Swimming With the Omnipod 5 Automated Insulin Delivery System: An Observational Investigation of Swimmers With Type 1 Diabetes in the Omnipod 5 Pivotal Trial

Abstract

Aims:

Limited studies have investigated glucose management around swimming in type 1 diabetes (T1D). This observational study of the Omnipod 5 Automated Insulin Delivery (AID) System assessed glycemic patterns and system operation with AID use during swimming.

Materials and Methods:

A total of seven swimmers (aged 11 ± 2 years, 57% female) with T1D recorded their swims using an activity tracker and completed online surveys after swims.

Results:

Participants reported 124 swims lasting 23–146 min. Median (IQR) starting glucose was 133 (93–178) mg/dL, with a change of −17 (−61 to 7) mg/dL during swims. Carbohydrate intake 2-h before swims was 20 (0–32) grams. Starting glucose ≥ 181 mg/dL showed the largest glucose decreases [−86 (−111 to −25) mg/dL] during swims. Activity feature was used in 87% of swims with a starting glucose < 90 mg/dL. Automated Mode was maintained for 98% (95%–100%) of swim time with 58% (24%–70%) of this time in Automated Mode: Limited.

Conclusions:

Swimming with AID was associated with generally modest declines in glucose. The Omnipod 5 System largely preserved automation in water (often via Automated Mode: Limited) without reverting to Manual Mode. Starting glucose dominated glycemic trajectory, with larger declines but low hypoglycemia risk at a higher starting glucose, and smaller changes, but more hypoglycemia at a lower starting glucose. Insulin suspension was not enough to reliably prevent hypoglycemia when starting glucose was < 180 mg/dL, highlighting the need for proactive carbohydrate intake with a lower starting glucose, even with AID.

1 |. Introduction

Adoption of diabetes technology has increased steadily over the past decade [1–3], and automated insulin delivery (AID) systems represent a fundamental advancement leading to improvements in glycemic outcomes for youth with type 1 diabetes (T1D) [4–6]. An important application of AID systems is in their use during exercise. Although consensus guidelines recommend youth with T1D should aim for at least 60 min of moderate-to-vigorous physical activity (MVPA) per day [7], many youth do not appear to follow exercise consensus guidelines [8, 9] and are not meeting these recommendations [10]. Exercise and regular physical activity are often avoided by individuals with T1D for various reasons including fear of hypoglycemia, disruption to glycemic management and lack of time [11, 12].

Glucose responses to exercise can vary depending on activity type, intensity and duration [7, 13]. Aerobic activities such as swimming that tend to be longer in duration and lower intensity increase the risk of hypoglycemia, including delayed or overnight episodes [13]. Recommended strategies to decrease this risk include reducing basal insulin by 50%–80%, setting a higher exercise target before exercise, temporarily suspending or disconnecting the insulin pump, reducing the prandial bolus by 25%–75% at the preceding meal, and/or consuming carbohydrates before or during exercise [7, 13, 14]. For prolonged exercise (> 2 h), pump suspension is typically discouraged due to the risk of rebound hyperglycemia and potential development of diabetes-associated ketoacidosis (DKA) [7, 13].

Although guidance and strategies exist on managing glycemia around exercise in T1D, few studies have specifically investigated it during swimming [15–17]. This represents a major knowledge gap, as water immersion introduces additional challenges for people using diabetes devices. In particular, water immersion can disrupt CGM-to-Pod communication, potentially shifting AID systems into constrained operating states. Real-world data are therefore needed to understand whether automation can be preserved and how glycemia behaves during aquatic activities when using AID.

When operating in Automated Mode, the Omnipod 5 System adjusts basal insulin delivery every 5 min based on delivery history as well as current and forecast CGM values [13]. Because Bluetooth communication between the CGM and Pod can be disrupted in water, CGM signal interruptions may occur during swimming. During prolonged CGM signal loss (≥ 20 min), the system will move into Automated Mode: Limited, in which delivery is constrained to a static basal rate: no more than the user’s pre-programmed basal rate that would be active during Manual Mode [18, 19]. Although we previously showed that maintaining a close proximity (≤ 13 cm, ideally ~10 cm) between the CGM and Pod improves inter-device communication in the water [20], it remains unclear how glycemia behaves during routine swimming with variable device placement and intermittent signal loss. The present analysis explores real-world glycemia during swimming from youth with T1D using the Omnipod 5 System.

