Real-World Effects of Second-Generation Versus Earlier Intermediate/Basal Insulin Analogues on Rates of Hypoglycemia in Adults with Type 1 and 2 Diabetes (iNPHORM, US)

Data were analyzed from the Investigating Novel Predictions of Hypoglycemia Occurrence using Real-World Models (iNPHORM) study: a US-wide, 12-wave ambidirectional panel survey. Complete details on the design and conduct of iNPHORM are published elsewhere [26]. The present article complies with ‘The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: guidelines for reporting observational studies’ [27]; it addresses Objective 3 of the overall iNPHORM study (see protocol [26]).

Self-assessed data on hypoglycemia and potential non-clinical/clinical factors (intrinsic, extrinsic, non-modifiable, and modifiable) were captured across 14 closed questionnaires (screening, baseline, 12-monthly follow-ups) that were pretested and piloted prior to fielding.

Participants were recruited across two subpanels (A and B) from a probability-based internet panel (> 65,000 Americans with diabetes). After screening, eligible panelists were required to provide consent and complete a baseline questionnaire to enroll. Panelists who were (1) 18–90 years old, (2) living in the US (past year), (3) self-reporting T1DM or T2DM, and (4) using insulin, secretagogues, or both (past year) were eligible. Those diagnosed with gestational diabetes, participating in a concurrent trial, or pregnant (at screening or year prior) were ineligible.

Subpanel A (n = 1257) was conveniently sampled from a random subset of the internet panel. Those who failed to complete the first follow-up were withdrawn (n = 488) and refreshed with 437 new eligible participants conveniently sampled from a different random subset of the internet panel (subpanel B). Individuals in subpanel A who completed the first follow-up and all those in subpanel B comprised the iNPHORM Longitudinal Panel (N = 1206). Quota sampling was used to ensure minimum representation by diabetes type, sex, age, and medication regimen.

Enrollees were managed by Ipsos Interactive Services (IIS), a global leader in real-world survey conduct. In total, participants in the iNPHORM Longitudinal Panel were emailed 12-monthly follow-up questionnaires by IIS. Participants were given 7 days to submit each follow-up using any internet-enabled device (e.g., computer, tablet, smartphone). Responses were automatically stored on the IIS platform. Token cash incentives, in line with ethical guidelines [28], were distributed based on the number and timing of completed questionnaires.

At screening and all monthly follow-ups, participants were asked to self-report on their current use of second-generation (insulin degludec and insulin glargine U-300) and earlier intermediate/basal insulin analogues (insulin detemir, insulin glargine U-100, premixed insulin, fixed-ratio [FRC] insulin, and intermediate [insulin isophane {Neutral Protamine Hagedorn; NPH} insulin]). To improve self-reported accuracy, participants were encouraged to review their medication packaging before reporting, prompted by medication lists of generic and brand names, provided information about their last reported antihyperglycemic regimen, and allowed to complete the survey over multiple days, if needed.

The causal effect of second-generation versus earlier intermediate/basal insulin analogues on real-world rates of hypoglycemia is plausibly confounded by various clinical and non-clinical factors. To identify possible confounding variables, we constructed a directed acyclic graph (DAG) using the daggity plotting tool [29] for all severities and timings (daytime and nocturnal) of hypoglycemia (Supplementary Fig. 1 in the electronic supplementary material).

A single DAG was created to ensure internal comparability between effect estimates for hypoglycemia severities and timings. Based on the DAG, we identified the following minimally sufficient adjustment set: age, employment status, household income, diabetes type, diabetes duration, insulin therapy duration, bolus insulin use, and background secretagogue therapy.

All variables were self-reported and collected via iNPHORM questionnaires. Age was determined at screening based on each participant’s date of birth (month and year). Current employment status (i.e., working full or part time (including self-employment), retired, unemployed, or a student) was measured (single response option) at baseline and updated at 4-month intervals. Gross annual household income was assessed at baseline according to intervals of $15,000. Diabetes type (T1DM or T2DM) was self-reported at screener and verified at baseline. Diabetes duration was determined at baseline based on age of diabetes diagnosis. Insulin therapy duration was assessed at baseline by asking participants to self-report how long in years and months they had been taking any insulin, regardless of brand, type, or delivery method. Bolus insulin use (i.e., bolus/prandial insulin, including rapid- and short-acting insulin) and background secretagogue therapy (i.e., short-, intermediate-, or long-acting sulfonylurea; meglitinide; meglitinide and biguanide fixed-dose combination or sulfonylurea and biguanide fixed-dose combination; and/or some other secretagogue) were assessed (multi-response option) at baseline and updated at every follow-up.

