Data issues

• 1. Data sources
• 2. Case definition
• 3. Age adjustment
• 4. Confidence intervals
• 5. Time series
• 6. Cause code aggregations
• 7. Data reliability

1. Data sources

Deaths data are from the Australian Bureau of Statistics (ABS) mortality unit record data collection, 1979-97. Population data are based on ABS estimates of resident population as at 30 June 1997.

2. Case definition

The cause of each death registered in Australia is classified by the ABS according to the International Classification of Diseases (ICD). The 9th revision (ICD9) has been used for death registrations beginning in 1979. All deaths given an ICD9 "External Cause" code by the ABS are included in this Bulletin.
Data are presented according to the year in which deaths were registered. Nine percent of deaths registered in 1997 occurred in an earlier year. A similar proportion of deaths which occurred in 1997 will not have been registered until after 1998. Information on these cases is not yet available. State-specific data are presented on the basis of the state or territory in which death was registered. This is normally the one in which death occurred.

3. Age adjustment

Most all-ages rates have been standardised to overcome the effect of differences in the proportions of people of different ages (and different injury risks) in the populations that are compared. Direct standardisation was employed, taking the Australian population in 1991 as the standard.

4. Confidence intervals

All (or nearly all) deaths are registered, so sampling errors do not apply to these data. However, the time periods used to group the cases (ie. calendar years) are arbitrary. Use of another period (eg. July to June) would result in different rates. Where case numbers are small, the effect of chance variation on rates can be large. Confidence intervals (95%, based on a Poisson assumption about the number of cases in a time period) have been placed around rates as a guide to the size of this variation. Chance variation alone would be expected to lead to a rate outside the interval only once out of 20 occasions. An extreme rate in a single period of enumeration should not be ignored simply because of a wide confidence interval - a time series may show such a rate to be part of a more significant pattern.

5. Time series

Time trends have been presented for the period 1988 to 1997. Australian deaths data have been classified according to the 9th revision of the International Classification of Diseases (ICD9).

6. Cause code aggregations

NISU statistical publications make use of standard aggregations of the ICD9 external cause (E-code) classification. The E-code equivalents of most groups presented in this Bulletin are noted in the text.

7. Data reliability

The chief question concerns the reliability of information about type of injury death. This depends principally on the information available in coroner's records, and on the reliability of the application of ICD9 E-codes, generally based on that information. Little empirical information is available. There is considerable potential for factors to do with information recording or coding to affect data in different ways for different states and territories. Hence, apparent differences between jurisdictions should be interpreted with caution. Beginning with 1993 registrations, coding has been centralised at the Brisbane office of the ABS.