2 |. Materials and Methods

A single-center observational sub-study was conducted from September 2021 to April 2022 among a subset of individuals at Stanford actively enrolled in the Omnipod 5 pivotal trial extension study (NCT04196140). Individuals eligible for sub-study participation were ≥ 6 years of age (no upper limit) and swimming at least two times per month for ≥ 30 min per swim. Stanford Institutional Review Board-approved consent and assent were obtained from each participant’s legal guardian and participant, respectively.

Participants were all using the Omnipod 5 System and were not provided any guidance on when or how to adjust AID settings or glucose levels around swims. They independently managed their device placement and glucose control without additional study team involvement. The Activity feature, a user-initiated mode that reduces insulin delivery and temporarily raises the glucose target to 150 mg/dL for a user-defined duration (1–24 h) [18], was optional and used by participants only based on personal preference.

Baseline demographic details were obtained including age, self-identified race/ethnicity and gender, body mass index (BMI), duration of T1D, most recent HbA1c and insulin total daily dose (TDD). Participants wore a study-provided physical activity tracker (Garmin Swim 2 activity tracker; Garmin Ltd., Olathe, KS) on swim days and were asked to log each swim both on the activity tracker and via post-swim log survey hosted on REDCap (Vanderbilt University, Nashville, TN). The online post-swim survey included swim time(s), timing and amount of carbohydrates consumed within 2 h of swimming and during swimming, subjective perception of swimming exertion (via a modified version of the BORG scale [21]), and body placement of devices (CGM and Pod) during the swim.

Swim times were collected from the activity tracker and verified with the times recorded on the online swim logs. Individual glucose data files (.csv) were downloaded from Dexcom Clarity, and detailed Pod insulin delivery data was provided by Insulet Corporation (Acton, MA, USA). The analysis included only swims with at least 70% CGM data availability. CGM metrics during swims were calculated as the percent of time spent < 54 mg/dL, 54–69 mg/dL, 70–180 mg/dL, 181–250 mg/dL and > 250 mg/dL. We defined the occurrence of hypoglycemia during swims (from the start to the end of a swim) as any sensor glucose reading < 70 mg/dL. Unpaired t-tests were used to compare group differences, and a multiple regression model was applied to evaluate major contributors to glucose change during swims across four broad categories, including starting glucose, insulin delivery, carbohydrate intake and exercise duration and intensity. Given the observational design, behavioural heterogeneity and clustering of swims within participants, these inferential analyses are considered exploratory and hypothesis-generating, intended to describe associations rather than infer causality.

3 |. Results

Of the 16 pivotal trial participants who met eligibility criteria, seven were enrolled. Their characteristics are summarised in Table 1 with participant-level demographics reported in Table S1. Across all participants, 137 swims were recorded; 124 (91%) had CGM data for ≥ 70% of the swim duration and were included in the analysis; 94 (76%) of these swims were accompanied by a completed swim log. Swim contributions were unevenly distributed across participants, with two participants accounting for 103 of the 124 swims (83%). Because swim counts varied between individuals, descriptive summaries are presented primarily at the swim-event level; participant-level contributions are available in Table S2.

TABLE 1 |.

Baseline participants demographics.

Participant characteristics (N = 7)
Female	4 (57%)
Race/Ethnicity
Non-Hispanic White	6 (86%)
Hispanic	1 (14%)
Age (years)	11 ± 2
Body mass index (BMI, kg/m2)	20.2 ± 3.5
BMI percentile	76.6 ± 18.0
Diabetes duration (years)	5.5 ± 2.1
HbA1c (%)	6.7 ± 0.5
Insulin total daily dose (units)	37.6 ± 16.6

Note: Data reported as N (%) or mean ± SD.