Frequencies of daytime and nocturnal non-severe and severe hypoglycemia were assessed at baseline and prospectively across 12 follow-ups. Definitions consistent with the International Hypoglycaemia Study Group and American Diabetes Association (ADA) Standards of Medical Care in Diabetes [30] were provided in all questionnaires. Specifically, non-severe hypoglycemia was defined as any event, identified via symptoms and/or blood glucose measurement, that the participant was physically able to self-treat. Severe hypoglycemia was defined as any Level 3 event that the participant was physically unable to self-treat and thus may have required external assistance for recovery (e.g., treatment by family/friend, ambulatory care, hospitalization) [31]. The response options “I recovered on my own without any kind of treatment,” “Other,” and “Unknown treatment” were provided. Participants reported whether their non-severe/severe hypoglycemia occurred while awake (daytime) or sleeping/attempting to sleep (nocturnal). Overall non-severe and overall severe hypoglycemia combined daytime and nocturnal events.

To ensure accurate responses and to prevent overlapping recall intervals, follow-up data on daytime and nocturnal non-severe hypoglycemia were captured “Within the past 30 days” (if the last scheduled questionnaire was not completed) or “Since the last time an iNPHORM survey was completed” (if the last scheduled questionnaire was completed). Conversely, given its relative infrequency and saliency, daytime and nocturnal severe hypoglycemia was captured “Since the last time an iNPHORM survey was completed.”

Analyses were restricted to T1DM or T2DM participants who reported taking intermediate/basal insulin (with or without a bolus regimen) at any time over follow-up and who responded to at least one follow-up questionnaire. Sample characteristics were summarized using frequencies and percentages for categorical variables and means and standard deviations (SD) or medians and interquartile ranges (IQR) for continuous variables. Incidence proportions of daytime and nocturnal non-severe and severe hypoglycemia were quantified over the duration of follow-up using Wilson’s confidence interval for binomial proportions. To account for over-dispersion, annualized rates were modeled using negative binomial regression with follow-up durations included as offsets.

We estimated the population-average rate ratio of hypoglycemia events comparing use of second-generation with earlier intermediate/basal insulin analogues using multivariable negative binomial regression with generalized estimating equations (GEEs). Models were adjusted for confounding variables identified by our DAG. Repeated and time-varying outcome and covariate measures were analyzed in our final models. We selected an exchangeable covariance structure to ensure all repeated observations within a participant had the same correlation over time. We estimated separate effects by severity (non-severe and severe), timing (overall, daytime, and nocturnal), and diabetes type. Finally, a sensitivity analysis was performed to test the potential time-varying confounding effect of hypoglycemia frequency on subsequent medication regimens (e.g., high hypoglycemia rates may have encouraged participants to initiate second-generation basal insulin analogue use).

Before recruitment, we obtained ethics approval from the Western University Health Sciences Research Ethics Board (Project ID: 112,986; December 17, 2019). The study was conducted in accordance with the Helsinki Declaration of 1964 and its later amendments. Consent to participate was obtained at screening and could be revoked by not completing future questionnaires. Consent to publication was obtained at screening.

Table 1 presents the baseline clinical and sociodemographic characteristics of our final cohort. Roughly 20% (n = 81) of participants had T1DM, while the median diabetes duration across the sample was 15 years (IQR: 16 years). Individuals were on average 53.0 (SD: 13.6) years old, and 55% (n = 226) were female. Mean body mass index (BMI) was 33.0 (SD: 9.6) kg/m², and 86% (n = 356) reported having one or more comorbidities. Roughly two-thirds (n = 271; 65.6%) of participants reported HbA1c values < 8%.

Overall, 23% (n = 93) reported using a second-generation basal insulin analogue, 64% (n = 264) a concomitant bolus insulin, and 22% (n = 92) a secretagogue therapy. Private insurance was more common among second-generation versus earlier intermediate/basal insulin users (44% vs. 36%; p = 0.07); contrariwise, public insurance (e.g., via government assistance plans) was more common among earlier versus second-generation insulin users (42% vs. 31%; p = 0.03). Participants treated with earlier intermediate/basal insulin analogues were also more likely to report annual gross household incomes < $25,000 (though not statistically significantly; 23% vs. 18%; p = 0.15) compared to second-generation insulin analogue users who more frequently reported incomes between $55,000 and $84,999 (28% vs. 19%; p = 0.03).

For non-severe and severe hypoglycemia, respectively, a total of 311.7 and 287.7 person-years were observed, with a mean of 274.8 and 253.6 days observed per participant and 30.5 and 28.1 days observed per questionnaire. Tables 2 and 3 summarize hypoglycemia incidence proportions and rates among second-generation and earlier intermediate/basal insulin users, by type of diabetes, event severity, and timing. Most participants reported at least one overall non-severe hypoglycemia event over follow-up (83.5%, 95% confidence interval [CI]: 79.7−86.8%), while the rate of overall non-severe hypoglycemia was 27.0 events per person-year (EPPY) (95% CI 23.5−30.9). Participants reported a rate of 2.9 overall severe hypoglycemia EPPY (95% CI 2.2−3.8 EPPY); 31.2% experienced one or more overall severe hypoglycemia events (95% CI 27.0−35.9%). Overall, daytime events were more frequent than nocturnal.

