1868 Care Volume 38, October 2015

Nisha Nigil Haroon,1 Peter C. Austin,2 Risk of Dementia in Seniors With Baiju R. Shah,1,2 Jianbao Wu,2 Newly Diagnosed Diabetes: A Sudeep S. Gill,2,3 and Gillian L. Booth1,2 Population-Based Study Diabetes Care 2015;38:1868–1875 | DOI: 10.2337/dc15-0491

OBJECTIVE To study whether diabetes onset in late life is a for dementia.

RESEARCH DESIGN AND METHODS We conducted a population-based matched cohort study using provincial health data from Ontario, Canada. Seniors with (n = 225,045) and without newly diag- nosed diabetes (n = 668,070) between April 1995 and March 2007 were followed until March 2012 for a new diagnosis of dementia. Cox proportional hazards modeling was used to compare the risk of dementia between groups after adjust- EPIDEMIOLOGY/HEALTH SERVICES RESEARCH ing for baseline cardiovascular , chronic kidney disease (CKD), hyperten- sion, and other risk factors.

RESULTS Over this period, we observed 169,114 new cases of dementia. Individuals with diabetes had a modestly higher incidence of dementia (2.68 vs. 2.62 per 100 person-years) than those without diabetes. In the fully adjusted Cox model, the risk of dementia was 16% higher among our subgroup with diabetes (hazard ratio [HR] 1.16 [95% CI 1.15–1.18]). Adjusted HRs for dementia were 1.20 (95% CI 1.17– 1.22) and 1.14 (95% CI 1.12–1.16) among men and women, respectively. Among seniors with diabetes, the risk of dementia was greatest in those with prior cerebrovascular disease (HR 2.03; 95% CI 1.88–2.19), peripheral vascular disease (HR 1.47; 95% CI 1.19–1.82), and CKD (HR 1.44; 95% CI 1.38–1.51), and those with one or more hospital visits for (HR 1.73; 95% CI 1.62–1.84). 1Department of Medicine, University of Toronto, CONCLUSIONS Toronto, Canada 2Institute for Clinical Evaluative Sciences (ICES), In this population-based study, newly diagnosed diabetes was associated with a Toronto, Canada 16% increase in the risk of dementia among seniors. Preexisting vascular disease 3Department of Medicine, Queen’s University, and severe hypoglycemia were the greatest risk factors for dementia in seniors Kingston, Canada with diabetes. Corresponding author: Gillian L. Booth, boothg@ smh.ca. Received 7 March 2015 and accepted 2 July The incidence and prevalence of diabetes have increased substantially in recent 2015. years (1). Diabetes is a risk factor for blindness, kidney failure, coronary artery This article contains Supplementary Data online disease (CAD), stroke, and some cancers. However, medical advances over the at http://care.diabetesjournals.org/lookup/ suppl/doi:10.2337/dc15-0491/-/DC1. past two decades have led to reductions in cardiovascular risk, resulting in longer life expectancy (2). This has led to the emergence of geriatric complications in A slide set summarizing this article is available online. elderly people with diabetes, including Alzheimer disease and related dementias. ’ © 2015 by the American Diabetes Association. With the aging of the world s population, both diabetes and dementia are expected Readers may use this article as long as the work is to become global epidemics. Currently, ;44 million people have dementia, and this properly cited, the use is educational and not for number is expected to reach 135 million by the year 2050 (3). profit, and the work is not altered. care.diabetesjournals.org Haroon and Associates 1869

There is a growing body of evidence Our secondary aim was to identify fac- those with evidence of physician-diag- supporting an association between dia- tors that predict a higher risk of demen- nosed dementia at baseline, based on phy- betes and dementia (4–11). Since the tia in elderly patients with diabetes, as sician billing claims and hospitalizations. two entities appear to share antecedent an aid to individualized risk assessment fi cardiometabolic risk factors, dementia and prevention. Outcome De nitions Individuals were followed until 31 may be yet another vascular complica- March 2012 for a new diagnosis of de- tion of diabetes. Cardiovascular risk fac- RESEARCH DESIGN AND METHODS mentia, our primary outcome. This was tors in midlife such as Setting and Data Sources defined based on one or more hospital- (HTN), dyslipidemia, and have We conducted a population-based ization records or two outpatient physi- been implicated in the risk of dementia matched cohort study to assess the as- cian billing claims (within 6 months) in later life, although few studies have sociation between new-onset diabetes listing a relevant ICD-9 code (prior to 1 had sufficient follow-up to draw firm and dementia using provincial health April 2002: 290.x [except 290.8 or 290.9, conclusions (4,7,9). The demonstration administrative databases from Ontario, indicating senile psychosis], 331.0, of selective central nervous system ab- ’ Canada. Virtually all of Ontario s13.5 331.2, or 797) or ICD-10 code (1 April normalities in signaling have led million residents receive coverage for 2002 onwards: F00, F01, F02.3, F03, to the notion that Alzheimer disease ’ health services under the province s F05.1, F09, G30, G31.1, or R54) (17– represents “type 3” diabetes (12–14). publicly funded universal health care 20). This algorithm was validated in a In animal models, at system. The Registered