Advertisement

The Impact of Dementia on Diabetes Control: An Evaluation of HbA1c Trajectories and Care Outcomes in Linked Primary and Specialist Care Data

Open AccessPublished:May 31, 2022DOI:https://doi.org/10.1016/j.jamda.2022.04.045

      Abstract

      Objectives

      Diabetes self-care may become increasingly challenging as cognition declines. We sought to characterize glycated hemoglobin A1c (HbA1c) trajectories, markers of diabetes-related management, health care utilization, and mortality in people with preexisting type 2 diabetes (T2D) with and without dementia and based on the extent of cognitive impairment at the time of dementia diagnosis.

      Design

      Retrospective matched cohort study.

      Setting and Participants

      Using a linkage between a primary care (Lambeth DataNet) and a secondary mental healthcare database, up to 5 individuals aged ≥65 y with preexisting T2D without dementia were matched to each individual with dementia based on age, sex, and general practice.

      Methods

      Comparisons were made for HbA1c trajectories (linear mixed effects models), markers of diabetes-related management and severity at dementia diagnosis (logistic regression), mortality (Cox regression), and health care utilization (multilevel mixed effects binomial regression).

      Results

      In 725 incident dementia and 3154 matched comparators, HbA1c trajectories differed by dementia status; HbA1c increased over time for mild dementia and non-dementia, but the increase was greater in the mild dementia group; for those with moderate-severe dementia, HbA1c decreased over time. Despite individuals with dementia having increased health care utilization around the time of dementia diagnosis, they were less likely to have had routine diabetes-related management. Patients with dementia had a higher prevalence of macrovascular complications and diabetes foot morbidity at dementia diagnosis and a higher mortality risk than those without dementia; these relationships were most marked in those with moderate-severe dementia.

      Conclusions and Implications

      Our study has highlighted important differences in the monitoring, management, and control of diabetes in people with dementia. The effects of frailty and the extent of cognitive impairment on the ability to self-manage diabetes and on glycemic control may need to be considered in treatment guidelines and by primary care.

      Keywords

      Dementia and type 2 diabetes (T2D) are common, chronic conditions that frequently coexist in older populations.
      • Bunn F.
      • Burn A.M.
      • Goodman C.
      • et al.
      Comorbidity and dementia: A scoping review of the literature.
      The number of people with both conditions is expected to increase because of population ageing and rising levels of obesity.
      • Gustafson D.R.
      • McFarlane S.I.
      Epidemiology of type 2 diabetes and dementia.
      There is evidence of shared causal genes and biological pathways, yet the relationship is intricate and potentially reciprocal
      • Hu Z.
      • Jiao R.
      • Wang P.
      • et al.
      Shared causal paths underlying Alzheimer's dementia and type 2 diabetes.
      ; although individuals with T2D have a higher risk of developing dementia,
      • Cheng G.
      • Huang C.
      • Deng H.
      • et al.
      Diabetes as a risk factor for dementia and mild cognitive impairment: A meta-analysis of longitudinal studies.
      the presence of cognitive impairment may compromise an individual's ability to self-manage her or his diabetes given the complexity of some diabetes treatment regimens.
      • Santos T.
      • Lovell J.
      • Shiell K.
      • Johnson M.
      • Ibrahim J.E.
      The impact of cognitive impairment in dementia on self-care domains in diabetes: A systematic search and narrative review.
      • Gadsby R.
      • Hope S.
      • Hambling C.
      • Carnegie A.
      Frailty, older people and type 2 diabetes.
      • Sinclair A.J.
      • Hillson R.
      • Bayer A.J.
      National Expert Working Group. Diabetes and dementia in older people: A Best Clinical Practice Statement by a multidisciplinary National Expert Working Group.
      This may have clinical implications.
      Poor glycemic control, resulting in elevated glycated hemoglobin A1c (HbA1c), is associated with diabetes-related complications (eg, micro- and macrovascular complications), and an increased risk of emergency hospitalizations and mortality
      • Gunn L.H.
      • McKay A.J.
      • Molokhia M.
      • et al.
      Associations between attainment of incentivised primary care indicators and emergency hospital admissions among type 2 diabetes patients: A population-based historical cohort study.
      • Critchley J.A.
      • Carey I.M.
      • Harris T.
      • et al.
      Variability in glycated hemoglobin and risk of poor outcomes among people with type2 diabetes in a large primary care cohort study.
      • Dailey G.
      Overall mortality in diabetes mellitus: Where do we stand today?.
      • Gebregziabher M.
      • Egede L.E.
      • Lynch C.P.
      • et al.
      Effect of trajectories of glycemic control on mortality in type 2 diabetes: A semiparametric joint modeling approach.
      ; these account for a substantial proportion of the global direct costs of diabetes.
      • Hex N.
      • Bartlett C.
      • Wright D.
      • et al.
      Estimating the current and future costs of type 1 and type 2 diabetes in the UK, including direct health costs and indirect societal and productivity costs.
      Despite the potential challenges of diabetes self-management and the potential clinical and economic consequences, there is a paucity of information concerning patterns of glycemic control in those with dementia. We therefore sought to compare trajectories of HbA1c among individuals with preexisting T2D, newly diagnosed with dementia, to those of a matched group of patients without dementia, using linked routine clinical data from an inner-city London borough.
      • Dorrington S.
      • Carr E.
      • Stevelink S.A.M.
      • et al.
      Demographic variation in fit note receipt and long-term conditions in south London.
      In addition, we sought to evaluate health care utilization, differences in survival, and time to insulin initiation between the groups. As diabetes management may become increasingly challenging as cognition declines, we also evaluated differences based on the extent of cognitive impairment among those with dementia.

      Methods

      Study Design

      This retrospective matched cohort study included individuals with a preexisting T2D diagnosis from a primary care database and compared groups with and without a clinical diagnosis of dementia. Individuals registered at primary care practices in the London borough of Lambeth, part of the South London and Maudsley NHS Foundation Trust (SLaM),
      • Perera G.
      • Broadbent M.
      • Callard F.
      • et al.
      Cohort profile of the South London and Maudsley NHS Foundation Trust Biomedical Research Centre (SLaM BRC) Case Register: Current status and recent enhancement of an electronic mental health record-derived data resource.
      were eligible for inclusion.

      Data Sources and Sample

      Primary care data were obtained from Lambeth DataNet, a general practice (GP) electronic health record database that contains pseudonymized patient data recorded as part of routine clinical care. Data are collected from all practices within the London borough of Lambeth, which provides care for approximately 400,000 patients. Structured, individual-level data on clinical diagnoses, consultations, prescriptions, laboratory tests, and public health initiatives, such as the Quality and Outcomes Framework diagnostic data, are recorded.
      • Dorrington S.
      • Carr E.
      • Stevelink S.A.M.
      • et al.
      Demographic variation in fit note receipt and long-term conditions in south London.
      Clinical data are recorded using Read and Systematized Nomenclature of Medicine (SNOMED) clinical coding vocabularies, in line with the recent national change in preferred ontology.
      • Wardle M.
      • Spencer A.
      Implementation of SNOMED CT in an online clinical database.
      The SLaM Trust is a provider of mental health care and dementia assessment and management services for a total catchment of >1.2 million residents,
      • Perera G.
      • Broadbent M.
      • Callard F.
      • et al.
      Cohort profile of the South London and Maudsley NHS Foundation Trust Biomedical Research Centre (SLaM BRC) Case Register: Current status and recent enhancement of an electronic mental health record-derived data resource.
      including the borough of Lambeth. The Clinical Record Interactive Search (CRIS) data resource provides research access to pseudonymized electronic health records; it allows both structured and unstructured data to be abstracted from patient records based on interactions within secondary care mental health services within SLaM. By using an established linkage between Lambeth DataNet and the Clinical Record Interactive Search data resource, we were able to identify individuals registered to GP practices in Lambeth with a confirmed diagnosis of dementia and obtain measures of cognitive assessment from routine clinical care. Anonymized primary care data to December 31, 2019, were extracted.

      Study Population

      The source sample comprised patients with ≥12 months of prior registration on Lambeth DataNet and a T2D duration of ≥12 months at the index date. Individuals with primary care codes for type 1 diabetes on or prior to the index date were excluded.

      Dementia

      Individuals aged ≥65 years with a diagnosis of dementia (International Classification of Diseases, Tenth Revision, codes F00∗-F03∗) between January 1, 2007, and December 31, 2018, were ascertained. The index date was the date of first dementia-coded diagnosis in SLaM; measures in Lambeth DataNet for primary care interactions were anchored around this date. Individuals with evidence of mild cognitive impairment based on an MMSE score of >27 were excluded.

