OBSTETRICS OBSTETRICS

Neonatal Mortality Risk Related to Birth Weight and Gestational Age in British Columbia William J. Kierans, BA, MA,1 Lorne A. Verhulst, MD, MPA, 2 Jemal Mohamed, BSc, MSc,3 Leslie T. Foster, BSc, MA, PhD4 1

British Columbia Vital Statistics Statistics Agency, Victoria BC

2

Chronic Disease Management and Primary Health Care Renewal Branch, Medical Services Division, British Columbia Ministry of Health, Victoria BC

3

Strategic Directions and Analysis Branch, Medical Services Division, British Columbia Ministry of Health, Victoria BC

4

School of Child and Youth Care, University of Victoria, Victoria BC

Abstract Objectives: To provide gender-specific neonatal mortality grids depicting relative risk in narrow birth weight ranges at each week of gestation. The grids will provide practitioners with clinically relevant information pertinent to pregnancy, delivery, and postnatal care in Canada. Methods: Records from the British Columbia Vital Statistics Agency birth and death registries from 1981 to 2000 were deterministically linked and resulted in a 99.86% linkage rate. Risk ratios were computed by dividing percent neonatal mortality in 250 g birth weight categories at each week of gestation by the overall gender-specific mortality rates for the full period. We adjusted random rate fluctuations that were due to low frequencies in the narrow birth weight and gestational age strata. Results: Females exhibited greater survival across the full spectrum of birth weight by gestational age strata, but their mortality configurations were noticeably different from those of males. In addition, there were demarcations in both grids that depicted relatively abrupt changes in risk ratios. Although the crude mortality rates in BC decreased during the study period, the use of risk ratios reduced the disparity in crude rates over time. Conclusion: These gender-specific mortality grids refine and enhance previously available comparisons by portraying neonatal mortality risks in narrow birth weight ranges at each week of gestation.

Résumé Objectifs : Fournir des grilles de mortalité néonatale par sexe décrivant le risque relatif, en ce qui concerne d’étroites plages de poids de naissance à chaque semaine de la gestation. Ces grilles fourniront aux praticiens des renseignements cliniques pertinents au sujet de la grossesse, de l’accouchement et des soins postnatals au Canada. Méthodes : Les dossiers des registres des naissances et des décès de la ritish Co umbia Vita Statistics gency , couvrant la période Key Words: Birth weight, gestational age, neonatal mortality, mortality grids Competing Interests: None declared. Received on November 27, 2006 Accepted on February 6, 2007

568

l JULY JOGC JUILLET 2007

de 1981 à 2000, ont été liés de façon déterministe, ce qui a mené à un taux d’appariement de 99,86 %. Les risques relatifs ont été calculés en divisant la mortalité néonatale (en %) dans les catégories de poids de naissance de 250 g à chaque semaine de gestation par les taux globaux de mortalité par sexe pour l’intégralité de la période. Nous avons neutralisé l’effet des fluctuations aléatoires des taux qui étaient attribuables à de faibles fréquences au sein de la strate restreinte « poids de naissance et âge gestationnel ». Résultats : Les filles ont présenté un taux de survie supérieur tout le long de la strate « poids de naissance par âge gestationnel », mais leurs configurations de mortalité étaient nettement différentes de celles des garçons. De plus, les deux grilles présentaient des démarcations décrivant des modifications relativement abruptes en ce qui concerne les risques relatifs. Bien que, en C.-B., les taux bruts de mortalité aient connu une baisse pendant la période d’étude, l’utilisation de risques relatifs a permis de réduire la disparité des taux bruts avec le temps. Conclusion : Ces grilles de mortalité par sexe améliorent et rehaussent les comparaisons qui étaient disponibles auparavant, en décrivant les risques de mortalité néonatale en fonction d’étroites plages de poids de naissance à chacune des semaines de gestation. J Obstet Gynaecol Can 2007;29(7):568–574

INTRODUCTION

irth weight and gestational age are well recognized determinants of neonatal mortality. Changes in population-based neonatal mortality, related to birth weight and gestational age, have been quantified regularly in reports from government vital event registries. However, those reports provide only a cursory understanding of the interaction of both factors on mortality risk. Grids showing greater detail have rarely been reported.1–5

B

Mortality grids provide pertinent information for medical practitioners when they are considering the possibility of postnatal problems, whether to induce delivery or prolong gestation, when to carry out in utero therapy, or when to transfer a pregnant patient to a higher care hospital. They

