Calibrating Dexcom G7 improves its performance in the context of repetitive recreational scuba diving in people with type 1 diabetes

Abstract

Background

The use of continuous glucose monitors (CGM) in scuba diving for patients with type 1 diabetes (T1DM) shows potential but faces challenges related to accuracy. Previous research has highlighted the poor accuracy of the Dexcom G7 (DG7) in repetitive diving contexts. This study investigates the impact of calibration on the accuracy of DG7, providing valuable insights for patients and clinicians.

Materials and methods

In August 2024, ‘Diabete Sommerso’ organised a 4-day diving cruise around Elba Island (Italy) with 15 participants, including individuals with T1DM. Each participant with diabetes wore two DG7 sensors (one on the arm and one on the abdomen), calibrated daily and compared the results to capillary glucose (Beurer GL50Evo as the reference). Accuracy was assessed using mean absolute relative difference (MARD)/median ARD, Food and Drug Administration (FDA) integrated continuous glucose monitoring (iCGM) criteria and Surveillance Error Grid (SEG) analysis. Hypoglycaemia detection and trends were also evaluated.

Results

Eight participants with T1DM completed the study using 16 DG7 sensors with no detachments or skin reactions. Analysis of 765 sensor-capillary glucose pairs during 68 dives showed an overall MARD of 13.7%, with arm sensors (11% MARD) outperforming abdomen sensors (16%, p=0.0001). SEG analysis revealed that more than 97% of readings fell within the no-risk zone; however, the FDA’s iCGM criteria for non-adjunctive use were not met.

Conclusions

Calibration improved the accuracy of DG7 in repetitive diving for patients with T1DM. However, capillary glucose checks remain essential, as non-adjunctive criteria were not met.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Scuba diving is now also possible for people with diabetes thanks to the development of dedicated protocols and the use of glucose sensors has proven useful in monitoring glycaemic trends and preventing hypoglycaemia.

WHAT THIS STUDY ADDS

• The sensor Dexcom G7 proved to underperform versus other competitors, with a possible effect of lack of calibration in previous study and which we explore in the present study, finding a significant improvement in accuracy. However, criteria for non-adjunctive use are not met.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Results of this study confirm that capillary glucose checks remain essential for people with type 1 diabetes for safe planning of scuba diving, underlining the importance of a conscious and contextualised use of technology as part of a broader educational and therapeutic framework. Future research may focus on further exploring factors impacting on glucose sensors accuracy in this peculiar high-risk setting.

Introduction

The use of continuous interstitial glucose sensors (CGMs) during physical activity is still hindered by problems concerning sensor adhesion and data accuracy to date. Even in the field of scuba diving, an activity which is currently possible also for people with type 1 diabetes mellitus (T1DM) adequately selected and trained according to specific protocols,¹,³ research has demonstrated the potential of CGM. The effectiveness and safety of retrospectively using glucose sensors have been validated, highlighting their great usefulness in dive preparation, particularly for identifying trends and preventing hypoglycaemia.⁴,⁹ This is not the case with real-time use. Beyond data transmission problems in water, sensor accuracy still needs to be evaluated with extreme caution due to the peculiar setting’s impact on sensor performance. This is the case of FreeStyle Libre 3 and Dexcom G7, as reported in our recent publication,¹⁰ where we explored the possibility of using these two continuous glucose monitoring models in the context of repetitive diving in patients with T1DM, analysing their performance in terms of accuracy. We have shown that, despite the peculiar condition characterised by the presence of the wetsuit and multiple dives per day, these devices can be used without any particular problem of skin reactions and/or detachment before the prescribed time, with a reinforcing patch applied at the time of their application. Concerning the accuracy of the real-time data provided by the two factory-calibrated sensors, data collected showed a significant superiority of the FreeStyle Libre 3 over the Dexcom G7 in terms of analytical and clinical accuracy (except for the hypoglycaemic range and on the day after the diving cycle), with a mean absolute relative difference (MARD) of 14.2% and 31%, respectively. Both sensors fail to achieve the standard of accuracy required for non-adjunctive use.^{11 12} Therefore, in our study, we conclude that sensors can represent an additional tool and support, especially in the early identification of hypoglycaemia; however, capillary tests remain the essential tool for safe dive planning, as recommended by international protocols. We did not have an explanation for the large difference in accuracy between the two sensors, except that study participants were expressly requested not to calibrate the Dexcom G7 to avoid biasing the results against the FreeStyle Libre 3, which does not allow calibration. We therefore speculate that the Dexcom G7, although factory-calibrated, may need, in this particular context, one or more calibrations. Furthermore, we found that only the sensor application site was significant in terms of performance for the Dexcom G7, favouring the abdomen over the arm.

