A population-based computer record-linkage study of infant births and deaths in 1978 and 1979in eight Canadian provinces (Quebec and Newfound- land were ...
Risk factors for perinatal mortality in Canada John Silins,* BSc Robert M. Semenciw,t BSc Howard I. Morrison,t BSc Joan Lindsay,* BA Greg J. Sherman,t PhD Yang Mao,t PhD Donald T. Wigle,t MD, PhD, MPH A population-based computer record-linkage study of infant births and deaths in 1978 and 1979 in eight Canadian provinces (Quebec and Newfoundland were excluded) was undertaken to permit analysis of perinatal mortality in relation to maternal and infant characteristics. Perinatal mortality rates were significantly higher in nonurban than in urban areas (p < 0.05). A logistic regression model was used to assess the effects on perinatal mortality of variables reported on birth and stillbirth records. This model included length of gestation, infant's birth weight and sex, number of previous births and number of previous stillbirths as well as an interaction term for length of gestation and birth weight. For early-neonatal mortality, odds ratios over 8 were observed for birth weight less than 2500 g or gestation less than 35 weeks. About 75% of early-neonatal mortality was attributable to low birth weight or fetal immaturity. Greater emphasis should be placed on the prevention of low birth weight. On a entrepris dans huit provinces du Canada (a l'exclusion du Quebec et de Terre-Neuve) I'etude informatisee reliant les dossiers de naissance et de mortalite perinatale en 1978 et en 1979 afin de mettre celle-ci en rapport avec des donnees concernant la mere et l'enfant. La mortalite perinatale est plus forte en milieu non-urbain qu'en milieu urbain (p < 0,05). On s'est servi de la regression logistique pour connaitre l'incidence sur cette mortalite des parametres suivants colliges des dossiers de naissance et de mortinatalite: duree de la gestation, poids de naissance, sexe de l'enfant, et nombre de naissances et de mortinaissances anterieures; on From *the Vital Statistics and Disease Registries Section, Health Division, and tthe Non-Communicable Disease Division, Bureau of Epidemiology, Health Protection Branch, Department of National Health and Welfare, Ottawa
Reprint requests to: Mr. Robert M. Semenciw, Non-Communicable Disease Division, Bureau of Epidemiology, Department of National Health and Welfare, Tunney's Pasture, Ottawa, Ont. KIA OL2 1214
CAN MED ASSOC J, VOL. 133, DECEMBER 15,1985
y a joint une formule d'interaction des deux premiers de ces facteurs. La mortalite neonatale precoce est majoree au moins huit fois en presence d'un poids natal inferieur a 2500 g ou d'une duree de gestation de moins de 35 semaines; elle est attribuable pour les 75% des cas au faible poids de naissance ou a l'immaturite foetale. II faut mettre encore plus d'accent qu'on ne le fait deja sur l'importance de prevenir l'insuffisance ponderale a la naissance.
P) erinatal deaths, which include stillbirths and deaths of infants up to 7 days of age, accounted for 285 000 potential years of life lost before age 75 in Canada during 1982 (H. Smith, Bureau of Epidemiology, Health Protection Branch, Department of National Health and Welfare, Ottawa: unpublished data, 1984). Only cancer, cardiovascular diseases and violent deaths ranked higher. Risk factors of obvious public significance (i.e., modifiable causal factors) for perinatal mortality include maternal diet' and use of tobacco2,3 or alcohol.4 Other risk factors or "markers" include maternal age,5'6 parity,6 previous stillbirths,6 marital status,7 social class,7 infant's sex,6 birth weight" and length of gestation.8 Linkage of infant death registrations with the corresponding birth records has been done in both Great Britain and the United States as a special exercise and, since 1975, routinely in England and Wales. To investigate, with Canadian data, any associations that may exist between variables reported on birth certificates, stillbirth records and infant death certificates, a population-based record-linkage study was undertaken. The purpose of the study was to relate overall and cause-specific perinatal mortality rates to vital statistical data available on birth certificates. Methods In Canada information on births and deaths is recorded and accumulated by registrars of vital statistics in each province. Under federal-provin-
cial agreement, copies of registration documents and, when applicable, machine-readable abstracts are transmitted to Statistics Canada for compilation. Quebec and Newfoundland were excluded from the study because of lack of certain information on the birth records supplied: Quebec does not supply names in machine-readable form for either births or deaths, and the birth certificates from Newfoundland did not include the statistical variables of interest. The data we studied consisted of information on 3118 stillbirths (28 weeks' or greater gestation) and 510 425 live births that occurred during 1978 and 1979 and on 3076 corresponding deaths in the first week of life during 1978, 1979 and 1980.
