Affective disorder and diabetes in Taiwan
In this study, we found that for Taiwanese subjects with T2DM aged 50 years or over, having no oral anti-hyperglycemic therapy is a risk factor for AD, with an overall increase in risk of 2.62-fold during the period 1996 to 2007. One reason our findings are not negative , unlike those of other studies [12–15], may be that they were not masked by therapeutic interventions for diabetes that we have previously found to reduce the risks of dementia  and Parkinson's disease . Another is that in Taiwan, as elsewhere, the burden of diabetes in an ageing population is increasing , and thus the phenomenon is more recognizable. However, in some community-based ageing populations such as that of Australia, older people do not necessarily show an increased prevalence of depression . It is possible that diabetes might play a greater role in the development of depression or AD in communities in which its prevalence increases to a greater extent in later life. Gender did not seem to influence the status of diabetes as a risk factor for AD, whether unipolar or bipolar. This does not diminish the importance of gender differences in risk for depression or AD that have been recognized for other factors , or of the pathogenetic and clinical differences between the major types of AD, but it does emphasize the possible commonalities through the diabetes linkage and raises questions about which of these might be relevant in prevention and management.
There has been no study to date of T2DM and AD on East Asian populations, where marked increases in T2DM prevalence have occurred in recent years, and where almost half of the world's population is exposed to diabetes, and consequently to a potential threat to mental health. Our study makes a contribution to the relevance of diabetes as a risk factor for AD in East Asian populations, as Taiwan has a dominantly Chinese ethnic population and is therefore an example of a large under-studied group.
Preventive or therapeutic benefit of oral anti-hyperglycemic agents for affective disorder
The available studies have not considered diabetes pharmacotherapy in relation to the effect of diabetes on the risk for depression or AD, which may have led to lower estimates of the possible reduction in risk of AD by diabetes therapy. In the meta-analysis on Europeans and North Americans carried out by Nouwen et al. , the increased risk for depression was 24% and in the study by Nefs et al. , it was 14%; given our findings, these were probably underestimates. We found the increased risk for incident AD in subjects with T2DM untreated by OAA to be 2.6-fold (both unipolar and bipolar AD HRs were similar, at 2.81 and 2.31, respectively), which suggests that the available studies have grossly underestimated the potential effect of diabetes in this area of mental health.
The combination of sulfonylurea and metformin significantly reduced both incidence and relative risk for AD in T2DM, irrespective of gender; there was no such change evident for sulfonylurea or metformin alone (Table 2, Table 3). It also did not matter in which order sulfonylurea or metformin were started. For any particular intervention, patients were extensively matched with their reference non-OAA subject, using counterparts with the same date of diabetes diagnosis, within the same calendar year. This reduced the possibility that a change in accepted diabetes management might have accounted for the observed effects, whether in men or women. Bias, especially from time, was minimized, and thus made this cohort study more similar to a clinical trial [29, 30]. Differences in referent subjects between treatment categories were addressed by matching for matters such as accessibility to care (by locality), level of care (by primary to tertiary or medical center) and associated health status (CCI) (Table 3).
Despite adjustment for co-morbidities and the exclusion of incident AD for up to 3 years after introduction of OAA therapy, the combination of sulfonylurea and metformin was still associated with HRs of less than 1.00 compared with no OAA therapy. This applied to both unipolar and, to a greater extent, bipolar AD. Thus, although differences between these two forms of AD are well documented and with agreed therapies, our findings raise the possibility that there are some common underlying pathogenetic factors, perhaps related to neurodegeneration.
The findings for the sulfonylurea/metformin combination rather than either alone would be consistent with synergistic or complimentary mechanisms of action of the two agents. Because advanced glycation end-products are found in the central nervous system (CNS), specifically in microglia , these two OAAs together might act via a more pronounced anti-hyperglycemic effect. Such an effect might be combined with specific AD-reduction actions of the drugs. As mentioned above, metformin might operate through mechanisms that are protective against neurodegeneration, if this is a shared pathogenesis with AD. When diabetes affects the CNS, AD might share underlying mechanisms with neurodegenerative diseases such as dementia and Parkinsonism [23, 24]. We previously reported that T2DM also increases the risk of dementia by 2.6-fold, and that the combination of sulfonylureas and metformin was able to decrease the risk of dementia in diabetes within 12-years , with similar findings in PD . These collective findings strengthen the case for a linkage between brain diabetes, neurodegenerative disease, and AD, a conjunction already in evidence from various pathophysiological studies [33, 34]. Chen et al.. found that metformin might increase the biogenesis of amyloid peptides in Alzheimer's disease, which could be relevant if there is a shared pathogenesis with depression, so our population-wide study supports this idea .
