Acceptability and Feasibility of a Primary Care Hypoglycemia Prevention Program: A Pilot Study

Abstract

OBJECTIVE

Hypoglycemia, a major cause of morbidity for people with diabetes, is underrecognized in primary care, leading to underuse of risk mitigation strategies. This study reports the development and acceptability/feasibility testing of the Hypoglycemia Prevention Program (HPP), an electronic health record–integrated toolkit for primary care.

RESEARCH DESIGN AND METHODS

The HPP was developed through stakeholder engagement and included a pre-visit hypoglycemia history survey, primary care provider (PCP) hypoglycemia toolkit, and patient self-management handout. The HPP pilot study was a pre-post clinical trial at one primary care practice. Participants were adults with diabetes treated with insulin or sulfonylureas, their PCPs, and clinic staff. The primary outcome was acceptability; exploratory outcomes included pre- to postintervention change in PCP hypoglycemia prevention actions, self-reported hypoglycemia, and optional blinded 10-day continuous glucose monitoring (CGM).

RESULTS

The intervention was completed by 31 of 35 enrolled patients with diabetes (89%) for whom it triggered frequent hypoglycemia alerts. The intervention was rated acceptable or completely acceptable by 87% of patients and 100% of participating PCPs (n = 12) and clinic staff (n = 5). PCPs increased indicated glucagon prescribing from pre- to postintervention (0% preintervention to 32% postintervention, P = 0.003); there were small increases in CGM initiation and deintensification of hypoglycemia-causing medications. Self-reported and CGM-detected hypoglycemia decreased from pre- to postintervention, although with incomplete ascertainment and low CGM participation (n = 11).

CONCLUSIONS

A hypoglycemia prevention program designed for primary care is feasible to implement and acceptable to clinical and patient stakeholders, with evidence of improved hypoglycemia prevention efforts. Further efficacy evaluation is needed.

Graphical Abstract

Introduction

Hypoglycemia remains a major cause of morbidity and lower quality of life among adults with diabetes. Approximately one-half of the 40 million U.S. adults with diabetes are treated with insulin or insulin secretagogues, agents that can cause hypoglycemia (1,2). While secretagogue use declined over the past two decades, insulin use increased substantially (2–4). Each year, one-third of adults treated with these medication classes experience level 3 hypoglycemia (a severe event requiring assistance of another person) (5), which may directly cause falls, motor vehicle accidents, and seizures and is associated with adverse cardiovascular events, cognitive impairment, and mortality (6–9). Hypoglycemia also negatively impacts psychological health, contributes to anxiety among care partners, and lowers health-related quality of life (10–12).

In the U.S., there are substantial gaps in the use of effective hypoglycemia risk reduction practices, including continuous glucose monitoring (CGM), glucagon prescription, appropriate treatment deintensification, and referral to diabetes self-management education and support (DSMES) (13–16). Most diabetes care occurs within primary care, where there are likewise missed opportunities for hypoglycemia prevention (17). Guidelines recommend hypoglycemia history assessment at each clinical visit for people treated with insulin or secretagogues as a critical first step in mitigating hypoglycemia risk (18). However, our work has demonstrated that this assessment occurs in less than one-third of primary care visits and that patients with diabetes do not routinely disclose hypoglycemia to their treatment team (19–21).

We hypothesized that increasing hypoglycemia assessment in primary care visits would result in higher use of evidence-based preventive actions and ultimately reduce hypoglycemic events. This hypothesis is supported by findings from a national survey in which physicians were substantially more likely to appropriately deintensify insulin or secretagogues when they were made aware of a history of hypoglycemia (22).

Therefore, we developed the primary care Hypoglycemia Prevention Program (HPP), an intervention that embeds routine pre-visit hypoglycemia assessment into the primary care workflow coupled with actionable electronic health record alerts and orders. We report the development of the HPP intervention and findings from the HPP pilot study at one primary care practice.