Neonatal Mortality Risk Related to Birth Weight and Gestational Age in British Columbia

are also a useful educational tool, because they graphically portray changes in risk in relation to intrauterine growth and gestational age. They may serve as a useful public health measure by providing detailed comparisons among populations. The few available grids have not been particularly applicable in Canada because the reported mortality rates were higher than Canadian rates.1–5 Therefore it is uncertain whether the relative risks associated with babies that are large for dates or small for dates could be extrapolated to the Canadian context. There are few actual neonatal deaths in the relatively smaller samples in published reports.1,2,4 These reports were based on regional or in-hospital samples,1–4 and thus were not representative of the entire population of newborns. Published grids were not gender-specific.1,2,4,5 Females are at lower risk for mortality and morbidity than males after accounting for lower birth weight at the same gestational age.3,6 To our knowledge, the first populationbased grid in Canada was published by the British Columbia Vital Statistics Agency (BCVSA),7 and the BCVSA subsequently published the first gender-specific grids.8 Most reports published in the last 10 years that analyzed gestational age-specific neonatal mortality were focused on small for gestational age (SGA) neonates.9–14 Studies that included immediate and longer term outcomes related to large for gestational age (LGA) neonates other than macrosomia are much less common.15–17 However the studies were not in a Canadian context, with the exception of the reports of Anath and Wen9 and Wen et al.17 Our data quantify the relative risk of neonatal death at both the upper and lower extremes of weight at all gestational ages in a large Canadian dataset. This report presents updated gender-specific, populationbased mortality grids based on deliveries in the context of Canadian obstetrical care. They have been modified from previous versions8 to show clinically relevant estimates of relative risk. We used data provided by BCVSA because of the virtually complete linkage rate between birth and death records and the established quality of gestational age estimates. METHODS

The data were based on live birth and death registrations and notices of birth detailing events from January 1, 1981 to December 31, 2000. The full BCVSA data file contained 882 390 records of live births to residents of British Columbia. The analyses file was restricted to live births at between 22 and 44 completed weeks of gestation and birth weights between 500 g and 6000 g. Records in which the weight, gestational age, or gender were missing, and four cases in which the weight was grossly at variance with the

gestational age (see below), were excluded. Those missing birth weight and gestational age values comprised < 0.25% of births in the full data file and were proportionally distributed across gestational age. The BCVSA makes a particular effort to include birth weight and gestational age values on all records. If either measure is not recorded on the notice of birth, the source is contacted before the record is processed. There were only nine cases with missing gender values. Births to BC residents occurring in Alberta were also included because previous work indicated the potential for skewed results if those events were not included,18 although such events were few (< 1%) and would not have affected the results reported here. This study was designed to provide reference charts that are representative of all live births; therefore, non-standard births (e.g., births with complications, multiple deliveries, or congenital anomalies) were not excluded from the analyses. That has also been the practice in other mortality grid studies.2–4 Infant death registration records from the BCVSA death registry were linked and added to the birth records (including any infant deaths in 2001 that occurred to infants born in 2000). There were 3661 neonatal deaths in the BCVSA death file for the period 1981 to 2000.19 Comprehensive searching yielded 3656 (99.86%) links to the birth file. Links were deterministically based on birth registration number, which appears on the infant death record. Subsequent exclusions reduced the number of neonatal deaths included in the final grids. In the case of infant deaths to former residents of British Columbia, inter-provincial agreements ensured that the death record was available for linkage. The confidentiality of BCVSA records was protected according to approved practices. Neonatal mortality (< 28 days) was calculated as a percentage of live births in birth weight groups of 250 g at each week of gestation for males and females separately. However, the narrow birth weight and gestational age ranges resulted in low birth and death frequencies in individual grid cells, which in turn produced random mortality rate fluctuations. To compensate for those random rate fluctuations in individual grid cells, a calculation strategy was used to produce the final male and female grids. In this strategy, cells with fewer than five births were excluded from further calculations. Mortality in percentage for each remaining cell was averaged with the percentage in the four cells immediately surrounding it. If four cells were not available (such as on the margins of the grid), excluding unavailable cells would produce a negative bias in the upper grid cells and a positive bias in the lower cells, so in those cases the percentage in the specific cell was averaged with the percentage in the cell immediately to the left and below (i.e., one week and birth weight stratum lower), and the percentage in the cell JULY JOGC JUILLET 2007 l

