Clinical characteristics and healthcare utilisation associated with undiagnosed cognitive impairment in elderly patients with diabetes in a primary care setting: a population-based cohort study


  • A large number of patients screened with comprehensive information retrieved from the Leumit Health Services-electronic medical record.

  • A prospective analysis of health service utilisation outcomes.

  • A relatively short duration of follow-up after diagnosis (12 months).

  • Cognitive dysfunction was defined by a low Montreal Cognitive Assessment score, which may not fully describe all aspects of cognitive impairment in diabetes.


Diabetes is a known risk factor for the development of cognitive impairment and various forms of dementia. In a pooled analysis of 2.3 million people, diabetes increased the risk of dementia (both vascular and non-vascular dementia) by 60%.1 In the Rotterdam study which included 6330 participants, patients with type 2 diabetes mellitus (T2DM) had an OR of 1.3 (95% CI 1.0 to 1.9) for dementia.2 The use of insulin was identified as an additional risk factor. A large population-based longitudinal cohort study in the UK, including 10 095 participants with a median follow-up of 31.7 years found that the increased risk for dementia was associated with the duration of diabetes and a younger age of onset of the disease.3 Another population-based cohort study including 431 178 participants and conducted over 12 years in Taiwan identified the coexistence of diabetes-related complications as risk factors for dementia.4 While an increase in the diagnosis of both vascular dementia and Alzheimer’s disease is noted in association with diabetes, the pathologic changes characteristic of the latter (ie, neuritic plaques and neurofibrillary tangles) are not.5 Correspondingly, diabetes-related cognitive impairment has unique clinical characteristics. It is associated with a greater reduction in executive function, attention, processing and memory when compared with dementia in participants without diabetes.6 These may have direct deleterious effects on the ability of an individual with diabetes to adhere to complex treatment regimens and lifestyle care.

The prevalence of undiagnosed cognitive impairment in elderly patients with diabetes, the clinical characteristics of these patients and how it may be associated with clinical outcomes (specifically focusing on healthcare utilisation) are still not well reported. To partially address this knowledge gap, a screening initiative to identify elderly patients with T2DM and cognitive or behavioural impairments was conducted in Leumit Health Services (LHS; a large health maintenance organisation in Israel) between 1 June 2015 and 31 May 2018. Cognitive impairment was assessed using the Montreal Cognitive Assessment (MoCA) questionnaire.7 We report here the initial results of this survey and the characteristics and clinical outcomes of the survey participants as obtained from the LHS-electronic medical records (EMR). Moreover, we focus on the unique clinical features of patients with previously undiagnosed cognitive impairment when compared with patients with normal cognitive function.

Materials and methods

LHS serves approximately 720 000 individuals in Israel with all members having similar general health insurance and equal access to health services. LHS has a comprehensive computerised database, which is continuously updated and includes demographics, medical visits, laboratory tests, hospitalisations and medication prescriptions. Prescription records are available from 1998 and include those refilled and purchased per patient. Clinical diagnosis is established during each physician visit according to the International Classification of Diseases-9/10.8 The validity of diagnoses in the registry has been previously examined and confirmed as high.9

Between 1 June 2015 (baseline) and 31 May 2018, a screening initiative (supported by the Israeli Ministry of Health and ESHEL-Joint Distribution Committee) was conducted in LHS to identify elderly patients with diabetes and cognitive or behavioural impairments. During the screening period, community-dwelling patients who had a diagnosis of T2DM, age >65 years, and had no prior diagnosis of dementia (including the codes of the international classification of diseases, 10th edition: dementia in Alzheimer’s disease-F00, vascular dementia-F01, dementia in Pick’s disease-F02, unspecified dementia-F03, and Alzheimer’s disease-G30) in their EMR were actively invited to undergo an initial assessment using the MoCA questionnaire administered by trained registered nurse practitioners. As part of the screening initiative, primary-care teams were educated about the importance of identifying functional and psychological impairments among elderly patients with diabetes.