Then, although our results have provided new useful information for clinical practice, several questions remain open. These include, in particular, the possibility of improved performance of the Dexcom G7 model when calibrated.

The present study aims to extend these findings by investigating the effects of calibration on the accuracy of the Dexcom G7 in a repetitive diving context in patients with T1DM, thereby complementing previous observations and providing further clarification and useful information for patients and clinicians. Indeed, the reliability of CGM data in this high-risk context becomes a critical factor for safe decision-making. Understanding device performance in such settings is therefore essential to support informed and personalised therapeutic choices.

Materials and methods

Case selection and study design

In August 2024, one of the activities promoted by the association ‘Diabete Sommerso’ was a short 4-day cruise around the island of Elba (Italy). 15 people, including patients with T1DM, from Italy and Switzerland participated. T1DM subjects who signed informed consent from the non-profit organisation were included in the study. Participants enrolled wore two Dexcom G7 sensors simultaneously, one on their arm (DG7-ar) and one on the abdomen (DG7-ab), placed 16 hours before the first dive. They were asked to calibrate each sensor at the end of the run-in phase, the following morning and every 24 hours thereafter, always fasting in the morning on a glycaemic stability time. For safe diving, the Diabete Sommerso protocol was followed,^{13 14} with multiple capillary glycaemic readings taken before and after each dive. All glucose capillary tests were performed with a Beurer GL50 glucometer. The participants were asked to keep the sensors on for the 5 days after the end of the cruise, continuing the daily calibration and capillary glycaemic checks. No patients or members of the public were involved in the design, conduct, reporting or dissemination plans of this research.

Reference standard for glucose data

Capillary blood glucose was used as a reference standard to assess the accuracy of Dexcom G7. Sensor glucose data were collected 1 min before or after the self-monitoring of blood glucose (SMBG) data.

Measures

Concerning analytical accuracy, considering the non-normal distribution of the data, but at the same time the clinical relevance of the outliers, we perform both median ARD (MeARD) and MARD on overall data, within the glycaemic ranges <70 mg/dL (<4 mmol/L), 70–180 mg/dL (4–10 mmol/L), >180 mg/dL (>10 mmol/L), 180–250 mg/dL (10–13.9 mmol/L), >250 mg/dL (>13. 9 mmol/L), in the safety protocol suggested range between 150 and 250 mg/dL (8.3–13.9 mmol/L), according to the participant gender, per single subject, on dive days only, on no dive days only and according to the breathed mixture (air vs nitrox).

An analysis of the trends identified from the capillary data and the sensor data in different time intervals was carried out, and the consistency of the sensor trend arrow with the capillary glucose value was assessed.

We also performed an analysis according to the Food and Drug Administration (FDA) criteria (points V, A–I) for integrated continuous glucose monitoring (iCGM).¹²

The clinical accuracy was analysed via Surveillance Error Grid (SEG) Analysis using a web-based tool available at https://www.diabetestechnology.org/seg/.

Finally, the ability of the two sensors to correctly identify hypoglycaemia was assessed by contingency tables.

Statistical analysis

A two-tailed Spearman’s test was conducted to evaluate the correlation between capillary and sensor data. Bland-Altman plots illustrated data distribution and the bias between sensor and capillary measurements. To determine the significance of MeARD and MARD values and analyses according to FDA criteria for integrated continuous glucose monitoring (iCGMs), both Wilcoxon and t-tests were applied, as well as the Mann-Whitney test. The capability of the two sensors to accurately detect hypoglycaemia was assessed by calculating sensitivity, specificity, predictive values and diagnostic OR. This analysis considered three hypoglycaemia levels: level 1 (mild, 54–69 mg/dL or 3–3.8 mmol/L), level 2 (moderate, <54 mg/dL or <3 mmol/L) and level 3 (severe, requiring assistance). All statistical analyses and visualisations were conducted using GraphPad Prism V.7 (GraphPad Software, Boston, Massachusetts, USA).

Results

Participants’ characteristics

We included eight participants with T1DM, six women and two men, with a median age of 39.5 years (IQR 13.5 years), median diabetes duration of 21.5 years (IQR 8.8 years), median body mass index 21.7 kg/m² (IQR 1.9 kg/m²), median glycated haemoglobin (HbA1c) 6.6% or 49 mmol/mol (IQR 0.6% or 6.5 mmol/mol). Patients were on closed-loop (6), pump+CGM (1) or pens+CGM (1) therapy. The median total daily dose was 38 units/day (IQR 15.3 UI/day). Regarding complications, one of them had stable and non-proliferative diabetic retinopathy. No patients had a regular intake of paracetamol >1 gram/day. Concerning dive licences, four had PADI Advanced Open Water Diver, two had PADI Rescue Diver, one had PADI Assistant Instructor and one had FIPSAS P2. No detachments, premature shutdowns or adverse skin events occurred on the 16 sensors deployed.

