Reliability of ICD-10 external cause of death codes in the National Coroners Information System

Transcription

1 Reliability of ICD-10 external cause of death codes in the National Coroners Information System Lyndal Bugeja, Angela J Clapperton, Jessica J Killian, Karen L Stephan & Joan Ozanne-Smith Abstract Availability of ICD-10 cause of death codes in the National Coroners Information System (NCIS) strengthens its value as a public health surveillance tool. This study quantifi ed the completeness of external cause ICD-10 codes in the NCIS for Victorian deaths (as assigned by the Australian Bureau of Statistics (ABS) in the yearly Cause of Death data). It also examined the concordance between external cause ICD-10 codes contained in the NCIS and a re-code of the same deaths conducted by an independent coder. Of 7,400 NCIS external cause deaths included in this study, 961 (13.0%) did not contain an ABS assigned ICD-10 code and 225 (3.0%) contained only a natural cause code. Where an ABS assigned external cause ICD-10 code was present (n=6,214), 4,397 (70.8%) matched exactly with the independently assigned ICD-10 code. Coding disparity primarily related to differences in assignment of intent and specifi city. However, in a small number of deaths (n=49, 0.8%) there was coding disparity for both intent and external cause category. NCIS users should be aware of the limitations of relying only on ICD-10 codes contained within the NCIS for deaths prior to 2007 and consider using these in combination with the other NCIS data fi elds and code sets to ensure optimum case identifi cation. Keywords (MeSH): Coroners and Medical Examiners; ICD-10; Mortality; Cause of Death Introduction Injury remains the leading cause of death of Australians aged 1 to 44 years of age (Kreisfeld, Newson & Harrison 2004). Data on the nature, extent and determinants of fatal injury can inform injury prevention policy, priority setting and the allocation of research and development funds (Mathers et al. 2005). All unexpected deaths, including those resulting from injury, are legally required to be reported to the coroner for investigation (Freckelton & Ranson 2006). Since July 2000 (January 2001 in Queensland), all deaths reported to and investigated by coronial offices in Australia have been recorded in the National Coroners Information System (NCIS) a world first, Internet-based data storage and retrieval system. The NCIS contains coded and free text fields, the police narrative from the report of death to the coroner, the autopsy report, toxicology report and coroner s finding (NCIS 2005). The availability of this information, particularly full text reports, varies between jurisdictions. For example, according to NCIS Operational Statistics, South Australia is yet to upload autopsy reports and Queensland is yet to upload both autopsy and toxicology reports. These reports are due to be uploaded into the NCIS in late Due to the unavailability of reports, cause of death information is limited to the statement of the medical cause of death and this is often insufficient to account for the presence of injury, disease and alcohol and drugs in the death. Many of the NCIS data items are based on the definitions outlined in the National Health Data Dictionary (NHDD) including: date of birth, sex and marital status ensuring consistency with other Australian data collections. The International Classification of External Causes of Injury (ICECI) is used for the coding of: intent; activity when injured (including sport and recreation); location; mechanism of injury; objects or substances producing injury; and transport vehicle details (mode, user, context and counterpart). Additional fields from the Australian Bureau of Statistics (ABS), and provided in the NCIS, include cause of death codes classified to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10). From an epidemiological perspective, ICD-10 cause of death codes are a key variable for the identification, extraction and analysis of injury death data. Chapter XX (External Causes of Morbidity and Mortality) of ICD-10 contains the group of alphanumeric codes 16 HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)

2 which can be applied to classify deaths resulting from injury according to many factors including the intent of the deceased person (unintentional, intentional selfharm, assault, undetermined intent), causal mechanism (such as transport, falls, drowning and submersion, etc.) and object involved in the injury event (such as fall involving playground equipment, drowning and submersion while in a swimming pool, etc.). The process for coding death certificates in Australia has been well documented by the ABS and by researchers using these data (ABS 2008b; Walker, Chen & Madden 2008; McKenzie, Chen & Walker 2009). For all coronial deaths, the ABS assigns ICD-10 cause of death codes. Until 2006, coding decisions were based on information from a number of sources including the medical cause of death determined by the forensic pathologist (subsequently registered by the Registry of Births, Deaths and Marriages (BDM)), information generated for the coronial investigation from coroners offices and records in the NCIS. The Mortality Medical Data System (MMDS) is used by the ABS to attempt to code