Variation in Hypoglycemia Risk During Real-World Physical Activity in Adults with Type 1 Diabetes: Insights from the Type 1 Diabetes Exercise Initiative

Abstract

Background:

Physical activity (PA) poses significant challenges in glucose management for individuals with type 1 diabetes (T1D). Real-world PA is more frequent than structured PA, but remains underexplored. We analyzed 8171 real-world PA sessions comprising 45 activity types from the Type 1 Diabetes Exercise Initiative, examining hypoglycemia risk correlations with PA-level and population-level factors.

Methods:

Hypoglycemia risk was measured by change in continuous glucose monitoring (ΔCGM) from PA onset to end, low blood glucose index (LBGI), and hypoglycemia event occurrence. Primary analyses used analysis of variance and Tukey’s range test to measure correlations. Secondary analyses compared risk across activity types and categories (aerobic, mixed, and anaerobic).

Results:

Higher hypoglycemia risk was associated with longer PA duration (median [Interquartile Range (IQR)] ΔCGM −24 [−60, 11] mg/dL for 60–120 min vs. −12 [−31, 5] mg/dL for 15–30 min), lower starting glucose (90% of sessions starting <50 mg/dL had hypoglycemia), and declining glucose rates before PA (all P < 0.05). Carbohydrate (CHO) intake 2–4 h before and during PA was associated with higher hypoglycemia risk (ΔCGM −37 [−67, −14] mg/dL with rescue CHO vs. −15 [−42, 8] mg/dL without, P < 0.05), but this paradoxical effect was explained by higher insulin on board (IoB) and lower starting glucose. Males had larger glucose drops (ΔCGM −20 [−46, 4] mg/dL vs. −16 [−44, 7] mg/dL in females, P < 0.05). Closed-loop users exhibited lower LBGI compared with open-loop users (P < 0.05). Secondary analyses showed significant glycemic variability across activity types (P < 0.05). Aerobic activities caused the greatest glucose drop, followed by mixed and anaerobic (P < 0.05), whereas LBGI differences were nonsignificant (P = 0.32).

Conclusions:

Real-world PA has a highly variable glycemic impact, with longer duration, lower starting glucose, and higher IoB increasing hypoglycemia risk. Glycemic responses differed significantly by activity type, with aerobic activities resulting in the greatest decline. These findings highlight the need for tailored strategies to mitigate PA-related hypoglycemia in T1D.

Keywords

continuous glucose monitoring exercise hypoglycemia physical activity type 1 diabetes

Introduction

Regular exercise and physical activity (PA) are strongly encouraged for individuals with type 1 diabetes (T1D) to improve glycemic management and enhance the overall quality of life.^1,2 However, with an increased risk of hypoglycemia often associated with PA,^3–9 many individuals with T1D do not meet the recommended 150 min of moderate-to-vigorous PA per week.¹⁰ Several factors, including activity duration, starting glucose levels, carbohydrate consumption, and insulin on board (IoB), have been found to impact glycemia during structured PA.^11–13 However, the influence of such factors on the hypoglycemia risk with PA in real-world settings has remained largely unexplored.

Previous research has investigated the glycemic effects of structured PA, demonstrating that aerobic PA leads to higher hypoglycemia risk than resistance and interval PA,^13,14 and that glycemic control during structured PA is impacted by various activity-level and population-level factors, such as IoB, starting glucose, glucose rate of change, daily step count, hypoglycemia awareness, and menstrual cycle.^15–18 Similar studies have been conducted on adolescents with T1D in the Type 1 Diabetes Exercise Initiative Pediatric study,^19–22 demonstrating that the drop in glucose during PA in youth with T1D is impacted by factors such as IoB and starting glucose.

However, structured PA typically follows a standard format (e.g., structured exercise for 30 min at the gym) that does not accurately reflect real-world scenarios, where individuals engage in a wide range of PA that often varies in duration and intensity, and is more spontaneous or unstructured. Due to the significant differences in duration and intensity, it remains uncertain whether insights from structured PA apply to real-world PA, and understanding of how real-world PA influences glucose variability is limited.

To address this gap, this study presents the first large-scale analysis of hypoglycemia risk associated with real-world PA in individuals with T1D. Drawing on data from 8171 activity sessions spanning 45 activity types and 404 participants in the Type 1 Diabetes Exercise Initiative (T1DEXI) dataset,¹³ the analysis examines multiple hypoglycemia risk markers to identify both activity-level and population-level factors that influence exercise-related hypoglycemia.

The TIDEXI was a real-world study aimed at understanding glycemic responses to at-home exercises in adults with T1D and included both structured exercise sessions and real-world PA.²³ Participants used continuous subcutaneous insulin infusion (CSII) via open- or closed-loop systems or multiple daily injections (MDI) for insulin administration. Participants were randomly assigned to complete six structured aerobic, interval, or resistance training exercise sessions over a 4-week period while also engaging in real-world PA. All of the PA, food intake, and insulin dosing (for MDI users) were self-reported. A custom smartphone application further provided pump (for pump users), heart rate, and continuous glucose monitoring (CGM) data. The study is described in more detail elsewhere.¹³

As such, this analysis examines real-world PA sessions from the T1DEXI dataset to assess how the hypoglycemia risk associated with PA is influenced by (1) activity-level factors (i.e., PA duration, starting glucose level, glucose rate of change, carbohydrate intake up to 4 h before and during PA, IoB, and time of day) and (2) population-level factors (i.e., age, sex, baseline HbA1c, body mass index [BMI], and T1D duration). In addition, the analysis assesses the variable impact of 45 different real-world PA types, highlighting the challenges of developing treatments that are effective outside of controlled settings.

Methods

Study design

A total of 512 adults with T1D enrolled in the T1DEXI study. Participants wore a CGM system and were assigned to perform at least six video-guided exercise sessions (∼30 min each) at home over a 4-week period. These structured sessions included aerobic, interval, or resistance activities. In addition, participants maintained their usual real-world PA routines, which were self-reported via the T1DEXI app. For each PA session, a participant needed to report the activity start time and duration and to select the activity type from a provided list. Other self-reported information included food intake and demographics. For participants on pumps, insulin information was extracted from the Tidepool Data Platform.

In this study, we only analyzed the real-world PA sessions and excluded all structured PA sessions. Due to the variability of the real-world PA sessions, several filtering criteria were used to select the sessions for this analysis. First, the analysis focused on real-world PA sessions performed by participants using either open-loop (i.e., standard pump therapy) or closed-loop CSII from the T1DEXI dataset. PA sessions for participants using MDI were excluded from this analysis. Second, real-world PA sessions were analyzed if they included at least one CGM reading within 5 min before and after the exercise and one reading during the exercise window. Third, the analysis included PA sessions lasting between 15 min and 2 h in duration, excluding PA extremes (both short and prolonged).

Measuring the glycemic response to exercise

The primary analysis assessed the hypoglycemia risk during real-world PA based on the change in glycemia and the low blood glucose index (LBGI)—a measure of the frequency and extent of low glucose (<70 mg/dL) readings²⁴ that has been identified as a clinically useful indicator of hypoglycemia risk.²⁵ Using LBGI as the metric offers the advantage that lower glucose values, particularly those below 54 mg/dL, are assigned a higher penalty. The change in glucose levels during each PA session was calculated using the difference in CGM readings from the onset of PA up to 5 min after PA, defined as “ΔCGM.” In addition, a session was flagged for the presence of hypoglycemia if there were at least 15 consecutive minutes of glucose readings below 70 mg/dL within the PA.

The secondary analysis compared hypoglycemia risk during PA based on self-reported PA type. While previous studies have demonstrated that hypoglycemia risk varies across different PA categories (e.g., aerobic, interval, and resistance exercise),^13,16,26,27 these findings were limited to structured PA sessions without a detailed analysis of specific real-world PA types.

Measuring activity and population-level factors

The activity-level factors (starting glucose level, glucose rate of change, IoB, carbohydrate intake, and PA time of day) were extracted from both recorded and self-reported data. The starting glucose level for each PA session was determined by the CGM reading at the onset of PA. The glucose rate of change before PA was estimated by the linear regression slope applied to glucose readings taken 15 min before the activity. The IoB at the start of each PA was calculated using the basal and bolus insulin administered within the 4 h leading up to the PA session. The insulin decay curve for the IoB calculation was adopted from a previous publication.²⁸ The carbohydrate intake timing before PA was defined by whether the last meal was consumed within 1, 2, 3, 4, or more than 4 h before PA. The carbohydrate intake during PA was considered rescue carbohydrates. The time of day of a PA session was defined by the starting time of the session and categorized into morning (3 am to <noon), afternoon (noon to <6 pm), evening (6 pm to <9 pm), and night (9 pm to <3 am). The population-level factors (age, sex, baseline HbA1c, BMI, T1D duration, race, ethnicity, insulin delivery mode, hours asleep per day) were directly obtained from the self-reported data.

