- Email: [email protected]
Fetal deaths in the United States, 1997 vs 1991
American Journal of Obstetrics and Gynecology (2005) 193, 489–95
www.ajog.org
Fetal deaths in the United States, 1997 vs 1991 Hongbo Yuan, MD, PhD,a,* Robert W. Platt, PhD,a,b Lucie Morin, MD,c K. S. Joseph, MD, PhD,d Michael S. Kramer, MDa,b Departments of Epidemiology and Biostatistics,a Pediatrics,b and Obstetrics and Gynecology,c McGill University Faculty of Medicine, Montreal, Quebec, Canada; the Perinatal Epidemiology Research Unit, Department of Pediatrics and of Obstetrics and Gynecology, Dalhousie University, Halifax, Nova Scotia, Canadad Received for publication March 8, 2004; revised November 15, 2004; accepted December 2, 2004
KEY WORDS Fetal death Stillbirth Labor induction Post-term pregnancy Vital statistics
Objective: The purpose of this study was to examine the temporal change in fetal death risk in the US from 1991 to 1997, and to assess the extent to which changes in registration practices and labor induction have contributed to that change. Study design: This was a cohort study of all singleton pregnancies 20 to 43 weeks of gestation in 1991 and 1997 in the US. Results: From 1991 to 1997, the overall fetal death rate fell from 77.6 to 67.8 per 10,000 total births. However, fetal deaths at 20 to 22 weeks as a proportion of total births increased from 14.5 to 16.9 per 10,000. In a Cox regression analysis, the crude period effect (1997 vs 1991) at 40 to 43 weeks was 0.87 (95% CI 0.80-0.94), and remained virtually unchanged (HR 0.88, 95% CI 0.81-0.96) after adjustment for maternal sociodemographic, medical, and lifestyle risk factors. In ecologic (Poisson regression) analysis based on states as the unit of analysis, the crude period effect in non-Hispanic whites (RR 0.79, 95% CI 0.74-0.84) disappeared (RR 0.98, 95% CI 0.82-1.16) after adjusting for induction of labor. The effect of induction in blacks was limited to 42 to 43 weeks in those at high risk. Conclusion: Increased registration is probably responsible for an increase in fetal death risk at 20 to 22 weeks of gestation, whereas the increasing trend toward routine labor induction at and after term appears to have reduced the risk of fetal death, especially among whites. Ó 2005 Elsevier Inc. All rights reserved.
Since the 1980s, the decline in fetal mortality (stillbirth) has been much slower than the decline in infant mortality, and fetal death has now become the leading Drs Kramer, Platt, and Joseph are career investigators of the Canadian Institutes of Health Research, and Dr Joseph is also a Clinical Research Scholar of the Dalhousie University Faculty of Medicine. * Reprint requests: Dr Hongbo Yuan, Toronto Western Hospital, University Health Network (UHN), Edith Cavell 2-046, 399 Bathurst St, Toronto ON M5T 2S8. E-mail: [email protected] 0002-9378/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.ajog.2004.12.002
contributor to perinatal (fetal C early neonatal) mortality in the US.1,2 Few studies have examined temporal trends in fetal mortality,3-6 however, and even fewer have investigated the determinants of those trends. Induction of labor has often been used in the management of post-term pregnancies, which are at high risk for fetal death.7 However, it is unclear whether the recent decline in fetal mortality can be attributed to increased use of labor induction. In addition, the impact of changes in registration practices (eg, more complete reporting of fetal death at the borderline of viability) on
490
Yuan et al Table I Fetal deaths in 3 early GA categories as a proportion per 10,000 total births, 50 states and DC, all ethnicities
Figure 1 D.C.
Fetal death rates in whites and blacks, 50 states and
the temporal trend remains unknown. We studied changes in gestational age-specific fetal death risk in the US from 1991 to 1997, and assessed the extent to which the observed changes can be attributed to changes in registration practices and in induction of labor.
Material and methods Our study is based on the US live birth and fetal death files for the years 1991 and 1997.8 No data were available on obstetric procedures (ie, labor induction) before the 1989 revision of US vital statistic files, and were incomplete until 1991. The 1997 file was the most recent available when the study was initiated. Because multiple gestations are associated with a high risk of fetal death, we restricted our analyses to singleton births. All singleton fetal deaths R20 weeks of gestation are included, regardless of birth weight. The National Center for Health Statistics computes gestational age (GA) in completed weeks by subtracting the date of birth from the date of the onset of the last menstrual period (LMP). When the date of the LMP is incomplete, eg, missing day when there is a valid month and year, GA is imputed from the value of the preceding record with a complete LMP date, the same computed month of gestation, and the same 500-g birth weight interval.9 A clinical estimate of gestational age is used when information on the date of LMP is missing (eg, no data on month or day), invalid, or inconsistent with birth weight. For about 95% of births, the GA assignment on the vital records is based on the woman’s LMP date.10 To assess the impact of changes in registration practices, we included all ethnic groups in the 50 states and District of Columbia. We examined fetal deaths at 20 to 22, 23 to 25, and 26 to 28 weeks as a proportion of total births R20 weeks to assess the change in registration of fetal deaths at early GAs. Note that this proportion cannot be interpreted as the gestational age-specific fetal death risk. For the analysis of registration artifact, records with implausible birth weight-GA data were not deleted, and all ethnicities were included. GA-specific fetal death risk is defined as the number of fetal deaths per 10,000 fetuses at risk at each GA. For this computation, the numerator is the number of fetal deaths
GA (wk)
1991
1997
RR (1997 vs 1991)
95% CI
20-22 23-25 26-28 Total (20-43)
14.5 10.5 6.9 77.6
16.9 10.1 6.6 67.8
1.17 0.96 0.95 0.87
1.12-1.21 0.92-0.99 0.90-0.99 0.86-0.88
that occurred at each completed week of gestation, and the denominator is the number of undelivered fetuses at the beginning of that specific GA week, ie, the number of fetal deaths plus live births delivered at that GA or later GAs. We calculated GA-specific fetal death risk for nonHispanic whites and blacks separately. To reduce the impact of GA measurement errors on the estimate of fetal death risk, we used Alexander’s approach11 to delete implausible birth weight-GA data. Because previous studies based on US live birth and fetal death files have noted incomplete data on maternal medical conditions, lifestyle risk factors, and obstetric procedures,12-15 we examined missing values for each state individually for 3 item headings: maternal medical conditions, lifestyle risk factors, and obstetric procedures. Each heading contains a list of individual items in checkbox format, including ‘‘none’’; if all checkboxes were left blank, then the item heading was treated as missing. We included only those states with R80% complete data on each of these 3 headings. Nine states were excluded due to lack of data on maternal cigarette smoking (California, Hawaii, Indiana, Louisiana, Maryland, Massachusetts, New York, Oklahoma, and South Dakota); Illinois was excluded because of a high missing rate for induction of labor (51%); and Texas was excluded because of the high missing rate for maternal smoking (47%). These analyses were restricted to nonHispanic whites and blacks from the remaining 39 states and the District of Columbia. Subjects with missing values on any of the analyzed variables were deleted. We used Cox proportional hazards model16 to analyze fetal death as a censored time-to-event outcome. Based on the 39 states with reasonably complete covariates (see above), we first analyzed a crude model containing a single dummy variable representing the period effect between 1991 and 1997, using 1991 as the reference. Next, potential determinants were sequentially added to the crude model to determine the extent to which temporal changes in maternal sociodemographic characteristics (ie, age !20, 20-34, or R35 years, and educational attainment %12 vs O12 years, medical conditions [diabetes, chronic hypertension, or pregnancy-induced hypertension]), fetal gender, onset of prenatal care during the first trimester, and maternal cigarette smoking contributed to the period effect at the level of individual women.
