3317.0.55.002 - Information Paper: ABS Causes of Death Statistics: Concepts, Sources and Methods, 2006  
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 14/03/2008  First Issue
   Page tools: Print Print Page Print all pages in this productPrint All

DATA PROCESSING


INTRODUCTION

Given the importance of assuring data quality at every stage of the statistical process (during collection, processing or dissemination), it is important to understand the business process from end to end. The diagram below summarises the statistical process used to produce cause of death statistics.


CAUSES OF DEATH, Deaths records process diagram, Deaths registered in a specific calendar year
Diagram:  Causes of death





DATA COLLECTION

Registrars pass information from death notifications and death certificates to the ABS (even if the registration is not finalised i.e. full cause of death details are not available). Data from death registrations is received monthly by the ABS in a standard electronic format from RBDMs. Processing of this data is carried out by the Health and Vitals Statistics Unit (HVSU), located in the Queensland Office of the ABS.


DATA PROCESSING - DEMOGRAPHIC DATA ITEMS

For death statistics collection data items are derived, coded and edited before they are deemed to be of sufficient quality to release to clients. The ABS Vitals Statistics System records all steps through these processes and provides a sound statistical platform for staff to intervene and resolve data quality errors as well as analyse the overall quality of the resulting dataset or a particular data item.

A generic outline of the process for turning the registration data into death statistics is briefly outlined below.
      1. Record is loaded
      2. Record has derivations applied i.e. some data items are used to “derive” data items that are not directly collected (for example “Date of Birth” and “Date of Death” are used to derive “Age at Death”)
      3. Record passes through the duplicate identification process. If a record is found to be a duplicate, action is taken to determine which record is correct and which record is a duplicate. Once the record has been determined to be correct, it will either continue on through the process it was in prior to the identification i.e. continue through derivation, coders and edits. Records found to be a duplicate are cancelled and play no further part in processing or analysis.
      4. Record passes through a range of coding processes including Birthplace coder, Geographical coder, occupation coder, medical terminology coder etc
      5. Record passes through the edit process
      6. If the record has triggered an edit manual intervention is undertaken to resolve the edit. This may involve contacting the RBDM for the relevant jurisdiction to seek additional information.
      7. For records that did not trigger an edit or for a record with all edits resolved, the records will pass through derivations, coders (and assigned a code if applicable) and edits again. If a record is clean (no edits) then the record has passed through all demographic processes and found to be of sufficient quality.
      8. A range of macro editing processes occurs in order to validate the entire data set

DATA PROCESSING - CAUSES OF DEATH DATA ITEMS (MORTALITY CODING)

A generic outline of the process for turning death data into causes of death statistics is briefly outlined below.
      1. Once a record is deemed to be “clean” from a demographic perspective, the record proceeds through to mortality coding
      2. If the record has been certified by a Doctor, the record will then proceed through to the Automatic Cause of Death coder to assign a ICD-10 code
      3. If the record has been certified by a Coroner, the record undertakes a matching process with data from the NCIS in order to obtain the coronial information required to code the record. Once the record is matched it will then proceed through to the Automatic Cause of Death coder to assign a ICD-10 code
      4. ICD- 10 coded records then pass a range of causes of death specific edits for quality assurance processes. These edits include (but limited to) edits to check for:
          • Gender versus cause restrictions
          • Age versus cause restrictions
          • Specificity of assigned underling cause
          • Trivial and rare causes
          • Improbable causes
      5. Records unable to be assigned a cause of death code automatically, are sent to a “query” stage where additional information is sought from the certifier. On receipt of additional information, records proceed through the Cause of Death coder and edit process again.
      6. If record is clean (no edits) then the record has passed through all causes of death processes and found to be of sufficient quality
      7. A range of macro editing processes occurs in order to validate the entire data set


MORTALITY CODING

Background to Mortality Coding in Australia

Prior to 1997, the ABS manually coded the 'underlying cause of death', assigned according to World Health Organisation (WHO) guidelines. The underlying cause of death is defined as:
        'the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury'. (ABS, 2006)
When more than one condition is entered on the death certificate, the underlying cause is selected using the coding rules of the ICD. Since its adoption in 1948, statistics based on the underlying cause concept have served the purpose of summarising international cause-specific mortality statistics into a single index which has been used to assess trends in causes of death.