Statistical analysis

An analysis of variance test was used to determine whether the differences in hypoglycemia risk and the glucose response against each factor were statistically significant. For all of the tests, a P value <0.05 was considered statistically significant. In addition, for each factor, pairwise post hoc testing was used to determine whether there was a statistical difference between all pairs. Tukey’s range test was used for post hoc testing.²⁹

Categorizing real-world activities

The real-world PA sessions were classified into categories of “aerobic,” “mixed,” or “anaerobic” based on the 2024 Adult Compendium of Physical Activities.³⁰

Results

After applying the filtering criteria described in the Study Design section, a total of 8171 sessions, covering 45 types of real-world PA, were analyzed from 404 adult T1D participants on CSII. With the filtering criteria, this analysis still covered a majority (78.9%) of the total 512 participants in the original T1DEXI study.

Participant characteristics are summarized in Table 1. The participants had a median age of 33 years (range: 18–70 years), a mean ± standard deviation HbA1c of 6.6 ± 0.7%, a BMI of 25.5 ± 3.9 kg/m², and an average T1D duration of 18.4 ± 13.0 years. The total daily dose of insulin was 40.1 ± 16.2 units. A total of 75% of participants self-reported as female, 91.3% identified as White, and 94.6% as not Hispanic or Latino. In addition, 55% used closed-loop automated insulin delivery (AID), whereas the remaining 45% used open-loop pump systems.

Table 1.

Study Participant Demographics (Total Participants: 404)

	mean ± stdmedian [IQR]
Age (Years)	36.55 (14.05) 33.0 [25.0, 46.0]
Sex	75% Female 25% Male
Race	91.3% White 2.2% Asian 2.2% Black/African American 1.7% Not reported 1.5% Multiple 0.5% Unknown 0.5% American Indian/Alaskan native
Ethnicity	94.6% Not Hispanic or Latino 2.5% Hispanic or Latino 2.5% Do not wish to answer 0.5% Don't know
HbA1c (%)	6.63 (0.72) 6.6 [6.18, 7.1]
BMI (kg/m²)	25.51 (3.94) 24.81 [22.99, 27.43]
T1D duration (years)	18.48 (13.02) 15.0 [9.0, 25.0]
TDD (Units)	40.07 (16.16) 37.68 [28.07, 47.57]
Insulin delivery mode	55% AID (Closed loop) 45% Pump (Open loop)

AID, automated insulin delivery; BMI, body mass index; T1D, type 1 diabetes; TDD, total daily dose.

The analysis included participants with a wide range of routine exercise intensities. We estimate the participants’ usual level of activity based on Metabolic Equivalent (MET) minutes calculated from the International Physical Activity Questionnaire (IPAQ), following the work introducing the T1DEXI dataset.¹³ IPAQ asks for the duration in 1 day of vigorous activities, moderate activities, walking, and sitting. MET minutes are calculated as a weighted sum of the 4 durations.³¹ Out of the 404 participants in the analysis, 361 participants completed the IPAQ. The MET minutes are 3011 ± 3055 min. The 25th, 50th (median), and 75th percentiles are 1356, 2184, and 3685 min. The statistics indicate that participants of various routine exercise intensities are included in the analysis.

Figure 1 presents the frequency and categorization of real-world PA sessions. Of the 45 activity types, eight were classified as predominantly “aerobic,” 26 were classified as “mixed,” and the remaining 11 were categorized as “anaerobic.”³⁰ In total, 5340 sessions (65.4%) involved aerobic PA and 2831 sessions (34.6%) included mixed/anaerobic PA. The four most common PA types included “walking/dog walking” (n = 3421), “biking” (n = 757), “jogging/running” (n = 675), and “strength training/weight lifting” (n = 637). Out of these, three were classified as aerobic (walking/dog walking, biking, jogging/running contributing to 59.3% of all PA logged), whereas strength training/weight lifting was classified as mixed/anaerobic (contributing to 7.8% of all PA logged).

FIG. 1.

Frequency of different real-world physical activities in the analysis.

For all sessions (N = 8171), the average duration of the PA was 40.8 ± 22.8 min. Participants engaged in an average of 20.2 ± 12.6 exercise sessions over a mean study duration of 24.9 ± 5.0 days. On average, participants did PA on 15.2 ± 6.3 days, resulting in a percentage of PA days of 60.3% ± 19.7%. For each PA day, participants completed an average of 1.3 ± 0.3 sessions, with an average duration of 53.6 ± 18.6 min.

Glycemic effect of real-world PA

Table 2 presents the change in glucose (ΔCGM) during PA, the PA-related LBGI, and the percentage of PA sessions associated with hypoglycemia for each factor, with corresponding post hoc analysis results visualized in Supplementary Figures S1–S16. Across all sessions, the mean ± standard deviation ΔCGM was −20 ± 45 mg/dL, LBGI was 1.2 ± 3.1, and a hypoglycemia event occurred during 6.2% of PA sessions.

Table 2.

Change in Continuous Glucose Monitoring, Low Blood Glucose Index, And Percentage of Sessions with Hypoglycemic Events

