The effects of diagnosed and undiagnosed diabetes type 2 on labour market outcomes
Diabetes is an expanding chronic disease, it has been estimated that there will be 783 million diabetic people worldwide in 2045, while only 536 million people were affected by diabetes in 2021 (International Diabetes Federation, 2021).
Diabetes has been studied both under a medical and a socio-economic lens. In the former case, it has been argued that diabetes is related to diseases such as depression (Roy and Lloyd, 2012), hypertension and cardiovascular diseases (Sowers et al., 2001). In the latter case, economists refer to the human capital theory, considering health as an endowment of human capital that has a positive influence over labour market outcomes, since healthier people can devote more time to their jobs (Grossman, 1972). The estimation of the costs caused by diabetes is of great interest to the economic academic literature. Expenses for diabetes mellitus (DM) type 2 are the most relevant for the Netherlands, as DM type 2 is the most common type of diabetes in the country. Indeed, 90% of Dutch diabetic people have DM type 2 (Peters et al., 2017). Therefore, in our analysis we will mainly focus on type 2.
The costs caused by this type of diabetes are both direct, such as health expenditures, as well as indirect, such as decreased labour market participation (Currie and Vogl, 2013). Furthermore, since one in two adults living with diabetes is undiagnosed (International Diabetes Federation, 2021), it is rather likely that the absence of a diagnosis increases the costs sustained by society. Long periods of time without a diagnosis, thus, without proper treatment, might lead to poorer health and labour market outcomes.
Focusing on labour market participation, many studies (a review of them can be found in Pedron et al. (2019)) have found a negative association between being
openly diabetic and labour market participation. Furthermore, the negative association of diabetes is stronger for men (Brown et al., 2004; Seuring et al., 2015). However, low- and medium-income countries seem to report different effects for women than high income ones (Zhang et al., 2009; Seuring et al., 2019; Koch and Thsehla, 2022).
Recently, thanks to the availability of blood sample data, researchers have been able to address people with undiagnosed DM, which constitute a large fraction of people affected by diabetes. This is an understudied population of diabetic people that is likely to differ in many ways from the diagnosed one, as an example diagnosed people might suffer from poorer health, hence why they got a diagnosis. Recent cross-sectional estimates indicate that undiagnosed people might not to suffer from lower labour market outcomes, as it is indeed the case for diagnosed ones (Minor and MacEwan, 2016; Seuring et al., 2019). However, the literature does not present results obtained from panel data or instrumental variable estimation for undiagnosed individuals, thus cannot account for unobserved time-invariant heterogeneity and does not follow individuals over time. Following undiagnosed people over time is likely to produce different estimates from those reported in the literature, as it has been previously found that undiagnosed diabetic people are more likely to experience poor clinical outcomes (Minor and MacEwan, 2016; Umpierrez et al., 2002).
The ability to follow undiagnosed people over time is the most innovative element of our research, as previous works were only able to gather cross-sectional data (Seuring et al., 2019). Furthermore, since Lifelines contains individuals that are related to each other, we can account for household fixed effects and/or presence of diabetes in the family. This is important as individual belonging to the same family are likely to share some traits and behaviours, such as dietary habits. To the best of our knowledge, this would be the first project to focus on undiagnosed diabetes and its labour market effects over time.
We expect that people that stay undiagnosed for a long time are more likely to experience complications, thus, lower labour market outcomes. Trying to assess health complications, we will control for the most usual health complications, such as hypertension and cardiovascular disease (Sowers et al., 2001). Furthermore, to capture the general health state of the individual, we will construct a measure of allostatic load (AL). AL is used to measure the “wear and tear on the body” (McEwen, 1998) in the medical literature and it is associated with physical and mental health (Guidi et al., 2020).
Future research may replicate our approach to investigate other pathologies that are frequently undiagnosed, such as depression, thanks to the richness of the dataset provided by Lifelines.
In our research we will pay careful attention to the endogeneity issue, such as omitted variable bias, wondering whether having diabetes was caused by individual
traits or behaviours that influence labour market outcomes as well. In addition there might be reverse causality issues, i.e. labour market events might influence diabetes (see Bergemann et al. 2020). We will uses a variety of methods in order to investigate the sensitivity of our results. We will be contrasting individual and family fixed effects models with models using instrumental variables based on familial diabetes as well as using Mandelian randomization using genetic data. Mandelian randomization is a well-established method used to study the socioeconomic effects of diseases (see for example Harrison et al. 2019). However, to the best of our knowledge it has not been applied yet with respect to our specific question. The richness of the Lifelines data makes such unique comparison of methods possible. This addresses the issue that most articles concerned with the causal relationship between diabetes and employment have not considered endogeneity issues sufficiently (see Pedron et al., 2019), they typically use very simple models to analyse the effects of diabetes on labour market outcomes. Regarding the few studies that have focused on endogeneity, overall results are quite inconsistent (Pedron et al., 2019). In our modelling approach we will hereby build upon existing knowledge of the association of socioeconomic position, genetic predisposition and diabetes prevalence (see for example van Zon et al. 2018).