competing risks, the proportion of SIDS cases that involved ELBW infants in663
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS MAY 2001
Table II. Analysis results (standard errors adjusted by delta method)
Race/ethnic group Non-Hispanic white African American Mexican American Puerto Rican Other Other Hispanic Mother didn’t smoke during pregnancy Mother smoked during pregnancy No available smoking data Mother completed <12 grades of schooling at birth Mother at least high school graduate at birth Mother college graduate at birth Mother not married at birth Mother married at birth Birth weight ≥2501 g 1501-2500 g 1001-1500 g 500-1000 g 1989 1996 Smoking interacted with 1996 African American interacted with 1996 Birth weight 1501-2500 g interacted with 1996 Birth weight 1000-1501 g interacted with 1996 Birth weight 500-1000 g interacted with 1996 Scale parameter in survival model
AOR from baseline survivor model (95% CI)
AOR from interactive survivor model (95% CI)
AOR from binary logistic regression (95% CI)
1.000 1.242 [1.173, 1.315] 0.549 [0.498, 0.604] 0.532 [0.427, 0.664] 1.176 [1.056, 1.308] 0.572 [0.492, 0.665] 1.000 2.209 [2.081, 2.345] 1.035 [0.980, 1.095] 1.000 0.651 [0.620, 0.684] 0.550 [0.500, 0.605] 1.000 0.628 [0.596, 0.662]
1.000 1.147 [1.071, 1.229] 0.559 [0.508, 0.616] 0.533 [0.428, 0.665] 1.185 [1.065, 1.319] 0.577 [0.496, 0.671] 1.000 2.058 [1.919, 2.208] 1.025 [0.969, 1.084] 1.000 0.655 [0.623, 0.688] 0.550 [0.505, 0.610] 1.000 0.627 [0.595, 0.661]
1.000 1.148 [1.072, 1.229] 0.559 [0.508, 0.615] 0.533 [0.428, 0.664] 1.185 [1.065, 1.319] 0.577 [0.497, 0.670] 1.000 2.063 [1.930, 2.204] 1.025 [0.969, 1.084] 1.000 0.655 [0.624, 0.687] 0.555 [0.506, 0.610] 1.000 0.627 [0.596, 0.660]
1.000 2.157 [2.013, 2.312] 3.215 [2.750, 3.759] 1.865 [1.476, 2.357] 1.000 0.628 [0.598, 0.660] — — — — — 0.549 [0.537, 0.561]
1.000 2.186 [2.010, 2.379] 3.231 [2.669, 3.911] 1.346 [0.973, 1.863] 1.000 0.555 [0.519, 0.593] 1.226 [1.107, 1.358] 1.259 [1.134, 1.397] 0.965 [0.842, 1.105] 0.990 [0.719, 1.364] 2.366 [1.483, 3.777] 0.549 [0.537, 0.561]
1.000 2.192 [2.022, 2.377] 3.247 [2.692, 3.917] 1.348 [0.974, 1.865] 1.000 0.553 [0.518, 0.589] 1.223 [1.104, 1.354] 1.261 [1.136, 1.398] 0.947 [0.827, 1.086] 0.925 [0.671, 1.274] 1.593 [0.999, 2.540] —
The final row indicates the scale factor, an estimated parameter of the log-logistic model required to convert the coefficients into AOR form.
creased from 0.70% to 1.22% between 1989 and 1996. Uniformly across the first year of life, the SIDS hazard rate—the probability of SIDS per live infant per day—declined by approximately one third between the 2 years. The timing of SIDS deaths appeared similar across categories of race/ethnicity, maternal marital status, and education groups across both years. For infants born weighing >1500 g, hazard rates were highest between 30 and 120 days of life. Approximately 90% of SIDS deaths occurred during the first 6 months. 664
Multivariate Analysis Adjusted odds ratios are shown to match standard epidemiologic presentations (Table II). Adjusting for known risk factors, SIDS incidence declined by more than one third between the 2 years (AOR = 0.628). Given 1996 baseline values, this corresponds to more than 1600 averted deaths annually. Other coefficients are also pertinent in documenting changing SIDS risks. We find an AOR of 2.209 associated with maternal smoking. Between 1989 and 1996, the prevalence of maternal smoking during pregnancy declined by
5.1 percentage points. According to the methods used in previous research,21 these point estimates imply that declining maternal smoking is associated with a decline of 0.061 SIDS deaths per 1000 in the 1996 birth cohort, or approximately 250 cases throughout the United States. The second set of coefficients in Table II shows results from the baseline model, augmented with timeinteraction terms to capture changing patterns of competing risk. Results from the survival model are shown side by side with a standard binary lo-
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 gistic regression. As shown, AORs are virtually identical in the 2 models for all main effects. Relative risks associated with African American births (AOR = 1.259) and maternal smoking during pregnancy (AOR = 1.226) slightly increased between 1989 and 1996. The most dramatic change occurred for infants born weighing between 500 and 1000 g, whose SIDS risks are altered by declining birth weight–specific mortality between the 2 years. These infants had more than double the relative SIDS risk in the survival model (AOR = 2.366).