Yuan et al
491
Table II Frequency distribution of maternal sociodemographic characteristics, medical, and lifestyle risk factors: live births and fetal deaths, 1991 vs 1997 1991 Characteristic (%) Maternal age (y)* %19 20-34 R35 Education (y) 0w8 9w12 13w15 R16 Missing Marital status* Unmarried Married Race* White Black Onset of prenatal care None 1st (trimester) 2nd or 3rd Missing Parity* 1 R2 Fetal gender* Male Female Cigarette smoking Yes No Missing Medical conditionsy Yes No Missing
1997
Live births n = 1,880,980
Fetal deaths n = 3171
Live births n = 1,806,406
Fetal deaths n = 2620
12.6 78.6 8.8
12.5 74.0 13.6
12.5 75.7 11.8
13.7 70.5 15.8
3.3 53.1 20.8 18.2 4.6
4.7 55.5 16.5 12.9 10.4
3.7 47.8 23.2 23.9 1.5
4.8 52.2 16.4 14.3 12.3
27.3 72.7
35.0 65.0
30.7 69.3
39.5 60.5
81.0 16.0
76.6 20.9
81.1 15.1
75.3 21.4
1.1 78.2 19.2 1.5
4.3 67.6 21.3 6.7
0.9 82.7 14.6 1.8
3.2 72.5 16.7 7.6
32.8 67.2
31.7 68.3
33.2 66.8
32.8 67.2
49.1 50.9
50.3 49.7
49.0 51.0
48.8 51.2
17.9 80.2 2.0
22.7 67.9 9.5
14.2 84.7 1.1
20.3 72.3 7.4
5.5 92.8 1.8
10.7 83.2 6.1
6.4 92.7 0.9
11.0 83.4 5.6
* No data were missing for maternal age, marital status, race, parity, or fetal gender. y Medical conditions include diabetes, chronic hypertension, and pregnancy-induced hypertension.
Unfortunately, no information is available from the birth registration on the indication for labor induction. Because labor induction is used both to deliver antepartum fetal deaths and also to hasten delivery in pregnancies with compromised fetuses (ie, to prevent fetal death), it is impossible to use individual-level data to examine the effect of labor induction on reducing the risk of fetal death. To avoid this ‘reverse causality’ problem, we used state-level ecologic analysis based on Poisson regression. The analysis is based on the frequency of induction in each state, but not on data about whether an individual delivery is induced or not. The Poisson regression results, thus, indicate whether an increase in induction rate between 2 time periods was
associated with a decrease in fetal death risk over the same period at the state level. To this end, the data were stratified according to state (49 states and DC, 50 categories), birth year (1997 vs 1991), and fetal gender, thereby resulting in 50 ! 2 ! 2 = 200 strata (Illinois was excluded because of the high missing rate for induction). The percent of induced deliveries was used for 1991 and 1997 in each state for each gender. The impact of induction of labor is represented by the change in point estimate for the period effect before and after adjustment for labor induction. The ecologic approach was repeated after stratification by low- vs high-risk pregnancy and in whites and blacks separately. Low-risk pregnancy is defined as
492 Table III
Yuan et al Crude fetal death risk by maternal sociodemographic characteristics, medical, and lifestyle risk factors, 1991 and 1997 1991 Risk/10,000
Risk factors (%) Maternal age (y) %19 20-34 R35 Education (y) %8 9-12 13-15 R16 Marital status Unmarried Married Race Black White Onset of prenatal care 1st (trimester) 2nd or 3rd Parity 1 R2 Fetal gender Male Female Cigarette smoking Yes No Diabetes Yes No Chronic hypertension Yes No PIH Yes No
1997 RR
95% CI
Risk/10,000
RR
95% CI
31.3 26.0 39.6
1.21 1.00 1.52
1.12-1.30 (reference) 1.40-1.65
30.7 23.0 32.4
1.33 1.00 1.41
1.23-1.44 (reference) 1.30-1,53
35.1 31.9 22.9 19.8
1.77 1.61 1.16 1.00
1.53-2.03 1.49-1.75 1.05-1.24 (reference)
32.6 30.3 21.3 16.6
1.97 1.83 1.28 1.00
1.70-2.25 1.68-1.98 1.16-1.38 (reference)
33.4 21.5
1.55 1.00
1.47-1.65 (reference)
31.3 20.4
1.53 1.00
1.44-1.62 (reference)
41.3 24.9
1.66 1.00
1.56-1.77 (reference)
38.2 22.5
1.70 1.00
1.59-1.81 (reference)
25.6 36.4
0.7 1.00
0.66-0.75 (reference)
23.2 34.9
0.66 1.00
0.62-0.71 (reference)
28.2 27.7
1.02 1.00
0.96-1.08 (reference)
25.5 24.9
1.02 1.00
0.97-1.09 (reference)
27.6 28.2
0.98 1.00
0.93-1.03 (reference)
24.3 25.9
0.94 1.00
0.89-0.99 (reference)
36.8 25.8
1.43 1.00
1.34-1.52 (reference)
39.2 22.6
1.73 1.00
1.62-1.85 (reference)
56.3 27.2
2.07 1.00
1.83-2.35 (reference)
51.0 24.4
2.09 1.00
1.85-2.37 (reference)
114.1 27.2
4.18 1.00
3.56-4.92 (reference)
82.1 24.6
3.33 1.00
2.78-4.00 (reference)
54.0 27.1
1.99 1.00
1.78-2.24 (reference)
37.2 24.6
1.51 1.00
1.34-1.71 (reference)
RR, Risk ratio; CI, confidence interval.
maternal age 20 to 34 years and the absence of diabetes, chronic hypertension, and pregnancy-induced hypertension; high-risk pregnancy is defined as maternal age !20 or R35 years and/or the presence of any of the above chronic conditions. All statistical analyses were carried out using SAS-PC for Windows, version 8.0 (SAS Institute, Inc, Cary, NC).