      Comparator Group

      The comparator group comprised T2D individuals without recorded dementia. Up to 5 comparator individuals per dementia case were matched on age (±5 years), sex, and GP practice at the index date. Individuals with recorded dementia prior to, or within 2 years of, the assigned index date in primary care were excluded. An individual could serve as a comparator for more than 1 case.

      Measurements

      Demographic characteristics included age at index date, sex, and self-reported ethnicity. The Index of Multiple Deprivation score,
      Ministry of Housing Communities and Local Government
      The English Indices of Deprivation 2019.
      a widely used measure of relative neighborhood-level deprivation, was derived from the last recorded address; a higher score indicates a more deprived area.
      NHS
      NHS data dictionary 2021.
      Baseline measures of health status within 12 months of, and closest to, the index date were extracted, including blood pressure (mm Hg), total cholesterol (mmol/L), body mass index (BMI), and Read codes indicating obesity and smoking status (never, ex-smoker, current). In the United Kingdom, older adults (aged ≥65 years) are eligible for an annual influenza vaccine; therefore, we included this as a measure of engagement with health care.
      The Charlson Comorbidity Index (CCI) was calculated for the 12-month period prior to the index date using published codes
      University College London
      CALIBER 2020.
      and weightings.
      • Metcalfe D.
      • Masters J.
      • Delmestri A.
      • et al.
      Coding algorithms for defining Charlson and Elixhauser co-morbidities in Read-coded databases.
      Dementia and uncomplicated diabetes were excluded from the overall score calculation. Other comorbidities included hypertension and dyslipidemia. Use of statins based on prescriptions in the year prior to the index date were also extracted.
      Selected diabetes-related health measures extracted or derived from Lambeth DataNet included diabetes duration (time from earliest T2D primary care code to index date) and age at first diabetes record. Diabetes therapy prescribed in the 3 month period up to index date was extracted and the following agents classified as binary variables: (1) metformin, (2) dipeptidyl peptidase-4 inhibitors (DDP4i), (3) glitazones, (4) glucagon-like peptide-1 agonists (GLP1), (5) insulin, (6) meglitinide, (7) sodium-glucose cotransporter-2 inhibitors (SGLT2i), (8) sulfonylureas; in addition, number of different therapies within these groups were estimated. HbA1c (mmol/mol), creatinine (μmol/L), and microalbuminuria (mg/L) measurements within 12 months of and closest to index date were extracted. Binary indicators of diabetes-related management (foot examination, retinal screening, and measurement of HbA1c) and diabetes complications (micro- or macrovascular complications and foot morbidity)
      Health Data Research UK
      HDR UK Phenotype Library 2021.
      were extracted within 12 months of index date.
      Cognitive function at dementia diagnosis was estimated from CRIS from Mini-Mental State Examination (MMSE) scores recorded within 12 months of and closest to the index date; otherwise, the closest Health of the Nation Outcome Scales (HoNOS) cognitive impairment subscale was selected for those without an MMSE measurement. HoNOS is a clinician-rated instrument, and its cognitive scale (0-4) correlates with MMSE measurement.
      • Pirkis J.E.
      • Burgess P.M.
      • Kirk P.K.
      • Dodson S.
      • Coombs T.J.
      • Williamson M.K.
      A review of the psychometric properties of the Health of the Nation Outcome Scales (HoNOS) family of measures.
      ,
      • Canuto A.
      • Weber K.
      • Gold G.
      • et al.
      Structured assessment of mental health status in psychogeriatrics: Validity of the French HoNOS65+.
      MMSE scores of 21-27, 10-20, or <10, or HoNOS cognitive impairment subscale scores of 2, 3, or 4, were used to represent mild, moderate, or severe dementia; moderate and severe categories were combined because of the relatively low number of individuals with severe dementia.

      Outcomes

      All measures of HbA1c from 3 years prior to 5 years following the index date were extracted and converted where necessary to millimoles per mole (mmol/mol) units using a standard formula.
      • Heinemann L.
      • Freckmann G.
      Quality of HbA1c measurement in the practice: the German perspective.
      For HbA1c measurements without units recorded, we used the approach taken by Rodgers et al.
      • Rodgers L.R.
      • Weedon M.N.
      • Henley W.E.
      • Hattersley A.T.
      • Shields B.M.
      Cohort profile for the MASTERMIND study: Using the Clinical Practice Research Datalink (CPRD) to investigate stratification of response to treatment in patients with type 2 diabetes.
      Values of <20 or >195 mmol/mol were considered outliers and excluded. Duplicate HbA1c measures on the same date were removed: where different results were recorded on the same date, an average was taken. Binary indicators of diabetes-related management and complications of diabetes were treated as outcomes in some analyses.
      Date of death was obtained from the primary care record. Individuals were censored at the earliest death, transfer out of practice, or December 31, 2019, up to a maximum of 5 years following the index date. Additionally, for comparators with evidence of dementia in their primary care record >2 years following the index date, HbA1c and health care encounters in the 2-year period prior to dementia diagnosis were not included in the models fitted. These exclusions were made to achieve consistency with the approach taken for dementia cases: because changes in cognition can occur years before a formal dementia diagnosis, data from such a period were deemed not to be appropriate for comparator purposes.
      Date of first insulin prescription among those individuals without evidence of use of insulin in the 3-month period up to the index date was extracted. Finally, GP consultations from 2 years before and up to 5 years after the index date were extracted; if several encounters were recorded on the same day, only 1 was included.

      Statistical Analyses

      Stata 15.0 was used for statistical analyses. Comparisons of baseline characteristics and binary outcome measures between the matched cohorts were made using conditional logistic regression. For the descriptive analyses, normally distributed continuous variables were presented as mean ± SD, and categorical variables were presented as percentages. For comparisons between those with mild and moderate-severe dementia, a t test or a χ2 test of association was performed to evaluate continuous and categorical variables, respectively. Adjusted and unadjusted logistic regressions were performed to compare binary outcomes in those with mild vs moderate-severe dementia.
      To evaluate HbA1c trajectories over time (with index date as time = 0), we used a linear mixed effects model with an exponential covariance structure, that is, a generalization of the autoregressive covariance model that allows for the differences in frequency and timing of repeated HbA1c measurements. An interaction term was included to test whether the slope of HbA1c level vs time differed depending on dementia status (no dementia vs dementia). This analysis accounted for the non-independence of data (ie, individuals could have repeated HbA1c measures) and for matching from the same GP. Both unadjusted and fully adjusted models (with terms for age, sex, CCI, diabetes-related measures at index date, diabetes duration, number of diabetes therapies, etc, as fixed effects) were fitted. The model with the lowest Akaike information criterion was chosen for further interpretation. The relationship between HbA1c and time was modeled with linear and quadratic terms; models with and without the quadratic term were compared using a likelihood ratio test. Predicted mean HbA1c and 95% CI were obtained at 6-month intervals. Lastly, to compare HbA1c trajectories by severity, models were fitted for those with mild dementia and those with moderate-severe dementia, in each case vs matched comparators without dementia.
      A Cox proportional hazards model with shared frailty, to account for matched case-comparator sets, was used to evaluate time from index date to death, and time to initiation of insulin among individuals not using insulin at baseline; estimates of the effect of dementia case vs comparator status were adjusted for baseline HbA1c, age, sex, social deprivation, diabetes duration, and number of classes of diabetes therapy at baseline. The proportional hazards (PH) assumption was tested using Schoenfeld residuals.
      Health care encounters were evaluated by quarter using a multilevel mixed effects binomial regression. To avoid artificially skewing the frequency of encounters toward the null hypothesis in individuals with partial follow-up in the first or last quarter, only complete quarters were evaluated.