DISCUSSION In this study a competing-risk survival model was used to explore SIDS incidence in the 1989 and 1996 birth cohorts. Controlling for important confounders, we find a large decline in SIDS hazard rates, with a baseline AOR of 0.647. Among whites, declining SIDS incidence accounted for 30% of the observed decline in overall infant mortality rate between the 2 study years. ELBW infants were a larger proportion of SIDS deaths in 1996 (1.22%) than they were in 1989 (0.70%). Estimated SIDS hazard rates declined among low birth weight and VLBW infants between the 2 years. However, observed SIDS incidence within these groups was stable over time because of declining competing risks. Because VLBW and ELBW infants experience high mortality from other causes, high SIDS incidence within these birth categories is easily overlooked in analyses that do not consider competing risks. SIDS deaths also occur significantly later within the ELBW group than among infants of higher birth weights.8 However, SIDS hazard rates rapidly increase during subsequent months. These patterns are difficult to detect but are striking within a national ELBW sample. Controversy remains regarding proper sleep position for VLBW and
ELBW infants. Various authors have documented improved respiratory function in the prone position.22 Revised AAP recommendations indicate that premature or low birth weight infants should generally sleep supine, because the risks of prone sleep outweigh potential benefits.7,14* The SIDS Global Strategy Task Force (Developmental Physiology Working Group) has also emphasized the importance of supine sleeping for preterm infants, including patients in the neonatal intensive care unit, except for those with specific contraindications to supine sleeping.23 Ottolini et al24 found that parents who initially placed their children supine may switch to the prone position because of the perception that the infants sleep better. This perception is based in reality: prone sleep does seem to increase sleep duration and decrease arousals.25 Ironically, this decreased arousal may itself pose SIDS risks. Health care professionals should frequently discuss these issues during well-child visits in the first 6 months of life, especially for premature infants. Premature infants present especially important opportunities for preventive intervention because they face high SIDS risk and have frequent contact with the health care delivery system. A small but important portion of the decline in SIDS incidence is attributable to declining smoking prevalence among pregnant women. Maternal smoking before, during, and after pregnancy is an important and preventable threat to infant health. Smoking cessation interventions, though imperfect, have been shown to be cost-effective in many studies, including studies of SIDS risk.26 This study includes important limitations. First, maternal smoking data are based on self-report and may therefore understate true prevalence.27 The size and direction of the resulting bias *We thank an anonymous referee and Dr Carl E. Hunt for this reference.
depends on the specific pattern of misreporting.16 Misdiagnosis raises a second potential problem. SIDS is a diagnosis of exclusion—a diagnosis sometimes made under adverse conditions without autopsy or recommended death scene investigation.28 We know of no evidence that misdiagnosis patterns have significantly changed in recent years. Data from England and Australia suggest that the pattern of SIDS diagnoses was unaltered by the introduction of the “Back to Sleep” campaigns.29,30 Ambiguous diagnoses might be more important for the small population of VLBW and ELBW infants with a diagnosis of SIDS during the 2 study years. These infants face high risks of mortality from other causes. Changing treatment patterns or diagnostic methods involving VLBW and ELBW infants might have influenced reported patterns of SIDS deaths. VLBW and ELBW infants highlight the third limitation of the current study. We estimate a competing risk model in which non-SIDS infant deaths are regarded as non-informative of subsequent SIDS risk. Some 1996 VLBW and ELBW infants were “marginal survivors”; that is, they survived the first year of life as members of the 1996 cohort, but they would have died from other causes had they been born with the same characteristics in the 1989 cohort. If marginal survivors faced higher but unobserved SIDS risks when compared with their 1996 peers, the current analysis would understate the role of VLBW and ELBW and would overstate the ELBW and VLBW interaction terms. ELBW infants also experience health problems that require close monitoring and that may therefore reduce SIDS risk for important periods. Infants in the neonatal intensive care unit or other intensive care settings are less vulnerable to SIDS than are infants in noninstitutional settings. We believe that such monitoring accounts for the extremely low SIDS incidence in early in665
POLLACK AND FROHNA
fancy among ELBW infants. Our reported coefficients may therefore understate true SIDS risks facing the same infants sleeping at home, in day care, or in other non-institutional settings. A fourth limitation speaks to tradeoffs inherent in large-sample, birth certificate data sets. Important variables such as family income are unavailable. Birth certificate data address prenatal maternal smoking but do not address post-natal smoking, smoking by other household members, or illicit substance use. These data do not address infant sleep position, co-sleeping, and other epidemiologic risk factors. Some of these variables are included in more detailed surveys such as the Pregnancy Risk Assessment and Monitoring System (PRAMS).
Implications Despite these limitations, our findings highlight the remarkable success of SIDS prevention. Improved prenatal care timing, reduced prevalence of maternal smoking, and other observed factors have helped to reduce the number of SIDS deaths. Yet these changes account for only a small fraction of the observed decline. The “Back to Sleep” campaign’s simple, consistent, and clear message provides an important model for other public health education interventions. Our findings also highlight shortcomings of SIDS prevention. Relative risks for African Americans failed to decline, and indeed slightly increased, between 1989 and 1996. These results are consistent with prior studies indicating that African Americans are less likely than non-Hispanic whites or Hispanic/Latinos to sleep supine.4,24,31 Young mothers and mothers with fewer years of completed education appear especially likely to place their infants prone.3,10,32 Improved outreach and clinical interventions are critical to reducing SIDS incidence within these high-risk populations.31 Parents of VLBW and ELBW infants require appropriate SIDS prevention 666
THE JOURNAL OF PEDIATRICS MAY 2001 education, including information about the dangers of environmental tobacco exposure and prone infant sleep. Nurses and physicians should explicitly recommend that infants be placed supine to sleep, because this advice has been the most predictive of proper sleep position.5,6 When babies near the time of discharge, hospital staff must model proper sleep position, because most mothers who saw their infants placed prone at the hospital reported that they would place their children prone at home.31 Even infants initially placed in the back or side position are at increased risk for being placed prone during the first 6 months of life as a means to improve infant comfort.24 Parents of neonatal intensive care unit graduates may therefore require more guidance regarding infant comfort and sleep patterns after discharge. As neonatal survival improves within this medically fragile group, SIDS prevention becomes an increasingly important public health concern. We thank Gary Freed, Myron Wegman, and Jack Wheeler for useful comments.