all deaths to ICD-10 for the year. Some external cause deaths cannot be processed because the software dictionary does not recognise the terms used on the certificate. These records are rejected and are instead manually coded by qualified ABS coders (National Centre for Classification in Health n.d.). Where the death remains under the coroner s investigation at the time of ABS final data processing, coding is based on the available medical and legal information, which may be limited (Walker, Chen & Madden 2008). Prior to 2007, the ABS did not update coding, even where additional information became available from the coroner. This may have resulted in the ABS applying default or non-specific codes to deaths, for example coding of a possible intentional self-harm death as unintentional, other or unspecified intent. This is in accordance with the ICD-10 mortality coding rules and techniques developed by the World Health Organization (WHO) and applied by the ABS for assigning external cause codes. These rules are described in further detail elsewhere (ABS 2007). The NCIS entered into a reciprocal data exchange agreement with the ABS to provide ICD-10 Underlying Cause of Death (UCoD) and Multiple Cause of Death (MCoD) codes for all deaths (external and natural cause) in the NCIS (Daking & Dodds 2007). NCIS inclusion of ICD-10 codes for coronial deaths from was performed retrospectively in 2004, and from 2005 upon completion of the ABS Annual Cause of Death Coding. ABS assignment of ICD-10 codes for deaths until the completion of the 2001 file were coded using data accessed directly from coroner s offices. Between 2002 and 2004, the ABS accessed both the coronial records and the NCIS (ABS 2008a). Since the commencement of the 2006 Cause of Death coding, the NCIS has been the main source of data utilised by the ABS for coronial deaths (ABS 2008a). To facilitate data exchange with the NCIS, the ABS developed a data matching process. The NCIS provided a data extract of variables to the ABS; these were then matched against the deceased s given name, surname, date of birth, date of death, age and sex (M. Wood (ABS) 2008, pers. comm. 5 Feb.). ICD-10 codes for matched deaths were provided back to the NCIS. Where a match between NCIS data and the ABS record was not found, no ICD-10 code was provided to the NCIS by the ABS. This method was prone to error and a more comprehensive process of matching was introduced for 2006 data onward. From 2006, deaths were matched probabilistically by the ABS using NCIS data provided to the ABS (demographic information, cause of death details, mechanism of injury and intent) and the ABS mortality records. This process involves an automated computerised algorithm followed by manual matching for deaths where some details match but not with sufficient confidence for the automated match. The NCIS is increasingly relied upon for identifying cases for death investigation, research and to monitor external cause deaths in Australia. The addition of ICD-10 cause of death codes strengthens the value of the NCIS for injury surveillance. These codes can be used as the primary variable for case identification and/or to validate other case identification methods. Previous studies have shown discrepancies between the ICD-10 codes assigned by the ABS in the NCIS and an independent recode (Daking & Dodds 2007) and between the classification of the NCIS variable Intent completion (Elnour & Harrison 2009). The reliability of external cause ICD-10 coding for deaths in the NCIS has not been studied for a total population case series. This study is limited to the Victorian population and aims to quantify: (a) the completeness of ABS assigned external cause ICD-10 codes in the NCIS; and (b) concordance between ABS assigned ICD-10 codes and independently re-coded data. It also highlights potential enhancements to the ICD-10 matching and coding process for coronial deaths. Method Research design The research design comprised a retrospective population case series comparison of external cause deaths reported to the Coroner in Victoria, coded by two HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE) 17

3 discrete processes. The study arose from a larger Victorian study examining coronial data for which ICD-10 cause of death codes was a key variable. It was beyond the scope of the study to examine a national case series. Case identification Coronial external cause deaths for the five years from 1 July 2000 to 30 June 2005, where the investigation had been completed by 31 December 2006, were identified and extracted from the NCIS. Deaths were included where the NCIS variable Case type completion (that is whether the death was determined by the Coroner as resulting from natural, external or unknown cause(s) on completion of their investigation), was classified as Death due to External Cause(s). Data collection A de-identified unit record data extract comprising the variables: age at death; intent completion; medical cause of death; activity; location; and ICD-10 UCoD and MCoD codes were provided by the NCIS. The Coroner s finding was reviewed via the NCIS for each death and a short structured narrative was generated comprising: intent; mechanism; detailed circumstances; and object/substance (for example unintentional motor