Results For both whites and blacks, the fetal death risk in 1991 and 1997 from 20 to 43 weeks is shown for the 50 states (plus DC) in Figure 1. From 1991 to 1997, the fetal death risk for whites and blacks combined in the 50 states and
DC increased by 18% at 20 to 22 weeks of gestation from 10.6 to 12.5 per 10,000 ongoing pregnancies (RR 1.18, 95% CI 1.13-1.23), whereas from 40 to 43 weeks, it decreased by 22% from 12.2 to 9.5 per 10,000 ongoing pregnancies (RR 0.78, 95% CI 0.73-0.83). For whites, the degree of reduction in fetal death risk increased from 40 to 43 weeks, while in blacks the decrease was only observed at 42 and 43 weeks. For both whites and blacks, the increase in fetal death risk at early GAs was limited to 20 to 22 weeks. Note that in calculating the fetal death risk, a substantial proportion of fetal death records were deleted based on Alexander’s approach, especially at 20 to 27 weeks (21.1% for 1991 and 13.1% for 1997). Table I shows the proportion of fetal deaths at 20 to 22, 23 to 25, and 26 to 28 weeks relative to total births
Yuan et al (at all GAs) for all ethnic groups in all 50 states and DC. The overall fetal death rate fell from 77.6 to 67.8 per 10,000 total births, a relative decrease of 14.6%. However, fetal deaths at 20 to 22 weeks as a proportion of total births increased from 14.5 to 16.9 per 10,000, a relative increase of 17%. No similar increase was observed in other early GA categories; instead, a slight but statistically significant decrease was seen. Note that the data shown in Table I cannot be interpreted as GAspecific fetal death risk because the denominator is total births, not all ongoing pregnancies at risk. Table II summarizes the maternal sociodemographic, medical, and lifestyle risk factors for 1991 and 1997, while Table III shows the crude fetal death risk according to these risk factors in the 2 study years, based on the 39 states and DC with reasonably complete data on the risk factors (see Material and methods). A statistically significantly higher risk of fetal death occurred among mothers who were !20 or R35 years of age, unmarried, or of black race. Compared with the reference group (R16 years), mothers who had %15 years of education were at higher risk of fetal death; the increased risk was more evident for mothers with %12 years of education. A monotonic trend was observed in fetal death risk by educational attainment: the lower the educational level, the higher the risk. Women who began prenatal care during the first trimester were at significantly lower risk for fetal death compared with those who began in the second or third trimester. Maternal cigarette smoking during pregnancy increased the risk of fetal death. Maternal diseases such as diabetes, chronic hypertension, and PIH showed strong associations with fetal death. Induction of labor increased sharply over the 6-year period in the US. The proportion of total live births delivered by induction increased from 13% to 22% (a 66% relative increase) in whites, and from 8.4% to 15% (a 79% relative increase) in blacks. In contrast, the cesarean section rate remained nearly unchanged (21% in 1991; 22% in 1997). Over time, the black-white gap in labor induction remained virtually unchanged; the induction rate for blacks was about 50% lower than for whites. Moreover, a significant increase rate of labor induction (per 1000 fetuses at risk) was observed with advancing GA in both racial groups (Figure 2). For both whites and blacks in 1991, women who delivered at 42 weeks had the highest rate of labor induction, whereas in 1997, the highest rate was observed at 41 weeks. The Cox proportional hazards model was used to analyze the crude and sequentially adjusted period effect (1997 vs 1991) in the 39 states (plus DC) with reasonably complete data on the studied covariates. At 40 to 43 weeks, the crude period effect (hazard ratio [HR], 1997 vs 1991) was 0.87 (95% CI 0.80-0.94), indicating a 13% reduction in fetal death risk. After sequentially adjusting for maternal age, race, fetal gender, maternal educa-
493
Figure 2 Induction of labor at each completed week of gestation.
tional attainment, first trimester onset of prenatal care, marital status, parity, cigarette smoking, diabetes, chronic hypertension, and PIH, the period effect remained virtually unchanged (HR 0.88, 95% CI 0.81-0.96), indicating that neither the individual nor combined effects of the temporal changes in these covariates could explain the observed decrease in fetal death risk at 40 to 43 weeks. Table IV summarizes the results of the Poisson regression analysis of the state-level (ecologic) effect of induction of labor on risk of fetal death at 40 to 43 weeks in whites and blacks (49 states and DC). Among whites, the crude period effect RR (1997 vs 1991) was 0.79 (95% CI 0.74-0.84). After adjusting for induction of labor, however, the period effect entirely disappeared (RR 0.98, 95% CI 0.82-1.16), suggesting that induction was responsible for the decrease. For white mothers, these results were similar among both risk strata. Among blacks, the crude period effect was 0.76 (95% CI 0.67-0.87), and actually became slightly stronger (RR 0.67, 95% CI 0.50-0.88) after adjusting for induction of labor. In both racial/ethnic groups, CIs widened after adjustment, as expected from the collinearity between year and induction at the state level (collinearity indicates the extent to which 2 terms in a linear statistical model are highly correlated). Because the reduction in fetal death risk in blacks was observed only at 42 and 43 weeks, we carried out a separate Poisson regression analysis restricted to these gestational ages. The results show that the crude period effect of 0.69 (95% CI 0.56-0.84) changed little (RR 0.69, 95% CI 0.51-0.94) after adjusting for induction of labor. Similar results were observed among black mothers at low risk (crude RR 0.58, 95% CI 0.43-0.79; adjusted RR 0.58, 95% CI 0.39-0.88). For black mothers at high risk, however, the period effect at 42 to 43 weeks fell from RR 0.81 (95% CI 0.58-1.13) before adjustment to RR 1.06 (95% CI 0.64-1.73) after adjustment. In white mothers at low risk, the induction effect was 0.96 (95% CI 0.94-0.98), representing a 4% reduction in fetal death risk for every 1% of increase in induction rate at each state. For black mothers, however, no such
494
Yuan et al
Table IV Poisson regression model for period effect (and 95% CI) on fetal death at 40 to 43 weeks in whites and blacks, before and after adjustment for induction of labor, 49 states and DC, 1997 versus 1991 Whites All pregnancies Low-risk pregnancies* High-risk pregnanciesy
Blacks
Crude RR
Adjusted RR
Crude RR
Adjusted RR
0.79 (0.74-0.84) 0.82 (0.75-0.90) 0.78 (0.70-0.87)
0.98 (0.82-1.16) 1.19 (0.95-1.49) 0.96 (0.71-1.27)
0.76 (0.67-0.87) 0.70 (0.58-0.85) 0.84 (0.68-1.03)
0.67 (0.50-0.88) 0.63 (0.44-0.93) 0.77 (0.49-1.17)
RR, Rate ratio (reference: 1991). * Low-risk pregnancy is defined as maternal age 20 to 34 years and absence of diabetes, chronic hypertension, and pregnancy-induced hypertension. y High-risk pregnancy is defined as maternal age !20 or 35 years and/or presence of 1 or more of the above chronic conditions.
a protective effect was observed (RR 1.01, 95% CI 0.971.06) in either risk stratum.