Persons Data- population of ;2,000 participants en- the level of the leads to tau hyper- base contains information on demo- rolled in the Canadian Study of Health phosphorylation and b-amyloid accu- graphics, location of residence, and and Aging and found to have positive and mulation, both of which may promote vital status of all individuals registered negative predictive values of 73 and the development of Alzheimer disease ’ under the province s health plan (i.e., all 94%, respectively, using results from ex- (15). Moreover, insulin may have direct Ontario residents, including new immi- tensive cognitive testing as the gold effects on cortical function. Hyperglyce- $ grants 3 months since arrival). Records standard (18). mia, itself, may exert direct effects can be linked anonymously across data- through advanced glycation end prod- bases in order to track each individual’s Baseline Variables ucts, oxidative , inflammation, health care utilization and health out- We collected data on baseline covari- and microvascular alterations such as comes over time. This study was ap- ates including age, sex, income, HTN, capillary basement membrane thicken- proved by the Institute for Clinical and history of CAD, cerebrovascular dis- ing, similar to abnormalities seen in the Evaluative Sciences and the institu- ease (CVD), peripheral vascular disease kidney and retina (13,14). tional review board of Sunnybrook (PVD), and chronic kidney disease (CKD) Epidemiological evidence also supports Health Sciences Centre in Toronto. in the 36 months prior to the index date, a direct link between diabetes and demen- using previously defined diagnostic al- tia (5–11). A meta-analysis published in Study Population and Eligibility gorithms (21–24) (Supplementary Table 2011, based on 19 longitudinal studies, Our study cohort consisted of Ontario 1). HTN was defined using an algorithm found a nearly 1.5-fold higher incidence seniors with newly diagnosed (incident) based on hospitalization and physicians’ of Alzheimer disease and a 2.5 times higher diabetes and a matched comparison co- claims data, similar to that of diabetes, likelihood of vascular dementia in those hort without diabetes, who were 66– which has a positive and negative pre- with preexisting diabetes (5). However, el- 105 years of age between 1 April 1995 dictive value of 87 and 88%, respectively igible studies had heterogeneous results and 31 March 2007. Individuals who (21). Prior CAD included hospitalizations and differed substantially with respect to had a new record in the Ontario Diabe- for acute myocardial infarction, percu- their definitions of exposure and outcome tes Database (ODD) during this window taneous transluminal coronary angio- events. Moreover, many studies were served as our exposure group. The ODD plasty, or coronary artery bypass surgery. limited by relatively small sample sizes, is a validated administrative data-derived Our definition of CVD included hospitaliza- insufficient numbers of outcome events, registry that has a sensitivity and spec- tion for stroke, transient ischemic attack, and lack of data on diabetes duration or ificity of 86 and 97% (respectively) for or carotid endarterectomy. We identified vascular risk factors. identifying individuals with physician- patients undergoing lower extremity ar- It is still unknown whether the onset diagnosed diabetes (16). Only individu- terial bypass surgery or percutaneous of diabetes in late life can accelerate als with health care coverage for a transluminal angioplasty as having PVD. In- cognitive decline. The primary aim minimum of 3 years were included to en- formation on individual-level income was of our study was to examine whether sure that new entries into the ODD were not available in our data sources; hence incident diabetes is a risk factor for truly incident cases. We then identified the median household income level of dementia in elderly individuals. We three matches without diabetes with the each individual’s neighborhood of resi- hypothesized that exposure to even same age, sex, and region of residence for dence, derived from the nearest Canadian short-term in late life every person with newly diagnosed diabe- census year (1996, 2001, or 2006) was at- can trigger or accelerate cognitive de- tes in our cohort, using their entry date into tributed to them using their postal code. cline and therefore that incident diabe- the ODD as the baseline (index) date for Among individuals with diabetes, we tes is a risk factor for dementia after both groups. We excluded individuals living also captured information on ethnicity accounting for differences in cardiovascular in long-term care facilities (given the high using a validated algorithm that uses disease and other common risk factors. prevalence of dementia in this setting) and surnames to identify those of South 1870 Dementia and Newly Diagnosed Diabetes Diabetes Care Volume 38, October 2015