vehicle accident car (driver) collision with tree). The structured narrative was prepared to de-identify and summarise the Coroner s finding. Ethical limitations prevented the provision of the identifying Coroners findings to the independent coder. The NCIS variables and structured narratives were provided to an independent Health Information Officer with qualifications and experience in ICD-10 mortality coding. This coder was instructed to assign a three digit ICD-10 UCoD code in accordance with the WHO coding rules. Data analysis For deaths where ICD-10 codes were absent in the NCIS, the frequency and proportion of independently assigned external cause ICD-10 codes was calculated for each external cause category (that is, transport, falls etc.). These were compared to the frequency and proportion of independently assigned external cause ICD-10 codes where ABS-assigned ICD-10 external cause codes were present in the NCIS. This was a mechanism to determine how deaths missing an ICD- 10 code in the NCIS were distributed across external cause categories to check and demonstrate any systematic bias. Chi square and Fisher s exact tests were used to examine if differences were statistically significant. Where ICD-10 codes were present in the NCIS, the first occurring external cause ICD-10 code (UCoD or MCoD) was compared to the independently assigned ICD-10 UCoD external cause code to determine levels of concordance. The method of analysis was adopted from a similar study examining ICD-10- AM concordance for morbidity data (McKenzie & McClure 2010). Agreement was assessed across three components of the code: 1) intent; 2) external cause category; and 3) specificity. These components were defined as follows: 1. Intent the role of the deceased in the death: unintentional; intentional self-harm; assault; undetermined; legal intervention; complications of medical or surgical care; or sequelae of external causes. 2. External cause category the injury leading to death was caused by: transport accidents; falls; inanimate mechanical forces; animate mechanical forces; drowning and submersion; other threats to breathing; electric current, radiation and extreme ambient air temperature and pressure; smoke/ fire/flames; heat and hot substances; venomous animals and plants; forces of nature; poisoning by and exposure to noxious substances; or other and unspecified factors. 3. Specificity injury leading to death was caused by: a specified mechanism; other specified mechanism; or unspecified mechanism (McKenzie & McClure 2010). The following categories were developed to quantify levels of agreement, to examine the nature of and possible explanation for disagreement and to assess the impact on case identification in the NCIS: 1. Different intent, different category. 2. Same intent, different category. 3. Different intent, same category, same specificity. 4. Same intent, same category, different specificity. 5. Same code (McKenzie & McClure 2010). Results The NCIS data extract contained 7,400 deaths, closed and classified as resulting from external causes reported to Victorian coroners during the five years from 1 July 2000 to 30 June For 961 (13.0%) deaths there was no ICD-10 code available in the NCIS. For 225 (3.0%) deaths, a natural cause of death ICD-10 code was present in the NCIS. With the removal of these deaths from the study, the final cohort included 6,214 deaths or 84.0% of the original extract. Before a comparison was undertaken between the external cause ICD-10 codes in the NCIS and those assigned by the independent coder, further analysis was conducted on deaths missing or assigned a natural cause ICD-10 in the NCIS. 18 HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)

4 Table 1: Independently assigned external cause ICD-10 UCoD category where an ABS assigned ICD-10 code was absent compared to present by ICD-10 external cause categories, Victoria July June 2005 INDEPENDENTLY ASSIGNED INDEPENDENTLY ASSIGNED EXTERNAL CAUSE ICD-10 EXTERNAL CAUSE ICD-10 UCoD CODE WHERE ABS UCoD CODE WHERE ABS ICD-10 EXTERNAL CAUSE CATEGORIES ASSIGNED CODE ABSENT ASSIGNED CODE PRESENT p n % n % Transport accidents Falls < 0.001* Inanimate mechanical forces Animate mechanical forces Drowning and submersion Other threats to breathing Electric current, radiation and extreme ambient air temperature and pressure Smoke / fi re / fl ames Heat and hot substances Venomous animals and plants Forces of nature Poisoning by and exposure to noxious substances Overexertion, travel and privation Exposure to other and unspecifi ed factors Intentional self-harm * Assault < 0.001* Undetermined Intent Legal intervention and operations of war Complications of medical and surgical care Sequelae of external causes of morbidity and mortality Total , Calculated using Fisher s exact test with Bonferroni adjustment for multiple comparisons: Bonferonni adjusted p-value for significance was.0025 * Statistically signifi cant p-values Missing ICD-10 Codes Table 1 shows the comparison between the independently assigned external cause ICD-10 codes for the 961 deaths where there was no ICD-10 code available in the NCIS and the independently assigned external cause ICD-10 codes for the 6,214 deaths where there was an ICD-10 code available in the NCIS. The overall chi square test