Comment Fetal deaths at 20 to 22 weeks increased relative to total births (at all GAs) between 1991 and 1997, although no similar increase was found at other early GA categories (23-25 or 26-28 weeks). A similar increase in the proportion of fetal deaths !500 g (relative to total births) was documented in the state of Alabama from 1974 to 1994, and in Canada from 1985 to 1995 (note: birth weight of 500 g approximately corresponds to a GA of 22 weeks), but was not observed in other lowbirth-weight categories (500-999 g or 1000-1499 g).16-19 It is unlikely that extrinsic risk factors (eg, exposure to environmental toxins) can account for this increase because the hazardous effect of such exposure should not be limited to 20 to 22 weeks. As did previous investigators,17-20 we attribute our finding of an increased risk in fetal death at 20 to 22 weeks to registration artifact, ie, more complete reporting of fetal deaths close to the 20-week cutoff. The marked decrease in fetal death risk at R40 weeks in whites appears to be the result of increased labor induction. We not only observed a dramatic increase in induction of labor over time, but the timing of induction appeared at earlier gestational ages in 1997 than in 1991. Given that fetal death risk increases sharply as gestational age advances to 43 weeks (see Figure 1), routine labor induction at or before 41 weeks may have enabled the induced fetuses to avoid the high-risk post-term period. The induction rate in blacks, however, might have been too low to produce a detectable overall impact; even after a substantial increase in 1997, the rate (15%) was similar to the level observed in whites (13%) in 1991. Our inference about the impact of labor induction is strengthened by the fact that the observed reduction in fetal death risk cannot be explained by contemporaneous changes in maternal sociodemographic characteristics, cigarette smoking, maternal medical conditions, or early prenatal care. Similarly, the marked decrease in fetal death risk at R40 weeks
cannot be attributed to more frequent delivery by cesarean section because the latter did not increase (22% in 1991 vs 21% in 1997) over the 6-year period. It is of interest that the protective effect of labor induction was observed among white pregnancies either at high or low risk for adverse pregnancy outcomes. In modern obstetric practice, high-risk pregnancies are under intense monitoring and special care. Once maternal or fetal complications are recognized, induction or (if the complications are emergent) cesarean section is often used. It is therefore not surprising to see a protective effect of labor induction among high-risk pregnancies. Our results also suggest a benefit among pregnancies apparently at low risk, at least in whites. This finding is consistent with those of a recent Canadian study, which reported that an increasing rate of induction of labor among post-term pregnancies was accompanied by a significant reduction in fetal deaths from 1980 to 1995.21 It is also consistent with the results of the Cochrane systematic review of 19 randomized trials assessing the effect routine vs selective induction of labor post-term on perinatal mortality (fetal deaths C early neonatal deaths).22 It is unlikely that the marked decrease in fetal deaths at R40 weeks can be attributed to the slight increase in use of ultrasound (which increased from 62% to 68% in whites and 54% to 62% in blacks). Improved dating of pregnancies with early ultrasound can reduce the rate of post-term pregnancies and, thus, affect the GA-specific risk of fetal death at post-term GAs, but the vast majority of NCHS’s gestational age estimates were based on the LMP. The increased use of electronic fetal monitoring (EFM) or stimulation of labor might also help reduce fetal death. However, given that EFM is primarily (and stimulation of labor, exclusively) applied during labor, such an effect would be limited to intrapartum fetal deaths, which represents only a small proportion of total fetal deaths (10-15%), too small to account for the marked decrease observed at R40 weeks. Cesarean section can theoretically prevent some intrapartum and antepartum fetal deaths but did not increase between 1991 and 1997. The quality of US vital statistics data has often been questioned, especially the completeness of data on
Yuan et al obstetric procedures and medical and lifestyle risk factors.12-15 Of particular concern for our study is the possible under-reporting of induction of labor. Missing data would indeed be a serious problem if the analysis was based on individual women, but our ecologic analysis was based on the percentage of all deliveries in each state. Thus, the impact of under-reporting should not be large unless the observed increase in labor induction was an artifact of better reporting. During the study period, the rate of amniocentesis remained nearly unchanged (31.7% in 1991 vs 30.7% in 1997). This procedure should have also been subject to a similar increase if a substantial improvement occurred in reporting. Moreover, if the temporal increase in labor induction were merely reporting artifact, no association should have been observed between states with increased induction rates and those with lower fetal death risk. In other words, the period effect for the fetal death hazard should have remained unchanged after adjustment. A potential problem of the ecologic approach, however, is the so-called ‘ecologic fallacy.’ The observed effect of induction at the state level may differ from its effect in individual women because the state-level ‘induction rate’ cannot specify which women did or did not receive labor induction, nor whether women whose labor was induced did or did not experience a fetal death. Finally, the GA measurement in this study is based primarily on the date of the last menstrual period, which is subject to error.11 In fact, after deleting inappropriate birth weight-GA records (using Alexander’s approach), the fetal death hazard at 20 to 25 decreased substantially; had we not deleted these errors, the hazard at early GAs would have been overestimated. Unfortunately, records with imputed gestational age cannot be identified (and thus, excluded from the analysis); nonetheless, their small number10 would be highly unlikely to affect our results and conclusions. We conclude that the increase in fetal death risk at 20 to 22 weeks between 1991 and 1997 in the US was probably due to more complete reporting of fetal deaths, but attribute the decrease at 40 to 43 weeks in whites to the increased use of labor induction. No statistically significant effect of induction of labor was observed in US blacks, except at 42 to 43 weeks for those at high risk of fetal death. Although investigation with individual-level data on indication for induction of labor would help strengthen our inference, the recent increase in routine induction at and after term appears to have succeeded in preventing many late fetal deaths.
References 1. National Vital Statistics Reports. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; 1999.
495 2. Gaudino JA, Hoyert DL, MacDorman MF, et al. Fetal deaths. In: Wilcox LS, Marks JS, editors. From data to action: CDC’s public health surveillance for women, infants, and children. Atlanta: U.S. Department of Health and Human Services; 1994. 3. Fretts RC, Boyd ME, Usher RH, Usher HA. The changing pattern of fetal death, 1961-1988. Obstet Gynecol 1992;79:35-9. 4. Ahlenius I, Thomassen P. The changing panorama of late fetal death in Sweden between 1984 and 1991. Acta Obstet Gynecol Scand 1999;78:408-14. 5. Hsieh H, Lee K, Khoshnood B, Herschel M. Fetal death rate in the United States, 1979-1990: trend and racial disparity. Obstet Gynecol 1997;89:33-9. 6. Goldenberg RL, Foster JM, Cutter GR, Nelson KG. Fetal deaths in Alabama, 1974-1983: a birth weight-specific analysis. Obstet Gynecol 1987;70:831-40. 7. Yudkin PL, Wood L, Redman CW. Risk of unexplained stillbirth at different gestational ages. Lancet 1987;I:1192-4. 8. National Center for Health Statistics. 1991 and 1997 linked birth/ infant death data set. NCHS CD-ROM series 20 no. 5, 7, and 15. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention. 9. National Center for Health Statistics. Instruction manual, computer edits for natality data, part 12. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; 1991. p. 34-6. 10. Demissie K, Rhoads GG, Ananth CV, Alexander GR, Kramer MS, Kogan MD, Joseph KS. Trends in preterm birth and neonatal mortality among blacks and whites in the United States from 1989 to 1997. Am J Epidemiol 2001;154:307-15. 11. Alexander GR, Himes JH, Kaufman RB, Mor J, Kogan M. A United States national reference for fetal growth. Obstet Gynecol 1996;87:163-8. 12. Piper JM, Mitchel EF, Snowden M, Hall C, Adams M, Taylor P. Validation of 1989 Tennessee birth certificates using maternal and newborn hospital records. Am J Epidemiol 1993;137:758-68. 13. Parrish KM, Holt AL. Variations in accuracy of obstetric procedures and diagnoses on birth records in Washington State, 1989. Am J Epidemiol 1993;138:119-27. 14. Buescher PA, Taylor KP, Davis MH. The quality of the new birth certificate data: a validation study in North Carolina. Am J Public Health 1993;83:1163-5. 15. Watkins ML, Edmonds L, McClean A. The surveillance of birth defects: the usefulness of the revised US standard birth certificate. Am J Public Health 1996;86:731-4. 16. Cox DR. Regression models and life tables. J R Stat Soc [B] 1972;34:187-220. 17. Goldenberg RL, Cutter GR, Nelson KG, Foster J. Effects of very low birth weight on fetal and neonatal mortality rates in Alabama. Public Health Rep 1989;104:487-92. 18. Phelan ST, Goldenberg RL, Alexander GR, Cliver SP. Perinatal mortality and its relationship to the reporting of low-birthweight infants. Am J Public Health 1998;88:1236-9. 19. Joseph KS, Allen A, Kramer MS, Cyr M, Fair M. Fetal-Infant Mortality Study Group of the Canadian Perinatal Surveillance System. Changes in the registration of stillbirths !500 g in Canada, 1985-95. Pediatr Perinat Epidemiol 1999;13:278-87. 20. Kramer MS, Platt RW, Yang H, Haglund B, Cnattingius S, Bergsjo P. Registration artifacts in international comparisons of infant mortality. Pediatr Perinat Epidemiol 2002;16:16-22. 21. Sue-A-Quan AK, Hannah ME, Cohen MM, Foster GA, Liston RM. Effect of labour induction on rates of stillbirth and cesarean section in post-term pregnancies. CMAJ 1999;160:1145-9. 22. Crowley P. Interventions for preventing or improving the outcome of delivery at or beyond term (Cochrane Review). In: The Cochrane Library, Issue 3 Chichester, UK: John Wiley & Sons, Ltd.; 2004.
