Objective: To determine the shape of the income-mortality association, before and after adjusting for confounding by other socioeconomic variables.
Methods: Poisson regression analyses were conducted on 11.7 million years of follow-up of 25-59 year old New Zealand census respondents spanning four separate cohort studies (1981-1984, 1986-1989, 1991-1994, and 1996-1999).
Results: Mortality among low-income people was approximately two times that among high-income people. Adjustment for potential socioeconomic confounders (marital status, education, car access, and neighbourhood socioeconomic deprivation) halved the strength of the income-mortality association, but did not appreciably change the shape of the association. Further adjustment for labour force status largely removed the income-mortality association. The association of non-transformed income with mortality was non-linear, with a flattening out of the slope at higher incomes. Both the logarithm and rank of income appeared to have a better linear fit with the mortality rate, although the association of mortality with the logarithm of income flattened out notably at low incomes.
Conclusions: Much, but not all, of the crude association of income with mortality could be due to confounding. Adjusting income-mortality associations for labour force status (also a proxy for health status) is problematic: on the one hand, it over-adjusts the association as poor health will be on the pathway from income to mortality; on the other hand, it appropriately adjusts for both confounding by labour force status and reverse causation whereby income changes as a result of poor health. Both logarithmic and rank transformations of income have a reasonable linear fit with income.