Rosiglitazone has a neutral effect on the risk of dementia in type 2 diabetes patients

This study investigated whether rosiglitazone might increase or reduce dementia risk. Taiwan’s National Health Insurance database was used to enroll a cohort of 1:1 matched-pairs of ever and never users of rosiglitazone based on propensity score from patients with new-onset type 2 diabetes during 1999-2006. The patients were alive on January 1, 2007 and were followed up for dementia until December 31, 2011. A total of 5,048 pairs of never users and ever users were identified. The incident case numbers were 127 and 121, respectively. The adjusted hazard ratio for ever versus never users was 0.895 (95% confidence interval: 0.696-1.151). The adjusted hazard ratios for the first (<12.1 months), second (12.1-25.1 months) and third (>25.1 months) tertiles of cumulative duration of rosiglitazone therapy were 0.756 (0.509-1.123), 0.964 (0.685-1.357) and 0.949 (0.671-1.341), respectively. When cumulative duration was treated as a continuous variable, the adjusted hazard ratio was 1.000 (0.992-1.008). Subgroup analyses conducted in ever users and never users of metformin and in patients diagnosed with diabetes during three different periods of time, i.e., 1999-2000, 2001-2003 and 2004-2006, all supported a neutral effect of rosiglitazone. In conclusion, rosiglitazone does not increase or redcue the risk of dementia.


INTRODUCTION
Dementia is a clinical presentation characterized by progressive deterioration of cognitive functions such as memory, thinking and reasoning and behavioral abilities for daily life and self-care. Dementia can result from a vascular etiology or a neurodegenerative disease known as Alzheimer's disease. Insulin resistance in the brain can be observed in patients with Alzheimer's disease [1][2][3]. Some experts coined the term "type 3 diabetes" in 2005 [1] to reflect the close link between diabetes mellitus and Alzheimer's disease, because they share potential common pathophysiological changes of impaired insulin expression and insulin resistance [1][2][3]. The increased risk of dementia in diabetes patients may be explained by some pathophysiological changes related to diabetes mellitus that lead to atherosclerosis and neurodegeneration, including insulin resistance, increased inflammation and oxidative stress, deposition of advanced glycation end-products and lipid dysregulation [4]. amyloid precursor protein by secretases [5]. Peroxisome proliferator-activator receptor gamma (PPARγ) is expressed in the brain [6] and it downregulates the expression of secretase and reduces Aβ deposition [7]. Knockdown of the PPARγ gene may also affect the expression of several other genes associated with Alzheimer's disease [8]. Therefore, drugs that improve insulin resistance or activate PPARγ in the brain can theoretically be beneficial in preventing Alzheimer's disease or dementia [2,9]. In our previous studies, two antidiabetic drugs, specifically metformin and pioglitazone (a PPARγ agonist), that improve insulin resistance and cross the blood-brain-barrier [10,11], reduced the risk of dementia in a dose-response pattern in patients with type 2 diabetes mellitus [12,13].

AGING
Rosiglitazone is another PPARγ agonist that had been commonly used as an oral antidiabetic drug before 2007 but is now rarely used because of a potential risk of cardiovascular disease [14]. Whether rosiglitazone might increase or reduce the risk of dementia has rarely been studied. Some in vitro, in vivo and animal studies suggested a neuroprotective effect of rosiglitazone [15][16][17][18][19][20][21][22][23]. An early pilot clinical trial conducted in 30 subjects (20 assigned to rosiglitazone and 10 assigned to placebo) with mild Alzheimer's disease or amnestic mild cognitive impairment suggested that rosiglitazone use for 6 months might have a potential for the treatment of cognitive decline [24]. However, this beneficial effect of rosiglitazone could not be confirmed by later clinical trials [25][26][27].
By using a cohort of 1:1 propensity score matched-pairs of rosiglitazone ever users and never users derived from the reimbursement database of Taiwan's National Health Insurance (NHI), the present study investigated whether rosiglitazone could increase or reduce dementia risk in patients with type 2 diabetes. In data analyses, ever users of pioglitazone were excluded and the potential confounding effect of metformin was addressed by subgroup analyses in ever users and never users of metformin.

Baseline characteristics
The characteristics of a selected cohort of 1:1 propensity score matched-pairs of never and ever users of rosiglitazone are shown in Table 1. The two groups were well matched and none of the calculated values of the standardized difference between ever and never users of rosiglitazone was found to be >10%.

