Volume 23
Number 4
2002

[Table of Contents]

The impact of missing birth weight in deceased versus surviving fetuses and infants in the comparison of birth weight-specific feto-infant mortality 

Shi Wu Wen, Li-Mei Chen, Chung-Yi Li, Michael S Kramer and Alexander C Allen, for the Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System 

Abstract 

Birth weight-specific is preferred to crude feto-infant mortality in epidemiologic studies comparing rates across jurisdictions, because it can help limit the bias arising from regional differences in the completeness of reporting of vital events and in classification of live versus stillbirth among extremely small and immature infants. The potential impact of missing birth weight information in deceased versus surviving fetuses and infants in the comparison of birth weight-specific feto-infant mortality has been seldom examined, however. The authors investigated this issue, using data collected from two nationwide surveys of all pregnancy outcomes occurring 15–17 May 1989 and 12–16 February 1996, respectively, in Taiwan and the 1989 and 1996 linked birth and infant death records in Canada (excluding Ontario and Newfoundland). The proportions with missing birth weight information in Taiwan in 1989 were 25.0%, 15.4%, 0%, and 0.6%, respectively, for stillbirths, neonatal deaths, post-neonatal deaths, and survivors, and in 1996 were 100%, 5.0%, 0%, and 0.2%. The proportions with missing birth weight information in Canada in 1989 were 5.8%, 2.6%, 1.2%, and 0.6% for fetal deaths, neonatal deaths, post-neonatal deaths, and survivors, respectively, and in 1996 were 5.0%, 2.4%, 1.1%, and 0.6%. Infant and (especially) fetal death rates were substantially higher in Taiwan than in Canada among births with missing birth weight. The authors concluded that differences in missing birth weight information between deaths and survivors can bias comparisons of birth weight-specific feto-infant mortality. 

Key words: bias; birthweight; infant mortality; stillbirth 

Introduction 

Infant mortality has been considered the single most comprehensive index for comparing health status in a society.^1–3 Many different sources of data, including routine vital statistics and records collected from perinatal health surveillance systems or health surveys, have been used for such comparisons.^4–10 An essential issue in the comparison of infant mortality across different countries and regions is ensuring that the observed differences reflect true differences in infant mortality among the countries/regions to be compared, rather than artifacts. 

The most frequently reported artifact is due to regional differences in registration and reporting practices. For example, countries and regions differ in the completeness of their reporting of vital events and in their classification of live births versus fetal deaths among extremely small and immature infants,^10-14 and such comparisons may therefore be biased. Analysis of birth weight- and age-at-death-specific infant mortality rates may help in this regard. If most of the observed difference in fetal and early neonatal mortality occurs in infants near the borderline of viability, registration practice is the probable explanation. Otherwise, other factors, such as maternal health and the quality of and access to maternal and infant health care services, should be considered. 

Theoretically, differences in any process of data collection, editing, and reporting among countries/regions may bias the comparison of infant mortality. For example, if birth weight information is missing more frequently among deaths than among survivors and if this differentially missing birth weight information varies among the countries or regions to be compared, birth weight-specific infant mortality comparison will be biased. The potential bias introduced by differential missing birth weight information between deaths and survivors has seldom been assessed. 

We recently initiated a project comparing birth weight- and age-at-death-specific feto-infant mortality between Taiwan and Canada. In the preliminary analysis, we found a substantially higher proportion of missing birth weight information among fetal deaths than in live births in Taiwan. This disproportionally higher missing birth weight information in Taiwan prevented us from further comparison of birth weight- and age-at-death-specific feto-infant mortality between Taiwan and Canada. Because a similar problem is likely to occur in other countries and regions, we highlight these observed differences and examine their impact on birth weight-specific feto-infant mortality. 

Materials and methods 

The 1989 and 1996 Statistics Canada Linked Birth and Death files were used for the Canadian analysis, and the data collected in two national surveys in Taiwan in 1989 and 1996 were used for the Taiwanese analysis. 

Birth and death information in Canadian data is collated from birth and death registrations.¹⁵ Information on birth weight in the birth registration is obtained from the responsible physician or the mother. These data have been extensively evaluated for quality, and the successful linkage rate is more than 97%.¹⁶ Ontario data were excluded because of documented problems in data quality.¹⁷ Births to mothers residing in Newfoundland were also excluded, because data from this province were not available prior to 1991. 

