Summary of evidence and interpretation
We identified 30 studies, which evaluated the impact of diabetes on labour market outcomes, which imply a complete absence of any occupation. The available studies were quite heterogeneous in terms of definition of outcomes, age of the population considered and statistical method used even within the four outcome clusters we identified. Generally, the studies included provide consistent evidence that diabetes is negatively associated with employment and that diabetes patients are more likely to retire early, be fully out of the labour force and to receive a full and permanent disability pension, although effects may vary across subgroups.
The studies included also show considerable differences in the methods used, which could significantly impact the results. Furthermore, evaluations are often based on an extremely simplified modelling of diabetes, its dynamics and its progression, resulting in potential sources of bias. In this context, the majority of data is based on self-reported diabetes status and often no heterogeneity factors or endogeneity of the labour market outcomes are considered, resulting in lower quality scores for several studies included.
Specifically, a stratified analysis using potential sources of heterogeneous effects, such as gender, age, age at retirement or diabetes type, was inconsistently carried out throughout the studies, limiting the comparison of results regarding different groups within the scope of this review. In fact, a consistent stratified analysis between genders is available only for the outcome “employment”. For the other outcomes, only isolated evidence with a high risk of bias could be found [29, 30, 34, 39]. As shown in many of the studies included [16, 19, 25, 27, 30, 32, 35, 39, 42] and in a previous review [7], both men and women suffering from diabetes have higher chances of adverse labour market outcomes, but within the same studies, the effect is generally higher for men than for women. However, no study furnished an evidence-based explanation of this result. The main interpretation is that, since the employment chances of elder females are already low due to several other factors (e.g. providing informal care, traditional household regimes), diabetes influences the employment chances of women in a less disruptive way than those of men. In this context, also the differences between studies from LMICs and other countries should be emphasized: the effect of diabetes for the employment and early retirement chances of women in LMICs is never significant, while the effect for men is in line with those observed in HIC [24, 31, 39]. The non-significant effect for women should be put in the right context and should be interpreted in the light of labour market differences, regarding most notably the social security systems and the role of women in society, which still characterize the divide between HIC and LMICs and which could significantly affect the employment chances of women in the first place. However, in line with previous studies [11], this review highlights also the paucity of evidence regarding the differences between HIC and LMICs, since only three of the included studies focused on the latter [24, 31, 39], and thus highlights the need for more research on these differences.
Most studies were based on large survey data, where diabetes status was self-reported (see Table 1). Although previous studies showed that there is a high correspondence between self-report and objective diagnosis [48, 49], this implies that most of the available evidence regarding the effect of diabetes on labour market outcomes bases its analysis and conclusions on a subjective measure of diabetes and is thus potentially open to bias. This bias is expected to be upwards, since the undiagnosed cases are probably those who also do not show any symptom or impairment from the disease, and as such are much less likely to leave the labour force due to diabetes. This potential pitfall is reflected in the lower quality score assigned to those studies based on self-report of diabetes and should be considered as an important limitation of the available evidence in this field.
Furthermore, in the same studies, no other information about age at onset, diabetes type, severity or medications was available, according to the publications identified. One important distinction in this context is that between T1DM and T2DM. Although the prevalence of T1DM is usually low [1], not controlling for this difference could cause a downward bias and, thus, an underestimation of the effect of T2DM on employment. In fact, the few studies that distinguish between the two diabetes types show that the negative effect of diabetes on employment is actually driven by T2DM, since the coefficients on T1DM are either insignificant or even significantly positive. Furthermore, T1DM and T2DM are two distinct conditions, with two different aetiologies and ways of coping with the illness. Therefore, this difference should be taken into account when modelling diabetes. For example, in absence of more detailed information, the age at onset could offer a good approximation, as already done in some of the studies included [25, 26].
Most studies also adopted a very simplified modelling of comorbidities and complications. These factors can play a crucial role in the ability to work of diabetes patients over the life course and, thus, should be considered when modelling diabetes and labour market outcomes. There is no consensus on how to take them into account. In most of the studies considered, they are either not taken into account or are modelled as confounders. However, as highlighted by some authors [25], simply adding them as confounders could be problematic, since they might be highly correlated with diabetes or a result of common unobserved factors. Therefore, including them as covariates into the model could result in biased estimates for the diabetes variable. In isolated cases comorbidities and complications are included [1] as confounders in different versions of the model as further specification [37, 43, 46], [2] as a way to differentiate the exposure variable (diabetes with/without complications) [41] or [3] as exposure in a further analysis focusing only on the diabetes group [27]. These three implementations show that adding such confounders leads to a change in the magnitude or in the significance of the coefficient on the diabetes variable [37, 43, 46]. In addition, Kraut et al. (2001) [41] showed that only diabetes with complications leads to a full labour market exit. Ng et al. (2001) [27] also revealed that people suffering from diabetes with complications have a higher chance of being out of the labour force than people suffering from diabetes without complications.
