The measurement of health outcomes for Aboriginal and Torres Strait Islander people is limited by inadequate identification of Indigenous status in data collected at state and national levels. Only about 56% of Indigenous deaths are reported to be registered accurately, and hospitalisation records are similarly unreliable, with information from some states considered inadequate for inclusion in the National Hospital Morbidity Database.1 A 1998 study of the quality of identification of Aboriginal and Torres Strait Islanders in Australian hospital data found correct recording of Indigenous status ranged from 55% to 100%, and varied according to the proportion of Aboriginal people living in the hospital catchment area,2 with accuracy highest in areas with large Indigenous populations. Western Australia was not included in this study. This variation in quality reduces the reliability of hospital administrative data as a source of data for Indigenous health research and makes ‘measuring the gap’ between health outcomes for Aboriginal and non-Aboriginal Australians difficult. The ‘gold standard’ against which the accuracy of recording of Indigenous status was measured was face-to-face interviews with hospitalised patients. This study was replicated in 2000 in Western Australia, where more than 10,000 patients were interviewed in hospitals across the state, of whom 754 (7.5%) identified themselves as Aboriginal.3 In this later study the accuracy was greater, ranging from 78.3% in the Perth metropolitan area to 93.5% in the Kimberley region. As there are indications that recording of Indigenous status has improved since 2000,4 we used information from the Perth Aboriginal Atherosclerosis Risk Study (PAARS) and the Western Australian Linked Data to update the information from WA and to examine trends in accuracy. The provided an opportunity to assess the accuracy of the recording of Indigenous status in the Western Australian hospital data over a 26 year period. Quantifying this aspect of validity of the administrative data will provide valuable information for other researchers using the database, and extend the opportunity to conduct high quality Indigenous and comparative health research.
Objective: To assess the sensitivity of the recording of Aboriginality in the Western Australia Linked Data.
Methods: This was a follow-up study using record linkage. Demographic data was obtained from 993 adult, urban-dwelling Aboriginal Australian participants in the Perth Aboriginal Atherosclerosis Risk study (PAARS). These were linked to the Western Australian Linked Data (State-wide hospital admissions and discharges, and deaths) to provide the number of admissions and Indigenous status coding from 1980 to 2006.
Results: There were 14,413 admissions for PAARS participants in the study period. The sensitivity of coding of Indigenous status in hospital admissions data significantly improved over time, exceeding 0.9 in every year since 2002. Prior to 2002 sensitivity was around 0.8, but poorer for males, with some anomalous years.
Conclusions: The coding of Indigenous status in the Western Australia Hospital Morbidity Database since 2002 has improved. The data from earlier decades must be approached with more caution.
Implications: The improved accuracy of identification of Indigenous status in the Western Australia Hospital Morbidity Database allows comparative studies of adult Aboriginal and non-Aboriginal population health outcomes to be undertaken with confidence.
To assess the sensitivity of the recording of Aboriginality in the Western Australian Hospital Morbidity Database between 1980 and 2006.
In 1998/99 we conducted a study of 993 self-identified urban Aboriginal people living in Perth. The PAARS participants were self-selected, but representative of the age structure of adult Aboriginal people in Perth. The study has been described in detail elsewhere.5 The ‘gold standard’ external dataset (the PAARS data) was electronically linked by the Western Australian Data Linkage Unit to administrative data including the Western Australian Hospital Morbidity Data, which includes hospital admissions and discharges for both public and private acute care hospitals State-wide in Western Australia, and to the registry of deaths within Western Australia. Probabilistic matching of records was based on name, date of birth, sex, and last known residential address. The validity of the matching for hospital morbidity has been assessed previously, with invalid links (false positives) and missed links (false negatives) estimated at 0.11% each.6 Linked data from January 1980 onwards was used to follow the participants' health outcomes to December 2006.
Each admission was searched to determine whether the patient was identified as Aboriginal or Torres Strait Islander in the ‘Aboriginality’ field. Codes 1-3 are used to identify patients as ‘Aboriginal’, ‘Torres Strait Islander’ or ‘Aboriginal and Torres Strait Islander’. Code 4 is assigned to patients who are not identified as Aboriginal or Torres Strait Islander. The identification of themselves as Aboriginal Australians by the participants in the PAARS was the gold standard against which the status recorded in the hospital data was measured.
