https://orcid.org/0000-0002-7030-0442
Abstract
Previous studies have shown hypoglycaemia to be associated with an increased risk of dementia; however, there are several design challenges to consider. The objective of this study is to assess the association between hypoglycaemia and dementia while addressing these challenges using a lag period, exposure density sampling (EDS) and inverse probability of treatment weighting (IPTW).
This was a population-based cohort using data (1996–2018) from British Columbia, Canada. From a cohort of incident type 2 diabetes patients aged 40–70 years, we created a dynamic sub-cohort of hypoglycaemia-exposed (≥1 episode requiring hospitalization or a physician visit) and unexposed individuals using EDS, in which four unexposed individuals per one exposed were randomly selected into risk sets based on diabetes duration and age. Follow-up was until dementia diagnosis, death, emigration or 31 December 2018. Those diagnosed with dementia within 2 years of follow-up were censored. We adjusted for confounding using IPTW and estimated the hazard ratio (HR, 95% CI) of dementia using weighted conditional cause-specific hazards risk models with death as a competing risk.
Among 13 970 patients with incident type 2 diabetes, 2794 experienced hypoglycaemia. There were 329 dementia events over a median (interquartile range: IQR) follow-up of 5.03 (5.7) years. IPTW resulted in well-balanced groups with weighted incidence rates (95% CI) of 4.59 (3.52, 5.98)/1000 person-years among exposed and 3.33 (2.58, 3.88)/1000 person-years among unexposed participants. The risk of dementia was higher among those with hypoglycaemia (HR, 1.83; 95% CI 1.31, 2.57).
After addressing several methodological challenges, we showed that hypoglycaemia contributes to an increased risk of all-cause dementia among patients with type 2 diabetes.
Among patients with type 2 diabetes, the risk of all-cause dementia is over 80% higher among individuals who experienced at least one hypoglycaemic episode, compared with those who did not.
The use of exposure density sampling, high-dimensional propensity scores, inverse probability of treatment weighting, and a range of lag periods was utilized to minimize threats to validity.
These findings add insight on the modifiable risk factors of dementia in type 2 diabetes.
Introduction
Hypoglycaemia is an acute complication of diabetes that often occurs as an adverse effect of exogenous insulin or other medications that increase endogenous insulin secretion, such as sulphonylureas.1 Although most hypoglycaemic episodes are mild and can be managed independently by patients, the InHypo-DM survey study found that approximately 38% of participants with type 2 diabetes had reported at least one hypoglycaemic event that required the assistance of a third party to administer treatment with glucose or glucagon.2 Additionally, an estimated 1% of type 2 diabetes patients treated with oral antihyperglycaemics and 7% of those treated with insulin experience at least one severe hypoglycaemic episode that requires an emergency department visit in their lifetime.3 Although these estimates might be low, the global proportion of hypoglycaemia-related deaths is 4.49/1000 of total diabetes deaths.4,5
Glucose is the brain’s primary source of energy, and the reduction of glucose supply from the peripheral circulation to the brain can negatively affect cognitive function.6 Although cognitive function is often restored when glucose supply is normalized, it has been hypothesized that severe hypoglycaemic episodes can lead to platelet aggregation, fibrinogen formation and irreversible damage, including neuronal cell death.6–14 The ACCORD-MIND trial showed that severe hypoglycaemic attacks were not associated with increases in brain atrophy or abnormal white matter volume which is indicative of damage.15 However, these findings are limited, given only 40 months of follow-up and a sample of 500 individuals.15
Given that hypoglycaemia is not amenable to randomization and a long-follow up period is necessary to capture dementia, the weight of evidence assessing the association between hypoglycaemia and dementia is from observational studies. In fact, findings from previous observational studies, assessing this association using different data sources, design elements, follow-up periods and populations, have consistently shown a higher risk of dementia, with relative risk estimates ranging between 1.20 and 4.4016–22; albeit one study with a small sample size and an older population at baseline did not support such findings.23
Although robust methodology is critical for all observational studies, some design considerations are more critical when assessing the complex association between hypoglycaemia and dementia. First, the relationship between hypoglycaemia and dementia appears to be bidirectional, and patients with unrecognized cognitive impairment may be more susceptible to severe hypoglycaemia. 9,24 Despite that, some studies did not consider a possible lag period to account for this reverse causality.16–20 That is, these studies did not require a specific censoring period of dementia events occurring after the hypoglycaemia to account for hypoglycaemia due to prodromal dementia. Additionally, a latency period between exposure to hypoglycaemia and the development of dementia needs to be considered. Second, hypoglycaemia is an adverse effect of diabetes therapies; therefore, exposure to hypoglycaemia should be time-dependent, wherein most exposed individuals start as unexposed. Whereas some studies did consider hypoglycaemia as a time-dependent exposure,16–19,21,22 the covariate assessment period was often at diabetes diagnosis or study enrolment. Therefore, groups might be imbalanced on several confounding variables at the time of hypoglycaemic episodes. For example at diabetes diagnosis, groups might be well-balanced on diabetes duration, therapy and complications; however, this balance is not necessarily maintained at the time of hypoglycaemia. In fact, those with more severe diabetes over time are more likely to receive insulin and therefore more likely to experience hypoglycaemia. Third, most previous studies only adjusted for a limited number of important confounders despite the complexity of the relationship, thus suffering from residual confounding.16,17,22
Herein, we use multiple design approaches to emulate a hypothetical trial and combat the aforementioned threats to validity, to assess the association between hypoglycaemic episodes and all-cause dementia using real-world data. Specifically, we use a lag period, exposure density sampling (EDS) and high-dimensional propensity scores with inverse-probability of treatment weights (IPTW) for confounding adjustment.
Methods
Study design and data source
This was a retrospective population-based cohort study using British Columbia’s (BC) health care data from 1 January 1996 to 31 December 2018, obtained from the administrative databases within Population Data BC [https://www.popdata.bc.ca/data]. This repository captures the encounters with the health care system for nearly all of BC’s population who receive universal health care coverage through the provincial government.25–29 These data have been validated and used extensively in health services research.30–34
We linked data across multiple databases using a de-identified personal health identification number. Several databases were used as follows.
The population registry (Consolidation File) to capture date of birth, sex and dates of health care coverage.25
The PharmaNet programme included drug dispensation date, name, drug identification number (DIN) and quantity.26 This database captures all prescription drugs dispensed by community pharmacies to BC residents regardless of the type of insurance coverage (government-sponsored, private or out-of-pocket), comprehensively capturing non-hospital drug use. The provincial Pharmacare programe provides complete coverage of eligible medications for residents after an income-based deductible has been met during the fiscal year.
The Medical Services Plan (MSP) database provided data on physician visits, including the service date and the International Classification of Diseases, 9th Revision [Clinical Modification] (ICD-9-CM) diagnosis code.27
The Discharge Abstract Database provided hospital admission and discharge dates and several diagnoses coded with ICD-10-Canadian Adaptation (CA) codes.28
The Vital Events Deaths database provided the date of death.29
We also acquired an area-level measure of socioeconomic status (SES) based on the first three characters of the postal code, and aggregated neighbourhood-level income data from Census Geodata.35
Study population
First, we identified a cohort of patients newly diagnosed with type 2 diabetes between 1 January 1998 through 31 December 2016. Using a washout period of 2 years, incident diabetes was defined based on the validated diabetes case-defining algorithm from the Canadian Chronic Disease Surveillance System, whereby diabetes is defined as the earliest occurrence of two physician claims (ICD-9 codes) or one hospitalization (ICD-10-CA) for diabetes within a 2-year period.36 This definition has a 89.3% sensitivity [95% confidence interval (CI) 88.9, 89.9], 97.6% specificity (95% CI 97.5, 97.7), 81.9% positive predictive value (95% CI 81.3, 82.4) and 98.7% negative predictive value (95% CI 98.6, 98.7).36,37
Second, we applied the following inclusion criteria:
aged between 40–70 years at the date of diabetes onset, with the lower limit set to allow enough follow-up time to capture incident dementia, and the upper limit set to account for a possible period of prodromal dementia and delayed diagnosis;
continuous registration in the population registry for at least 2 years prior to diabetes onset;
no receipt of any anti-hyperglycaemic agents prior to diabetes onset;
no presence of a diagnostic code indicating type 1 diabetes at any time or receipt of insulin monotherapy as first-line treatment;
no previous record of diagnostic codes indicating dementia or any cognitive impairment or a dispensation record for a cholinesterase inhibitor before diabetes diagnosis;
no diagnosis of Down’s syndrome, due to the high risk of diabetes and dementia in Down’s syndrome with genetic variation that we were unable to assess.
ICD codes used to identify diabetes and inclusion criteria are reported in Supplementary Table S1 (available as Supplementary data at IJE online).
Exposure definition and exposure density sampling
Hypoglycaemia was defined as at least one hospitalization or a physician claim indicating hypoglycaemia. The date of the first hypoglycaemic episode was defined as the index date. ICD codes used to identify hypoglycaemia are reported in Supplementary Table S1.