      Results

      In total, 725 individuals with T2D and dementia and 3154 matched T2D comparators without dementia were included. Individuals with dementia were older, living in more socially deprived neighborhoods, had a lower BMI, lower proportion with a diagnosis of obesity, lower systolic blood pressure, and a higher CCI at index date than the matched comparator group (Table 1).
      Table 1Baseline Characteristics: Dementia Cases and Matched Nondementia Comparator Group
      Nondementia (n = 3154)All Dementia (n = 725)OR (95% CI)
      Conditional logistic model with dementia status as the dependent variable (nondementia as reference group). For continuous explanatory variables, OR is in relation to a 1-unit change in the explanatory variable.
      P
      Age at index date, y, mean (SD)78.7 (6.5)80.3 (6.5)1.14 (1.1, 1.17)<.001
      Sex (% female)
      Dementia and nondementia cases were matched on sex.
      55.456.3
      Ethnic group, %
       White British or European4036.4Reference
       Black African or Caribbean37.5411.24 (1.02, 1.51).032
       Asian10.510.11.1 (0.81, 1.49).55
       Mixed or other4.83.70.85 (0.55, 1.33).49
       Not recorded7.28.8
      IMD score, mean (SD)29.5 (9.8)30.5 (9.8)1.01 (1.00, 1.02).009
      BMI, mean (SD)29.0 (5.6)26.9 (5.6)0.93 (0.91, 0.94)<.001
      Obesity diagnosis, %2.40.80.32 (0.14, 0.75).009
      Registration prior to index, y, mean (SD)19.2 (13.9)18.2 (14.5)0.99 (0.99, 1.00).07
      Smoking status, %
       Never smoker5050.9Reference
       Ex-smoker19.719.30.97 (0.76, 1.23).80
       Current smoker12.610.50.89 (0.66, 1.21).47
       Not recorded17.719.3
      Systolic BP, mm Hg, mean (SD)136.7 (17.1)133.2 (18.2)0.99 (0.98, 0.99)<.001
      Diastolic BP, mm Hg, mean (SD)73.0 (10.2)72.5 (11.0)1 (0.99, 1.01).76
      Cholesterol, mmol/L, mean (SD)4.1 (1.0)4.3 (4.1)1.04 (0.99, 1.09).16
      Hypertension, %14.313.10.94 (0.73, 1.21).64
      Dyslipidemia, %3.32.80.85 (0.51, 1.4).52
      Statin use, %76.678.11.14 (0.93, 1.4).20
      Flu vaccine, %29.830.61.14 (0.88, 1.5).32
      CCI, mean (SD)0.8 (1.2)1.0 (1.4)1.15 (1.08, 1.22)<.001
      BP, blood pressure; IMD, Index of Multiple Deprivation; OR, odds ratio.
      Boldface indicates significance (P < .05).
      Conditional logistic model with dementia status as the dependent variable (nondementia as reference group). For continuous explanatory variables, OR is in relation to a 1-unit change in the explanatory variable.
      Dementia and nondementia cases were matched on sex.
      Individuals with dementia had a longer diabetes duration (13.4 vs 11.3 years, P < .001) and a higher mean HbA1c at the index date than those without dementia (58.3 and 56.6 mmol/mol respectively; P = .015) after accounting for the matching variables, social deprivation and CCI (Table 2). Dementia cases and comparators were prescribed a similar number of classes of antidiabetic medications in the 3-month period up to the index date, but cases were less likely to have received metformin (40.6% vs 43.7%, P = .007) and more likely to have received insulin, although this was of borderline significance (20.6% vs 15.8%, P = .052). Higher prevalences of macrovascular complications (9.4% vs 3.5%, P < .001) and diabetes foot morbidity (1.8% vs 0.4%, P = .003) were recorded in those with dementia. A lower proportion had recorded retinal screening (53.1% vs 68.2%, P < .001) or foot examination (69.1% vs 72.6%, P = .002) within 12 months of the index date than the matched comparator group. Demographic and diabetes-related comparisons by baseline dementia severity are described in Supplementary Tables 1 and 2. Those with mild dementia were living in less socially deprived neighborhoods, had a higher BMI, and greater use of statins, but fewer had received an influenza vaccine than those with moderate-severe dementia (Supplementary Table 1). Individuals with mild dementia had longer diabetes duration and a greater number of classes of diabetes medications prescribed in the 3-month period prior to the index date than those with moderate-severe dementia. Those with moderate-severe dementia were less likely to have had retinal screening, a foot examination, or an HbA1c test than those with mild dementia in the year prior to the index date (Supplementary Table 2).
      Table 2Baseline Diabetes-Related Measures Comparisons
      Conditional logistic model with dementia status as the dependent variable (nondementia as reference group). For continuous explanatory variables, this refers to the change in OR per unit change in the explanatory variable.
      Between Dementia Cases and Matched Nondementia Comparator Group
      Status at Index DateNondementia (n = 3154)All Dementia (n = 725)UnadjustedAdjusted
      Adjusted for age, sex, GP, social deprivation, and Charlson Comorbidity Index.
      OR (95% CI)POR (95% CI)P
      Diabetes duration, y, mean (SD)11.3 (7.9)13.4 (8.9)1.03 (1.02, 1.04)<.0011.03 (1.02, 1.04)<.001
      Age at diabetes diagnosis, y, mean (SD)67.9 (9.7)67.9 (10.55)0.99 (0.98, 1).080.98 (0.97, 0.99).001
      Number of classes of antidiabetic therapy prescribed in 3 mo up to index
      214 of 725 (29.5%) dementia cases and 1086 of 3154 (34.4%) comparators were not prescribed any antidiabetic medications in the 3-month period prior to the index date within the primary care setting.
      , mean (SD)
      1.0 (0.9)1.0 (0.9)1.07 (0.98, 1.18).141.07 (0.97, 1.17).20
      Antidiabetic therapies prescribed in the 3-mo period prior to index
      Not mutually exclusive; patients may have been prescribed multiple therapies at any given point or switched therapies in the 3-month period prior to index.
      , %
       Metformin43.740.60.66 (0.53, 0.82)<.0010.73 (0.58, 0.92).007
       DPP4i6.88.31.1 (0.78, 1.55).591.15 (0.81, 1.64).44
       Glitazone1.42.11.43 (0.74, 2.75).281.13 (0.55, 2.35).73
       GLP10.60
       Insulin15.820.61.32 (1.05, 1.67).0191.27 (1, 1.63).05
       Meglitinide0.20
       SGLT2i0.10
       Sulfonylureas29.330.60.91 (0.73, 1.12).360.92 (0.74, 1.15).46
       Other diabetes agents (gua gum, acarbose)0.40.30.95 (0.2, 4.6).950.75 (0.15, 3.7).73
      HbA1c mmol/mol, mean (SD)56.6 (15.7)58.3 (20.3)1.01 (1, 1.01).0051.01 (1, 1.01).015
      Creatinine μmol/L, mean (SD)102.6 (62.0)106.4 (69.8)1.0 (1.0, 1.0).151.0 (1.0, 1.0).86
      Microalbuminuria (mg/L), mean (SD)74.3 (168.9)78.0 (145.0)1.0 (1.0, 1.0).331.0 (1.0, 1.0).32
      Diabetes complications (1 y prior to index date), %
       Microvascular complications16.118.51.21 (0.97, 1.51).090.92 (0.7, 1.22).57
       Macrovascular complications3.59.42.94 (2.14, 4.04)<.0012.54 (1.81, 3.57)<.001
       Diabetes foot morbidity0.41.84.2 (1.94, 9.12)<.0013.45 (1.54, 7.72).003
      Measures related to ongoing diabetes management (within 1 y of index date), %
       At least 1 HbA1c measurement92.692.61.04 (0.76, 1.43).820.98 (0.7, 1.36).91
       Foot examination performed72.669.10.66 (0.51, 0.85).0010.65 (0.5, 0.85).002
       Retinal screening performed68.253.10.48 (0.4, 0.58)<.0010.48 (0.4, 0.58)<.001
      DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1, glucagon-like peptide-1 agonists; OR, odds ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitors.
      Boldface indicates significance (P < .05).
      Conditional logistic model with dementia status as the dependent variable (nondementia as reference group). For continuous explanatory variables, this refers to the change in OR per unit change in the explanatory variable.
      Adjusted for age, sex, GP, social deprivation, and Charlson Comorbidity Index.
      214 of 725 (29.5%) dementia cases and 1086 of 3154 (34.4%) comparators were not prescribed any antidiabetic medications in the 3-month period prior to the index date within the primary care setting.
      § Not mutually exclusive; patients may have been prescribed multiple therapies at any given point or switched therapies in the 3-month period prior to index.

