DIVE/DPV registries: benefits and risks of analog insulin use in individuals 75 years and older with type 2 diabetes mellitus

Data collection

Prospective longitudinal, standardized routine data were obtained from the prospective, multicenter German patient databases on diabetes mellitus: DPV (Diabetes-Patienten-Verlaufsdokumentation) and DIVE (DIabetes Versorgungs-Evaluation).

DPV data on patients with diabetes mellitus are collected during routine examinations using DPV software and the anonymized data are sent to Ulm University for inclusion into the database every 6 months. Detailed information on the documentation systems has been published previously.10 11 This analysis includes data from 159 DPV centers.

The DIVE registry was established in Germany in 2011.12 Consecutive patient data are collected from German centers and these patients continue to be followed up. Data are entered into an online database using DPV software. All patients provided written informed consent. This study comprises data of 101 DIVE study sites.

Patient population for the analysis

Patients were sampled in September 2020 and were included in the current analysis if they had at least one clinic visit per year and their documented therapy was available. We selected patients of at least 75 years with T2DM with basal insulin (human or analog) or NPH insulin treatment. Patients aged <75 years, with other forms of diabetes, using an insulin pump or with prandial insulin only were excluded. The final study comprised 8022 propensity score-matched (PSM) subjects for baseline and 3836 PSM subjects for follow-up analysis. The mean follow-up period was 2.3±2.1 years for the HI/NPH group and 2.6±2.1 years for BIA.

Definitions

Hypertension was defined as ≥145 mm Hg for systolic blood pressure and/or a diastolic blood pressure of ≥85 mm Hg and/or antihypertensive treatment.13 14 Body mass index (BMI) was defined as bodyweight divided by the square body height in kg/m². Dyslipidemia was defined as a low-density lipoprotein cholesterol of ≥100 mg/dL (≥2.6 mmol/L) without further risk factors and ≥70 mg/dL (≥1.8 mmol/L) in patients with CVD or chronic kidney disease or those receiving lipid-lowering drug treatment.13 15 Comorbidities were grouped into (patient-reported or physician-reported) microvascular and macrovascular diseases. The former included any form of retinopathy, blindness, nephropathy, renal failure, dialysis or neuropathy. The latter included transient ischemic attack/prolonged reversible ischemic neurologic deficit, stroke, coronary artery disease, myocardial infarction and peripheral arterial disease. Polypharmacy was determined as a numerical-only definition of >10 medications that were used concomitantly.16 The DIVE/DPV databases were searched with focus on drug categories according to the PRISCUS listing.17 Medications comprised prescription medications and routine use of over-the-counter medications.

HbA1c values were standardized to the Diabetes Control and Complications Trial.18 19 For severe hypoglycemia, the definition of the American Diabetes Association Workgroup on Hypoglycemia (‘an event requiring assistance by another person to actively administer carbohydrates, glucagon or other resuscitative actions’) was applied.20 Hypoglycemia with coma was defined as loss of consciousness or occurrence of seizures.10 DKA was defined as pH <7.3 and/or bicarbonate <15 mmol/mol or hospitalization for DKA. Insulin regimen was categorized as follows: (1) basal-supported oral therapy (BOT) (basal insulin only); (2) supplementary insulin therapy (SIT) (prandial insulin only); (3) conventional insulin therapy (CT) (prandial and basal insulin combined, up to three injection time points); (4) intensified conventional insulin therapy (ICT) (prandial and basal insulin combined, >3 injection time points); and (5) continuous subcutaneous insulin infusion (CSII).10 To stratify patients using basal insulin, patients on SIT and CSII were excluded. Patients on BOT, CT or ICT were identified because these therapy regimens use a long-acting (basal) insulin or premixed/co-formulation insulin. Patients on basal insulin were divided by use of HI/NPH and BIA. Insulin treatment initiation had to be documented between 2000 and June 2020, with HbA1c and BMI data available ±3 months from the time of insulin initiation. Baseline data for patients were aggregated within ±3 months of insulin initiation. Follow-up data were aggregated to the most recent documented treatment year without therapy switch (excluding the baseline period). Treatment was considered continuous if there was a maximum gap of 120 days between two clinic visits, with the same treatment being documented before and after the gap.

