1 Key concepts, data, methods and results Index Trends in cancer survival by ethnic and socioeconomic group, New Zealand, Soeberg M, Blakely T, Sarfati D, Tobias M, Costilla R, Carter K, Atkinson J A study published by the University of Otago and Ministry of Health, 2012 CancerTrends A study funded by the Health Research Council and the Ministry of Health

Structure of this presentation 2 Current knowledge and gaps in knowledge Measuring cancer survival Data and methods Results and interpretation

Current knowledge, and gaps in knowledge 3

Current New Zealand evidence 4 Cancer survival is improving over time But little is know about the magnitude of these changes over time, including for each ethnic and socioeconomic group.

Current New Zealand evidence 5 Ethnic and socioeconomic inequalities in cancer survival exist But little is know about whether these inequalities are narrowing or widening over time.

Study objectives 6 To present cancer survival trends for 21 adult cancer sites in New Zealand from with follow-up to 2006 for: –Ethnic groups (Māori and non-Māori separately) –Income groups (low income and high income patients separately) And to assess gaps in survival between: –Māori and non-Māori averaged over time, and for any change in time –Income groups averaged over time, and for any change in time.

Study objectives 7 Changes over time in cancer survival by ethnic and socioeconomic group This study measured changes over time in cancer survival for each ethnic and socioeconomic group.

Study objectives 8 Cancer survival inequalities, averaged over time This study measures the gap between ethnic and socioeconomic groups, averaged over time. This study also measured ethnic and socioeconomic cancer survival inequalities, averaged over time.

Study objectives 9 Changes over time in cancer survival inequalities This study also measured changes over time in ethnic and socioeconomic cancer survival inequalities.

Measuring trends in cancer survival 10

Measuring cancer survival 11 Time-to-event studies In this study, we were interested in the time from cancer diagnosis to the event (in this case death). Cancer diagnosis Death Time

Measuring cancer survival 12 Time-to-event studies, where death from a specific cancer is of interest Some studies in NZ have looked at the time from a cancer diagnosis to death from the diagnosed cancer (cause-specific survival). Breast cancer diagnosis Death from breast cancer where deaths from all other causes are censored Time but the quality of cause of death data in New Zealand is poor.

Measuring cancer survival 13 Time-to-event studies, where deaths from any cause are of interest An alterative method is relative survival where deaths from any cause are the event of interest, but where all other causes of death are accounted for. Breast cancer diagnosis Death from any cause taking into account all other causes of death Time

Measuring cancer survival 14 Relative survival The relative survival ratio is commonly used in population- based cancer survival studies. RSR of 0.80 = 0.75 (observed survival) / 0.92 (expected survival)

Measuring cancer survival 15 Key disadvantage of relative survival Non-comparability bias is introduced in relative survival analyses where the mortality rates in the cancer and non- cancer populations are not comparable. Mortality rates in the Māori cancer population Mortality rates in the total non-cancer population

Measuring cancer survival 16 Key disadvantage of relative survival Using simulated data, it was possible to consider the impact of non-comparability bias for the research questions in this study. Five-year RSR for breast cancer Using total population life tables Using social group-specific life tables Difference Most advantaged group % Least advantaged group %

Measuring cancer survival 17 Non-comparability bias leads to: Modest to moderate under-estimation of relative survival for Māori and the most deprived groups Slight over-estimation of relative survival for non-Māori and the least deprived groups Over-estimation of ethnic and socioeconomic inequalities in cancer survival, at each calendar period Little impact on trends in ethnic and socioeconomic cancer survival inequalities Key disadvantage of relative survival

Measuring cancer survival 18 Sparseness of data Relative survival is bound by the values of 0 and 1 Does not allow for simulatenous consideration of multiple factors associated with cancer survival, e.g. age, stage at diagnosis, follow-up time since cancer diagnosis Other disadvantages of relative survival

Measuring cancer survival 19 Survival and mortality scales Relative survival can also be presented on an excess mortality rate scale (mirror image of relative survival). Relative survival scaleEquivalent annual excess mortality rate scale

Measuring cancer survival 20 Regression methods have been developed to model cancer excess mortality Scale is bound between 0 and positive infinity Allows for the various factors associated with trends and inequalities in cancer survival to be accounted for, e.g. age sex ethnicity socioeconomic position calendar period follow-up time since cancer diagnosis interaction terms. Modelling excess cancer mortality rates

Measuring differences in cancer survival 21 Cancer survival varies by calendar period Cancer survival varies by ethnic and socioeconomic group Cancer survival varies by combinations of calendar period and ethnic and socioeconomic group (allowing for investigation of trends in ethnic and socioeconomic inequalities in cancer survival) Reasons to measure differences in cancer survival

Measuring differences in cancer survival 22 Absolute and relative differences On the relative survival ratio (RSR) scale On the excess mortality rate (EMR) scale Ways to measure differences in cancer survival

Measuring cancer survival 23 A framework for absolute and relative differences in cancer survival MeasureScale AbsoluteRelative Relative survival Relative survival ratio difference (RSRD) Ratio of relative survival ratios (RSRR) Excess mortality rate Excess mortality rate differences (EMRD) Excess mortality rate ratio (EMRR) Cancer survival inequalities can be assessed using absolute or relative measures calculated on the RSR or EMR scales.

Measuring differences in cancer survival 24 Different conclusions from the same data ScaleCancer siteAbsolute measure Relative measure Five-year relative survival scale RSRDRSRR Breast Colorectal Lung Annual excess mortality rate scale EMRDEMRR Breast Colorectal Lung In this study, we have mostly measured the RSRDs and the EMRRs.