      HbA1c Trajectory

      A total of 714 dementia and 3048 comparator patients contributed 5828 and 31,817 HbA1c measurements, respectively. Table 3 shows results of the linear mixed effects model examining the effect of dementia status (dementia: all, mild, and moderate-severe) on HbA1c levels over time, before and after covariate adjustment, compared with the matched comparator groups. In the first model, HbA1c in those with dementia was higher at all time points than in the nondementia group; however, HbA1c levels showed an increase over time in the nondementia group and no increase in those with dementia. Thus, the 2 groups converged (Figure 1A). HbA1c increased over time in both the mild dementia and the nondementia groups; however, HbA1c at the index date was higher, and the increase was greater, in those with mild dementia (positive interaction term). In contrast, for those with moderate-severe dementia, HbA1c did not significantly differ at the index date but diminished (negative interaction term) over time relative to the nondementia group (Figure 1B and C).
      Table 3Effects of Dementia Status on HbA1c Levels Over Time in People With Diabetes
      Dementia, All (vs Nondementia Comparators)Mild Dementia (vs Matched Nondementia Comparators)Moderate-Severe Dementia (vs Matched Nondementia Comparators)
      UnadjustedAdjusted
      Models were additionally adjusted for age at index, sex, neighborhood deprivation, CCI, diabetes duration, insulin use 3 months prior to the index date and self-reported ethnicity.
      UnadjustedAdjusted
      Models were additionally adjusted for age at index, sex, neighborhood deprivation, CCI, diabetes duration, insulin use 3 months prior to the index date and self-reported ethnicity.
      UnadjustedAdjusted
      Models were additionally adjusted for age at index, sex, neighborhood deprivation, CCI, diabetes duration, insulin use 3 months prior to the index date and self-reported ethnicity.
      Coefficient (SE)PCoefficient (SE)PCoefficient (SE)PCoefficient (SE)PCoefficient (SE)PCoefficient (SE)P
      Dementia (reference category = nondementia)1.66 (0.55).0031.80 (0.62).0033.55 (0.87)<.0012.49 (0.94).0080.15 (0.71).830.84 (0.84).32
      Time
      Coefficient indicates the change in HbA1c over a 12-month period in the nondementia reference category.
      0.36 (0.04)<.0010.32 (0.05)<.0010.287 (0.065)<.0010.22 (0.08).010.40 (0.05)<.0010.39 (0.07)<.001
      Dementia × time−0.37 (0.12).002−0.36 (0.14).0020.10 (0.17).570.36 (0.20).08−0.86 (0.16)<.001−1.21 (0.21)<.001
      Intercept56.93<.00163.69<.00156.93<.00164.08<.00156.93<.00163.17<.001
      Effects are estimated from a linear mixed effects model examining the relationship between dementia status and change in HbA1c levels over time. Boldface indicates significance (P < .05).
      Models were additionally adjusted for age at index, sex, neighborhood deprivation, CCI, diabetes duration, insulin use 3 months prior to the index date and self-reported ethnicity.
      Coefficient indicates the change in HbA1c over a 12-month period in the nondementia reference category.
      Figure thumbnail gr1
      Fig. 1HbA1c trajectories in people with diabetes from 3 years prior to index to up to 5 years following index date, by dementia status. (A) Dementia (all categories pooled) vs matched nondementia. (B) Mild dementia vs matched nondementia. (C) Moderate-severe dementia vs matched nondementia. The figure presents results from a linear mixed effects model of the relationship between dementia status and HbA1c level over time, relative to the date of diagnosis. The model included an HbA1c level × time interaction term to take into account the difference in slope depending on dementia status. The model used an exponential covariance structure, that is, a generalization of the autoregressive covariance model that allows for the differences in frequency and timing of repeated HbA1c measurements. This model takes into account the nonindependence of data (ie, individuals could have repeated HbA1c measures) and for matching from the same GP. Additionally, models were adjusted for age at index, sex, neighborhood deprivation, CCI, diabetes duration, insulin use 3 months prior to index and self-reported ethnicity (as fixed effect terms). Up to 5 comparators per dementia case were matched on age (±5 years at the index date), sex, and GP.

      Time-To-Event Analyses

      Mortality

      Overall, 17.7% (n = 558/3154) of the nondementia comparators and 32.8% (n = 238/725) of the dementia cases died during follow-up. The hazard ratio (HR) of mortality for dementia relative to nondementia was 2.75 (95% CI 2.35, 3.22) in the unadjusted model, and remained significant in the fully adjusted model (2.16; 1.78, 2.62). Fully adjusted HRs for mild and moderate-severe dementia, compared to controls, were 1.54 (1.16, 2.06) and 2.82 (2.23, 3.55), respectively (Supplementary Table 3 and Supplementary Figure 1). The most marked changes were observed where neighborhood deprivation, ethnicity, CCI, and smoking status were included in models. In a secondary analysis to evaluate the potential effect of survival bias on HbA1c trajectories, higher age-adjusted HRs were seen in the highest baseline HbA1c category (≥75 mmol/mol) relative to the lowest (≤58 mmol/mol), but no difference in those with a baseline HbA1c of 59 to 74.9 mmol/mol relative to the lowest category; the magnitude of difference also did not vary substantially by dementia severity (Supplementary Table 4).

      Time to insulin initiation

      In 565 individuals with dementia and 2108 nondementia comparator patients with no insulin prescriptions in the 3-month period prior to the index date, we observed no significant differences in time to insulin initiation in either unadjusted or adjusted analyses (Supplementary Table 5).

      Health care encounters

      The multilevel negative binomial model was adjusted for age at index date, sex, social deprivation, CCI, and diabetes duration; time included as a quadratic term resulted in a statistically significant improvement in model fit relative to a model with time as a linear term only [LR χ2(3) = 120.5, P < .001]. Health care utilization increased in the lead up to the index date for those with dementia and declined thereafter but remained constant over time in the nondementia group (Figure 2A and B). Those with mild dementia had a greater number of encounters at all time points than the moderate-severe dementia group, with statistically significant divergence between the groups by the end of follow-up. Model coefficients are presented in Supplementary Table 6.
      Figure thumbnail gr2
      Fig. 2GP consultations by quarter (2 years prior and up to 5 years post index date) in people with diabetes by dementia status. (A) Dementia (all categories pooled) vs matched nondementia. (B) Mild and moderate-severe dementia. The estimated mean numbers of GP consultations over successive time periods are presented. Time was modeled as a quadratic effect. Estimates were obtained from multilevel mixed effects negative binomial models adjusted for age at index date, sex, social deprivation, CCI, and diabetes duration as fixed effects terms. The random effects model terms take into account the nonindependence of data (ie, individuals could have repeated health care encounters) and of matching of patients from the same GP.

      Discussion

      In this retrospective cohort of individuals with preexisting T2D, we report different HbA1c trajectories based on dementia status and stage. Specifically, HbA1c levels showed a greater increase over time in those with mild vs no dementia, but a decrease in those with moderate-severe vs no dementia. Despite individuals with dementia having increased health care utilization around the dementia diagnosis, they were less likely to have had routine diabetes-related management. Patients with dementia had a higher prevalence of macrovascular complications and diabetes foot morbidity and a higher mortality risk than those without dementia, particularly in those with moderate-severe dementia. To our knowledge, this is the first study to examine longitudinal changes in glycemic control associated with dementia as well as the extent of cognitive impairment.

      Diabetes-Related Measures at Baseline

      Even at the time of dementia diagnosis, differences in diabetes-related clinical measures were apparent by dementia status and stage. Individuals with dementia had a greater odds of macrovascular complications and diabetes foot morbidity but were less likely to have had routine retinal screening or foot examination than those without dementia, particularly in those with more advanced dementia. Other studies have also reported less frequent diabetes monitoring and an increased risk of diabetes complications in those with dementia compared to those without dementia,
      • Wargny M.
      • Gallini A.
      • Hanaire H.
      • Nourhashemi F.
      • Andrieu S.
      • Gardette V.
      Diabetes care and dementia among older adults: A nationwide 3-year longitudinal study.
      ,
      • Thorpe C.T.
      • Thorpe J.M.
      • Kind A.J.
      • Bartels C.M.
      • Everett C.M.
      • Smith M.A.
      Receipt of monitoring of diabetes mellitus in older adults with comorbid dementia.
      including a survey of US nursing home medical directors that reported less routine diabetes monitoring in nursing home residents with cognitive impairment than residents without.
      • McNabney M.K.
      • Pandya N.
      • Iwuagwu C.
      • et al.
      Differences in diabetes management of nursing home patients based on functional and cognitive status.
      In our study, patients with dementia had significantly more health care encounters around the time of dementia diagnosis than the nondementia group; thus, dementia may serve as a competing demand in the lead up to dementia diagnosis to the detriment of diabetes management,
      • Piette J.D.
      • Kerr E.A.
      The impact of comorbid chronic conditions on diabetes care.
      particularly in those with more cognitive impairment.