Statistics

Data from the two registries were combined and analyzed as a single data set. Data were then aggregated as medians in the year of intensification with insulin (baseline). Categorical variables are presented as percentages and continuous variables are presented as mean±SD. Rates and event rates are presented per 100 patient-years (PY). Differences of non-PSM variables were analyzed using Wilcoxon test for continuous variables or χ² test for binary variables. P values were adjusted for multiple testing using the Bonferroni stepdown method.

Propensity score matching was used to ensure that both the HI/NPH therapy regimen group and the BIA group had similar baseline characteristics. Differences between the treatment groups and propensity score for insulin analogs were estimated according to a former analyses of the DIVE/DPV registries.21 The multivariable logistic regression model includes age, sex, duration of diabetes, BMI and HbA1c level as covariates. For each patient, the probability (propensity score) for HI/NPH and BIA was estimated from the logistic model based on the patient’s specific covariate values. Matching was conducted with a one-to-one matching process (greedy-matching algorithm) and caliper width 0.2. To evaluate balancing of matched variables, standardized differences were assessed. A standardized difference of <10% indicates a negligible difference in the mean.22 In order to compare hospitalization times and lengths of stay between the matched groups, a standard Poisson regression model with the logarithm of the individual time under risk as offset was used. Event rates of severe hypoglycemia, hypoglycemic coma, DKA and severe DKA were calculated and compared using negative binomial regression with the logarithm of the individual time under risk as offset. For HbA1c a linear regression model was used. Further, incidence rate ratios and 95% CIs were calculated for DKA, hypoglycemia and hypoglycemia with coma.

A p value <0.05 was considered statistically significant. Statistical analysis was performed using SAS V.9.4.

Patient population

The DIVE/DPV registries included 167 300 patients with T2DM aged ≥75 years at the time of data export. Of these, 29 800 (17.8%) patients initiated insulin therapy between 2000 and June 2020, and 11 108 (6.6%) used therapy regimens with basal insulin. The final analysis comprised 9601 patients in the entire cohort (n=5583 on regular insulin and n=4018 on insulin analogs) and 8022 PSM patients (n=4011 in both treatment groups) (figure 1).

Request permission

Patient baseline characteristics

For analysis of baseline characteristics 4011 matched pairs were available. The standardized differences for the matched variables were highest for HbA1c (4.5% for the entire cohort and 2.0% for the matched cohort) and age (−6.1% and 0.6%) but remained largely under 10%, demonstrating a negligible difference between both treatment groups (table 1). The mean patient age was 80.3 years in both matched treatment groups, half of the patients were male, the mean diabetes duration was approximately 12 years, and the mean baseline HbA1c was 8.6% for both insulins. There were only minor significant differences in the entire cohort as well as in the matched cohort for non-matched variables at time of insulin therapy initiation.

FBG did not differ between the groups (10.6 mmol/L vs 10.7 mmol/L, p=0.50 for the entire cohort; 10.6 mmol/L vs 10.8 mmol/L for the PMS cohort, p=0.42). The bodyweight of BIA patients in the entire cohort was slightly higher (79.4 kg vs 80.6 kg; p=0.003), but this difference was not seen in the PMS cohort (80.2% vs 80.6%; p=1.00). In the PMS cohort, the proportion of patients with dyslipidemia was higher in the BIA group (85.9% vs 89.3%; p=0.001). The proportions of other risk factors such as hypertension (71.0% vs 73.5%) and smoking (3.5% vs 3.4%) were largely similar. The only exception was the mean number of medications, which was significantly higher in the BIA group compared with the HI/NPH group in (both cohorts 3.7 vs 4.3; p<0.001), whereas the proportion of patients with polypharmacy was borderline significantly lower in the HI/NPH group of the entire cohort (0.5% vs 1.1%; p=0.05), but not in the matched cohort (0.6% vs 1.1%; p=0.26). Furthermore, in the PSM cohort depression was more prevalent with BIA treatment (6.5% vs 8.4%; p=0.01).