569

OBSTETRICS

immediately to the right and above (i.e., one week and birth weight stratum higher). This latter procedure was intended to balance a progressive reduction in mortality upwards and to the right in the whole grid. Cells that did not conform to both averaging procedures were excluded from the final grids. Risk ratios were computed for each birth weight by gestational age crossing in the male and female grids to provide a clinically relevant assessment of comparative risk. The ratios were calculated by dividing the percentages in each cell by the overall gender-specific percentage mortality8 during the study period. The accuracy of the gestational age estimates in the BCVSA data file was examined prior to analysis. Birth weights more than four standard deviations from the mean birth weight at each week of gestation were identified and confirmed or corrected by accessing original documents, which were found for all but four cases; these four were excluded from further analysis. Since the early 1980s, ultrasound assessment has been routinely performed in British Columbia early in the second trimester. The main source of BCVSA gestational age data prior to 1993 was the notice of birth completed by the attending physician (containing the gestational age as recorded by the physician, which is usually based on an early ultrasound estimate20), rather than the birth registration completed by the mother, and since 1993 the notice of birth has been the sole source. RESULTS

After the exclusions reported previously (see Methods), there were 450 742 male births (1706 neonatal deaths) and 427 196 female births (1214 neonatal deaths) included in the final grids. Available infant and maternal characteristics of male and female cases included in the grids and in the full analysis file are shown in the Table. The multi-staged exclusion and averaging process had a greater proportional effect on the neonatal deaths because excluded cases were on the extreme margins of the grids where there was greater risk of mortality. The male and female grids each had 205 cells, but the mortality configurations were noticeably different. Risk ratios for each birth weight/gestational age cell are shown for males and females in Figure 1 and Figure 2, respectively. Larger, shaded areas depict relatively abrupt changes in risk ratios. Risk ratios were categorized into six groups: Very High, Moderately High, High, Low, Moderately Low, and Very Low. All the risks in the very high risk area were more than 100 times the overall risk, whereas the apparent risks in the moderately high area were almost all well below that level. Similarly, there was an immediate reduction in risk at the transition from the high risk to the moderately high areas. The original 570

l JULY JOGC JUILLET 2007

gender-specific grids showing percent mortality are available on the BCVSA website.8 Although the mortality rates in previous versions showed a clearer demarcation between the shaded areas, the use of risk ratios altered the size and shape of those areas very little. The most obvious difference between the male and female grids was the relative size of the Very Low Mortality area. Females not only had a greater number of cells in that area (38 cells, 18.5% of the full grid) compared with males (25 cells, 12.2%), but the numerical values of the ratios were generally lower among females as well. Another obvious difference in the two grids was the relative size of the Moderately Low areas. There were fewer female cells (35 cells, 17.1%) in the area compared with males (42 cells, 20.5%), and the female Very Low area appeared to have displaced or extended into the Moderately Low area relative to the male grid, particularly in the cells depicting higher weights. The sizes of the remaining risk areas were quite similar in the male and female grids. A central line of least risk is apparent in both grids if the lowest ratio in each gestational age column is followed. The risk ratios in each gestational age column were progressively greater in the cells below and above the lowest ratio. Although the crude mortality rates in BC decreased from 1981 to 2000,8 the use of risk ratios—based on the crude rates in the whole period as denominator—reduced the disparity in crude rates over time. Male and female birth frequencies were insufficient in some cells to compare full grids for 1981 to 1990 and 1991 to 2000, so the neonatal death rates in the six risk categories were used to compare the two time periods. For each period and for each gender, we calculated the ratio of the mortality in each risk category to the overall percent mortality in the period. The period differences (not shown) fluctuated because of the reduced frequencies, but were within limits that assure reliability of the full-period risk estimates. DISCUSSION

We present mortality grids that represent improvements to previously available grids1–5,7,8 by showing clinically relevant newborn mortality risks that are both gender-specific and applicable to the Canadian care setting. In addition, we describe an easily replicable averaging process that adjusts for random variation in risk but retains the characteristics of the changes across the grid. This may prove useful in other populations with typically low Canadian neonatal mortality frequencies. The demonstrated quality of the BCVSA gestational age estimates assures the reliability of the assessments for clinical decision making. Compared with the sample sizes used in some previous grids1,2,4 the 3656 neonatal deaths in our initial data file

Neonatal Mortality Risk Related to Birth Weight and Gestational Age in British Columbia

Maternal and infant characteristics in mortality grids and analysis file Births included in grids Infant and Maternal Characteristics Infant Death Neonatal Death