Research Design and Methods

Study Design

The HPP pilot study was a single-site pre-post clinical trial to test the acceptability and feasibility of the HPP intervention (ClinicalTrials.gov identifier: NCT06353217). Participants were recruited from 22 May 2024 to 1 February 2025; data collection was completed on 1 February 2026. This trial was approved by the Johns Hopkins University School of Medicine Institutional Review Board, Baltimore, MD, and all participants provided written informed consent.

Setting and Population

This study was conducted at Johns Hopkins General Internal Medicine at Green Spring Station, a community-based suburban primary care practice. The study enrolled adults with diabetes at risk for iatrogenic hypoglycemia. Major inclusion criteria were age ≥18 years, diabetes of any subtype, and treatment with insulin, sulfonylureas, and/or meglitinides. Exclusion criteria were significant cognitive impairment or dementia, residence in a long-term-care facility, receipt of hospice or end-of-life care, serious illness impeding study participation, or planning to leave the area during the study period. The study also enrolled primary care providers (PCPs) delivering care to people with diabetes at the included practice and clinic staff who routinely participated in the patient check-in process.

Study Intervention

The HPP intervention was a set of tools integrated into the Epic electronic health record system (Epic Systems, Verona, WI) that were designed to assess pre-visit hypoglycemia history and generate guideline-concordant alerts and orders for the PCP within the clinical visit to facilitate hypoglycemia prevention (18,23,24). The intervention was developed using a participatory action research process in which an intervention development team met in five 2-h virtual meetings to define priorities and iteratively review and refine study materials and protocols (25). The intervention development team included three people with diabetes who had experienced hypoglycemia, two PCPs, one endocrinologist and clinical director, two primary care clinic staff, and one Epic physician builder. Key decisions included the scope of hypoglycemia history assessment, the survey instruments used, the content of the PCP-facing alerts and orders, and the hypoglycemia patient education materials to incorporate. Based on intervention development team input, hypoglycemia history was considered broadly to include history of hypoglycemic events, risk factors, and psychosocial measures critical for making diabetes treatment decisions. To assess these factors, we used patient-reported outcome measures validated in people with diabetes where available (26–30); otherwise, we created our own questions to meet the needs of the target population (Supplementary Table 1).

The final intervention had three components (Supplementary Fig. 1). First was the Hypoglycemia Patient Profile (Supplementary Table 2), a 25-question pre-visit patient survey with the following domains: hypoglycemia frequency and severity, hypoglycemia monitoring and treatment, impaired hypoglycemia awareness, fear of hypoglycemia, diabetes distress, food insecurity, and diabetes medication problems. This survey was designed to be completed during pre-visit registration on the Epic MyChart patient portal or on a tablet in the clinic waiting area. Second was the PCP Hypoglycemia Toolkit, which comprised alerts in the Epic OurPractice Advisory Storyboard (located in the left-hand column of the Epic clinician screen) and orders in the form of an Epic SmartSet. The Storyboard and SmartSet features are core Epic user interface components that are broadly available. The PCP alerts and orders were generated only when applicable based on responses to the Hypoglycemia Patient Profile and clinical data (Supplementary Table 3). SmartSet orders included blood glucose monitor prescription, CGM prescription, glucagon prescription, endocrinology referral, and case management referral. The order set also linked to the FindHelp Epic platform for patients who screened positive for food insecurity, which provided local resources (31). During intervention development, the orders also included referral to DSMES, but this was removed, as the service was unavailable in our system during the study period. Third was a hypoglycemia self-management handout populated into the patient’s after-visit summary, which was developed by the Association of Diabetes Care & Education Specialists, used with permission for this study (32). PCPs and clinic staff were trained in the use of the intervention in one 30-min staff meeting and provided a two-page user guide.

Participant Recruitment and Flow

PCPs and clinic staff were recruited by direct outreach from the study team. Patients with diabetes were recruited by identifying all individuals with an upcoming visit to an enrolled PCP and who were potentially eligible based on insulin, sulfonylurea, or meglitinide treatment. PCPs were provided a list of their potentially eligible patients and could indicate that they were not appropriate for the study. Otherwise, a letter was mailed to the potentially eligible patient, followed by a phone call from study staff to confirm eligibility and offer study participation.