There were small differences in the prevalence of comorbidities such as dementia (HI/NPH 7.9% vs BIA 8.8%), cancer (HI/NPH 5.0% vs BIA 5.8%) and microvascular and macrovascular diseases (HI/NPH 85.3% vs BIA 83.6% and HI/NPH 44.6% vs BIA 44.4%), but these were not statistically significant.

The proportion of care-dependent patients (HI/NPH 5.2% vs BIA 7.1%) was low and there were no differences observed between both treatment groups. Marginally more than one-third of the patients participated in a diabetes management program (HI/NPH 37.1% vs BIA 37.9%).

Concomitant antidiabetic drugs at follow-up

Detailed information on concomitant antidiabetic treatment was available for 49.5% of the HI/NPH subjects and 46.1% of the BIA group for the most recent treatment year (follow-up) (table 2). Most of the patients on regular insulin were treated with CT (56.2% vs 13.2%; p<0.001), whereas BOT and ICT were more frequent with BIA (14.4% vs 42.6% and 29.4% vs 44.3%; both p<0.001).

Rapid-acting insulin was added to basal insulin concomitantly in 79.4% of HI/NPH and 53.5% of BIA patients. Rapid-acting HI was preferred in the HI/NPH group versus the BIA group (61.3% vs 19.3%). The reverse was true for rapid-acting analogs (18.1% vs 34.1%) (both p<0.001).

Oral antidiabetic drugs and glucagon-like peptide-1 (GLP-1) agonists were more frequently used in patients using insulin analogs. Usage proportions were highest for metformin (23.3% vs 28.7%), dipeptidyl peptidase-4 (DPP-4) inhibitors (10.0% vs 25.8%) and sulfonylureas (10.8% vs 11.2%). With the exception of the latter and glucosidase inhibitors, differences in non-insulin antidiabetic therapy were highly significant between both groups. However, percentages of patients taking glucosidase inhibitors (0.8% vs 0.9%), GLP agonists (0.8% vs 2.2%) and sodium/glucose cotransporter 2 (SGLT-2) inhibitors (0.6% vs 3.6%) were low.

Efficacy and safety parameters

A comparison of follow-up data in 1984 matched patients on HI/NPH to 1852 patients on insulin analog was used to determine differences in glycemic control, rates of severe hypoglycemia and DKA (table 3).

Glycemic control and hospitalizations

Patients in the BIA group needed less total daily insulin (0.30 IU/kg/day vs 0.22 IU/kg/day; p<0.001). The prandial to total insulin ratio was higher in the BIA group compared with the regular insulin group (0.83 vs 0.70; p<0.001), demonstrating a higher use of prandial insulin in the BIA group. The resulting median HbA1c was similar with HI/NPH treatment and BIA treatment (7.3% vs 7.4%; p=0.06). FBG did not differ between the groups (both 8.4 mmol/L; p=0.85).

The mean number of hospitalizations per PY was slightly lower in the BIA compared with the HI/NPH group, but did not reach statistical significance (0.7 vs 0.8; p=0.15). The mean length of stay per PY did not differ between the groups (HI/NPH 16.1 days vs BIA 16.3 days; p=0.53).

Hypoglycemia and DKA

Of the HI/NPH patients 48.6% had HbA1c values below the median HbA1c and no severe hypoglycemia or hypoglycemia with coma was observed. The same was true for 51.5% of the BIA group (p=0.09 HI/NPH vs BIA). Overall, 2.8% of subjects on regular insulin and 2.4% on insulin analogs suffered at least one severe hypoglycemia event. Event rates for severe hypoglycemia, however, tended to be lower in the BIA group but did not reach statistical significance (HI/NPH 4.4 and BIA 4.1 per 100 PY; p=0.88). Severe hypoglycemia with coma was documented in DPV only and observed in 2.8% of the HI/NPH group and 3.0% of the BIA group with event rates of 3.3 and 3.6 per 100 PY (p=0.88). The percentage of patients with DKA was low and the event rates did not differ (HI/NPH 0.3% vs 0.1% of patients and 0.28 vs 0.05 per 100 PY; p=0.36).