      HbA1c Trajectories

      Both mean HbA1c and trajectory of HbA1c are clinically important. Higher mean levels of HbA1c
      • Walraven I.
      • Mast M.R.
      • Hoekstra T.
      • et al.
      Distinct HbA1c trajectories in a type 2 diabetes cohort.
      and increasing, decreasing, or more variable patterns of glycemic control in T2D over time have been associated with higher risks of diabetic complications, mortality,
      • Critchley J.A.
      • Carey I.M.
      • Harris T.
      • et al.
      Variability in glycated hemoglobin and risk of poor outcomes among people with type2 diabetes in a large primary care cohort study.
      ,
      • Gebregziabher M.
      • Egede L.E.
      • Lynch C.P.
      • et al.
      Effect of trajectories of glycemic control on mortality in type 2 diabetes: A semiparametric joint modeling approach.
      ,
      • Laiteerapong N.
      • Karter A.J.
      • Moffet H.H.
      • et al.
      Ten-year hemoglobin A1c trajectories and outcomes in type 2 diabetes mellitus: The Diabetes & Aging Study.
      • Luo M.
      • Lim W.Y.
      • Tan C.S.
      • et al.
      Longitudinal trends in HbA1c and associations with comorbidity and all-cause mortality in Asian patients with type 2 diabetes: A cohort study.
      • Hemo B.
      • Geva D.
      • Shahar D.R.
      • Golan R.
      • Heymann A.D.
      Distinct trajectories in HbA1c are associated with different all-cause mortality and morbidity in newly diagnosed patients with type 2 diabetes.
      emergency hospital admissions,
      • Critchley J.A.
      • Carey I.M.
      • Harris T.
      • et al.
      Variability in glycated hemoglobin and risk of poor outcomes among people with type2 diabetes in a large primary care cohort study.
      and potentially avoidable hospitalizations.
      • Zaslavsky O.
      • Yu O.
      • Walker R.L.
      • et al.
      Incident dementia, glycated hemoglobin (HbA1c) levels and potentially preventable hospitalizations in people age 65 and older with diabetes.
      Our cohort had an increasing HbA1c trajectory overall, consistent with age-related increases in HbA1c.
      • Masuch A.
      • Friedrich N.
      • Roth J.
      • Nauck M.
      • Müller U.A.
      • Petersmann A.
      Preventing misdiagnosis of diabetes in the elderly: Age-dependent HbA1c reference intervals derived from two population-based study cohorts.
      ,
      • Dubowitz N.
      • Xue W.
      • Long Q.
      • et al.
      Aging is associated with increased HbA1c levels, independently of glucose levels and insulin resistance, and also with decreased HbA1c diagnostic specificity.
      Challenges in diabetes self-management because of cognitive impairment,
      • Santos T.
      • Lovell J.
      • Shiell K.
      • Johnson M.
      • Ibrahim J.E.
      The impact of cognitive impairment in dementia on self-care domains in diabetes: A systematic search and narrative review.
      • Gadsby R.
      • Hope S.
      • Hambling C.
      • Carnegie A.
      Frailty, older people and type 2 diabetes.
      • Sinclair A.J.
      • Hillson R.
      • Bayer A.J.
      National Expert Working Group. Diabetes and dementia in older people: A Best Clinical Practice Statement by a multidisciplinary National Expert Working Group.
      or poor glycemic control as a cause of cognitive decline, could contribute to the steeper increase in HbA1c in those with mild dementia. In contrast, declining HbA1c trajectory in those with moderate-severe dementia may be due to weight loss and/or frailty, which are more common in dementia,
      • Clegg A.
      • Young J.
      • Iliffe S.
      • Rikkert M.O.
      • Rockwood K.
      Frailty in elderly people.
      particularly in its later stages.
      • White H.
      • Pieper C.
      • Schmader K.
      The association of weight change in Alzheimer's disease with severity of disease and mortality: A longitudinal analysis.
      Loss of appetite, changes in energy consumption, olfactory changes, psychobehavioral disturbances, and dysphagia may contribute to weight loss in advanced dementia.
      • Morley J.E.
      Anorexia of aging: Physiologic and pathologic.
      In frail older people with diabetes, weight loss may increase insulin sensitivity, leading to normoglycemia and an increased risk of hypoglycemia, resulting in a reduced need for hypoglycemic medications.
      • Abdelhafiz A.H.
      • Koay L.
      • Sinclair A.J.
      The effect of frailty should be considered in the management plan of older people with Type 2 diabetes.
      Other studies have shown that a lower baseline cognitive function is associated with an increased risk of hypoglycemia.
      • Punthakee Z.
      • Miller M.E.
      • Launer L.J.
      • et al.
      Poor cognitive function and risk of severe hypoglycemia in type 2 diabetes: Post hoc epidemiologic analysis of the ACCORD trial.
      ,
      • de Galan B.E.
      • Zoungas S.
      • Chalmers J.
      • et al.
      Cognitive function and risks of cardiovascular disease and hypoglycaemia in patients with type 2 diabetes: the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial.
      These processes may be particularly pertinent in advanced stages of dementia and within the institutional care sector, and would benefit from further, more focused research. Of note, in our analysis, dementia patients had higher insulin use at baseline, but no difference in insulin receipt by dementia severity, and time to insulin initiation did not differ between people with and without dementia.

      Mortality

      We observed a dose-dependent relationship between increasing mortality risk and greater severity of dementia, even after adjusting for known risk factors for mortality in those with dementia
      • Sachs G.A.
      • Carter R.
      • Holtz L.R.
      • et al.
      Cognitive impairment: An independent predictor of excess mortality: A cohort study.
      ,
      • Garcia-Ptacek S.
      • Farahmand B.
      • Kareholt I.
      • et al.
      Mortality risk after dementia diagnosis by dementia type and underlying factors: A cohort of 15,209 patients based on the Swedish Dementia Registry.
      or diabetes.
      • Dailey G.
      Overall mortality in diabetes mellitus: Where do we stand today?.
      ,
      • Gebregziabher M.
      • Egede L.E.
      • Lynch C.P.
      • et al.
      Effect of trajectories of glycemic control on mortality in type 2 diabetes: A semiparametric joint modeling approach.
      Dementia itself is associated with a higher risk of mortality. For example, a UK primary care study reported a 2.5-fold greater mortality rate,
      • Rait G.
      • Walters K.
      • Bottomley C.
      • Petersen I.
      • Iliffe S.
      • Nazareth I.
      Survival of people with clinical diagnosis of dementia in primary care: Cohort study.
      which is broadly consistent with our findings of a 2.2-fold higher mortality, and therefore underlying factors may not be diabetes-specific.

      Health Care Utilization in Primary Care

      The higher number of primary care consultations around the time of dementia diagnosis and subsequent decline is consistent with prior studies.
      • Chen L.
      • Reed C.
      • Happich M.
      • et al.
      Health care resource utilisation in primary care prior to and after a diagnosis of Alzheimer's disease: A retrospective, matched case-control study in the United Kingdom.
      Greater health care utilization potentially reflects higher morbidity and increased interactions in the iterative steps involved in a dementia diagnosis. The subsequent decline relative to comparators might be due to an increased role of carers, access to home health care, or move to more supportive accommodation, better equipped for ambulatory care conditions,
      • Fillenbaum G.
      • Heyman A.
      • Peterson B.L.
      • Pieper C.F.
      • Weiman A.L.
      Use and cost of hospitalization of patients with AD by stage and living arrangement: CERAD XXI.
      particularly as health care utilization of the controls was relatively stable over time. Alternatively, reduced GP contact in those with dementia may be indicative of difficulties accessing or seeking care, particularly in those with moderate or severe dementia. This is clearly a concern in the context of less adequate monitoring of T2D and its consequences.

      Strengths and Limitations

      To our knowledge, this is the first longitudinal study of changes in glycemic control associated with dementia as well as extent of cognitive impairment; important differences in trajectories would have been missed if cognitive impairment was not accounted for. All available HbA1c measurements were used, permitting us to characterize trajectories from 3 years prior to up to 5 years after a dementia diagnosis. Considering limitations, measures of cognitive impairment were not available for the comparator group, and we cannot exclude missed/undiagnosed cases. Further, factors such as mobility, functional impairment, and lack of social support may contribute to the observed associations of diabetes-related complications such as foot morbidity and dementia severity; these were not captured in our data set. Details on whether patients were living in supported accommodation and had support from carers or access to home health care was not available to us. We considered survival bias as an alternative explanation of the declining HbA1c trajectory observed in those with moderate-severe dementia (eg, survivors being more likely to have lower HbA1c); however, in secondary analyses evaluating mortality by baseline HbA1c, there was no evidence that the positive association increased for those with dementia or by dementia severity; thus, this explanation is unlikely. Cause of death was not available in this data set, and it was therefore not possible to determine whether the underlying cause was due to diabetes-related factors. Changes in or adherence to diabetes medication were not evaluated; however, HbA1c can be viewed as a measure of treatment effect irrespective of adherence. Other unevaluated factors that may influence HbA1c trajectories or their interpretation include weight loss or frailty, hypoglycemic episodes, anemia, chronic renal disease, and certain nondiabetes medications
      • Gallagher E.J.
      • Le Roith D.
      • Bloomgarden Z.
      Review of hemoglobin A(1c) in the management of diabetes.
      ; furthermore, community and institutional settings were not evaluated in relation to health care use. Additionally, although there is considerable national-level standardization around diabetes management in primary care within the United Kingdom, there is potential for between-practitioner variation; our view is that such variation is more likely to obscure than exaggerate underlying associations. Some diabetes-related health care indicators may have changed during the time span of this study; for example, wording exempting patients with moderate or severe frailty from clinical target achievement was included in the 2019-2020 Quality and Outcomes Framework. Finally, we pooled moderate and severe categories of dementia; larger samples could elucidate whether there are further differences in HbA1c trajectories for severe dementia.