Multiple Cause Coding

The leading causes of death have changed over time from infectious and parasitic diseases to chronic and degenerative diseases. As the population ages the focus on chronic diseases and understanding their co-morbidities becomes increasingly important. Selecting only a single underlying cause leads to the loss of other valuable information that is often important in understanding disease processes.

Multiple cause coding is best defined as:
        'the coding of all morbid conditions, diseases and injuries entered on the death certificate, including those involved in the morbid train of events leading to the death which were classified as either the underlying cause, the intermediate cause, or any intervening causes, and those conditions which contributed to death but were not related to the disease or condition causing death.'

Major benefits of multiple cause coding include:
      • an increase in the type and variety of data available for analysis
      • an improved product for matching mortality and morbidity data
      • an improved product for internationally comparable data
      • further details on deaths from external causes (including nature of injuries)
Multiple cause coding allows researchers to focus on other levels of variables. Over four in every five deaths registered in Australia provide more than one condition on the death certificate. Increasing analysis on the contributory causes, as well as traditional analysis of underlying causes, may increase the available knowledge on these conditions and offer alternatives in terms of treatment and/or prevention.

International Classification of Diseases (ICD)

The International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) is used by the ABS to code causes of death. The International Classification of Diseases (ICD), produced by the World Health Organisation (WHO), is the international standard diagnostic classification used to classify diseases and other health problems recorded on many types of health and vital records including death certificates and hospital records (WHO 2007).

In 1994 the Australian Health Ministers Advisory Committee (AHMAC) established a committee to oversee the implementation of the tenth revision of the ICD in Australia. The ABS worked in conjunction with the Australian Institute of Health & Welfare (AIHW) and the National Reference Centre for Classification in Health (now known as National Centre for Classification in Health - NCCH) to ensure standard and appropriate implementation of the classification throughout mortality data. The ABS continues to work closely with NCCH and the Australian Institute of Health and Welfare to ensure correct and consistent interpretation of the classification by the ABS. The use of the ICD as the standard classification contributes greatly to the overall coherence of the ABS causes of death collection.

Automated Coding System

The ABS processes around 140 000 deaths annually, using an automated coding system (ACS). The Mortality Medical Data System (MMDS) allows the classification of multiple causes of death in accordance with the current version of the International Classification of Diseases (ICD). This result in the coding of every condition mentioned on a death certificate as contributing to the death.

The ABS implemented the MMDS system during 1997, after considerable research and testing using ICD-9 (WHO, 1975) and the subsequent introduction from 1999 for ICD-10 (WHO, 1984). The two previous years of data (1997 and 1998) were “back coded” in ICD-10. All deaths occurring within Australia, registered from 1 January 1997, have been multiple causes coded, in ICD-10.

Three main programs make up the MMDS software suite: SuperMICAR, MICAR200, and ACME/TRANSAX.
    • SuperMICAR is a text searching application designed to automatically encode medical cause of death data into numeric entity reference numbers (ERN). The software takes "cause" text and splits it into separate causes ignoring noise words, such as "massive", "terminal", and "life threatening". For example, if "myocardial infarction due to hypertension" is recorded on one line of the certificate, it will be separated into two causes, i.e. the hypertension and the myocardial infarction. Each cause is then analysed, and a unique numeric code (ERN) for each accepted cause of death term is assigned

        SuperMICAR contains a thesaurus, dictionary and word list. The dictionary holds valid descriptions for each ERN, while the thesaurus contains synonyms for words which cannot be matched in the dictionary. For example, the word "narrowing" may be recorded instead of the medical term stricture. Finally, the word list contains words which are to be ignored and words which separate causes. For example, the words "left" and "right" are important in some cases for coding but superfluous in others. Alternatively, in the above case of myocardial infarction due to hypertension, the "due to" are words which are used to separate causes. The ERNs represent a much more detailed classification than the various versions of the International Classification of Diseases. The design of this system enables a smoother transition between versions of the ICD.

