Volume 23
Number 1
2002

[Table of Contents]

Under-reporting of maternal mortality in Canada: A question of definition

Abstract

In Canada, maternal mortality reporting is based on information contained on death certificates. To examine the extent to which maternal deaths are under-reported in Canada and whether this is likely to change under the 10th revision of the International Classification of Diseases (ICD), we linked live birth and stillbirth registrations to death registrations of women aged 10 to 50 for 1988 through 1992. We reviewed the death certificates of women found to have died while pregnant or within a year of the termination of pregnancy. The officially reported maternal mortality ratio for the study years was 3.7 deaths per 100,000 live births. Depending on whether we included deaths where the certifying physician did not list pregnancy as a contributing factor on the death certificate, revised ratios under ICD-9 ranged from 4.9 to 5.1 per 100,000 live births for deaths from direct obstetric causes and from 0.5 to 1.2 per 100,000 live births for deaths from indirect obstetric causes. Reflecting changes in classification criteria, revised ratios under ICD-10 were lower than those under ICD-9 for deaths from direct obstetric causes - ranging from 3.9 to 4.1 per 100,000 live births - and higher for deaths from indirect obstetric causes - ranging from 2.0 to 3.0 per 100,000 live births. Of deaths from direct obstetric causes, those from cerebrovascular disease were the most numerous and also the most likely to be underreported. Deaths from pulmonary embolism and indirect obstetric causes were the next most likely to be underreported. In a companion article we report an investigation on whether deaths from causes not directly related to pregnancy - such as injury, infectious disease and epilepsy - are more or less likely to occur among pregnant and recently pregnant women.

Key words: definition of maternal mortality, maternal mortality, surveillance of maternal mortality

Introduction

Less than 100 years ago in North America, childbirth was a leading cause of death among young women, second only to tuberculosis.¹ Over the course of the last century, maternal deaths have become rare events in Canada and other industrialized countries. However, maternal mortality continues to be a key health indicator around the world,² and maternal mortality ratios are still routinely compared, as shown in Figure 1. Low levels of maternal mortality in Canada reflect the general good health of our population, our universal access to medical care, and the status we accord to women and their health care needs.

Despite our good record, every maternal death is a cause for concern. Approximately 15 maternal deaths are reported in Canada each year. Those that occur in hospital are usually the subject of thorough investigation by review committees in the hospitals where the deaths occurred, but some deaths that could be "maternal" may not be included in the officially reported counts. Researchers in other countries, often using definitions of maternal death that are broader than the definition used by vital registrars, have suggested that up to twice as many maternal deaths occur as are reported.^3-7

Classification of deaths as maternal is based on information contained on death certificates. Reasons for underreporting of maternal deaths include improper completion of death certificates and errors in coding the underlying cause of death. Death certification and classification by cause involve many steps (Appendix 1).

Before 2000, deaths in Canada were classified according to the 9th revision of the International Classification of Diseases (ICD-9);⁸ since 2000, deaths have been classified according to the 10th revision (ICD-10).⁹ Reflecting changes proposed over the years, the ICD-10 definition of maternal mortality is more comprehensive than its predecessors:^10,11 more causes of death are classified under maternal mortality, and two new categories have been added - "late maternal," which includes deaths that occur beyond the traditionally defined 42-day postpartum period, and "pregnancy-related," which includes all deaths around the time of pregnancy regardless of cause. These changes and additions are described in more detail below and in Appendix 2.

The primary purpose of the investigation reported here was to determine whether maternal mortality is under-reported in Canada and to determine reasons for any omissions. Another purpose was to explore the effect of changes in the definition of maternal death under ICD-10 on the understanding and reporting of maternal mortality.

FIGURE 1
Maternal mortality ratios in selected countries (1999 estimates)²

Methods

Ascertainment of previously unreported maternal deaths using record linkage

To identify deaths that occurred among pregnant or recently pregnant women, we linked live birth and stillbirth records to records of deaths among women of reproductive age that occurred within 365 days of a registered birth. Some deaths that occur around the time of pregnancy do not link to a birth registration, however. There will be no birth registration if the death occurred very early in pregnancy or the woman died undelivered. In other cases, a birth may not be registered or the birth registration may be missing. We therefore also requested the death certificates of women whose death had been classified as maternal (i.e., assigned an ICD-9 Chapter 11 code, see Appendix 1) but whose death records did not link to a birth. We were not able to identify women who died while pregnant or within 365 days of the termination of pregnancy if the death had not been classified as maternal and there was no birth registration.