Patients with diabetes were followed through two regularly scheduled PCP visits (Supplementary Fig. 2). The first visit was a preintervention visit where they received usual care. For the second visit, the HPP intervention was active. Nonroutine visits (e.g., urgent visits, preoperative evaluations) were excluded. The HPP intervention was activated after all participants’ preintervention visits were completed to avoid contamination. Participating patients and PCPs were informed when the intervention was active. Patients who completed the study received a $100 gift card. PCPs and clinic staff who completed the study received a $50 gift card.

Outcomes

The primary outcome was overall acceptability measured on a five-point Likert scale using an intervention evaluation survey adapted from the Theoretical Framework of Acceptability Questionnaire (33). Acceptability was assessed among patients with diabetes immediately after the clinic visit where they used the intervention and among PCPs and clinic staff at the end of the study.

Six additional domains of acceptability were assessed on the intervention evaluation survey. Feasibility outcomes were recruitment and retention rates, time needed to complete the Hypoglycemia Patient Profile (assessed by direct observation in the clinic waiting area), and the duration of PCP visits. Fidelity outcomes were completion of Hypoglycemia Patient Profile items and PCP use of each PCP Hypoglycemia Toolkit component. Outcomes related to clinical effectiveness were exploratory and were evaluated to understand feasibility of data collection and for estimating potential effect size and variability. Pre- and postintervention differences in PCP hypoglycemia prevention actions were ascertained by electronic health record review, which included glucagon prescription among patients with indications (insulin use, level 2 or 3 hypoglycemia, or impaired hypoglycemia awareness, with no current glucagon prescribed), CGM prescription among those with indications (insulin use or level 2 or 3 hypoglycemia, with no current CGM use), and deintensification (dose reduction, discontinuation, or switching) of insulin, sulfonylurea, or meglitinide therapy. We assessed pre- and postintervention differences in self-reported hypoglycemia frequency (within the past 3 months) among patients with diabetes by comparing their Hypoglycemia Patient Profile responses at the time of their intervention visit and a repeat survey 4–12 weeks later. The 4- to 12-week interval was used to allow time for changes made at the intervention visit to take effect. Hemoglobin A_1c (A1C) was ascertained from the electronic health record, with preintervention defined as the most recent value within 6 months before the intervention visit and postintervention defined as the next value >60 days after the intervention visit. Patient-reported outcomes were measured at baseline and after the intervention visit. These included the Patient Satisfaction Questionnaire Short Form (PSQ-18) (score range 18–90, with a higher score indicating greater satisfaction with medical care) (34) and the Patient Activation Measure Short Form (PAM-10) (score range 0–100, with a higher score indicating greater patient skill and confidence in self-management) (35).

Participating patients with diabetes were invited to complete pre- and postintervention CGM, which was not required for study participation. We used the Dexcom G6 Pro, a blinded CGM system (participants could not see glucose values) that measures glucose every 5 min for up to 10 days. Trained research staff placed the CGM on the patient’s abdomen at baseline and 4–12 weeks after the intervention visit. After a 10-day wear period, participants removed the device and returned it to the study team by mail. Participants received a $25 gift card for each CGM wear. CGM outcomes included the number of minutes per day glucose was <70 and <54 mg/dL, time in range, mean glucose, and glucose coefficient of variation.

Other Measurements

Self-reported race, ethnicity, sex, socioeconomic measures, and diabetes duration were ascertained by questionnaire. Diabetes subtype, treatment, comorbidities, and CGM use were ascertained from the electronic health record.

Statistical Analysis

Our target sample size of 35 people with diabetes was selected based on feasibility and to allow sufficient experience with study procedures. The primary acceptability outcome was descriptive, with >80% of participants reporting acceptable or completely acceptable responses considered to be high acceptability (33). Pre- and postintervention differences in outcomes were compared using nonparametric statistical tests for paired data, including McNemar χ² test for binary outcomes and Wilcoxon signed rank test for multicategorical or continuous outcomes. We performed post hoc exploratory analyses stratified by baseline CGM use. Although this pilot study was not powered to detect differences, we report P values for descriptive purposes, with P < 0.05 considered statistically significant.