The median and interquartile range of the total number of admissions for all PAARS participants in each year was calculated. Frequencies for correct identification were counted as a percentage of total hospital admissions for each individual. The proportion of individuals in whom identification was correct in 100%, or less, instances was recorded in increments of 10%.
Sensitivity was assessed by dividing the number of admissions with correct identification of Indigenous status in the Western Australian Hospital Morbidity Database by the total number of hospital admissions. Trends over time were measured by dividing the study period into calendar half-decades (1980-84, 1985-89, 1990-94, 1995-99, 2000-2004 and 2005/6) and recording the proportion correctly identified in each period. The chi-square test for trend was used to assess whether there was significant linear relationship between time and the proportion of admissions in which Indigenous status was correctly coded. A ‘p’ value of ≤0.05 was considered of statistical significance. STATA V.10 was used for all analyses.
Ethical approval for this study was obtained from the Western Australian Aboriginal Health Information and Ethics Committee, the Human Research Ethics Committee at the University of Western Australia and the Western Australian Confidentiality in Health Information Committee (Data Linkage).
There was a total of 14,413 admissions among 922 PAARS participants from 1980 to 2006. Of those, 69 patients (7.5%) had a single admission while 395 patients had 10 or more admissions (42.7% of total admissions), those with the highest number being patients undergoing renal dialysis. There were seven admissions with a missing value (no code for ATSI status assigned); these were not included in the analyses. There were 2,086 admissions (14.5%) coded as ‘4’ (not Aboriginal nor Torres Strait Islander), including 556 for 92 individuals (10.0%) who were not coded as ‘1-3’ in any admission. Less than half of the PAARS participants (n=369) were identified as an Aboriginal person at every hospitalisation during the study period. The proportions of admissions in which patients were coded as Aboriginal are shown in Table 1. No significant association was found between the proportion correctly identified and the number of admissions, or age.
|% of admissions coded as ATSI for a PAARS participant||N (%) of PAARS||Median number of admissions (inter-quartile range)||Minimum and maximum admissions per participant|
|100||369 (39.9)||6 (3-11)||1-154|
|90-99||135 (14.6)||21 (14-33)||10-755|
|80-89||100 (10.8)||11 (7-18)||5-230|
|70-79||69 (7.5)||9 (7-14)||4-78|
|60-69||40 (4.3)||5 (3-10.5)||3-31|
|50-59||38 (4.1)||4 (2-9)||2-140|
|40-49||23 (2.5)||7 (5-13)||5-26|
|30-39||21 (2.3)||3 (3-6)||3-23|
|20-29||23 (2.5)||5 (4-9)||4-33|
|10-19||12 (1.3)||9 (6.5-14)||6-17|
|<10||92 (10.1)||3 (1-6)||1-326|
|Total||922 (100)||8 (4-15)||1-755|
Admissions increased over time, especially among men, with more than double the admissions for 2000-04 than in any previous five-year period. There was a significant trend in the proportion of PAARS participants correctly identified as Aboriginal over time, with a noticeable increase in the accuracy of the HMD in the periods from the year 2000, compared with the previous half-decades (males χ2 trend18.9, df 4, p<0.001, females χ2 trend 10.8, df 4, p=0.001) (Table 2).
|Males (n=402)||Females (n=520)|
|Period||Total admissions (n)||Sensitivity||Total admissions (n)||Sensitivity|
Compared to women there was greater variability in the accuracy of the coding of Aboriginality for men over time (Figure 1) with a low point of 39% in 1997. From 2000 the accuracy of coding for men and women in the Western Australian Hospital Morbidity Database has exceeded 90% in each year.