Males

Full analysis file

Females Rate1

N 2963

6.57

N 2059

Male Rate1 4.82

N 3032

Both genders2

Female Rate1 6.72

N 2108

Rate1 4.93

N

Rate1

51440

5.85

1706

3.78

1214

2.84

1760

3.90

1253

2.93

3017

3.43

Low Birth Weight

20722

45.97

22475

52.61

21104

46.74

22689

53.05

43796

49.81

Preterm

29870

66.27

24613

57.62

30150

66.77

24880

58.17

55031

Average Birth Weight (g)

3360.00

3483.00

3359.00

3423.00

27.90

27.90

27.90

27.90

27.90

450742.00

427196.00

451525.00

427676.00

879210.00

Average Maternal Age (y) Live Births

62.59

3484.00

¹ Rate /1000 live births in gender category ² Gender was missing for 9 cases Data source: British Columbia Vital Statistics Agency

permitted statistically rigorous separate grids for males and females. Although the lower mortality rates for female newborns compared with those for males have long been established,3 these grids have refined the pattern of the gender-specific differences in 250 g birth weight cells at single weeks of gestational age. The relatively abrupt change in mortality risks depicted by the six areas in these grids were not as apparent as those depicted in the original versions,8 but the abrupt changes remained clear and the size and shape of the areas was unaltered. As with all narrow classifications of birth weight by gestational age,1–5,7,8 the survival patterns are complex and are not easily detailed. However, the size and shape of the risk areas show an apparent interaction of birth weight and gestational age. Thresholds of viability are often expressed separately in terms of birth weight or gestational age. Notably, birth weights of 1500 g and 2500 g, and gestational ages of 25, 30, and 37 weeks are commonly used,3 although advances in perinatal care continue to reduce those thresholds, particularly in developed countries.3,9,21 The risk areas depicted in these grids refine those thresholds in terms of both dimensions, particularly at earlier stages of depicted development, where survival is a crucial issue. Focusing on the male grid as an example, mortality risk decreased substantially above 1250 g at 29 weeks. However, the abrupt 1250 g demarcation line extended beyond 29 weeks, suggesting that the 1250 g threshold is maintained despite increases in gestational age. The same pattern is apparent in the female grid although the birth weight threshold is lower at 29 weeks.

Relatively abrupt changes in viability are apparent at other risk area transitions, but the patterns are quite different for males and females. The most obvious difference between the genders was the relative size of the Very Low Mortality area and the generally lower absolute values of ratios among females in that area. Further, the size of the Moderately Low Mortality area was greater for males than for females. In females, survival rates higher than those of males apparently extended into lower and higher birth weight categories and weeks of gestation. For females, the mortality risk was substantially less than one quarter of overall female risk at birth weights above 3000 g from 39 to 41 weeks of gestation. In contrast, mortality risk for males at the same birth weight and gestational age were often greater than one quarter of overall male risk, particularly in the higher weight categories. According to these data, female growth above 3000 g at term resulted in progressive decreases in mortality, whereas weight increases among males at term showed increased mortality. These differences may partially explain why Seeds and Pang16 reported only minor mortality increases associated with augmented fetal growth (> 90th percentile) in a cohort of combined gender live births. Although survival in the Very Low and Moderately Low risk areas, including mild and moderately preterm births above 2500 g, is comparatively high, the number of births included in those areas has a substantial effect on the absolute number of deaths in our analysis file.22 The other areas of the grids (i.e., Low to Very High mortality risk areas), encompassing mostly preterm and low birth JULY JOGC JUILLET 2007 l

571

OBSTETRICS

Figure 1. Male birth weight and gestational age neonatal mortality risk ratios*

* Risk ratios: percent neonatal mortality in cell / Overall male neonatal mortality 1981–2000 Data source: British Columbia Vital Statistics Agency

weight infants, were approximately the same size for males and females. However, females tended to have risk ratios similar to, and often lower than, males along the central line of least risk but showed a greater increase in lower and higher birth weight strata at each week of gestation. That is, females tended to have similar mortality risk in the central birth weight categories at each gestational age but progressively higher risk than males as their birth weight decreased, and to a lesser extent increased, at each gestational age. This suggests that females fare worse with deviations from “normal” fetal development, particularly preterm. Jones et al.23 found no consistent male and female mortality differences in separate analyses of very low birth weight (< 1500 g) and early preterm (< 30 weeks) infants admitted to Canadian neonatal intensive care units. However, at a given birth weight, females tended to be more mature as a consequence of greater gestational age. Those results point to the advantage of combining birth weight with gestational age. The use of risk ratios rather than crude rates (e.g., percentage mortality) provides clinically relevant indications of risk unaffected by ambient variation in the population under consideration. The denominators in this instance were the overall male and female percentage mortality during the study period; the grid ratios therefore adjust for underlying trends in mortality compared with the rates in the full population of male and female newborns. The higher crude rates in 1981 to 1990 than in 1991 to 2000 made little difference in the calculated ratios when the overall crude male and female rates in each period were used as denominators. In 572