Because Taiwan's official vital statistics are known to severely under-report perinatal deaths,^4,5 we analyzed data collected from two nationwide surveys of all pregnancy outcomes occurring at >=20 weeks' gestation in 15–17 May 1989 and 12–16 February 1996, respectively. Both surveys were designed and coordinated by a research team led by the local principal investigator (Prof. Li-Mei Chen). A two-stage data collection was designed. At the first stage of data collection, a birth event recording form specifically developed for the surveys was issued by each county and city health bureau to the medical facilities (including hospitals, clinics, health stations, and midwife's offices in the corresponding administrative area) before the start of the full-scale data collection to record all pregnancy outcomes occurring at >=20 weeks gestation during the study period. After receiving the completed recording forms from the corresponding medical facilities, pre-assigned administrative staff in each county and city health bureau carefully reviewed the completeness of the recording forms and then transferred the completed forms to the corresponding public health nurses for preparation of the second stage of data collection. In the second stage, public health nurses from each health station conducted face-to-face interviews in the women's homes to collect additional information. All completed forms were carefully checked by the associated administrative staff in each health bureau. Finally, all completed forms were compiled and sent back to the research centre for data entering and editing. 

To ensure accuracy and consistency of data collection, the research coordinating centre provided an intensive one-day training course for all of the participating administrative staff and public health nurses before the full-scale data collection. The training course included an explanation of the purpose of the study, the techniques of household interview, and detailed instructions for each questionnaire item. All study procedures were compiled into a field operations manual to further standardize data collection. Information on birth weight was obtained from the delivering hospitals, and mortality was ascertained during the second stage data collection (home visit by public health nurses). The response rate was high (>95%) and ascertainment of pregnancy events was complete in both surveys.^4,5 No systematic differences were detected between those who responded and those lost to follow-up.^4,5 Because there is no seasonal variation in feto-infant mortality and morbidity in Taiwan,⁵ the data from these two nationwide surveys can be considered as representative of all births for the corresponding years in Taiwan. 

Fetal death reporting in Taiwan is restricted to >=20 weeks of gestation, while in Canada the restriction is >=500 g or >=20 weeks of gestation, depending on the province. The definition of live birth in both Taiwan and Canada follows the World Health Organization: the complete expulsion or extraction from the mother of a production of conception, irrespective of duration of pregnancy, which after such separation breathes or shows any other evidence of life. Infant death is defined as death among live births in the first year (0 to 364 days) of life. Neonatal death is defined as death among live-born infants in the first 28 days (0 to 27 days) of life. Post-neonatal death is defined as death among survivors at 28 days of age in the post-neonatal period (28 to 364 days). To avoid the impact of temporal trends in fetal and infant mortality and their determinants in the two countries, we analyzed the two years' data separately. 

We first compared the distribution of maternal age, parity, plurality, and birth weight between Taiwan and Canada. To assess the magnitude of missing birth weight by survival status, we compared the proportion of missing birth weight among fetal deaths, neonatal deaths, post-neonatal deaths, and survivors in the two countries. 

We hypothesized that the missing birth weight information in deceased cases may have different effects on the Taiwan-Canada comparison within different categories of birth weight- and age-specific fetal and infant mortality. Specifically, we expected a major impact of missing birth weight in the comparison of fetal and neonatal mortality at the lower birth weights, with a much smaller impact on comparisons of post-neonatal mortality and at higher birth weights. We therefore compared fetal and infant mortality in the overall study sample and in three sub-samples: 

• births with missing birth weight information; 
• births with recorded weight <500 g; and 
• births with recorded weight >=500 g. 

We then compared fetal, neonatal, and post-neonatal mortality among births with birth weight of 500 to 1499 g, 1500 to 2499 g, and >=2500 g in Taiwan and Canada. Relative risks and 95% confidence intervals were used for comparison of mortality rates, with Canada as the reference. All analyses were performed using SAS-PC 6 version statistical software. 

Results 

The proportions of women of young and older ages and high parity were higher in Canada than in Taiwan (Table 1). While the rate of multiple birth was lower in Taiwan in Cohort I, the rate was higher in Cohort II (Table 1). The low birth weight rate was slightly lower in Taiwan, whereas the proportion of births with missing birth weight information was slightly higher (Table 1). The proportion of women of older age increased in Cohort II in both countries (Table 1). The proportion of women of high parity increased in Canada but decreased in Taiwan (Table 1). 

The proportions with missing birth weight information were substantially higher among deaths than among survivors in both Canada and Taiwan, but there were substantial variations in missing birth weight according to age at death and country. In Canada, missing birth weight occurred most frequently among fetal deaths and least frequently among survivors, with an apparent gradient in both years (Table 2). In Taiwan, the pattern was more complicated, with the highest frequency of missing birth weight among fetal deaths, the second highest among neonatal deaths, and the third among survivors, with no missing birth weight among post-neonatal deaths in either year (Table 2). Overall, missing birth weight among deaths was more frequent in Taiwan than in Canada. However, because the sample sizes in some cells in the Taiwan cohorts were very small, the results were unstable. 

Fetal and infant mortality rates were somewhat higher in Taiwan than in Canada (Table 3). When the comparison was restricted to the three sub-samples, different patterns emerged, however; infant and (especially) fetal death rates were substantially higher in Taiwan than in Canada among births with missing birth weight but smaller among infants with birth weight <500 g and those with birth weight >=500 g (Table 3). 