Since nearly all diabetes patients start as unexposed to hypoglycaemia, we used EDS with replacement to create a dynamic sub-cohort.38 EDS, a technique of dynamic matching at the time of exposure, allows for the estimation of the effect of a time-dependent exposure with minimal loss in precision and improved interpretability of the exposure effect, when compared with a full cohort analysis.39 Importantly, time-dependent bias, which can lead to an underestimation of risk in a standard survival analysis with exposure as a time-dependent covariate, is avoided.39,40
Specifically, we randomly selected four unexposed individuals for each exposed individual within risk sets based on diabetes duration and age. Those selected as unexposed to hypoglycaemia (controls) at one point were eligible to be exposed (cases) in the future (Figure 1). Index date was the date of hypoglycaemia for those exposed and the date equivalent to that in diabetes duration for those unexposed (Figure 2). The latest index date allowed was 31 December 2016, to allow for a minimum follow-up time of 2 years to capture dementia based on the validated algorithm used.
Illustration of exposure density sampling. Patients A and C meet the exposure criteria and are considered exposed. Patients C–E can be selected as unexposed for patient A. Patient E only can be selected as unexposed for patient C
Illustration of study design, including cohort entry, index date, covariate assessment period, and lag period
Outcome definition
Incident all-cause dementia was defined using a validated algorithm that requires one hospitalization code, three physician claims codes (at least 30 days apart in a 2-year period) or a prescription filled for a cholinesterase inhibitor.41 This definition has 79.3% sensitivity, 99.1% specificity, 80.4% positive predictive value and 99.0% negative predictive value.41 The outcome was restricted to all-cause dementia, due to difficulty in ascertaining subtypes using administrative data.41 ICD codes used to identify all-cause dementia are reported in Supplementary Table S1.
To account for a dementia latency period and minimize possible reverse causality (i.e. hypoglycaemic episodes due to prodromal cognitive impairment prior to dementia diagnosis), a lag period between exposure and the development of dementia was required (Figure 2). Specifically, those who received a dementia diagnosis within 2 years of index date were censored. This approach has been used previously to minimize reverse causality or protopathic bias in multiple observational studies assessing the risk of dementia.42–45
Confounding mitigation
First, we used the high dimensional propensity score (hdps) algorithm to identify relevant potential confounders based on five dimensions (hospitalizations, procedures, medical diagnoses, medical services and prescription medication claims) during the year before index date (Figure 2).46 We identified the 200 most prevalent variables in each dimension and ranked them according to their frequency as once, sporadic or frequent. Then, we selected the top 500 variables for inclusion in the model to estimate the propensity score, in addition to a list of 43 predefined variables, including:
demographic variables [age, sex and socioeconomic status (SES), defined as quintiles based on an area‐level measure of SES based on the first three characters of the postal code and aggregated neighbourhood-level income data35];
indicators of health care use [number of distinct medications dispensed, hospitalizations, physician visits];
indicators of diabetes severity such as macrovascular complications [ischaemic heart disease, heart failure, hypertension, dyslipidaemia, stroke, and peripheral vascular disease], microvascular complications [nephropathy, neuropathy, and retinopathy], anti-hyperglycaemic agents [metformin, sulphonylureas, thiazolidinediones, glucagon-like peptide receptor agonists (GLP1-RA), dipeptidyl peptidase-4 (DPP-4) inhibitors, sodium glucose co-transporter-2 (SGLT-2) inhibitors, insulins, meglitinides, and acarbose] and treatment for macrovascular complications [angiotensin-converting enzyme (ACE) inhibitors, angiotensin receptor blockers (ARBs), loop diuretics, thiazide diuretics, beta blockers, calcium channel blockers (CCB) and other antihypertensives, including methyldopa, hydralazine, and alpha blockers];
other morbidities [Parkinson disease, Huntington’s disease, delirium, and anxiety/mood disorder];
other prescription drug use [antidepressants, antipsychotics, opioids, migraine medications, Parkinson medications, and antacids];
index year.
A multivariable logistic regression model was used to estimate the likelihood of experiencing a hypoglycaemic episode.42 Then, propensity scores were used to compute the inverse probability of treatment weight (IPTW) to balance possible confounding variables.47 We used stabilized weights, as they are preferred over raw weights to reduce the variance associated with any extreme weights.48 No further truncation of weights was needed. Last, balance of baseline covariates after weighting was assessed using absolute standardized differences (ASD), with ASD >10% considered as an imbalance.49 Since we used EDS, individuals may appear more than once with different index dates. For those, hdps and IPTW were updated at each appearance.
Statistical analysis
Patients were followed from index date until the date of dementia diagnosis, death, emigration, end of provincial health coverage or end of study period (31 December 2018), whichever occurred first. A conditional weighted cause-specific hazards model with death as a competing risk was used to estimate a hazard ratio (HR) and 95% CI of dementia associated with the hypoglycaemic event.50,51 Model assumptions including the proportional hazards assumption were tested using Schoenfeld residuals.52 Two additional models were run, wherein we added interaction terms between the exposure variable and biological sex (female and male) or SES (quintiles) to assess for any effect modification. We further addressed the possible impact of the introduction of a government‐sponsored reimbursement policy for cholinesterase inhibitors in October 2007, which has affected the number of physician visits with a diagnosis of Alzheimer's disease in BC. Specifically, we created and adjusted for a ‘before/after’ variable to indicate if follow-up ended before or after October 2007.53 We used robust variance (sandwich estimator) to calculate a confidence interval for all models.
Secondary and sensitivity analyses
As a secondary analysis we repeated the primary analysis within a high-risk population. Specifically, we restricted the population to those using diabetes medications that have a high risk of inducing hypoglycaemia (sulphonylureas, meglitinides or insulin). Those who experienced a hypoglycaemic event before the initiation of any of these medications were excluded.
We also conducted several sensitivity analyses, wherein we varied the age of the included population, the exposure definition and the lag period. First, we repeated the primary analysis using a cohort of patients aged 50–60 years with incident diabetes. Increasing the lower age limit at cohort entry requires less follow-up time to capture incident dementia, and decreasing the upper limit helps account for a longer prodromal period or delayed diagnoses. Second, hypoglycaemic episodes that result in a hospitalization are more clinically severe than those reported in a physician visit; therefore, we changed the exposure definition wherein we stratified the hypoglycaemia composite exposure definition to either hypoglycaemia captured solely from hospitalizations or solely captured from physician claims. Third, we used six lag periods (1 year, 3–7 years), wherein those diagnosed with dementia were censored. Last, we calculated the E-value to quantify the minimum strength of amount of association between an unmeasured confounder, such as smoking, and the exposure/outcome for unmeasured confounding, to explain away the main result.54,55
All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC).
Results
Primary analysis
There was a total of 286 006 individuals with incident type 2 diabetes between 1998 and 2016, of whom 278 812 met the inclusion criteria (Figure 3). Our dynamic sub-cohort comprised 13 970 dementia-free patients, of whom 2794 experienced a serious hypoglycaemic episode between January 1998 and December 2016. There was a total of 116 all-cause dementia events over a median (IQR) follow-up period of 5.01 (5.55) years among those who experienced a hypoglycaemic episode, and 213 events over 5.07 (6.53) years among those who did not experience any hypoglycaemic episodes. The unadjusted incidence rate (95% CI) of dementia was 7.19 (6.00, 8.60) per 1000 person-years for those exposed and 3.21 (2.80, 3.67) per 1000 person-years for those unexposed. Before any confounding mitigation, the risk of all cause-dementia was more than twice as high among those who experienced hypoglycaemia compared with those who did not (crude HR, 2.73; 95% CI 2.12, 2.57).
Flow chart of the cohort study. a) Based on the Canadian Chronic Disease Surveillance System (one hospitalization or two physician claims within 2 years), without a code indicating type 1 diabetes mellitus. b) At any time before diabetes diagnosis with a minimum of 2 years; patients may belong to more than one group. ChEIs, cholinesterase inhibitors
Before ITPW, those who experienced a hypoglycaemic episode were more likely to be on multiple medications and more likely to be on certain medications such as insulin, antidepressants and opioids (Table 1). Additionally, they were more likely to have a more severe diabetes, as indicated by the higher frequency of several diabetes complications (Table 1). IPTW resulted in well-balanced groups across all the included potential confounders (Table 1). The mean age (SD) was 62.97 (8.38) years for those exposed and 63.48 (9.53) for those unexposed. Distribution of other sociodemographic characteristics, such as sex and SES as well as several clinical characteristics, were also well-balanced (Table 1). Although diabetes duration was no longer well-balanced after weighting with absolute standardized difference >10%, those exposed to hypoglycaemia had a shorter duration (∼8 months) compared with unexposed. Therefore, confounding by diabetes duration, a possible indicator of diabetes severity, is towards the null. The weighted incidence rate (95% CI) of all-cause dementia was 4.59 (3.52, 5.98) per 1000 person-years for those exposed and 3.33 (2.58, 3.88) per 1000 person-years for those unexposed (Table 2).