REFERENCES 1. AAP Task Force on Infant Positioning and SIDS. Positioning and SIDS. Pediatrics 1992;89:1120-6. 2. Willinger M, Hoffman HJ, Harford RB. Infant sleep position and risk for sudden infant death syndrome: report of meeting held January 13 and 14, 1994, National Institutes of Health, Bethesda, MD. Pediatrics 1994;93:814-9. 3. Willinger M, Hoffman H, Wu K, Hou J, Kessler RC, Ward SL, et al. Factors associated with the transition to nonprone sleep positions of infants in the United States: The national infants sleep position study. JAMA 1998;280:329-35. 4. Centers for Disease Control and Prevention. Progress in reducing risky infant sleeping positions—13 states, 1996-1997. MMWR Morb Mortal Wkly Rep 1999;48:878-82. 5. Willinger M, James LS, Catz C. Defining the sudden infant death syndrome (SIDS): deliberations of an expert panel convened by the National Institute of Child Health and Human Development. Pediatr Pathol 1991;11:677-84.
6. Willinger M, Ko C, Hoffman H, Kessler R, Corwin M. Factors associated with caregivers’ choice of infant sleep position, 1994-1998: the National Infant Sleep Position Study. JAMA 2000;283:2135-42. 7. AAP Task Force on Infant Positioning and SIDS. Positioning and sudden infant death syndrome (SIDS): update. Pediatrics 1996;98:1216-8. 8. Malloy MH, Hoffman HJ. Prematurity, sudden infant death syndrome, and age of death. Pediatrics 1995;96:464-71. 9. Oyen N, Markestad T, Skjaerven R, Irgens LM, Helweg-Larsen K, Alm B, et al. Combined effects of sleeping position and prenatal risk factors in sudden infant death syndrome: The Nordic epidemiological SIDS study. Pediatrics 1997;100:613-21. 10. Bigger HR, Silvestri JM, Shott S, Weese-Mayer DE. Influence of increased survival in very low birth weight, low birth weight, and normal birth weight infants on the incidence of sudden infant death syndrome in the United States: 1985-1991. J Pediatr 1998;133:73-8. 11. Bigger HR, Silvestri JM, WeeseMayer DF. Sudden infant death syndrome (SIDS) death rates in normal birth weight (NBW > 2500g), low birth weight (LBW 1499-2500), and very-low birth weight (VLBW 5001499g) infants: comparison of rates before and after implementation of the Back to Sleep initiative [abstract]. Pediatr Res 2000;47:143A. 12. Bigger HR, Silvestri JM, Shott SB, Weese-Mayer DF. Changes in sudden infant death syndrome (SIDS) rates: gender comparisons by birth weight before and after implementation of the Back to Sleep initiative [abstract]. Pediatr Res 2000;47:143A. 13. Sowler B, Doyle LW, Morley CJ, Altmann A, Halliday J. Is sudden infant death syndrome still more common in very low birthweight infants in the 1990s? Med J Aust 1999;171:411-3. 14. AAP Task Force on Infant Positioning and SIDS. Changing concepts of sudden infant death syndrome: implications for infant sleeping environment and sleep position. Pediatrics 2000; 105:650-6. 15. National Center for Health Statistics. Perinatal mortality file supporting documentation. Hyattsville (MD): National Center for Health Statistics; 1999. 16. Pollack H, Lantz P, Frohna J. Maternal smoking and adverse birth outcomes among singletons and twins. Am J Public Health 2000;90:395-400.
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 17. Alexander G, Tompkins M, Petersen D, Hulsey T, Mor J. Discordance between LMP-based and clinically estimated gestational age: implications for research, programs, and policy. Public Health Rep 1995;110:395-402. 18. National Center for Health Statistics. 1995 Perinatal mortality file supporting documentation, 1998. Hyattsville (MD): National Center for Health Statistics; 1998. 19. Allison P. Survival analysis using the SAS system. Cary (NC): SAS Institute; 1995. 20. Lancaster T. The econometric analysis of transition data. New York: Cambridge University Press; 1990. 21. Pollack HA. Sudden infant death syndrome, maternal smoking during pregnancy, and the cost-effectiveness of smoking cessation intervention. Am J Public Health 2001;91:432-6. 22. Martin RJ, DiFiore JM, Korenke CB, Randal H, Miller MJ, Brooks LJ. Vulnerability of respiratory control in
23.
24.
25.
26.
27.
healthy preterm infants placed supine. J Pediatr 1995;127:609-14. Hunt C. Expanded ‘Back-to-Sleep’ recommendations: hospital-based safe sleeping practices. J SIDS Infant Mortal 1997;2:223. Ottolini MC, Davis BE, Patel K, Sachs HC, Gershon NB, Moon RY. Prone infant sleeping despite the “Back to Sleep” campaign. Arch Pediatr Adolesc Med 1999;153:512-7. Kahn A, Groswasser J, Sottiaux M, Rebuffat E, Franco P, Dramaix M. Prone or supine body position and sleep characteristics in infants. Pediatrics 1993;91:1112-5. Lightwood J, Phibbs C, Glantz S. Short-term health and economic benefits of smoking cessation: low birth weight. Pediatrics 1999;104:1312-20. Floyd R, Rimer B, Giovino G, Mullen P, Sullivan S. Review of smoking in pregnancy: effects on pregnancy outcomes and cessation efforts. Ann Rev Public Health 1993;14:379-411.
28. Byard R, Krous H. Suffocation, shaking or sudden infant death syndrome: can we tell the difference? J Paediatr Child Health 1999;35:432-3. 29. Leach CEA, Blair PS, Fleming PJ, Smith IJ, Platt MW, Berry PJ, et al. Epidemiology of SIDS and explained sudden infant deaths. Pediatrics 1999; 104:e43. 30. Byard RW, Beal SM. Has changing diagnostic preference been responsible for the recent fall in incidence of sudden infant death syndrome in South Australia? J Paediatr Child Health 1995;31:197-9. 31. Brenner RA, Simons-Morton BG, Bhaskar B, Mehta N, Melnick VL, Revenis M, et al. Prevalence and predictors of the prone sleep position among inner-city infants. JAMA 1998; 280:341-6. 32. Lesko SM, Corwin MJ, Vezina RM, Hunt CE, Mandell F, McClain M, et al. Changes in sleep position during infancy. JAMA 1998;280:336-40.