Deaths of infants before 1 year of age were linked to live births by means of the generalized iterative record-linkage system of Statistics Canada.9 All but 3% of deaths in the first week of life were linked to the appropriate births. Provincial and territorial linkage rates were tabulated and are available on request. The maternal and infant variables used in this study were mother's province and county of residence, age and marital status, length of gestation (in weeks), number of previous births, number of previous stillbirths, number of infants born at the birth, infant's sex and birth weight, and, for deaths and stillbirths, the underlying cause of death, coded to ICD-810 for 1978 and ICD-9'1 for 1979. Perinatal, stillbirth and early-neonatal (first week of life) mortality rates were calculated for urban and nonurban areas. The rates were per 1000 total births for perinatal deaths and stillbirths and per 1000 live births for early-neonatal deaths. Urban rates were based on counties whose populations were classified as at least 85% urban in the 1976 census12 and that were at least partial components of census metropolitan areas. All other counties were considered nonurban. Cause-specific mortality rates were calculated for the 1979 birth cohort only. Births and deaths for 1978 were excluded because for some causes of death there was difficulty in converting from ICD-8 to ICD-9. As a result of the large number of observations, all variables were categorized, with the number of strata compressed as required. Attributable risks or the proportion of early-neonatal deaths attributable to various biologic factors were determined.13 Logistic regression analysis was used to assess the effects of risk factors. This method modelled the logarithm of the odds of death as a linear regression of the independent variables (including interaction terms, if necessary), with the regression coefficients providing estimates of the odds ratios for the independent variables.14,5 The resulting odds ratios were adjusted for the effects of certain other variables. For example, an odds ratio of 1.49 for the risk factor "unmarried" can be interpreted to mean that when adjustment is done for certain other risk factors, offspring of unmarried women have roughly a 50% increased risk of perinatal mortality. Odds ratios with 95% confi-
dence limits that included 1.0 were not statistically significant at the 5% level. Interaction occurs when the effect of one variable changes according to the level of another. In such cases odds ratios are given according to different levels of the interacting variable. For example, primipara status was found to be a risk factor for offspring of those aged 20 or older only, with a nonsignificant decrease in risk found for offspring of those under age 20. Primipara status interacted with maternal age; that is, the effect of primipara status was not consistent between age groups. Consequently odds ratios for primipara status are presented separately for offspring of women under age 20 and for those of women aged 20 or older. Data quality
Distributions of missing values for the variables maternal age, marital status, province and county of residence, length of gestation and birth weight were tabulated for live births and stillbirths (data available on request). Among live births, the proportion of records with any of these variables missing ranged from 0.03% in Ontario to 10.0% in the Northwest Territories. Results
The urban perinatal mortality rate was 11.6, significantly lower (p < 0.05) than the nonurban rate of 12.4. The lower rate was largely the result of lower early-neonatal mortality: the urban rate of 5.6 was significantly lower (p < 0.05) than the nonurban rate of 6.3. The stillbirth rates were comparable. Causes of perinatal mortality are presented in Table I. Of stillbirths 42.6% resulted from complications of the placenta, cord or membranes, while respiratory distress syndrome accounted for 15.6% of early-neonatal deaths. Congenital anomalies were reported as the cause of 11.2% of stillbirths and 25.9% of early-neonatal deaths. Birth-weight-specific early-neonatal mortality