Using the LHS-EMR system, we conducted a population-based cohort study and identified all patients who underwent screening and had at least one recorded follow-up visit during the 12 months after the screening visit. Data collected included clinical characteristics, diabetes-related complications, laboratory results, purchased medications and health service utilisation in primary and hospital settings.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.


SES—socioeconomic status was defined according to the patient’s home address, using the Israeli Central Bureau of Statistics classification of 20 subgroups, where 1 is the lowest level and 20 is the highest.10 Education was defined as above or below 12 years referring to patients who obtained or did not obtain an education level higher than high school level education. Depression was diagnosed based on a score ≥10 obtained from the patient health questionnaire-9 (depression module). A diagnosis of diabetes-related complications (including ischaemic heart disease, cerebrovascular accident, heart failure, peripheral artery disease, neuropathy, retinopathy, dialysis and amputation or diabetic foot ulcer) was based on ICD-9 code diagnoses according to medical history and review of patient files. Diabetic nephropathy was defined as a baseline creatinine>1.5 mg/dL or albumin/creatinine ratio>30 mg/gr. We chose the Usual Provider of Care (UPC) index, as a measure of continuity of care.11 The usual provider care index was calculated as the ratio between the number of visits made to the main—primary care physician (PCP) and the number of total visits to any PCP during 12 months of follow-up. Nurse visits, foot examinations, endocrinologist/diabetologist visits, participation in a diabetes education workshop and participation in a smoking cessation workshop are presented as the number and per cent of patients with at least one event during 12 months of follow-up. A subject with an emergency room (ER) visit or hospitalisation was defined as having at least one event during 12 months of follow-up.

Statistical analysis

Statistical analysis was conducted using STATA V.12 software (StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX: StataCorp LP). Categorical variables were expressed as numbers (per cent). Continuous variables were expressed as mean±SD. Baseline variables included demographic data, pre-existing diabetes-related complications, laboratory test results obtained prior to screening, and medication purchased prior to screening (variables were considered as baseline variables if measured within 12 months before the cognitive tests were performed; a complete list of baseline variables is presented in table 1). Statistical testing was done only when at least four subjects exhibited the tested characteristic in any treatment group. Statistical analysis included descriptive statistics of the complete cohort and of three groups based on the screening MoCA score (MoCA>26, MoCA 19–26 and MoCA<19). Assumptions were two-sided with an α of 0.05. The initial analysis compared demographic and clinical characteristics using Student’s t-test and Fischer’s exact χ2 test for continuous and categorical variables, respectively, based on the normal distribution and variable characteristics. Differences between the groups with MoCA>26, MoCA 19–26 and MoCA<19 were analysed by one-way analysis of variance, and post-analysis was verified by Scheffé’s post-hoc test. The most recent measurement was used.

Table 1

Baseline patient characteristics, diabetes-related complications, laboratory results and medications*

Multivariable linear regression models were used to assess the association between numeric outcomes and having a previously undiagnosed cognitive impairment. Covariates included age, sex and SES. Numeric outcomes in the models included MoCA score, UPC, number of PCP visits, number of personal PCP visits, minutes spent with the personal physician or any physician during a visit, nurse visits, foot examinations conducted by nurses, ER visits, hospitalisations and days spent in the hospital during hospitalisation.

A multivariate logistic regression was used to estimate the ORs and 95% CI for the independent association between the following nonlinear clinical characteristics:

  • Model 1 controlling for multiple clinical and laboratory characteristics including age, sex, education >12 years, SES (low SES defined as the reference group), depression, last haemoglobin A1c (A1C), last serum creatinine, smoking, diabetic retinopathy, ischaemic heart disease, cerebrovascular accident, congestive heart failure and diabetic foot ulcer.