To prepare for the linkage, we extracted to a separate file all records of deaths in the Canadian Mortality Data Base that occurred between January 1, 1988, and December 31, 1992, in females 10 to 50 years of age. We then extracted from the Canadian Birth Data Base all records of live births and stillbirths between January 1, 1987, and December 31, 1992. We excluded deaths from 1988 through 1990 and births from 1987 through 1990 in Newfoundland, due to under-counting of births in Newfoundland for those years.

To link the birth files to the death file, we used the mainframe version of the Generalized Record Linkage System (GRLS V1).¹² We generated an alternative entry if the mother's maiden surname field was different from the infant's. Surnames and maiden surnames were also assigned phonetic codes in case the names were misspelled. The mother's identifying information on the live birth/stillbirth file was compared with the decedent's identifying information on the death file. Identifying information included surname, maiden surname, given names or initials, date of birth, place of birth, marital status, place of event, and place of residence. We also assigned "weights" to each pair of records to reflect the probability that the computerized pairs of records represented the same person.¹³ Pairs of records with weights above a pre-determined threshold were considered potentially good links. Computer printouts listing linkage identifiers and other items (e.g., street address, spouse's given names) were generated for all potentially good links and were manually reviewed. If we could not determine from the information on the computer printout that the decedent was the same person as the mother listed on the birth registration, additional sets of identifying information contained on the birth and death registration forms were compared. False-positive links identified in this way were removed.

Definition of maternal death under ICD-9 and ICD-10

Shown in Appendix 2 are the ICD-9 and ICD-10 definitions of maternal death and its two subcategories, termed "direct" and "indirect" obstetric death. ICD-9 Chapter 11 (Complications of Pregnancy, Childbirth and the Puerperium) comprises codes between 630.0 and 676.9. All deaths for which the underlying cause has been assigned a code in this range are designated "direct" obstetric deaths, with the exception of deaths assigned codes between 647.0 and 648.9. These codes are designated "indirect" obstetric deaths and include deaths from causes that would otherwise be coded to other ICD-9 chapters (see Appendix 3). Chapter 11 includes codes for conditions such as eclampsia, postpartum hemorrhage and amniotic fluid embolism. These deaths are clearly obstetric deaths: they could only occur during pregnancy or around the time of childbirth. However, ICD-9 Chapter 11 also specifies that deaths from pulmonary embolism and cerebrovascular disorders - conditions that are not uniquely associated with pregnancy and childbirth - are classifiable as "direct" obstetric deaths if they occur during pregnancy or the postpartum period (within 42 days of the termination of pregnancy). More specifically, a death from pulmonary embolism is to be assigned a Chapter 11 code only if pregnancy has been listed as a contributing factor in Part 1 or Part 2 of the medical certificate of death, whereas any death from a cerebrovascular disorder during pregnancy or the postpartum period is to be classified as a direct obstetric death, regardless of whether the certifying physician listed pregnancy as a contributing factor.

Changes in the classification and designation of death as maternal death under ICD-10

Several changes in ICD-10 affect reporting of deaths as maternal deaths. First, the list of underlying causes of death included under the rubric "indirect" has been expanded to include all causes other than perinatal conditions, and injury and poisoning. Second, deaths from cerebrovascular disease during pregnancy or the postpartum period are classified as indirect rather than direct obstetric deaths.

Two new categories of "maternal mortality" are defined in ICD-10 (see Appendix 2). The first, "late maternal death," includes deaths from direct or indirect obstetric causes occurring more than 42 days but less than one year postpartum. A second, very broadly defined category, "pregnancy-related death," includes all deaths that occur during pregnancy or the postpartum period regardless of cause.⁸

Identification of maternal deaths

Two obstetricians (RK and RL) and a third medical expert reviewed the death certificates, but in cases where the underlying cause of death was given as cancer, injury or poisoning, they reviewed only those certificates on which there was a notation referring to pregnancy. Reviewers were informed of the number of days after the pregnancy outcome that each death occurred.