The median swim duration was 62 min (interquartile range [IQR], 48–86 min), with a median active energy expenditure of 301 kcal (IQR, 204–377 kcal) per swim (Table 2). The mean heart rate during swims was 125 ± 11 beats per minute, corresponding to 60% ± 5% of the estimated maximum heart rate (Table 2), consistent with moderate-intensity aerobic activity.

TABLE 2 |.

Physical activity tracker and swim characteristics (N = 137 individual swimming events).

Swim characteristics (N = 137)
Swims with available pump and CGM data	124 (91%)
Pod communication and insulin delivery	124
Total cumulative swim time analysed (hours)	143
Duration (mins); median [IQR]	62 [48, 86]
Distance travelled (metres); mean ± SD	1828 ± 806
Average heart rate; mean ± SD	125 ± 11
Relative heart rate (%); mean ± SD	60 ± 5
Max heart rate; mean ± SD	166 ± 18
Cadence (strokes per minute); mean ± SD	22 ± 3
Active kilocalories burned; median [IQR]	301 [204, 377]
Average pace (minutes/100 m); mean ± SD	1:42 ± 0:08

Note: Data reported as mean ± SD or median [IQR].

During swims, the AID system was maintained in Automated Mode for a median of 98% (IQR: 95%–100%) of swim time. The system was in Automated Mode: Limited for a median of 58% (IQR: 24%–70%) of this automated operation time. Hypoglycemia frequency did not increase with greater exposure to Automated Mode: Limited, occurring in 13.2% (5/38), 33.3% (16/48) and 15.8% (6/38) of swims with < 20, 20–40 and > 40 min in Automated Mode: Limited, respectively. Median swim time spent in Manual Mode was 2% (IQR: 0%–5%).

Figure 1 depicts swim-level data and vertically aligns the starting glucose (rank-ordered from lowest to highest, mg/dL), the corresponding change in glucose during the swim, the percent of time the Omnipod 5 System was suspended during the swim, the percent of time the system was in Automated Mode: Limited, and the incidence of hypoglycemia. Hypoglycemia frequency decreased with increasing starting glucose. Hypoglycemia occurred in 48% (15/31) of swims starting < 90 mg/dL, 24% (6/25) starting 90–125 mg/dL, 13% (5/39) starting 126–180 mg/dL and 3% (1/29) starting ≥ 181 mg/dL. Median (IQR) glucose change during swims was modest (−4.0 mg/dL; −16.5 to 7.5 mg/dL) for lower starting glucose (< 90 mg/dL) and greatest (−86 mg/dL; −111 to −25 mg/dL) among those starting ≥ 181 mg/dL (Figure 2). When starting glucose was < 90 mg/dL, 48% (15/31) of swims resulted in a glucose rise and 52% (16/31) in a glucose fall. The median (IQR) percentage of swim time that the system spent suspended across the same starting glucose strata was 100% (91%–100%) at < 90 mg/dL, 68% (36%–96%) at 90–120 mg/dL, 53% (13%–90%) at 126–180 mg/dL and 8% (0%–40%) at ≥ 181 mg/dL, respectively.

FIGURE 1 |.

Individual rank order of glucose levels at onset of swims in seven adolescents with T1D (panel a), change in glucose levels from start to end of swims (panel b), percent of time the system was in auto-suspend mode during swims (panel c), and percent of time the system was in Automated Mode: Limited during swims (panel d). Red denotes hypoglycemia (< 70 mg/dL) occurred during exercise. Box whisker plots represent the median (IQR) and 5–95 percentiles for the (a) pre-exercise glucose, (b) change in glucose, (c) grams of carbohydrates consumed 2 h pre-exercise, and (d) the % auto-suspend for participants that developed (red) and did not develop (black) hypoglycemia. * indicates that starting glucose, change in glucose, and % time the system was suspended during swims were significantly different in youth that developed hypoglycemia as compared to those that did not. Figure concept from Riddell et al. Pediatric Diabetes, 2019 [22].

FIGURE 2 |.