Asian or Chinese heritage, and on immi- using SAS software (version 9.4, SAS In- latter could not be ascertained reliably gration status using Canadian Citizenship stitute Inc., Cary, NC). from our databases prior to 1 April 2002, and Immigration data (25,26). We also Sensitivity Analyses we restricted this analysis to members captured information on the use of sta- Because unrecognized dementia may be of the cohort with diabetes entering the tins and antihypertensive drugs (ACE in- unmasked by the complexity of diabetes ODD on or after this date (n = 110,816). hibitors, angiotensin receptor blockers, management (i.e., the need for lifestyle We also looked at the effect of various diuretics, and calcium channel blockers changes, regular administra- medication classes. First, we examined [CCBs]) and antidiabetic treatment (met- tion, and monitoring), we the cumulative duration (number of formin, sulphonylureas, or insulin) from conducted a sensitivity analysis to ex- days received) of each class of medica- fi the Ontario Drug Bene t Database, amine whether detection bias could tion between diagnosis and the end of which contains records for all prescrip- explainanelevatedriskofdementia follow-up, based on the number of pre- fi tions lled by Ontario seniors. Last, we among patients with incident diabetes. scriptions dispensed and the number of included information about hospital ad- To do so, we allowed the effect of di- days supplied per prescription (adding missions and emergency department abetes status to vary between earlier 50% to the number of days dispensed (ED) visits for hypoglycemia. (,180 days) and later (.180 days) pe- for the last prescription observed). For riods of follow-up among the entire insulin use, we counted the total num- Statistical Analysis ber of prescriptions from the index date Primary Analysis sample. Similar analyses were con- until date of censoring. Second, we We conducted a time-to-event analysis ducted using earlier time thresholds of 30, 60, and 90 days. treated medication exposure as a time- using Cox proportional hazards model- varying covariate, allowing the use of ing to study the relationship between We further examined whether the proportional hazards assumption held each class of drug to change over time newly diagnosed diabetes and incident and denoting whether each individual dementia, stratifying by matched clus- by running an additional model in the overall sample that included an interac- was currently exposed (yes/no) on a ters of exposed and unexposed individ- given day of follow-up. Because of the uals. Models were adjusted for baseline tion term for diabetes and time. Again, we hypothesized that if detection bias size of our data, this analysis could not income and each of the comorbidities be performed on the entire sample but defined above, including HTN, CKD, were a major driver of the relationship between diabetes and dementia, then was conducted on a 5% random subset and vascular disease of various etiolo- (11,252). gies. We used cause-specific hazard the effect of incident diabetes would functions to account for the competing lessen over time. risk of , rather than Fine-Gray Secondary Analysis RESULTS regression models (27). Cumulative in- Cox proportional hazards modeling was Patient Characteristics cidence functions (CIFs), which also ac- used to identify risk factors for dementia We identified 225,045 seniors with count for death as a competing risk, among individuals with diabetes. Cova- newly diagnosed diabetes who were were used to estimate the probability riates included age, sex, income, ethnic- matched to 668,070 individuals without of the occurrence of dementia. As a ity, recent immigration (,10 years since diabetes (from a pool of 1,016,256 po- comparison, CIF curves were also gener- landing), baseline CAD, CVD, PVD, HTN, tentially eligible matches). Their base- ated for the incidence of death from any or CKD, and hospitalizations or ED visits line characteristics are shown in Table cause, and dementia or death from any for hypoglycemia during follow-up (as a 1. The median age of our cohort was cause. All analyses were performed time-varying covariate). Because the 73 years (interquartile range 69–78).

Table 1—Baseline characteristics of study cohort with newly diagnosed diabetes and matched comparison group without diabetes Characteristic Newly diagnosed diabetes, n = 225,045 No diabetes, n = 668,070 Total, n = 893,115 Age (years), median (IQR) 73 (69–78) 73 (69–78) 73 (69–78) Women 114,203 (50.8) 341,557 (51.1) 455,760 (51.0) Income quintile (Q) Q1 (lowest) 49,527 (22.0) 128,826 (19.3) 178,353 (20.0) Q2 50,732 (22.5) 141,088 (21.1) 191,820 (21.5) Q3 44,811 (19.9) 132,641 (19.9) 177,452 (19.9) Q4 40,658 (18.1) 126,697 (19.0) 167,355 (18.7) Q5 (highest) 38,723 (17.2) 137,243 (20.5) 175,966 (19.7) CAD 13,181 (5.9) 18,410 (2.8) 31,591 (3.5) CVD 6,652 (3.0) 9,147 (1.4) 15,799 (1.8) PVD 1,189 (0.5) 1,981 (0.3) 3,170 (0.4) HTN 158,151 (70.3) 371,313 (55.6) 529,464 (59.3) CKD 25,863 (11.5) 47,516 (7.1) 73,379 (8.2) Data are n (%) unless indicated otherwise. IQR, interquartile range. care.diabetesjournals.org Haroon and Associates 1871