Baseline characteristics of the exposure groups before and after inverse probability of treatment weighting
| Characteristic | Before weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Hypoglycaemia | No hypoglycaemia | ASD | Hypoglycaemia | No hypoglycaemia | ASD | |
| Age, years, mean (SD) | 63.25 (9.37) | 63.25 (9.36) | <0.001 | 62.97 (8.38) | 63.48 (9.53) | 0.058 |
| Female, n (%) | 1279 (45.78) | 5042 (45.11) | 0.020 | 1037.14 (45.41) | 5225.54 (45.04) | <0.001 |
| Diabetes duration, years, mean (SD) | 6.68 (4.69) | 6.68 (4.69) | 0.000 | 6.28 (4.09) | 6.96 (4.86) | 0.152 |
| Socioeconomic status quintile, n (%) | ||||||
| 1 (highest) | 773 (27.67) | 2425 (21.70) | 0.243 | 502.98 (22.02) | 2877.26 (24.80) | 0.093 |
| 2 | 653 (23.37) | 2417 (21.63) | 528.22 (23.13) | 2549.36 (21.98) | ||
| 3 | 521 (18.65) | 2278 (20.38) | 454.89 (19.92) | 2243.08 (19.34) | ||
| 4 | 433 (15.50) | 2063 (18.46) | 397.92 (17.42) | 2003.77 (17.27) | ||
| 5 (lowest) | 348 (12.46) | 1862 (16.66) | 350.77 (15.36) | 1779.81 (15.34) | ||
| Missing | 66 (2.36) | 131 (1.17) | 49.15 (2.15) | 147.65 (1.27) | ||
| Health care utilization | ||||||
| Number of hospitalizations in year before index date, n (%) | ||||||
| Zero | 1630 (58.34) | 8860 (79.28) | 0.509 | 1645.33 (72.04) | 8408.39 (72.48) | <0.001 |
| One | 558 (19.97) | 1545 (13.82) | 385.08 (16.86) | 1934.07 (16.67) | ||
| Two | 295 (10.56) | 495 (4.43) | 149.62 (6.55) | 847.39 (7.30) | ||
| Three or more | 311 (11.13) | 276 (2.47) | 103.89 (4.55) | 411.08 (3.54) | ||
| Number of physician visits in year before index date, n (%) | ||||||
| Zero | S | 284 (2.54) | 0.324 | 33.82 (1.48) | 229.58 (1.98) | 0.135 |
| One | S | 91 (0.81) | 14.29 (0.63) | 76.75 (0.66) | ||
| Two | S | 92 (0.82) | 10.11 (0.44) | 76.69 (0.66) | ||
| Three or more | 2782 (99.57) | 10709 (95.82) | 2225.7 (97.45) | 11217.90 (96.70) | ||
| Number of distinct drugs in year before index date, n (%) | ||||||
| Zero | 51 (1.83) | 631 (5.65) | 0.351 | 92.49 (4.05) | 539.59 (4.65) | 0.048 |
| One | 47 (1.68) | 606 (5.42) | 97.79 (4.28) | 5220.54 (4.49) | ||
| Two | 67 (2.40) | 684 (6.12) | 104.01 (4.55) | 597.20 (5.15) | ||
| Three or more | 2629 (94.09) | 9255 (82.81) | 1989.64 (87.12) | 9943.6 (85.71) | ||
| Comorbidities in year before index date, n (%) | ||||||
| Parkinson disease | 17 (0.61) | 45 (0.40) | 0.029 | 12.66 (0.55) | 47.09 (0.41) | 0.021 |
| Huntington’s disease | 0 (0) | S | 0.019 | 0 | S | 0.016 |
| Delirium | 91 (3.26) | 32 (0.29) | 0.227 | 22.61 (0.99) | 99.24 (0.86) | 0.014 |
| Anxiety/mood disorder | 1673 (59.88) | 3536 (31.64) | 0.591 | 954.70 (41.80) | 4556.19 (39.27) | 0.051 |
| Hypertension | 1136 (40.66) | 4193 (37.52) | 0.064 | 880.26 (38.54) | 4606.45 (39.71) | 0.024 |
| Ischaemic heart disease | 563 (20.15) | 1376 (12.31) | 0.021 | 33.13 (14.59) | 1653.94 (14.26) | 0.009 |
| Dyslipidaemia | 401 (14.35) | 1321 (11.82) | 0.075 | 286.00 (12.52) | 1394.91 (12.02) | 0.015 |
| Heart failure | 332 (11.88) | 418 (3.74) | 0.307 | 130.57 (5.72) | 795.13 (6.85) | 0.047 |
| Stroke | 172 (6.16) | 297 (2.66) | 0.171 | 86.65 (3.97) | 494.14 (4.26) | 0.024 |
| Nephropathy | 400 (14.32) | 630 (5.64) | 0.293 | 183.62 (8.04) | 936.96 (8.08) | 0.001 |
| Neuropathy | 128 (4.58) | 181 (1.62) | 0.172 | 61.71 (2.70) | 234.35 (2.02) | 0.045 |
| Retinopathy | 116 (4.15) | 200 (1.79) | 0.139 | 55.38 (2.42) | 272.92 (2.35) | 0.005 |
| Peripheral vascular disease | 367 (13.14) | 262 (2.34) | 0.412 | 130.28 (5.70) | 761.69 (6.57) | 0.036 |
| Use of medications in year before or on index date, n (%) | ||||||
| Antidepressants | 844 (30.21) | 2072 (18.54) | 0.274 | 554.06 (24.26) | 2471.79 (21.31) | 0.070 |
| Antipsychotics | 955 (34.18) | 2085 (18.66) | 0.358 | 541.63 (23.71) | 2787.43 (24.03) | 0.007 |
| Opioids | 1097 (39.26) | 2468 (22.08) | 0.379 | 624.07 (27.32) | 2947.77 (25.41) | 0.043 |
| Migraine medications | 39 (1.40) | 79 (0.79) | 0.068 | 17.10 (0.75) | 85.13 (0.73) | 0.002 |
| Parkinson’s medications | 60 (2.15) | 135 (1.21) | 0.073 | 38.87 (1.70) | 151.75 (1.31) | 0.036 |
| Antacids | 1093 (39.12) | 2530 (22.64) | 0.363 | 630.42 (27.60) | 3189.60 (27.49) | 0.002 |
| Metformin | 1756 (62.85) | 4998 (44.72) | 0.370 | 1167.16 (51.10) | 5804.37 (50.03) | 0.021 |
| Sulphonylurea | 1363 (48.78) | 1909 (17.08) | 0.716 | 590.45 (25.85) | 3062.32 (26.40) | 0.012 |
| Thiazolidinedione | 132 (4.72) | 262 (2.34) | 0.129 | 66.75 (2.92) | 356.84 (3.08) | 0.009 |
| GLP1-RA | 40 (1.43) | 131 (1.17) | 0.023 | 26.56 (1.16) | 145.18 (1.25) | 0.008 |
| DPP-4 inhibitor | 144 (5.15) | 504 (4.51) | 0.030 | 113.36 (4.96) | 543.39 (4.68) | 0.013 |
| SGLT-2 inhibitor | 30 (1.07) | 206 (1.84) | 0.064 | 20.32 (0.89) | 208.76 (1.80) | 0.079 |
| Insulin | 560 (20.04) | 588 (5.26) | 0.456 | 221.42 (9.69) | 1436.76 (12.38) | 0.086 |
| Meglitinides | 19 (0.68) | 52 (0.47) | 0.028 | 12.83 (0.56) | 56.05 (0.48) | 0.011 |
| Acarbose | 21 (0.75) | 47 (0.42) | 0.043 | 8.89 (0.39) | 49.29 (0.42) | 0.005 |
| Statins | 1476 (52.83) | 5232 (46.81) | 0.120 | 1092.01 (47.81) | 5664.86 (48.83) | 0.020 |
| ACE inhibitors | 1262 (45.17) | 3907 (34.96) | 0.209 | 851.31 (37.27) | 4407.62 (37.90) | 0.015 |
| ARBs | 495 (17.72) | 1976 (17.68) | <0.001 | 403.15 (17.65) | 2059.23 (17.75) | 0.003 |
| Loop diuretics | 475 (17.00) | 611 (5.47) | 0.371 | 187.74 (8.22) | 1016.69 (8.76) | 0.019 |
| Thiazide diuretics | 543 (19.43) | 1912 (17.11) | 0.060 | 436.39 (19.11) | 1964.53 (16.93) | 0.057 |
| Beta blockers | 766 (27.42) | 2191 (19.60) | 0.185 | 487.28 (21.34) | 2500.06 (21.55) | 0.005 |
| CCB | 656 (23.48) | 2071 (18.53) | 0.122 | 431.81 (18.91) | 2474.37 (21.33) | 0.060 |
| Other antihypertensives | 91 (3.26) | 150 (1.34) | 0.128 | 41.33 (1.81) | 231.41 (1.99) | 0.014 |
| Characteristic | Before weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Hypoglycaemia | No hypoglycaemia | ASD | Hypoglycaemia | No hypoglycaemia | ASD | |
