There were 1288 subjects who passed the preliminary screening, expressed interest in participating, were approached by a study physician and signed informed consent. Of them, 282 (21.1%) were excluded from the study due to incompatibility with eligibility criteria (based on physician assessment) and 109 (8.5%) refused to continue their participation in the study, so 897 subjects remained active participants. The study consists of 835 subjects who had complete data on sociodemographic, cardiovascular, and diabetes-related covariates as well as GDS score and had at least 2 HbA1c measurements. Table 1 describes the sample characteristics and compares the HbA1c trend categories using analysis of variance for continuous outcomes, and Pearson’s chi square for dichotomous outcomes. The mean age at entry into the IDCD was 72.75 (4.63) years, with a majority of males (60%). The average levels of total, HDL, and LDL cholesterol, diastolic blood pressure, and GFR were consistent with the general elderly population in this age range. Systolic blood pressure was higher than in the general population but similar to that observed in other T2D samples [22]. Most subjects were treated with oral anti-diabetic medications (79%), 1% were treated with insulin only, 9% were treated with a combined therapy of oral anti-diabetic medications and insulin and 11% were controlled by diet only. The mean number of HbA1c measurements per subject was 17.9 (SD = 9.6). Mean duration of inclusion in the DR was 8.70 (2.64) years, mean HbA1c levels at entry into the DR and at follow up were 6.95% (1.32) and 6.96% (1.01) respectively.

The PROC TRAJ analysis identified six trajectories, with each characterized by (1) the HbA1c level at entry into the DR (intercept), and (2) the trend over time in the registry (figure 1). Three types of trends (stable, increasing and decreasing) in HbA1c over time were observed, and within each type there were two groups with nearly parallel trends but with different intercepts (HbA1c levels at entry to the Registry). For convenience purposes, the groups are referred to by (1) whether it has a Higher or Lower intercept within each type of trend (the terms higher and lower were used for descriptive purposes rather than for clinical purposes, in order to differentiate, within each trend in HbA1c, those who entered with a relatively higher from those who entered with a relatively lower HbA1c into the DR), and (2) the type of trend in HbA1c over time, either Stable, Increasing or Decreasing. Groups differed in number of years in the DR (longest duration for the groups whose HbA1c decreased, followed by the groups whose HbA1c increased and finally by the groups that remained stable); HbA1c levels at baseline was highest for the groups whose HbA1c decreased, followed by the groups whose HbA1c increased and lowest for the groups that remained stable; mean HbA1c during follow up was highest for the Higher Decreasing and Higher Increasing groups and lowest for the two groups that remained stable. Differences were found for anti-diabetic medications use such that the Higher Decreasing group had the lowest use of oral anti-diabetics only and the highest use of both oral anti-diabetics and insulin. Additional differences were observed for age at entry into the IDCD, LDL cholesterol, and diastolic blood pressure.

The overall models were significant for overall cognition, semantic categorization, and executive functions (see Table 2). For descriptive purposes, Table 3 presents comparisons among the different trajectories. Mean overall cognitive z-score was significantly lower in the Higher Decreasing group compared to all other groups except Higher Increasing- in which the difference approached significance. Similarly, for semantic categorization, mean z-scores were lower for the Higher Decreasing group compared to all other groups (approaching significance for the Lower Decreasing group). The Lower Decreasing group had poorer z-scores than the other groups except for the Higher Increasing group. Scores for executive function were lower for Higher Decreasing compared to all other groups except the Higher Increasing group which by itself had poorer function compared to the two stable groups. The mean z-score for the Lower Increasing group was significantly lower than that of the Lower Stable group and approached significance compared to the Higher Stable group (tables 2 and 3). Finally, no significant differences were observed among groups in the episodic memory and attention/working memory domains.

After applying the Bonferroni-Holms step down correction to account for multiple comparisons, the following pairwise comparisons remained significant: for overall cognition, the comparison of the Higher Decreasing group to the Lower stable group (p = 0.0021) and to the Higher Stable group (p = 0.0029), for semantic categorization, the comparison of the Higher Decreasing group to the Lower Increasing group (p = 0.0031) and for executive function, the comparison of the Higher Decreasing group with the Lower Stable group (0.0015).

A secondary analysis demonstrated that higher mean HbA1c was associated with lower scores in overall cognitive score (p = 0.01) and executive functions (p = 0.0003) and with a trend for lower scores in semantic categorization (p = 0.06). Higher standard deviation in HbA1c was associated with lower scores in executive function (p = 0.025) and a trend towards lower scores in overall cognitive score (p = 0.07).

Exclusion of the handful of cases in which there was a gap between the CDR score and the MMSE scores did not affect the results.

The study was approved by the IRB committees of the Sheba Medical Center, Israel, the Maccabi Health Services (MHS), Israel and the Icahn School of Medicine at Mount Sinai, NY. Informed consent was signed by all study participants.

This study consisted of elderly (≥65 years old) T2D subjects in the IDCD, a collaboration between the Icahn School of Medicine at Mount Sinai, NY, the Sheba Medical Center, Israel, and the Maccabi Health Services (MHS), Israel. Detailed methods have been presented [29]. IDCD subjects were randomly selected from the approximately 11,000 T2D individuals aged ≥65 years that are in the Diabetes Registry (DR) of the MHS, the second largest HMO in Israel. The MHS DR is an integral part of the MHS Electronic Patient Record system, which was established in 1998 to facilitate disease management and to improve treatment. The DR has collected detailed laboratory, medication, and diagnoses information since 1998 [30]. The present analysis assesses the relationship of long term trajectories in HbA1c, since the subject’s entry into the MHS DR diabetes registry (the earliest time was 1998, when the diabetes registry was established) until the IDCD baseline cognitive evaluation (2010–2011).