Individuals with diabetes had a higher hazards assumption, we did detect a sig- 0.59–0.78 and 0.82; 95% CI 0.74–0.90, prevalence of HTN, vascular disease, nificant interaction between diabetes respectively, compared with other eth- and CKD at baseline relative to those status and time. The HR associated nic groups). Paradoxically, HTN also ap- in the comparison group. with diabetes status was 1.19 (95% CI peared to be mildly protective against 1.16–1.22), and this effect increased by the development of dementia (HR 0.95; Primary Analysis about 1% per year (HR for diabetes 3 95% CI 0.91–0.99). The average duration of follow-up was year interaction: 1.009; 95% CI: 1.004– We examined the impact of various 7.2 years, during which 169,114 individ- 1.014). Thus, after 10 years of duration, classes of on dementia uals were diagnosed with dementia. In- diabetes was associated with a nearly risk (glucose-, pressure–,and dividuals with diabetes experienced a 30% higher incidence of dementia. At cholesterol-lowering agents) in individ- modestly higher incidence of dementia any time point, the likelihood of death uals with diabetes. When we modeled compared with those without diabetes greatly outweighed the likelihood of de- the impact of cumulative medication (2.68 vs. 2.62 per 100 person-years) mentia (Fig. 1). In this older population, use, all drug classes were associated (Table 2). Based on the Cox model, diabe- the likelihood of dementia or death from with a significantly lower risk of demen- tes was associated with a 19% increased any cause approached 50% within tia, with adjusted HRs ranging from 0.90 risk of dementia (unadjusted hazard ratio 10 years of diabetes diagnosis. to 0.95 per 180 days of use (P , 0.0001 – [HR] 1.19; 95% CI 1.18 1.21). The HR re- for each). In models that treated medi- mained similar in magnitude after adjust- cation use as time-varying exposures, Predictors of Dementia Among Those ing for baseline covariates, including HTN, the risk of dementia was significantly With Diabetes CAD, CVD, PVD, and CKD (adjusted HR lower only with current statin (HR – Factors associated with an increased 1.16; 95% CI 1.15 1.18). Both men (HR 0.78; 95% CI 0.71–0.86) and CCB (HR – risk of dementia among individuals 1.20; 95% CI 1.17 1.22]) and women 0.80; 95% CI 0.73–0.89) use, whereas – with diabetes are shown in Fig. 2. De- (HR 1.14; 95% CI 1.12 1.16) with diabetes insulin use was associated with a sub- mentia risk increased by 12% per year experienced elevated risks of dementia stantially higher risk (HR 1.74; 95% CI with advancing age. Individuals living in compared with their counterparts with- 1.37–2.21). In these models, HTN was the lowest income areas were 16% more out diabetes. no longer associated with dementia likely compared with those in the (HR 0.99; 95% CI 0.89–1.09). Sensitivity Analysis wealthiest area (adjusted HR 1.17; 95% TheCIFcurve(Fig.1)showedavery CI 1.12–1.24). Prior vascular disease and early separation in dementia risk be- CKD were associated with significantly CONCLUSIONS tween those with and without diabetes, increased risks of dementia; however, Our study found a significantly higher occurring in the first 6 months. The risk the strongest predictor was previous risk of dementia among seniors with associated with diabetes remained un- CVD, associated with a doubling in risk newly diagnosed diabetes. This is of se- changed from very early (,180 days) to (adjusted HR 2.03; 95% CI 1.88–2.19). rious concern given the aging popula- later time periods (.180 days) (P . 0.5 Hospitalization and ED visits for hypogly- tion, increasing prevalence of diabetes, for differences in HR between the very cemia were also significant predictors of andthelimitedeffectivetreatment early and later periods), and the results dementia (adjusted HR 1.73; 95% CI currently available for dementia. This in- weresimilarwhenevenearliertime 1.62–1.84 for one or more vs. zero epi- creased risk was independent of pre- thresholds were used (30, 60, or sodes). Factors associated with a lower existing CAD or CVD, HTN, and CKD, 90 days). However, only a small per- risk of dementia included recent immigra- conditions thought to mediate the asso- centage of outcome events occurred tion (adjusted HR 0.63; 95% CI 0.55–0.72) ciation between diabetes and dementia. in the first 6 months of follow-up andbothSouthAsianandChinese Our results are supported by prior meta- (2.8%). When testing the proportional ethnicity (adjusted HRs 0.68; 95% CI analyses of largely smaller observational

Table 2—Risk of dementia among seniors with newly diagnosed diabetes and matched comparison group without diabetes† Incidence rate of Number of new Total person-years dementia per Unadjusted HR Fully adjusted HR‡ Group n dementia cases of follow-up 100 person-years* (95% CI) (95% CI) Men and women No diabetes 668,070 126,085 4,811,236 2.62 1.00 1.00 Diabetes 225,045 43,029 1,602,844 2.68 1.19 (1.18–1.21) 1.16 (1.15–1.18) Men No diabetes 326,513 53,819 2,283,649 2.36 1.00 1.00 Diabetes 110,842 18,865 769,158 2.45 1.23 (1.20–1.25) 1.20 (1.17–1.22) Women No diabetes 341,557 72,266 2,527,587 2.86 1.00 1.00 Diabetes 114,203 24,164 833,685 2.90 1.17 (1.15–1.19) 1.14 (1.12–1.16) †Groups with and without diabetes matched by age, sex, and region of residence. *Event rates are number of new dementia cases per 100 person- years. ‡Adjusted for CAD, CVD, or PVD hospitalizations and procedures #36 months prior to baseline entry, as well as HTN, CKD, and neighborhood socioeconomic status at baseline entry. 1872 Dementia and Newly Diagnosed Diabetes Diabetes Care Volume 38, October 2015