667
Sudden infant death syndrome is the most common cause of post-neonatal infant death in the United States. In 1992, the American Academy of Pediatrics recommended that “healthy term
infants” be placed on their backs to sleep.1 The decline in the prevalence of prone infant sleep was associated with declining SIDS incidence.2 The prevalence of prone sleeping has since de-
From the University of Michigan School of Public Health, Ann Arbor.
Submitted for publication May 17, 2000; revisions received Aug 17, 2000, and Sept 29, 2000; accepted Oct 11, 2000. Supported in part (data collection) by grant No. 031605 from the Robert Wood Johnson Foundation Substance Abuse Policy Research Program. Any opinions expressed or policy recommendations are those of authors and do not reflect those of the Robert Wood Johnson Foundation. Reprint requests: Harold A. Pollack, MPP, PhD, Assistant Professor of Health Management and Policy, University of Michigan School of Public Health, 109 Observatory, Ann Arbor, MI 48109-2029. Copyright © 2001 by Mosby, Inc. 0022-3476/2001/$35.00 + 0 9/21/112248 doi:10.1067/mpd.2001.112248
clined from 70% in 1992 to 24% in 1996.3,4 In subsequent analyses of 1994-98 trends, Willinger et al5,6 documented that the prevalence of prone sleeping declined from 44% to 17% among whites and from 53% to 32% among African Americans. Sleep position for premature and low birth weight infants has been a matter of greater uncertainty. Partly out of concern that premature infants might incur special risks from supine sleeping, the initial AAP recommendation did not apply to “premature infants experiencing respiratory distress.”1 However, in 1996, the Academy removed the specific exception for premature infants, stating that the benefits of supine sleeping outweigh the potential risks, except in cases of relatively rare specific diagnoses.7 AAP AOR ELBW LBW SIDS VLBW
American Academy of Pediatrics Adjusted odds ratio Extremely low birth weight Low birth weight Sudden infant death syndrome Very low birth weight
Low birth weight infants have a significantly higher risk of SIDS, a risk tightly correlated with prematurity. The period of greatest risk for SIDS occurs 4 to 6 weeks later in premature infants than in term infants.8 The Nordic Epidemiological SIDS study suggested important interaction effects between low birth weight and both prone and side sleeping. Compared with normal birth weight infants sleeping in the supine position, premature infants sleeping prone had an adjusted odds ratio for SIDS of 48.8 (95% CI 19-128).9 High SIDS incidence among low birth weight, very low birth weight, 661
POLLACK AND FROHNA
and extremely low birth weight infants persists despite increased overall survival within these infant populations.10 The decrease in SIDS incidence for VLBW infants lags behind that for other birth weight groups.11,12 Although SIDS incidence has decreased among VLBW infants, SIDS incidence within this population remains much higher than for infants of normal birth weight.13 The greater risk of SIDS for preterm infants was noted in the recent AAP statement, and the supine sleep position is now explicitly recommended for these infants.14 Given the increased survival of VLBW and ELBW infants, we hypothesize that the relative risk of SIDS for these infants may be even higher now than it was before the “Back to Sleep” campaign. We have used an explicit survival model to test the alternative possibility that higher relative incidence of SIDS is merely a statistical artifact that reflects declining VLBW and ELBW infant mortality from other causes. We examined the impact of birth weight and other SIDS risk factors for nearly 8 million singleton infants born within the 1989 and 1996 birth cohorts, using a competing risk analysis to account for increased survival among low birth weight infants.
METHODS Population Data are from the National Center for Health Statistics’ public-use birth certificate files. Data for the 1989 birth cohort are from the 1989 Linked Birth Infant Death File. This file contains matched individual-level data regarding almost all US births and, when applicable, infant death records. A linked birth—infant death file was unavailable for the 1996 US birth cohort. However, analogous data were contained in the 1996 and 1997 National Center for Health Statisitics Perinatal Mortality files.15 The 1996 file contained data on known births from 1996 and on infant deaths during the same 662
THE JOURNAL OF PEDIATRICS MAY 2001 year. The 1997 Perinatal Mortality data are needed for infants born during 1996 who died during 1997. Because infants who died during 1997 appear as live births in the 1996 data set and again as 1997 infant deaths, we followed the algorithm of Pollack, Lantz, and Frohna16 to find and remove redundant data entries. We matched infants across the data sets based on state of residence, birth weight in grams, gestation, maternal race, and maternal age at delivery to eliminate double-counting. Such matching is especially important for ELBW infants, whose high mortality rate would otherwise produce frequent double-counting. In particular, 3300 infants born during calendar year 1996 were recorded as 1997 deaths. We could match 2514 of these infants to a unique 1996 birth record. The remaining 786 were matched to multiple 1996 births. In such cases, the first matching live birth record was retained for the analysis. We exclude twins and higher plurality births because these infants raise issues outside the current analysis. We also exclude infants weighing <500 g at birth because the non-SIDS infant mortality rate is virtually 100% in this group and because infant and fetal deaths may be difficult to distinguish at such low birth weights. The resulting files include individuallevel birth certificate data for virtually all US births. These files provide many data items pertinent to SIDS risk. Maternal characteristics such as age, marital status, race/ethnicity, and years of completed education are available in these birth certificate data. Infant characteristics such as birth weight and gestational age are also provided. We chose to focus on birth weight, because this appears to be more reliably measured than gestational age in national birth certificate data.17 Data regarding maternal smoking during pregnancy—a key risk factor for SIDS—are also available for most US births. Unfortunately, several states
including New York and California do not provide smoking data in public-use data. We have included a dummy variable to mark births for which maternal smoking data are unavailable. No data are available to measure paternal smoking during pregnancy or postnatal tobacco exposure. In cases of infant death, these data files provide linked birth certificate and death certificate information for most infants. Supporting documentation indicates that more than 97% of infant deaths were successfully linked with accompanying birth certificate data.15,18 To address competing risks and changes in the timing of SIDS deaths, we have also analyzed age at infant death, as recorded on the infant death certificate.