l JULY JOGC JUILLET 2007

addition, risk ratios will compensate for any rate differences in other Canadian settings and sub-populations. Although the averaging procedure reduced random variation, there were still irregularities in some cells, particularly on the margins of the grids, so mortality in individual cells should not be viewed as definitive. Nonetheless, the mortality pattern can be considered reliable for the purposes of obstetric and neonatal decision making. As in all studies based on birth and death registrations, errors can occur in the recording or computer entry of birth weight or gestational age, and the linkage of live births and infant deaths. The deterministic linkage method used in BC should minimize the latter type of error, and detailed analyses of the quality of the birth weight and gestational age data from this file have been reported.8 Nonetheless, a few errors at very early preterm gestations can have relatively large effects on birth weight for gestational age and survival at those gestations. We did not exclude births known to have reduced survival potential (i.e., multiples, congenital anomalies) because it has been previously shown that they each represented less than 2% of all live births and the inclusion of deaths with those attributes would have contributed minimally7,8 to the mortality rates across the birth weight and gestational age spectrum analyzed here. From these data we were unable to examine further physiological, pathological, developmental, or theoretical reasons underlying the variation portrayed in these grids. Despite these limitations, we believe our results provide reliable neonatal risk patterns within narrow birth weight and gestational age strata.

Neonatal Mortality Risk Related to Birth Weight and Gestational Age in British Columbia

Figure 2. Female birth weight and gestational age neonatal mortality risk ratios*

* Risk ratios: percent neonatal mortality in cell / Overall female neonatal mortality 1981–2000 Data source: British Columbia Vital Statistics Agency

The limited differences in neonatal mortality in the Canadian provinces and territories24 and our comparisons of cell mortality as referenced to overall gender mortality will enable health care professionals in other Canadian jurisdictions to apply these grids with the assurance that they represent local mortality risk patterns.

Executive Officer of BCVSA, and personnel at BCVSA for their review of this paper. Preparation of this paper was supported by the British Columbia Ministry of Health. REFERENCES 1. Lubchenco LO, Searls DT, Brazie JV. Neonatal mortality rate: Relationship to birth weight and gestational age. J Pediatrics 1972;81(4):814–22.

CONCLUSION

2. Koops BL, Morgan LJ, Battaglia FC. Neonatal mortality in relation to birth weight and gestational age: update. J Pediatrics 1982;101(6):969–77.

These gender-specific mortality grids enhance previously available comparisons by portraying neonatal mortality in narrow birth weight ranges at each week of gestation. They should provide health care professionals providing obstetrical care in Canada with clinically applicable, gender-specific neonatal risk indicators in terms of birth weight by gestational age.

3. The newborn infant. In: WHO Expert Committee on Physical Status: the use and interpretation of anthropometry: report of a WHO expert committee. Geneva: World Health Organization; 1995. WHO Technical Report Series: 854. Available at: http://whqlibdoc.who.int/trs/ WHO_TRS_854.pdf. Accessed April 7, 2007.

ACKNOWLEDGEMENTS

The authors wish to thank R. J. Danderfer, formerly Executive Director of BCVSA and currently Assistant Deputy Minister of Knowledge, Management and Technology, Ministry of Health, under whose auspices the original study was conducted. The authors also wish to recognize other authors of the 2004 BCVSA report for their contributions to the original versions of these grids. We are indebted to the Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System, particularly Dr M. S. Kramer and Dr R. Liston for their comments and advice. Appreciation is also extended to A.K. McBride, current Chief