showed there was a statistically significant difference (χ 2 (19) = 84.1, p < 0.001) between the distribution of the independently assigned external cause ICD-10 codes where an ICD-10 code was missing in the NCIS compared to where an ICD-10 code was present in the NCIS. The Fisher s exact test, with Bonferroni adjustment for multiple comparisons, was used to determine for which external cause categories the differences were significant. There was a statistically significant difference for the following external cause categories: falls; intentional self-harm; and assault. The categories falls and intentional self-harm were significantly less likely to be missing an ICD-10 code while the category assault was significantly more likely to be missing an ICD-10 code in the NCIS. Deaths missing ICD-10 codes in the NCIS were likely due to an inability to match an ABS record to an NCIS case with a significant enough degree of confidence. As stated above, information to assist ICD-10 coding for deaths up to and including the 2005 file were accessed by the ABS both directly from the coroner s offices and using the NCIS. This warrants further examination. Natural Cause of Death Code/s The 225 deaths with natural cause ICD-10 codes in the NCIS were examined by ICD-10 cause of death category. Table 2 shows that almost 40% (n=88) were classified as resulting from diseases of the circulatory system. A comparison with the independently assigned external cause ICD-10 UCoD code was also conducted to determine possible reasons for the discrepancy. In 108 deaths (48.0%), natural cause ICD-10 code/s may have been assigned because the medical cause of death determined by the forensic pathologist did not include the presence of injury required to prompt external cause code assignment. It is possible that the medical HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE) 19

5 Table 2: Frequency and proportion of ABS assigned natural ICD-10 UCoD codes, Victoria July June 2005 ICD-10 CAUSE CATEGORIES n % Certain infectious and parasitic diseases Neoplasms Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism Endocrine, nutritional and metabolic diseases Mental and behavioural disorders Diseases of the nervous system Diseases of the circulatory system Diseases of the respiratory system Diseases of the digestive system Diseases of the musculoskeletal system and connective tissue Diseases of the genitourinary system Pregnancy, childbirth and the puerperium Certain conditions originating in the perinatal period Congenital malformations, deformations and chromosomal abnormalities Symptoms, signs and abnormal clinical and laboratory fi ndings NEC Total cause of death was the only information available for coding because either the coronial investigation was not completed or the final cause of death had not been entered into the NCIS. Common themes explaining these disparities include the medical cause of death stated as unascertained or the death occurred following mismanaged or mistreated disease or illness. In these deaths natural cause ICD-10 code/s may have been assigned because only the medical cause of death was relied on for classification or because an operation was reported without details of the condition for which it was performed or any mention of the complication. In these instances the WHO mortality rules indicate that a disease be recorded as the UCoD (World Health Organization 2005). In 78 deaths (34.7%), natural cause ICD-10 code/s were assigned despite external cause being stated in the cause of death. Examples are shown in Table 3. In the remaining 39 deaths (17.3%), natural cause ICD-10 code/s were assigned when the cause of death was attributed only to external causes, most likely a result of matching error. Examples are shown in Table 4. Table 3: Case examples where ABS assigned natural cause code(s) despite external cause present in the medical cause of death NCIS ICD-10 UCoD INDEPENDENT MEDICAL CAUSE OF DEATH AND MCoD CODES ICD-10 UCoD CODE 1A. Cardiomegaly in a man with evidence of amphetamine abuse UCoD = I517 UCoD = X41 ICD-10 Level 1 = I517 ICD-10 Level 2 = F151 1A. Cardiac arrest secondary to respiratory failure resulting from UCoD = J80 UCoD = V44 adult respiratory distress syndrome in a female involved ICD-10 Level 1 = I469 in a motor vehicle accident ICD-10 Level 2 = J969 ICD-10 Level 3 = J80 Table 4: Case examples where ABS assigned a natural cause code(s) despite only external cause stated in the medical cause of death NCIS ICD-10 UCoD INDEPENDENT MEDICAL CAUSE OF DEATH AND MCoD CODES ICD-10 UCoD CODE 1A. Stab injury to neck UCoD = A419 UCoD = X99 ICD-10 Level 1 = I469 ICD-10 Level 2 = E889 ICD-10 Level 3 = A419 ICD-10 Level 4 = L930 ICD-10 Level 5 = N180 1A. Stab wounds to chest and gunshot wounds to head UCoD = I608 UCoD = X95 ICD-10 Level 1 = G935 ICD-10 Level 2 = I614 ICD-10 Level 3 = I608 1A. Consistent with disseminated intravascular UCoD = E888 UCoD = X20 secondary to a brown snake bite ICD-10 Level 1 = J969 ICD-10 Level 2 = J189 ICD-10 Level 3 = G729 ICD-10 Level 4 = R629 ICD-10 Level 5 = E HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)