Impact of Birth Weight An important feature of this analysis is examination of changing patterns of SIDS incidence among low birth weight infants (birth weight <2501 g), VLBW infants (birth weight <1501 g), and ELBW infants (birth weight <1001 g). The populations of VLBW and ELBW infants, although small, are important because these infants face high mortality risks from SIDS and other causes. Because VLBW and ELBW infants represent a small fraction of all births, large national data sets are required to examine SIDS risks. In the 1996 birth cohort, we identified fewer than 100 SIDS deaths among infants weighing between 500 and 1500 g. Failure to model improved neonatal survival might lead researchers to conclude that SIDS risks had increased for VLBW and ELBW infants, even if this pattern is entirely an artifact of declining competing risks. Over time, VLBW and ELBW infants are increasingly likely to survive into the postneonatal period of peak SIDS risk. Because of this increased exposure, the number of SIDS deaths among VLBW and ELBW infants might have increased between 1989 and 1996, even
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 if the probability of SIDS death per infant per day had actually declined over the same period. We use survival analysis to account for these competing risks.
Table I. Descriptive statistics
Multivariate Analysis Our main statistical approach is to estimate a survival model of SIDS risk. Within this framework, the dependent variable is an infant’s reported age at death, measured in days. We estimate a log-logistic specification, which provides the simplest and most readily interpreted extension of binary logistic regression to account for competing risk.19 In the absence of competing risk, the log-logistic survival model and the standard binary logistic model yield equivalent results. We also analyze an interaction specification to examine the impact of changing competing risks over the study period. In our competing risk approach, we follow standard methodology in considering non-SIDS mortality to be a form of non-informative right-censoring. Accounting for observed risk factors, an infant death from other causes on day T is considered to reveal no additional information about that infant’s probability of subsequent SIDS death, had he or she survived to day T + 1. It is possible that SIDS-vulnerable infants are also vulnerable to other life-threatening events. In principle, a more elaborate parametric model might explore this possibility.19,20 Unfortunately, current data do not allow reliable investigation of these effects. Within the log-logistic model, SEs accompanying adjusted odds ratios are adjusted by using the delta method to reflect SEs in the estimated scale parameters.
RESULTS Population SIDS incidence declined by almost half between the 2 study years, from 1.35 per 1000 live births in 1989 to 0.76 per 1000 in 1996 (Table I). The infant mortality rate in our study sam-
Race/ethnic group Non-Hispanic white African American Mexican Puerto Rican Other Other Hispanic Mother smoked during pregnancy Mother not married at birth Mother married at birth Mother completed <12 grades of schooling at birth Mother at least high school graduate at birth Mother college graduate at birth Birth weight ≥2501 g 1501-2500 g 1001-1500 g 500-1000 g
1996 Singleton mean (% of sample)
1989 SIDS incidence, singleton sample (per 1000)
1996 SIDS incidence, singleton sample (per 1000)
67.9 14.8 12.8 1.4 5.1 3.7 13.1
1.2 2.3 0.9 0.7 1.5 0.9 3.0
0.7 1.5 0.5 0.9 0.7 0.4 2.2
32.5 67.5 23.7
2.4 1.0 2.1
1.4 0.5 1.3
76.3
1.2
0.6
21.6
0.5
0.2
94.1 5.0 0.5 0.5
1.2 3.6 5.4 2.2
0.7 2.0 2.9 2.1
Percentages do not add to 100.0 because of rounding.
ple is slightly lower than reported rates for the overall population of US infants, reflecting the restricted group of singleton infants with birth weights that exceed 500 g. Overall SIDS incidence declined among African Americans. However, the relative risk slightly increased over the study period when compared with non-Hispanic whites. In the 1996 cohort, African Americans experienced 2.5 times the SIDS incidence of nonHispanic whites. Our point estimates suggest that most, though not all, of this racial disparity is attributable to group differences in observed risk factors such as birth weight, maternal educational attainment, and marital status. SIDS incidence was extremely low among Mexican Americans and Puerto Ricans.