www.ajog.org
Fetal deaths in the United States, 1997 vs 1991 Hongbo Yuan, MD, PhD,a,* Robert W. Platt, PhD,a,b Lucie Morin, MD,c K. S. Joseph, MD, PhD,d Michael S. Kramer, MDa,b Departments of Epidemiology and Biostatistics,a Pediatrics,b and Obstetrics and Gynecology,c McGill University Faculty of Medicine, Montreal, Quebec, Canada; the Perinatal Epidemiology Research Unit, Department of Pediatrics and of Obstetrics and Gynecology, Dalhousie University, Halifax, Nova Scotia, Canadad Received for publication March 8, 2004; revised November 15, 2004; accepted December 2, 2004
KEY WORDS Fetal death Stillbirth Labor induction Post-term pregnancy Vital statistics
Objective: The purpose of this study was to examine the temporal change in fetal death risk in the US from 1991 to 1997, and to assess the extent to which changes in registration practices and labor induction have contributed to that change. Study design: This was a cohort study of all singleton pregnancies 20 to 43 weeks of gestation in 1991 and 1997 in the US. Results: From 1991 to 1997, the overall fetal death rate fell from 77.6 to 67.8 per 10,000 total births. However, fetal deaths at 20 to 22 weeks as a proportion of total births increased from 14.5 to 16.9 per 10,000. In a Cox regression analysis, the crude period effect (1997 vs 1991) at 40 to 43 weeks was 0.87 (95% CI 0.80-0.94), and remained virtually unchanged (HR 0.88, 95% CI 0.81-0.96) after adjustment for maternal sociodemographic, medical, and lifestyle risk factors. In ecologic (Poisson regression) analysis based on states as the unit of analysis, the crude period effect in non-Hispanic whites (RR 0.79, 95% CI 0.74-0.84) disappeared (RR 0.98, 95% CI 0.82-1.16) after adjusting for induction of labor. The effect of induction in blacks was limited to 42 to 43 weeks in those at high risk. Conclusion: Increased registration is probably responsible for an increase in fetal death risk at 20 to 22 weeks of gestation, whereas the increasing trend toward routine labor induction at and after term appears to have reduced the risk of fetal death, especially among whites. Ó 2005 Elsevier Inc. All rights reserved.
Since the 1980s, the decline in fetal mortality (stillbirth) has been much slower than the decline in infant mortality, and fetal death has now become the leading Drs Kramer, Platt, and Joseph are career investigators of the Canadian Institutes of Health Research, and Dr Joseph is also a Clinical Research Scholar of the Dalhousie University Faculty of Medicine. * Reprint requests: Dr Hongbo Yuan, Toronto Western Hospital, University Health Network (UHN), Edith Cavell 2-046, 399 Bathurst St, Toronto ON M5T 2S8. E-mail: [email protected] 0002-9378/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.ajog.2004.12.002
contributor to perinatal (fetal C early neonatal) mortality in the US.1,2 Few studies have examined temporal trends in fetal mortality,3-6 however, and even fewer have investigated the determinants of those trends. Induction of labor has often been used in the management of post-term pregnancies, which are at high risk for fetal death.7 However, it is unclear whether the recent decline in fetal mortality can be attributed to increased use of labor induction. In addition, the impact of changes in registration practices (eg, more complete reporting of fetal death at the borderline of viability) on
490
Yuan et al Table I Fetal deaths in 3 early GA categories as a proportion per 10,000 total births, 50 states and DC, all ethnicities
Figure 1 D.C.
Fetal death rates in whites and blacks, 50 states and
the temporal trend remains unknown. We studied changes in gestational age-specific fetal death risk in the US from 1991 to 1997, and assessed the extent to which the observed changes can be attributed to changes in registration practices and in induction of labor.
Material and methods Our study is based on the US live birth and fetal death files for the years 1991 and 1997.8 No data were available on obstetric procedures (ie, labor induction) before the 1989 revision of US vital statistic files, and were incomplete until 1991. The 1997 file was the most recent available when the study was initiated. Because multiple gestations are associated with a high risk of fetal death, we restricted our analyses to singleton births. All singleton fetal deaths R20 weeks of gestation are included, regardless of birth weight. The National Center for Health Statistics computes gestational age (GA) in completed weeks by subtracting the date of birth from the date of the onset of the last menstrual period (LMP). When the date of the LMP is incomplete, eg, missing day when there is a valid month and year, GA is imputed from the value of the preceding record with a complete LMP date, the same computed month of gestation, and the same 500-g birth weight interval.9 A clinical estimate of gestational age is used when information on the date of LMP is missing (eg, no data on month or day), invalid, or inconsistent with birth weight. For about 95% of births, the GA assignment on the vital records is based on the woman’s LMP date.10 To assess the impact of changes in registration practices, we included all ethnic groups in the 50 states and District of Columbia. We examined fetal deaths at 20 to 22, 23 to 25, and 26 to 28 weeks as a proportion of total births R20 weeks to assess the change in registration of fetal deaths at early GAs. Note that this proportion cannot be interpreted as the gestational age-specific fetal death risk. For the analysis of registration artifact, records with implausible birth weight-GA data were not deleted, and all ethnicities were included. GA-specific fetal death risk is defined as the number of fetal deaths per 10,000 fetuses at risk at each GA. For this computation, the numerator is the number of fetal deaths
GA (wk)
1991
1997
RR (1997 vs 1991)
95% CI
20-22 23-25 26-28 Total (20-43)
14.5 10.5 6.9 77.6
16.9 10.1 6.6 67.8
1.17 0.96 0.95 0.87
1.12-1.21 0.92-0.99 0.90-0.99 0.86-0.88
that occurred at each completed week of gestation, and the denominator is the number of undelivered fetuses at the beginning of that specific GA week, ie, the number of fetal deaths plus live births delivered at that GA or later GAs. We calculated GA-specific fetal death risk for nonHispanic whites and blacks separately. To reduce the impact of GA measurement errors on the estimate of fetal death risk, we used Alexander’s approach11 to delete implausible birth weight-GA data. Because previous studies based on US live birth and fetal death files have noted incomplete data on maternal medical conditions, lifestyle risk factors, and obstetric procedures,12-15 we examined missing values for each state individually for 3 item headings: maternal medical conditions, lifestyle risk factors, and obstetric procedures. Each heading contains a list of individual items in checkbox format, including ‘‘none’’; if all checkboxes were left blank, then the item heading was treated as missing. We included only those states with R80% complete data on each of these 3 headings. Nine states were excluded due to lack of data on maternal cigarette smoking (California, Hawaii, Indiana, Louisiana, Maryland, Massachusetts, New York, Oklahoma, and South Dakota); Illinois was excluded because of a high missing rate for induction of labor (51%); and Texas was excluded because of the high missing rate for maternal smoking (47%). These analyses were restricted to nonHispanic whites and blacks from the remaining 39 states and the District of Columbia. Subjects with missing values on any of the analyzed variables were deleted. We used Cox proportional hazards model16 to analyze fetal death as a censored time-to-event outcome. Based on the 39 states with reasonably complete covariates (see above), we first analyzed a crude model containing a single dummy variable representing the period effect between 1991 and 1997, using 1991 as the reference. Next, potential determinants were sequentially added to the crude model to determine the extent to which temporal changes in maternal sociodemographic characteristics (ie, age !20, 20-34, or R35 years, and educational attainment %12 vs O12 years, medical conditions [diabetes, chronic hypertension, or pregnancy-induced hypertension]), fetal gender, onset of prenatal care during the first trimester, and maternal cigarette smoking contributed to the period effect at the level of individual women.