rates are presented in Table II. Mortality rates were inversely related to birth weight for all causes examined, except in the highest birth-weight class (4500 g or more). For the highest-risk birth weight (less than 1000 g) the leading cause of death was respiratory distress syndrome, while for the lowest-risk birth weight (3000 to 4499 g) the leading cause of death was intrauterine hypoxia or anoxia. The proportion of early-neonatal mortality attributable to birth weight less than 2500 g was 75%. The attributable risk percents for length of gestation less than 32 weeks and less than 38 weeks were 55 and 74 respectively. To obtain estimates for singleton births of the relation between the effects of birth weight and length of gestation on early-neonatal mortality, a CAN MED ASSOC J, VOL. 133, DECEMBER 15,1985
1215
logistic regression model was constructed in which deaths due to congenital anomalies and the corresponding births were excluded. This statistical model included length of gestation, infant's birth weight and sex, number of previous births and number of previous stillbirths plus an interaction term with 1 degree of freedom for birth weight and length of gestation. Odds ratios and case fatality rates by birth weight and length of gestation are presented in Table III; a weight of 3000 g or more and 38 weeks' or more gestation were the baselines. Estimated odds ratios were above 8 for birth '
bl
1216
t,
trh and
ea
iV- eui:i"
rti7-.vveight spect::
l
.
3
;
ta}it ..
t
y
CAN MED ASSOC J, VOL. 133, DECEMBER 15,1985
weight less than 2500 g or gestation less than 35 weeks. Case fatality rates greater than 10o were found for birth weight less than 1500 g and gestation less than 32 weeks. Perinatal mortality rates for singleton births by maternal age are presented in Fig. 1. Infants of women less than 20 years old or over 30 years old had significantly increased perinatal mortality rates (chi-square test of association). To assess the importance of biologic age per se, perinatal mortality rates of infants of unmarried adolescent (aged 15 to 19 years) primiparas a tes
-(974
!r
t'-
,,
lirE J
.,
were compared with those of infants of unmarried adult (20 to 34 years) primiparas. The perinatal mortality rate in the former group, 15.2, was not significantly different from that in the latter group, 15.6. Logistic regression odds ratios for risk factors other than birth weight and length of gestation are presented in Table IV. These two variables were not included as potential confounding variables (except in the analysis of the effects of sex) because they are intermediate factors in the causal chain, and controlling for them would be expected to decrease the odds ratio. Risk factors found to be associated with significantly increased perinatal mortality included infant's sex (male), number of
previous births (none, or two or more), number of previous stillbirths, marital status (unmarried) and maternal age (less than 20, or 35 or more). Maternal age less than 20 years was a risk factor for multiparas only. Conversely, primipara status was a risk factor only for those aged 20 years or older. Discussion The concept of a "perinatal period", which originated during the 1920s and 1930s, was based on the premise that stillbirths and early-neonatal deaths spring from a common complex of causes.6 This is clearly true of certain conditions, such as
Table Ill-Estimated odds ratios and case fatality rates for early-neonatal mortality, by birth weight and length of gestation
Odds ratio (and case fatality rate, %);* length of gestation, wk
Weight, g
26-27
26 NS
NS
(99.5)t
(100.0)
< 500
7490
1770
(78.8)
(44.9) 576 (23.8)
500-999
1000-1499
2940
(50.0) 1500-1999 2000-2499
2500-2999 > 3000
no survival, odds ratio not calculated;
- =
32-34
1270 (35.9) 340
781 (25.6) 173
128
(13.7)
(8.4) 78.3
(3.4)
186 (8.5) 95.7 (7.9)
381
(15.0)
*NS
28-31
35-37
34.3
8.1
47.7 (2.2) 16.8 (0.6) 6.2 (0.4) 2.8
(0.0)
(0.8)
(0. 1)
(3.6) 32.3
(1.4)
52.0
14.9
(2.3)
(0.7)
>
38
97.4 (6.4) 30.1 (1.9) 8.7 (0.4) 2.7 (0.1) 1.0 (0.05)
fewer than 10 births.
tTwo babies survived the early-neonatal period but died between 6 and 28 days after birth.