Results

Study Participants

The study enrolled 35 patients with diabetes, 12 PCPs, and 5 clinic staff. The mean patient age was 68 years, 46% were female, and multiple races and ethnicities were represented (Table 1). Most patients (86%) had type 2 diabetes, 9% had type 1 diabetes, and 6% had another diabetes subtype (one with posttransplantation diabetes and one with latent autoimmune diabetes in adults). At baseline, approximately two-thirds of patients used long-acting insulin, one-third used rapid-acting insulin, and 57% used CGM. PCP and clinic staff characteristics can be found in Supplementary Table 4.

Table 1.

Baseline characteristics of enrolled patients with diabetes (n = 35)

Characteristic	Finding
Age, years, mean (SD)	68.1 (12.0)
Aged ≥65 years, n (%)	22 (62.9)
Female sex, n (%)	16 (45.7)
Race, n (%)
Asian	2 (5.7)
Black or African American	10 (28.6)
White	21 (60.0)
Other	2 (5.7)
Hispanic ethnicity, n (%)	1 (2.9)
Diabetes subtype, n (%)
Type 2	30 (85.7)
Type 1	3 (8.6)
Other	2 (5.7)
Years since diabetes diagnosis, mean (SD)	17.6 (9.8)
A1C, %, mean (SD)	7.4 (1.7)
A1C, mmol/mol, mean (SD)	57 (19)
CGM use, n (%)	20 (57.1)
Diabetes medication use, n (%)
Long-acting insulin	24 (68.6)
Rapid-acting insulin (scheduled)	10 (28.6)
Rapid-acting insulin (sliding scale)	11 (31.4)
Sulfonylurea	9 (25.7)
Metformin	18 (51.4)
Thiazolidinedione	1 (2.9)
Dipeptidyl peptidase 4 inhibitor	3 (8.6)
Sodium–glucose cotransporter 2 inhibitor	11 (31.4)
Glucagon-like peptide 1 receptor agonist	11 (31.4)
Tirzepatide	7 (20.0)
Total number of medications, mean (SD)	11.6 (4.3)
Comorbidities, n (%)
Coronary artery disease	13 (37.1)
Heart failure	4 (11.4)
Stroke	5 (14.3)
Diabetic eye disease	7 (20.0)
Diabetic neuropathy	16 (45.7)
Chronic kidney disease	18 (51.4)
Highest educational attainment, n (%)
Some high school	1 (2.9)
Completed high school	5 (14.7)
Some college	7 (20.6)
Completed college degree (BA, BS)	7 (20.6)
Postgraduate work	14 (41.2)
Employment status, n (%)
Full time	8 (22.9)
Part time	4 (11.4)
Retired	17 (48.6)
Disabled/unable to work	6 (17.1)
Annual household income, $, n (%)
<14,999	0 (0)
15,000–29,999	6 (17.6)
30,000–44,999	1 (2.9)
45,000–59,999	4 (11.8)
60,000–74,999	7 (20.6)
75,000–89,999	2 (5.9)
≥90,000	14 (41.2)

BA, bachelor of arts; BS, bachelor of science.

Intervention Acceptability

The intervention was found to be acceptable or completely acceptable by 87% of patients, 100% of PCPs, and 100% of clinic staff (Fig. 1). Acceptability was high across most domains, although lower for PCP confidence in the ability to use the intervention (67% confident or highly confident), and for fairness, which had a higher response indicating no opinion compared with other acceptability domains (Fig. 1 and Supplementary Table 5). All PCPs agreed or strongly agreed that the intervention was effective for hypoglycemia reduction and improved their clinical care.

Figure 1.

Intervention acceptability to patients with diabetes, PCPs, and clinic staff. The proportion of respondents indicating that the intervention was acceptable or completely acceptable overall (primary outcome) and in each acceptability domain. Only PCPs were asked whether the intervention improved their clinical care.