Forest plot of factors affecting the change in glucose during swims. The dotted line represents the zero-point, and the blue line represents the baseline-adjusted mean. Data reported as median with 95% CI.

Figure 2 summarises how median glucose changes varied across insulin-related, behavioural, and exercise-related factors, relative to the overall median decline of −18 mg/dL. Carbohydrates were consumed within 2 h of most swims (89%) and less frequently during swims (33%).

In light of the notable impact of starting glucose, Figure 3 additionally examines several variables based on starting glucose strata, including: (a) the proportion of swim time spent in standard glycemic ranges [23], (b) the percentage of swims using Activity feature and (c) the starting IOB (units) calculated by the device at swim onset. Hypoglycemia and hyperglycemia burdens are inversely related. Of particular note, the proportions of time spent in level 1 hypoglycemia (54–69 mg/dL) and level 2 hypoglycemia (< 54 mg/dL) ranges were 43.5% (~26 min) and 13% (~8 min) compared to 0.3% and 0.0% when starting glucose levels were < 90 mg/dL and ≥ 181 mg/dL, respectively (Figure 3). The highest percentage of Activity feature use (87%) was when the average starting glucose concentration was the lowest (< 90 mg/dL). Median IOB was higher in the lowest starting glucose group (< 90 mg/dL) compared to the next group (90–125 mg/dL).

FIGURE 3 |.

Panel (a) represents the overall CGM % of time spent < 54 mg/dL, 54–69 mg/dL, 70–180 mg/dL, 181–250 mg/dL and > 250 mg/dL during swims; Panel (b) represents the % of swims using Activity feature, and; Panel (c) represents the device-reported insulin on board (IOB) in the 2-h pre-exercise. Data reported in median (IQR).

4 |. Discussion

This observational sub-study provides novel, real-world insight into glycemic responses to swimming in youth with T1D using the Omnipod 5 System. Three key insights emerged. First, automated operation was preserved through most swim time, often via Automated Mode: Limited during periods of reduced CGM signal availability. Second, starting glucose was the dominant determinant of glycemic trajectory. Third, insulin suspension alone was insufficient to reliably prevent hypoglycemia when starting glucose was < 180 mg/dL, reinforcing that behavioural strategies remain important even when AID remains active during swimming.

Despite the known limitations of Bluetooth communication in the water, the system maintained Automated Mode for the vast majority of swim time, frequently via Automated Mode: Limited during periods of reduced CGM signal availability. This indicates that AID can remain operational during aquatic activity without routine reversion to Manual Mode. Although device placement was not standardised in this study, our prior work demonstrated that close CGM-to-Pod proximity (≤ 13 cm) improves underwater communication, whereas greater separation markedly reduces communication reliability [20]. Even with variable device placement and intermittent communication gaps, adaptive insulin adjustments and suspension were still observed, suggesting that the system retains partial functionality during reduced sensor connectivity, a feature with meaningful clinical implications.

Swims beginning with glucose levels < 90 mg/dL did not consistently result in large glucose declines—nearly half (15/31 swims; 48%) were stable or even increased. The high frequency of Activity feature use in this group (87%) suggests user recognition of hypoglycemia risk and anticipatory mitigation. Median device-reported IOB was also higher among these swims, suggesting that residual insulin exposure may have contributed to these lower pre-swim values and influenced subsequent glycemic trends. Interpretation of actual insulin exposure is limited by reliance on device-reported IOB, which depends on user-selected active insulin time and may thereby underestimate physiologic insulin action. Because we are unable to reconstruct a physiology-based estimate of insulin exposure from insulin delivery history, associations involving IOB should be interpreted with this limitation in mind.

Exploratory analyses further supported starting glucose levels and insulin-related factors (e.g., IOB and Activity feature use) as the variables most consistently associated with glycemic change in this dataset. Higher starting glucose was associated with larger glucose declines, whereas swims with lower starting levels showed smaller changes but greater hypoglycemia burden. Insulin exposure was also influential, as swims with IOB showed consistent declines, whereas those without IOB demonstrated more variable responses. Activity feature use was associated with smaller declines compared to swims without it. Similarly, Turner and colleagues [24] demonstrated that enabling the Activity feature 60 or 30 min before exercise reduced insulin delivery and attenuated glucose drops compared to leaving the Omnipod 5 System in Automated Mode without the Activity feature enabled. In contrast, exercise duration, intensity and carbohydrate intake showed more variable and inconsistent associations. These findings align with previous literature demonstrating that starting glucose levels and circulating insulin (IOB) are key predictors of glycemic responses to exercise [22, 25, 26].