Provincial registrars provided revised certificates of death for 29 of 33 certificates marked "interim," indicating that an underlying cause had not been determined at the time the certificate was originally sent to Statistics Canada. A medical coder from Statistics Canada assigned a code to the underlying cause for 27 of these deaths. A specific underlying cause had not been determined for two deaths, although these deaths were known to have been caused by trauma. We included revised certificates in the review process if they qualified according to the above criteria.

To assess the reliability of judging relatedness to pregnancy using only death certificate information and knowledge of the timing of death in relation to pregnancy, two of the three reviewers independently made an initial assessment of each selected death certificate. Reviewers agreed in 92% of the cases as to whether the death certificates contained sufficient information to make a judgement; in 97% of cases judged to have sufficient information they further agreed as to whether the death was related to pregnancy.

Reviewers assigned corrected codes of underlying causes where they deemed appropriate. The record linkage and review of confidential death data were carried out at Statistics Canada.

Results

We identified 633 deaths that had occurred within 365 days of a pregnancy outcome. During the study years, 72 deaths had been assigned codes under ICD-9 Chapter 11 and reported as maternal deaths. We found that two of these had occurred more than 42 days postpartum and therefore were not classifiable as maternal deaths under ICD-9. Our capture strategy also missed three deaths that had been reported as maternal. For 11 of the 70 correctly reported maternal deaths, we found no corresponding birth registration.

The results are summarized separately for the two categories of reportable maternal death: direct and indirect obstetric deaths. Because the causes of death classified as direct and indirect obstetric death differ under ICD-9 and ICD-10, differences in case ascertainment under each of these two classification systems are given.

For interest, we also report the numbers of deaths that would be included in the two new ICD-10 categories "late maternal" and "pregnancy-related" death. Numbers of deaths in these categories, however, are not reported by the vital records system.

Identification of unreported direct obstetric deaths and reclassification of reported direct obstetric deaths

Shown in Table 1 are all direct obstetric deaths that occurred during the study years by cause and source of ascertainment. Of the 70 correctly reported maternal deaths, 66 were classifiable as direct obstetric deaths under ICD-9 and are shown in Table 1. Four of the 70 were classifiable as indirect obstetric deaths and are included in Table 3. Our review process netted 33 additional deaths that reviewers agreed should have been classified as direct obstetric deaths under ICD-9. Note that deaths from cerebrovascular disorders would be classified as indirect obstetric deaths under ICD-10.

Reviewers also judged that 16 deaths originally assigned a cause of death code in Chapter 11 had been miscoded. They therefore assigned a new code, but all newly assigned codes were within the range of codes designating obstetric deaths. This code re-assignment resulted in fewer deaths in some categories and more deaths in other categories. These category shifts are shown in column two of Table 1.

Table 2 summarizes possible reasons why the 33 newly ascertained direct obstetric deaths had not been originally so classified.

Identification of unreported indirect obstetric deaths

Table 3 lists the deaths that reviewers judged to be indirect obstetric deaths if the death was eligible to be so classified under ICD-9 or ICD-10. Only four of the 70 correctly reported maternal deaths had been originally classified and reported as indirect obstetric deaths.

Identification of deaths under the newly defined ICD-10 category "late maternal" death and the newly defined concept "pregnancy-related" death

As stated above, two of the 72 deaths reported by the vital records system as maternal deaths occurred more than 42 days postpartum. These deaths are actually over-counts according to ICD-9 but would be included in the new ICD-10 category "late maternal" death. Our review process identified two additional late maternal deaths from direct obstetric causes and four from indirect obstetric causes.

The newly created category under ICD-10, "pregnancy-related death," included more deaths than the categories "direct" and "indirect" obstetric death. Of the total 633 deaths identified, 187 (29.5%) occurred (or were assumed to have occurred) during pregnancy or within 42 days of its outcome. All 187 would be included in this new ICD-10 category, including 34 deaths from injury or poisoning and 12 deaths from cancer (see Appendix 2).