Yuan et al
491
Table II Frequency distribution of maternal sociodemographic characteristics, medical, and lifestyle risk factors: live births and fetal deaths, 1991 vs 1997 1991 Characteristic (%) Maternal age (y)* %19 20-34 R35 Education (y) 0w8 9w12 13w15 R16 Missing Marital status* Unmarried Married Race* White Black Onset of prenatal care None 1st (trimester) 2nd or 3rd Missing Parity* 1 R2 Fetal gender* Male Female Cigarette smoking Yes No Missing Medical conditionsy Yes No Missing
1997
Live births n = 1,880,980
Fetal deaths n = 3171
Live births n = 1,806,406
Fetal deaths n = 2620
12.6 78.6 8.8
12.5 74.0 13.6
12.5 75.7 11.8
13.7 70.5 15.8
3.3 53.1 20.8 18.2 4.6
4.7 55.5 16.5 12.9 10.4
3.7 47.8 23.2 23.9 1.5
4.8 52.2 16.4 14.3 12.3
27.3 72.7
35.0 65.0
30.7 69.3
39.5 60.5
81.0 16.0
76.6 20.9
81.1 15.1
75.3 21.4
1.1 78.2 19.2 1.5
4.3 67.6 21.3 6.7
0.9 82.7 14.6 1.8
3.2 72.5 16.7 7.6
32.8 67.2
31.7 68.3
33.2 66.8
32.8 67.2
49.1 50.9
50.3 49.7
49.0 51.0
48.8 51.2
17.9 80.2 2.0
22.7 67.9 9.5
14.2 84.7 1.1
20.3 72.3 7.4
5.5 92.8 1.8
10.7 83.2 6.1
6.4 92.7 0.9
11.0 83.4 5.6
* No data were missing for maternal age, marital status, race, parity, or fetal gender. y Medical conditions include diabetes, chronic hypertension, and pregnancy-induced hypertension.
Unfortunately, no information is available from the birth registration on the indication for labor induction. Because labor induction is used both to deliver antepartum fetal deaths and also to hasten delivery in pregnancies with compromised fetuses (ie, to prevent fetal death), it is impossible to use individual-level data to examine the effect of labor induction on reducing the risk of fetal death. To avoid this ‘reverse causality’ problem, we used state-level ecologic analysis based on Poisson regression. The analysis is based on the frequency of induction in each state, but not on data about whether an individual delivery is induced or not. The Poisson regression results, thus, indicate whether an increase in induction rate between 2 time periods was
associated with a decrease in fetal death risk over the same period at the state level. To this end, the data were stratified according to state (49 states and DC, 50 categories), birth year (1997 vs 1991), and fetal gender, thereby resulting in 50 ! 2 ! 2 = 200 strata (Illinois was excluded because of the high missing rate for induction). The percent of induced deliveries was used for 1991 and 1997 in each state for each gender. The impact of induction of labor is represented by the change in point estimate for the period effect before and after adjustment for labor induction. The ecologic approach was repeated after stratification by low- vs high-risk pregnancy and in whites and blacks separately. Low-risk pregnancy is defined as
492 Table III
Yuan et al Crude fetal death risk by maternal sociodemographic characteristics, medical, and lifestyle risk factors, 1991 and 1997 1991 Risk/10,000
Risk factors (%) Maternal age (y) %19 20-34 R35 Education (y) %8 9-12 13-15 R16 Marital status Unmarried Married Race Black White Onset of prenatal care 1st (trimester) 2nd or 3rd Parity 1 R2 Fetal gender Male Female Cigarette smoking Yes No Diabetes Yes No Chronic hypertension Yes No PIH Yes No
1997 RR
95% CI
Risk/10,000
RR
95% CI
31.3 26.0 39.6
1.21 1.00 1.52
1.12-1.30 (reference) 1.40-1.65
30.7 23.0 32.4
1.33 1.00 1.41
1.23-1.44 (reference) 1.30-1,53
35.1 31.9 22.9 19.8
1.77 1.61 1.16 1.00
1.53-2.03 1.49-1.75 1.05-1.24 (reference)
32.6 30.3 21.3 16.6
1.97 1.83 1.28 1.00
1.70-2.25 1.68-1.98 1.16-1.38 (reference)
33.4 21.5
1.55 1.00
1.47-1.65 (reference)
31.3 20.4
1.53 1.00
1.44-1.62 (reference)
41.3 24.9
1.66 1.00
1.56-1.77 (reference)
38.2 22.5
1.70 1.00
1.59-1.81 (reference)
25.6 36.4
0.7 1.00
0.66-0.75 (reference)
23.2 34.9
0.66 1.00
0.62-0.71 (reference)
28.2 27.7
1.02 1.00
0.96-1.08 (reference)
25.5 24.9
1.02 1.00
0.97-1.09 (reference)
27.6 28.2
0.98 1.00
0.93-1.03 (reference)
24.3 25.9
0.94 1.00
0.89-0.99 (reference)
36.8 25.8
1.43 1.00
1.34-1.52 (reference)
39.2 22.6
1.73 1.00
1.62-1.85 (reference)
56.3 27.2
2.07 1.00
1.83-2.35 (reference)
51.0 24.4
2.09 1.00
1.85-2.37 (reference)
114.1 27.2
4.18 1.00
3.56-4.92 (reference)
82.1 24.6
3.33 1.00
2.78-4.00 (reference)
54.0 27.1
1.99 1.00
1.78-2.24 (reference)
37.2 24.6
1.51 1.00
1.34-1.71 (reference)
RR, Risk ratio; CI, confidence interval.
maternal age 20 to 34 years and the absence of diabetes, chronic hypertension, and pregnancy-induced hypertension; high-risk pregnancy is defined as maternal age !20 or R35 years and/or the presence of any of the above chronic conditions. All statistical analyses were carried out using SAS-PC for Windows, version 8.0 (SAS Institute, Inc, Cary, NC).