Among births with a birth weight of 500 to 1499 g in Taiwan, the fetal death rate was lower than in Canada, while the neonatal and post-neonatal death rates were higher. Among infants with a birth weight of 1500 to 2499 g, no fetal deaths were observed in Taiwan, but both neonatal and post-neonatal death rates were higher in Taiwan. Among births weighing >=2500 g, the fetal death rate was lower in Taiwan, whereas the neonatal and post-neonatal death rates were higher, although substantial variations in rates were observed in the two cohorts, owing to limited sample sizes (Table 4). 

Discussion 

Our original purpose in comparing birth weight- and age-at-death-specific feto-infant mortality between Taiwan and Canada was to assess the reasons why the overall infant mortality rate was 30% higher in Taiwan than in Canada.^4,5,10 We hypothesized that the higher infant mortality in Taiwan was caused by gaps in perinatal care. Previous studies have shown that because of their more favourable birth weight distribution (with fewer low birth weight infants) and lower maternal exposure to risk factors for perinatal death,^6–9 perinatal mortality rates are actually slightly lower in ethnic Chinese than in ethnic whites in societies with comparable perinatal care. We sought to test this hypothesis by analysing birth weight- and age-at-death-specific feto-infant mortality. 

Although we have limited available information on risk factors, the lower proportions of mothers of young and older age, high parity, and low birth weight infants in Taiwan (Table 1) support our hypothesis. However, our attempt to compare birth weight- and age-at-death-specific feto-infant mortality between the two countries was impaired, at least in part, by the problem of missing birth weight. 

Major differences were observed in missing birth weight information between deceased and surviving infants, with substantially more frequent missing birth weight information among the deaths than in the survivors both in Canada and Taiwan. In Canada, there was an apparent gradient in the proportion of missing birth weight information according to survival status, with the highest proportion among fetal deaths and the lowest among survivors through infancy. The nine-fold increase in missing birth weight among fetal deaths versus survivors in Canada in both years of study, even after rigorous scrutiny of data quality, raises serious concern in comparing birth weight- and age-specific feto-infant mortality. 

The situation was even more complicated in Taiwan. While the proportion of missing birth weight information among fetal deaths was 5- to 20-fold higher in Taiwan than in Canada, this proportion was actually lower than in Canada among survivors and (especially) post-neonatal deaths. Because births with missing birth weight cannot be included in the calculation of birth weight-specific mortality, disproportionally missing birth weight information between deaths and survivors can substantially bias all comparisons. 

As expected, fetal and infant death rates were much higher in Taiwan among births with missing birth weight information. On the other hand, fetal and infant mortality rates in Taiwan among birth weight <500 g or >=500 g may be artificially reduced, because some of the otherwise eligible deaths were excluded in the calculation of these rates. Among births weighing 1500–2499 g or >=2500 g, both neonatal and post-neonatal death rates were higher in Taiwan, suggesting true differences rather than artifacts. However, substantial variations in these rates made any attempt to draw inferences from the comparison of birth weight-and age-specific mortality difficult. Although limited sample size in Taiwan data and differences in data collection between the two countries undoubtedly affected the comparison of birth weight- and age-at-death-specific feto-infant mortality, there is no doubt that the substantial variation in missing birth weight information according to survival status also undermines any such attempt. 

We do not know why a higher proportion of birth weight information was missing in deaths than in survivors. Parents' attitudes may play a role here. Information on birth weight in Canada is provided by the responsible physician or parents who register the birth. When parents register their infant, they may be more careful about details such as birth weight and gestational age if their infant is healthy, while less attention may have been paid to these details if their infant dies or is seriously ill. The situation in Taiwan may be different. Because of apparent problems in official vital statistics, we have used information obtained from the two surveys. The two cohorts represented the most complete and accurate fetal and infant mortality statistics on the island. In the 1989 Taiwan survey, Chen et al found that the survey-derived infant mortality rate was 9.72 per 1000 live births, whereas the official statistic for the same calendar year was 5.71 per 1000 live births.⁵ Cultural factors may have played a role here. Most of the under-registration of infant deaths in Taiwan occurs during the first 27 days of life, because parents appear reluctant to register an infant who dies soon after birth.⁵ It is known that in Taiwan, no record of birth weight was kept for some births born in local hospitals or home-births, especially for stillbirths or infants who died early in life. Although the research team for the Taiwan surveys made every effort to identify the deaths, it was impossible for them to obtain birth weight information if it was not recorded in the first place. We do not know why information on birth weight was missing for all 18 cases of fetal death in the 1996 cohort in Taiwan. It should be noted that this is a highly unstable estimate because of the small number. 