Incidence rate ratios and 95% CIs show no differences in risk of severe hypoglycemia (0.95, 95% CI 0.34 to 2.61), hyperglycemic events with coma (1.16, 95% CI 0.25 to 5.27) and DKA (0.19, 95% CI 0.01 to 6.80) between regular insulins compared with insulin analogs (figure 2).

Request permission

Patient population

Patients with BIA tended to have more risk factors at baseline. This became significant in the slightly higher number of medications taken (4.3 vs 3.7) and higher proportion of patients with dyslipidemia in the BIA group. Although lipid targets are more likely to be met in the elderly, it is assumed that diabetic dyslipidemia in the elderly is undertreated.23 24 T2DM and obesity are also linked with depression,25 which also was more prevalent in the BIA group. Interestingly, the proportion of patients with polypharmacy (defined in this study as >10 medications) was only 0.6% and 1.1% in the HI/NPH and BIA groups, respectively. Although some medications are documented only in the DPV registry and not in DIVE, data indicate that this risk factor plays a minor role in the patient population.

Among the insulin-treated patients in this study the proportion of patients with hypertension was high (71.0% for HI/NPH and 73.5% for BIA). The dysregulation of neurohumoral and neuroimmune pathways contributes to the pathophysiology of both T2DM and hypertension; thus, there is a bidirectional association between macrovascular and microvascular systems.26 In line with this, we observed in this insulin-treated population a higher proportion of microvascular disease (85.3% vs 83.6%) compared with our previous DIVE/DPV analysis (also including insulin-naïve subjects), which revealed proportions from 59.8% of patients aged 70–79 years old to 50.4% of those aged >90 years.2

In Germany care-dependent elderly live either at home (assisted by family, caregivers or home care providers) or in ‘shared housing arrangements’ and nursing homes.27 The prevalence of care dependency rises from 8% among those aged 75–79 years old to 76% among those aged >90 years old.28 Surprisingly, in our study only 5.2% and 5.5% of patients with T2DM at comparable age were care-dependent. A previous study reported that care-dependent patients with T2DM are more likely to be treated with insulin compared with independent patients.29 However, a large proportion of nursing home residents are treated by general practitioners and not diabetologists and are therefore probably not documented in the registries.29 Furthermore, the support of self-management and therapy adherence and the intensified lifestyle training (eg, 37% of patients included in disease management programs) may lead to a higher proportion of independent elderly patients compared with the above-mentioned German statistics.28

Antidiabetic pharmacotherapy and insulin regimen

Rather than focusing on glycemic control in elderly patients with T2DM, it is important to maintain quality of life and patients’ ability to self-manage their diabetes. Current guidelines recommend simplifying treatment (eg, by using basal insulin and avoiding the additional use of regular and rapid-acting insulin) and to use drug classes with low hypoglycemic effect.13 30–32 BOT was used three times more frequently in the BIA group compared with HI/NPH (42.6% vs 14.4%) and the BIA group needed lower total daily insulin levels. Further, in our study additional rapid-acting insulins were needed less often in the BIA group compared with the HI/NPH group. Thus, data indicate that current guidelines advising a simple patient-approached therapy regimen to maintain self-management abilities in the elderly were better reflected by using a BOT regimen with insulin analogs.13

One study indicates a longer persistence on BOT therapy in elderly patients with T2DM with BIA before intensifying to ICT compared with patients with NPH.33 Further, with BOT additional antidiabetic drugs are necessary,34 explaining the higher percentages of concomitant medication in the BIA group compared with HI/NPH.