| Age, years, mean (SD) | 63.25 (9.37) | 63.25 (9.36) | <0.001 | 62.97 (8.38) | 63.48 (9.53) | 0.058 |
| Female, n (%) | 1279 (45.78) | 5042 (45.11) | 0.020 | 1037.14 (45.41) | 5225.54 (45.04) | <0.001 |
| Diabetes duration, years, mean (SD) | 6.68 (4.69) | 6.68 (4.69) | 0.000 | 6.28 (4.09) | 6.96 (4.86) | 0.152 |
| Socioeconomic status quintile, n (%) | ||||||
| 1 (highest) | 773 (27.67) | 2425 (21.70) | 0.243 | 502.98 (22.02) | 2877.26 (24.80) | 0.093 |
| 2 | 653 (23.37) | 2417 (21.63) | 528.22 (23.13) | 2549.36 (21.98) | ||
| 3 | 521 (18.65) | 2278 (20.38) | 454.89 (19.92) | 2243.08 (19.34) | ||
| 4 | 433 (15.50) | 2063 (18.46) | 397.92 (17.42) | 2003.77 (17.27) | ||
| 5 (lowest) | 348 (12.46) | 1862 (16.66) | 350.77 (15.36) | 1779.81 (15.34) | ||
| Missing | 66 (2.36) | 131 (1.17) | 49.15 (2.15) | 147.65 (1.27) | ||
| Health care utilization | ||||||
| Number of hospitalizations in year before index date, n (%) | ||||||
| Zero | 1630 (58.34) | 8860 (79.28) | 0.509 | 1645.33 (72.04) | 8408.39 (72.48) | <0.001 |
| One | 558 (19.97) | 1545 (13.82) | 385.08 (16.86) | 1934.07 (16.67) | ||
| Two | 295 (10.56) | 495 (4.43) | 149.62 (6.55) | 847.39 (7.30) | ||
| Three or more | 311 (11.13) | 276 (2.47) | 103.89 (4.55) | 411.08 (3.54) | ||
| Number of physician visits in year before index date, n (%) | ||||||
| Zero | S | 284 (2.54) | 0.324 | 33.82 (1.48) | 229.58 (1.98) | 0.135 |
| One | S | 91 (0.81) | 14.29 (0.63) | 76.75 (0.66) | ||
| Two | S | 92 (0.82) | 10.11 (0.44) | 76.69 (0.66) | ||
| Three or more | 2782 (99.57) | 10709 (95.82) | 2225.7 (97.45) | 11217.90 (96.70) | ||
| Number of distinct drugs in year before index date, n (%) | ||||||
| Zero | 51 (1.83) | 631 (5.65) | 0.351 | 92.49 (4.05) | 539.59 (4.65) | 0.048 |
| One | 47 (1.68) | 606 (5.42) | 97.79 (4.28) | 5220.54 (4.49) | ||
| Two | 67 (2.40) | 684 (6.12) | 104.01 (4.55) | 597.20 (5.15) | ||
| Three or more | 2629 (94.09) | 9255 (82.81) | 1989.64 (87.12) | 9943.6 (85.71) | ||
| Comorbidities in year before index date, n (%) | ||||||
| Parkinson disease | 17 (0.61) | 45 (0.40) | 0.029 | 12.66 (0.55) | 47.09 (0.41) | 0.021 |
| Huntington’s disease | 0 (0) | S | 0.019 | 0 | S | 0.016 |
| Delirium | 91 (3.26) | 32 (0.29) | 0.227 | 22.61 (0.99) | 99.24 (0.86) | 0.014 |
| Anxiety/mood disorder | 1673 (59.88) | 3536 (31.64) | 0.591 | 954.70 (41.80) | 4556.19 (39.27) | 0.051 |
| Hypertension | 1136 (40.66) | 4193 (37.52) | 0.064 | 880.26 (38.54) | 4606.45 (39.71) | 0.024 |
| Ischaemic heart disease | 563 (20.15) | 1376 (12.31) | 0.021 | 33.13 (14.59) | 1653.94 (14.26) | 0.009 |
| Dyslipidaemia | 401 (14.35) | 1321 (11.82) | 0.075 | 286.00 (12.52) | 1394.91 (12.02) | 0.015 |
| Heart failure | 332 (11.88) | 418 (3.74) | 0.307 | 130.57 (5.72) | 795.13 (6.85) | 0.047 |
| Stroke | 172 (6.16) | 297 (2.66) | 0.171 | 86.65 (3.97) | 494.14 (4.26) | 0.024 |
| Nephropathy | 400 (14.32) | 630 (5.64) | 0.293 | 183.62 (8.04) | 936.96 (8.08) | 0.001 |
| Neuropathy | 128 (4.58) | 181 (1.62) | 0.172 | 61.71 (2.70) | 234.35 (2.02) | 0.045 |
| Retinopathy | 116 (4.15) | 200 (1.79) | 0.139 | 55.38 (2.42) | 272.92 (2.35) | 0.005 |
| Peripheral vascular disease | 367 (13.14) | 262 (2.34) | 0.412 | 130.28 (5.70) | 761.69 (6.57) | 0.036 |
| Use of medications in year before or on index date, n (%) | ||||||
| Antidepressants | 844 (30.21) | 2072 (18.54) | 0.274 | 554.06 (24.26) | 2471.79 (21.31) | 0.070 |
| Antipsychotics | 955 (34.18) | 2085 (18.66) | 0.358 | 541.63 (23.71) | 2787.43 (24.03) | 0.007 |
| Opioids | 1097 (39.26) | 2468 (22.08) | 0.379 | 624.07 (27.32) | 2947.77 (25.41) | 0.043 |
| Migraine medications | 39 (1.40) | 79 (0.79) | 0.068 | 17.10 (0.75) | 85.13 (0.73) | 0.002 |
| Parkinson’s medications | 60 (2.15) | 135 (1.21) | 0.073 | 38.87 (1.70) | 151.75 (1.31) | 0.036 |
| Antacids | 1093 (39.12) | 2530 (22.64) | 0.363 | 630.42 (27.60) | 3189.60 (27.49) | 0.002 |
| Metformin | 1756 (62.85) | 4998 (44.72) | 0.370 | 1167.16 (51.10) | 5804.37 (50.03) | 0.021 |
| Sulphonylurea | 1363 (48.78) | 1909 (17.08) | 0.716 | 590.45 (25.85) | 3062.32 (26.40) | 0.012 |
| Thiazolidinedione | 132 (4.72) | 262 (2.34) | 0.129 | 66.75 (2.92) | 356.84 (3.08) | 0.009 |
| GLP1-RA | 40 (1.43) | 131 (1.17) | 0.023 | 26.56 (1.16) | 145.18 (1.25) | 0.008 |
| DPP-4 inhibitor | 144 (5.15) | 504 (4.51) | 0.030 | 113.36 (4.96) | 543.39 (4.68) | 0.013 |
| SGLT-2 inhibitor | 30 (1.07) | 206 (1.84) | 0.064 | 20.32 (0.89) | 208.76 (1.80) | 0.079 |
| Insulin | 560 (20.04) | 588 (5.26) | 0.456 | 221.42 (9.69) | 1436.76 (12.38) | 0.086 |
| Meglitinides | 19 (0.68) | 52 (0.47) | 0.028 | 12.83 (0.56) | 56.05 (0.48) | 0.011 |
| Acarbose | 21 (0.75) | 47 (0.42) | 0.043 | 8.89 (0.39) | 49.29 (0.42) | 0.005 |
| Statins | 1476 (52.83) | 5232 (46.81) | 0.120 | 1092.01 (47.81) | 5664.86 (48.83) | 0.020 |
| ACE inhibitors | 1262 (45.17) | 3907 (34.96) | 0.209 | 851.31 (37.27) | 4407.62 (37.90) | 0.015 |
| ARBs | 495 (17.72) | 1976 (17.68) | <0.001 | 403.15 (17.65) | 2059.23 (17.75) | 0.003 |
| Loop diuretics | 475 (17.00) | 611 (5.47) | 0.371 | 187.74 (8.22) | 1016.69 (8.76) | 0.019 |
| Thiazide diuretics | 543 (19.43) | 1912 (17.11) | 0.060 | 436.39 (19.11) | 1964.53 (16.93) | 0.057 |
| Beta blockers | 766 (27.42) | 2191 (19.60) | 0.185 | 487.28 (21.34) | 2500.06 (21.55) | 0.005 |
| CCB | 656 (23.48) | 2071 (18.53) | 0.122 | 431.81 (18.91) | 2474.37 (21.33) | 0.060 |
| Other antihypertensives | 91 (3.26) | 150 (1.34) | 0.128 | 41.33 (1.81) | 231.41 (1.99) | 0.014 |
ASD, absolute standardized difference; GLP1-RA, glucagon-like peptide-1 receptor agonist; DPP-4, dipeptidyl-peptidase 4; SGLT, sodium-glucose cotransporter; ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; CCB, calcium channel blocker; S, suppressed number <5 as per data provider requirements to ensure patient confidentiality/health privacy is maintained.