diabetes in cohort studies, and therefore of dementia, underscoring the impor- higher observed effect sizes. One other tance of stroke prevention. There is population-based study conducted using some evidence supporting a role for the Taiwan National Health Database vascular dysfunction and cerebral hypo- found a 76% higher HR for Alzheimer perfusion in the of Alz- disease in those with incident diabetes; heimer disease (30), and for stroke however, participants were from a broad prevention as a means of preventing fur- age-group (11). To the best of our knowl- ther cognitive decline (31). In addition, edge, ours is the first large, population- our findings demonstrated a harmful ef- wide study to report an association fect of severe hypoglycemia in seniors, between diabetes and dementia among a factor that may underlie the link we elderly people with newly diagnosed di- observed between insulin use and abetes, and the first of its kind in North dementia in this population. In the America. The large sample size afforded ACCORD-MIND trial, tight glycemic us sufficient statistical power to study control was associated with a lower like- this association and to identify subsets lihood of cerebral atrophy but had no of the elderly population with diabetes effect on (32). Some have the- that are most vulnerable to developing orized that potential benefits may have dementia. We were also able to test for been partly negated by a greater expo- the role of detection bias as a potential sure to severe hypoglycemia in the in- mediator in the association between di- tensively treated group. On the contrary, abetes and dementia and accounted for patients with cognitive dysfunction may the large competing risk of death in this be more susceptible to hypoglycemia, population. This adds further credence leading to a bidirectional relationship to our findings and those of others between severe hypoglycemia and de- demonstrating a clear link between mentia (33,34). Another important ob- these two entities. servation was that the association Figure 1—Cumulative incidence of demen- Our results are congruent with exist- between diabetes and dementia risk in- tia, death from any cause, and dementia or ing data that vascular risk factors and creased over time, possibly because of death from any cause among the study pop- CKD aggravate the risk of dementia cumulative exposure to microvascular in- ulation with and without diabetes. A: Cumu- (13,28,29). In our study, CVD was asso- sults, hypoglycemia, and other diabetes- lative incidence of dementia. B: Cumulative incidence of death from any cause. C: Cumu- ciated with a twofold increase in the risk related effects. However, mortality rates lative incidence of dementia or death from any cause. Group with diabetes, dotted line; group without diabetes, solid line.

and cohort studies that reported a 1.5- and 2.5-fold higher risk of Alzheimer disease and vascular dementia (res- pectively) in patients with established diabetes (5–8). There was considerable heterogeneity across studies with re- spect to exposure and outcome ascer- tainment, and many failed to account for preexisting vascular disease. Our risk estimates may have been lower than in these latter studies because our population was earlier in their course of diabetes. However, it is also possible that a survival effect led to lower event rates in our diabetes population than expected, since we limited our analysis to seniors. Individuals who develop dia- betes at an older age may be relatively healthier in comparison with their coun- Figure 2—Risk factors for dementia in seniors with newly diagnosed diabetes (n = 110,816). terparts without diabetes than those Cohort was restricted to individuals with diabetes entering the ODD 1 April 2002 through 31 March 2007. †Lowest vs. highest income quintile based on median household income level of who develop diabetes in midlife. Con- neighborhood of residence; ‡one or more hospitalizations or ED visits during follow-up, treated versely, selection bias may have resulted as time-varying covariate. Model includes all of the above covariates, as well as sex, ethnicity, in healthier control subjects without and immigration status. care.diabetesjournals.org Haroon and Associates 1873

remain high among seniors with diabe- risk we observed in association with any individuals with both conditions, since tes; thus, those who survive longer will drug class, analyses using time-varying incentive fees, linked to monetary com- also have more opportunity to develop drug exposures were designed to mini- pensation, are provided for diabetes dementia, whereas those who succumb mize such effects. Last, we found a par- management. Therefore, we may have earlier will not. Thus, if recent gains in adoxically lower risk of dementia in undercaptured cases of dementia in the life expectancy continue to occur among patients with established HTN based on population with diabetes. Last, we have patients with diabetes, the burden of diagnostic codes from hospitalization re- not accounted for confounders such as dementia in this population may para- cords and physicians’ service claims. abdominal obesity or other correlates of doxically rise despite overall improve- However, individuals who were captured insulin resistance, HbA1c levels, vascular ments in care. as having HTN may have had greater