  • Model 2 controlling for multiple clinical characteristics, laboratory characteristics and medications, including age, sex, education >12 years, SES (low SES defined as the reference group), depression, UPC, last A1C, last serum creatinine, smoking, diabetic retinopathy, ischaemic heart disease, cerebrovascular accident, congestive heart failure, diabetic foot ulcer, and medication use (metformin, glucagon-like peptide-1 (GLP-1) receptor agonists, sodium-glucose transport protein 2 (SGLT-2) inhibitors, dipeptidyl peptidase-4 inhibitor (DPP-4) inhibitors, long-acting insulin, rapid-acting insulin, sulfonylureas, statins, fibrates, ezetimibe, ACE-inhibitors and calcium channel inhibitors).


Three hundred and fifty patients with diabetes and age ≥65 years were screened during the screening initiative (mean age 73.8±5.8 years, 157 (44.9%) women). Of these, 130 (37.1%) had a MoCA score greater than 26, 152 (43.4%) had a score between 19 and 26, and 68 (19.4%) had a MoCA score below 19 (table 1). Patients with a MoCA score below 19 had similar age, socioeconomic status, and body mass index but significantly lower levels of education >12 years, and higher prevalence of smoking and depression when compared with patients with a MoCA>26. They had more leg amputations or foot ulcers but other diabetes-related complications were similar between the groups. Glycaemic and lipid control was poorer in patients with a low MoCA score. Subjects with MoCA<19 had higher mean HbA1C, fasting plasma glucose, LDL cholesterol, triglycerides, creatinine and albumin-to-creatinine ratio levels, and lower HDL cholesterol. Medication use was different between subjects with a high versus a low MoCA score. The use of metformin, GLP-1 agonists, SGLT-2 inhibitors and DPP-4 agonists was less common, and the use of sulfonylureas, meglitinides, thiazolidinedione and insulin (both long and short-acting) was more common in the subjects with a low MoCA score.

Patients with a low MoCA score had fewer visits to their main PCP (7.3±4.2 vs 3.9±3.2, p=0.008; with a lower UPC index; 0.9±0.2 vs 0.5±0.4, p=0.001; table 2) and a shorter duration of PCP visits (8.3±4.5 vs 4.0±3.5, p=0.007). They had more yearly foot examinations (19 (14.6%) vs 20 (29.4%), p=0.002), but fewer nutritionist visits (40 (30.8%) vs 8 (11.8%),p=0.003), and endocrinologist/diabetologist visits (93 (71.54%) vs 25 (36.76%), p=0.001), and lower participation in diabetes education (23 (17.7%) vs 4 (5.9 %), p=0.001) or smoking cessation workshops (3 (30.0%) vs 1 (7.7%), p=0.033). Using a multiple linear regression model adjusting for age, sex and SES, a significant association between the baseline MoCA score and UPC (0.026±0.004, 95% CI 0.179 to 0.035, p=0.001), the number of visits to the main PCP (0.271±0.046, 95% CI 0.181 to 0.362, p=0.001), and the duration of visits to the main (2.871±0.497, 95% CI 1.894 to 3.849, p=0.001) or any PCP (0.362±0.05, 95% CI 0.264 to 0.461, p=0.001) was found (table 3).

Table 2

Health service utilisation during a 12-month follow-up period*

Table 3

The association between reduced baseline MoCA score and parameters of utilisation of outpatient care and inpatient care during a 12-month follow-up period*

Patients with a low MoCA score were more likely to have ER visits (15 (11.5%) vs 16 (23.5%), p=0.019) and hospitalisations (8 (6.2%) vs 22 (32.4%), p=0.001) at least once during the post-screening year of follow-up (table 2). They had more ER visits and hospitalisations per patient and if hospitalised, had a longer duration of hospitalisation (4.3±3.2 vs 14.5±9.8, p=0.001). A low MoCA score was associated with a significantly increased crude risk of an ER visit, hospitalisation, or long hospital stay (≥5 days).