Intervention Feasibility and Fidelity

The recruitment rate for patients with diabetes from telephone outreach was 33% (35 enrolled of 106 called) (Supplementary Fig. 3). All 12 eligible PCPs and 5 eligible clinic staff at the practice were enrolled. Of the 35 patients enrolled, 31 (89%) completed the intervention clinic visit, which included use of the Hypoglycemia Patient Profile and completion of the primary acceptability outcome. All PCPs and clinic staff (100%) completed the primary acceptability outcome.

We evaluated process measures indicating the use and fidelity of intervention components. All 31 patients who completed an intervention clinic visit used the Hypoglycemia Patient Profile (65% via MyChart, 35% on a tablet in the clinic waiting area). Individual item completeness was high, with a mean (SD) 23.9 (4.1) of the 25 items completed. The completion time was a mean (SD) of 6.0 (1.8) min by direct observation of those who completed it in the waiting area. PCPs reported different rates of using the individual components of the Provider Hypoglycemia Toolkit (Supplementary Table 6). Most PCPs reported reviewing OurPractice Advisory alerts and Hypoglycemia Patient Profile responses for at least one-half of their patients. There was lower reported use of the hypoglycemia SmartSet orders, with a majority of PCPs reporting using it sometimes. The duration of the PCP visit was similar pre- and postintervention (33.4 vs. 34.7 min; P = 0.34).

Responses to the Hypoglycemia Patient Profile Triggering Alerts

At the intervention clinic visits, Hypoglycemia Patient Profile responses frequently triggered clinical alerts, and all patients had one or more alert value (Fig. 2). Notably, 36% of patients reported level 2 hypoglycemia (blood glucose <54 mg/dL) within the past 3 months, and 7% reported level 3 (severe) hypoglycemia. More than one-third of patients were not routinely carrying carbohydrates for hypoglycemia treatment, and more than one-third had no access to glucagon when indicated. There were also high rates of impaired hypoglycemia awareness (47%), diabetes distress (46%), and diabetes medication problems (43%), which included missing or skipping doses or taking diabetes medications not as instructed. CGM users, compared with nonusers, had more frequent alerts for hypoglycemia, impaired hypoglycemia awareness, and diabetes medication problems (Fig. 2).

Figure 2.

Frequency of responses to the Hypoglycemia Patient Profile triggering clinical alerts overall and by baseline CGM use. The proportion of Hypoglycemia Patient Profile responses triggering each clinical alert among all patients who completed the intervention (n = 31), CGM users (n = 18), and CGM nonusers (n = 13).

Exploratory Glycemic and Patient-Reported Outcomes

Among the 31 patients who completed the intervention, there was complete ascertainment of self-reported hypoglycemia for 74% (Table 2). Eleven patients (36%) completed both pre- and postintervention optional CGM. Among CGM completers, the median CGM wear time was 9.8 days preintervention and 9.9 days postintervention.

Table 2.

Completeness and pre- to postintervention change in exploratory process and clinical outcomes among patients with diabetes who completed the intervention (n = 31)