Baseline characteristics of the exposure groups before and after inverse probability of treatment weighting
| Characteristic | Before weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Hypoglycaemia | No hypoglycaemia | ASD | Hypoglycaemia | No hypoglycaemia | ASD | |
| Age, years, mean (SD) | 63.25 (9.37) | 63.25 (9.36) | <0.001 | 62.97 (8.38) | 63.48 (9.53) | 0.058 |
| Female, n (%) | 1279 (45.78) | 5042 (45.11) | 0.020 | 1037.14 (45.41) | 5225.54 (45.04) | <0.001 |
| Diabetes duration, years, mean (SD) | 6.68 (4.69) | 6.68 (4.69) | 0.000 | 6.28 (4.09) | 6.96 (4.86) | 0.152 |
| Socioeconomic status quintile, n (%) | ||||||
| 1 (highest) | 773 (27.67) | 2425 (21.70) | 0.243 | 502.98 (22.02) | 2877.26 (24.80) | 0.093 |
| 2 | 653 (23.37) | 2417 (21.63) | 528.22 (23.13) | 2549.36 (21.98) | ||
| 3 | 521 (18.65) | 2278 (20.38) | 454.89 (19.92) | 2243.08 (19.34) | ||
| 4 | 433 (15.50) | 2063 (18.46) | 397.92 (17.42) | 2003.77 (17.27) | ||
| 5 (lowest) | 348 (12.46) | 1862 (16.66) | 350.77 (15.36) | 1779.81 (15.34) | ||
| Missing | 66 (2.36) | 131 (1.17) | 49.15 (2.15) | 147.65 (1.27) | ||
| Health care utilization | ||||||
| Number of hospitalizations in year before index date, n (%) | ||||||
| Zero | 1630 (58.34) | 8860 (79.28) | 0.509 | 1645.33 (72.04) | 8408.39 (72.48) | <0.001 |
| One | 558 (19.97) | 1545 (13.82) | 385.08 (16.86) | 1934.07 (16.67) | ||
| Two | 295 (10.56) | 495 (4.43) | 149.62 (6.55) | 847.39 (7.30) | ||
| Three or more | 311 (11.13) | 276 (2.47) | 103.89 (4.55) | 411.08 (3.54) | ||
| Number of physician visits in year before index date, n (%) | ||||||
| Zero | S | 284 (2.54) | 0.324 | 33.82 (1.48) | 229.58 (1.98) | 0.135 |
| One | S | 91 (0.81) | 14.29 (0.63) | 76.75 (0.66) | ||
| Two | S | 92 (0.82) | 10.11 (0.44) | 76.69 (0.66) | ||
| Three or more | 2782 (99.57) | 10709 (95.82) | 2225.7 (97.45) | 11217.90 (96.70) | ||
| Number of distinct drugs in year before index date, n (%) | ||||||
| Zero | 51 (1.83) | 631 (5.65) | 0.351 | 92.49 (4.05) | 539.59 (4.65) | 0.048 |
| One | 47 (1.68) | 606 (5.42) | 97.79 (4.28) | 5220.54 (4.49) | ||
| Two | 67 (2.40) | 684 (6.12) | 104.01 (4.55) | 597.20 (5.15) | ||
| Three or more | 2629 (94.09) | 9255 (82.81) | 1989.64 (87.12) | 9943.6 (85.71) | ||
| Comorbidities in year before index date, n (%) | ||||||
| Parkinson disease | 17 (0.61) | 45 (0.40) | 0.029 | 12.66 (0.55) | 47.09 (0.41) | 0.021 |
| Huntington’s disease | 0 (0) | S | 0.019 | 0 | S | 0.016 |
| Delirium | 91 (3.26) | 32 (0.29) | 0.227 | 22.61 (0.99) | 99.24 (0.86) | 0.014 |
| Anxiety/mood disorder | 1673 (59.88) | 3536 (31.64) | 0.591 | 954.70 (41.80) | 4556.19 (39.27) | 0.051 |
| Hypertension | 1136 (40.66) | 4193 (37.52) | 0.064 | 880.26 (38.54) | 4606.45 (39.71) | 0.024 |
| Ischaemic heart disease | 563 (20.15) | 1376 (12.31) | 0.021 | 33.13 (14.59) | 1653.94 (14.26) | 0.009 |
| Dyslipidaemia | 401 (14.35) | 1321 (11.82) | 0.075 | 286.00 (12.52) | 1394.91 (12.02) | 0.015 |
| Heart failure | 332 (11.88) | 418 (3.74) | 0.307 | 130.57 (5.72) | 795.13 (6.85) | 0.047 |
| Stroke | 172 (6.16) | 297 (2.66) | 0.171 | 86.65 (3.97) | 494.14 (4.26) | 0.024 |
| Nephropathy | 400 (14.32) | 630 (5.64) | 0.293 | 183.62 (8.04) | 936.96 (8.08) | 0.001 |
| Neuropathy | 128 (4.58) | 181 (1.62) | 0.172 | 61.71 (2.70) | 234.35 (2.02) | 0.045 |
| Retinopathy | 116 (4.15) | 200 (1.79) | 0.139 | 55.38 (2.42) | 272.92 (2.35) | 0.005 |
| Peripheral vascular disease | 367 (13.14) | 262 (2.34) | 0.412 | 130.28 (5.70) | 761.69 (6.57) | 0.036 |
| Use of medications in year before or on index date, n (%) | ||||||
| Antidepressants | 844 (30.21) | 2072 (18.54) | 0.274 | 554.06 (24.26) | 2471.79 (21.31) | 0.070 |
| Antipsychotics | 955 (34.18) | 2085 (18.66) | 0.358 | 541.63 (23.71) | 2787.43 (24.03) | 0.007 |
| Opioids | 1097 (39.26) | 2468 (22.08) | 0.379 | 624.07 (27.32) | 2947.77 (25.41) | 0.043 |
| Migraine medications | 39 (1.40) | 79 (0.79) | 0.068 | 17.10 (0.75) | 85.13 (0.73) | 0.002 |
| Parkinson’s medications | 60 (2.15) | 135 (1.21) | 0.073 | 38.87 (1.70) | 151.75 (1.31) | 0.036 |
| Antacids | 1093 (39.12) | 2530 (22.64) | 0.363 | 630.42 (27.60) | 3189.60 (27.49) | 0.002 |
| Metformin | 1756 (62.85) | 4998 (44.72) | 0.370 | 1167.16 (51.10) | 5804.37 (50.03) | 0.021 |
| Sulphonylurea | 1363 (48.78) | 1909 (17.08) | 0.716 | 590.45 (25.85) | 3062.32 (26.40) | 0.012 |
| Thiazolidinedione | 132 (4.72) | 262 (2.34) | 0.129 | 66.75 (2.92) | 356.84 (3.08) | 0.009 |
| GLP1-RA | 40 (1.43) | 131 (1.17) | 0.023 | 26.56 (1.16) | 145.18 (1.25) | 0.008 |
| DPP-4 inhibitor | 144 (5.15) | 504 (4.51) | 0.030 | 113.36 (4.96) | 543.39 (4.68) | 0.013 |
| SGLT-2 inhibitor | 30 (1.07) | 206 (1.84) | 0.064 | 20.32 (0.89) | 208.76 (1.80) | 0.079 |
| Insulin | 560 (20.04) | 588 (5.26) | 0.456 | 221.42 (9.69) | 1436.76 (12.38) | 0.086 |
| Meglitinides | 19 (0.68) | 52 (0.47) | 0.028 | 12.83 (0.56) | 56.05 (0.48) | 0.011 |
| Acarbose | 21 (0.75) | 47 (0.42) | 0.043 | 8.89 (0.39) | 49.29 (0.42) | 0.005 |
| Statins | 1476 (52.83) | 5232 (46.81) | 0.120 | 1092.01 (47.81) | 5664.86 (48.83) | 0.020 |
| ACE inhibitors | 1262 (45.17) | 3907 (34.96) | 0.209 | 851.31 (37.27) | 4407.62 (37.90) | 0.015 |
| ARBs | 495 (17.72) | 1976 (17.68) | <0.001 | 403.15 (17.65) | 2059.23 (17.75) | 0.003 |
| Loop diuretics | 475 (17.00) | 611 (5.47) | 0.371 | 187.74 (8.22) | 1016.69 (8.76) | 0.019 |
| Thiazide diuretics | 543 (19.43) | 1912 (17.11) | 0.060 | 436.39 (19.11) | 1964.53 (16.93) | 0.057 |
| Beta blockers | 766 (27.42) | 2191 (19.60) | 0.185 | 487.28 (21.34) | 2500.06 (21.55) | 0.005 |
| CCB | 656 (23.48) | 2071 (18.53) | 0.122 | 431.81 (18.91) | 2474.37 (21.33) | 0.060 |
| Other antihypertensives | 91 (3.26) | 150 (1.34) | 0.128 | 41.33 (1.81) | 231.41 (1.99) | 0.014 |
| Characteristic | Before weighting | After weighting | ||||
|---|---|---|---|---|---|---|