6 Table 5: Intent and external cause category matches between ABS external cause ICD-10 (UCoD or MCoD) codes and independently assigned external cause ICD-10 UCoD codes, Victoria July June 2005 INDEPENDENTLY ASSIGNED EXTERNAL CAUSE ICD-10 UCoD CODES TOTAL TRANSPORT ACCIDENTS FALLS INANIMATE MECHANICAL FORCES ANIMATE MECHANICAL FORCES DROWNING OTHER THREATS TO BREATHING ELECTRICITY SMOKE / FIRE / FLAMES HEAT AND HOT SUBSTANCES VENOMOUS ANIMALS & PLANTS FORCES OF NATURE POISONING OTHER & UNSPECIFIED INTENTIONAL SELF-HARM ASSAULT UNDETERMINED INTENT LEGAL INTERVENTION COMPLICATIONS OF MEDICAL OR SURGICAL CARE SEQUELAE OF EXTERNAL CAUSES INDEPENDENT % MATCHED ABS ASSIGNED EXTERNAL CAUSE ICD-10 UCoD OR MCoD CODES # Transport accidents 1, , Falls Inanimate mechanical forces Animate mechanical forces Drowning Other threats to breathing Electricity Smoke / fire / flames Heat and hot substances Venomous animals & plants Forces of nature Poisoning Other & unspecified Intentional self-harm , , Assault Undetermined intent Legal intervention Complications of medical or surgical care Sequelae of external causes Total 1, , ,214 ABS % Matched HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE) 21

7 Concordance of ICD-10 External Cause of Death Codes After excluding the deaths with missing and natural cause of death codes, there were 6,214 deaths remaining in the study extract with an ICD-10 external cause code, either as the UCoD or MCoD. Table 5 shows the concordance with the independently assigned ICD-10 codes for intent and external cause categories. Although low in frequency, 100% concordance was found for electricity (n=7), animate mechanical forces (n=5), venomous animals and plants (n=2) and legal intervention (n=3). Less than 80% concordance was found for: inanimate mechanical forces; other threats to breathing; heat and hot substances; forces of nature; exposure to other and unspecified factors; and sequelae. Particularly low levels of concordance were found for ABS assigned exposure to other and unspecified factors (3.9% concordance) and independently assigned undetermined intent (40.5% concordance). The majority of ABS assigned deaths due to exposure to other and unspecified factors were classified as falls (n=32) and transport (n=22) by the independent coder. A large number of independently assigned undetermined intent deaths (n=69) were classified by the ABS as unintentional poisoning. Examination of levels of agreement by intent and external cause categories revealed an exact match in 4,397 (70.8%) deaths with UCoD (n=4,242) or MCoD (n=155) ICD-10 codes in the NCIS. Causes of deaths with a frequency of over 100 and at least 75% exact matches were: intentional self-harm (n=1,921, 84.5%); transport (n=1,367, 79.5%) and assault (n=114, 75.0%) (Table 6). The most common reason for coding disparity was in the assignment of specificity (n=1,293, 20.8%). Categories with a high frequency and/or proportion of specificity disparity included: falls (n=185, 43.1%); unintentional poisoning (n=347, 39.0%); complications of medical and surgical care (n=39, 37.1%); drowning and submersion (n=29, 25.0%); transport (n=303, 17.6%); and intentional self-harm (n=306, 13.5%). Table 6: Levels of coding agreement by ABS assigned intent and external cause category, Victoria July June 2005 ABS INTENT AND EXTERNAL CAUSE CATEGORY 1. DIFFERENT INTENT, DIFFERENT CATEGORY 2. SAME INTENT, DIFFERENT CATEGORY 3. DIFFERENT INTENT, SAME CATEGORY, SAME SPECIFICITY 4. SAME INTENT, SAME CATEGORY, DIFFERENT SPECIFICITY 5. SAME CODE TOTAL n % n % n % n % n % n Transport accidents , ,719 Falls Inanimate mechanical forces Animate mechanical forces Drowning and submersion Other threats to breathing Electric current, radiation and extreme ambient air temperature and pressure Smoke / fire / flames Heat and hot substances Venomous animals and plants Forces of nature Poisoning by and exposure to noxious substances Exposure to other and unspecified factors Intentional self-harm , ,273 Assault Undetermined intent Legal intervention and operations of war Complications of medical or surgical care Sequelae of external causes of morbidity and mortality Total , , , HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)

8 Further examination of the nature of specificity disparity showed that a large proportion of discrepancy for unintentional poisoning and drowning and submersion, resulted from the coding of deaths to other or unspecified poisoning/drowning (n=317, 91.4% and n=15, 51.4% respectively), by ABS, whilst the independently assigned code provided specific detail about the poisoning/drowning. Likewise in regard to deaths involving falls, 60% of the specificity disparity involved the assignment of codes other or unspecified (n=60, 32.4% and n=51, 27.6% respectively). In addition a further 18.9% of the disparity in relation to falls related to the height of the fall. For intentional self-harm, specificity disparity primarily related to the means, particularly between types of intentional self-poisoning (n=156, 51.0%) and firearms (n=142, 46.4%). Deaths related to transport injuries demonstrated specificity disparity in areas such as the counterpart involved (n=134, 43.9%) or different vehicle user or mode of transport (n=54, 17.7%), as well as the assignment to other or unspecified (n=76, 24.9%). Approximately 50% of specificity disparity for deaths involving complications of medical and surgical care were between ABS