To further examine the association between dementia and the risk of ER visits, hospitalisation, or long hospital stay, we conducted a multivariate logistic regression analysis with two models as described above (table 4, online supplemental tables 1 and 2). The odds of having ER visits, hospitalisation, or a long hospital stay were more significant when adjusting for multiple clinical and laboratory parameters in model 1 (table 4 and online supplemental table 1). However, when adding adjustments for concomitant medications and UPC in model 2, only a MoCA score below 19 was still associated with a significant risk for hospitalisation or a long hospital stay (table 4 and online supplemental table 2). Interestingly, in model 1, a diagnosis of diabetic retinopathy or diabetic foot ulcer was associated with an increased risk of hospitalisation and a longer hospital stay (online supplemental table 1). In model 2, a high UPC index and GLP-1 agonist use were associated with a lower risk of ER visits, and the use of SGLT-2 inhibitors was associated with a more significant risk (online supplemental table 2). The use of sulfonylureas or a diagnosis of diabetic retinopathy was strongly associated with a longer hospital stay.

Supplemental material

Table 4

The association between reduced baseline MoCA score and the probability of an ER visit, need for hospitalisation and long hospitalisation (>5 days) during a 12-month follow-up period*

Lastly, we analysed the association of MoCA as a continuous variable with parameters of the utilisation of inpatient care (table 3). Using a multiple linear regression model adjusted for age, sex and SES, we observed a significant association between MoCA, the number of hospitalisations (−0.035±0.009, 95% CI −0.053 to −0.018, p=0.001), and the length of hospital stays (−0.237±0.049, 95% CI −0.334 to −0.140, p=0.001).


In this study, we report the results of a screening initiative to detect impaired cognitive function and dementia in elderly patients with diabetes. Our findings suggest a very high prevalence of severe cognitive impairment (as depicted by a MoCA score—below 19) in almost 20% of elderly patients with diabetes who did not have a prior diagnosis of cognitive impairment or dementia.

In accordance with previous reports,12–14 patients with a low MoCA score were less educated, smoked more and had more depression. However, no association was noted with a low SES despite it being previously identified as a risk factor for cognitive impairment.15 This may be due to the fact that SES was determined according to residency, or of an inherent bias in Israeli society in which certain subpopulations, such as ultra-orthodox Jewish men, often do not work but rather commit to religious academic studies therefore belonging to a lower SES by choice of lifestyle rather than by circumstance.

Patients with severe cognitive impairment had more diabetes-related complications and worse glycaemic and lipid control. They were more likely to use insulin and sulfonylureas, thus exposing them to hypoglycaemia, and less likely to use SGLT-2 inhibitors and GLP-1 agonists and potentially benefit from the cardiorenal protection they exert (despite having reduced renal function).

Preventive interventions by the primary healthcare team and PCP visits were shorter and less frequent in patients with severe cognitive impairment. Using linear regression, we observed a significant association between a lower MoCA score, a lower UPC, and fewer visits to the main PCP. Moreover, lower MoCA was associated with a shorter duration of visits to both the main PCP and any PCP.

On the contrary, these patients were more likely to use hospital-based services with more frequent ER visits, hospitalisations and longer hospital stays if admitted. Using two logistic regression models adjusting for multiple clinical characteristics with and without medications and the UPC index and a linear regression model (addressing MoCA as a continuous parameter), we identify a low MoCA score as an independent risk factor for increased use of hospital-based treatment—especially more hospitalisations and longer hospital stays.

Similar screening initiatives among patients with diabetes have reported a high prevalence of cognitive impairment in patients with diabetes. A meta-analysis of observational studies suggested that up to 45% of patients with diabetes have mild cognitive impairment.16 Likewise, in a cross-sectional study in Mysuru, India, patients with diabetes had a lower mean MoCa score than patients without diabetes (18.99±0.48 vs 26.21±0.46). However, this study excluded elderly patients (>60 years) and did not exclude patients with a prior diagnosis of cognitive impairment.17 In contrast, our study is unique as we demonstrate a previously unreported very high prevalence of severe cognitive impairment in elderly patients with diabetes and without a previous diagnosis of cognitive impairment.

