Background
High-quality ethnicity data is essential for monitoring health trends by ethnicity. It is also needed to obtain quality information about Māori health and disparities to inform planning, development and evaluation of policies and interventions.
The Ministry publishes ethnicity data protocols for the health and disability sector that outline procedures to be used for the standardised collection, recording and output of ethnicity data for the sector (see Ethnicity Data Protocols).
Ethnicity classification in the cancer registration
According to previous research findings from the New Zealand Census Mortality Study (NZCMS), the ethnicity records in the death registrations for the years 2001–2004 showed no net undercount of Māori deaths. However, cancer registration data sets in the years 1981–2004 have been shown to undercount Māori cancer registrations.
Since 2009, the methodology used to assign ethnicity to a cancer registration examines the ethnicity recorded on the corresponding death registration, national health index (NHI) and hospitalisation record(s). A cancer registration is automatically assigned the ethnicity on a death registration and NHI (unless the ethnicity is not stated, or identified as ‘Other’). The ethnicity recorded on the hospital records is assigned to the cancer registration only if this particular ethnicity has been recorded on at least 20% of an all hospitalisation records for this person. For example, if a person is recorded as Māori on 9 hospitalisation records and is recorded as Samoan on 1 hospitalisation record it is likely that the recording of Samoan is incorrect, therefore Samoan would not be recorded as an ethnicity on the person’s cancer registration.
The Tatau Kahukura Māori health statistics do not adjust for an undercount, so cancer registration rates for Māori could still be underestimated.
Using this methodology means that when there are different ethnic groups recorded for an individual on different source data sets, multiple ethnicities are recorded on the cancer register. Further information about the current methodology used to assign ethnicity to cancer registrations can be requested from the Ministry of Health: email [email protected]
Ethnicity classification in the public hospital event records
Hospitalisation statistics have been shown to undercount Māori (see Tatau Kura Tangata: Health of Older Māori Chart Book 2011, Mātātuhi Tuawhenua: Rural Hospitalisations 2007 to 2011). To examine whether the quality of ethnicity data has improved in hospital event records, the hospital data set was linked to mortality data.[1]
For the purposes of these health statistics, death registration ethnicity was assumed to be a reliable count of Māori ethnicity data. Using encrypted NHIs, public hospital event records were linked to death registrations for those people who had both been admitted to hospital and died in the period 2008–12. Death records were only available up to 2012, whereas hospitalisation data was available up to 2014. The time period 2008–12 was chosen because it was the closest period to the period of interest for hospitalisations (2012–14), and it was wide enough to provide enough data to calculate reliable adjusters. The number of Māori hospitalisations using hospital event ethnicity was compared with the number of Māori hospitalisations using death registration ethnicity. Anyone recorded as Māori (either alone or in combination with another ethnic group or groups) was classified as Māori. Everyone else was classified as non-Māori.
Age group (in years) | Māori recorded at 2008–12 death registration | Māori recorded at 2008–12 public hospital admission [2] | Ratio (death/ hospital) |
Smoothed ratio [3] | Linked hospital and mortality data 2000–2004 (from Hauora IV) [4] |
---|---|---|---|---|---|
0–4 | 1680 | 1636 | 1.027 | 0.990 | 1.144 |
5–9 | 307 | 268 | 1.146 | 0.990 | 1.084 |
10–14 | 332 | 437 | 0.760 | 0.991 | 1.309 |
15–19 | 701 | 777 | 0.902 | 0.991 | 1.192 |
20–24 | 950 | 916 | 1.037 | 0.991 | 1.132 |
25–29 | 841 | 759 | 1.108 | 0.991 | 1.167 |
30–34 | 716 | 760 | 0.942 | 0.990 | 1.059 |
35–39 | 1571 | 1533 | 1.025 | 0.992 | 0.999 |
40–44 | 2147 | 2190 | 0.980 | 0.993 | 1.009 |
45–49 | 3975 | 4024 | 0.988 | 0.999 | 1.084 |
50–54 | 5520 | 5481 | 1.007 | 1.005 | 1.068 |
55–59 | 8241 | 8425 | 0.978 | 1.011 | 1.048 |
60–64 | 11,599 | 11,772 | 0.985 | 1.018 | 1.046 |
65–69 | 9017 | 8972 | 1.005 | 1.025 | 1.040 |
70–74 | 10,047 | 10,170 | 0.988 | 1.033 | 1.125 |
75–79 | 8598 | 8110 | 1.060 | 1.040 | 1.137 |
80–84 | 4901 | 4652 | 1.054 | 1.048 | 1.153 |
85+ | 2938 | 2635 | 1.115 | 1.056 | 1.161 |
Table A4.1 presents linked hospital and mortality data for the time period 2008–12. The ratios (death registration ethnicity/hospital event ethnicity) are all very close to 1, and increase with age. A ratio greater than 1 indicates more Māori hospitalisations using death registration ethnicity for that age group compared with Māori hospitalisations using hospital event ethnicity, and therefore suggests an undercount of Māori in the hospitalisation data. Undercounting of Māori is more likely to be seen among older age groups.
The data in younger age groups may be less reliable, due to the smaller numbers of deaths and the fact that there are therefore fewer linkages. Age-specific ratios were smoothed to account for the effect of low numbers in younger age groups and potential unreliability. Smoothing ratios were created using local regression with the LOESS procedure in SAS (version 9.1, SAS Institute Inc, Cary NC).
Table A4.1 shows the ratios developed for Hauora IV for comparison. The ratios of this chart book are generally closer to 1 than those found in Hauora IV, this is an indication of improved data quality for ethnicity records in the hospital data set.
Summary
For the purposes of Tatau Kahukura Māori health statistics, and assuming the death registration data records ethnicity data accurately, there is no undercount of Māori in the public hospital event records. Therefore, there is no need to adjust for hospitalisations. However, cancer registration rates for Māori could still be underestimated since Tatau Kahukura does not adjust for an undercount.
1. The linkage method described here was developed in Hauora: Māori Standards of Health IV: A study of the years 2000–2005 (Robson and Harris 2007); the authors would like to acknowledge those authors’ contribution to this report.
2. Short stay Emergency Department (ED) events were excluded.
3. The ratios were smoothed using local regression with LOESS procedure in SAS.
4. Appendix 3 of Hauora: Māori Standards of Health IV. A study of the years 2000–2005, from Harris et al.