Incidences of dementia and hazard ratios by rosiglitazone exposure
The incidence rates of dementia and the hazard ratios by rosiglitazone exposure are shown in Table 2. After a median follow-up of 4.8 years in either the ever users or never users of rosiglitazone, there were 127 incident cases of dementia in never users and 121 incident cases in ever users. The incidence rates of dementia were 616.79 and 537.54 per 100,000 person-years, respectively. The adjusted hazard ratio for ever versus never users of rosiglitazone was 0.895 (95% confidence interval: 0.696-1.151). Analyses with cumulative duration of rosiglitazone therapy categorized into tertiles or treated as a continuous variable all favored a neutral effect of rosiglitazone. Table 3 shows the results analyzed separately in ever users and never users of metformin. All analyses showed a non-significant effect of rosiglitazone in terms of overall hazard ratios, hazard ratios for the tertiles and hazard ratios for the cumulative duration of rosiglitazone therapy being treated as a continuous variable.   Table 4 shows the overall hazard ratios for ever versus never users of rosiglitazone estimated for patients who were enrolled during three different periods of time (i.e., 1999-2000, 2001-2003 and 2004-2006). A null association between rosiglitazone use and dementia risk was consistently observed.

DISCUSSION
The findings suggested that rosiglitazone use in patients with type 2 diabetes mellitus had a neutral effect on the risk of dementia (Tables 2-4).
It is interesting that the risk of dementia was decreased in association with the use of metformin [12] and pioglitazone [13] in our previous studies but not associated with rosiglitazone in the present study. Metformin [10] and pioglitazone [11] both cross the blood-brain barrier and may therefore reduce insulin resistance and inflammation in the brain. The reduced risk associated with pioglitazone may also result from the regulation of PPARγ on genes associated with Alzheimer's disease [8]. One of the possible explanations for a null effect associated with rosiglitazone is that rosiglitazone does not cross the blood-brain barrier [3] or only in a small proportion of 9-14% [3,28]. Different genetic backgrounds may also affect the responses to a drug treatment. For example, an earlier study suggested that patients with mild to moderate Alzheimer's disease and without ApoE4 allele might response to rosiglitazone treatment while those with such an allele would not response to the treatment [3]. Since we did not have genetic polymorphisms of ApoE for additional analyses, whether this might have explained the lack of a protective effect of rosiglitazone on dementia in our patients with type 2 diabetes mellitus remains to be answered. Following PPARγ activation, pioglitazone and rosiglitazone may affect the transcriptions of different sets of genes and such a discrepancy might have also explained their different effects on dementia risk and other clinical events. For example, pioglitazone has a beneficial effect on lipid profile [29,30] and cardiovascular events [31,32], but rosiglitazone may adversely affect lipid profile [30] and cardiovascular risk [14]. Therefore, it is not known whether different sets of genes with opposite effects on dementia may be regulated by pioglitazone and rosiglitazone, respectively. The different effects on dementia exerted by rosiglitazone and pioglitazone are worthy of more in-depth investigation for their potential clinical implications. A recent study showed that the brain concentration of pioglitazone is affected by Pglycoprotein, a drug efflux transporter [33]. While (+)pioglitazone is more resistant to this efflux transporter and accumulates in higher concentrations in the brain tissue, (-)-pioglitazone is less resistant to the efflux transporter and accumulates less in the brain. It is also worthy to further investigate whether the brain concentration of rosiglitazone may be affected by the stereospecific types of rosiglitazone compounds.