Moreover, our study suggests an association between severe cognitive impairment and quality and quantity of inpatient/outpatient care. While we do not know why patients with a low MoCA received poorer outpatient care (including a lower UPC index, shorter visits, less frequent use of newer diabetes- related medications, etc), we hypothesise that the lower uptake of SGLT-2 inhibitors and GLP-1 agonists may relate to an inherent bias of the PCPs or the patients against treatment regimen changes. Changing to these newer agents may be perceived as more complex (due to their unique adverse event profile and complex treatment regimen) and may therefore be avoided by the PCP or the patient. This in turn may lead to shorter visit duration—relying on previous treatment regimens and avoiding changes during visits. The reduced use of newer agents with proven cardiovascular and renal protective effects may contribute eventually to the observed increase in inpatient care. Other biases may exist and necessitate further study.

The strengths of this study include a large number of patients screened, the comprehensive information that was retrieved from the LHS-EMR, and the prospective nature of the health service utilisation outcomes. However, our analysis has several limitations including its descriptive nature, a lack of a more detailed understanding of the pre-existing conditions leading to cognitive impairment, the relatively short duration of follow-up after diagnosis (12 months), and the absence of actual morbidity and mortality data in this population. Furthermore, our definition of cognitive impairment was based only on the MoCA test, which may not fully describe all aspects of cognitive impairment in diabetes. The MoCa questionnaire was chosen as a screening tool since it is a relatively simple test to administer, available in local languages and was suggested to be a more sensitive tool for diagnosing cognitive impairment in older patients and patients with high risk for dementia or diabetes, when compared with other tests such as the standardised mini-mental state exam.18–20 Other tests such as the Hopkins verbal Test (HVLT),21 Addenbrooke’s Cognitive Examination III Learning Test (ACE-III),22 the Hayling Sentence Completion Test and the Brixton Spatial Anticipation Test,23 retinal microperimetry,24 Boston Naming Test,25 and complex attention tests (such as the Digit Span Test or the D2 Attentiveness Endurance Test),26 and so on, may have enhanced the screening sensitivity for assessing cognitive impairment while focusing on specific cognitive function domains beyond the capabilities of the MoCA test (such as executive function, memory, lexical access, etc). However, these were not used as these could not be used as part of a busy clinic visit (in which cognitive screening was performed in order to adapt a diabetes treatment plan accordingly). In addition, some are not available in local languages or lack validation in the current population. Moreover, when compared with the MoCA test as a single screening test, these tests have several limitation; the HVLT focuses on the memory domain without examining other domains, the ACE-III covers fewer cognitive domains and takes twice as long to administer as the MoCA test,27 Retinal microperimetry may be affected by retinopathy and necessitates special equipment,24 while other tests focus on a single domain—such as executive function with the Hayling and Brixton tests and lexical access with the Boston naming test (this test is also affected by bilingualism28 which is very common in the Israeli society).

In conclusion, this study highlights the very high prevalence of undiagnosed cognitive impairment in elderly patients with diabetes. Moreover, it emphasises the importance of identifying cognitive impairment as a risk factor for poor outpatient care and increased utilisation of hospital services. Our results call for a prospective interventional study to examine the efficacy of actively augmenting outpatient care in patients with a low-MoCA score in an effort to prevent ER visits and hospitalisations. Nevertheless, increasing our vigilance in diagnosing cognitive impairment in older patients with diabetes through the routine implementation of active screening assessments such as the MoCA test is crucial in this high-risk population.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information. No additional data are available.

Ethics statements

Patient consent for publication

Ethics approval

The study protocol was approved by the Institutional Review Board of the Shamir Medical Center and the Research Committee of the LHS, approval number- LEU-0010–22.

This post was originally published on