ex- risk factors (e.g., blood pressure, serum We found that certain factors ap- posure to protective therapies (CCBs and , or smoking), or genetic factors peared to be protective with respect statins), since this effect became neutral such as APOE4 mutations, since admin- to the risk of dementia. Fewer South once medication use was accounted for istrative databases lack information Asians developed dementia despite be- in our analysis. Physicians who recognize about these. ing thought to have a higher risk of and submit claims for HTN management Our results suggest that newly diag- cardiovascular complications. Varied may also treat blood pressure to lower nosed diabetes is an important risk fac- cultural and genetic factors may contrib- targets; however, our databases lacked tor for the diagnosis of dementia in ute to the differential susceptibility to information on actual blood pressure older adults. Severe hypoglycemia, dementia across racial groups; however, levels and therefore we were unable to stroke, and other vascular the “healthy immigrant effect” caused separate out the effects of blood pres- may represent modifiable risk factors by health screening during immigration, sure itself from its treatment. for dementia. Clinical trials suggest immigrant self-selection, and the return Our study has many strengths, that current therapeutic measures to of recent immigrants with poor health including a large sample size and well- control hyperglycemia, blood pressure, to their native countries may also play a validated diagnostic algorithms to and dyslipidemia are unable to prevent role (26,35). In contrast, low income define diabetes, dementia, and other the onset of dementia in older adults was a risk factor for dementia, as it is relevant covariates. Furthermore, our with diabetes. More research is needed for other diabetes-related outcomes population-based design reduces selec- to identify strategies that can effectively (36,37). Impaired health literacy, poorer tion bias and makes the results more ward off dementia in this growing seg- self-management, and adverse health generalizable. However, there are cer- ment of the population. behaviors, such as smoking, have been tain limitations that merit discussion, linked to low income and could explain including ambiguity about the time of this association. From epidemiologic onset of dementia. Although we had ex- Funding. This work was funded by the Canadian data, smoking appears to be a risk factor cluded previous dementia as compre- Institutes of HealthResearch.P.C.A.andG.L.B.are – supported by Career and Mid-Career Investigator for both diabetes and dementia (38 40); hensively as we could, some patients Awards (respectively) from the Heart and Stroke however, levels of current smoking are might have had some degree of cogni- FoundationofOntario.B.R.S.andS.S.G.arefunded comparable between the populations tive decline at baseline. Thus we are by New Investigator Awards from the Canadian with and without diabetes and fall sig- unable to comment on whether newly Institutes of Health Research. G.L.B. holds a Mid- nificantly over age 65 years (23,41). diagnosed diabetes led to the develop- CareerInvestigatorAwardfromtheDepartmentof MedicineattheUniversityofToronto.TheOntario Whether prior smoking predicts demen- ment of dementia or accelerated its Ministry of Health and Long-Term Care provides tia (or significantly differed across groups) progression. Detection bias may have financial support to the Institute for Clinical Eval- is unclear. contributed to dementia in our popula- uative Sciences (ICES) and provides health-related In our study, many commonly used tion with diabetes; however, only a very and other databases to ICES under a data sharing vascular and antidiabetic medications small proportion of events occurred agreement. None of the funding or supporting agencies showed no association with the risk of within a short time window of diabetes participated in the following aspects of this dementia, with the exception of statins diagnosis. Furthermore, the effect of di- research: the design and conduct of the study; and CCBs. Although such treatments abetes grew over time rather than less- the collection, management, analysis, and in- have been postulated to be protective ened. However, we cannot entirely rule terpretation of the data; or the preparation, against dementia, numerous trials have out an element of reverse causation review, and approval of the manuscript. The fi opinions, results, and conclusions reported in failed to identify any bene cial role whereby individuals with early cognitive this article are those of the authors and are of glucose-, blood pressure–, or - decline experienced a change in diet or independent from the funding sources. No en- lowering agents on cognitive decline activity pattern leading to a greater like- dorsement by ICES or the Ontario Ministry of (31,32,42–45), as suggested by previous lihood of gaining weight and developing Health and Long-Term Care is intended or should observational data (11,46,47). The latter diabetes. Further, we could not examine be inferred. Duality of Interest. N.N.H. is supported by an may be vulnerable to numerous biases, the impact of diabetes on different sub- unrestricted