Outcome	Preintervention	Postintervention	P
Hypoglycemia prevention processes
Complete outcome data		31 (100.0)
Glucagon prescription with indications, n (%)*	0 (0.0)	9 (32.1)	0.003
CGM initiation with indications, n (%)*	0 (0.0)	1 (14.3)	0.32
Deintensified insulin or sulfonylurea, n (%)	2 (6.5)	4 (12.9)	0.41
Self-reported hypoglycemia frequency, n (%)†
Complete outcome data		23 (74.2)
Level 1 hypoglycemia			0.51
No episodes	6 (26.1)	9 (39.1)
Once every few months	7 (30.4)	3 (13.0)
Once per month	2 (8.7)	3 (13.0)
Once per week	4 (17.4)	5 (21.7)
More than once per week	4 (17.4)	3 (13.0)
Level 2 hypoglycemia			0.092
No episodes	14 (60.9)	18 (78.3)
Once every few months	3 (13.0)	2 (8.7)
Once per month	2 (8.7)	2 (8.7)
Once per week	2 (8.7)	1 (4.4)
More than once per week	2 (8.7)	0 (0.0)
Level 3 hypoglycemia			0.56
No episodes	20 (90.9)	21 (95.5)
Once every few months	2 (9.1)	1 (4.6)
CGM outcomes (participation optional), mean (SD)‡
Complete outcome data		11 (35.5)
Minutes per day <70 mg/dL	11.5 (24.2)	1.1 (2.6)	0.15
Minutes per day <54 mg/dL	2.0 (4.8)	0 (0)	0.084
TBR <70 mg/dL, %	0.8 (1.7)	0.1 (0.2)	0.15
TBR <54 mg/dL, %	0.1 (0.3)	0 (0)	0.084
TIR 70–180 mg/dL, %	63.8 (28.4)	62.4 (23.6)	0.79
TAR >180 mg/dL, %	35.0 (28.9)	37.5 (23.6)	0.66
Glucose, mg/dL	174 (41)	174 (35)	0.86
CV glucose, mg/dL	27 (8)	25 (4)	0.37
Glycemic outcomes, mean (SD)§
Complete outcome data		17 (54.8)
A1C, %	7.9 (2.4)	8.4 (2.7)	0.14
A1C, mmol/mol	63 (26)	68 (30)	0.14
Patient-reported outcomes, mean (SD)
Complete outcome data		31 (100.0)
Patient Satisfaction Questionnaire Short Form (PSQ-18) score	73.7 (8.5)	76.5 (8.2)	0.040
Patient Activation Measure Short Form (PAM-10) score	69.4 (16.8)	68.5 (15.7)	0.63

CV, coefficient of variation; TAR, time above range; TBR, time below range; TIR, time in range.

*There were 28 patients with indications for glucagon: insulin use, level 2 or 3 hypoglycemia, or impaired hypoglycemia awareness with no current glucagon prescribed. There were seven patients with indications for CGM: insulin use or level 2 or 3 hypoglycemia with no current CGM use.

†Comparing self-reported hypoglycemia within the past 3 months ascertained at the intervention visit (preintervention) and 4–12 weeks later (postintervention).

‡Comparing 10-day CGM at baseline (preintervention) and 4–12 weeks after the intervention visit (postintervention).

§Comparing the most recent value within 6 months before the intervention visit (preintervention) to the next value >60 days after the intervention visit (postintervention).

All PCP actions to prevent hypoglycemia increased from pre- to postintervention (Table 2), with a substantial increase in glucagon prescription when indicated (0% preintervention to 32% postintervention; P = 0.003). There were small increases in CGM initiation when indicated (0% preintervention to 14.3% postintervention; P = 0.32) and deintensification of insulin or sulfonylureas (7% preintervention to 13% postintervention; P = 0.41). There were no significant differences in pre- and postintervention self-reported hypoglycemia, although the proportion reporting hypoglycemia in the past 3 months decreased across all levels of severity. This was especially notable for level 2 hypoglycemia, which decreased from 39% reporting hypoglycemia preintervention to 22% postintervention and was driven by transitions from frequent hypoglycemia (once per week or more) to lower frequencies (Supplementary Fig. 4). There was a nonsignificant pre- to postintervention increase in A1C from a mean of 7.9% (63 mmol/mol) to 8.4% (68 mmol/mol) (P = 0.14). Change in A1C was driven by two high outlier patients (Supplementary Fig. 5), neither of whom had changes to their hypoglycemia-causing medications at the intervention visit. Patient satisfaction scores increased significantly from pre- to postintervention; patient activation scores were similar (Table 2). Pre- to postintervention changes in self-reported hypoglycemia and A1C were similar between CGM users and nonusers (Supplementary Table 7).

Exploratory CGM Outcomes

Among the 11 patients who completed both optional CGM measurements, there was a nonsignificant decrease in the mean minutes per day <70 mg/dL (from 11.5 to 1.1 min; P = 0.15) and minutes per day <54 mg/dL (from 2.0 to 0.0 min; P = 0.084). CGM time in range and CGM mean glucose were similar pre- and postintervention.