The definition of OAA usage in our models was a prescription for at least 3 months. Minimal doses for metformin were equivalent to about 500 mg per day and, for sulfonylureas, one standard tablet or capsule per day. These therapeutic regimens are feasible for most people with T2DM, and may largely remove the risk of AD posed by diabetes within 12 years of starting treatment. Owing to the limited number of subjects with T2DM on insulin therapy in this study, we did not treat them separately. Controlling for or exclusion of insulin usage in the models made no difference to our findings about OAAs and AD.
Issues in the diagnosis of affective disorder in diabetes
The Taiwanese NHI claim data rely on the medical service utilization of patients and on medical diagnoses. Underdiagnosis of AD and DM is likely if the reach of a healthcare system is limited, but NHI covers more than 98% of the population in Taiwan and, with the use of our matching protocol for accessibility (that is, region, level of service, and income), underdiagnosis should have been minimized. For social reasons, AD may be under-reported, but this is less likely with diabetes because of frequent and more intensive contact with the health system. Nevertheless, clinicians may have a tendency to simplify the diagnostic list when patients have multisystem diseases such as diabetes, in favor of the 'less pressing' diagnoses, which might therefore mean less recording of AD.
We have no information about the diagnosis of either diabetes or AD before 1996, when the NHI started, but it is unlikely that the presence of either of these would not have been recorded by a medical practitioner at some subsequent consultation. However, because the diagnosis of bipolar AD depends on at least two visits, such conditions may have been under-diagnosed as a consequence. Thus, there are several factors which may affect AD diagnosis rates, and thus their net effect on the differentials with regard to antecedent diabetes remains unclear.
A valuable insight into the validity of these diagnostic rates comes from the relevance of diagnostic methods, whether questionnaire or psychiatric diagnosis . Differences in estimates have been found to be dependent on year of publication of the report, but also to be probably increasing with time. Thus, there are questions about how contemporary our work, using 1996 datasets, might be regarded, because, as an NHI system requirement, we used ICD-9 for the diagnosis of AD rather then ICD-10 or DSM-IV. The ICD-9 codes used in the present study were 296.0 to 296.9 and its Taiwanese coding counterpart was A212 (before year 2000); nevertheless, these do cover ICD-10 codes F30 to F33 and F38 to F39, and thus, our study should be replicable.
We recognize that using the DSM-IV would have been a more useful approach to the diagnosis of depression and, more particularly, mood or AD, because it includes clinical symptoms and other axes that would have provided more analytical scope. However, it is reassuring that there can be some convergence of the risk assessment for diabetes on depression or AD, even using different methods . At the same time, our large and representative population-based study has allowed a more detailed understanding of the association of diabetes with different mood disorders than would otherwise have been possible.
In our study of AD, we considered both its unipolar and bipolar forms, but not dysthymic disorders, thus our findings are not applicable to all forms of depression. With regard to diabetes as a risk factor for AD, there seems to be no gender difference. Nor did we find any apparent difference between unipolar and bipolar AD, although sample size may have been limiting. Because incident AD was progressively excluded in the models, sample size diminished. Therefore, we may not have been able to discern more substantial and longer-term differences between unipolar and bipolar AD insofar as OAA effects are concerned.
T2DM and depression or AD seem to be at least bidirectional, if not mutually reinforcing [10, 36]. This process may begin at an early stage in the evolution of diabetes from pre-diabetic states and in the early stages of depressive illness. We did not study pre-diabetes subjects, but we did track individuals through what will have been that phase, from being diabetes-free in the 1996 to 2007 cohort. From other evidence [10, 36], we cannot fully exclude effects on depression-induced diabetes, which may have, in turn, increased AD incidence.
The study cohort is an administrative sample for which measures of diabetes severity are not available. Therefore, we do not have dose-response data for the studied OAAs and glycemic control. However, we can say that this particular OAA combination changes the risk for AD in T2DM, what the minimal dosages are likely to be for protection against AD, and how long it might take for these effects to be evident, as these informed the method we used to establish exposures. Matching for CCI, which partly represents diabetes-related complications and its severity, make it more likely that we observed a genuine OAA effect rather than one simply of glycemic control, although the latter is still possible.
Although ours is not an intervention study, we found at least one OAA combination which could enhance risk reduction for AD. This is unlikely to be a surrogate for other potential determinants of risk like body mass index. Moreover, were this combination of sulfonylurea and metformin to operate to reduce AD risk through a possible surrogate like body mass index, it might well be that these OAA act in directions opposite from each other, given their known opposite effects on body mass index. Unfortunately, these covariates are unavailable.