      Conclusions and Implications

      Our findings highlight unmet needs in older people with diabetes and comorbid dementia; we report differences in the monitoring, management, and control of diabetes based on the presence of dementia and the extent of cognitive impairment. The distinct HbA1c trajectories we observed in people with mild dementia and moderate-severe dementia have been associated with poorer clinical outcomes (diabetes complications and mortality) and a higher risk of emergency hospitalizations by other studies, in nondementia samples. Our findings indicate the need for caution when interpreting lower HbA1c values as indicating good glycemic control: instead, it should be considered in the context of HbA1c changes over time, extent of cognitive impairment, and presence of frailty including weight loss. The greater use of insulin at baseline in those with dementia, despite their having lower mean BMI than the nondementia group, is consistent with studies in frail older people with diabetes that report unnecessarily tight glycemic control and overtreatment
      • Lipska K.J.
      • Ross J.S.
      • Miao Y.
      • et al.
      Potential overtreatment of diabetes mellitus in older adults with tight glycemic control.
      despite current guidelines to relax glycemic targets and reduce use of hypoglycemic medications in these population.
      • Gadsby R.
      • Hope S.
      • Hambling C.
      • Carnegie A.
      Frailty, older people and type 2 diabetes.
      ,
      • Abdelhafiz A.H.
      • Koay L.
      • Sinclair A.J.
      The effect of frailty should be considered in the management plan of older people with Type 2 diabetes.
      ,
      American Diabetes Association
      12. Older adults: Standards of Medical Care in Diabetes-2020.
      ,
      International Diabetes Federation
      Managing Older People With Type 2 Diabetes: IDF Global Guideline.
      Lastly, despite high levels of health care utilization around the time of dementia diagnosis, those with dementia received less routine monitoring of their diabetes, suggesting the need to ensure that diabetes care is not compromised if dementia is present. This is particularly important given the higher prevalence of macrovascular complications, diabetes foot morbidity, and mortality risk that we observed, especially in those with moderate-severe dementia. Tailoring diabetes management and adapting diabetes care–related indicators in the primary care setting, to account for the effects of frailty and the extent of cognitive impairment, may help improve clinical outcomes, quality, access, and efficiency of care in this vulnerable population.

      Supplementary Data

      Supplementary Table 1Baseline Characteristics: Mild and Moderate-Severe Dementia Case Comparisons
      Mild (n = 290)Moderate-severe (n = 435)Test Statistic (Degrees of Freedom)P value
      Age at index date, y, mean (SD)79.91 (6.06)80.52 (6.78)t(723) = −1.253.211
      Sex, %
       Female55.5256.78χ2(1) = 0.113.737
       Male44.4843.22
      Ethnic group, %
       White British or European36.936.09χ2(5) = 2.854.723
       Black African or Caribbean40.6941.15
       Asian11.728.97
       Mixed or other3.453.91
       Not recorded7.249.89
      IMD score, mean (SD)29.08 (9.88)31.42 (9.64)t(711) = −3.150.002
      BMI, mean (SD)27.61 (5.18)26.29 (5.75)t(561) = 2.805.005
      Obesity, %1.380.46χ2(1) = 1.7923.181
      Registration prior to index, y, mean (SD)18.7 (13.15)17.81 (15.26)t(690) = 0.795.427
      Smoking status, %
       Never smoker51.7250.34χ2(3) = 9.519.023
       Ex-smoker23.116.78
       Current smoker10.6910.34
       Not recorded14.4822.53
      Frailty index∗, mean (SD)0.4 (0.14)0.39 (0.09)t(98) = 0.373.710
      Hypertension, %15.1711.72χ2(1) = 1.817.178
      Systolic BP, mm Hg, mean (SD)134.65 (17.87)132.16 (18.32)t(711) = 1.8.072
      Diastolic BP, mm Hg, mean (SD)72.33 (10.32)72.63 (11.39)t(711) = −0.363.717
      Cholesterol, mmol/L, mean (SD)4.1 (1.04)4.47 (5.3)t(667) = −1.131.258
      Dyslipidemia, %1.383.68χ2(1) = 3.428.064
      Statin use, %82.175.4χ2(1) = 4.517.034
      Flu vaccine, %23.535.4χ2(1) = 11.704.001
      CCI, mean (SD)1.02 (1.41)1.04 (1.39)t(723) = −0.228.5901
      BMI, body mass index; BP, blood pressure; CCI, Charlson Comorbidity Index; IMD, Index of Multiple Deprivation.
      Supplementary Table 2Baseline Diabetes-Related Measures: Mild and Moderate-Severe Dementia Cases Comparisons
      Mild (n = 290)Moderate-Severe (n = 435)Test Statistic (df), PUnadjustedAdjusted
      Adjusted for age at index date, sex, GP practice, social deprivation and Charlson Comorbidity Index.
      Odds Ratio (95% CI)POdds Ratio (95% CI)P
      Diabetes duration, y, mean (SD)14.2 (9.2)12.9 (8.6)t(723) = 1.9, P = .0560.98 (0.97, 1).0580.98 (0.97, 1).043
      Age at diabetes diagnosis, y, mean (SD)67.0 (9.9)68.6 (11.0)t(723) = −2.0, P = .0441.01 (1, 1.03).0441.02 (1, 1.03).08
      Number of classes of antidiabetic therapy prescribed in 3 mo up to index date
      214 of 725 (29.5%) dementia cases and 1086 of 3154 (34.4%) comparators were not prescribed any anti-diabetic medications in the 3-month period prior to index date within the primary care setting.
      , mean (SD)
      1.2 (0.9)0.9 (0.8)t(723) = 4.4, P < .0010.68 (0.57, 0.81)<.0010.69 (0.57, 0.82)<.001
      Antidiabetic therapies prescribed in the 3 mo period
      Not mutually exclusive; patients may have been prescribed multiple therapies at any given point or switched therapies in the 3-month period prior to index.
      prior to index date, %
       Metformin47.635.9χ2(1) = 1.0, P = .310.83 (0.58, 1.18).310.87 (0.61, 1.26).47
       DPP4i11.76χ2(1) = 3.7, P = .050.59 (0.34, 1.01).060.6 (0.35, 1.05).08
       Glitazone2.12.1χ2(1) = 0.2, P = .691.24 (0.43, 3.52).691.03 (0.32, 3.3).95
       GLP100
       Insulin22.819.1χ2(1) = 0.04, P = .841.04 (0.71, 1.53).841.04 (0.7, 1.55).83
       Meglitinide00
       SGLT2i00
       Sulfonylureas34.827.8χ2(1) = 0.04, P = .850.97 (0.68, 1.37).850.96 (0.67, 1.38).83
       Other diabetes agents (gua gum, acarbose)0.30.2χ2(1) = 0.02, P = .890.82 (0.05, 13.15).890.52 (0.03, 9.03).66
       HbA1c, mmol/mol, mean (SD)59.7 (19.8)57.3 (20.6)t(669) = 1.5, P = .140.99 (0.99, 1).140.99 (0.99, 1).20
       Creatinine, μmol/L, mean (SD)101.4 (48.1)109.9 (81.4)t(692) = −1.6, P = .111 (1, 1).121 (1, 1).16
