—from Epidemiological Bulletin, Vol. 24 No. 4, December 2003—

On the Estimation of Mortality Rates For Countries of the Americas

Introduction
It is well known that statistics derived from registered mortality can be affected during any of the phases in their production: from collection of data and completion of forms, coding, data processing, to their subsequent enumeration. Indicators produced from this information (such as numbers of death and distribution of cases by cause) that have a role in the creation of rates can be altered in both the numerator and the denominator. Therefore, knowledge of the environment in which mortality statistics are produced and the problems that arise when producing them is indispensable for their correct interpretation and use. This knowledge allows for application of procedures to correct problems and improve the quality and credibility of the statistics.

Errors in collecting and processing databases can also give rise to problems that can be apparent only when data comparisons and their trends are studied. This implies a certain degree of knowledge in the field and a regular use of data. Estimation of rates requires a denominator that corresponds to the population by age groups on the one hand and to the registered live births, which are a part of maternal and child mortality rates, on the other hand. The population estimate for inter-census years is taken from projections, which could inadequately represent migration problems faced by some countries. Live births statistics also have some problems, the most important of which is extemporaneous registration of births. Consequently, observed maternal and child mortality rates will differ from actual rates if late registration of births and non-registration of births and deaths are not accounted for.

The quality of cause-specific mortality data is also affected by limitations in current medical knowledge, diagnostic errors, deficiencies of certification, and perhaps to a lesser extent, coding and other processing errors. The validity of the distribution by cause also is affected by under-registration of deaths. Cause of death certification, even when done by attending physicians, is often incomplete or of low quality for reasons such as lack of training on proper certification and insufficient understanding of the uses made of the information provided on the death certificate. Another problem frequently encountered is that physicians may prefer certain kinds of diagnoses, such as the ones in their specialty area; this bias may vary from country to country and over time. In many developing countries a sizable segment of the population lacks access to medical care. Consequently, non-attending physicians, who may have insufficient information for a diagnosis, may sign death certificates and non-medical witnesses may provide death reports. Both developing and developed countries face some of the same problems. For example, legal, societal, and other reasons may lead to the under-reporting of causes of a sensitive nature, such as suicide or HIV/AIDS, on the death certificate. Moreover, physicians often do not understand how to adequately fill out the death certificate, especially in relation to the identification of direct, intervening, and underlying causes. Furthermore, the selection of a single underlying cause of death is often problematic in elderly decedents, who often suffer from several chronic diseases that concurrently lead to death.

Clearly, there is a real need to educate the public, physicians, and health sector decision-makers about both the importance of accurate and complete reporting on the death certificate and the impact of erroneous reporting on aggregate mortality statistics. Practices differ from country to country as to whether deaths without medical certification are included or not on tabulations of deaths by cause. A World Health Organization (WHO) provision specifies that when deaths without medical certification constitute less than 2% of the total, they should be included in such tabulations under the category “ill-defined cause;” when they exceed this percentage, they should be tabulated separately. Countries sometimes apply different criteria, however. Deaths without medical certification are sometimes included in the national cause of death tabulations as follows: under codes 798.9 [International Classification of Diseases, Ninth Revision (ICD-9)](1) or R98 (ICD-10)(2), “unattended death,” when the cause of death is not external but is unknown due to the lack of medical care at death or during the illness or condition leading to death; or under codes 799.9 (ICD-9) or R99 (ICD-10), “other unknown and unspecified cause of mortality”. For medically certified cause of death data, the simplest indicator of quality is the proportion of deaths assigned to “symptoms, signs, and ill-defined conditions” (SSI), codes 780-799 (ICD-9) and R00-R99 (ICD-10). The “unknown” causes of death assigned to 798.9 and R98, or 799.9 and R99 account for a large proportion of deaths attributed to SSI, since most of these are without medical certification. Where registration coverage is incomplete, however, the proportion of deaths assigned to SSI will usually increase as coverage increases, without there having been a real drop in the quality of medical cause of death certification. In fact, under both ICD-9 and ICD-10, the magnitude of the proportion of deaths assigned to SSI is a lower bound estimate on the proportion of deaths from ill-defined causes, because a number of “defined” ICD-9 and ICD-10 categories, such as cardiac arrest and heart failure, lack diagnostic meaning. It should also be noted that deaths from “defined” causes are not necessarily “well” defined; they are subject to diagnostic, certification, and coding errors that cannot be detected after statistics are compiled. For most countries the proportion of deaths assigned to the category SSI, in combination with the proportion of deaths certified by attending and non-attending physicians, is useful for monitoring trends and differentials in access to medical care. Table 1 shows, by country, the total number of registered deaths and the percentage of deaths assigned to SSI around 2000 (or for the latest 3 data years available). In 21 countries of the Americas, less than 5.0 % of registered deaths were assigned to SSI around 2000.

