Estimating the direct and indirect pathways between education and diabetes incidence among Canadian men and women: a mediation analysis

Abstract

Purpose

To estimate the direct and indirect pathways between education and diabetes.

Methods

We examined the relative contribution of eight different pathways between education and diabetes incidence over a 9-year period in Ontario, Canada. Our data source was respondents (35–60 years of age) to the 2000–2001 Canadian Community Health Survey individually linked with physician and hospital administrative data. Our sample contained 11,899 participants with no previous diagnosis of diabetes. The direct and indirect effects of education level on incident diabetes were estimated using Aalen additive hazard models.

Results

Not having completed secondary education was associated with 120 extra diabetes cases per 10,000 men per year and 43 additional diabetes cases per 10,000 women per year, compared with having Bachelors education or higher. Body mass index accounted for 13 of the 120 extra diabetes cases among men, and 24 of the 43 additional diabetes cases for women.

Conclusions

Of the mediating pathways examined in this paper, body mass index was the pathway through which the largest number of diabetes cases was mediated among men and women. A substantial number of excess diabetes cases among respondents with lower education levels, in particular among men, were not mediated through any of the eight pathways examined.

Introduction

Diabetes is a growing public health concern in most developed countries. In Ontario, Canada, the prevalence of diabetes in 2005 had already surpassed the predicted global rate for 2030, almost doubling between 1995 and 2005 [1]. Like many health outcomes diabetes, prevalence is inversely distributed across socioeconomic position whether measured by education, occupation, or income [2], [3], [4], [5].

A precise understanding of the pathways through which lower socioeconomic position impacts on diabetes risk is unclear. Modifiable risk factors that could potentially mediate the relationship between education level and diabetes include body mass index (BMI) [6], [7], depression [8], smoking [9], physical inactivity [10], alcohol consumption [11], psychosocial work conditions [12], [13], and household income [14]. Unfortunately, many papers examining the relationship between socioeconomic position and diabetes have not distinguished between potential confounders and potential mediators in their examinations of education and diabetes risk. However, this distinction is critical, because the examination of mediators helps to open the “black box” between education level of diabetes risk [15], [16].

A handful of studies have examined the potential contribution of mediators in the socioeconomic position to diabetes relationship [17], [18], [19], [20], [21], [22]. In each of these papers, the amount of mediation attributed to each pathway between socioeconomic position and diabetes was assessed by examining the change in coefficients for socioeconomic position in a model with and without mediators (referred to as the Baron and Kenny approach [23]). However, in the case of a time-to-event or dichotomous outcome (as was the case in all papers), interpreting the amount of mediation by comparing the change in the hazard or odds ratio of the socioeconomic position measure between models is mathematically inconsistent [24], [25], [26]. This is because the ratio estimates in these models do not give an absolute unit-level size of effect, and so models with and without a mediator could be based on different unit levels. In the case of the Cox model, because the baseline hazard function is not estimated in the modeling process, the baseline hazard function in the models with and without the mediator could be different, because one model may have more undefined confounders than the other, also leading to incomparable models [26].

The objective of this paper is to further the research in understanding the relationship between education and diabetes risk among men and women. We sought to examine the potential importance of eight different modifiable pathways that may mediate the relationship between lower levels of education and risk of diabetes over a 9-year period in Ontario, Canada. To do this we implemented an Aalen additive hazard model [24], [27], using a marginal structural model approach (separately modeling the effect of education on each mediator and on the outcome). The marginal structural modeling approach is mathematically appropriate to assess mediation when the outcome is not normally distributed. Further, the Aalen model also allows specific estimates to be generated of the proportion, or number, of excess diabetes cases that can be specifically attributed to each mediating pathway, which is not possible with the Baron and Kenny approach. For example, how many of the additional diabetes cases in respondents with low education levels can be attributed to educational differences in BMI? Understanding the relative impact of each of these different pathways provides valuable research knowledge that can be integrated into strategies to reduce educational inequalities in diabetes in Canada.