Introduction Ethnicity is difficult to collect in a standard way between health and census datasets, meaning routinely calculated rates are prone to numerator-denominator bias.

Methods Census and mortality data have been routinely linked in the New Zealand Census-Mortality Study, creating time series data from 1981. More recently, the Cancer trends study has linked cancer registrations. The proportion of eligible mortality / cancer records linked to a census record is now over 80%. Weighting is used to adjust for any linkage bias.

Results Previously calculated mortality rates for Māori and Pacific people were shown to be underestimated by a quarter due to numerator-denominator bias. Corrected trends in mortality rates since 1981 demonstrate little if any improvement in Māori mortality rates during the 1980s–90s, concurrent with major reforms that impacted more on Māori than non-Māori (eg, unemployment rates among Māori reached 25% in 1992, compared with 7% among Europeans).

Other analyses demonstrate that CVD is the major cause of death giving rise to ethnic inequalities in mortality, but as CVD rates fall for all ethnic groups inequalities in mortality are “relocating” to cancer. Inequalities in cancer incidence are less prominent than inequalities in cancer mortality, but nevertheless there are curious findings, such as: higher (and faster rising) breast cancer incidence among Māori, despite a more favourable risk factor profile; and particularly high endometrial and thyroid cancer incidence among Pacific women. There has also been a tremendous reduction in cervical cancer inequalities.

Conclusion This linkage has allowed accurate research on trends that was hitherto impossible.