grant from Amgen. No other po- including the healthy user effect, since types of dementia separately, as the da- tential conflicts of interest relevant to this article patients with cognitive decline may be tabases we used do not adequately were reported. less willing to take multiple therapies or discriminate between them. Primary Author Contributions. N.N.H. and G.L.B. con- ceptualized the study, designed the analyses, less likely to be prescribed them. Al- care physicians in Ontario may be and wrote the manuscript. P.C.A., B.R.S., and though this phenomenon may explain more likely to list a diagnosis of diabetes S.S.G. contributed to the design of the analyses, the nonspecific reduction in dementia than dementia on a billing claim for contributed to the discussion, and edited the 1874 Dementia and Newly Diagnosed Diabetes Diabetes Care Volume 38, October 2015

manuscript. J.W. researched data, contributed to resistance, IRS-1 dysregulation, and cognitive disease: is Alzheimer’s a vascular disorder? Am J the discussion, and edited the manuscript. G.L.B. decline. J Clin Invest 2012;122:1316–1338 Cardiovasc Dis 2013;3:197–226 is the guarantor of this work and, as such, had full 16. Hux JE, Ivis F, Flintoft V, Bica A. Diabetes in 31. Tzourio C, Anderson C, Chapman N, et al.; access to all the data in the study and takes Ontario: determination of prevalence and inci- PROGRESS Collaborative Group. Effects of blood responsibility for the integrity of the data and dence using a validated administrative data al- pressure lowering with perindopril and indapa- the accuracy of the data analysis. gorithm. Diabetes Care 2002;25:512–516 mide therapy on dementia and cognitive decline 17. St Germaine-Smith C, Metcalfe A, Pringsheim in patients with cerebrovascular disease. Arch References T, et al. Recommendations for optimal ICD codes Intern Med 2003;163:1069–1075 1. Lipscombe LL, Hux JE. Trends in diabetes to study neurologic conditions: a systematic re- 32. Launer LJ, Miller ME, Williamson JD, et al.; – prevalence, incidence, and mortality in Ontario, view. 2012;79:1049 1055 ACCORD MIND investigators. Effects of inten- Canada 1995-2005: a population-based study. 18. Gill SS, Camacho X, Poss JW. Community- sive glucose lowering on brain structure and Lancet 2007;369:750–756 dwelling older adults with dementia: tracking function in people with 2. Lind M, Garcia-Rodriguez LA, Booth GL, et al. encounters with the health system. In Health (ACCORD MIND): a randomised open-label sub- – Mortality trends in patients with and without System Use by Frail Ontario Seniors: An In-Depth study. Lancet Neurol 2011;10:969 977 diabetes in Ontario, Canada and the UK from Examination of Four Vulnerable Cohorts. 33. Whitmer RA, Karter AJ, Yaffe K, 1996 to 2009: a population-based study. Diabe- Bronskill SE, Camacho X, Gruneir A, Ho MM, Quesenberry CP Jr, Selby JV. Hypoglycemic epi- tologia 2013;56:2601–2608 Eds. Toronto, ON, Institute for Clinical Evalua- sodes and risk of dementia in older patients 3. Policy Brief for Heads of Government: tive Sciences, 2011 with type 2 diabetes mellitus. JAMA 2009;301: The Global Impact of Dementia 2013–2050. 19. Al-Omran M, Mamdani MM, Lindsay TF, 1565–1572 Alzheimer’s Disease International (ADI), London, Melo M, Juurlink DN, Verma S; Systematic As- 34. Yaffe K, Falvey CM, Hamilton N, et al.; December 2013. Available from http://www sessment of Vascular Risk Investigators. Subop- Health ABC Study. Association between hypo- .alz.co.uk/research/GlobalImpactDementia2013 timal use of statin therapy in elderly patients glycemia and dementia in a biracial cohort of .pdf. Accessed 31 August 2014 with atherosclerosis: a population-based study. older adults with diabetes mellitus. JAMA Intern 4. Daviglus ML, Plassman BL, Pirzada A, et al. J Vasc Surg 2008;48:607–612 Med 2013;173:1300–1306 Risk factors and preventive interventions for 20. Seitz DP, Gill SS, Bell CM, et al. Postopera- 35. Llibre Rodriguez JJ, Ferri CP, Acosta D, et al.; Alzheimer disease: state of the science. Arch tive medical complications associated with an- 10/66 Dementia Research Group. Prevalence of Neurol 2011;68:1185–1190 esthesia in older adults with dementia. J Am dementia in Latin America, India, and China: 5. Cheng G, Huang C, Deng H, Wang H. Diabetes Geriatr Soc 2014;62:2102–2109 a population-based cross-sectional survey. Lan- as a risk factor for dementia and mild cognitive 21. Tu K, Campbell NR, Chen ZL, Cauch-Dudek cet 2008;372:464–474 impairment: a meta-analysis of longitudinal KJ, McAlister FA. Accuracy of administrative da- 36. Booth GL, Bishara P, Lipscombe LL, et al. studies. Intern Med J 2012;42:484–491 tabases in identifying patients with hyperten- Universal drug coverage and socioeconomic dis- 6. Lu FP, Lin KP, Kuo HK. Diabetes and the risk of sion. Open Med 2007;1:e18–e26 parities in major diabetes outcomes. Diabetes multi-system aging phenotypes: a systematic 22. Booth GL, Kapral MK, Fung K, Tu JV. Relation Care 2012;35:2257–2264 review and meta-analysis. PLoS One 2009;4: between age and cardiovascular disease in men 37. Booth GL, Creatore MI, Moineddin R, et al. e4144 and women with diabetes compared with non- Unwalkable neighborhoods, poverty, and the 7. Biessels GJ, Staekenborg S, Brunner E, diabetic people: a