    • MICAR200 performs editing and validation functions by accessing mortality coding rules, which are held in look-up files, together with a dictionary. These rules are applied to the ERNs assigned during the SuperMICAR process, to produce ICD multiple cause codes.
    • ACME/TRANSAX applies World Health Organisation (WHO) rules to the ICD codes determined by MICAR200. ACME uses a number of look-up tables to allocate underlying cause codes appropriately. ACME makes implicit linkages between multiple causes in assigning ICD codes for underlying cause. TRANSAX identifies the relationship between conditions mentioned on the death certificate, and then assigns an ICD code for any significant combinations.
In addition to the benefit of allowing for all the conditions on the death certificate to be coded, this system offers several other advantages, including:
      • The potential for substantially improving the cost effectiveness of mortality coding.
      • Removal of the subjectivity inherent in manual systems, through the use of pre-coded logic. (A percentage of coding still requires manual intervention)
      • More internationally comparable mortality statistics.

Coding of external causes of death

Most deaths from natural causes require little intervention in the automated coding process. However, due to the complexity of assigning multiple and underlying causes for deaths due to external causes, the MMDS is unable to automatically code these deaths. For deaths from external causes, coding is still undertaken within MMDS, but substantial manual intervention by ABS coders is required. In total, the ABS manually codes around 25,000 deaths each year, including virtually all deaths from external causes.

Following ICD-10 rules, deaths from external causes are classified according to the intent of death (e.g. accidental, suicide, assault, unspecified intent etc) and also according to the mechanism of death (e.g. suffocation, drowning, transport etc) using codes from Chapter XX 'External causes of morbidity and mortality' . The terms included in this chapter are not medical conditions but descriptions of the circumstances in which the violence occurred (e.g. fire, explosion, fall, assault, collision, and submersion). In addition, codes from Chapter XIX 'Injury, poisoning and certain other consequences of external causes' are used to classify the nature of injury or injuries involved (e.g. asphyxiation, injury to head, multiple injuries etc).

The primary axis of Chapter XX 'External causes of morbidity and mortality' is based on intent of death, represented by the following blocks of codes:
      • accidental (V01-X59)
      • intentional self harm (suicide) (X60-84)
      • assault (X85-Y09)
      • event of undetermined intent (Y10-Y34)
      • legal intervention and operation of war (Y35-Y36)
      • complications of medical and surgical care (Y40-Y84)
      • sequelae of external causes of morbidity and mortality (Y85-Y89).

Query process

Death records which are certified by a coroner have little or no information in the fields relating to causes of death when supplied to the ABS by Registrars of Births, Deaths and Marriages, because at the time the death is required to be registered the information necessary to certify the causes of death is not available. Causes of death information become available over time as coronial cases are closed and the information is posted on the NCIS.

During processing at the ABS, any records which contain insufficient information for coding are flagged for further investigation. A large proportion of these records are 'reportable' deaths. Records which are regarded as having sufficient information continue through the processing system.

Some examples of entries on death records in the fields relating to causes of death which contain insufficient information to accurately code the causes of death are as follows:
      • 'multiple injuries' (with no further information)
      • 'hanging' (with no further information)
      • 'head injury' (with no further information)
      • blank entry.
In cases such as these, and in many similar instances involving both external and natural causes of death, the ABS must seek further information to accurately code the causes of death, as well as other information on the death record. This leads to an extensive query process involving liaison with Registrars' offices and Cancer Registries, query letters to doctors, further checking of the NCIS, as well as access to paper files in individual coroner's offices. The ABS undertakes a query process for all records requiring resolution, particularly for 'reportable' deaths. The query process continues until the ABS finishes compiling the annual causes of death statistics, after which time information subsequently becoming available is not coded and no further changes are made to the data file.

In total, ABS query action is performed on around 25,000 death records annually. The effectiveness of the ABS query process depends on the goodwill and resources available in other organisations. The ABS receives substantial assistance in this regard from the NCIS and coroners' offices, and the ABS continues to work with these organisations to improve relevant processes.


VALIDATION

Causes of death data are subject to a large number and range of validation checks, including checks on specificity of coding, coding logic, and age/gender consistency. The data are also compared with previous years to ensure consistency and to aid in the identification of trends.