Diabetes technology use in U.S. hospitals: A survey of user experiences

1 |. INTRODUCTION

Continuous glucose monitors (CGMs) and automated insulin delivery (AID) systems have transformed outpatient diabetes management, improving glycemic outcomes and quality of life.^1,2 In the hospital setting, they offer several advantages over conventional management involving multiple daily injections (MDI) and intermittent glucose monitoring.³ This, combined with their outpatient success and growing adoption, has prompted societal support for inpatient use.^4–6 However, current real-world use in hospitals remains relatively unknown and poorly understood. This study aimed to ascertain the patterns of CGM and AID system continuation during hospital encounters in the United States (U.S.) and assess their impact on self-reported glycemic control.

2 |. RESEARCH DESIGN AND METHODS

We conducted an online survey from May to November 2023, hosted on Qualtrics. Participants were recruited via diabetes-related networks and social media platforms (Supporting Information S1). Eligibility criteria, assessed in the first three survey questions, included individuals aged ≥18 years with (or caregivers of individuals with) any type of diabetes who presented to the hospital in the U.S. within the prior year and used a CGM, insulin pump, or both at the time of hospital encounter.

The survey (Supporting Information S2) included demographics, diabetes characteristics, and hospital encounter details. Participants also reported on the extent and duration of device use during their most recent encounter, reasons for discontinuation, and perceptions of glycemic control. Device groups (CGM, manual pump, AID) were based on baseline use and were not mutually exclusive; individuals could be included in multiple groups if applicable. The nine individuals using predictive low glucose suspend (PLGS) systems were classified as AID users. Analyses were conducted in GraphPad Prism. Fisher’s exact tests were used to compare glycemic control perceptions among those who used and did not use their devices during the encounters. Respondents did not receive compensation for their participation. Approval for survey distribution and analysis was provided by Stanford University’s Institutional Review Board.

3 |. RESULTS

Of the 551 survey responses received, 257 were included for analysis after excluding 173 ineligible responses, 94 incomplete entries, and 27 duplicates. Reasons for ineligibility included no hospital encounter in the prior year (n = 112), residing outside the U.S. (n = 32), lack of diabetes device use at baseline (n = 23), and absence of a diabetes diagnosis (n = 6). Among eligible responses, the majority (93%) had Type 1 diabetes and 69% identified as female. The mean age was 49 ± 20 years. Participants were predominantly white (95%), with nearly two-thirds covered by private insurance and 33% on public insurance, such as Medicare or Medicaid.

Nearly all respondents reported CGM outpatient use at the time of presentation (98%), most commonly the Dexcom G6 (82%). AID systems were used by 69%, of which Tandem Control IQ (56%) and Insulet Omnipod 5 (24%) were the most frequently reported. A smaller proportion (14%) were using manual insulin pumps (without automation). Detailed participant characteristics and baseline technology use are available in Supporting Information S3.

Hospital encounters were reported from 45 states (Supporting Information S4) and a variety of settings, including 57% urban, 36% suburban, and 7% rural facilities. Care was received at over 190 distinct hospitals, most of which were represented by only one or two participants. Seven hospitals had slightly higher representation: one hospital with five participants, two with four, and four with three. Among respondents, 52% had overnight hospital admissions (median length of stay 3.8 days, IQR 2.0–5.3), 11% had ER stays lasting ≥12 h, and 37% were admitted for same-day procedures. The most common reasons for hospital presentation included surgical procedures (31%), diabetes-related complications (21%), infections (12%) and gastrointestinal issues (10%) (Supporting Information S5).

Figure 1 illustrates the proportions of respondents who actively used their CGMs, non-AID insulin pumps, and AID systems during their hospital encounter. The CGM continuation rate was 91%, typically active for most of the admission. Among AID users, 79% continued their systems, in most cases functioning in automated mode for more than 90% of the stay. Despite the commonness of CGM use, nearly half of respondents reported receiving three or more fingerstick blood glucose (FSBG) checks per day, while 26% had 1–2 checks, and 25% reported no FSBG checks at all.

FIGURE 1.

Proportions of respondents who actively used their CGM, insulin pump, and/or AID system during hospital encounters. Survey participants reported whether they used each diabetes device during their hospital stay. If used, they indicated the proportion of time the device was active: <50%, 50–90%, or >90% of their stay. The percentages above the bars show overall device usage, while the stacked bars represent the proportion of time each device was used. For AID systems, participants were specifically asked about whether the automated mode for their system was used and to indicate its usage as a proportion of the encounter. Categories (e.g., CGM, AID) are not mutually exclusive; individuals are included in each category for which they reported using the applicable device(s).