Glycaemic data

Data were collected between dives, both at times specified by the safety protocol and at other intervals, including day and night, regardless of the diving schedule. CGM traces and capillary data were collected from a total of 68 dives, with an average duration of 50 min, at an average maximum depth of 24.8 m. For the accuracy analysis, 382 (for DG7-ab) and 383 (for DG7-ar) sensor-capillary glucose data pairs were evaluated. Data distribution was not normal, and global correlation analysis between CGM and SMBG data showed a Spearman’s r of 0.8473 (p<0.0001) for Dexcom G7 placed on the abdomen and 0.9120 (p<0.0001) for those placed on the arm. The average blood glucose value measured by the glucometer was 190±70 mg/dL (10.6±3.9 mmol/L) and the mean bias versus capillary data was 22 mg/dL (1.2 mmol/L) for DG7-ab and 11 mg/dL (0.6 mmol/L) for DG7-ar.

MARD and MeARD data

Data analysis revealed a statistically significant difference in overall MARD between DG7-ar and DG7-ab (11% vs 16%, p=0.0001; MeARD: 8.3% and 10.2%, respectively, p=0.0015). If considered without distinction according to the site of application, the overall MARD results are 13.7% and MeARD 9%.

Data correlation, distribution and overall MARD for DG7-ar and DG7-ab are shown in figure 1.

Figure 1. Data distribution (Bland-Altman plots), linear correlation (two-tailed Spearman’s), MeARD and MARD of Dexcom G7 placed on arm (DG7-ar)/abdomen (DG7-ab) versus SMBG Beurer GL50 Evo. MARD, mean absolute relative difference, MeARD, median ARD; SMBG, self-monitoring of blood glucose.

Regarding glucose ranges, the only clinical and statistically significant difference between the two sensor sites was found in the normoglycaemic range (70–180 mg/dL; 3.9–10 mmol/L), where sensors placed on the arms performed better than those placed on the abdomen (MARD 11.7% vs 20.6%, p<0.0001 and MeARD 7.9 vs 11.8%, p<0.0001). The same was observed in female subjects (MARD 12.2% vs 17.5%, p=0.0002 and MeARD 8.8 vs 10.7%, p=0.0016), while in males’ results, there was no significant divergence.

A significant performance difference between application sites was observed during diving days (DG7-ar MARD 10.9% vs DG7-ab MARD 15.9%, p<0.0001; DG7-ar MeARD 8.3% vs DG7-ab MeARD 10.7%, p=0.0003), whereas this gap narrowed and was no longer significant in the 5 days post-diving (DG7-ar MARD 15% vs DG7-ab 16.2%; DG7-ar MeARD 6.3% vs DG7-ab 7.2%).

Analysis by breathing gas mixture revealed a significant performance advantage for sensors applied to the arm, both in air and nitrox dives. Additionally, single-subject analysis showed markedly poorer performance of DG7-ab in subjects 1, 2, 4 and 8, reaching statistical significance in subjects 1, 4 and 8.

Seven episodes of hypoglycaemia were identified at capillary control (0.097 episodes/person/day), with no severe (level 3) episodes and no episodes occurring while diving or requiring dive interruption. Within this glycaemic range, the MARD was 50.1% for DG7-ab (n=7) and 17.6% for DG7-ar (n=6).

Table 1 presents the MARDs/MeARDs values of the two sensors placed at the two sites, overall, and in the different subcategories of gender, glycaemic range, dive versus no-dive days, air versus nitrox and per single patient.

Table 1. Dexcom G7 and FreeStyle Libre 3 MARD and MeARD (overall and subgroup analysis).