The magnitude of under-reporting of maternal mortality under ICD-9 versus ICD-10

Table 4 shows ranges of maternal mortality ratios from causes classifiable as direct and indirect obstetric deaths and the differences in each under ICD-9 and ICD-10. Maximum values include all deaths that reviewers retrospectively judged to be obstetric deaths eligible under each ICD version. Minimum values exclude unreported deaths retrospectively judged to have been maternal deaths but for which the certifying physician had not listed pregnancy as a contributing factor in Part 1 or Part 2 of the medical certificate of death, a requirement for the death to be classified as maternal by medical coders.

Discussion

Our main findings were a striking under-reporting of deaths from cerebrovascular disorders, pulmonary embolism and causes indirectly related to pregnancy. Approximately two-thirds of direct obstetric deaths that were not reported were associated with cerebrovascular disorders or pulmonary embolisms. In fact, cerebrovascular disorders became the most frequent category of obstetric death, as well as the most likely to be under-reported. Classification of deaths from this cause under ICD-9 is unclear.⁴ Moreover, deaths from cerebrovascular disorders are to be classified as indirect obstetric deaths under ICD-10. If we do not include deaths from cerebrovascular disorders as direct obstetric deaths, then direct obstetric deaths were under-reported by approximately 20%, within the range of major classification errors reported for death certification in general.¹⁴ In a companion article, we discuss our further findings with respect to whether deaths during pregnancy and the postpartum period from cerebrovascular disorders should be classified as direct obstetric deaths.

The category "indirect obstetric death" was introduced in ICD-9. Although this revision was published in 1975, few deaths are so classified, even when physicians clearly list pregnancy as a contributing factor on the death certificate. Given that few are reported and that the judgement on what constitutes an indirect obstetric death is subjective - in contrast to most direct obstetric deaths, which clearly would not have occurred had the woman not been pregnant - the utility of this category is questionable.

Deaths from cerebrovascular disorders may not have been captured as direct obstetric deaths because physicians who certify death certificates and medical coders may have been unaware that, under ICD-9, all deaths from this cause during pregnancy or the postpartum period were to be classified as direct obstetric deaths. Deaths from pulmonary embolism may not be classified as maternal deaths because physicians are not aware that in Part 1 or Part 2 of the medical certificate of death they must clearly state, if they so believe, that pregnancy was a contributing factor in the death.

We also found that deaths from direct obstetric causes tended to be somewhat misclassified within the major ICD-9 heading for pregnancy and childbirth. However, after reclassifying these deaths and including newly ascertained direct obstetric deaths, we found that the most frequent causes of direct obstetric death (other than deaths from cerebrovascular disorders) remained essentially the same - hypertensive disorders of pregnancy, amniotic fluid embolism, and postpartum hemorrhage. Deaths from other types of pulmonary embolism within 42 days of a pregnancy outcome ranked in the top four direct obstetric cause-of-death categories after our review, although only two such deaths had been reported previously for the study years. Deaths from this cause tended not to have been reported as maternal deaths or to have been misclassified under other subcategories of direct obstetric deaths.

The magnitude of under-reporting of maternal deaths depended in part on whether we included deaths retrospectively judged to have been maternal deaths even though the physician who completed the death certificate had not listed pregnancy as a contributing factor. In some cases, the physician may not have been aware that the woman had recently been pregnant. In other cases, however, a separate question on the death certificate as to whether the decedent had been pregnant within the preceding 42 days had been appropriately answered "yes". In these latter cases, we could not determine whether the certifying physician omitted to list pregnancy as a contributing factor in Part 1 or Part 2 of the medical certificate of death because of neglect, or because he or she did not believe pregnancy to have been a contributing factor in the death.