assignment as abnormal reaction and independent assignment as misadventure during surgical medical care. The second most common reason for coding disparity was in the assignment of intent (n=285, 4.6%). Two categories with a high frequency and/or proportion of intent disagreement were: other threats to breathing (n=37, 23.3%); and poisoning (n=131, 14.7%). Further examination of the nature of intent discrepancies showed that for the other threats to breathing deaths, the ABS had assigned an intent of unintentional while the independent coder had assigned an intent of intentional self-harm. For poisoning, the ABS had again assigned an intent of unintentional whereas the independent coder assigned intents of undetermined intent (n=69) or intentional self-harm (n=60). A third reason for coding disparity was differences in the external cause category assigned, which occurred for 190 (3.1%) deaths. The category with a high frequency and/or proportion of external cause category disagreement was exposure to other and unspecified factors (n=58, 76.3%). Whilst ABS has been unable to further define or clarify these deaths, the independent coder was able to assign the majority of these deaths as either falls (n=32) or transport (n=22). The remaining four deaths were coded to complications of medical or surgical care (n=2), assault (n=1) and inanimate mechanical forces (n=1). In the remaining 49 (0.8%) deaths, coding disparity related to both intent and external cause category. The category with a high frequency and/or proportion of intent and external cause category disparity was exposure to other and unspecified factors (n=15, 19.7%). Further examination showed that these deaths were classified by the independent coder as intentional self-harm (n=5), undetermined intent (n=5), assault (n=4), and complications of medical and surgical care (n=1). Discussion This study examined the concordance between the external cause ICD-10 codes available in the NCIS, as assigned by the ABS, and the same deaths recoded by an independent ICD-10 trained coder. In summary, an extract of 7,400 deaths were utilised for this study, although not all deaths were useable; 961 (13.0%) deaths did not actually contain any ICD-10 data from ABS and 225 (3.0%) deaths contained only natural cause ICD-10 code/s. This resulted in the comparable subset of 6,214 deaths. Analysis illustrated that independently assigned external cause ICD-10 UCoD codes matched exactly with the ABS assigned codes in 4,397 (70.8%) deaths. The proportion of exact matches for external cause deaths for this study was higher than reported in a previous study (Daking & Dodds 2007), which compared a smaller subset of national data to the 4 th character level. A further 1,293 deaths (20.8%) matched on intent and external cause category but not specificity, 285 (4.6%) on external cause category and specificity but not intent and 190 (3.1%) matched on intent but not external cause category. The remaining 49 deaths (0.8%) did not match on intent or external cause category. Areas of coding disparity related to the assignment of specificity by ABS which included the use of other and unspecified and the coding of intentional self-harm deaths to an unintentional code, in particular to other threats to breathing and unintentional poisoning. These coding disparities likely reflect the timing of the coding activity being completed whilst the death is still under investigation by the coroner. Prior to completion of the coroners investigation, information about the circumstances of death, including the external cause of death may be limited. The time taken to complete a coronial investigation often exceeds the deadline requirements for ABS to deliver the Cause of Death data file. As a result, in accordance with WHO coding rules, coders are instructed to default code to an unintentional cause. This in turn may result in miscoding or non-specific coding by the ABS (Walker et al. 2008; ABS 2007). It is likely that a significant number of these deaths would have matched if this HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE) 23

9 issue were rectified. Changes to coding procedures implemented by the ABS for deaths occurring from 2007 will address this issue, with updates to the UCoD and MCoD being made as coronial investigations are finalized and changes to the decision making process for the coding of intentional self-harm. This is discussed further below. A further limitation relates specifically to the coding of intent. Intent is formally determined by the Coroner, based on the medical and legal evidence gathered during the death investigation. However, coroners are not legally required to make a determination of intent in their finding. For suspected intentional self-harm deaths, a high standard of legal evidence is required for the coroner to make a determination of suicide (Freckelton & Ranson 2006). This may result in coroners varying their approach to the determination of intentional self-harm both within and between jurisdictions. This impacts on coding decisions (by data entry staff in the Coroners Court or by the ABS coder) as information on intent is not always explicit. For the coding of deaths prior to the 2007 file the WHO ICD-10 coding rules stated that coders were to assign an