In contrast to that, more than half of the HI/NPH group used a CT insulin regimen (56.2%). CT necessitates fixed meal times and is useful in less active or care-dependent patients.34 It is likely that patients who have used this treatment strategy for long periods of time were not switched as they aged. In contrast to CT, the more complex ICT is used in active patients with T2DM with good mental and functional status but poor glycemic control.34 The higher proportion of ICT in the BIA group indicates that a higher proportion of patients had difficulty stabilizing their diabetes and explains the higher prandial to total insulin ratio. Another reason for the higher ICT proportion with BIA might be the higher risk of hypoglycemia with ICT and physicians might have prescribed BIA due to its promoted antihypoglycemic properties. Furthermore, the BIA group needed lower total daily insulin levels, but a higher percentage of patients were treated with newer non-insulin antidiabetic agents. For example, the proportion of DPP-4 inhibitors was twice as high compared with the HI/NPH group (10.0% vs 25.8%). In studies, DPP-4 inhibitors, GLP agonists and SGLT-2 inhibitors showed a reduced risk of hypoglycemia, and DPP-4 inhibitors are often used in elderly patients and in patients with renal impairment.34 However, GLP-1 agonists have to be injected and overall rates of GLP-1 agonists and SLGT-2 inhibitors were low (<5%).

Glycemic control and safety

In older adults with few coexisting chronic illnesses and good mental and functional status, the glycemic target of HbA1c should be <7.5%. Multimorbid patients, however, should have less strict glycemic goals with HbA1c values between 8.0% and 8.5%.13 24 31 32 In our study HbA1c decreased clinically relevant with insulin therapy to a mean follow-up HbA1c of 7.3% and 7.4%, suggesting that there might still be a slight antiglycemic overtreatment in this elderly population at high risk of diabetic and cardiovascular complications. Antiglycemic overtreatment of older patients with T2DM has been shown to be common and the possibly resulting harm is cited in guidelines.31 34 35

In our study, 2.8% and 2.4% of the patients were documented with severe hypoglycemia and the event rates per 100 PY were low (HI/NPH 4.40 vs BIA 4.07) and even lower for hypoglycemia with coma (HI/NPH 3.26 vs BIA 3.64), tending toward lower rates with BIA, but without significant differences. There is no clear evidence of studies comparing the two insulin types with regard to the risk of hypoglycemic events. Studies in elderly patients are lacking and severe hypoglycemia is often unrecognized in elderly patients with T2DM.30 Most of the available studies in patients with T2DM show comparable effects for HbA1c and a lower risk for patients experiencing hypoglycemia for insulin analogs when compared with NPH insulin.36 However, results of recent studies comparing insulin analogs with premixed or split regular insulin regimens are incongruent with regard to the frequency and severity of hypoglycemia and the occurrence of nocturnal hypoglycemia.8 37–41 This was also true for rapid-acting analogs versus regular HI.42 A recent observational study using Medicare claims data examined the association between long-acting insulin analog use and emergency department visits or hospitalizations for hypoglycemia in more than 575 000 T2DM insulin users over 65 years.43 Bradley et al43 showed that the initiation of glargine and detemir use was associated with a reduced risk of hypoglycemia compared with NPH insulin use (HR for glargine vs NPH 0.71, 95% CI 0.63 to 0.80; HR for detemir vs NPH 0.72, 95% CI 0.63 to 0.82). The protective association was not seen with concomitant prandial insulin use. In a real-world setting as well as in our study long-acting insulin and prandial insulin are often used together. Therefore, potential benefits of BIA use in our study might not have been that clear. Overall, the split evidence evokes a controversial expert discussion about the clinically relevant differences of the two insulin types.34 44

Limitations

The strength of this analysis is the large number of elderly patients aged ≥75 years with a matched group comprising over 8000 patients in a real-world setting and a mean age of 80.3 years. However, in an observational study residual selection bias despite the effect of propensity score matching is possible, for example by the sole involvement of specialized diabetes centers in the DIVE/DPV registries. Newer concepts such as frailty and sarcopenia are rarely or not reported in the registries due to lack of consensus on diagnostic criteria; for example, International Classification of Diseases-10 for sarcopenia was only established in 2016.32 45 Furthermore, the analysis included follow-up data from 2000. This may lead to an under-representation of newer therapy options, such as non-insulin antidiabetic drugs (SGLT-2 inhibitors, GLP-1 agonists) and the use of insulin analogs. Finally, we did not distinguish between premixed and split usage of insulins, which might also have an implication on outcomes.