Maternal smoking during pregnancy declined by almost one third between the 2 study years. Because maternal smoking is associated with more than a doubling of SIDS risk, declining maternal smoking is estimated to produce a significant decline in SIDS incidence. Overall infant mortality declined from 7.7 per 1000 to 5.36 per 1000, although the reported incidence of low birth weight, VLBW, and ELBW was virtually unchanged. Non-SIDS neonatal mortality significantly declined within these categories between 1989 and 1996. Within the 1996 cohort, some of these marginal survivors survived into the post-neonatal period when they subsequently died of SIDS. Partly because of such declining competing risks, the proportion of SIDS cases that involved ELBW infants in663
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS MAY 2001
Table II. Analysis results (standard errors adjusted by delta method)
Race/ethnic group Non-Hispanic white African American Mexican American Puerto Rican Other Other Hispanic Mother didn’t smoke during pregnancy Mother smoked during pregnancy No available smoking data Mother completed <12 grades of schooling at birth Mother at least high school graduate at birth Mother college graduate at birth Mother not married at birth Mother married at birth Birth weight ≥2501 g 1501-2500 g 1001-1500 g 500-1000 g 1989 1996 Smoking interacted with 1996 African American interacted with 1996 Birth weight 1501-2500 g interacted with 1996 Birth weight 1000-1501 g interacted with 1996 Birth weight 500-1000 g interacted with 1996 Scale parameter in survival model
AOR from baseline survivor model (95% CI)
AOR from interactive survivor model (95% CI)
AOR from binary logistic regression (95% CI)
1.000 1.242 [1.173, 1.315] 0.549 [0.498, 0.604] 0.532 [0.427, 0.664] 1.176 [1.056, 1.308] 0.572 [0.492, 0.665] 1.000 2.209 [2.081, 2.345] 1.035 [0.980, 1.095] 1.000 0.651 [0.620, 0.684] 0.550 [0.500, 0.605] 1.000 0.628 [0.596, 0.662]
1.000 1.147 [1.071, 1.229] 0.559 [0.508, 0.616] 0.533 [0.428, 0.665] 1.185 [1.065, 1.319] 0.577 [0.496, 0.671] 1.000 2.058 [1.919, 2.208] 1.025 [0.969, 1.084] 1.000 0.655 [0.623, 0.688] 0.550 [0.505, 0.610] 1.000 0.627 [0.595, 0.661]
1.000 1.148 [1.072, 1.229] 0.559 [0.508, 0.615] 0.533 [0.428, 0.664] 1.185 [1.065, 1.319] 0.577 [0.497, 0.670] 1.000 2.063 [1.930, 2.204] 1.025 [0.969, 1.084] 1.000 0.655 [0.624, 0.687] 0.555 [0.506, 0.610] 1.000 0.627 [0.596, 0.660]
1.000 2.157 [2.013, 2.312] 3.215 [2.750, 3.759] 1.865 [1.476, 2.357] 1.000 0.628 [0.598, 0.660] — — — — — 0.549 [0.537, 0.561]
1.000 2.186 [2.010, 2.379] 3.231 [2.669, 3.911] 1.346 [0.973, 1.863] 1.000 0.555 [0.519, 0.593] 1.226 [1.107, 1.358] 1.259 [1.134, 1.397] 0.965 [0.842, 1.105] 0.990 [0.719, 1.364] 2.366 [1.483, 3.777] 0.549 [0.537, 0.561]
1.000 2.192 [2.022, 2.377] 3.247 [2.692, 3.917] 1.348 [0.974, 1.865] 1.000 0.553 [0.518, 0.589] 1.223 [1.104, 1.354] 1.261 [1.136, 1.398] 0.947 [0.827, 1.086] 0.925 [0.671, 1.274] 1.593 [0.999, 2.540] —
The final row indicates the scale factor, an estimated parameter of the log-logistic model required to convert the coefficients into AOR form.
creased from 0.70% to 1.22% between 1989 and 1996. Uniformly across the first year of life, the SIDS hazard rate—the probability of SIDS per live infant per day—declined by approximately one third between the 2 years. The timing of SIDS deaths appeared similar across categories of race/ethnicity, maternal marital status, and education groups across both years. For infants born weighing >1500 g, hazard rates were highest between 30 and 120 days of life. Approximately 90% of SIDS deaths occurred during the first 6 months. 664
Multivariate Analysis Adjusted odds ratios are shown to match standard epidemiologic presentations (Table II). Adjusting for known risk factors, SIDS incidence declined by more than one third between the 2 years (AOR = 0.628). Given 1996 baseline values, this corresponds to more than 1600 averted deaths annually. Other coefficients are also pertinent in documenting changing SIDS risks. We find an AOR of 2.209 associated with maternal smoking. Between 1989 and 1996, the prevalence of maternal smoking during pregnancy declined by
5.1 percentage points. According to the methods used in previous research,21 these point estimates imply that declining maternal smoking is associated with a decline of 0.061 SIDS deaths per 1000 in the 1996 birth cohort, or approximately 250 cases throughout the United States. The second set of coefficients in Table II shows results from the baseline model, augmented with timeinteraction terms to capture changing patterns of competing risk. Results from the survival model are shown side by side with a standard binary lo-
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 gistic regression. As shown, AORs are virtually identical in the 2 models for all main effects. Relative risks associated with African American births (AOR = 1.259) and maternal smoking during pregnancy (AOR = 1.226) slightly increased between 1989 and 1996. The most dramatic change occurred for infants born weighing between 500 and 1000 g, whose SIDS risks are altered by declining birth weight–specific mortality between the 2 years. These infants had more than double the relative SIDS risk in the survival model (AOR = 2.366).
DISCUSSION In this study a competing-risk survival model was used to explore SIDS incidence in the 1989 and 1996 birth cohorts. Controlling for important confounders, we find a large decline in SIDS hazard rates, with a baseline AOR of 0.647. Among whites, declining SIDS incidence accounted for 30% of the observed decline in overall infant mortality rate between the 2 study years. ELBW infants were a larger proportion of SIDS deaths in 1996 (1.22%) than they were in 1989 (0.70%). Estimated SIDS hazard rates declined among low birth weight and VLBW infants between the 2 years. However, observed SIDS incidence within these groups was stable over time because of declining competing risks. Because VLBW and ELBW infants experience high mortality from other causes, high SIDS incidence within these birth categories is easily overlooked in analyses that do not consider competing risks. SIDS deaths also occur significantly later within the ELBW group than among infants of higher birth weights.8 However, SIDS hazard rates rapidly increase during subsequent months. These patterns are difficult to detect but are striking within a national ELBW sample. Controversy remains regarding proper sleep position for VLBW and
ELBW infants. Various authors have documented improved respiratory function in the prone position.22 Revised AAP recommendations indicate that premature or low birth weight infants should generally sleep supine, because the risks of prone sleep outweigh potential benefits.7,14* The SIDS Global Strategy Task Force (Developmental Physiology Working Group) has also emphasized the importance of supine sleeping for preterm infants, including patients in the neonatal intensive care unit, except for those with specific contraindications to supine sleeping.23 Ottolini et al24 found that parents who initially placed their children supine may switch to the prone position because of the perception that the infants sleep better. This perception is based in reality: prone sleep does seem to increase sleep duration and decrease arousals.25 Ironically, this decreased arousal may itself pose SIDS risks. Health care professionals should frequently discuss these issues during well-child visits in the first 6 months of life, especially for premature infants. Premature infants present especially important opportunities for preventive intervention because they face high SIDS risk and have frequent contact with the health care delivery system. A small but important portion of the decline in SIDS incidence is attributable to declining smoking prevalence among pregnant women. Maternal smoking before, during, and after pregnancy is an important and preventable threat to infant health. Smoking cessation interventions, though imperfect, have been shown to be cost-effective in many studies, including studies of SIDS risk.26 This study includes important limitations. First, maternal smoking data are based on self-report and may therefore understate true prevalence.27 The size and direction of the resulting bias *We thank an anonymous referee and Dr Carl E. Hunt for this reference.