Results For both whites and blacks, the fetal death risk in 1991 and 1997 from 20 to 43 weeks is shown for the 50 states (plus DC) in Figure 1. From 1991 to 1997, the fetal death risk for whites and blacks combined in the 50 states and
DC increased by 18% at 20 to 22 weeks of gestation from 10.6 to 12.5 per 10,000 ongoing pregnancies (RR 1.18, 95% CI 1.13-1.23), whereas from 40 to 43 weeks, it decreased by 22% from 12.2 to 9.5 per 10,000 ongoing pregnancies (RR 0.78, 95% CI 0.73-0.83). For whites, the degree of reduction in fetal death risk increased from 40 to 43 weeks, while in blacks the decrease was only observed at 42 and 43 weeks. For both whites and blacks, the increase in fetal death risk at early GAs was limited to 20 to 22 weeks. Note that in calculating the fetal death risk, a substantial proportion of fetal death records were deleted based on Alexander’s approach, especially at 20 to 27 weeks (21.1% for 1991 and 13.1% for 1997). Table I shows the proportion of fetal deaths at 20 to 22, 23 to 25, and 26 to 28 weeks relative to total births
Yuan et al (at all GAs) for all ethnic groups in all 50 states and DC. The overall fetal death rate fell from 77.6 to 67.8 per 10,000 total births, a relative decrease of 14.6%. However, fetal deaths at 20 to 22 weeks as a proportion of total births increased from 14.5 to 16.9 per 10,000, a relative increase of 17%. No similar increase was observed in other early GA categories; instead, a slight but statistically significant decrease was seen. Note that the data shown in Table I cannot be interpreted as GAspecific fetal death risk because the denominator is total births, not all ongoing pregnancies at risk. Table II summarizes the maternal sociodemographic, medical, and lifestyle risk factors for 1991 and 1997, while Table III shows the crude fetal death risk according to these risk factors in the 2 study years, based on the 39 states and DC with reasonably complete data on the risk factors (see Material and methods). A statistically significantly higher risk of fetal death occurred among mothers who were !20 or R35 years of age, unmarried, or of black race. Compared with the reference group (R16 years), mothers who had %15 years of education were at higher risk of fetal death; the increased risk was more evident for mothers with %12 years of education. A monotonic trend was observed in fetal death risk by educational attainment: the lower the educational level, the higher the risk. Women who began prenatal care during the first trimester were at significantly lower risk for fetal death compared with those who began in the second or third trimester. Maternal cigarette smoking during pregnancy increased the risk of fetal death. Maternal diseases such as diabetes, chronic hypertension, and PIH showed strong associations with fetal death. Induction of labor increased sharply over the 6-year period in the US. The proportion of total live births delivered by induction increased from 13% to 22% (a 66% relative increase) in whites, and from 8.4% to 15% (a 79% relative increase) in blacks. In contrast, the cesarean section rate remained nearly unchanged (21% in 1991; 22% in 1997). Over time, the black-white gap in labor induction remained virtually unchanged; the induction rate for blacks was about 50% lower than for whites. Moreover, a significant increase rate of labor induction (per 1000 fetuses at risk) was observed with advancing GA in both racial groups (Figure 2). For both whites and blacks in 1991, women who delivered at 42 weeks had the highest rate of labor induction, whereas in 1997, the highest rate was observed at 41 weeks. The Cox proportional hazards model was used to analyze the crude and sequentially adjusted period effect (1997 vs 1991) in the 39 states (plus DC) with reasonably complete data on the studied covariates. At 40 to 43 weeks, the crude period effect (hazard ratio [HR], 1997 vs 1991) was 0.87 (95% CI 0.80-0.94), indicating a 13% reduction in fetal death risk. After sequentially adjusting for maternal age, race, fetal gender, maternal educa-
493
Figure 2 Induction of labor at each completed week of gestation.
tional attainment, first trimester onset of prenatal care, marital status, parity, cigarette smoking, diabetes, chronic hypertension, and PIH, the period effect remained virtually unchanged (HR 0.88, 95% CI 0.81-0.96), indicating that neither the individual nor combined effects of the temporal changes in these covariates could explain the observed decrease in fetal death risk at 40 to 43 weeks. Table IV summarizes the results of the Poisson regression analysis of the state-level (ecologic) effect of induction of labor on risk of fetal death at 40 to 43 weeks in whites and blacks (49 states and DC). Among whites, the crude period effect RR (1997 vs 1991) was 0.79 (95% CI 0.74-0.84). After adjusting for induction of labor, however, the period effect entirely disappeared (RR 0.98, 95% CI 0.82-1.16), suggesting that induction was responsible for the decrease. For white mothers, these results were similar among both risk strata. Among blacks, the crude period effect was 0.76 (95% CI 0.67-0.87), and actually became slightly stronger (RR 0.67, 95% CI 0.50-0.88) after adjusting for induction of labor. In both racial/ethnic groups, CIs widened after adjustment, as expected from the collinearity between year and induction at the state level (collinearity indicates the extent to which 2 terms in a linear statistical model are highly correlated). Because the reduction in fetal death risk in blacks was observed only at 42 and 43 weeks, we carried out a separate Poisson regression analysis restricted to these gestational ages. The results show that the crude period effect of 0.69 (95% CI 0.56-0.84) changed little (RR 0.69, 95% CI 0.51-0.94) after adjusting for induction of labor. Similar results were observed among black mothers at low risk (crude RR 0.58, 95% CI 0.43-0.79; adjusted RR 0.58, 95% CI 0.39-0.88). For black mothers at high risk, however, the period effect at 42 to 43 weeks fell from RR 0.81 (95% CI 0.58-1.13) before adjustment to RR 1.06 (95% CI 0.64-1.73) after adjustment. In white mothers at low risk, the induction effect was 0.96 (95% CI 0.94-0.98), representing a 4% reduction in fetal death risk for every 1% of increase in induction rate at each state. For black mothers, however, no such
494
Yuan et al
Table IV Poisson regression model for period effect (and 95% CI) on fetal death at 40 to 43 weeks in whites and blacks, before and after adjustment for induction of labor, 49 states and DC, 1997 versus 1991 Whites All pregnancies Low-risk pregnancies* High-risk pregnanciesy
Blacks
Crude RR
Adjusted RR
Crude RR
Adjusted RR
0.79 (0.74-0.84) 0.82 (0.75-0.90) 0.78 (0.70-0.87)
0.98 (0.82-1.16) 1.19 (0.95-1.49) 0.96 (0.71-1.27)
0.76 (0.67-0.87) 0.70 (0.58-0.85) 0.84 (0.68-1.03)
0.67 (0.50-0.88) 0.63 (0.44-0.93) 0.77 (0.49-1.17)
RR, Rate ratio (reference: 1991). * Low-risk pregnancy is defined as maternal age 20 to 34 years and absence of diabetes, chronic hypertension, and pregnancy-induced hypertension. y High-risk pregnancy is defined as maternal age !20 or 35 years and/or presence of 1 or more of the above chronic conditions.
a protective effect was observed (RR 1.01, 95% CI 0.971.06) in either risk stratum.