| Hypoglycaemia | No hypoglycaemia | ASD | Hypoglycaemia | No hypoglycaemia | ASD | |
| Age, years, mean (SD) | 63.25 (9.37) | 63.25 (9.36) | <0.001 | 62.97 (8.38) | 63.48 (9.53) | 0.058 |
| Female, n (%) | 1279 (45.78) | 5042 (45.11) | 0.020 | 1037.14 (45.41) | 5225.54 (45.04) | <0.001 |
| Diabetes duration, years, mean (SD) | 6.68 (4.69) | 6.68 (4.69) | 0.000 | 6.28 (4.09) | 6.96 (4.86) | 0.152 |
| Socioeconomic status quintile, n (%) | ||||||
| 1 (highest) | 773 (27.67) | 2425 (21.70) | 0.243 | 502.98 (22.02) | 2877.26 (24.80) | 0.093 |
| 2 | 653 (23.37) | 2417 (21.63) | 528.22 (23.13) | 2549.36 (21.98) | ||
| 3 | 521 (18.65) | 2278 (20.38) | 454.89 (19.92) | 2243.08 (19.34) | ||
| 4 | 433 (15.50) | 2063 (18.46) | 397.92 (17.42) | 2003.77 (17.27) | ||
| 5 (lowest) | 348 (12.46) | 1862 (16.66) | 350.77 (15.36) | 1779.81 (15.34) | ||
| Missing | 66 (2.36) | 131 (1.17) | 49.15 (2.15) | 147.65 (1.27) | ||
| Health care utilization | ||||||
| Number of hospitalizations in year before index date, n (%) | ||||||
| Zero | 1630 (58.34) | 8860 (79.28) | 0.509 | 1645.33 (72.04) | 8408.39 (72.48) | <0.001 |
| One | 558 (19.97) | 1545 (13.82) | 385.08 (16.86) | 1934.07 (16.67) | ||
| Two | 295 (10.56) | 495 (4.43) | 149.62 (6.55) | 847.39 (7.30) | ||
| Three or more | 311 (11.13) | 276 (2.47) | 103.89 (4.55) | 411.08 (3.54) | ||
| Number of physician visits in year before index date, n (%) | ||||||
| Zero | S | 284 (2.54) | 0.324 | 33.82 (1.48) | 229.58 (1.98) | 0.135 |
| One | S | 91 (0.81) | 14.29 (0.63) | 76.75 (0.66) | ||
| Two | S | 92 (0.82) | 10.11 (0.44) | 76.69 (0.66) | ||
| Three or more | 2782 (99.57) | 10709 (95.82) | 2225.7 (97.45) | 11217.90 (96.70) | ||
| Number of distinct drugs in year before index date, n (%) | ||||||
| Zero | 51 (1.83) | 631 (5.65) | 0.351 | 92.49 (4.05) | 539.59 (4.65) | 0.048 |
| One | 47 (1.68) | 606 (5.42) | 97.79 (4.28) | 5220.54 (4.49) | ||
| Two | 67 (2.40) | 684 (6.12) | 104.01 (4.55) | 597.20 (5.15) | ||
| Three or more | 2629 (94.09) | 9255 (82.81) | 1989.64 (87.12) | 9943.6 (85.71) | ||
| Comorbidities in year before index date, n (%) | ||||||
| Parkinson disease | 17 (0.61) | 45 (0.40) | 0.029 | 12.66 (0.55) | 47.09 (0.41) | 0.021 |
| Huntington’s disease | 0 (0) | S | 0.019 | 0 | S | 0.016 |
| Delirium | 91 (3.26) | 32 (0.29) | 0.227 | 22.61 (0.99) | 99.24 (0.86) | 0.014 |
| Anxiety/mood disorder | 1673 (59.88) | 3536 (31.64) | 0.591 | 954.70 (41.80) | 4556.19 (39.27) | 0.051 |
| Hypertension | 1136 (40.66) | 4193 (37.52) | 0.064 | 880.26 (38.54) | 4606.45 (39.71) | 0.024 |
| Ischaemic heart disease | 563 (20.15) | 1376 (12.31) | 0.021 | 33.13 (14.59) | 1653.94 (14.26) | 0.009 |
| Dyslipidaemia | 401 (14.35) | 1321 (11.82) | 0.075 | 286.00 (12.52) | 1394.91 (12.02) | 0.015 |
| Heart failure | 332 (11.88) | 418 (3.74) | 0.307 | 130.57 (5.72) | 795.13 (6.85) | 0.047 |
| Stroke | 172 (6.16) | 297 (2.66) | 0.171 | 86.65 (3.97) | 494.14 (4.26) | 0.024 |
| Nephropathy | 400 (14.32) | 630 (5.64) | 0.293 | 183.62 (8.04) | 936.96 (8.08) | 0.001 |
| Neuropathy | 128 (4.58) | 181 (1.62) | 0.172 | 61.71 (2.70) | 234.35 (2.02) | 0.045 |
| Retinopathy | 116 (4.15) | 200 (1.79) | 0.139 | 55.38 (2.42) | 272.92 (2.35) | 0.005 |
| Peripheral vascular disease | 367 (13.14) | 262 (2.34) | 0.412 | 130.28 (5.70) | 761.69 (6.57) | 0.036 |
| Use of medications in year before or on index date, n (%) | ||||||
| Antidepressants | 844 (30.21) | 2072 (18.54) | 0.274 | 554.06 (24.26) | 2471.79 (21.31) | 0.070 |
| Antipsychotics | 955 (34.18) | 2085 (18.66) | 0.358 | 541.63 (23.71) | 2787.43 (24.03) | 0.007 |
| Opioids | 1097 (39.26) | 2468 (22.08) | 0.379 | 624.07 (27.32) | 2947.77 (25.41) | 0.043 |
| Migraine medications | 39 (1.40) | 79 (0.79) | 0.068 | 17.10 (0.75) | 85.13 (0.73) | 0.002 |
| Parkinson’s medications | 60 (2.15) | 135 (1.21) | 0.073 | 38.87 (1.70) | 151.75 (1.31) | 0.036 |
| Antacids | 1093 (39.12) | 2530 (22.64) | 0.363 | 630.42 (27.60) | 3189.60 (27.49) | 0.002 |
| Metformin | 1756 (62.85) | 4998 (44.72) | 0.370 | 1167.16 (51.10) | 5804.37 (50.03) | 0.021 |
| Sulphonylurea | 1363 (48.78) | 1909 (17.08) | 0.716 | 590.45 (25.85) | 3062.32 (26.40) | 0.012 |
| Thiazolidinedione | 132 (4.72) | 262 (2.34) | 0.129 | 66.75 (2.92) | 356.84 (3.08) | 0.009 |
| GLP1-RA | 40 (1.43) | 131 (1.17) | 0.023 | 26.56 (1.16) | 145.18 (1.25) | 0.008 |
| DPP-4 inhibitor | 144 (5.15) | 504 (4.51) | 0.030 | 113.36 (4.96) | 543.39 (4.68) | 0.013 |
| SGLT-2 inhibitor | 30 (1.07) | 206 (1.84) | 0.064 | 20.32 (0.89) | 208.76 (1.80) | 0.079 |
| Insulin | 560 (20.04) | 588 (5.26) | 0.456 | 221.42 (9.69) | 1436.76 (12.38) | 0.086 |
| Meglitinides | 19 (0.68) | 52 (0.47) | 0.028 | 12.83 (0.56) | 56.05 (0.48) | 0.011 |
| Acarbose | 21 (0.75) | 47 (0.42) | 0.043 | 8.89 (0.39) | 49.29 (0.42) | 0.005 |
| Statins | 1476 (52.83) | 5232 (46.81) | 0.120 | 1092.01 (47.81) | 5664.86 (48.83) | 0.020 |
| ACE inhibitors | 1262 (45.17) | 3907 (34.96) | 0.209 | 851.31 (37.27) | 4407.62 (37.90) | 0.015 |
| ARBs | 495 (17.72) | 1976 (17.68) | <0.001 | 403.15 (17.65) | 2059.23 (17.75) | 0.003 |
| Loop diuretics | 475 (17.00) | 611 (5.47) | 0.371 | 187.74 (8.22) | 1016.69 (8.76) | 0.019 |
| Thiazide diuretics | 543 (19.43) | 1912 (17.11) | 0.060 | 436.39 (19.11) | 1964.53 (16.93) | 0.057 |
| Beta blockers | 766 (27.42) | 2191 (19.60) | 0.185 | 487.28 (21.34) | 2500.06 (21.55) | 0.005 |
| CCB | 656 (23.48) | 2071 (18.53) | 0.122 | 431.81 (18.91) | 2474.37 (21.33) | 0.060 |
| Other antihypertensives | 91 (3.26) | 150 (1.34) | 0.128 | 41.33 (1.81) | 231.41 (1.99) | 0.014 |
ASD, absolute standardized difference; GLP1-RA, glucagon-like peptide-1 receptor agonist; DPP-4, dipeptidyl-peptidase 4; SGLT, sodium-glucose cotransporter; ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; CCB, calcium channel blocker; S, suppressed number <5 as per data provider requirements to ensure patient confidentiality/health privacy is maintained.