depends on the specific pattern of misreporting.16 Misdiagnosis raises a second potential problem. SIDS is a diagnosis of exclusion—a diagnosis sometimes made under adverse conditions without autopsy or recommended death scene investigation.28 We know of no evidence that misdiagnosis patterns have significantly changed in recent years. Data from England and Australia suggest that the pattern of SIDS diagnoses was unaltered by the introduction of the “Back to Sleep” campaigns.29,30 Ambiguous diagnoses might be more important for the small population of VLBW and ELBW infants with a diagnosis of SIDS during the 2 study years. These infants face high risks of mortality from other causes. Changing treatment patterns or diagnostic methods involving VLBW and ELBW infants might have influenced reported patterns of SIDS deaths. VLBW and ELBW infants highlight the third limitation of the current study. We estimate a competing risk model in which non-SIDS infant deaths are regarded as non-informative of subsequent SIDS risk. Some 1996 VLBW and ELBW infants were “marginal survivors”; that is, they survived the first year of life as members of the 1996 cohort, but they would have died from other causes had they been born with the same characteristics in the 1989 cohort. If marginal survivors faced higher but unobserved SIDS risks when compared with their 1996 peers, the current analysis would understate the role of VLBW and ELBW and would overstate the ELBW and VLBW interaction terms. ELBW infants also experience health problems that require close monitoring and that may therefore reduce SIDS risk for important periods. Infants in the neonatal intensive care unit or other intensive care settings are less vulnerable to SIDS than are infants in noninstitutional settings. We believe that such monitoring accounts for the extremely low SIDS incidence in early in665
POLLACK AND FROHNA
fancy among ELBW infants. Our reported coefficients may therefore understate true SIDS risks facing the same infants sleeping at home, in day care, or in other non-institutional settings. A fourth limitation speaks to tradeoffs inherent in large-sample, birth certificate data sets. Important variables such as family income are unavailable. Birth certificate data address prenatal maternal smoking but do not address post-natal smoking, smoking by other household members, or illicit substance use. These data do not address infant sleep position, co-sleeping, and other epidemiologic risk factors. Some of these variables are included in more detailed surveys such as the Pregnancy Risk Assessment and Monitoring System (PRAMS).
Implications Despite these limitations, our findings highlight the remarkable success of SIDS prevention. Improved prenatal care timing, reduced prevalence of maternal smoking, and other observed factors have helped to reduce the number of SIDS deaths. Yet these changes account for only a small fraction of the observed decline. The “Back to Sleep” campaign’s simple, consistent, and clear message provides an important model for other public health education interventions. Our findings also highlight shortcomings of SIDS prevention. Relative risks for African Americans failed to decline, and indeed slightly increased, between 1989 and 1996. These results are consistent with prior studies indicating that African Americans are less likely than non-Hispanic whites or Hispanic/Latinos to sleep supine.4,24,31 Young mothers and mothers with fewer years of completed education appear especially likely to place their infants prone.3,10,32 Improved outreach and clinical interventions are critical to reducing SIDS incidence within these high-risk populations.31 Parents of VLBW and ELBW infants require appropriate SIDS prevention 666
THE JOURNAL OF PEDIATRICS MAY 2001 education, including information about the dangers of environmental tobacco exposure and prone infant sleep. Nurses and physicians should explicitly recommend that infants be placed supine to sleep, because this advice has been the most predictive of proper sleep position.5,6 When babies near the time of discharge, hospital staff must model proper sleep position, because most mothers who saw their infants placed prone at the hospital reported that they would place their children prone at home.31 Even infants initially placed in the back or side position are at increased risk for being placed prone during the first 6 months of life as a means to improve infant comfort.24 Parents of neonatal intensive care unit graduates may therefore require more guidance regarding infant comfort and sleep patterns after discharge. As neonatal survival improves within this medically fragile group, SIDS prevention becomes an increasingly important public health concern. We thank Gary Freed, Myron Wegman, and Jack Wheeler for useful comments.
REFERENCES 1. AAP Task Force on Infant Positioning and SIDS. Positioning and SIDS. Pediatrics 1992;89:1120-6. 2. Willinger M, Hoffman HJ, Harford RB. Infant sleep position and risk for sudden infant death syndrome: report of meeting held January 13 and 14, 1994, National Institutes of Health, Bethesda, MD. Pediatrics 1994;93:814-9. 3. Willinger M, Hoffman H, Wu K, Hou J, Kessler RC, Ward SL, et al. Factors associated with the transition to nonprone sleep positions of infants in the United States: The national infants sleep position study. JAMA 1998;280:329-35. 4. Centers for Disease Control and Prevention. Progress in reducing risky infant sleeping positions—13 states, 1996-1997. MMWR Morb Mortal Wkly Rep 1999;48:878-82. 5. Willinger M, James LS, Catz C. Defining the sudden infant death syndrome (SIDS): deliberations of an expert panel convened by the National Institute of Child Health and Human Development. Pediatr Pathol 1991;11:677-84.