Although a similar magnitude of under-reporting of maternal mortality has been documented in the United States and Europe^3-7 even before implementation of ICD-9,¹⁵ previous studies have been inconsistent with respect to what constitutes maternal death. Previous investigators have included as unreported many deaths not classifiable as obstetric under the ICD-9 definition, including deaths from injuries sustained in motor vehicle collisions, suicide, and deaths occurring more than 42 days postpartum.^4-6,16 Moreover, most investigators have categorized unreported deaths from cerebrovascular disorders as indirect rather than direct obstetric deaths, despite specification in the ICD-9 coding manual.^3,6,17-20

Limitations and generalizability of the findings

Unless the underlying cause of death was initially classified as obstetric, we were not able to identify deaths among women who may have been pregnant at the time of death if the pregnancy did not result in a birth that was registered. This would apply to most women who died before 20 weeks' gestation from causes other than those directly related to pregnancy and to women who died outside of hospital and there was no attempt to deliver the fetus. Additionally, some births may not have been registered or the registration may have been missing. We found no birth registration corresponding to 11 of the 72 reported maternal deaths.

The problems we encountered with respect to death certificate completion and ambiguities of classification, including uncertainties about what constitutes maternal death, may be generalizable to other countries. These problems have been cited previously as a challenge not only to maternal death reporting²¹ but also to classification and reporting of underlying causes of death generally, with recommendations for more attention to physician training at the postgraduate level in the completion of death certificates.¹⁴

Future surveillance of maternal mortality in Canada

The existing reporting system appears satisfactory for annual reporting of deaths from direct obstetric causes except for deaths from cerebrovascular disorders. The problems we detected - some misclassification and delays in replacing interim death certificates with the final version - may be correctable.

Ascertainment of late maternal deaths and most indirect obstetric deaths requires a labour-intensive process of record linkage and expert review of death certificates. Ellerbrock and colleagues have suggested that it is important to capture deaths of women who experience catastrophic events during childbirth but who die while on life support beyond 42 days postpartum.¹¹ We found only one such death, however. In total, only six deaths judged to be obstetric that occurred beyond 42 days postpartum were unreported. To find these six deaths required obtaining and partially or fully reviewing 446 death certificates in addition to the 187 death certificates for women who died within the 42-day postpartum period.

The other newly introduced category of maternal death under ICD-10, "pregnancy-related," requires record linkage but not expert review, but resulting rates or ratios would include an unknown number of deaths clearly unrelated to pregnancy.

For deaths from indirect causes, judgements of relatedness to pregnancy are necessarily subjective, both for physicians who certify death certificates and for reviewers making retrospective judgements. Unlike deaths from most direct obstetric causes, deaths from indirect causes may have been coincidental to the pregnancy. As already explained, the ICD-9 definition of indirect obstetric death includes deaths in only some cause-of-death categories, whereas under ICD-10 all cause-of-death categories other than injury and poisoning have been included. We could find no published description of the decision-making process that led to this more inclusive definition or to the introduction of the concept of indirect obstetric death under ICD-9. Demonstrating that death is more likely to occur from certain causes among pregnant or recently pregnant women would contribute to an evidence-based rationale for monitoring particular causes of death among pregnant and recently pregnant women. In a companion article, we explore whether pregnant or recently pregnant Canadian women were more or less likely to die of specific causes than women of the same age not known to have been pregnant during the same period.

In spite of low maternal mortality ratios in industrialized countries, surveillance of maternal mortality continues to be of interest. There is a need, however, to resolve current misunderstandings with respect to classification of deaths indirectly related to pregnancy. Until this is accomplished, comparisons of maternal mortality ratios among countries might be most appropriately limited to comparisons of direct obstetric deaths.

Acknowledgements

The authors wish to acknowledge the contribution of the provincial and territorial vital statistics registrars

References

1. King CR. The New York maternal mortality study: a conflict of professionalization. Bull Hist Med 1991;65:476-502.

2. World Health Organization and UNICEF. Revised 1990 estimates of maternal mortality, a new approach by WHO and UNICEF. Geneva: WHO, 1996.

3. Ziskin LZ, Gregory M, Kreitzer M. Improved surveillance of maternal deaths. Int J Gynaecol Obstet 1979;16:281-6.

4. Bouvier-Colle M-H, Varnoux N, Costes P, Hatton F. Reasons for the underreporting of maternal mortality in France, as indicated by a survey of all deaths among women of childbearing age. Int J Epidemiol 1991;20:717-21.