unintentional code where intent had not been determined (ABS 2007). As a consequence of this, intentional self-harm deaths may have been under-reported by the ABS. As stated above, initiatives implemented for the coding of the 2007 data may resolve some of this issue. Starting with the 2007 mortality data file the ABS will commence revising open coronial deaths upon completion (Wood 2008). As further information becomes available to the ABS about the cause of coronial deaths, it will publish revised data on the basis of year of registration and year of occurrence (ABS 2009). An additional explanation for mismatching data may involve limitations in the process of matching ABS assigned ICD-10 codes to coroner reported external cause deaths. Two unique identification numbers are created during the coroner certification process: the local case number assigned by the Coroners Court of Victoria and the death registration number assigned by the Registry of Births, Deaths and Marriages (BDM). There is no reciprocal recording of both of these unique death identification numbers by the Coroners Court of Victoria, BDM, the NCIS and the ABS. Following death registration with BDM, the Coroners Court of Victoria does not record the Death Registration Number. The Registry and the ABS do not record the local case number allocated by the Coroners Court of Victoria, though it is recorded in the NCIS. As a result, an onerous process of unit record data matching is required between the NCIS and the ABS to assign ICD-10 cause of death codes. A final limitation is the difference in the definition of external cause between the coroners system, which is subsequently coded in the NCIS, and the ICD-10 classification system. An example specific to Victoria is mesothelioma, developed as a result of exposure to asbestos. From 2003, asbestos-related deaths were considered reportable to the Coroner and coded as resulting from external causes in the NCIS. In ICD-10, mesothelioma is classified in Chapter II Malignant neoplasms. For the purposes of consistency, it would be helpful for the NCIS classification of these deaths to be consistent with ICD-10. A limitation of the current study was with the method of case selection. Deaths were included in the study where the NCIS Case type completion variable was classified as Death due to external cause(s). That is, it is possible that deaths may have been misclassified as natural causes instead of external causes, which would lead to under-reporting of external cause deaths in the current study. Particular deaths where this may have occurred include complications of medical or surgical care coded as natural cause deaths. These deaths are particularly difficult for coroners data entry clerks as often the medical cause of death is natural, however the Coroner indicates in the finding that it involved complications of medical or surgical care. It is a potentially grey area that may benefit from the development of detailed coding instructions. The independent professional re-coding of external causes may also have included some level of error. Firstly, neither the ABS nor the independently assigned ICD-10 coding represents a gold standard. ICD-10 coding is dependent on the level of detail available and may be subject to interpretation. The reliability of the independently assigned ICD-10 codes could have been improved by re-coding a random sample of the deaths and calculating the level of agreement. Due to resource constraints, this was not possible. Conclusion The current study showed that just over 15% of deaths classified as resulting from external causes in the five-year study extract were either missing an ICD-10 code or were assigned natural cause code/s. Where an external cause ICD-10 code was assigned by the ABS, the level of full agreement with independently re-coded data was 70.8%. To identify all possible deaths of interest, ICD-10 cause of death codes should not be relied on exclusively for the period under study. The recent quality initiatives introduced by the ABS will go some way to improving consistency and ensuring appropriately coded data is available in the NCIS. UCoD and MCoD ICD-10 codes should be searched, and NCIS users should employ additional 24 HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)

10 search methods using other variables such as intent, mechanism of injury and object. The accuracy and specificity of ICD-10 coding of deaths (both for external and natural causes) reported to the Coroner has implications for the more precise determination of the nature, extent and determinants of injury mortality. This is particularly important for distinguishing between unintentional and intentional deaths and for deaths resulting from poisoning and other threats to breathing. The study results showed that these areas are particularly difficult to code when the presence of the injury or poisoning is not specified and a standard determination of intent is not made by the Coroner. Another area of potential improvement would be to record the presence of injury or poisoning in the medical cause of death to ensure that its contribution is accounted for when assigning ICD-10 codes. Both the injury and the circumstances surrounding the injury should be recorded in the correct sequence (Wood 2008). To