Comment Fetal deaths at 20 to 22 weeks increased relative to total births (at all GAs) between 1991 and 1997, although no similar increase was found at other early GA categories (23-25 or 26-28 weeks). A similar increase in the proportion of fetal deaths !500 g (relative to total births) was documented in the state of Alabama from 1974 to 1994, and in Canada from 1985 to 1995 (note: birth weight of 500 g approximately corresponds to a GA of 22 weeks), but was not observed in other lowbirth-weight categories (500-999 g or 1000-1499 g).16-19 It is unlikely that extrinsic risk factors (eg, exposure to environmental toxins) can account for this increase because the hazardous effect of such exposure should not be limited to 20 to 22 weeks. As did previous investigators,17-20 we attribute our finding of an increased risk in fetal death at 20 to 22 weeks to registration artifact, ie, more complete reporting of fetal deaths close to the 20-week cutoff. The marked decrease in fetal death risk at R40 weeks in whites appears to be the result of increased labor induction. We not only observed a dramatic increase in induction of labor over time, but the timing of induction appeared at earlier gestational ages in 1997 than in 1991. Given that fetal death risk increases sharply as gestational age advances to 43 weeks (see Figure 1), routine labor induction at or before 41 weeks may have enabled the induced fetuses to avoid the high-risk post-term period. The induction rate in blacks, however, might have been too low to produce a detectable overall impact; even after a substantial increase in 1997, the rate (15%) was similar to the level observed in whites (13%) in 1991. Our inference about the impact of labor induction is strengthened by the fact that the observed reduction in fetal death risk cannot be explained by contemporaneous changes in maternal sociodemographic characteristics, cigarette smoking, maternal medical conditions, or early prenatal care. Similarly, the marked decrease in fetal death risk at R40 weeks
cannot be attributed to more frequent delivery by cesarean section because the latter did not increase (22% in 1991 vs 21% in 1997) over the 6-year period. It is of interest that the protective effect of labor induction was observed among white pregnancies either at high or low risk for adverse pregnancy outcomes. In modern obstetric practice, high-risk pregnancies are under intense monitoring and special care. Once maternal or fetal complications are recognized, induction or (if the complications are emergent) cesarean section is often used. It is therefore not surprising to see a protective effect of labor induction among high-risk pregnancies. Our results also suggest a benefit among pregnancies apparently at low risk, at least in whites. This finding is consistent with those of a recent Canadian study, which reported that an increasing rate of induction of labor among post-term pregnancies was accompanied by a significant reduction in fetal deaths from 1980 to 1995.21 It is also consistent with the results of the Cochrane systematic review of 19 randomized trials assessing the effect routine vs selective induction of labor post-term on perinatal mortality (fetal deaths C early neonatal deaths).22 It is unlikely that the marked decrease in fetal deaths at R40 weeks can be attributed to the slight increase in use of ultrasound (which increased from 62% to 68% in whites and 54% to 62% in blacks). Improved dating of pregnancies with early ultrasound can reduce the rate of post-term pregnancies and, thus, affect the GA-specific risk of fetal death at post-term GAs, but the vast majority of NCHS’s gestational age estimates were based on the LMP. The increased use of electronic fetal monitoring (EFM) or stimulation of labor might also help reduce fetal death. However, given that EFM is primarily (and stimulation of labor, exclusively) applied during labor, such an effect would be limited to intrapartum fetal deaths, which represents only a small proportion of total fetal deaths (10-15%), too small to account for the marked decrease observed at R40 weeks. Cesarean section can theoretically prevent some intrapartum and antepartum fetal deaths but did not increase between 1991 and 1997. The quality of US vital statistics data has often been questioned, especially the completeness of data on
Yuan et al obstetric procedures and medical and lifestyle risk factors.12-15 Of particular concern for our study is the possible under-reporting of induction of labor. Missing data would indeed be a serious problem if the analysis was based on individual women, but our ecologic analysis was based on the percentage of all deliveries in each state. Thus, the impact of under-reporting should not be large unless the observed increase in labor induction was an artifact of better reporting. During the study period, the rate of amniocentesis remained nearly unchanged (31.7% in 1991 vs 30.7% in 1997). This procedure should have also been subject to a similar increase if a substantial improvement occurred in reporting. Moreover, if the temporal increase in labor induction were merely reporting artifact, no association should have been observed between states with increased induction rates and those with lower fetal death risk. In other words, the period effect for the fetal death hazard should have remained unchanged after adjustment. A potential problem of the ecologic approach, however, is the so-called ‘ecologic fallacy.’ The observed effect of induction at the state level may differ from its effect in individual women because the state-level ‘induction rate’ cannot specify which women did or did not receive labor induction, nor whether women whose labor was induced did or did not experience a fetal death. Finally, the GA measurement in this study is based primarily on the date of the last menstrual period, which is subject to error.11 In fact, after deleting inappropriate birth weight-GA records (using Alexander’s approach), the fetal death hazard at 20 to 25 decreased substantially; had we not deleted these errors, the hazard at early GAs would have been overestimated. Unfortunately, records with imputed gestational age cannot be identified (and thus, excluded from the analysis); nonetheless, their small number10 would be highly unlikely to affect our results and conclusions. We conclude that the increase in fetal death risk at 20 to 22 weeks between 1991 and 1997 in the US was probably due to more complete reporting of fetal deaths, but attribute the decrease at 40 to 43 weeks in whites to the increased use of labor induction. No statistically significant effect of induction of labor was observed in US blacks, except at 42 to 43 weeks for those at high risk of fetal death. Although investigation with individual-level data on indication for induction of labor would help strengthen our inference, the recent increase in routine induction at and after term appears to have succeeded in preventing many late fetal deaths.
References 1. National Vital Statistics Reports. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; 1999.
495 2. Gaudino JA, Hoyert DL, MacDorman MF, et al. Fetal deaths. In: Wilcox LS, Marks JS, editors. From data to action: CDC’s public health surveillance for women, infants, and children. Atlanta: U.S. Department of Health and Human Services; 1994. 3. Fretts RC, Boyd ME, Usher RH, Usher HA. The changing pattern of fetal death, 1961-1988. Obstet Gynecol 1992;79:35-9. 4. Ahlenius I, Thomassen P. The changing panorama of late fetal death in Sweden between 1984 and 1991. Acta Obstet Gynecol Scand 1999;78:408-14. 5. Hsieh H, Lee K, Khoshnood B, Herschel M. Fetal death rate in the United States, 1979-1990: trend and racial disparity. Obstet Gynecol 1997;89:33-9. 6. Goldenberg RL, Foster JM, Cutter GR, Nelson KG. Fetal deaths in Alabama, 1974-1983: a birth weight-specific analysis. Obstet Gynecol 1987;70:831-40. 7. Yudkin PL, Wood L, Redman CW. Risk of unexplained stillbirth at different gestational ages. Lancet 1987;I:1192-4. 8. National Center for Health Statistics. 1991 and 1997 linked birth/ infant death data set. NCHS CD-ROM series 20 no. 5, 7, and 15. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention. 9. National Center for Health Statistics. Instruction manual, computer edits for natality data, part 12. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention; 1991. p. 34-6. 10. Demissie K, Rhoads GG, Ananth CV, Alexander GR, Kramer MS, Kogan MD, Joseph KS. Trends in preterm birth and neonatal mortality among blacks and whites in the United States from 1989 to 1997. Am J Epidemiol 2001;154:307-15. 11. Alexander GR, Himes JH, Kaufman RB, Mor J, Kogan M. A United States national reference for fetal growth. Obstet Gynecol 1996;87:163-8. 12. Piper JM, Mitchel EF, Snowden M, Hall C, Adams M, Taylor P. Validation of 1989 Tennessee birth certificates using maternal and newborn hospital records. Am J Epidemiol 1993;137:758-68. 13. Parrish KM, Holt AL. Variations in accuracy of obstetric procedures and diagnoses on birth records in Washington State, 1989. Am J Epidemiol 1993;138:119-27. 14. Buescher PA, Taylor KP, Davis MH. The quality of the new birth certificate data: a validation study in North Carolina. Am J Public Health 1993;83:1163-5. 15. Watkins ML, Edmonds L, McClean A. The surveillance of birth defects: the usefulness of the revised US standard birth certificate. Am J Public Health 1996;86:731-4. 16. Cox DR. Regression models and life tables. J R Stat Soc [B] 1972;34:187-220. 17. Goldenberg RL, Cutter GR, Nelson KG, Foster J. Effects of very low birth weight on fetal and neonatal mortality rates in Alabama. Public Health Rep 1989;104:487-92. 18. Phelan ST, Goldenberg RL, Alexander GR, Cliver SP. Perinatal mortality and its relationship to the reporting of low-birthweight infants. Am J Public Health 1998;88:1236-9. 19. Joseph KS, Allen A, Kramer MS, Cyr M, Fair M. Fetal-Infant Mortality Study Group of the Canadian Perinatal Surveillance System. Changes in the registration of stillbirths !500 g in Canada, 1985-95. Pediatr Perinat Epidemiol 1999;13:278-87. 20. Kramer MS, Platt RW, Yang H, Haglund B, Cnattingius S, Bergsjo P. Registration artifacts in international comparisons of infant mortality. Pediatr Perinat Epidemiol 2002;16:16-22. 21. Sue-A-Quan AK, Hannah ME, Cohen MM, Foster GA, Liston RM. Effect of labour induction on rates of stillbirth and cesarean section in post-term pregnancies. CMAJ 1999;160:1145-9. 22. Crowley P. Interventions for preventing or improving the outcome of delivery at or beyond term (Cochrane Review). In: The Cochrane Library, Issue 3 Chichester, UK: John Wiley & Sons, Ltd.; 2004.