Characteristic	Bins	Number of sessions	Number of participants	ΔCGM	P value	LBGI	P value	Sessions with hypoglycemia (%)
(a) Activity-level factors
Duration (min)	15 to <30	2527	340	−13.76 (30.0)	<0.05	1.05 (2.94)	<0.05	2.81
	15 to <30	2527	340	−12.0 [−31.0, 5.0]		0.0 [0.0, 0.42]		2.81
	30 to <60	3900	396	−22.39 (45.53)		1.17 (3.11)		5.82
	30 to <60	3900	396	−20.0 [−49.0, 6.0]		0.0 [0.0, 0.64]		5.82
	60 to ≤120	1744	334	−25.49 (59.59)		1.48 (3.13)		12.16
	60 to ≤120	1744	334	−24.0 [−60.0, 11.0]		0.08 [0.0, 1.67]		12.16
Starting glucose (mg/dL)	<50	20	18	32.8 (38.91)	<0.05	20.1 (12.13)	<0.05	90
	<50	20	18	24.5 [0.0, 53.25]		18.48 [10.13, 32.04]		90
	50 to <100	1317	326	8.63 (32.06)		4.36 (4.98)		21.34
	50 to <100	1317	326	4.0 [−12.0, 24.0]		2.66 [0.85, 5.9]		21.34
	100 to <150	3506	390	−11.24 (33.57)		0.81 (1.97)		4.11
	100 to <150	3506	390	−12.0 [−34.0, 9.0]		0.02 [0.0, 0.59]		4.11
	150 to <200	2160	379	−33.26 (42.41)		0.31 (1.24)		2.36
	150 to <200	2160	379	−31.5 [−63.0, −6.0]		0.0 [0.0, 0.0]		2.36
	200 to <250	817	287	−51.55 (51.78)		0.17 (0.86)		1.71
	200 to <250	817	287	−49.0 [−83.0, −18.0]		0.0 [0.0, 0.0]		1.71
	250 to <300	264	158	−67.1 (62.08)		0.07 (0.47)		0.38
	250 to <300	264	158	−61.5 [−106.25, −22.75]		0.0 [0.0, 0.0]		0.38
	300 to <350	66	53	−87.26 (86.96)		0.1 (0.79)		1.52
	300 to <350	66	53	−71.0 [−152.75, −16.0]		0.0 [0.0, 0.0]		1.52
	≥350	21	18	−83.9 (110.09)		0.05 (0.21)		0
	≥350	21	18	−31.0 [−111.0, 0.0]		0.0 [0.0, 0.0]		0
Glucose change rate	<−1.0	1212	345	−38.78 (47.6)	<0.05	2.59 (4.67)	<0.05	14.44
	<−1.0	1212	345	−37.0 [−62.0, −9.0]		0.34 [0.0, 3.23]		14.44
	−1.0 to <−0.5	1130	354	−27.02 (41.77)		1.86 (3.88)		9.91
	−1.0 to <−0.5	1130	354	−24.0 [−47.0, −4.0]		0.1 [0.0, 2.0]		9.91
	−0.5 to <0.5	3278	397	−17.04 (40.15)		0.99 (2.63)		4.79
	−0.5 to <0.5	3278	397	−13.0 [−38.0, 6.0]		0.0 [0.0, 0.63]		4.79
	0.5 to <1.0	1048	325	−12.1 (43.32)		0.63 (1.76)		2.86
	0.5 to <1.0	1048	325	−10.0 [−34.0, 14.0]		0.0 [0.0, 0.23]		2.86
	≥1.0	1490	356	−13.37 (52.28)		0.44 (1.53)		2.42
	≥1.0	1490	356	−10.0 [−42.0, 19.75]		0.0 [0.0, 0.03]		2.42
IoB	>0 and <1 U	659	181	5.63 (34.49)	<0.05	0.91 (2.55)	<0.05	3.03
	>0 and <1 U	659	181	5.0 [−13.0, 23.5]		0.0 [0.0, 0.6]		3.03
	1–<2U	2230	337	−7.51 (38.75)		1.09 (2.9)		5.34
	1–<2U	2230	337	−7.0 [−28.0, 14.0]		0.0 [0.0, 0.72]		5.34
	2–<3U	1727	359	−22.65 (40.98)		1.19 (2.74)		6.14
	2–<3U	1727	359	−20.0 [−45.0, 2.0]		0.0 [0.0, 0.98]		6.14
	≥3U	3517	384	−32.3 (48.41)		1.33 (3.39)		7.48
	≥3U	3517	384	−30.0 [−58.0, −3.0]		0.0 [0.0, 0.79]		7.48
Hours since last meal	<1 h	3472	387	−16.92 (46.77)	<0.05	1.52 (3.57)	<0.05	8.09
	<1 h	3472	387	−15.0 [−44.0, 11.0]		0.01 [0.0, 1.22]		8.09
	2 h	1853	359	−30.47 (46.09)		1.19 (2.99)		6.64
	2 h	1853	359	−26.0 [−54.0, −3.0]		0.0 [0.0, 0.85]		6.64
	3 h	919	302	−25.74 (44.07)		0.94 (2.58)		4.35
	3 h	919	302	−22.0 [−51.0, −1.0]		0.0 [0.0, 0.6]		4.35
	4 h	369	191	−27.83 (40.51)		0.83 (2.04)		4.61
	4 h	369	191	−24.0 [−47.0, −2.0]		0.0 [0.0, 0.47]		4.61
	>4 h	1558	302	−11.18 (39.22)		0.76 (2.24)		3.15
	>4 h	1558	302	−7.0 [−28.0, 10.0]		0.0 [0.0, 0.3]		3.15
Rescue carbs	Yes	604	205	−40.11 (46.91)	<0.05	3.27 (4.86)	<0.05	19.54
	Yes	604	205	−37.0 [−67.25, −14.0]		1.29 [0.06, 4.28]		19.54
	No	7567	404	−18.81 (44.73)		1.04 (2.81)		5.18
	No	7567	404	−15.0 [−42.0, 8.0]		0.0 [0.0, 0.56]		5.18
Activity time of day	Morning	2965	352	−14.08 (44.07)	<0.05	0.84 (2.4)	<0.001	3.91
	(3 am to <noon)	2965	352	−11.0 [−37.0, 12.0]		0.0 [0.0, 0.4]		3.91
	Afternoon	3432	382	−24.87 (44.51)		1.27 (3.1)		7.2
	(noon to <6 pm)	3432	382	−21.0 [−49.0, 2.0]		0.0 [0.0, 0.96]		7.2
	Evening	1443	311	−23.52 (46.82)		1.72 (3.8)		8.59
	(6 pm to <9 pm)	1443	311	−21.0 [−50.0, 4.0]		0.01 [0.0, 1.6]		8.59
	Night	331	133	−16.69 (49.08)		1.4 (3.93)		6.95
	(9 pm to <3 am)	331	133	−14.0 [−41.5, 8.0]		0.0 [0.0, 0.64]		6.95
(b) Population-level factors
Age	18 to <25	1375	89	−18.93 (49.65)	0.11	1.15 (2.8)	<0.05	6.84
	18 to <25	1375	89	−18.0 [−46.0, 9.0]		0.0 [0.0, 0.66]		6.84
	26 to <45	4172	206	−19.99 (44.17)		1.17 (3.13)		5.9
	26 to <45	4172	206	−17.0 [−44.0, 7.0]		0.0 [0.0, 0.74]		5.9
	45 to <60	1500	68	−22.8 (46.11)		1.4 (3.49)		7.6
	45 to <60	1500	68	−17.0 [−47.0, 6.0]		0.0 [0.0, 1.0]		7.6
	≥60	1124	41	−20.39 (42.14)		1.11 (2.47)		4.98
	≥60	1124	41	−18.0 [−43.0, 5.0]		0.01 [0.0, 0.96]		4.98
Sex	Female	6071	304	−19.76 (45.9)	<0.05	1.13 (3.01)	<0.05	5.75
	Female	6071	304	−16.0 [−44.0, 7.0]		0.0 [0.0, 0.64]		5.75
	Male	2100	100	−22.19 (43.22)		1.41 (3.22)		7.67
	Male	2100	100	−20.0 [−46.0, 4.0]		0.03 [0.0, 1.15]		7.67
HbA1c	<6.0	1593	69	−19.95 (42.27)	0.48	1.5 (3.31)	<0.05	7.34
	<6.0	1593	69	−17.0 [−44.0, 5.0]		0.04 [0.0, 1.44]		7.34
	6.0 to <6.5	2080	106	−21.4 (42.65)		1.18 (2.9)		5.91
	6.0 to <6.5	2080	106	−17.0 [−43.25, 4.0]		0.0 [0.0, 0.87]		5.91
	6.5 to <7.0	2132	103	−18.83 (43.18)		1.09 (3.01)		5.86
	6.5 to <7.0	2132	103	−16.0 [−42.0, 7.0]		0.0 [0.0, 0.55]		5.86
	7.0 to <7.5	1396	74	−20.95 (48.42)		1.07 (3.07)		5.52
	7.0 to <7.5	1396	74	−17.0 [−47.0, 9.0]		0.0 [0.0, 0.51]		5.52
	7.5 to <8.0	723	33	−21.5 (49.82)		1.25 (3.22)		6.78
	7.5 to <8.0	723	33	−19.0 [−49.5, 10.0]		0.0 [0.0, 0.64]		6.78
	≥8.0	247	19	−21.47 (65.14)		1.06 (2.68)		7.69
	≥8.0	247	19	−16.0 [−51.0, 10.5]		0.0 [0.0, 0.18]		7.69
BMI	<25	4638	213	−21.24 (45.61)	<0.05	1.15 (2.95)	0.065	5.86
	<25	4638	213	−18.0 [−45.0, 6.0]		0.0 [0.0, 0.73]		5.86
	25 to <30	2615	137	−20.47 (45.59)		1.22 (3.16)		6.46
	25 to <30	2615	137	−17.0 [−46.0, 6.5]		0.0 [0.0, 0.79]		6.46
	≥30	918	54	−15.82 (41.99)		1.41 (3.35)		7.52
	≥30	918	54	−13.0 [−38.0, 8.0]		0.01 [0.0, 1.14]		7.52
T1D duration (years)	<5	758	40	−19.76 (40.82)	0.064	1.2 (2.66)	<0.05	5.67
	<5	758	40	−16.0 [−42.0, 5.0]		0.01 [0.0, 1.08]		5.67
	5 to <10	1264	66	−17.77 (41.55)		0.86 (2.41)		3.8
	5 to <10	1264	66	−15.0 [−40.0, 5.25]		0.0 [0.0, 0.47]		3.8
	≥10	6149	298	−21.0 (46.45)		1.27 (3.23)		6.81
	≥10	6149	298	−18.0 [−46.0, 7.0]		0.0 [0.0, 0.85]		6.81
Race	White	7648	369	−20.68 (45.3)	<0.05	1.19 (3.05)	<0.05	6.18
	White	7648	369	−17.0 [−44.0, 6.0]		0.0 [0.0, 0.79]		6.18
	Asian	110	9	−7.88 (42.83)		1.07 (2.2)		4.55
	Asian	110	9	−7.5 [−32.75, 16.75]		0.0 [0.0, 1.27]		4.55
	Black/African American	92	9	−7.18 (34.45)		0.51 (1.63)		2.17
	Black/African American	92	9	−5.0 [−27.25, 16.25]		0.0 [0.0, 0.0]		2.17
	American Indian/Alaskan native	22	2	−12.86 (36.42)		2.13 (5.34)		13.64
	American Indian/Alaskan native	22	2	−13.0 [−30.25, 9.0]		0.0 [0.0, 0.72]		13.64
	Multiple	91	6	−13.24 (55.84)		0.86 (3.37)		6.59
	Multiple	91	6	−8.0 [−39.5, 13.0]		0.0 [0.0, 0.06]		6.59
	Unknown	25	2	−23.48 (48.33)		1.27 (2.19)		4
	Unknown	25	2	−9.0 [−55.0, 1.0]		0.21 [0.0, 1.46]		4
	Not reported	183	7	−26.22 (41.08)		2.2 (4.02)		10.93
	Not reported	183	7	−24.0 [−50.0, −2.5]		0.18 [0.0, 2.62]		10.93
Ethnicity	Not Hispanic or Latino	7802	382	−20.39 (44.96)	0.045	1.21 (3.08)	0.057	6.31
	Not Hispanic or Latino	7802	382	−17.0 [−44.0, 6.0]		0.0 [0.0, 0.81]		6.31
	Hispanic or Latino	186	10	−20.02 (45.43)		0.63 (1.78)		1.61
	Hispanic or Latino	186	10	−13.0 [−45.75, 8.0]		0.0 [0.0, 0.24]		1.61
	Do not wish to answer	156	10	−16.71 (55.02)		1.42 (3.43)		8.97
	Do not wish to answer	156	10	−19.0 [−53.0, 16.5]		0.0 [0.0, 0.85]		8.97
	Do not know	27	2	−43.44 (54.85)		1.5 (4.53)		3.7
	Do not know	27	2	−45.0 [−80.0, 7.5]		0.0 [0.0, 0.6]		3.7
Insulin delivery mode	AID (closed loop)	4413	222	−20.33 (45.47)	0.906	1.12 (2.95)	<0.05	5.71
	AID (closed loop)	4413	222	−17.0 [−45.0, 8.0]		0.0 [0.0, 0.61]		5.71
	Pump (open loop)	3758	182	−20.45 (44.96)		1.3 (3.2)		6.87
	Pump (open loop)	3758	182	−17.0 [−43.0, 5.0]		0.0 [0.0, 1.02]		6.87
Hours of sleep	4	88	2	−6.59 (35.83)	<0.05	0.83 (2.05)	<0.05	2.27
	4	88	2	−3.0 [−23.25, 14.25]		0.01 [0.0, 0.8]		2.27
	5	203	16	−10.07 (46.23)		1.53 (3.95)		8.87
	5	203	16	−8.0 [−31.0, 17.0]		0.0 [0.0, 1.12]		8.87
	6	1364	65	−23.25 (46.04)		1.42 (3.34)		7.48
	6	1364	65	−20.0 [−48.25, 4.0]		0.0 [0.0, 1.25]		7.48
	7	3466	164	−19.73 (42.2)		1.22 (2.99)		5.83
	7	3466	164	−16.0 [−42.0, 5.0]		0.0 [0.0, 0.92]		5.83
	8	2635	131	−20.95 (48.76)		1.1 (3.04)		6.34
	8	2635	131	−17.0 [−47.0, 8.0]		0.0 [0.0, 0.54]		6.34
	9	332	21	−18.88 (42.7)		0.99 (2.81)		5.42
	9	332	21	−16.0 [−39.0, 5.25]		0.0 [0.0, 0.54]		5.42
	10	83	5	−28.51 (48.18)		0.46 (1.24)		1.2
	10	83	5	−22.0 [−52.0, 4.5]		0.0 [0.0, 0.18]		1.2