Some participants reported advocating unsuccessfully for continued device use during their stay. Requests to use CGMs, manual insulin pumps, and AID systems during an encounter were made by 180 participants (71% of CGM users), 28 participants (76% of manual pump users), and 98 participants (56% of AID system users), respectively. These requests were denied in 11 cases for CGMs (6.1%), 2 cases for manual pumps (7.1%), and 16 cases for AID systems (16%). Hospital policies or protocols were the most frequently cited reasons for the non-use of CGMs (9 occurrences) and AID systems (6 occurrences), often reflecting institutional restrictions on device use (Supporting Information S6). Other notable barriers included lack of device availability, staff unfamiliarity, clinical procedures/conditions, individual preferences, and technical device issues. Discontinuation rates of CGM and AID across key participant and hospital characteristics of interest are displayed in Supporting Information S7.

4 |. PERCEIVED GLYCEMIC CONTROL

People using AID systems reported significantly more favourable perceptions of glycaemic control than those using CGMs without AID or no CGM (Figure 2).

FIGURE 2.

Comparison of respondent perceptions of glycemic control based on device use. Likert scale responses to the statement, “My blood sugar was well controlled in the hospital,” were categorized in three groups, based on respondent-reported device use: No CGM use, CGM use without an AID system, and CGM use with an AID system in automated mode. Participants rated their agreement as Strongly disagree (dark orange), Disagree (light orange), Neither agree nor disagree (yellow), Agree (light green), or Strongly agree (dark green). The proportions of respondents who agreed (Agree/Strongly Agree) were compared across groups using Fisher’s exact test. Significant differences were identified between those using AID and those using CGM without AID (p = 0.0004) as well as those not using CGM (p = 0.004). Differences with p < 0.05 are indicated by an asterisk (*) in the figure. Group numbers are slightly reduced from the overall cohort due to incomplete responses. Manual insulin pump users were included in the non-CGM or CGM without AID groups, as their numbers were too small for separate analysis. Bars are visually centred on the “Neither agree nor disagree” (neutral) category to make it easier to visually compare the distributions of positive and negative responses. NS, not significant.

5 |. CONCLUSIONS

In this study, the continuation of personal diabetes devices during hospital encounters was common, and AID use was associated with more favourable perceptions of glycemic control.

Respondents frequently advocated for these device continuations, likely reflecting their familiarity with and trust in the technologies. However, some encountered barriers in hospitals without clear policies. While standardised protocols and staff training could support broader use, developing and implementing these is not a straightforward task for an institution.⁷ This is especially true regarding inpatient AID, where published guidance is sparse, and major gaps include when to permit automated mode use and how to document actual insulin delivery during automation.³ In the absence of clear protocols, staff may rely on familiar practice, and patients may feel ignored. This variability is also reflected in the wide range of fingerstick check frequencies reported by respondents, likely a consequence of differing local protocols (or lack thereof). While recent guidance recommends a “hybrid” approach to assessing CGM accuracy in the hospital,⁶ it stops short of specifying how that validation should be operationalised. These findings highlight the need for implementable guidance and further evaluation of how patients’ diabetes technologies can be used in the hospital.

Although our ability to detect individual-level differences is limited by the respondent population, several interesting discontinuation trends (Supporting Information S7) warrant further investigation: CGM and AID discontinuation were more common among respondents under 18, those with lower educational attainment, lower income, and public insurance. Hospital encounters at urban hospitals and overnight admissions also saw higher rates of device discontinuations–patterns which may reflect the influence of more structured or restrictive hospital policy environments.

This study has several important limitations, and therefore the results should be interpreted with some caution. The online survey was distributed through diabetes networks and social media, potentially introducing selection bias toward individuals more engaged with diabetes management and technology. Requiring baseline diabetes technology use may have added to this. Most respondents were white, female, and had Type 1 diabetes, which may limit the generalisability, particularly since T1D only accounts for about 5% of inpatient diabetes cases in the U.S.⁸ Additionally, the self-reported nature of the survey introduces potential recall and interpretation biases that could affect the accuracy of reported experiences.

Another limitation is the lack of more detailed clinical context, such as clinical acuity, hospital unit type, or specific institutional policies, which could have simultaneously and bi-directionally influenced both device use (even per protocol) and perceptions of control. For example, certain factors not systematically captured, such as diabetes-related ketoacidosis (DKA), cognitive impairment, and imaging, were mentioned in a few open-text responses (Supporting Information S6) but cannot be assessed across the full cohort.

Despite these limitations, the findings offer valuable insights into the broader use of personal diabetes devices in hospital settings—an area that is largely unreported and unexplored. Notably, the findings suggest that use is not rare, but is instead relatively common and widespread. While the study was limited to U.S. respondents, the challenges observed are likely not unique to the U.S. and may reflect broader issues relevant to inpatient care in other countries as well. Clear guidance for hospitals is needed. Future research should focus on the impact of inpatient CGM and AID system use on clinical outcomes, patient satisfaction, and healthcare costs and should examine how these technologies can be scaled across diverse hospital settings to support equitable access.³

Supplementary Material

Additional supporting information can be found online in the Supporting Information section at the end of this article.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.