4. Thomas P, Peabody J, Turnier V, Clark R. A new look at intrauterine growth and the impact of race, altitude, and gender. Pediatrics 2000;106(2):e21–6. 5. Alexander GR, Kogan M, Bader D, Carlo W, Allen M, Mor J. US birth weight/gestational age-specific neonatal mortality: 1995–1997 rates for whites, hispanics, and blacks. Pediatrics 2003;111(1):e61–6. 6. Joseph KS, Wilkins R, Dodds L, Allen VM, Ohlsson A, Marcoux S, et al. Customized birth weight for gestational age standards: Perinatal mortality patterns are consistent with separate standards for males and females but not for blacks and whites. BMC Pregnancy Childbirth. 2005;5(1):3. 7. Kierans WJ, Collison MA, Foster LT, Uh S-H. Charting birth outcome in British Columbia: determinants of optimal health and ultimate risk. Victoria, BC: British Columbia Vital Statistics Agency; 1993. 8. Kierans WJ, Kramer MS, Wilkins R, Liston R, Foster L, Uh S-H, et al. Charting birth outcome in British Columbia: determinants of optimal health and ultimate risk—an expansion and update. Victoria, BC: British Columbia Vital Statistics Agency; 2004. Available at: http://www.vs.gov.bc.ca/stats/ features/index.html. Accessed September 10, 2006. 9. Anath CV, Wen SW. Trends in fetal growth among singleton gestations in the United States and Canada, 1985 through 1998. Semin Perinatol 2002;26(4):260–7.

JULY JOGC JUILLET 2007 l

573

OBSTETRICS

10. Simchen MJ, Beiner ME, Strauss-Liviathan N, Dulitzky M, Kuint J, Mashiach S, et al. Neonatal outcome in growth-restricted versus appropriately grown preterm infants. Am J Perinatol 2000;17(4):187–92. 11. Boulet SL, Alexander GR, Salihu HM, Kirby RS, Carlo WA. Fetal growth risk curves: defining levels of fetal growth restriction by neonatal death risk. Am J Obstet Gynecol 2006;195(6):1571–7. 12. Doctor BA, O’Riordan MA, Kirchner HL, Shah D, Hack M. Perinatal correlates and neonatal outcomes of small for gestational age infants born at term gestation. Am J Obstet Gynecol 2001;185(3):652–9. 13. Garite TJ, Clark R, Thorp JA. Intrauterine growth restriction increases morbidity and mortality among premature neonates. Am J Obstet Gynecol 2004;191(2):481–7. 14. Piper JM, Xenakis EM, McFarland M, Elliott BD, Berkus MD, Langer O. Do growth-retarded premature infants have different rates of perinatal morbidity and mortality than appropriately grown premature infants? Obstet Gynecol 1996;87(2):169–74. 15. Hediger ML, Overpeck MD, McGlynn A, Kuczmarski RJ, Maurer KR, Davis WW. Growth and fatness at three to six years of age of children born small- or large-for-gestational age. Pediatrics 1999;104(3):e33. 16. Seeds JW, Peng TC. Does augmented growth impose an increased risk of fetal death? Am J Obstet Gynecol 2000;183(2):316–23. 17. Wen SW, Kramer MS, Platt R, Demissie K, Joseph KS, Liu S, et al. Secular trends of fetal growth in Canada, 1981 to 1997. Paediatr Perinat Epidemiol 2003;17(4):347–54.

574

l JULY JOGC JUILLET 2007

18. Burr KF, McKee B, Foster LT, Nault F. Interprovincial Data Requirements for Local Health Indicators: The British Columbia Experience. Health Reports 1995;7(2),17–24. 19. Danderfer RJ. Selected Vital Statistics and Health Status Indicators: One Hundred Twenty-ninth Annual Report 2000 Division of Vital Statistics. Victoria BC: British Columbia Vital Statistics Agency;2000. 20. Kramer M, McLean F, Boyd M, Usher R. The validity of gestational age estimation by menstrual dating in term, preterm, postterm gestations. JAMA 1988;260:3306–8. 21. Joseph KS, Kramer MS, Allen AC, Cyr M, Fair M, Ohlsson A, et al. Gestational age- and birth weight-specific declines in infant mortality in Canada, 1985–94. Paediatr Perinat Epidemiol 2000;14:332–9. 22. Kramer MS, Demissie K, Yang H, Platt RW, Sauve R, Liston R. The contribution of mild and moderate preterm birth to infant mortality. Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System. JAMA 2000;284(7):843–9. 23. Jones HP, Karuri S, Cronin CM, Ohlsson A, Peliowski A, Synnes A, et al. Actuarial survival of a large Canadian cohort of preterm infants. BMC Pediatr 2005;5:40. 24. Statistics Canada, Canadian Vital Statistics 2003, Birth and Death Databases. Ottawa: CANSIM Table 102–0507.