Importantly, the present data demonstrate that insulin suspension alone does not reliably protect against hypoglycemia when starting glucose is below 180 mg/dL. Even in swims characterised by extensive insulin suspension and frequent Activity feature use, hypoglycemia remained common at lower starting glucose levels. This should underscore that AID does not eliminate the need for proactive behavioural strategies (e.g., carbohydrate intake, setting higher glucose targets) when the glucose at swim start is already near the lower end of the target range.

Although the largest glucose declines occurred when starting glucose was elevated (≥ 181 mg/dL), hypoglycemia was uncommon in this range. This suggests that both algorithmic modulation and user adjustments helped limit low-glucose exposure, making insulin delivery during higher starting-glucose states relatively safer. However, as algorithms become more aggressive in correcting post-meal hyperglycemia, carbohydrate intake before exercise may increase insulin delivery and potentially raise the risk of glucose drops unless insulin action is tempered, for example, through a user-initiated exercise announcement. Taken together, these findings suggest that glycemic responses during swimming reflect the combined influence of starting glucose, residual insulin activity, algorithm-driven insulin adjustments and behavioural strategies such as carbohydrate intake and use of the Activity feature.

There are several study limitations that warrant consideration. The small sample size and observational study design limit generalisability. Swim contributions were unevenly distributed across participants, with two participants contributing 83% of the total swim sessions. The findings may therefore disproportionately reflect behaviours, insulin dose adjustment strategies and glycemic responses of these participants. Pre-swim behaviours, including carbohydrate intake and insulin adjustments, were not standardised. Additionally, this analysis focused on pre-swim and during swim periods to characterise real-world glucose outcomes and system performance specifically during swimming, when device communication and AID may be affected by water exposure. Future studies evaluating post-swim glucose patterns would provide additional insight into delayed glycemic effects. Despite these limitations, to our knowledge, this study provides the first real-world description of AID performance during swimming—an activity that poses unique challenges in T1D management and with diabetes technology reliant upon Bluetooth communication. Further studies should explore structured exercise protocols, algorithm-level insulin delivery patterns, CGM performance metrics and post-exercise glycemic trends to refine practical guidance for swimming with AID.

5 |. Conclusion

These observations reinforce previous findings linking starting glucose to hypoglycemia risk during exercise [22, 25] and extend this relationship to swimming in youth with T1D. Lower starting glucose levels were associated with higher incidences of hypoglycemia but smaller overall glucose declines during swims, suggesting that both the AID system and the user are responding to the increased risk of hypoglycemia at lower starting glucose levels.

Although most swims involved CGM-to-Pod distances greater than the 13 cm line-of-sight threshold previously shown to optimise underwater communication [19], resulting in reduced and intermittent connectivity, the system still demonstrated robust insulin suspension in response to hypoglycemia risk. However, this insulin suspension alone was not sufficient to reliably prevent hypoglycemia, especially when swims began at lower starting glucose levels, reinforcing that AID alone does not eliminate the need for proactive behavioural strategies for swimming.

Maintaining AID operation in the water may offer advantages over disconnecting or manually suspending insulin delivery. Future studies incorporating more detailed measures of carbohydrate intake, insulin exposure and exercise intensity may further clarify glycemic patterns during swims and help refine practical guidance for swimming with AID systems and T1D.

Supplementary Material

Supporting Information

Additional supporting information can be found online in the Supporting Information section. Table S1: dom70715-sup-0001-FigureS1.png. Table S2: dom70715-sup-0002-FigureS2.png.

Data Availability Statement:

The data that support the findings of this study are available from the corresponding author upon reasonable request.