5. Rubin G, McCarthy B, Shelton J, Rochat RW, Terry J. The risk of childbearing re-evaluated. Am J Public Health 1981;71:712-6.

6. Rochat RW, Koonin LM, Atrash HK, Jewett JF, and the maternal mortality collaborative. Maternal mortality in the United States: report from the maternal mortality collaborative. Obstet Gynecol 1988;72:91-7.

7. Dye TD, Gordon H, Held B, Tolliver NJ, Holmes AP. Retrospective maternal mortality case ascertainment in West Virginia, 1985 to 1989. Am J Obstet Gynecol 1992; 167:72-6.

8. World Health Organization. Manual of the international statistical classification of diseases, injuries, and causes of death, 9th revision, Vol. 1. Geneva: WHO, 1977.

9. World Health Organization. Manual of the international statistical classification of diseases, injuries, and causes of death, 10th revision, Vol. 1. Geneva: WHO, 1993.

10. Rochat RW, Rubin SL, Selik R, Sachs BP, Tyler CW. Changing the definition of maternal mortality: a new look at the postpartum interval. Lancet 1981;1(8224):831.

11. Ellerbrock TV, Atrash HK, Hogue CJR, Smith JC. Pregnancy mortality surveillance: a new initiative. Contemp OB/Gyn 1988;June:23-34.

12. Newcombe HB. Handbook of record linkage: methods for health and statistical studies, administration and business. Oxford, U.K.: Oxford University Press, 1988.

13. Fellegi IP, Sunter AB. A theory of record linkage. JASA 1969;40:1183-210.

14. Myers KA, Farquhar DRE. Improving the accuracy of death certification. Can Med Assoc J 1998;158:1317-23.

15. Barno A, Freeman DW, Bellville TP. Minnesota maternal mortality study: five-year general summary. Obstet Gynecol 1957; 9:336-44.

16. Steele R. Why mothers die: confidential enquiry into maternal deaths. RCM Midwives Journal 1999;2:80-1.

17. Henry OA, Sheedy MT, Beischer NA. When is a maternal death a maternal death? A review of maternal deaths at the Mercy Maternity Hospital, Melbourne. Med J Aust 1989;151:628-31

18. Högberg U, Innala E, Sandström A. Maternal mortality in Sweden, 1980-1988. Obstet Gynecol 1994;84:240-4.

19. Gissler M, Kauppila R, Meriläinen, Toukomaa H, Himminki E. Pregnancy-associated deaths in Finland 1987-1994 - definition problems and benefits of record linkage. Acta Obstet Gynaecol Scand 1997;76:651-7.

20. Benedetti TJ, Starzyk P, Frost F. Maternal deaths in Washington State. Obstet Gynecol 1985;66:99-101.

21. Zemach R. What the vital statistics system can and cannot do. Am J Public Health 1984;74:756-58.

Author References

Linda A Turner, Bureau of Reproductive and Child Health, Health Canada, Ottawa, Ontario

Margaret Cyr, Martha Fair, Occupational and Environmental Health Research Section, Health Statistics Division, Statistics Canada, Ottawa, Ontario

Robert A H Kinch, Department of Obstetrics and Gynecology, McGill University, Montreal, Quebec

Robert Liston, BC Women's Hospital, Vancouver, BC

Michael S Kramer, Department of Pediatrics and Department of Epidemiology and Biostatistics, McGill University Faculty of Medicine, Montreal, Quebec; Distinguished Scientist of the Canadian Institutes of Health Research

Maureen Heaman, Faculty of Nursing, University of Manitoba, Winnipeg, Manitoba for the Maternal Mortality and Morbidity Study Group of the Canadian Perinatal Surveillance System

Correspondence: Dr Linda A Turner, Canadian Institute for Health Information, 90 Eglinton Ave E, Suite 300, Toronto, Ontario, M4P 2Y3; Fax: (416) 481-2950

[Previous] [Table of Contents] [Next]