R A; T 30 X
I
P E R
25-~ / /
I
20
/
1
/
--
1 5-
7
A
-
1
"Ir
_~~~~~~p
_
_-
_
_
0-,
B
Ii 5a
R
T H
I
5 -l II
0
-r--t--t--
1 5
I
20
,
7---7--T--T'
25
I
30
MOTHER'S
I
40
35
45
AGE, YR
Fig. 1 -Perinatal mortality rate for singleton births, by maternal age. P
=
primiparous, M
=
multiparous.
CAN MED ASSOC J, VOL. 133, DECEMBER 15,1985
1217
"marker". However, its effects disappeared when birth weight was controlled for. This finding, along with the significant association found between marital status and birth weight, indicates that the effects associated with being unmarried are mediated through a downward shift in the birth-weight distribution. Similar findings have been reported from Sweden.19 Male infants were at greater risk of perinatal death than female infants. Controlling for birth weight and length of gestation increased the size of the odds ratio, a reflection of the tendency of male infants to be heavier. Five or more previous births was also found to be a significant risk factor, although controlling for confounding variables reduced the odds ratio. Perinatal mortality rates are declining throughout the Western world, including Canada. The improvement is largely the result of lower birth-weight-specific mortality rates, not a change in birth-weight distributions.20 Given the high cost of medical technology, greater emphasis should be placed on the prevention of low birth weight. The comparatively low neonatal mortality rate in Sweden is at least partially the result of a more favourable birth-weight distribution.21 Birth weight per se is not a modifiable risk factor but should be considered as a health outcome for the infant. One of the most important modifiable risk factors for low birth weight is maternal cigarette smoking: the Ontario perinatal mortality study22 estimated that 31% of the risk for low birth weight is attributable to this factor. From the public health perspective, the risk factors of major interest are the modifiable factors that result in retarded fetal growth, birth defects or premature onset of labour. Low birth weight and fetal immaturity can be viewed as an intermediate step in the overall
congenital anomalies or anomalies resulting from maternal complications. It would be arbitrary, for example, to separate stillborn infants with anencephaly from those with the condition who are live-born. However, it seems less appropriate to class together stillbirths and early-neonatal deaths due to other conditions. Early-neonatal mortality is largely due to the effects of low birth weight and fetal immaturity, with roughly 75% of all deaths attributable to these conditions. In contrast, when congenital anomalies are excluded, one half of stillbirths result from complications of the placenta, membranes or cord. Low birth weight is more likely to be an effect than a cause of this condition. It is for this reason that, unlike in other studies,6,6 analysis by birth weight in this study was restricted to early-neonatal mortality. Early-neonatal mortality rates were significantly lower in urban than in nonurban counties. This finding may be attributable in part to better access to neonatal care facilities in the former areas. As in other cross-sectional studies, the uncontrolled relation between maternal age and perinatal mortality was found to be U-shaped. Bakketeig and colleagues,6 in a longitudinal study, found a steady increase in risk in firstborn infants from a maternal age of under 20 years to over 35 years. For second and third births, rates were lowest in infants of women aged 20 to 24 years, with a slight increase thereafter for second births and no increase for third births. Our observation of no difference in perinatal mortality rates between infants of unmarried adolescent primiparas and those of their adult counterparts has also been reported from Australia.'7 The increased risk in offspring of adolescents is likely a reflection of low socioeconomic status and high prevalence of cigarette smoking among adolescent mothers.18 Being unmarried was a significant risk ,
1218
:.
CAN MED ASSOC J, VOL. 133, DECEMBER 15,1985
.;.
..
.4
,
't :.:.
"k.
'-.
'.I
.'.