population-based retrospec- risk of diabetes among recent immigrants to Brayne C, Scheltens P. Risk of dementia in di- tive cohort study. Lancet 2006;368:29–36 Canada compared with long-term residents. Di- abetes mellitus: a systematic review. Lancet 23.BoothGL,LipscombeLL,DeganiN,etal. abetes Care 2013;36:302–308 Neurol 2006;5:64–74 Diabetes. In Project for an Ontario Women’s 38. Rosella LC, Manuel DG, Burchill C, Stukel 8. Vagelatos NT, Eslick GD. Type 2 diabetes as a Health Evidence-Based Report. Vol. 2. Bierman TA; PHIAT-DM team. A population-based risk risk factor for Alzheimer’s disease: the con- AS, editor. Toronto, St. Michael’sHospitaland algorithm for the development of diabetes: founders, interactions, and neuropathology as- the Institute for Clinical Evaluative Sciences, development and validation of the Diabetes sociated with this relationship. Epidemiol Rev 2010 Population Risk Tool (DPoRT). J Epidemiol Com- 2013;35:152–160 24. Ronksley PE, Tonelli M, Quan H, et al.; Al- munity Health 2011;65:613–620 9. Kimm H, Lee PH, Shin YJ, et al. Mid-life and berta Kidney Disease Network. Validating a case 39. Baumgart M, Snyder HM, Carrillo MC, Fazio late-life vascular risk factors and dementia in definition for chronic kidney disease using ad- S, Kim H, Johns H. Summary of the evidence Korean men and women. Arch Gerontol Geriatr ministrative data. Nephrol Dial Transplant 2012; on modifiable risk factors for cognitive decline 2011;52:e117–e122 27:1826–1831 and dementia: a population-based perspective. 10. Wang KC, Woung LC, Tsai MT, Liu CC, Su 25. Shah BR, Victor JC, Chiu M, et al. Cardiovas- Alzheimers Dement 2015;11:718–726 YH, Li CY. Risk of Alzheimer’sdiseaseinre- cular complications and mortality after diabetes 40. Di Marco LY, Marzo A, Munoz-Ruiz~ M, et al. lation to diabetes: a population-based co- diagnosis for South Asian and Chinese patients: Modifiable lifestyle factors in dementia: a sys- hort study. Neuroepidemiology 2012;38: a population-based cohort study. Diabetes Care tematic review of longitudinal observational 237–244 2013;36:2670–2676 cohort studies. J Alzheimers Dis 2014;42:119– 11. Huang CC, Chung CM, Leu HB, et al. Diabe- 26. Okrainec K, Booth GL, Hollands S, Bell CM. 135 tes mellitus and the risk of Alzheimer’s disease: Impact of barriers on complications 41. Janz T. Statistics Canada. Health at a glance: a nationwide population-based study. PLoS One and mortality among immigrants with diabetes: current smoking trends. Available from http:// 2014;9:e87095 a population-based cohort study. Diabetes Care www.statcan.gc.ca/pub/82-624-x/2012001/ 12. Pilcher H. Alzheimer’sdiseasecouldbe 2015;38:189–196 article/11676-eng.htm. Accessed 2 June 2015 “type 3 diabetes”. Lancet Neurol 2006;5:388– 27. Lau B, Cole SR, Gange SJ. Competing risk 42. McGuinness B, Craig D, Bullock R, 389 regression models for epidemiologic data. Am Passmore P. Statins for the prevention of de- 13. Strachan MW, Reynolds RM, Marioni RE, J Epidemiol 2009;170:244–256 mentia. Cochrane Database Syst Rev 2009;15: Price JF. Cognitive function, dementia and 28. Exalto LG, Biessels GJ, Karter AJ, et al. Risk CD003160 type 2 diabetes mellitus in the elderly. Nat Rev score for prediction of 10 year dementia risk in 43. Anderson C, Teo K, Gao P, et al.; ONTARGET Endocrinol 2011;7:108–114 individuals with type 2 diabetes: a cohort study. and TRANSCEND Investigators. Renin-angiotensin 14. Schrijvers EM, Witteman JC, Sijbrands EJ, Lancet Diabetes Endocrinol 2013;1:183–190 system blockade and cognitive function in pa- Hofman A, Koudstaal PJ, Breteler MM. Insulin 29. Bruce DG, Davis WA, Starkstein SE, Davis tients at high risk of cardiovascular disease: and the risk of Alzheimer disease: TM. Mid-life predictors of cognitive impairment analysis of data from the ONTARGET and the Rotterdam Study. Neurology 2010;75:1982– and dementia in type 2 diabetes mellitus: the TRANSCEND studies. Lancet Neurol 2011;10: 1987 Fremantle Diabetes Study. J Alzheimers Dis 43–53 15. Talbot K, Wang HY, Kazi H, et al. Demon- 2014;42(Suppl. 3):S63–S70 44. Williamson JD, Launer LJ, Bryan RN, et al.; strated brain insulin resistance in Alzheimer’s 30. Kelleher RJ, Soiza RL. Evidence of endothelial Action to Control Cardiovascular Risk in Diabe- disease patients is associated with IGF-1 dysfunction in the development of Alzheimer’s tes in Diabetes Investigators. Cognitive care.diabetesjournals.org Haroon and Associates 1875

function and brain structure in persons with type in the Very Elderly Trial cognitive function as- and metformin. J Alzheimers Dis 2011;24:485– 2 diabetes mellitus after intensive lowering of sessment (HYVET-COG): a double-blind, pla- 493 blood pressure and lipid levels: a randomized clin- cebo controlled trial. Lancet Neurol 2008;7: 47. Peters R, Booth A, Peters J. A systematic ical trial. JAMA Intern Med 2014;174:324–333 683–689 review of calcium channel blocker use and 45. Peters R, Beckett N, Forette F, et al.; 46. Hsu CC, Wahlqvist ML, Lee MS, Tsai HN. In- cognitive decline/dementia in the elderly. HYVET investigators. Incident dementia and cidence of dementia is increased in type 2 di- J Hypertens 2014;32:1945–1957; discussion blood pressure lowering in the Hypertension abetes and reduced by the use of sulfonylureas 1957–1958