The measures are reported as mean (std) followed by median [IQR]. The table contains two subtables, where the first subtable (a) includes activity-level factors and the next subtable (b) contains population-level factors.

ΔCGM, change in continuous glucose monitoring; IoB, insulin on board; LBGI, low blood glucose index.

Activity-level factors

Longer duration PA sessions (60 to ≤120 min) had a greater drop in glucose (median [IQR] ΔCGM −24 [−60, 11] mg/dL) and higher LBGI (0.08 [0.0, 1.67]) compared with shorter sessions (ΔCGM −12 [−31, 5] mg/dL and LBGI 0.0 [0.0, 0.42] for 15 to <30-min sessions, P < 0.05). Hypoglycemia occurred in 12.2% of the longer PA sessions (60 to ≤120 min) compared with 2.8% and 5.8% for PA lasting 15 to <30 min and 30 to <60 min.

The drop in glucose levels was greater with higher starting glucose levels (P < 0.05). Sessions with a starting glucose between 300 and <350 mg/dL (n = 66) had the largest drop (ΔCGM −71 [−153, −16] mg/dL), whereas those starting <50 mg/dL (n = 20) were associated with an increase in glucose (ΔCGM 25 [0, 53] mg/dL). Sessions with a lower starting glucose level (<50 mg/dL and 50 to <100 mg/dL) had a significantly higher LBGI (18.48 [10.13, 32.04] and 2.66 [0.85, 5.9], respectively) compared with PA with higher starting glucose values (P < 0.05). The percentage of sessions with hypoglycemia events increased consistently from 0% to 90% as starting glucose levels decreased from ≥350 to <50 mg/dL.

A decreasing glucose before PA (rate of change <−0.5 mg/dL per minute) correlated with a greater drop in glucose and higher LBGI (P < 0.05). The percentage of hypoglycemia sessions increased from 2.42% to 14.44% as the glucose change rate decreased from ≥1 mg/dL per minute to <−1 mg/dL.

PA sessions with the largest IoB (≥3 U) had the greatest drop in glucose (ΔCGM −30 [−58, −3] mg/dL), the highest LBGI (0.0 [0.0, 0.79]), and the highest percentage of PA sessions with hypoglycemia (7.5%).

When carbohydrates were consumed within 1 h of PA, there was a smaller drop in glucose (ΔCGM −15 [−44, 11] mg/dL), compared with the drop when the last meal was consumed 2–4 h before PA (ΔCGM −26 [−54, −3] mg/dL, −22 [−51, −1] mg/dL, and −24 [−47, −2] mg/dL for last meal consumption 2, 3, and 4 h before PA), P < 0.05 (Table 2). The smallest drop in glucose occurred when the last meal was consumed >4 h before PA (ΔCGM −7 [−28, 10] mg/dL, P < 0.05). Conversely, sessions with meal intake within 1 h before PA had the highest LBGI of 0.01 [0.0, 1.22] (P < 0.05) and the greatest hypoglycemia risk rate (8.09%).

PA sessions with rescue carbohydrates (i.e., prevention/treatment of hypoglycemia) consumed had a greater drop in glucose (ΔCGM −37 [−67, −14] mg/dL) and higher LBGI (1.29 [0.06, 4.28], respectively) compared with PA without (ΔCGM −15 [−42, 8] mg/dL and LBGI 0.0 [0.0, 0.56]), P < 0.05. In addition, 19.5% of sessions with rescue carbohydrate intake had a hypoglycemia event compared with 5.2% of those without.

The lowest pre-PA glucose (138 ± 45 mg/dL) was found when the last meal was consumed >4 h before PA, followed by PA with a meal taken <1 h before PA (145 ± 51 mg/dL). Furthermore, PA with rescue carbohydrate intake had a lower starting glucose level (133 ± 44 mg/dL) compared with PA without rescue carbohydrate intake (148 ± 52 mg/dL).

The smallest drop in glucose (ΔCGM −11 [−37, 12] mg/dL) and lowest LBGI (0.0 [0.0, 0.4]) were found when the PA was performed in the morning (between 3 am and 12 pm). The percentage of sessions with hypoglycemia was also the lowest (3.91%) for PA performed within this period. The PA performed in the afternoon (noon to <6 pm), evening (6 pm to <9 pm), and night time (9 pm to <3 am) had comparatively larger drop in glucose (ΔCGM −21 [−49, 2] mg/dL, −21 [−50, 4] mg/dL, and −14 [−42, 8] mg/dL, respectively), higher LBGI (0.0 [0.0, 0.96], 0.01 [0.0, 1.6], and 0.0 [0.0, 0.64], respectively), and a larger percentage with hypoglycemia events (7.20%, 8.59%, and 6.95%, all P < 0.05).