improve matching of deaths, the Coroners Court of Victoria, as well as other coronial offices, should consider recording the death registration number assigned by BDM on their electronic case management system (which is linked to the NCIS). This information could be uploaded to the existing field in the NCIS and provided to the ABS as a variable in their data file for regular ICD-10 coding. In addition, BDM could record the coroner s local case number and provide this information to the ABS. This would ensure crosschecking and may eliminate the need for probabilistic data matching using names, dates of birth and death and other possibly non-unique variables. The Coroners Court of Victoria should consider the addition of a section to their finding to reflect case type, intent and / or ICD-10 cause of death. This approach may assist with the interpretation and classification of the Coroner s finding by the Coroners Court, which will flow through to the ABS coding of ICD-10. Finally, more timely completion of Coroners findings would reduce the need for the ABS to code deaths still under investigation by the coroner. Acknowledgements An Australian Postgraduate Award supported the first and third authors for their PhD candidature. The Monash University Accident Research Centre Doctoral Student Fund supported funding for independent coding of ICD-10 cause of death. Many thanks to Jessica Pearse (Manager, NCIS) and Leanne Daking (Quality Manager, NCIS) for their expert advice and review and comments on the manuscript. Thanks also to Dr Kirsten McKenzie (Deputy Director and Senior Research Fellow, National Centre for Health Information Research and Training Queensland University of Technology) for providing information on the ABS ICD-10 coding processes and valuable comment on the manuscript. References Australian Bureau of Statistics (ABS) (2007). External Causes of Death data quality, ABS Cat. No Canberra, Australian Bureau of Statistics. Australian Bureau of Statistics (ABS) (2008a). Causes of Death: Australia ABS Cat. No Canberra, Australian Bureau of Statistics. Australian Bureau of Statistics (ABS) (2008b). Information paper: Cause of Death certification Australia ABS Cat. No Canberra, Australian Bureau of Statistics. Australian Bureau of Statistics (ABS) (2009). Causes of Death: Australia ABS Cat. No Canberra, Australian Bureau of Statistics. Daking, L. and Dodds, L. (2007). ICD-10 mortality coding and the NCIS: a comparative study. Health Information Management Journal 26: Elnour, A. A. and Harrison, J. (2009). Suicide decline in Australia: where did the cases go? Australian and New Zealand Journal of Public Health 33: Freckelton, I. and Ranson, D. (2006). Death investigation and the coroner s inquest, Melbourne, Oxford University Press. Kreisfeld, R., Newson, R. and Harrison, J. (2004). Injury deaths, Australia Injury Research and Statistics Series. Adelaide, AIHW. Mathers, C. D., Ma Fat, D., Inoue, M., Rao, C. and Lopez, A. D. (2005). Counting the dead and what they died from: an assessment of the global status of cause of death data. Bulletin of the World Health Organization 83: McKenzie, K., Chen, L. and Walker, S. (2009). Correlates of undefined cause of injury mortality data in Australia. Health Information Management Journal 38(1): McKenzie, K. and McClure, R. J. (2010). Sources of coding discrepancies in injury morbidity data: implications for injury surveillance. International Journal of Injury Control and Safety Promotion. 17(1): National Centre for Classification in Health (n.d.). ICD-10 Mortality Coding. Available at: new/3.3.aspx (accessed 30 September 2009). National Coroners Information System (NCIS) (2005). NCIS coding manual and user guide, Version 2.5. Melbourne, Victorian Institute of Forensic Medicine. Walker, S., Chen, L. and Madden, R. (2008). Deaths due to suicide: the effects of certification and coding practices in Australia. Australian and New Zealand Journal of Public Health, 32: World Health Organization (2005). ICD-10 Second Edition. Rules and guidelines for mortality and morbidity coding, Operations. Geneva, World Health Organization. HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE) 25

11 Corresponding author: Lyndal Bugeja BA(Hons) Adjunct Research Fellow Accident Research Centre, Monash University Building 70 Monash University Clayton VIC 3800 AUSTRALIA Tel: lyndal.bugeja@monash.edu Angela J Clapperton BSc(Behav), GradDipEdPsych, MCounsel Research Fellow Accident Research Centre, Monash University Building 70 Monash University Clayton VIC 3800 AUSTRALIA Tel: angela.clapperton@monash.edu Jessica J Killian BSc, GradDipRep Sci, MRepSci PhD Candidate Accident Research Centre, Monash University Building 70 Monash University Clayton VIC 3800 AUSTRALIA Tel: jessica.killian@monash.edu Karen L Stephan BSc(Hons), MPH, GradCertBiostats Research Fellow and PhD Candidate Accident Research Centre, Monash University Building 70 Monash University Clayton VIC 3800 AUSTRALIA Tel: karen.stephan@monash.edu Joan Ozanne-Smith MBBS, MA(Prelim), MPH, MD, FAFPHM Head, Prevention Research Services Department of Forensic Medicine, Monash University Kavanagh Street Southbank VIC 3006 AUSTRALIA Tel: joan.ozanne-smith@monash.edu 26 HEALTH INFORMATION MANAGEMENT JOURNAL Vol 39 No ISSN (PRINT) ISSN (ONLINE)