Effect of the change of ICD revisions on mortality data
The introduction of the Tenth Revision of the ICD in the Americas, starting in 1996, marked the most sweeping changes in the Classification since the Sixth Revision was introduced in 1949 and reflects a conceptual shift in structure and content from previous revisions. Although each revision has produced some breaks in the comparability of cause of death statistics, the change from the Ninth Revision, in use since 1979, to the Tenth Revision, has had many consequences on the coding of mortality. The ICD-10 has considerably greater detail than ICD-9 (almost twice the number of codes); and includes shifts of inclusion terms and titles from one category, section, or chapter to another; new cause of death titles and corresponding cause of death codes and sections; regroupings of diseases; and changes in the coding rules to select the underlying cause of death. All of these result in a number of discontinuities in the comparability of cause of death statistics over time or in historical series. These discontinuities are best assessed at the national level from the analysis of the results of double-coding (or bridge-coding) studies on national data and observing comparability ratios.

Comparability ratios are derived from the dual classification of the underlying cause of death on mortality records for a single year, classified under the new revision and under the previous revision. They are calculated by dividing the number of deaths for a selected cause classified under the new revision by the number of deaths to the most comparable cause classified under the previous revision. A ratio of 1.0 indicates that the same number of deaths was classified to a particular cause or combination of causes regardless of the revision used; it does not necessarily mean that the cause was unaffected by changes in classification and coding procedures but that there was no net change. A ratio greater than 1.0 indicates that more deaths were assigned to a cause in ICD-10 than the comparable cause in ICD-9 and a ratio less than 1.0 indicates fewer deaths were assigned to a cause in ICD-10 than the comparable cause in ICD-9.

Completeness of Data
In many countries of the Americas, the coverage of the civil registration system is incomplete, and in some countries the population covered by available mortality data needs to be further clarified. Within countries, the completeness of registration is known to vary according to geographic area and age group. Registration of vital events is less complete in rural areas than in cities and, in general, is worse in areas with poor living conditions. Table 1 shows the estimated under registration of deaths in countries of the Americas around 2000. The estimates are based on a comparison of the crude death rates obtained using registered mortality, as reported to PAHO for the three-year period indicated, and the death rates estimated by using abridged life table central death rates (see section on estimation of death rates by cause, age and sex), where available, or from death rates estimated by the Population Division of the United Nations.(3)

Differences among countries in the time period used for calculation of registered death rates reflect differences in the availability of data from countries at the time the table was prepared. Country-wide registered mortality data are not available from Bolivia, Honduras, Netherlands Antilles and only for recent years and with limited coverage from Haiti. The estimates shown in Table 1 provide an indication of the magnitude of the existing under registration problem in the countries. The characteristics of, and underlying reasons for, under registration of deaths vary greatly among countries and also within each country. As can be seen in the table, there is little or no under registration in Anguilla, Antigua, Argentina, Barbados, Belize, Canada, Chile, Costa Rica, Cuba, Dominica, Guadeloupe, Martinique, Saint Lucia, Saint Vincent, and Trinidad and Tobago, the United States, Uruguay, Venezuela, and the Virgin Islands (USA). In these countries, the registered rate for the period shown is identical to, and sometimes greater than, the estimated rate for the quinquennium that contains the period. Under-registration is low in Puerto Rico (5.1%) and intermediate in Brazil, Guatemala, Mexico, Panama, and Suriname, which have estimated under-registration ranging between 13% and 19%, and appear to be on the way to achieving satisfactory levels of death registration. Another 11 countries continue to have serious under-registration problems, with estimates ranging between 20% and 92%. The level of under registration is unknown in 7 countries – Bermuda, Cayman Islands, Grenada, Montserrat, St. Kitts and Nevis, Turks and Caicos Islands and the Virgin Islands (UK). No data from civil registration sources are available for Bolivia, Honduras and Netherlands Antilles in recent years. Under registration is greater for infant deaths than for deaths occurring at older ages. Infants who live just a few hours or days may not be registered as either live births or infant deaths. At advanced ages there tends to be overstatement of age, which contributes to under estimation of mortality for some adult age groups and over estimation for older groups. Clustering of deaths in certain ages due to reporting preferences (such as ages ending in 0 or 5) is another well-known phenomenon that affects the age distribution of registered deaths.

Estimation of Death Rates by Cause, Age and Sex
In view of the above limitations in the coverage of civil registration systems and in the “quality” of mortality data as indicated by the proportion of deaths assigned to the category “signs, symptoms and ill-defined conditions,” a general method to more accurately estimate mortality rates that addressed these limitations was required.

Estimation of mortality rates in PAHO is based on an estimation procedure first presented in the 1992 edition of Health Statistics from the Americas.(4) This procedure was updated to proportionately re-assign deaths not stated by age and sex and is described in the following paragraphs as well as in the 2003 edition of that publication, which is available on-line at www.paho.org.(5)

Assumptions and methodology
The procedure uses registered mortality data available in the PAHO regional mortality database. The data is tabulated for selected year(s), causes of death, age groups, and sex. The estimates of the central death rates (nMx) for the corresponding age groups and sex are obtained from life tables for 20 Latin American countries prepared and published by the Latin American and Caribbean Demographic Center (CELADE)(3) [For English speaking countries of the Caribbean, Canada, Puerto Rico, and United States, registered rates available from the PAHO database were used]; and corresponding annual population estimates by age groups and sex. The registered mortality data is first adjusted for deaths unknown by age and sex. The number of deaths unknown by age are redistributed into known age groups by multiplying the number of deaths for each sex and age group by an adjustment factor, fa = D/Da, where D is the total number of deaths and Da is the number of deaths stated by age. A similar adjustment factor is used to redistribute the number of deaths in each age group not stated by sex.