Population-level factors

Analyses were conducted on population-level factors, including age group, sex, baseline HbA1c levels, BMI, duration of T1D, race, ethnicity, insulin delivery mode, and hours asleep per day. No significant differences in ΔCGM or LBGI were found across ages, baseline HbA1c, races, ethnicities, and hours asleep per day. However, PA performed by males had a greater drop in glucose (ΔCGM −20 [−46, 4] mg/dL), higher LBGI (0.03 [0.0, 1.15]), and a higher percentage of PA sessions with hypoglycemia (7.7%) compared with those by females (ΔCGM −16 [−44, 7] mg/dL, LBGI 0.0 [0.0, 0.64], and 5.8% sessions with hypoglycemia events), P < 0.05.

PA performed by participants using AID closed-loop systems was associated with a significantly lower LBGI (0.0 [0.0, 0.61]) compared with those using open-loop pumps (0.0 [0.0, 1.02]), P < 0.05. Closed-loop users also experienced a lower percentage of sessions with hypoglycemia (5.71%) than open-loop users (6.87%). However, ΔCGM did not differ significantly between the two groups.

While LBGI did not vary significantly with BMI (P = 0.07), the drop in glucose was smaller (P < 0.05) for PA completed by participants with a BMI ≥30 kg/m² (−13 [−38, 8] mg/dL) compared with sessions done by participants with a BMI <25 kg/m² (−18 [−45, 6] mg/dL). Participants with ≥10 years and <5 years of T1D duration had higher LBGI (0.0 [0.0, 0.85] and 0.01 [0.0, 1.08], respectively) and percentage of sessions with hypoglycemia events (6.8% and 5.7%, respectively) compared with participants with T1D duration of 5–10 years (LBGI = 0.0 [0.0, 0.47]; 3.8% sessions with hypoglycemia events; all P < 0.05).

Table 3 compares the characteristics of PA with at least one hypoglycemia event (n = 510) to those with no hypoglycemia events (n = 7661) during PA, at both the activity and population level. Sessions with hypoglycemia events were of longer duration (median [IQR]: 50 [35.0, 61.75] min vs. 32 [25.0, 50.0] min), had a lower starting glucose (93 [72, 122] mg/dL vs. 140 [113, 177] mg/dL), higher IoB (3.09 [1.91, 4.81] vs. 2.59 [1.61, 4.22]), and a lower glucose rate change (−0.7 [−1.4, −0.1] vs. 0.0 [−0.6, 0.7]).

Table 3.

Sessions With Versus Without TBR and Hypoglycemia Events

	Sessions with hypoglycemic event	Sessions without hypoglycemic event
N	510	7661
Duration (min)	54.38 (27.32)	39.86 (22.15)
	50.0 [35.0, 61.75]	32.0 [25.0, 50.0]
Starting glucose (mg/dL)	102 (41)	149 (51)
	93 [72, 122]	140 [113, 177]
Age	38.79 (14.09)	38.53 (14.62)
	36.0 [26.0, 50.0]	35.0 [26.0, 49.0]
Sex	68.43% F, 31.57% M	74.69% F, 25.31% M
HbA1c	6.55 (0.78)	6.57 (0.71)
	6.5 [6.1, 7.0]	6.5 [6.1, 7.0]
Carb intake during activity	23.14% took	6.34% took
Hours since last meal	<1 h: 39.22%	<1 h: 29.84%
	2 h: 30.78%	2 h: 26.08%
	3 h: 10.98%	3 h: 14.75%
	4 h: 6.67%	4 h: 5.86%
	>4 h: 12.35%	>4 h: 23.47%
BMI	25.28 (3.73)	25.16 (3.69)
	24.87 [22.44, 27.43]	24.39 [22.71, 26.63]
IoB	3.71 (2.52)	3.28 (2.5)
	3.09 [1.91, 4.81]	2.59 [1.61, 4.22]
Glucose rate change	−0.72 (1.19)	0.06 (1.23)
	−0.7 [−1.4, −0.1]	0.0 [−0.6, 0.7]
T1D duration (years)	20.07 (13.11)	19.73 (14.05)
	16.0 [11.0, 26.0]	16.0 [9.0, 27.0]
Race	White: 92.75%	White: 93.66%
	Asian: 0.98%	Asian: 1.37%
	Black/African American: 0.39%	Black/African American: 1.17%
	American Indian/Alaskan native: 0.59%	American Indian/Alaskan native: 0.24%
	Multiple: 1.18%	Multiple: 1.11%
	Unknown: 0.20%	Unknown: 0.31%
	Not reported: 3.92%	Not reported: 2.13%
Ethnicity	Not Hispanic or Latino: 96.47%	Not Hispanic or Latino: 95.42%
	Hispanic or Latino: 0.59%	Hispanic or Latino: 2.39%
	Do not wish to answer: 2.75%	Do not wish to answer: 1.85%
	Do not know: 0.20%	Do not know: 0.34%
Insulin delivery mode	AID (closed loop): 50.59%	AID (closed loop): 54.31%
	Pump (open loop): 49.41%	Pump (open loop): 45.69%
Hours asleep per day (h)	7.12 (0.92)	7.19 (0.95)
	7 [7, 8]	7 [7, 8]

The duration, starting glucose, age, HbA1c, and hours asleep per day are reported as mean (std) followed by median [IQR]. TBR, time below range.

Impact of type of PA

The secondary analysis investigated how different types of real-world PA influence glycemic responses by measuring the ΔCGM, LBGI, and percentage of sessions with hypoglycemia events for the 45 types of PA.

Figure 2a–c shows the distribution of ΔCGM, LBGI, and percentage of sessions with hypoglycemia events for the 45 real-world PA, sorted in ascending order by the median values. The analysis demonstrates a significant variation in all of the outcomes across different activity types (P < 0.05), underscoring the considerable variability in the glycemic response to real-world physical activities.

FIG. 2.

(a) Top: Distribution of ΔCGM between the end and start of the PA session for different activities. (b) Middle: Distribution of LBGI for different activities. (c) Bottom: Distribution of the percentage of sessions with hypoglycemia for different activities. “n” refers to the number of sessions for each activity. For each of the three metrics, ANOVA revealed significant variation across different activity types (P < 0.05). ANOVA, analysis of variance; ΔCGM, change in continuous glucose monitoring; LBGI, low blood glucose index.

The PA in Figure 2 is further color-coded by their category. The figures do not reveal a clear clustering of activities, as activities from different categories are dispersed throughout the plots rather than forming distinct groupings by the risk metrics.

Table 4 shows the distribution of the three metrics for activity sessions grouped by their category. Aerobic activities exhibited the largest glucose reduction (median [IQR] ΔCGM: −21.0 [−50.0, 3.0] mg/dL), followed by mixed (−11.5 [−38.0, 13.0] mg/dL) and anaerobic activities (−9.0 [−31.0, 12.5] mg/dL), P < 0.05. In contrast, the highest percentage of sessions with hypoglycemic events occurred during mixed activities (7.56%), followed by aerobic (6.01%) and anaerobic (5.28%) activities. No significant difference was observed in LBGI across activity categories (P = 0.32).

Table 4.

The Change in Continuous Glucose Monitoring, Low Blood Glucose Index, and Percentage of Sessions with Hypoglycemia for Aerobic, Mixed, and Anaerobic Activities

Activity type	Number of sessions	Number of participants	ΔCGM	P value	LBGI	P value	Sessions with hypoglycemia (%)
Aerobic	5340	391	−24.54 (45.22)−21.0 [−50.0, 3.0]	<0.05	1.2 (3.08)0.0 [0.0, 0.8]	0.318	6.01
Mixed	1732	307	−14.03 (46.51)−11.5 [−38.0, 13.0]		1.27 (3.26)0.0 [0.0, 0.87]		7.56
Anaerobic	1099	226	−10.18 (40.24)−9.0 [−31.0, 12.5]		1.09 (2.69)0.0 [0.0, 0.64]		5.28

Discussion

This analysis of the real-world PA sessions of the T1DEXI study demonstrated that hypoglycemia risk associated with real-world PA varied significantly with PA duration, starting glucose, CHO intake, activity time, IoB, and the preactivity rate of glucose change. At the population level, risk differed significantly by sex, BMI, and T1D duration, but did not vary by age.

In contrast to structured PA sessions, real-world PA exhibited substantial variability in duration (40.77 ± 22.78 min). Due to the unstructured nature of these sessions, their daily patterns, including intra- and interday frequency, also varied. While previous studies on structured PA have primarily focused on sessions of fixed duration (e.g., 30 min),^13,32 this analysis leveraged the variability in real-world PA duration, finding that longer sessions (60 to ≤120 min) were associated with a higher risk of hypoglycemia, characterized by a significantly greater glucose drop, higher LBGI, and a greater percentage of sessions with hypoglycemic events. These findings highlight the need for additional precautions to manage glycemia effectively during prolonged real-world PA.