The most notable finding was of the significant improvement in the accuracy of coding of Aboriginal status in the Western Australian Hospital Morbidity Data since the year 2000 for adults living in Perth. More than 90% of PAARS participants were identified as Aboriginal people in each admission assessed to the end of 2006. This increase in accuracy is heartening, and likely to be a true reflection of the quality of the hospital morbidity data. The cross-sectional study from 2000 in Western Australia3 found that Aboriginal males were less likely than females to be correctly recorded in hospital records (72% vs 82%), as we found in the years up to 2000. Neither study has information which might explain this disparity. The overall proportion of correctly identified admissions for Aboriginal people in Perth was 72% for the period 1995-1999, similar to that in other states of Australia in 1998, where correct identification was reported to be around 66% in areas with low proportions of Indigenous people, and 78.5% in capital cities.2 The improvement in Western Australia might indicate a trend to improved accuracy of health statistics in the first five years of this century, as the identification of Indigenous status among children aged 12-14 months in Australia is reported to have improved from 42% of the estimated national cohort in 2002 to 95% in 2005.4
We did not have sufficient information to assess social, institutional or other factors that may account for the inaccuracies and variability in coding of Aboriginality in Western Australia since 1980. For example, failure of identification of Indigenous status has been associated in the past with reluctance on the part of clerical admissions staff to ask patients about their status as an Aboriginal or Torres Strait Islander or otherwise.2,7 Greater awareness of the value of the identification of Indigenous status for health research purposes, and to the development of health policy aimed at narrowing the gap between Aboriginal and non-Aboriginal Australians may have increased clerical accuracy in more recent years. Other factors, such as the growth in Indigenous health services also may have aided the identification of Aboriginals on admission to hospital, as referral from these services is associated with a greater likelihood of being correctly identified.7 The PAARS participants were drawn from the Perth electoral region where Aboriginal and Torres Strait Islander peoples comprise less than 2% of the population. The accuracy of identification of Indigenous status in rural health services in Western Australia should be at least equivalent, as areas with greater concentrations of Aboriginal people are reported to have more accurate coding in medical records.2,3
The strengths of this study are the use of the PAARS participants, one of few large, well-characterised populations of self-identified urban Aboriginal Australians, the length of the time over which hospital admission data were accumulated and the availability of these data at the State population level. This provided a sound basis for establishing the accuracy of the recording of Indigenous status over time for adult Aboriginals in the Western Australian health administrative database.
A limitation to this study was that non-Aboriginal people were not included; therefore we could assess sensitivity, but not other aspects of validity (specificity and predictive value) of the Western Australian Hospital Morbidity Database. However, the value for comparative studies is not diminished, as the earlier cross-sectional study from Western Australia reported that non-Indigenous people were correctly identified in 98.5% to 99.7% cases across metropolitan and regional areas.3
The results from this study cannot be generalised to other states of Australia without confirmation that coding of Indigenous status among adults has similarly improved since the 1998 study.2 The study was conducted among adults so cannot provide reliable estimates of the accuracy of coding for Aboriginal children admitted to Australian hospitals.
The Western Australia Hospital Morbidity Database has become a more accurate tool for the identification of adult Aboriginal people since 2002. Indigenous status for 926 Aboriginal PAARS participants was recorded accurately in more than 90% of hospital admissions. Prior to 2002, sensitivity was around 80%, but poorer for males, with some anomalous years, so the data from the earlier decades must be approached with caution.
Implications for population health research
The improvement in the accuracy of identification of Indigenous status over time means that comparative studies of morbidity resulting in admission to hospital can be undertaken with confidence among Aboriginal and non-Aboriginal adults in Western Australia. The Western Australian Hospital Morbidity Database is an important tool for health research, with the capacity for true population studies.
This study was supported by a grant-in-aid from the Heart Foundation of Australia. The Aboriginal Advisory Committee comprised Lyn Dimer (Heart Foundation of Australia), Francine Eades (Derbarl Yerrigan Health Service Inc.), Tana Dunshire (Aboriginal Health Council, WA), Darryl Kickett (Aboriginal Health Council of WA), Colin Garlett (CEO, Derbarl Yerrigan Health Service Inc.), Roz Yarran (Board member of DYHS & Canning Division of General Practice). The Community Reference Group members are Darryl Kickett, David Kelly, Debbie Stobie, Roz Yarran, Tracey Kickett, Sharon Bushby, Lyn Dimer, Francine Eades and Tiew-Hwa Katherine Teng. Thanks also to Janice Reidy from DiabetesWA.