Any of the following criteria should be met in order to be included in the MHS DR: 1) HbA1c >7.25%, 2) Glucose >200 mg/dl on two exams more than three months apart, 3) purchase of anti-diabetic medication twice within three months supported by a HbA1c >6.5% or Glucose >125 mg/dl within half a year, 4) diagnosis of T2D (ICD9 code) by a general practitioner, internist, endocrinologist, ophthalmologist, or T2D advisor, supported by a HbA1c >6.5% or Glucose >125 mg/dl within half a year. These criteria have been validated by twenty physicians in MHS against their own practice records [30]. Additionally, age specific prevalence rates were similar to those of a DR of another large HMO in Israel²⁸.

(1) In the MHS DR, i.e. diagnosed with T2D (2) lives in the central area of Israel (near Tel Aviv), (3) 65 years of age or above, (4) cognitively normal (not suffering from dementia or MCI) at entry into the IDCD (based on a multidisciplinary consensus conference), (5) two or more HbA1c measurements in the DR, (6) has an informant, and (7) speaks Hebrew fluently. Potential subjects were excluded if they had an ICD code for dementia or its subtypes, treatment with prescribed cholinesterase inhibitors, or had a major psychiatric or neurological condition (such as schizophrenia, stroke or Parkinson’s disease) that could affect cognitive performance.

An algorithm including the eligibility criteria that were available in the MHS Electronic Patient Record was used to randomly select the subjects. After random selection of subjects, letters were sent by MHS to the primary care physicians, asking for permission to contact each patient regarding the study. If the doctor agreed, a letter was sent to the patient briefly describing the study and saying he or she would be contacted by phone in the following two weeks. The study coordinator then called the patient and invited participation in the study, after determining fluency in Hebrew and that there was an informant willing to provide information about the subject’s health. The social structure of Israel is such that most elderly individuals live with or near their extended family, so few are excluded for this reason. A second informant was sought to be a replacement if necessary. Subjects were assessed in two phases, typically at their residence, or at the Sheba Medical Center memory clinic, according to their preference. In the first meeting, a study physician obtained signed informed consent; performed medical, neurological, geriatric and nutritional (Food Frequency Questionnaire- FFQ) assessments; and drew blood for inflammatory markers (Il-6, CRP), and haptoglobin and ApoE genotypes. The second meeting, conducted within two weeks after the physician’s assessment, involved a neuropsychologist administering to the subject a cognitive battery, and both to the subject and an informant questionnaires for cognitive and functional impairment, and for depression and behavioral disturbances characteristic of dementia.

The present analysis examined the association between long term trends in glycemic control (using all HbA1c measurements present in the subjects’ DR from their entry into the registry until IDCD initiation) with cognitive outcomes based on cognitive assessment that was performed at entry into the IDCD (baseline cognitive assessment). The cognitive assessment includes the scales and neuropsychological tests described below and takes approximately 2 hours to be administered. The study is ongoing and follow-up cognitive assessments have recently begun and do not enable, at this phase, report of the association of trajectories of glycemic control and changes in cognition over time.

A factor analysis with varimax rotation of the comprehensive neuropsychological battery of the IDCD subjects was used to identify cognitive domains in this study. Four domains were identified. For each domain, a summary was calculated as the sum of z-scores (test scores transformed to mean zero and standard deviation one, reversed if necessary for positive values to indicate good cognition) of tests with high loadings: episodic memory (immediate and delayed recall, and recognition word list), semantic categorization [39] (letter and category fluency, and similarities), attention/working memory (diamond cancellation test, digit span forward and backward), and executive functions (Trails making A and B and the DSST). Finally, an overall cognition measure was calculated by summing the four domain summaries.

Sociodemographic factors (age, sex, and years of education) were collected at entry into the IDCD. Cardiovascular covariates (glomerular filtration rate calculated by the MDRD formula, total cholesterol, HDL, LDL, diastolic and systolic blood pressure) were defined as means of all measurements available in the DR. Diabetes-related covariates were the estimated duration of T2D (based on time in the DR) and diabetes medications (which was categorized as oral anti-diabetic medications only, insulin only, combination of oral anti-diabetic and insulin and no medication). Since depression has been associated both with T2D [42], degree of glycemic control [43], risk for T2D- related complications [44] and with dementia [45], we also included the geriatric depression scale (GDS) score as a potential confounder. HbA1c levels at entry into the DR were defined as the third HbA1c recorded in the registry after excluding the first two HbA1c measurements, since these might reflect instability of glycemic control prior to diagnosis of T2D and treatment initiation.

We identified distinctive trajectory groups of HbA1c using a SAS macro named PROC TRAJ [46]. This approach applies a multinomial modeling strategy to identify relatively homogenous clusters of developmental trajectories within a sample population, that is, the modeling strategy allows for the emergence of more than two trajectories. Trajectory parameters are derived by latent class analysis using maximum likelihood estimation. In particular, the distinctive trajectories of HbA1c were derived by modeling HbA1c as a function of the number of years in the DR. Quadratic curves were used to model the trends over time. The number of trajectories was determined using the guidelines suggested by Jones et al [46]. The output of PROC TRAJ includes the equations for the different trajectories along with the assignment of each patient to one of the trajectory groups. This group assignment was then used in an analysis of covariance to estimate and compare mean cognitive domain z-scores among the trajectory groups while adjusting for sociodemographic, cardiovascular, diabetes-related covariates (years in the DR and anti-diabetic medications), and GDS score.

In order to provide analyses that are comparable with those of other studies [47], we assessed the relationships of mean and standard deviations of HbA1c with cognitive function in secondary analyses, using the GLM procedure.