6. Willinger M, Ko C, Hoffman H, Kessler R, Corwin M. Factors associated with caregivers’ choice of infant sleep position, 1994-1998: the National Infant Sleep Position Study. JAMA 2000;283:2135-42. 7. AAP Task Force on Infant Positioning and SIDS. Positioning and sudden infant death syndrome (SIDS): update. Pediatrics 1996;98:1216-8. 8. Malloy MH, Hoffman HJ. Prematurity, sudden infant death syndrome, and age of death. Pediatrics 1995;96:464-71. 9. Oyen N, Markestad T, Skjaerven R, Irgens LM, Helweg-Larsen K, Alm B, et al. Combined effects of sleeping position and prenatal risk factors in sudden infant death syndrome: The Nordic epidemiological SIDS study. Pediatrics 1997;100:613-21. 10. Bigger HR, Silvestri JM, Shott S, Weese-Mayer DE. Influence of increased survival in very low birth weight, low birth weight, and normal birth weight infants on the incidence of sudden infant death syndrome in the United States: 1985-1991. J Pediatr 1998;133:73-8. 11. Bigger HR, Silvestri JM, WeeseMayer DF. Sudden infant death syndrome (SIDS) death rates in normal birth weight (NBW > 2500g), low birth weight (LBW 1499-2500), and very-low birth weight (VLBW 5001499g) infants: comparison of rates before and after implementation of the Back to Sleep initiative [abstract]. Pediatr Res 2000;47:143A. 12. Bigger HR, Silvestri JM, Shott SB, Weese-Mayer DF. Changes in sudden infant death syndrome (SIDS) rates: gender comparisons by birth weight before and after implementation of the Back to Sleep initiative [abstract]. Pediatr Res 2000;47:143A. 13. Sowler B, Doyle LW, Morley CJ, Altmann A, Halliday J. Is sudden infant death syndrome still more common in very low birthweight infants in the 1990s? Med J Aust 1999;171:411-3. 14. AAP Task Force on Infant Positioning and SIDS. Changing concepts of sudden infant death syndrome: implications for infant sleeping environment and sleep position. Pediatrics 2000; 105:650-6. 15. National Center for Health Statistics. Perinatal mortality file supporting documentation. Hyattsville (MD): National Center for Health Statistics; 1999. 16. Pollack H, Lantz P, Frohna J. Maternal smoking and adverse birth outcomes among singletons and twins. Am J Public Health 2000;90:395-400.
POLLACK AND FROHNA
THE JOURNAL OF PEDIATRICS VOLUME 138, NUMBER 5 17. Alexander G, Tompkins M, Petersen D, Hulsey T, Mor J. Discordance between LMP-based and clinically estimated gestational age: implications for research, programs, and policy. Public Health Rep 1995;110:395-402. 18. National Center for Health Statistics. 1995 Perinatal mortality file supporting documentation, 1998. Hyattsville (MD): National Center for Health Statistics; 1998. 19. Allison P. Survival analysis using the SAS system. Cary (NC): SAS Institute; 1995. 20. Lancaster T. The econometric analysis of transition data. New York: Cambridge University Press; 1990. 21. Pollack HA. Sudden infant death syndrome, maternal smoking during pregnancy, and the cost-effectiveness of smoking cessation intervention. Am J Public Health 2001;91:432-6. 22. Martin RJ, DiFiore JM, Korenke CB, Randal H, Miller MJ, Brooks LJ. Vulnerability of respiratory control in
23.
24.
25.
26.
27.
healthy preterm infants placed supine. J Pediatr 1995;127:609-14. Hunt C. Expanded ‘Back-to-Sleep’ recommendations: hospital-based safe sleeping practices. J SIDS Infant Mortal 1997;2:223. Ottolini MC, Davis BE, Patel K, Sachs HC, Gershon NB, Moon RY. Prone infant sleeping despite the “Back to Sleep” campaign. Arch Pediatr Adolesc Med 1999;153:512-7. Kahn A, Groswasser J, Sottiaux M, Rebuffat E, Franco P, Dramaix M. Prone or supine body position and sleep characteristics in infants. Pediatrics 1993;91:1112-5. Lightwood J, Phibbs C, Glantz S. Short-term health and economic benefits of smoking cessation: low birth weight. Pediatrics 1999;104:1312-20. Floyd R, Rimer B, Giovino G, Mullen P, Sullivan S. Review of smoking in pregnancy: effects on pregnancy outcomes and cessation efforts. Ann Rev Public Health 1993;14:379-411.
28. Byard R, Krous H. Suffocation, shaking or sudden infant death syndrome: can we tell the difference? J Paediatr Child Health 1999;35:432-3. 29. Leach CEA, Blair PS, Fleming PJ, Smith IJ, Platt MW, Berry PJ, et al. Epidemiology of SIDS and explained sudden infant deaths. Pediatrics 1999; 104:e43. 30. Byard RW, Beal SM. Has changing diagnostic preference been responsible for the recent fall in incidence of sudden infant death syndrome in South Australia? J Paediatr Child Health 1995;31:197-9. 31. Brenner RA, Simons-Morton BG, Bhaskar B, Mehta N, Melnick VL, Revenis M, et al. Prevalence and predictors of the prone sleep position among inner-city infants. JAMA 1998; 280:341-6. 32. Lesko SM, Corwin MJ, Vezina RM, Hunt CE, Mandell F, McClain M, et al. Changes in sleep position during infancy. JAMA 1998;280:336-40.
667