Table 2.

Summary of eight studies using Australian data linkage systems to examine Indigenous identification in administrative data

1stAuthor, Year,State	Issue/Question	Study Approach	Findings
Mak 2008 WA	• 26% of 7,619 notifications of STI/BBVs in WA in 2004 were missing information on Aboriginality.	Infectious disease notification data for STIs/BBVs from 2004, WADLS link to mortality, hospitalisation, midwives and mental health datasets. Utilised various definitions for Aboriginal	• Aboriginality able to be assigned for 74% of cases with a missing Aboriginal identifier
	• Can data linkage help with better estimation of the true rates of disease?	Sensitive – any Aboriginal identifier	• Determining Aboriginality via data linkage was significantly and independently associated with sex and disease
		Specific – identified as Aboriginal on notification form or consistently on data linkage	• Indigenous disease rates and the Indigenous /non-Indigenous rate ratios decreased with improved Indigenous identification
Bradshaw 2009 WA	• Has ascertainment of Indigenous status improved in hospital separation data over time?	1998/99 Aboriginal cohort, WADLS link, identification in hospital data in admissions 1980 to 2006	• Substantial variation in Indigenous coding
			• Improved Indigenous coding since 2002; sensitivity since 2002 of >90%
Draper 2009 WA	• Over the years 1997–2002, the proportion of deaths in WA increased steadily from 0.6% to 6.6%	Linked deaths of unknown Indigenous status in mortality records through WADLS to hospital, mental health and midwives data.	• Indigenous status assigned to most people.
	• Could data linkage be used to estimate missing Indigenous status and what were the effects on life expectancy and mortality rates?	M1 = most frequent count	• “Unknowns” proportionately more likely to be Indigenous
		M2 = any Indigenous identification	• Under-ascertainment leads to elevated life expectancy and lower mortality rates
Briffa 2010 WA	• What is the effect of different algorithms for identifying Indigenous status through data linkage?	Patients hospitalised with CVD, WADLS linked 20 yr history.	• Modest increases by linkage using 50% of admissions
	• What are the demographic factors most associated with under-ascertainment	1. Index admission (baseline comparator)	• 20.8% increase if identified on one admission or death record
		2. Index admission or subsequent death record flag	• Older, less disadvantaged and urban living more likely to be under-identified
		3. At least 50% of admissions recorded as Indigenous
		4. Indigenous on at least 1 admission or death record5. Indigenous on at least 1 admission or death record
Katzenellenbogen 2010 WA	• Individuals may have inconsistent coding of Indigenous status in hospital records and/or be coded different on mortality data	Index = Acute MI in hospital or death data	For Indigenous identified patients on inclusive definition
	• What is the impact of 2 different methods for ascertainment of Indigenous status on acute myocardial infarction and 28-day case fatality rates in Indigenous people?	WADLS - admissions since 1980 and death data	• age standardised rates and age-standardised rate ratios were higher
		Sensitivity analysis	• case fatality was reduced
		Inclusive = ever identified Indigenous
		Restricted = coded as Indigenous on incident or death data
Neville 2011 NSW	• Can reporting of deaths among Indigenous people in NSW on the ABS mortality be improved by record linkage with the NSW Admitted Patient Data Collection (APDC)?	ABS mortality data for 2002–2006 were linked with APDC. Six algorithms were developed	• Maximised enhancement occurred with “Indigenous ever” but Algorithms 5 and 6 were considered most methodologically sound.
	• To investigate specific sources of bias caused by record linkage	1. Baseline = reporting based on ABS mortality	• Algorithms 3–5 relatively similar enhancement and relatively unaffected by the number of years of APDCs linked.
		2. Ever reported as Indigenous	• Enhancement in identification:
		3. Proportional record criterion (>50%)	○ varied by age (most in children 5–9 years and those> 85 years)
		4. Proportional facility-level criterion or ABS mortality data	○ was higher in females
		5. Proportional records and proportional facility criteria (50% of record in 50% of facilities) or ABS mortality data	○ was greater in urban areas
		6. Two or more records in two or more facilities in the APDC or ABS mortality data
		Also used different years of APDC and explored impact of age, sex and remoteness
Xu 2011 NSW	• To improve the statistical ascertainment of Indigenous mothers in NSW by linking the Midwives Data Collection (MDC) (which records the Indigenous status of the mother only) and the Indigenous identity as recorded on the Registry of Births Deaths and Marriages (RBDM)	Births in NSW 2001–2005	• The mother’s Indigenous status was highly consistent between the MDC and the RBDM
		An Aboriginal Statistical Variable (ASV) was created using the Indigenous identification in both datasets.	• The sensitivity was low in both data collections. At least one third of Indigenous mothers were not identified in the MDC and one-seventh in the RBDM
		The ASV was assessed by comparing numbers and percentages of births to Aboriginal mothers with the estimates by capture-recapture analysis	• Indigenous babies were more likely to be unregistered
Randall 2012 NSW	• Study examining mortality in Aboriginal and non-Aboriginal people after acute myocardial infarction in NSW	Used Aboriginal identifier based on the most recent public hospital admission for most of the analysis given recent improvements in Indigenous identification (88%), but undertook a sensitivity analysis based upon ‘ever identified’ and ‘all admissions’	• ‘Most recent’ identified 1183 (2.0%) of patients as Aboriginal; ‘ever identified’ identified 1479 (2.5%) and ‘all admissions’ identified 631 (1.1%) AMI patients as Aboriginal
			• In the fully-adjusted individual-level models, the ‘ever identified’ definition produced similar results to the ‘most recent’ definition, but the ‘all admissions’ definition resulted in higher odds of both 30-day and 365-day mortality for Aboriginal compared with non-Aboriginal patients

*For all studies linkage was based on probabilistic linkage. In WA this was undertaken by the WA Data Linkage Service and in NSW studies by the Centre for Health Record Linkage in NSW.