-I
9. Howe G, Lindsay J: A generalized iterative record linkage computer system for use in medical follow-up studies. Comput Biomed Res 1981; 14: 327-340 10. International Classification of Diseases, 8th rev, WHO, Geneva, 1969 11. International Classification of Diseases, 9th rev, WHO, Geneva, 1978 12. Statistics Canada: 1976 Census of Canada Population: Geographic Distribution. Urban and Rural Distribution, 1 (cat no 92-807), Minister of Supply and Services, Ottawa, 1978 13. Kleinbaum D, Kupper L, Morgenstern H: Epidemiologic Research: Principles and Quantitative Methods (Research Methods ser), Lifetime Learn, Belmont, Calif, 1982 14. Baker R, Nelder J: The GLIM System Release 3, Numerical Algorithms Group, Oxford, 1978 15. Breslow NE, Day NE: Statistical Methods in Cancer Research, vol 1: The Analysis of Case-Control Studies, International Agency for Research on Cancer, Lyon, 1980 16. Alberman E: Factors influencing perinatal wastage. Clin Obstet Gynaecol 1974; 1: 1-15 17. Correy JF, Kwok PC, Newman NM et al: Adolescent pregnancy in Tasmania. Med J Aust 1984; 141: 150-154 18. Meyer MB, Tonascia JA, Buck C: The interrelationship of maternal smoking and increased perinatal mortality with other risk factors. Am J Epidemiol 1974; 100: 443-452 19. Social and Biological Effects of Perinatal Mortality, WHO, Budapest, 1978 20. Dunn HG: Social aspects of low birth weight. Can Med Assoc J 1984; 130: 1131-1140 21. Guyer B, Wallach LA, Rosen SL: Birth-weight-standardized mortality rates and the prevention of low birth weight: How does Massachusetts compare with Sweden? N Engl J Med
pathway between causal risk factors and most early-neonatal deaths. To determine which risk factors could best be reduced in pregnant women, cohort or case-control studies are needed to obtain precise data on maternal physiology and risk factors such as smoking, drug abuse and diet that are not available on a population basis. References 1. Susser M, Stein Z: Prenatal nutrition and subsequent development. In Reed D, Stanley F (eds): The Epidemiology of Prematurity, Urban & S, Baltimore, 1977: 177-192 2. The Health Consequences of Smoking for Women: a Report of the Surgeon General, US Dept of Health and Human Services, Washington, 1980: 224-279 3. Naeye RL: Relationship of cigarette smoking to congenital anomalies and perinatal death. Am J Pathol 1978; 90: 289293
4. Kaminski M, Rumeau C, Schwartz D: Alcohol consumption in pregnant women and the outcome of pregnancy. Alcoholism (NY) 1978; 2: 155-163 5. Cabrera R: The influence of maternal age, birth order and socio-economic status on infant mortality in Chile. Am J Public Health 1980; 70: 174-177 6. Bakketeig L, Hoffman H, Oakley A: Perinatal mortality. In Braken M (ed): Perinatal Epidemiology, Oxford U Pr, New York, 1984: 99-151 7. Office of Population Censuses and Surveys: Social and Biological Factors in Infant Mortality 1975-6 (Occasional Paper 12), HMSO, London, 1978: 1-20 8. Koops B, Morgan LJ, Battaglia FC: Neonatal mortality risk in relation to birthweight and gestational age: update. J Pediatr 1982; 101: 969-977
1982; 306: 1230-1233
22. Meyer MB, Jones BS, Tonascia JA: Perinatal events associated with maternal smoking during pregnancy. Am J Epidemiol 1976; 100: 464-476
3 a new concept
~
*I1 *l
I
~
~
~
~
~
S11
0
lli^'0 _~~~~~~~~~~~l
se
d
it
mm
'l
oeinfr
m
_-
6 ~
~
~
_*
* _ ** _.*** _* Is3|S *.. _** * -e s
iil
i *iI
E
FC
N
C.
D
01
av.: Vitoi
.,j -.z-ONE No;
iE
_6
B O
ain
sen me mor inomain BISL 0AAAIC 0 v.Vcoi St-Lambert, Quebec, Canada J4R 1P8 (514) 465-9010 NAME
__
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
ADDRESSS
I~CITY
PRC.
CAN MED ASSOC J, VOL. 133, DECEMBER 15, 1985
1219