The rate calculations make the following assumptions about the cause distribution of registered mortality data:

(a) All registered deaths coded to an external cause were in fact due to an external cause, and none of the registered deaths coded to other cause categories, including SSI, were really due to external causes. Consequently, all deaths assigned to SSI can be proportionately redistributed among other non-external cause categories, age groups, and sex, under the assumption that the SSI deaths follow the same distribution as that observed among registered deaths from non-external “defined” causes.

(b) An estimate of the total number of deaths that actually occurred in a given year or time period is obtained by applying the corresponding quinquennial central death rates for each age and sex group from the life table to the population estimates and totaling the number of deaths in each age group by sex. By subtracting the number of registered deaths, an estimate of the number of unregistered deaths is obtained. It is further assumed that the distribution of unregistered deaths into cause categories, by age group and sex, is the same as that among registered deaths. Accordingly, unregistered deaths, including unregistered deaths due to external causes, are redistributed into corresponding cause categories by age and sex in the same proportions as the registered deaths.

Estimated age and sex specific rates are calculated by accumulating the estimated total deaths (registered and unregistered) in a given year or time period, by cause category and dividing by the sum of the corresponding estimated populations. The infant mortality rate is calculated using the estimated number of live births, if available. Otherwise, the estimated population under 1 year of age is used in the denominator.

The estimated number of deaths for a selected age-sex group, d’i and the country’s total estimated deaths, D’ annually or for a given time period are defined in Box 1, as well as the estimated number of unregistered deaths, d’iU in the ith age-sex group. The proportion of unregistered deaths due to external causes for the ith age-sex group d”iex and the estimated total number of deaths due to external causes in the ith age-sex group d’iex are also shown.

The estimated total number of deaths, d’ic, for a selected cause category, c and age-sex group i, can be calculated from the above. The second expression in the equation for d’ic presented in box 1 reflects the proportionate redistribution of registered SSI deaths and unregistered deaths due to non-external causes in the ith age-sex group that will be re-assigned to cause category c. By accumulating the estimated deaths in each age-sex cause grouping, the total estimated number of deaths can be determined.

Some limitations
In some instances, the number of registered deaths for a given year or time period was greater than the estimate obtained from the CELADE life tables. This indicates that the central death rate estimates of the life table for that country and time period do not adequately reflect the observed age patterns of mortality. In those instances and in countries where life table estimates are not available, the registered mortality data, adjusted for unknown age and sex, is used in estimating the rates. In effect, this assumes that there is no under registration present in that year or time period.

Since PAHO uses CELADE as its primary source for life tables, this information is not available for the English-speaking countries of the Caribbean, Canada, Puerto Rico and United States. Other sources of life table information could be consulted including the use of national life tables and model life tables and the feasibility of their use studied. The US Census Bureau’s International database (www.census.gov/ipc/www/idbacc.html) also has this data for a few Caribbean countries (Guadeloupe, Martinique, St. Kitts and Nevis, Saint Lucia, and Trinidad and Tobago) but only for a year around 1980.

The estimation of rates utilizing this methodology is dependent on having suitable life tables that accurately account for a country’s mortality patterns and can be used to assess the level of completeness of a country’s vital registration system. It also is dependent on the accuracy in selecting and coding the underlying cause of death and on assumptions for the re-distribution of the cause category SSI and “unregistered” deaths to the cause of death structure for registered deaths. It is assumed that the registered deaths have negligible misclassification of the underlying cause of death.

References
(1) World Health Organization. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death, Ninth Revision (1975), Geneva, WHO, 1975. (Vol. 1 & 2).
(2) World Health Organization. International Statistical Classification of Diseases and Related Health Problems. Tenth Revision. Vols. 1-3. Geneva, WHO, 1992-1994.
(3) CELADE. Latin America: Life Tables 1950-2025. Demographic Bulletin (Santiago), 2001(Jan); 67.
(4) Pan American Health Organization. Health Statistics from the Americas, 1992 edition. Washington, D.C.:PAHO, 1992 (Scientific Publication 542).
(5) Pan American Health Organization. Health Statistics from the Americas, 2003 Edition. (Scientific Publication 591) [Web page]. Available at: http://www.paho.org/english/am/pub/SP_591.htm.

Source: Prepared by Mr. John Silvi from PAHO’s Area of Health Analysis and Information Systems, and presented at the II Meeting of the Regional Advisory Committee on Health Statistics (CRAES) in September 2003.

Return to Index
Epidemiological Bulletin, Vol. 24 No. 4, December 2003