Consistent with previous findings,^13,33 higher starting glucose levels correlated with a larger drop in glucose. However, the hypoglycemia risk, as measured by LBGI and the percentage of sessions with hypoglycemic events, was significantly higher for low starting glucose values (<100 mg/dL), indicating that individuals beginning PA with lower glucose levels are more vulnerable to hypoglycemia despite experiencing smaller absolute glucose declines.

The hypoglycemia risk, as indicated by all three metrics, also increased with a decreasing rate of glucose change before activity, in line with previous research.³³ While LBGI did not significantly vary based on starting IoB, higher IoB was associated with significantly larger glucose drops and a greater percentage of sessions with hypoglycemic events.

Although carbohydrate intake within 1 h of activity was associated with smaller glucose declines, it also coincided with higher LBGI and more frequent hypoglycemia. These paradoxical findings that higher carbohydrate intake is associated with greater hypoglycemia risk likely reflect higher IoB in situations where carbohydrates are consumed or behavioral responses to perceived or actual hypoglycemia risk. As expected, sessions with rescue carb intake had larger glucose drops, higher LBGI, and a greater likelihood of hypoglycemia, which suggests a reversed causality (i.e., carbohydrate intake occurs in the setting of high hypoglycemia risk or when hypoglycemia occurs). Because carbohydrate intake is known to lower hypoglycemia risk if insulin is not coadministered,¹³ this pattern suggests that carbs were consumed in response to these events rather than causing them. To explore this further, we conducted a follow-up analysis (Fig. 3), where sessions were classified based on the timing of carbohydrate intake relative to the activity. Intake was considered before activity if the last meal occurred within 1 h prior or if a snack was reported beforehand and during activity if rescue carbs were consumed mid-activity. Accordingly, sessions were labeled as “Before Activity Only” if carbs were taken only before activity, “During Activity Only” if taken only during, “Both Before and During” if taken at both times, and “Not Around Activity” if neither condition was met. The percentage of sessions in each category was then plotted against varying starting glucose levels (Fig. 3a) and IoB (Fig. 3b). The trends indicate that carbohydrate intake—both before and during activity—was more common when starting glucose levels were low or when IoB was high, conditions associated with elevated risk of exercise-related hypoglycemia. This suggests that the increased hypoglycemia observed in sessions with rescue carb use or meal intake within 1 h before activity may reflect preventive or corrective behaviors in response to perceived risk. This behavior may also explain why glucose declines were somewhat mitigated despite higher LBGI and a greater percentage of hypoglycemic events, a pattern similarly seen in sessions starting with lower glucose. In addition, higher IoB following recent meal consumption may contribute to this effect. Taken together, these findings imply that carbohydrate intake is not the cause of increased hypoglycemia risk, but rather a response to sessions characterized by lower starting glucose and higher IoB, both of which predispose to hypoglycemia.

FIG. 3.

Distribution of carbohydrate intake timing for varying (a) starting blood glucose levels (top) and (b) IoB (bottom). The legend categories indicate whether carbohydrates were consumed “Before Activity Only” (last meal within 1 h or snack before activity), “During Activity Only” (rescue carbs taken mid-activity), “Both Before and During,” or “Not Around Activity” (no carb intake near the activity). IoB, insulin on board.

Real-world PA sessions performed in the morning were associated with the lowest hypoglycemia risk, a trend also observed in structured PA sessions performed by adults,¹³ but not in real-world PA sessions performed by adolescents.¹⁹ The potential causes of the lower hypoglycemia risk in the morning than in the afternoon include lower insulin sensitivity,³⁴ less suppression of endogenous glucose production,³⁴ lipolysis-enhancing (glucose-sparing) effect of higher growth hormone levels,³⁵ higher plasma cortisol levels,^36,37 and an absence of insulin doses before PA in the morning.³⁷ This finding supports the recommendation to reduce bolus insulin after PA in the afternoon or evening,¹² periods found to be associated with a higher risk of nocturnal hypoglycemia.^37,38

Out of the population-level factors, sex, BMI, and T1D duration correlated with the risk of hypoglycemia.

First, sessions conducted by male participants showed a slightly higher risk of hypoglycemia compared with female participants (2 percentage points more sessions with hypoglycemia), a result that is consistent with a previous report on resistance exercise.³⁹ Further analysis indicated that sessions by male participants tended to have a lower starting glucose (140 ± 48 mg/dL) compared with female participants (149 ± 52 mg/dL). In addition, beyond the top four most popular activities overall, the next most common activity for female participants was “Pilates, Yoga, or PiYo,” which resulted in a smaller glucose drop and LBGI (−9 ± 36 mg/dL and 1.35 ± 3.19, respectively) compared with the next most common activity for male participants, “Outdoor Chores,” with a ΔCGM and LBGI of −21 ± 45 mg/dL and 1.68 ± 3.30, respectively. These factors may help explain the observed higher risk for PA-related hypoglycemia in males than in females.

Second, participants with a higher BMI (≥30 kg/m²) exhibited a smaller glucose drop compared with those with a lower BMI (<25 kg/m²), which has been previously found for real-world PA sessions performed by adolescents,¹⁹ but not for structured PA sessions performed by adults.¹³

Third, participants with a longer duration of T1D (≥10 years) had a higher LBGI compared with those with 5 to <10 years of T1D duration. This is consistent with the previous finding that the group of adult participants with the longest duration of T1D (≥10 years) has the largest number of hypoglycemia events on both sedentary days and days with structured PA.¹³ This previous study also suggested that participants with shorter duration of T1D have higher TIR on both sedentary days and days with structured PA. For adolescents, in contrast, the glucose drop for real-world PA sessions has been found to be greater for those with shorter duration of T1D.¹⁹

Finally, participants using AID closed-loop systems demonstrated lower LBGI values and a reduced percentage of sessions with hypoglycemia compared with those using open-loop pump systems. These findings suggest that closed-loop algorithms may offer a protective effect against exercise-induced hypoglycemia, likely due to their ability of real-time glucose monitoring and automated insulin adjustments.^40,41 This supports prior findings indicating that closed-loop systems can improve TIR and reduce exercise-related hypoglycemia in people with T1D.^40,42

The baseline HbA1c group did not correlate with glucose drop or LBGI in the real-world PA sessions, whereas this factor correlated with glucose drop for structured PA sessions performed by adults¹³ and real-world PA sessions performed by adolescents.¹⁹ In both structured and real-world PA sessions performed by adults, those with a lower HbA1c level experienced a greater drop in glucose.

Age did not correlate with glucose drop or LBGI for real-world exercise sessions. This has also been observed for both structured exercise sessions performed by adults¹³ and real-world exercise sessions performed by adolescents,¹⁹ despite the different age bins.

Prior research has shown that aerobic exercise lowers blood glucose in individuals with T1D,⁴³ whereas high-intensity interval training or resistance exercise leads to an increase.^44–46 An analysis of structured PA sessions in the T1DEXI dataset found the largest glucose drop with aerobic exercise, followed by interval and resistance.¹³ Expanding on these findings, our work performs a novel analysis measuring the impact of 45 types of real-world activities, revealing significant variation in glucose drops, LBGI, and the percentage of sessions with hypoglycemic events (Fig. 2a). Fushimi et al.¹⁴ showed that their model failed to categorize real-world activities that led to an increase in ΔCGM as aerobic, interval, or resistance, leaving them unclassified. Similarly, our analysis showed an average glucose increase for some real-world activities, such as “horseback riding” and “stretching.”

Contrary to previous findings, the activities did not always follow the expected trend of aerobic exercise leading to the highest glucose drop and hypoglycemia risk. For example, sessions from the mixed activity category “outdoor chores” showed a greater median glucose drop (−20 mg/dL) and LBGI (0.17) than the aerobic activities “walking, dog walking” (median ΔCGM −18 mg/dL, LBGI 0.0) and “swimming” (median ΔCGM −7 mg/dL, LBGI 0.0).

When grouped into aerobic, mixed, and anaerobic categories, aerobic activities exhibited the largest glucose drop, followed by mixed and then anaerobic, consistent with prior research. However, mixed activities had the highest percentage of sessions with hypoglycemic events, whereas LBGI did not significantly differ across activity categories. Furthermore, the analysis showed that glycemic variation is influenced by several critical factors beyond activity type, including baseline glucose levels, activity duration, carbohydrate intake, and IoB. These findings underscore the need for fine-grained insulin adjustments to optimize glycemic control across diverse real-world activity types.