Adverse Events

There was one study-related adverse event in which one patient had a self-limited skin rash from the study CGM.

Conclusions

This pilot study demonstrated that the HPP was highly acceptable to patients with diabetes and clinical stakeholders. The intervention was feasible to implement with high fidelity within a primary care workflow. There was a substantial pre- to postintervention increase in glucagon prescribing. These findings indicate that the HPP is a promising intervention to promote hypoglycemia risk reduction in primary care practice and should be evaluated for clinical efficacy.

This study is the first intervention to our knowledge that focuses on increasing comprehensive hypoglycemia assessment and prevention in primary care. Recently, a 6-month clinical pharmacist intervention aimed at optimizing diabetes pharmacotherapy was effective at decreasing insulin and sulfonylurea treatment and reducing hypoglycemic events (36). While our intervention has not yet been tested for clinical efficacy, this work adds to the growing understanding that practice-based interventions could potentially address major gaps in hypoglycemia and improve diabetes outcomes.

Based on stakeholder input, our Hypoglycemia Patient Profile included a broad assessment of hypoglycemia history, risk factors, and psychosocial impact. In addition to uncovering a high frequency of clinically significant hypoglycemia, this survey found a substantial proportion of patients with major gaps in hypoglycemia risk reduction practices, including carrying carbohydrates for hypoglycemia treatment and access to glucagon. Given that the study population was relatively affluent, some risk factors, such as food insecurity, were less common, but we expect these aspects to be highly relevant in lower resourced populations (37). Our study population also had a high baseline use of CGM, and CGM users had higher rates than nonusers of hypoglycemic events, impaired hypoglycemia awareness, and diabetes medication problems. These findings support the need for comprehensive hypoglycemia history assessment, including among CGM users, for revealing important opportunities to reduce hypoglycemia risk.

To facilitate dissemination of the HPP tools, we may need to balance comprehensive assessment versus survey burden. Our Hypoglycemia Patient Profile contained 25 items and took, on average, 6 min to complete in the clinic waiting area; completion time may have been shorter for people who responded on the MyChart patient portal, but we were not able to measure this. Future work will consider a short form version of the Hypoglycemia Patient Profile, embedding skip logic to reduce the total number of questions, and asking certain items at longer intervals (e.g., annually rather than at every visit).

The overall acceptability of the HPP intervention was high, especially for perceived effectiveness to reduce hypoglycemia and improve clinical care. On acceptability surveys, a number of PCPs and patients had neutral responses to the intervention’s fairness, which was a domain on the Theoretical Framework of Acceptability Questionnaire. We suspect that this was due to the fairness domain being less relevant to the intervention.

Feasibility evaluation of the HPP indicated areas for potential optimization. There was high use of the Hypoglycemia Patient Profile by enrolled patients. PCPs frequently reviewed the Hypoglycemia Patient Profile and associated alerts, but used SmartSet orders inconsistently. While PCP glucagon prescribing increased substantially, there were only small increases in CGM prescription and high-risk medication deintensification. We plan postintervention qualitative research to identify opportunities to optimize use of the SmartSet and associated orders, and improve PCP training. We will also explore barriers external to this intervention, such as lack of insurance coverage for diabetes technology, that can be targeted in future interventions.