A further issue, only addressed in a few studies, is the problem of reverse causality or endogeneity of diabetes in labour market outcome models. Typical ways for taking this problem into account include recursive multivariate probit approaches [20, 21, 23, 24] or the use of genetic instrumental variables [25, 31]. Results from studies taking endogeneity into account generally differed in two aspects: (i) the actual endogeneity of the diabetes variable and (ii) the direction of the bias in the regression coefficients with respect to the basic model without endogeneity. Overall, diabetes was not found to be consistently endogenous in each study considered and for every gender subgroup. Furthermore, while comparing the results from models with and without endogeneity within the same study, no clear direction of the bias of the coefficients could be highlighted (see Table 2). Therefore, since the pattern of presence and effect is not clear, endogeneity should always be tested for in this context and the limitations of results should be discussed carefully.
Strengths and limitations
This review specifically gathered evidence regarding the effect of diabetes on all labour market outcomes involving the complete absence of occupation. Hence, it complements related reviews, which focused on other productivity outcomes [9] or reviewed part of the included outcomes as a secondary aim [11]. Furthermore, in the present review, we paid specific attention to the methods used, providing ground for an evidence-based discussion on how to produce credible and robust findings both from an economic and a statistical point of view.
However, our study may suffer from some limitations. First, we have adopted rather restrictive inclusion criteria. We searched three databases and we included only articles already published in peer-reviewed journals, starting from the year 2000. Therefore, the review might suffer from publication bias. However, the large number of studies initially retrieved after an independent screening by two researchers and a comprehensive reference check allowed us to apply such restrictive criteria in order to report the most robust evidence available. Second, we based our quality and risk of bias assessment on the Newcastle-Ottawa Scale [17], as already done in similar reviews [9, 10]. Besides the transparent procedure of evaluation, the scale had to be modified for our specific case, which prevents comparability to a certain extent (for detailed explanation see Additional file 2). Furthermore, the scale is actually suitable for evaluating epidemiological studies involving clinical outcomes but could still be adapted to our specific question and context. Although the scale represents the best instrument available to our knowledge, this problem should be taken into account in further studies, aiming at improving also quality and risk of bias assessment.
Implications for practice, policy, and research
The aggregated evidence available reveals that generally, individuals suffering from type 2 diabetes mellitus are more likely to fully exit the labour market early, retire early and receive a permanent disability pension. Both men and women are affected, but the probability of employment of men is affected stronger than that of women. Diabetes can be endogenous in the labour market outcomes, but it is not clear why and in which cases it is present and how coefficients are influenced.
Maintaining and possibly also extending the ability to work of older workers is one of the primary goals of current pension reforms. This study shows, however, that chronically ill individuals suffering from T2DM, might not be able to maintain their employment status and will therefore exit the labour market earlier. Since T2DM prevalence is rising, not only in high- but also in low- and middle-income countries [1], a considerable effort should be undertaken to improve and prolong the ability to work of diabetes individuals. Specific attention should be paid to developing and increasing the efficacy of evidence-based prevention and management programs.
Finally, the existing evidence should be improved, specifically investigating the underlying dynamics and establishing and strengthening the link to practice. First, future cost studies investigating the indirect costs of diabetes should take the complete absence of an occupation due to diabetes or its complications into account. Failing to consider this aspect could lead to a severe underestimation of the burden this condition is imposing. Second, future studies will need to differentiate between gender and/or diabetes type, while also checking specifically for the endogeneity of diabetes. These methods should be applied for every outcome, not only for the presence versus absence of employment. Third, the issue of diabetes endogeneity should be discussed for each study, since no pattern of presence and effect could be found. Understanding how the underlying processes and effects work, being it through reverse causality or through unobserved factors, could also prove helpful in understanding how a chronic life-style illness impacts the outcomes considered. Lastly, the available studies adopt an extremely simplified definition and modelling of diabetes, its progression, its severity and its complications and comorbidities. Further research should rely on more objective ways to determine diabetes. Also, it should improve the understanding of which factors and dynamics actually lead to adverse labour market outcomes and should include different modelling strategies on how comorbidities and complications actually work. Furthermore, additional aspects of the illness, such as efficiency of management, health literacy, and medication adherence [50, 51], should be included in the analysis, to gather further understanding on underlying factors and allow for the individualisation of concrete starting points for practical intervention.