Risk estimates of all-cause dementia associated with exposure to hypoglycaemia from all analyses
| Exposure | No. of patients | No. of events | Median follow-up in years (IQR) | Crude incidence ratea (95% CI) | Weighted incidence ratea (95% CI) | Crude HR (95% CI) | Weighted HRb (95% CI) | Adjusted HRc (95% CI) |
|---|---|---|---|---|---|---|---|---|
| Primary analysis | ||||||||
| Hypoglycaemia | 2794 | 116 | 5.01 (5.55) | 7.19 (6.00, 8.60) | 4.59 (3.52, 5.98) | 2.73 (2.12, 2.57) | 1.83 (1.31, 2.57) | 1.78 (1.27, 2.49) |
| No hypoglycaemia | 11176 | 213 | 5.07 (6.35) | 3.21 (2.80, 3.67) | 3.33 (2.58, 3.88) | 1.00 | 1.00 | 1.00 |
| Secondary analysis (high-risk population) | ||||||||
| Hypoglycaemia | 1824 | 90 | 4.48 (5.21) | 9.25 (7.54, 11.34) | 7.78 (4.51, 11.34) | 2.73 (2.08, 3.57) | 1.98 (1.41, 2.78) | 1.96 (1.39, 2.77) |
| No hypoglycaemia | 7296 | 212 | 5.48 (6.67) | 4.57 (3.97, 5.27) | 4.95 (4.18, 5.86) | 1.00 | 1.00 | 1.00 |
| Sensitivity analyses | ||||||||
| Population restricted to those with incident type 2 diabetes aged 50, 60 years | ||||||||
| Hypoglycaemia | 1111 | 38 | 5.13 (5.69) | 5.83 (4.25, 7.98) | 3.18 (1.43, 7.01) | 4.43 (2.68, 7.30) | 2.78 (1.50, 5.15) | 3.05 (1.65, 5.63) |
| No hypoglycaemia | 4444 | 46 | 5.22 (6.27) | 1.72 (1.28, 2.31) | 2.06 (1.47, 2.89) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on hospitalizations only | ||||||||
| Hypoglycaemia | 290 | 15 | 2.89 (7.84) | 10.83 (6.56, 17.88) | 10.64 (6.43, 17.61) | 3.11 (1.59, 6.09) | 3.13 (1.50, 6.96) | 2.88 (1.34, 6.21) |
| No hypoglycaemia | 1160 | 47 | 6.04 (7.40) | 5.87 (4.43, 7.77) | 5.87 (4.43, 7.77) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on physician visits only | ||||||||
| Hypoglycaemia | 2570 | 104 | 5.09 (5.39) | 6.90 (5.71, 8.35) | 4.57 (3.39, 6.16) | 2.48 (1.91, 3.23) | 1.55 (1.07, 2.23) | 1.55 (1.07, 2.26) |
| No hypoglycaemia | 10280 | 196 | 4.83 (6.15) | 3.31 (2.87, 3.81) | 4.42 (3.27, 5.97) | 1.00 | 1.00 | 1.00 |
| Exposure | No. of patients | No. of events | Median follow-up in years (IQR) | Crude incidence ratea (95% CI) | Weighted incidence ratea (95% CI) | Crude HR (95% CI) | Weighted HRb (95% CI) | Adjusted HRc (95% CI) |
|---|---|---|---|---|---|---|---|---|
| Primary analysis | ||||||||
| Hypoglycaemia | 2794 | 116 | 5.01 (5.55) | 7.19 (6.00, 8.60) | 4.59 (3.52, 5.98) | 2.73 (2.12, 2.57) | 1.83 (1.31, 2.57) | 1.78 (1.27, 2.49) |
| No hypoglycaemia | 11176 | 213 | 5.07 (6.35) | 3.21 (2.80, 3.67) | 3.33 (2.58, 3.88) | 1.00 | 1.00 | 1.00 |
| Secondary analysis (high-risk population) | ||||||||
| Hypoglycaemia | 1824 | 90 | 4.48 (5.21) | 9.25 (7.54, 11.34) | 7.78 (4.51, 11.34) | 2.73 (2.08, 3.57) | 1.98 (1.41, 2.78) | 1.96 (1.39, 2.77) |
| No hypoglycaemia | 7296 | 212 | 5.48 (6.67) | 4.57 (3.97, 5.27) | 4.95 (4.18, 5.86) | 1.00 | 1.00 | 1.00 |
| Sensitivity analyses | ||||||||
| Population restricted to those with incident type 2 diabetes aged 50, 60 years | ||||||||
| Hypoglycaemia | 1111 | 38 | 5.13 (5.69) | 5.83 (4.25, 7.98) | 3.18 (1.43, 7.01) | 4.43 (2.68, 7.30) | 2.78 (1.50, 5.15) | 3.05 (1.65, 5.63) |
| No hypoglycaemia | 4444 | 46 | 5.22 (6.27) | 1.72 (1.28, 2.31) | 2.06 (1.47, 2.89) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on hospitalizations only | ||||||||
| Hypoglycaemia | 290 | 15 | 2.89 (7.84) | 10.83 (6.56, 17.88) | 10.64 (6.43, 17.61) | 3.11 (1.59, 6.09) | 3.13 (1.50, 6.96) | 2.88 (1.34, 6.21) |
| No hypoglycaemia | 1160 | 47 | 6.04 (7.40) | 5.87 (4.43, 7.77) | 5.87 (4.43, 7.77) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on physician visits only | ||||||||
| Hypoglycaemia | 2570 | 104 | 5.09 (5.39) | 6.90 (5.71, 8.35) | 4.57 (3.39, 6.16) | 2.48 (1.91, 3.23) | 1.55 (1.07, 2.23) | 1.55 (1.07, 2.26) |
| No hypoglycaemia | 10280 | 196 | 4.83 (6.15) | 3.31 (2.87, 3.81) | 4.42 (3.27, 5.97) | 1.00 | 1.00 | 1.00 |
Per 1000 person years.
Inverse probability of treatment weighted model (IPTW).
IPTW adjusted for the impact of policy change in cholinesterase inhibitor coverage in British Columbia.
Risk estimates of all-cause dementia associated with exposure to hypoglycaemia from all analyses
| Exposure | No. of patients | No. of events | Median follow-up in years (IQR) | Crude incidence ratea (95% CI) | Weighted incidence ratea (95% CI) | Crude HR (95% CI) | Weighted HRb (95% CI) | Adjusted HRc (95% CI) |
|---|---|---|---|---|---|---|---|---|
| Primary analysis | ||||||||
| Hypoglycaemia | 2794 | 116 | 5.01 (5.55) | 7.19 (6.00, 8.60) | 4.59 (3.52, 5.98) | 2.73 (2.12, 2.57) | 1.83 (1.31, 2.57) | 1.78 (1.27, 2.49) |
| No hypoglycaemia | 11176 | 213 | 5.07 (6.35) | 3.21 (2.80, 3.67) | 3.33 (2.58, 3.88) | 1.00 | 1.00 | 1.00 |
| Secondary analysis (high-risk population) | ||||||||
| Hypoglycaemia | 1824 | 90 | 4.48 (5.21) | 9.25 (7.54, 11.34) | 7.78 (4.51, 11.34) | 2.73 (2.08, 3.57) | 1.98 (1.41, 2.78) | 1.96 (1.39, 2.77) |
| No hypoglycaemia | 7296 | 212 | 5.48 (6.67) | 4.57 (3.97, 5.27) | 4.95 (4.18, 5.86) | 1.00 | 1.00 | 1.00 |
| Sensitivity analyses | ||||||||
| Population restricted to those with incident type 2 diabetes aged 50, 60 years | ||||||||
| Hypoglycaemia | 1111 | 38 | 5.13 (5.69) | 5.83 (4.25, 7.98) | 3.18 (1.43, 7.01) | 4.43 (2.68, 7.30) | 2.78 (1.50, 5.15) | 3.05 (1.65, 5.63) |
| No hypoglycaemia | 4444 | 46 | 5.22 (6.27) | 1.72 (1.28, 2.31) | 2.06 (1.47, 2.89) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on hospitalizations only | ||||||||
| Hypoglycaemia | 290 | 15 | 2.89 (7.84) | 10.83 (6.56, 17.88) | 10.64 (6.43, 17.61) | 3.11 (1.59, 6.09) | 3.13 (1.50, 6.96) | 2.88 (1.34, 6.21) |
| No hypoglycaemia | 1160 | 47 | 6.04 (7.40) | 5.87 (4.43, 7.77) | 5.87 (4.43, 7.77) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on physician visits only | ||||||||
| Hypoglycaemia | 2570 | 104 | 5.09 (5.39) | 6.90 (5.71, 8.35) | 4.57 (3.39, 6.16) | 2.48 (1.91, 3.23) | 1.55 (1.07, 2.23) | 1.55 (1.07, 2.26) |
| No hypoglycaemia | 10280 | 196 | 4.83 (6.15) | 3.31 (2.87, 3.81) | 4.42 (3.27, 5.97) | 1.00 | 1.00 | 1.00 |
| Exposure | No. of patients | No. of events | Median follow-up in years (IQR) | Crude incidence ratea (95% CI) | Weighted incidence ratea (95% CI) | Crude HR (95% CI) | Weighted HRb (95% CI) | Adjusted HRc (95% CI) |
|---|---|---|---|---|---|---|---|---|
| Primary analysis | ||||||||
| Hypoglycaemia | 2794 | 116 | 5.01 (5.55) | 7.19 (6.00, 8.60) | 4.59 (3.52, 5.98) | 2.73 (2.12, 2.57) | 1.83 (1.31, 2.57) | 1.78 (1.27, 2.49) |
| No hypoglycaemia | 11176 | 213 | 5.07 (6.35) | 3.21 (2.80, 3.67) | 3.33 (2.58, 3.88) | 1.00 | 1.00 | 1.00 |
| Secondary analysis (high-risk population) | ||||||||
| Hypoglycaemia | 1824 | 90 | 4.48 (5.21) | 9.25 (7.54, 11.34) | 7.78 (4.51, 11.34) | 2.73 (2.08, 3.57) | 1.98 (1.41, 2.78) | 1.96 (1.39, 2.77) |
| No hypoglycaemia | 7296 | 212 | 5.48 (6.67) | 4.57 (3.97, 5.27) | 4.95 (4.18, 5.86) | 1.00 | 1.00 | 1.00 |
| Sensitivity analyses | ||||||||
| Population restricted to those with incident type 2 diabetes aged 50, 60 years | ||||||||
| Hypoglycaemia | 1111 | 38 | 5.13 (5.69) | 5.83 (4.25, 7.98) | 3.18 (1.43, 7.01) | 4.43 (2.68, 7.30) | 2.78 (1.50, 5.15) | 3.05 (1.65, 5.63) |
| No hypoglycaemia | 4444 | 46 | 5.22 (6.27) | 1.72 (1.28, 2.31) | 2.06 (1.47, 2.89) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on hospitalizations only | ||||||||
| Hypoglycaemia | 290 | 15 | 2.89 (7.84) | 10.83 (6.56, 17.88) | 10.64 (6.43, 17.61) | 3.11 (1.59, 6.09) | 3.13 (1.50, 6.96) | 2.88 (1.34, 6.21) |
| No hypoglycaemia | 1160 | 47 | 6.04 (7.40) | 5.87 (4.43, 7.77) | 5.87 (4.43, 7.77) | 1.00 | 1.00 | 1.00 |
| Hypoglycaemia defined based on physician visits only | ||||||||
| Hypoglycaemia | 2570 | 104 | 5.09 (5.39) | 6.90 (5.71, 8.35) | 4.57 (3.39, 6.16) | 2.48 (1.91, 3.23) | 1.55 (1.07, 2.23) | 1.55 (1.07, 2.26) |
| No hypoglycaemia | 10280 | 196 | 4.83 (6.15) | 3.31 (2.87, 3.81) | 4.42 (3.27, 5.97) | 1.00 | 1.00 | 1.00 |
Per 1000 person years.