This study was not powered for demonstrating effects on clinical hypoglycemia outcomes; findings yielded valuable information for planning future efficacy evaluation. We measured hypoglycemia by both self-report and blinded CGM, which each have advantages and disadvantages. There is a growing understanding that self-reported and CGM-detected hypoglycemia represent different constructs that do not entirely overlap but are both clinically important (38). Self-reported hypoglycemia is strongly associated with patient outcomes and quality of life, is clinically relevant to diabetes treatment decisions, and was the primary outcome in major trials (10–12,18,39). However, self-report may miss events among people with impaired awareness of hypoglycemia, which was prevalent in this study population, or with suboptimal glycemic monitoring (40). There is also the possibility of recall bias, which may be reduced by adjudicating hypoglycemic events (39). Furthermore, ascertainment of self-reported hypoglycemia was incomplete in this study, indicating a need to devote more resources to participant retention for this outcome. CGM detects all hypoglycemia, including asymptomatic events, but is less accurate in lower ranges and may have false positives from compression of the sensor site (23). In addition, only a minority of participants completed the optional CGM measurements in this study, indicating that requiring CGM for participation would likely limit enrollment. Given the high prevalence of baseline CGM use, a potential alternative strategy may be to ascertain outcome data from patients’ own CGM devices.

We observed a pre- to postintervention increase in A1C that was not statistically significant. This finding was driven by two participants whose PCPs did not deintensify their diabetes medications, suggesting that the change in A1C was unrelated to study interventions. It is possible that changes in patient behavior to prevent or manage hypoglycemia resulted in higher glucose levels. However, prior interventions to reduce hypoglycemia did not increase A1C, so we suspect that harmful glycemic effects from hypoglycemia prevention strategies are unlikely (18,36); this should be confirmed in future efficacy studies. Given that complete A1C data were only available in 55% of participants through the electronic health record, longer follow-up and/or study-directed A1C measurement may be needed in a future efficacy trial.

This pilot study demonstrated the importance of intervention adaptability for broader dissemination. While the intervention was designed to facilitate DSMES referral where available, our site did not have DSMES, so this order was deactivated. It is likely that other clinical orders in the intervention will need to be changed to reflect available resources, which we will explore in a future implementation evaluation. Notably, the HPP was designed to provide patient hypoglycemia education, given that DSMES is not available in many practices nationally (16).

Strengths of this study included its development through an evidence-driven and stakeholder-informed process incorporating input from people with diabetes with a history of hypoglycemia. The intervention was fully integrated into Epic, the most commonly used electronic health record system, to facilitate dissemination. All PCPs and clinic staff at the study site were enrolled, minimizing the potential for selection bias that could occur if PCPs with a greater interest in diabetes preferentially joined the study. There was high retention, with 89% of enrolled patients and all participating PCPs and clinic staff completing the primary outcome.

There were also several notable limitations. As a pre-post study, there was no evaluation of long-term intervention use or sustainability. The study was conducted at a single practice site, and future work will be needed to ensure that the intervention can be translated to other practices and health systems. The study should not be interpreted as demonstrating the effects of the intervention on clinical hypoglycemia outcomes for which the study was underpowered, had incomplete ascertainment, and had a potential for confounding by temporal trends or regression to the mean.

Overall, this pilot trial successfully demonstrated high acceptability of a stakeholder-informed primary care hypoglycemia prevention intervention. The findings provide a promising roadmap for enhancing hypoglycemia prevention in primary care.

This article contains supplementary material online at https://doi.org/10.2337/figshare.32794980.

Funding Statement

This study was supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grant K23DK128572. S.J.P. received research funding from NIDDK grants R03DK144349, R01DK126825, and R01DK139324 and Resolve to Save Lives. R.G.M. and A.R.-L. were supported in part by National Institute on Aging (NIA) grant P30AG097158 and are investigators at the University of Maryland Institute for Health Computing, which is supported by funding from Montgomery County, MD, and the University of Maryland Strategic Partnership: MPowering the State, a formal collaboration between the University of Maryland, College Park, and the University of Maryland, Baltimore. A.R.K. was supported by NIA grant K01AG084971. N.N.M. received research funding from Breakthrough T1D, Type 1 Diabetes Exchange, Samsung Research America, and NIDDK grants R01DK125780 and R01DK134955. N.M.M. received research funding from NIDDK grants R01DK126825, 1U01DK134975, R03DK135898, and R03DK144349; the American Diabetes Association; and Maryland Cigarette Restitution Fund grant M00B5600171.

The funder had no role in the study design; collection, analysis, and interpretation of data; writing the report; or the decision to submit the report for publication.