       Microalbuminuria, mg/L, mean (SD)74.2 (141.0)81.4 (148.9)t(266) = −0.403, P = .691 (1, 1).851 (1, 1).81
      Diabetes complications (1 y prior to index date), %
       Microvascular complications2017.5χ2(1) = 0.7, P = .390.85 (0.58, 1.24).390.76 (0.47, 1.24).28
       Macrovascular complications7.910.3χ2(1) = 1.2, P = .281.34 (0.79, 2.27).281.43 (0.83, 2.47).20
       Diabetes foot morbidity2.11.6χ2(1) = 0.2, P = .650.77 (0.26, 2.33).650.67 (0.22, 2.06).49
      Measures related to ongoing diabetes management (within 1 y of index date), %
       At least 1 HbA1c measurement95.290.8χ2(1) = 4.8, P = .0280.5 (0.27, 0.94).0310.54 (0.29, 1.02).06
       Foot examination performed78.662.8χ2(1) = 20.5, P < .0010.46 (0.33, 0.64)<.0010.43 (0.3, 0.61)<.001
       Retinal screening performed60.748.1χ2(1) = 11.2, P = .0010.6 (0.44, 0.81).0010.62 (0.46, 0.85).003
      df, degrees of freedom; DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1, glucagon-like peptide-1 agonists; HbA1c, glycated hemoglobin A1c; OR, odds ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitors.
      Adjusted for age at index date, sex, GP practice, social deprivation and Charlson Comorbidity Index.
      214 of 725 (29.5%) dementia cases and 1086 of 3154 (34.4%) comparators were not prescribed any anti-diabetic medications in the 3-month period prior to index date within the primary care setting.
      Not mutually exclusive; patients may have been prescribed multiple therapies at any given point or switched therapies in the 3-month period prior to index.
      Supplementary Table 3Unadjusted and Adjusted Hazard Ratios for Mortality in People With Diabetes by Dementia Presence and Severity (Relative to No Dementia)
      Analysis based on 714 dementia (288 mild, 426 moderate-severe) and 3154 nondementia comparators.
      Model AdjustmentsHazard Ratios (95% CI) by Baseline Dementia Severity (Nondementia Reference)
      Dementia, AllMild DementiaModerate-Severe Dementia
      A. Unadjusted2.75 (2.35, 3.22)1.94 (1.52, 2.48)3.47 (2.89, 4.17)
      B. Age and gender adjusted2.52 (2.16, 2.95)1.86 (1.46, 2.36)3.09 (2.57, 3.71)
      C. Plus IMD, ethnicity, CCI, smoking status2.25 (1.83, 2.76)1.65 (1.24, 2.19)2.95 (2.36, 3.7)
      D. Plus baseline HbA1c
      HbA1c: ≤58 mmol/mol (reference), 59-74.9 mmol/mol, and ≥75 mmol/mol.
      2.19 (1.81, 2.66)1.59 (1.19, 2.12)2.83 (2.24, 3.56)
      E. Plus diabetes duration, number of classes of diabetes therapies used in 3 mo prior to index date2.16 (1.78, 2.62)1.54 (1.16, 2.06)2.82 (2.23, 3.55)
      F: Plus microvascular or macrovascular complications2.17 (1.78, 2.63)1.55 (1.16, 2.06)2.83 (2.24, 3.57)
      G. Model E plus BMI and baseline systolic blood pressure
      BMI and changes in blood pressure may be secondary to dementia.
      2.14 (1.73, 2.64)1.51 (1.11, 2.05)2.82 (2.18, 3.63)
      BMI, body mass index; CCI, Charlson Comorbidity Index; HbA1c, glycated hemoglobin A1c; IMD, Index of Multiple Deprivation.
      Overall, 17.7% (n = 558) nondementia, 26.4% of mild (n = 76), and 36.2% of moderate-severe dementia (n = 154) died during follow-up.
      Analysis based on 714 dementia (288 mild, 426 moderate-severe) and 3154 nondementia comparators.
      HbA1c: ≤58 mmol/mol (reference), 59-74.9 mmol/mol, and ≥75 mmol/mol.
      BMI and changes in blood pressure may be secondary to dementia.
      Supplementary Table 4Age-Adjusted Hazard Ratios for Mortality by Baseline HbA1c
      Baseline HbA1c (≤58 mmol/mol Reference)Hazard Ratio (95% CI) for Mortality by Case Status
      NondementiaMild DementiaModerate-Severe Dementia
      59-74.9 mmol/mol0.99 (0.8, 1.23)1.27 (0.72, 2.22)0.91 (0.55, 1.49)
      ≥75 mmol/mol1.41 (1.09, 1.83)1.49 (0.84, 2.65)1.2 (0.76, 1.9)
      HbA1c, glycated hemoglobin A1c.
      Supplementary Table 5Unadjusted and Adjusted Hazard Ratios for Insulin Initiation by Dementia Presence and Severity (Relative to Nondementia)
      Hazard Ratio (95% CI) for Insulin Initiation by Case Status (Nondementia Reference)
      Dementia, AllMild DementiaModerate-Severe Dementia
      A. Unadjusted0.84 (0.51, 1.39)0.97 (0.49, 1.92)0.74 (0.38, 1.47)
      B. Age and gender adjusted0.86 (0.52, 1.43)0.99 (0.5, 1.96)0.77 (0.39, 1.52)
      C. Plus baseline HbA1c, diabetes duration, CCI, ethnic group0.65 (0.37, 1.14)0.67 (0.31, 1.41)0.63 (0.28, 1.38)
      CCI, Charlson Comorbidity Index; HbA1c, glycated hemoglobin A1c.
      Analysis sample was composed of individuals without evidence of insulin use in the 3-month period prior to the index date (n = 222 mild dementia, 343 moderate/severe dementia, and 2108 nondementia comparators).
      Supplementary Table 6Results From the Mixed Effects Negative Binomial Regression
      Mixed effects negative binomial regression adjusted for age at index date, sex, social deprivation, Charlson Comorbidity Index, and diabetes duration; time included as a quadratic term.
      Evaluating GP Encounters Over Time
      CoefficientSEzP Value
      Dementia (nondementia reference)0.26010.024010.85<.001
      Time0.00210.00073.08.002
      Case status × Time (quarter)−0.00150.0016−0.89.37
      Case status × Quadratic time−0.00020.0001−3.21.001
      Time × Time−0.00120.0002−7.93<.001
      Age at index0.00380.00162.3.021
      Sex0.05460.02192.5.012
      IMD−0.00330.0010−3.41.001
      CCI0.10460.007513.96<.001
      Diabetes duration0.00760.00126.51<.001
      Intercept0.61480.13584.53<.001
      Log-transformed overdispersion parameter−1.2980.0122
      Between group variance component0.02540.0046
      Between-subject (within-group) variance component0.26000.0080
      CCI, Charlson Comorbidity Index; GP, general practice; IMD, Index of Multiple Deprivation.
      LR test vs negative binomial model: χ2(2) = 18,779.59, P < .001 (favors a mixed effects negative binomial regression over negative binomial regression without random effects). Boldface indicates significance (P < .05).
      Mixed effects negative binomial regression adjusted for age at index date, sex, social deprivation, Charlson Comorbidity Index, and diabetes duration; time included as a quadratic term.
      Figure thumbnail fx1
      Supplementary Fig. 1Kaplan-Meier curve for mortality in people with diabetes by dementia status and time (months). Analysis adjusted for demographics, Charlson Comorbidity Index, baseline glycated hemoglobin A1c (HbA1c) and number of classes of diabetes therapies used in the 3 months prior to the index date (model E in ).