Interestingly, 127 real-world exercise sessions of 15 activities (1.6% of total sessions) led to low LBGI and no hypoglycemia events (Fig. 2b, c). Previous literature^13,34,47,48 has demonstrated that activities predominantly conducted in the morning, fasted state, or anaerobic intensity activities may lead to a rise in glucose concentrations and may carry lower hypoglycemia risk. Because these activities have fewer than 20 sessions each, we cannot conclude that these activities carry no risk of hypoglycemia, even if no cases were observed in the available data.

This analysis aims to provide insights to guide improvements in AID systems for better management of PA, where each session-level or population-level factor is analyzed independently. However, implementation of specific AID settings for each activity type observed here cannot yet be implemented until larger and longer observational and interventional studies are conducted. Future work will focus on leveraging these results to further enhance AID system performance during PA, while exploring interdependence of all factors in greater depth.

One limitation of our work is that we leave the sessions performed by MDI users for further analysis. Due to the difference in this insulin delivery modality from CSII, those sessions require a separate and dedicated investigation of the activity-level factors. Another limitation is that some activity types, such as archery, skating, and violin performing, are underrepresented in the T1DEXI dataset. These activity types all have only a few available sessions, so the activity type effect may be misrepresented. In addition, our analysis is based on self-reported activity types, which might introduce personal biases when the boundaries between activity types are ambiguous (e.g., running and jogging). Moreover, to examine session-level factors and activity types, we treated each activity session as independent rather than computing per-participant averages for glycemia metrics. The statistical significance of our results should therefore be interpreted in the context of this approach. In addition, the sample is skewed toward individuals who identify as female, White, and non-Hispanic or Latino, with good glycemic control (low incidence of CGM-measured hypoglycemic events) and high levels of PA, which may limit the generalizability of the findings. Finally, for participants using closed-loop insulin delivery systems, the dataset does not indicate whether exercise mode or custom activity profiles were activated during nonstructured activity sessions. As a result, we are unable to account for the influence of these system features on hypoglycemia risk during real-world PA.

Conclusions

An analysis was conducted for 8171 real-world PA sessions comprising 45 different activity types from the T1DEXI dataset, with the goal of exploring how PA-related hypoglycemia risk correlates with activity-level and population-level factors. The results from the primary analysis show a highly variable hypoglycemia risk, with increased risk associated with sessions of longer duration, lower baseline glucose levels, and high IoB. The primary analysis also revealed that participants frequently consumed carbohydrates, likely as a preventative measure, for activities with lower starting glucose and higher IoB levels. Despite this, sessions with carb intake before and during the activity were associated with larger glucose drops and higher hypoglycemia risk, highlighting the need for more nuanced treatment adjustments considering a wide range of factors. Furthermore, closed-loop AID systems were associated with lower LBGI and fewer hypoglycemic events compared with open-loop pumps, indicating the potential benefits of AID adjustments during PA. The secondary analysis revealed significant glycemic variability across different activity types. While aerobic activities resulted in the largest glucose drops, followed by mixed and anaerobic activities, mixed activities had the highest percentage of sessions with hypoglycemia. Notably, LBGI did not differ significantly across activity categories. These findings emphasize the importance of tailored treatment strategies for real-world PA guidelines/decision support systems so that individuals with T1D can safely increase their daily PA and harness its significant health benefits, even outside structured exercise.

Authors’ Contributions

M.D.: Methodology, software, formal analysis, investigation, data curation, writing—original draft, writing—review and editing, and visualization. K.T.: Methodology, software, formal analysis, investigation, data curation, writing—original draft, writing—review and editing, and visualization. E.M.A.: Conceptualization, writing—review and editing, and supervision. D.P.Z.: Conceptualization, validation, writing—review and editing, supervision, and funding acquisition. R.A.L.: Conceptualization, validation, writing—review and editing, and supervision. C.S.: Conceptualization, methodology, supervision, and project administration. B.A.: Conceptualization, supervision, project administration, and funding acquisition. K.W.: Conceptualization and supervision. M.J.C.: Methodology, resources, and writing—review and editing. L.E.F.: Writing—review and editing. I.B.: Writing—review and editing. A.L.C.: Writing—review and editing. R.S.K.: Writing—review and editing and project administration. B.S.: Writing—review and editing. M.C.R.: Conceptualization, validation, writing—review and editing, and supervision. Y.Q.: Conceptualization, methodology, validation, formal analysis, investigation, resources, data curation, writing—original draft, writing—review and editing, supervision, project administration, and funding acquisition.

Footnotes

Author Disclosure Statement

E.M.A. was with the University of Trento, Trento, Italy. She is now with the University of Pavia. D.P.Z. has received honoraria for speaking engagements from Ascensia Diabetes, Insulet Canada, and Medtronic Diabetes and serves on the advisory board of DexCom Inc. and the Diabetes Research Hub. R.A.L. consults for Abbott Diabetes Care, Adaptyx Biosciences, Biolinq, Capillary Biomedical, Deep Valley Labs, Gluroo, Portal Diabetes, Sanofi, and Tidepool. He has served on advisory boards for Provention Bio and Lilly. He receives research support from his institution from Insulet, Medtronic, Sinocare, and Tandem. C.S. is employed by Tidepool. B.A. is employed by Tidepool and is a shareholder of Eli Lilly stock. K.W. is employed by Tidepool. M.J.C. is employed by Tidepool. M.C.R. has received consulting fees from Eli Lilly, Embecta, the Jaeb Center for Health Research, Zealand Pharma, and Zucara Therapeutics; speaker fees from Dexcom Canada, Eli Lilly, Novo Nordisk, and Sano Diabetes; and stock options from Zucara Therapeutics. Y.Q. is with the University of California, Santa Barbara and Google. No other potential conflicts of interest relevant to this article were reported.

Funding Information

Research reported in this publication was supported by the Leona M. and Harry B. Helmsley Charitable Trust (G-2404-06905).

Supplemental Material

References

American Diabetes Association. Standards of medical care in diabetes—2019 abridged for primary care providers. Clin Diabetes, 2019; 37(1):11–34.

Caspersen

, Powell

, Christenson

. Physical activity, exercise, and physical fitness: Definitions and distinctions for health-related research. Public Health Rep, 1985; 100(2):126–131.

Zaharieva

, Addala

. Current and novel strategies to reduce fear of hypoglycemia as a barrier to physical activity in adults and youth with type 1 diabetes. Can J Diabetes, 2022; 46(1):1–2.

Brennan

, Brown

, Leslie

, Ntoumanis

. Acceptability of self-management group education to reduce fear of hypoglycemia as a barrier to physical activity in adults with type 1 diabetes: A mixed methods approach. Can J Diabetes, 2022; 46(1):16–25.e2.

Cigrovski Berkovic

, Bilic-Curcic

, La Grasta Sabolic

, et al. Fear of hypoglycemia, a game changer during physical activity in type 1 diabetes mellitus patients. World J Diabetes, 2021 May 15;12(5):569–577.

Roberts

, Taplin

, Isom

, et al. Association between fear of hypoglycemia and physical activity in youth with type 1 diabetes: The SEARCH for diabetes in youth study. Pediatr Diabetes, 2020; 21(7):1277–1284.

Brennan

, Albrecht

, Brown

, et al. Self-Management group education to reduce fear of hypoglycemia as a barrier to physical activity in adults living with type 1 diabetes: A pilot randomized controlled trial. Can J Diabetes, 2021; 45(7):619–628.

Brazeau

, Rabasa-Lhoret

, Strychar

, Mircescu

. Barriers to physical activity among patients with type 1 diabetes. Diabetes Care, 2008; 31(11):2108–2109.

Brennan

, Brown

, Ntoumanis

, Leslie

. Barriers and facilitators of physical activity participation in adults living with type 1 diabetes: A systematic scoping review. Appl Physiol Nutr Metab, 2021; 46(2):95–107.

10.

Huerta-Uribe

, Hormazábal-Aguayo

, Izquierdo

, García-Hermoso

. Youth with type 1 diabetes mellitus are more inactive and sedentary than apparently healthy peers: A systematic review and meta-analysis. Diabetes Res Clin Pract, 2023; 200:110697.

11.

Colberg

, Sigal

, Yardley

, et al. Physical activity/exercise and diabetes: A position statement of the American diabetes association. Diabetes Care, 2016; 39(11):2065–2079.

12.

Riddell

, Gallen

, Smart

, et al. Exercise management in type 1 diabetes: A consensus statement. Lancet Diabetes Endocrinol, 2017; 5(5):377–390.