Previous studies of the impact of registration practices on international comparisons of fetal and infant mortality rates have focussed on incomplete reporting of death events and variation in classification of live births and fetal deaths.^5,10–14 This study adds to this literature by highlighting an additional source of bias when comparing birth weight- and age-at-death-specific feto-infant mortality: differences in missing birth weight information between deaths and survivors. 

Acknowledgments 

We thank the vital statistics registrars in the Canadian provinces and territories who gave us access to their data files. This study was supported in part by a grant from the National Science Council of Taiwan (NSC 89-2320-B-030-003), and in part by the Canadian Perinatal Surveillance System. 

References 

1.    Brzezinski J. Mortality indicators and health-for-all strategies in the WHO European Region. Wld Hlth Stat Q 1986;39:365–378. 

2.    Centers for Disease Control. Censensus set of health status indicators for the general assessment of community health status – United States MMWR 1991;40:449–451. 

3.    Wise PH, Pursley DWM. Infant mortality as a social mirror. N Engl J Med 1992;326: 1558–1560. 

4.    Chen L-M. Epidemiologic study of infant mortality, neonatal mortality, and perinatal mortality in Taiwan. Annual project report. Taipei, Taiwan: Department of Health, 1997. 

5.    Chen L-M, Sun C-A, Wu D-M, Shen M-H, Lee W-C. Underregistration of neonatal deaths: an empirical study of the accuracy of infantile vital statistics in Taiwan. J Epidemiol Community Health 1998;52:289–292. 

6.    Fung KP, Wong TW, Lau SP. Ethnic determinants of perinatal statistics of Chinese: demography of China, Hong Kong and Singapore. Int J Epidemiol 1989;18:127–31. 

7.    Wang X, Strobino DM, Guyer B. Differences in cause-specific infant mortality among Chinese, Japanese, and white Americans. Am J Epidemiol 1992;135:1382–93. 

8.    Shiono PH, Klebanoff MA, Graubard BI, Berendes HW, Rhoads GG. Birth weight among women of different ethnic groups. JAMA 1986;255:48–52. 

9.    Wen SW, Kramer MS. A comparison of perinatal mortality between ethnic Chinese and ethnic whites: why the Chinese rate was lower? Ethnicity & Health 1997;2:177–82. 

10.    Joseph KS, Kramer MS. Recent trends in Canadian infant mortality rates: effect of changes in registration of live newborns weighing less than 500 g. CMA J 1996;155: 1047–52. 

11.    Cartlidge PHT, Stewart JH. Effect of changing the stillbirth definition on evaluation of perinatal mortality rate. Lancet 1995;346: 486–88. 

12.    Howell EM, Blondel B. International infant mortality rates: bias from reporting differences. Am J Public Health 1994;84:850–852. 

13.    Sepkowitz S. Why infant very low birthweight rates have failed to decline in the United States vital statistics. Int J Epidemiol 1994;23:321–326. 

14.    Sachs BP, Fretts RC, Gardner R, Hellerstein S, Wampler NS, Wise PH. The impact of extreme prematurity and congenital anomalies on the interpretation of international comparisons of infant mortality. Obstet Gynecol 1995;85:941–946. 

15.    Fair M, Cyr M. The Canadian Birth Data Base: a new research tool to study reproductive outcomes. Health Rep 1993;5: 281–90. 

16.    Fair M, Cyr M, Allen A, Guyon G, Wen SW, McDonald R. Assessment of the validity of record linkage of births and infant deaths in Canada. Chronic Dis Can 2000;21:8–13. 

17.    Joseph KS, Kramer MS. Recent trends in infant mortality rates and proportions of low-birth-weight live births in Canada. CMAJ 1997;157:535–541.

Author References 

Shi Wu Wen, Health Surveillance and Epidemiology Division, Centre for Healthy Human Development, Health Canada, Ottawa, Ontario, and Department of Obstetrics and Gynecology and Clinical Epidemiology Program, University of Ottawa, Ottawa, Ontario 

Li-Mei Chen, Chung-Yi Li, Department of Public Health, College of Medicine, Fu-Jen Catholic University, Taipei County, Taiwan, ROC 

Michael S Kramer, Departments of Pediatrics and of Epidemiology and Biostatistics, McGill University, Montreal, Quebec 

Alexander C Allen, Department of Pediatrics, Dalhousie University, Halifax, Nova Scotia, for the Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System* 

^*Contributing members: K S Joseph (Dalhousie University), Robert Platt (McGill University), Reg Sauve (University of Calgary) 

Correspondence: Dr Shi Wu Wen, The Ottawa Hospital – General Campus, 501 Smyth Rd, Box 201, Ottawa, Ontario, Canada K1H 8L6, Fax: (613) 739-6266;
E-mail: swwen@ohri.ca 

[Previous] [Table of Contents] [Next]