Inverse probability of treatment weighted model (IPTW).
IPTW adjusted for the impact of policy change in cholinesterase inhibitor coverage in British Columbia.
The risk of all-cause dementia was higher for those exposed to hypoglycaemia compared with those who were not (weighted HR, 1.83; 95% CI 1.31, 2.57). As previously mentioned, we further adjusted for any potential impact on dementia diagnoses in BC due to the introduction of a cholinesterase inhibitor reimbursement policy in 2007. This further adjustment led to a similar risk estimate (dajusted HR, 1.78; 95% CI 1.27, 2.49) (Table 2). Additionally, results from multiplicative interaction models were not statistically significant (P >0.06 for all) and did not suggest any effect modification by sex or SES.
Secondary and sensitivity analyses
From those who were using sulphonylurea, meglitinides or insulin, we created a secondary sub-cohort of 9120 individuals of whom 1824 experienced a hypoglycaemic episode. The incidence rates and hazard ratio were slightly higher compared with the primary sub-cohort (weighted HR, 1.98; 95% CI 1.41, 2.78 and adjusted HR, 1.96; 95% CI 1.27, 2.77) (Table 2).
The overall conclusion of an increased risk of dementia associated with hypoglycaemia was similar using both exposure definitions; however, it was higher when hypoglycaemia was defined based on hospitalization records only, albeit with wider confidence intervals due to a smaller number of events (Table 2). Similarly, the risk of all-cause dementia among those who experienced hypoglycaemia was higher compared with those unexposed when we used a restricted age range at cohort entry (Table 2).
Importantly, both crude and weighted hazard ratios were consistent when we used lag periods up to 4 years. However, estimates were attenuated using longer lag periods (5–7 years) with weighted hazard ratios not reaching statistical significance (Figure 4).
Crude (A) and weighted (B) hazard ratios (95% confidence intervals) of main analysis using different lag periods
Finally, the minimum strength of association on the risk ratio scale required for an unmeasured confounder associated with the exposure as well as the outcome to explain away the association (i.e. the E-value) was 3.06 (Figure 5).
Joint values of the minimum strength of association between an unmeasured confounder and hypoglycaemia and an unmeasured confounder and all-cause dementia, to fully explain away the observed point estimate of the main analysis
Discussion
Our study found an increased risk of all-cause dementia associated with hypoglycaemia. This conclusion was consistent across secondary and sensitivity analyses. These findings are broadly consistent with previous studies that have assessed this association16–22; however, we used several design and analysis techniques in an effort to combat multiple threats to validity.
Importantly, this study is the first to use data from Canada and utilize the EDS approach to handle the time-dependent nature of the exposure. This approach addresses the limitations of previous studies by anchoring the index date and ascertainment period for the adjustment of confounders. Specifically, unlike previous studies, our covariate assessment period was after diabetes diagnosis but within 365 days before the index date (i.e. exposure to hypoglycaemia or equivalent date for controls within the risk set). There at least three advantages to our approach.
First, this allowed enough time for diabetes complications to develop and exposure to several diabetes medications, particularly insulin, to take place. For example, in studies where covariates were assessed before diabetes diagnosis, the proportion of those using diabetes medications, particularly insulin, and those with a diabetes-related microvascular complication was low.21
Second, this approach emulates a randomized controlled trial (RCT), wherein we were able to mimic randomization by modelling the exposure and the outcome separately. Although theoretically the causal inference positivity assumption is not violated, as individuals can experience a hypoglycaemic event before or at the time of diabetes diagnosis, it is highly unlikely. Therefore, modelling the probability of exposure at the time of diabetes diagnosis is suboptimal. The EDS approach allowed us to anchor the index date at the time of exposure to hypoglycaemia, and therefore we were able to model the exposure using the hdps algorithm wherein we included >500 covariates.
Third, we were able to use the hdps algorithm to derive propensity scores to calculate IPTW, which resulted in well-balanced exposure groups on several important covariates. This was evident by the reduction in the absolute standardized difference (ASD) between exposure groups for a wide range of potential confounders, such as the number of distinct medications used, a measure of polypharmacy that can increase the risk of both hypoglycaemia and dementia. A similar reduction in ASD was observed with the use of insulin and other medications, such as antidepressants, psychotropics, opioids and Parkinson disease medications, some of which have anticholinergic properties. Moreover, several other macro- and micro-complications, including stroke, ischaemic heart disease, heart failure, nephropathy, retinopathy, neuropathy and peripheral vascular disease, all of which are indicative of diabetes severity, were more likely to be present among those who experienced hypoglycaemia before IPTW but not after. Indeed, this is clearly evident in the distribution of insulin use at baseline, which was much higher among the hypoglycaemia group in most of the previous studies that adjusted for insulin use, but not in our analysis.16,17,19,22
Additionally, as a sensitivity analysis, we explored the effect of different lag periods on the association between hypoglycaemia and dementia. Importantly, when a lag period was not considered, both the crude and weighted hazard ratios were higher compared with hazard ratios when a lag period was considered. This may be explained by potential reverse causality wherein hypoglycaemia is an early manifestation of dementia that is yet to be diagnosed. This can occur due to several clinical scenarios including incorrect dosing of insulin or lack of dose adjustment despite weight loss or frailty. Weighted hazard ratios were consistently above 1 when all lag periods were used (1 to 7 years); however, estimates were no longer significant with the lower limit below 1, using longer lag periods (5–7 years). Although reverse causality remains possible, the risk estimates become less precise with longer lag periods occurring after 5 years of index date. Additionally, given the uncertainty on an optimal latency period needed for hypoglycaemia to impact on the development of dementia, longer lag periods can lead to an underestimation of the effect of hypoglycaemia. Therefore, neuroimaging studies as well as observational studies with longer follow-up times are needed for a more definitive conclusion.
In addition to these design nuances, our findings provide important clinical insights. Interestingly, the risk of dementia seems to be higher with more severe hypoglycaemic episodes that required hospitalizations, compared with those captured using physician visits. This signals the need for further work to detail how the severity of the hypoglycaemic episode can affect brain structure and lead to cognitive impairment. Moreover, despite plausible effect modification by SES, as access to health care after the occurrence of hypoglycaemia might affect its cognitive consequences, we did not observe any evidence of difference in risk across quintiles of SES. However, Canada has a universal health care system and therefore this finding should not be generalized to populations with less accessible healthcare.
Our study has some limitations. First, we used data collected for administrative purposes; therefore, misclassification of type 2 diabetes, hypoglycaemia and dementia is possible. Validated algorithms and specific eligibility criteria were used to minimize misclassification bias. We were also only able to assess serious hypoglycaemic events that require medical attention (hospitalization or a physician visit) and we are not able to capture milder hypoglycaemic events. In addition, our outcome was limited to all-cause dementia and we were not able to accurately differentiate between subtypes. Second, although we used multiple lag periods up to 7 years after exposure index date, reverse causality remains possible due to the bidirectional nature of the relationship between hypoglycaemia and dementia. Third, we were not able to include important clinical indicators such as haemoglobin A1c; however, we included several indicators for diabetes severity including macrovascular and microvascular complications and diabetes therapies. Our data also lacked information on lifestyle-related covariates, such as smoking, alcohol consumption and education. Fourth, as with all observational studies, residual and unmeasured confounding remains possible, despite the use of hdps and IPTW. However, to fully explain the observed HR of 1.83, a confounder would have to be associated with both hypoglycaemia and with dementia, each by a risk ratio of at least 3.06 in addition to the confounders that we were able to measure and adjust for. Last, we did not test for a dose-response relationship among subjects with recurrent hypoglycaemic episodes. Future studies should investigate this issue, as it would improve our understanding of the underlying pathophysiology.
Conclusion
Using longitudinal population-level real-world data from over 20 years, we found that serious hypoglycaemic episodes contribute to an increased risk of all-cause dementia. These findings add to the existing body of evidence and provide clinical and public health insight on the modifiable risk factors of dementia in type 2 diabetes. Importantly, this study provides an illustration of several design elements that need to be considered when studying this complex association.
Ethics approval
We obtained ethics approval from the University of Waterloo. All data were de-identified and no personal information was available at any point of the study. Access to data provided by the Data Steward(s) is subject to approval, but can be requested for research projects through the Data Steward(s) or their designated service providers. All inferences, opinions and conclusions drawn in this publication are those of the author(s) and do not reflect the opinions or policies of the Data Steward(s).
Data availability
The data that support the findings of this study are available from Population Data BC, but restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. Data are available from Population Data BC through a data access request [[email protected]].
Supplementary data
Supplementary data are available at IJE online.
Author contributions
W.A., J.M.G., C.J.M. developed the study idea. W.A. conducted all analyses and wrote the first draft of the manuscript. J.M.G. supervised the work. All authors contributed to design, methodology and the final draft of the manuscript. All authors approved the submitted version of the manuscript.
Funding
This project is funded by the Mike & Valeria Rosenbloom Foundation Research Award at the Alzheimer’s Society of Canada.
Conflict of interest
None declared.