13.

Riddell

, Li

, Gal

, et al.; T1DEXI Study Group. Examining the acute glycemic effects of different types of structured exercise sessions in type 1 diabetes in a real-world setting: The Type 1 Diabetes and Exercise Initiative (T1DEXI). Diabetes Care, 2023; 46(4):704–713.

14.

Fushimi

, Aiello

, Cho

, et al. Online Classification of Unstructured Free-Living Exercise Sessions in People with Type 1 Diabetes. Diabetes Technol Ther, 2024; 26(10):709–719.

15.

Turner

, Marak

, Gal

, et al.; T1DEXI Study Group. Associations between daily step count classifications and continuous glucose monitoring metrics in adults with type 1 diabetes: Analysis of the Type 1 Diabetes Exercise Initiative (T1DEXI) cohort. Diabetologia, 2024; 67(6):1009–1022.

16.

Kamimoto

JLJ

, Li

, Gal

, et al. Effect of impaired awareness of hypoglycemia on glucose decline during and after exercise in the T1DEXI study. J Clin Endocrinol Metab, 2024; 109(9):2233–2241.

17.

, Calhoun

, Rickels

, et al. Factors affecting reproducibility of change in glucose during exercise: Results from the type 1 diabetes and exercise initiative. J Diabetes Sci Technol, 2025; 19(4):958–970.

18.

, Yardley

, Zaharieva

, et al. Changing glucose levels during the menstrual cycle as observed in adults in the type 1 diabetes exercise initiative study. Can J Diabetes, 2024; 48(7):446–451.

19.

Riddell

, Gal

, Bergford

, et al. The acute effects of real-world physical activity on glycemia in adolescents with type 1 diabetes: The Type 1 Diabetes Exercise Initiative Pediatric (T1DEXIP) study. Diabetes Care, 2024; 47(1):132–139.

20.

Patton

, Bergford

, Sherr

, et al. Postprandial glucose variability following typical meals in youth living with type 1 diabetes. Nutrients, 2024; 16(1):162.

21.

Sherr

, Bergford

, Gal

, et al. Exploring factors that influence postexercise glycemia in youth with type 1 diabetes in the real world: The Type 1 Diabetes Exercise Initiative Pediatric (T1DEXIP) Study. Diabetes Care, 2024; 47(5):849–857.

22.

Bergford

, Riddell

, Gal

, et al. Predicting hypoglycemia and hyperglycemia risk during and after activity for adolescents with type 1 diabetes. Diabetes Technol Ther, 2024; 26(10):728–738.

23.

Riddell

, Li

, Beck

, et al. More time in glucose range during exercise days than sedentary days in adults living with type 1 diabetes. Diabetes Technol Ther, 2021; 23(5):376–383.

24.

Clarke

, Kovatchev

. Statistical tools to analyze continuous glucose monitor data. Diabetes Technol Ther, 2009; 11(Suppl 1):S45–S54.

25.

Kovatchev

, Cox

, Gonder-Frederick

, et al. Assessment of risk for severe hypoglycemia among adults with IDDM: Validation of the low blood glucose index. Diabetes Care, 1998; 21(11):1870–1875.

26.

Jacobs

, Chase Marak

, Calhoun

, et al. An evaluation of how exercise position statement guidelines are being used in the real world in type 1 diabetes: Findings from the type 1 diabetes exercise initiative (T1DEXI). Diabetes Res Clin Pract, 2024; 217:111874.

27.

Bergford

, Riddell

, Jacobs

, et al. The type 1 diabetes and exercise initiative: Predicting hypoglycemia risk during exercise for participants with type 1 diabetes using repeated measures random forest. Diabetes Technol Ther, 2023; 25(9):602–611.

28.

Gondhalekar

, Dassau

, Doyle

. Velocity-weighting & velocity-penalty MPC of an artificial pancreas: Improved safety & performance. Automatica (Oxf), 2018; 91:105–117.

29.

Hochberg

, Tamhane

. Multiple Comparison Procedures [Internet]. 1st ed.. Wiley; 1987. (Wiley Series in Probability and Statistics). Available from: https://onlinelibrary-wiley-com-443.web.bisu.edu.cn/doi/book/10.1002/9780470316672 [Last accessed: August 1, 2024].

30.

Herrmann

, Willis

, Ainsworth

, et al. 2024 Adult compendium of physical activities: A third update of the energy costs of human activities. J Sport Health Sci, 2024; 13(1):6–12.

31.

Craig

, Marshall

, Sjöström

, et al. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc, 2003; 35(8):1381–1395.

32.

Cho

, Aiello

, Ozaslan

, et al. Design of a real-time physical activity detection and classification framework for individuals with type 1 diabetes. J Diabetes Sci Technol, 2024; 18(5):1146–1156.

33.

Riddell

, Zaharieva

, Tansey

, et al. Individual glucose responses to prolonged moderate intensity aerobic exercise in adolescents with type 1 diabetes: The higher they start, the harder they fall. Pediatr Diabetes, 2019; 20(1):pedi.12799–p106.

34.

Toghi-Eshghi

, Yardley

. Morning (Fasting) vs afternoon resistance exercise in individuals with type 1 diabetes: A randomized crossover study. J Clin Endocrinol Metab, 2019; 104(11):5217–5224.

35.

Yardley

. Fasting may alter blood glucose responses to high-intensity interval exercise in adults with type 1 diabetes: A randomized, acute crossover study. Can J Diabetes, 2020; 44(8):727–733.

36.

Riddell

, Turner

, Patton

. Is there an optimal time of day for exercise? A commentary on when to exercise for people living with type 1 or type 2 diabetes. Diabetes Spectr, 2023 May 15;36(2):146–150.

37.

Gomez

, Gomez

, Aschner

, et al. Effects of performing morning versus afternoon exercise on glycemic control and hypoglycemia frequency in type 1 diabetes patients on sensor-augmented insulin pump therapy. J Diabetes Sci Technol, 2015 May;9(3):619–624.

38.

Davey

, Howe

, Paramalingam

, et al. The effect of midday moderate-intensity exercise on postexercise hypoglycemia risk in individuals with type 1 diabetes. J Clin Endocrinol Metab, 2013 Jul;98(7):2908–2914.

39.

Brockman

, Sigal

, Kenny

, et al. Sex-Related differences in blood glucose responses to resistance exercise in adults with type 1 diabetes: A secondary data analysis. Can J Diabetes, 2020; 44(3):267–273.e1.

40.

Eckstein

, Weilguni

, Tauschmann

, et al. Time in range for closed-loop systems versus standard of care during physical exercise in people with type 1 diabetes: A systematic review and meta-analysis. J Clin Med, 2021; 10(11):2445.

41.

Peyser

, Dassau

, Breton

, Skyler

. The artificial pancreas: Current status and future prospects in the management of diabetes. Ann N Y Acad Sci, 2014; 1311:102–123.

42.

American Diabetes Association Professional Practice Committee. Summary of Revisions: Standards of Care in Diabetes-2024. Diabetes Care, 2024; 47(Supplement_1):S5–S10.

43.

Tuominen

, Karonen

, Melamies

, et al. Exercise-induced hypoglycaemia in IDDM patients treated with a short-acting insulin analogue. Diabetologia, 1995; 38(1):106–111.

44.

Aronson

, Brown

, Li

, Riddell

. Optimal insulin correction factor in post–high-intensity exercise hyperglycemia in adults with type 1 diabetes: The FIT study. Diabetes Care, 2019; 42(1):10–16.

45.

Turner

, Luzio

, Gray

, et al. Algorithm that delivers an individualized rapid‐acting insulin dose after morning resistance exercise counters post‐exercise hyperglycaemia in people with Type 1 diabetes. Diabet Med, 2016; 33(4):506–510.

46.

Harmer

, Chisholm

, McKenna

, et al. High-intensity training improves plasma glucose and acid-base regulation during intermittent maximal exercise in type 1 diabetes. Diabetes Care, 2007; 30(5):1269–1271.

47.

McClure

, Alcántara-Cordero

, Weseen

, et al. Systematic review and meta-analysis of blood glucose response to high-intensity interval exercise in adults with type 1 diabetes. Can J Diabetes, 2023; 47(2):171–179.

48.

Fahey

, Paramalingam

, Davey

, et al. The effect of a short sprint on postexercise whole-body glucose production and utilization rates in individuals with type 1 diabetes mellitus. J Clin Endocrinol Metab, 2012; 97(11):4193–4200.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.82 MB