The rate of non-severe hypoglycemia was higher among T1DM versus T2DM participants (63.5 vs. 17.9 EPPY, p < 0.001); however, for severe hypoglycemia, rates were statistically comparable (3.3 [T1DM] vs. 2.8 [T2DM] EPPY, p = 0.55). Second-generation basal insulin users reported lower crude rates of daytime and nocturnal non-severe and severe hypoglycemia compared to earlier intermediate/basal insulin users. Notably, the rate of non-severe nocturnal hypoglycemia among second-generation insulin users was nearly half that reported by earlier intermediate/basal insulin users (5.2 vs. 8.6 EPPY, p = 0.04); we observed a similar trend for severe nocturnal events (0.4 vs. 0.9 EPPY, p = 0.14).

Table 4 reports population-average adjusted hypoglycemia rate ratios for second-generation versus intermediate/basal insulin use (by type of diabetes, event severity, and timing). Supplementary Table 1 in the Supplementary Material summarizes all parameter estimates.

The estimated population-average nocturnal non-severe hypoglycemia rate was statistically significantly lower among second-generation versus earlier intermediate/basal insulin analogue users (rate ratio [RR]: 0.57 [95% CI 0.44−0.74], p < 0.001); this trend was statistically non-significant for the estimated population-average daytime non-severe hypoglycemia rate (RR: 0.91 [95% CI 0.76−1.10], p = 0.32). Participants on second-generation versus earlier intermediate/basal insulin analogues also experienced statistically significantly fewer severe nocturnal events (RR: 0.56 [95% CI 0.35−0.90], p = 0.02); this association was more salient among participants with T1DM (RR: 0.23 [95% CI 0.06−0.93], p = 0.04) compared to those with T2DM (RR: 0.63 [95% CI 0.38–1.06], p = 0.08). Other estimated effects of second-generation compared to earlier intermediate/basal insulin analogue use on hypoglycemia rates were consistent between T1DM and T2DM participants as well as across event severities and timings (Table 4).

Prior to iNPHORM, no studies had assessed the real-world effect of second-generation versus earlier intermediate/basal insulin types on rates of hypoglycemia. Instead, earlier evidence stemmed from poorly generalizable trials that excluded individuals at greatest risk of hypoglycemia [62] and health administrative databases [44, 63, 64] prone to severe outcome ascertainment bias [16, 65].

Addressing this gap, we administered a screener, baseline, and 12-monthly questionnaires to an online panel reflective of the US public with diabetes. Primary self-reported data collection optimized complete and honest hypoglycemia reporting [21]; it also allowed us to capture information on potential confounding variables often unavailable or poorly documented in routine health records (e.g., employment, household income).

The use of internet panel surveys enabled us to reach participants with T1DM and T2DM across the US while ensuring feasibility and strong participant retention. Quota sampling ensured sufficient representation of key groups (i.e., people with T1DM and T2DM, older people, people assigned female sex at birth, and people treated with insulin, secretagogues, and their combination). Nonetheless, despite efforts to ensure sample representativeness, iNPHORM participants were often insured and unemployed, retired, or in school.

Prospective collection of hypoglycemia events at monthly intervals maximized participant recall while minimizing burden of data collection on participants. Needless to say, we could not assess events undetected or unreported by participants. For example, perception bias may have caused some individuals treated with second-generation insulins to under-estimate their hypoglycemia frequencies—particularly non-severe events. Consequent measurement error could have attenuated self-reported non-severe event frequencies among second-generation versus earlier intermediate/basal insulin users, resulting in a conservative estimate of the corresponding hypoglycemia rate reduction. Severe hypoglycemia was less likely affected by perception bias; nevertheless, we did not corroborate the percentage of events resulting in healthcare (e.g., using health service records).

Compared with glycemic measurements obtained by real-time/continuous glucose monitoring (rt-C/FGM), self-reported hypoglycemia events must be recalled by participants—risking measurement error when events are unable to be recalled (e.g., nocturnal events are less likely to be recalled if the participant was asleep while experiencing the event). While guideline-defined severe hypoglycemia does not stipulate a glycemic threshold[30], non-severe hypoglycemia detection could be improved using rt-C/FGM. However, rt-C/FGM use remains low: in our sample only 11.6% reported using a rt-C/FGM device overall and 10.5% among participants with T2DM. Simlarly, the Behavioral Risk Factor Surveillance System observed low adoption of rt-C/FGM (4.1%) among Americans with diabetes in 2020 [66]. Relying on self-report ensured data collection among a large sample was feasible and mitigated the risk of selection bias due to rt-C/FGM inaccessibility.

The broad range of self-reported participant characteristics allowed us to control for confounding factors unmeasured in earlier observational studies. Adjustments for confounding balanced feasibility with granularity; however, residual confounding remains possible. We also did not account for multiple comparisons, which may have increased type I error. Instead, we based inferences on the estimated directions and magnitudes of effect and confidence intervals.

Finally, owing to the design of our questionnaire, we were unable to isolate the effect of second- versus first-generation or intermediate insulins specifically. Additionally, we could not ascertain potential differences between glargine U-300 and insulin degludec on rates of hypoglycemia, despite some research suggesting that their effects vary [6, 7, 9, 10, 13‐15, 63].