      References

        • Bunn F.
        • Burn A.M.
        • Goodman C.
        • et al.
        Comorbidity and dementia: A scoping review of the literature.
        BMC Med. 2014; 12: 192
        • Gustafson D.R.
        • McFarlane S.I.
        Epidemiology of type 2 diabetes and dementia.
        in: Srikanth V. Arvanitakis Z. Type 2 Diabetes and Dementia. Elsevier Science & Technology, 2018
        • Hu Z.
        • Jiao R.
        • Wang P.
        • et al.
        Shared causal paths underlying Alzheimer's dementia and type 2 diabetes.
        Sci Rep. 2020; 10: 4107
        • Cheng G.
        • Huang C.
        • Deng H.
        • et al.
        Diabetes as a risk factor for dementia and mild cognitive impairment: A meta-analysis of longitudinal studies.
        Intern Med J. 2012; 42: 484-491
        • Santos T.
        • Lovell J.
        • Shiell K.
        • Johnson M.
        • Ibrahim J.E.
        The impact of cognitive impairment in dementia on self-care domains in diabetes: A systematic search and narrative review.
        Diabetes Metab Res Rev. 2018; 34: e3013
        • Gadsby R.
        • Hope S.
        • Hambling C.
        • Carnegie A.
        Frailty, older people and type 2 diabetes.
        Diabetes Prim Care. 2017; 19: 18-22
        • Sinclair A.J.
        • Hillson R.
        • Bayer A.J.
        National Expert Working Group. Diabetes and dementia in older people: A Best Clinical Practice Statement by a multidisciplinary National Expert Working Group.
        Diabet Med. 2014; 31: 1024-1031
        • Gunn L.H.
        • McKay A.J.
        • Molokhia M.
        • et al.
        Associations between attainment of incentivised primary care indicators and emergency hospital admissions among type 2 diabetes patients: A population-based historical cohort study.
        J R Soc Med. 2021; 114: 299-312
        • Critchley J.A.
        • Carey I.M.
        • Harris T.
        • et al.
        Variability in glycated hemoglobin and risk of poor outcomes among people with type2 diabetes in a large primary care cohort study.
        Diabetes Care. 2019; 42: 2237-2246
        • Dailey G.
        Overall mortality in diabetes mellitus: Where do we stand today?.
        Diabetes Technol Ther. 2011; 13: S65-S74
        • Gebregziabher M.
        • Egede L.E.
        • Lynch C.P.
        • et al.
        Effect of trajectories of glycemic control on mortality in type 2 diabetes: A semiparametric joint modeling approach.
        Am J Epidemiol. 2010; 171: 1090-1098
        • Hex N.
        • Bartlett C.
        • Wright D.
        • et al.
        Estimating the current and future costs of type 1 and type 2 diabetes in the UK, including direct health costs and indirect societal and productivity costs.
        Diabet Med. 2012; 29: 855-862
        • Dorrington S.
        • Carr E.
        • Stevelink S.A.M.
        • et al.
        Demographic variation in fit note receipt and long-term conditions in south London.
        Occup Environ Med. 2020; 77: 418-426
        • Perera G.
        • Broadbent M.
        • Callard F.
        • et al.
        Cohort profile of the South London and Maudsley NHS Foundation Trust Biomedical Research Centre (SLaM BRC) Case Register: Current status and recent enhancement of an electronic mental health record-derived data resource.
        BMJ Open. 2016; 6: e008721
        • Wardle M.
        • Spencer A.
        Implementation of SNOMED CT in an online clinical database.
        Future Healthc J. 2017; 4: 126-130
        • Ministry of Housing Communities and Local Government
        The English Indices of Deprivation 2019.
        Ministry of Housing Communities and Local Government, 2019
        • NHS
        NHS data dictionary 2021.
        https://datadictionary.nhs.uk/index.html
        Date accessed: February 21, 2021
        • University College London
        CALIBER 2020.
        • Metcalfe D.
        • Masters J.
        • Delmestri A.
        • et al.
        Coding algorithms for defining Charlson and Elixhauser co-morbidities in Read-coded databases.
        BMC Med Res Methodol. 2019; 19: 115
        • Health Data Research UK
        HDR UK Phenotype Library 2021.
        • Pirkis J.E.
        • Burgess P.M.
        • Kirk P.K.
        • Dodson S.
        • Coombs T.J.
        • Williamson M.K.
        A review of the psychometric properties of the Health of the Nation Outcome Scales (HoNOS) family of measures.
        Health Qual Life Outcomes. 2005; 3: 76
        • Canuto A.
        • Weber K.
        • Gold G.
        • et al.
        Structured assessment of mental health status in psychogeriatrics: Validity of the French HoNOS65+.
        Can J Psychiatry. 2007; 52: 37-45
        • Heinemann L.
        • Freckmann G.
        Quality of HbA1c measurement in the practice: the German perspective.
        J Diabetes Sci Technol. 2015; 9: 687-695
        • Rodgers L.R.
        • Weedon M.N.
        • Henley W.E.
        • Hattersley A.T.
        • Shields B.M.
        Cohort profile for the MASTERMIND study: Using the Clinical Practice Research Datalink (CPRD) to investigate stratification of response to treatment in patients with type 2 diabetes.
        BMJ Open. 2017; 7: e017989
        • Wargny M.
        • Gallini A.
        • Hanaire H.
        • Nourhashemi F.
        • Andrieu S.
        • Gardette V.
        Diabetes care and dementia among older adults: A nationwide 3-year longitudinal study.
        J Am Med Dir Assoc. 2018; 19: 601-606.e2
        • Thorpe C.T.
        • Thorpe J.M.
        • Kind A.J.
        • Bartels C.M.
        • Everett C.M.
        • Smith M.A.
        Receipt of monitoring of diabetes mellitus in older adults with comorbid dementia.
        J Am Geriatr Soc. 2012; 60: 644-651
        • McNabney M.K.
        • Pandya N.
        • Iwuagwu C.
        • et al.
        Differences in diabetes management of nursing home patients based on functional and cognitive status.
        J Am Med Dir Assoc. 2005; 6: 375-382
        • Piette J.D.
        • Kerr E.A.
        The impact of comorbid chronic conditions on diabetes care.
        Diabetes Care. 2006; 29: 725-731
        • Walraven I.
        • Mast M.R.
        • Hoekstra T.
        • et al.
        Distinct HbA1c trajectories in a type 2 diabetes cohort.
        Acta Diabetol. 2015; 52: 267-275
        • Laiteerapong N.
        • Karter A.J.
        • Moffet H.H.
        • et al.
        Ten-year hemoglobin A1c trajectories and outcomes in type 2 diabetes mellitus: The Diabetes & Aging Study.
        J Diabetes Complications. 2017; 31: 94-100
        • Luo M.
        • Lim W.Y.
        • Tan C.S.
        • et al.
        Longitudinal trends in HbA1c and associations with comorbidity and all-cause mortality in Asian patients with type 2 diabetes: A cohort study.
        Diabetes Res Clin Pract. 2017; 133: 69-77
        • Hemo B.
        • Geva D.
        • Shahar D.R.
        • Golan R.
        • Heymann A.D.
        Distinct trajectories in HbA1c are associated with different all-cause mortality and morbidity in newly diagnosed patients with type 2 diabetes.
        Primary Care Diabetes. 2020; 14: 413-419
        • Zaslavsky O.
        • Yu O.
        • Walker R.L.
        • et al.
        Incident dementia, glycated hemoglobin (HbA1c) levels and potentially preventable hospitalizations in people age 65 and older with diabetes.
        J Gerontol A Biol Sci Med Sci. 2021; 76: 2054-2061
        • Masuch A.
        • Friedrich N.
        • Roth J.
        • Nauck M.
        • Müller U.A.
        • Petersmann A.
        Preventing misdiagnosis of diabetes in the elderly: Age-dependent HbA1c reference intervals derived from two population-based study cohorts.
        BMC Endocr Disord. 2019; 19: 20
        • Dubowitz N.
        • Xue W.
        • Long Q.
        • et al.
        Aging is associated with increased HbA1c levels, independently of glucose levels and insulin resistance, and also with decreased HbA1c diagnostic specificity.
        Diabet Med. 2014; 31: 927-935
        • Clegg A.
        • Young J.
        • Iliffe S.
        • Rikkert M.O.
        • Rockwood K.
        Frailty in elderly people.
        Lancet. 2013; 381: 752-762
        • White H.
        • Pieper C.
        • Schmader K.
        The association of weight change in Alzheimer's disease with severity of disease and mortality: A longitudinal analysis.
        J Am Geriatr Soc. 1998; 46: 1223-1227
        • Morley J.E.
        Anorexia of aging: Physiologic and pathologic.
        Am J Clin Nutr. 1997; 66: 760-773
        • Abdelhafiz A.H.
        • Koay L.
        • Sinclair A.J.
        The effect of frailty should be considered in the management plan of older people with Type 2 diabetes.
        Future Sci OA. 2016; 2: FSO102
        • Punthakee Z.
        • Miller M.E.
        • Launer L.J.
        • et al.
        Poor cognitive function and risk of severe hypoglycemia in type 2 diabetes: Post hoc epidemiologic analysis of the ACCORD trial.
        Diabetes Care. 2012; 35: 787-793
        • de Galan B.E.
        • Zoungas S.
        • Chalmers J.
        • et al.
        Cognitive function and risks of cardiovascular disease and hypoglycaemia in patients with type 2 diabetes: the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial.
        Diabetologia. 2009; 52: 2328-2336
        • Sachs G.A.
        • Carter R.
        • Holtz L.R.
        • et al.
        Cognitive impairment: An independent predictor of excess mortality: A cohort study.
        Ann Intern Med. 2011; 155: 300-308
        • Garcia-Ptacek S.
        • Farahmand B.
        • Kareholt I.
        • et al.
        Mortality risk after dementia diagnosis by dementia type and underlying factors: A cohort of 15,209 patients based on the Swedish Dementia Registry.
        J Alzheimers Dis. 2014; 41: 467-477
        • Rait G.
        • Walters K.
        • Bottomley C.
        • Petersen I.
        • Iliffe S.
        • Nazareth I.
        Survival of people with clinical diagnosis of dementia in primary care: Cohort study.
        BMJ. 2010; 341: c3584
        • Chen L.
        • Reed C.
        • Happich M.
        • et al.
        Health care resource utilisation in primary care prior to and after a diagnosis of Alzheimer's disease: A retrospective, matched case-control study in the United Kingdom.
        BMC Geriatr. 2014; 14: 76
        • Fillenbaum G.
        • Heyman A.
        • Peterson B.L.
        • Pieper C.F.
        • Weiman A.L.
        Use and cost of hospitalization of patients with AD by stage and living arrangement: CERAD XXI.
        Neurology. 2001; 56: 201-206
        • Gallagher E.J.
        • Le Roith D.
        • Bloomgarden Z.
        Review of hemoglobin A(1c) in the management of diabetes.
        J Diabetes. 2009; 1: 9-17
        • Lipska K.J.
        • Ross J.S.
        • Miao Y.
        • et al.
        Potential overtreatment of diabetes mellitus in older adults with tight glycemic control.
        JAMA Intern Med. 2015; 175: 356-362
        • American Diabetes Association
        12. Older adults: Standards of Medical Care in Diabetes-2020.
        Diabetes Care. 2020; 43: S152-S162
        • International Diabetes Federation
        Managing Older People With Type 2 Diabetes: IDF Global Guideline.
        International Diabetes Federation, 2013