Data and methods 25

Data and methods 26 Cancer population data (linked Census, cancer and mortality records) Non-cancer population data (ethnic- and income-specific life tables) Relative survival analyses for 3 calendar periods Excess mortality rate analyses for all patients diagnosed Observed and expected survival data and analyses

Data and methods 27 Linked Census, cancer and mortality data Cancer cases 1991* – * * Dx 2. DxDied 3. DxDied 4. Dx 1991 Mortality follow up period 2006 * 1991, 1996 and 2001 were Census years

Observed survival data 28 Approximately 80% of cancer registrations were linked to Census records, with 95% of those being true links. Between 11% and 15% of records were excluded because their income was missing, but only approximately 1% were excluded because of missing ethnicity data. Between 6% and 9% of records were excluded because they had zero survival time (mostly their basis of cancer diagnosis was from death certificate). Stage at diagnosis was not included as a variable in analyses due to large variations in the quality of reporting stage over time. Linked Census, cancer and mortality records

Observed survival data 29 A total of 147,344 patients were included in relative survival analyses by ethnic group for patients diagnosed A total of 127,305 patients were included in relative survival analyes by income group for patients diagnosed A total of 125,567 patients were included in excess mortality analyses for patients diagnosed Total number of patients included in analyses

Expected survival data 30 Life tables are an essential input in relative survival and excess mortality analyses Life tables provide data on the expected survival and the mortality from all other (non-cancer) causes of death Ethnic-, income- and combined ethnic- and income-specific life tables were constructed for this study for the periods 1991, 1996 and 2001 Minimising the impact of non-comparability bias

Expected survival data 31 Example of data from life tables Probability of a person aged x surviving to age x + 1

Statistical analyses 32 Estimation of relative survival ratios (RSRs) –1-year and 5-year RSRs by ethnic and income group for patients diagnosed , , –Ethnic-specific and income-specific life tables used –RSRDs calculated for ethnic and income group differences at each calendar period Relative survival and excess mortality analyses

Statistical analyses 33 Excess mortality rate (EMR) modelling –Four EMR models run for each cancer site to estimate a) ethnic trends in cancer survival and b) income trends in cancer survival –EMRRs derived from EMR models to assess a) trends in survival, b) inequalities in survival, and c) trends in survival inequalities –Pooled EMRRs estimated across cancer sites –Combined ethnic- and income-specific life tables used Relative survival and excess mortality analyses

Results 34

Trends in cancer survival 35 Cancer excess mortality rates reduced by 26% per decade Equivalent to a 3% reduction per annum in excess mortality rates

Trends in cancer survival 36 Changes in the date of diagnosis and/or the date of death through improvements in treatment, and/or advances in diagnosis, and/or the introduction of cancer screening. Possible explanations

Ethnic inequalities in cancer survival 37 Māori had 29% greater excess mortality compared to non-Māori Māori had 29% greater excess mortality compared to non-Maori

Income inequalities in cancer survival 38 Low income had 12% greater excess mortality compared to high income Low income patients had 12% greater excess mortality compared high income patients

Inequalities in cancer survival 39 Differences between ethnic and socioeconomic groups in: stage at diagnosis (not adjusted for in this study) quality and timing of treatment patient factors, such as co-morbidities (and possibly tumour biology) Possible explanations

Trends in ethnic inequalities in cancer survival 40 % changes per decade in absolute and relative differences MeasureScale AbsoluteRelative Relative survival RSRD Possible 18% decrease to a possible 41% increase per decade RSRR 20-24% decrease per decade Excess mortality rate EMRD 25% decrease per decade, with a possible 13% to 35% decrease EMRR 4% increase per decade with a possible 6% decrease to 14% increase There was little change over time in ethnic inequalities when looking at the change in the EMRR. but a narrowing of ethnic inequalities over time when looking at the EMRD and RSRR.

Trends in income inequalities in cancer survival 41 % changes per decade in absolute and relative differences MeasureScale AbsoluteRelative Relative survival RSRD Possible 14% decrease to a possible 40% increase per decade RSRR 20-23% decrease per decade Excess mortality rate EMRD 24% decrease per decade, with a possible 17% to 30% decrease EMRR 9% increase per decade with a possible 1% to 17% increase There was a 9% widening over time in income inequalities over time when looking at the per decade change in the EMRR. but a narrowing of income inequalities over time when looking at the EMRD and RSRR.

Trends in cancer survival inequalities 42 Different rates by ethnic and socioeconomic group over time in the receipt of cancer detection, diagnosis and treatment services (the ‘inverse equity’ hypothesis) Differences over time in the recording of ethnicity Use of absolute and relative measures on the RSR and EMR scales Changes in the income gap distribution between Māori and non-Māori driving changes in ethnic inequalities in cancer survival Possible explanations

Conclusions 43 Cancer survival is improving over time for all cancer sites, with variation by cancer site in the magnitude of those improvements Ethnic and, to a lesser extent, socioeconomic inequalities in cancer survival were reported for the majority of cancer sites There was evidence of a relative increase per decade in excess mortality comparing low- to high-income groups

Acknowledgements 44 This work was supported by the Health Research Council of New Zealand and the Ministry of Health. Access to the data used in this study was provided by and sourced from Statistics New Zealand under conditions designed to give effect to the security and confidentiality provisions of the Statistics Act The results presented in this study are the work of the authors, not Statistics New Zealand.