While patients with atrial fibrillation may have a higher risk of dementia, some recent studies suggested that the use of oral anticoagulants may provide a protective effect against dementia [34,35]. An estimated risk reduction of 21% (relative risk 0.79, 95% confidence interval: 0.67-0.93) was associated with the use of oral anticoagulants in a meta-analysis that includes 1 randomized controlled trial and 5 observational studies [34]. An observational study also suggested that the protective effect might be more remarkable for the use of non-vitamin K antagonist oral anticoagulants (adjusted hazard ratio 0.48, 95% confidence interval: 0.40-0.58) than for vitamin K antagonist (adjusted hazard ratio 0.62, 95% confidence interval: 0.60-0.64) [35]. Because non-vitamin K antagonist oral anticoagulants were not available during the study period up to 2011 in Taiwan, only warfarin, a vitamin K antagonist, could be included as a potential confounder in the present study (Table 1). In secondary analyses we found that atrial fibrillation was significantly associated with an increased risk of dementia (adjusted hazard ratio 2.270, 95% confidence interval: 1.332-3.869, P = 0.0026) but the use of warfarin had a neutral effect after adjustment for all covariates including atrial fibrillation (adjusted hazard ratio 0.949, 95% confidence interval: 0.488-1.846, P = 0.8777). Therefore, whether the use of oral anticoagulants may reduce the risk of dementia requires additional analyses. In additional subgroup analyses in patients with and without a diagnosis of atrial fibrillation and in patients who used and did not use warfarin, respectively, the risk of dementia remained neutral and insignificant for ever versus never users of rosiglitazone (data not shown). These secondary analyses still supported a neutral effect of rosiglitazone and did not change the conclusions of the study.
This study may have some clinical and research significance even though rosiglitazone is no longer widely used in clinical practice. First, although insulin resistance may increase the risk of dementia [1][2][3], this study strongly supports that not all drugs that improve insulin resistance and lower blood glucose may have a beneficial effect on the prevention of dementia in humans. Second, neuroprotective findings of rosiglitazone observed in in vitro, in vivo and animal studies should not be readily interpreted as potential protection against dementia in humans without consideration of its accessibility to the brain. This may have important implications in the future development of insulin sensitizers for the treatment of dementia. Third, taking into account the potentially higher risk of cardiovascular disease associated with rosiglitazone use [14] and the lack of a beneficial effect on dementia (Tables 2-4), the usefulness of rosiglitazone as an oral antidiabetic drug requires further justification.
The study has some merits that deserve mentioning. Because the NHI database covers >99% of Taiwan's population, the findings can be generalized to the whole AGING population. The potential bias resulted from selfreporting could be much reduced by using the medical records. Detection bias due to different socioeconomic status is less likely in our healthcare system because drug cost-sharing is low and can always be waived in patients with low-income, in veterans and when the patients receive prescription refills for chronic disease.
Study limitations include a lack of blood levels of glucose and insulin and a lack of indicators of insulin resistance (especially in the brain) and β-cell function for more in-depth analyses. Furthermore, the information of some confounders like nutritional status, dietary pattern, lifestyle, exercise, anthropometric factors, smoking, alcohol drinking, family history, and genetic parameters was not available.
In summary, unlike our previous study that showed a beneficial effect of pioglitazone on the risk of dementia in patients with type 2 diabetes mellitus, the present study finds a neutral effect of rosiglitazone. The discrepant effects between rosiglitazone and pioglitazone are worthy of more in-depth investigation.

NHI reimbursement database
Taiwan's NHI is a unique healthcare system implemented since March 1995. It covers >99.6% of the population. The Bureau of NHI has contracts with all in-hospitals and 93% of all medical settings throughout the nation. The database keeps all records of disease diagnoses, medication prescriptions and performed procedures and can be used for academic research after ethics review. The database was described in more details in our previous papers [36,37] and this retrospective cohort study used the longitudinal reimbursement database of the NHI for analyses with an approved number of 99274. Figure 1 shows the procedures followed in the creation of a cohort of 1:1 matched-pairs of ever and never users of rosiglitazone used in the study. Because a metaanalysis published in 2007 suggested a potential risk of cardiovascular disease associated with rosiglitazone use [14], the prescription of rosiglitazone had been withdrawn in many patients in Taiwan not based on clinical judgment but because of psychological impacts.

Selection of a propensity score-matched study cohort
To avoid the potential impact of some unknown factors following this event, the present study restricted the enrollment of ever users of rosiglitazone to patients who used the drug before 2007. At first, 444,750 patients with new-onset diabetes mellitus during 1999-2006 and having been prescribed antidiabetic drugs for 2 or more times were identified from the outpatient clinics.
Patients with a diagnosis of diabetes mellitus between 1996 and 1998 were not included to ensure a new diagnosis after 1999. The following ineligible patients were then excluded: 1) patients who died or had a diagnosis of dementia before January 1, 2007 (n=25,593), 2) patients who were initiated with rosiglitazone use after 2007 (n=27,569), 3) type 1 diabetes mellitus (n=2,352), 4) missing data (n=557); 5) ever users of pioglitazone (n=85,683), 6) rosiglitazone use for <180 days (n=2,681), 7) diagnosis of any cancer before entry or within 6 months of diabetes diagnosis (n=34,720, cancer patients might have a shortened lifespan and were excluded because they might have distorted follow-up time and dementia could be misdiagnosed from the clinical presentations of malignancy), 8) age <25 years (n=1,110), 9) age >75 years (n=33,952) and 10) follow-up duration <180 days (n=6,278). As a result, 5,048 ever users and 219,207 never users of rosiglitazone were identified as the unmatched original cohort. A cohort of 1:1 matchedpairs of 5,048 ever users and 5,048 never users (the matched cohort) was created by matching on propensity score (PS) based on the Greedy 81 digit match algorithm [38]. Logistic regression was used to create the PS with all characteristics listed in Table 1 being treated as independent variables. Potential confounders were divided into the following categories: demographic data, major comorbidities associated with diabetes mellitus, diabetes-related complications, some major risk factors of dementia, potential risk factors of cancer, antidiabetic drugs, medications commonly used in diabetes patients and use of warfarin. Demographic data included age, sex, diabetes duration, occupation and living region (classified as Taipei, Northern, Central, Southern, and Kao-Ping/Eastern). Occupation was classified as class I (civil servants, teachers, employees of governmental or private businesses, professionals and technicians), class II (people without a specific employer, self-employed people or seamen), class III (farmers or fishermen) and class IV (low-income families supported by social welfare, or veterans). Atrial fibrillation was defined by ICD-9-CM code of 427.31 and the ICD-9-CM codes for other potential confounders for major comorbidities AGING associated with diabetes mellitus (i.e., hypertension, dyslipidemia and obesity), diabetes-related complications (i.e., nephropathy, eye disease, stroke, ischemic heart disease and peripheral arterial disease), some major risk factors of dementia (i.e., head injury, Parkinson's disease and hypoglycemia) and potential risk factors of cancer (chronic obstructive pulmonary disease, tobacco abuse and alcohol-related diagnoses) can be found in a previously published paper [13]. Antidiabetic drugs included insulin, sulfonylureas, metformin, meglitinide, and acarbose. Commonly used medications in diabetes patients included angiotensin converting enzyme inhibitors/angiotensin receptor blockers, calcium channel blockers, statins, fibrates, and aspirin. The only oral anticoagulant included was warfarin because all non-vitamin K antagonist oral anticoagulants were not available in Taiwan during the study period.

Follow-up and calculation of incidence
Cumulative duration of rosiglitazone therapy was calculated in months from the database. Incidence density of dementia was calculated for the following subgroups of rosiglitazone use: never users, ever users and the tertiles of cumulative duration of therapy. The numerator was the case number of newly diagnosed dementia identified during follow-up and the denominator was the follow-up duration in personyears. Follow-up started on January 1, 2007 and ended on December 31, 2011, at the time of a new diagnosis of dementia, or on the date of death or the last reimbursement record, whichever occurred first.

Main analyses
Standardized difference was calculated for each potential confounder as recommended by Austin and Stuart, who also suggested the use of a cutoff value of >10% as an indicator of potential confounding from the variable [39].
All potential confounders were adjusted for in Cox proportional hazards regression models, which were used to estimate the hazard ratios and their 95% confidence intervals for ever users and for each tertile of cumulative duration of rosiglitazone therapy in comparison to a referent group of never users. Cumulative duration of rosiglitazone therapy was also treated as a continuous variable for estimating hazard ratios for every 1-month increment of use. P < 0.05 was considered statistically significant for the hazard ratios.

Subgroup analyses
To further examine whether the effect of rosiglitazone could be independent of metformin use, the above Cox proportional hazards regression models were also performed separately in subgroups of patients of ever and never users of metformin.
Before 1995, only metformin and sulfonylureas were available as oral antidiabetic drugs in Taiwan. In consideration that more antidiabetic drugs were available after 1995 and the guidelines for the use of antidiabetic drugs have evolved following the introduction of newer classes of drugs, the hazard ratios for dementia for ever versus never users of rosiglitazone were also estimated for patients whose diabetes was diagnosed during three different periods of time: 1999-2000, 2001-2003 and 2004-2006.

Statistical software
Analyses were conducted using SAS statistical software, version 9.4 (SAS Institute, Cary, NC).

AUTHOR CONTRIBUTIONS
Tseng CH researched data and wrote manuscript.