Category		Dexcom G7 - abdomen					Dexcom G7 - arm					P value
		n	MARD (%)	SD (%)	MeARD (%)	IQR	n	MARD (%)	SD (%)	MeARD (%)	IQR
Overall		382	16	22	10.2	14.6	383	11.4	11.4	8.3	11.5	0.0001 (MARD) 0.0015 (MeARD)
Gender	Male	69	9	8.4	6.1	11.5	72	8.1	5.9	7.9	10.1	0.2955 (MARD) 0.4696 (MeARD)
	Female	313	17.5	23.6	10.7	16.2	311	12.2	12.2	8.8	12	0.0002 (MARD) 0.0016 (MeARD)
Glucose range	<70 mg/dL (<3.9 mmol/L)	7	50.1	57.6	45.5	150	6	17.6	5.5	20.8	9.5	0.2373 (MARD) 0.2812 (MeARD)
	70–180 mg/dL (3.9–10 mmol/L)	174	20.6	27	11.8	21.3	176	11.7	12.8	7.9	13.4	<0.0001
	180–250 mg/dL (10–13.9 mmol/L)	127	10.1	9	7.9	11.7	127	12	10.2	10.1	10.2	0.0618 (MARD) 0.1119 (MeARD)
	>180 mg/dL (>10 mmol/L)	201	10.7	9.8	7.8	12.3	201	11.1	10.1	9	10.1	0.8781 (MARD) 0.9633 (MeARD)
	>250 mg/dL (>13.9 mmol/L)	74	11.8	11	7.7	15.4	74	9.5	9.8	6.8	9.9	0.0303 (MARD) 0.0532 (MeARD)
	Protocol range 150–250 mg/dL (8.3–13.9 mmol/L)	192	11.4	12.4	8	12.2	191	12.1	11.9	9.6	11.9	0.7025 (MARD) 0.4668 (MeARD)
Dives	Yes (days 1–4)	325	15.9	18.8	10.7	14.7	329	10.9	9.7	8.3	11.3	< 0.0001 (MARD) 0.0003 (MeARD)
	No (days 5–9)	57	16.2	34.8	6.3	11.65	54	15	18.5	7.2	17.9	0.6719 (MARD) 0.7793 (MeARD)
Gas	Air	66	17.5	25.1	10	13.6	66	8.7	7.6	7	8.9	0.0065 (MARD) 0.0116 (MeARD)
	Nitrox	175	13.8	18	7.8	13.4	174	10	9.1	7.5	11.9	0.0083 (MARD) 0.1251 (MeARD)
Patient	1	69	19.2	14.6	15.5	23.4	69	13.9	10.4	12.1	11.5	0.0082 (MARD) 0.0149 (MeARD)
	2	36	16.6	11.6	13.4	13.2	37	12.8	10.9	11.3	14.4	0.2645 (MARD) 0.0656 (MeARD)
	3	29	7	6.4	4.4	10.1	27	8.6	6.7	5.9	12	0.4017 (MARD) 0.2942 (MeARD)
	4	71	8.3	8.5	5.3	11.7	65	13.2	12.4	10.1	15.1	0.0064 (MARD) 0.0074 (MeARD)
	5	35	10.4	8.3	7.6	13	37	8.7	11.5	5.6	7	0.0145 (MARD) 0.0216 (MeARD)
	6	48	9	10	6.6	9.6	48	8.2	7	6.6	9.4	0.5139 (MARD) 0.8223 (MeARD)
	7	40	10.5	9.4	7.5	11.7	45	7.8	5.5	7.9	9.1	0.0609 (MARD) 0.0724 (MeARD)
	8	54	40.2	44.1	22.7	56.3	55	14.3	17.3	7.5	13.3	0.0005 (MARD) 0.0004 (MeARD)

MARD, mean absolute relative difference; MeARD, median ARD.

Other accuracy analysis

We conducted an analysis of glycaemic trends detected by SMBG and the two glucose sensors (DG7-ar and DG7-ab) during specific intervals within the protocol: between −60 and −10 min before the dive, and from −10 min before the dive to immediately after the dive. This analysis was based on 242 glycaemic data points. Considering the first interval (−60 vs −10 min), SMBG showed a glycaemic variation of +4.6% (9 mg/dL, 0.5 mmol/L). The DG7-ar showed a similar trend with an increase of +4.1% (8 mg/dL, 0.44 mmol/L), while DG7-ab detected no glycaemic change. Regarding the second interval (−10 min vs post-dive), SMBG showed no changes in glycaemic levels. DG7-ar showed a decrease of −3.5% (7 mg/dL, 0.39 mmol/L), and DG7-ab showed a slight increase of +0.5% (1 mg/dL, 0.06 mmol/L).

The sensor trend arrow was consistent with the capillary glucose value, accounting for the interstitial glucose at the time, in 73% of cases for G7 sensors applied on the arm and 71% for those on the abdomen.

Both sensors failed to meet the required thresholds at points (V, A–I) outlined in the FDA iCGM criteria, except for points H (no sensor value <70 mg/dL corresponding to an SMBG>180 mg/dL) and I (no sensor value >180 mg/dL corresponding to an SMBG<70 mg/dL), which were satisfied by both sensors. Additionally, DG7-ar met the requirement for point F (>99% of CGM measurements >180 mg/dL within ±40% of the corresponding blood glucose value). This analysis is reported in table 2.

Table 2. Analysis according to iCGM FDA 2017 criteria (point V, A–I).

CGM data range (mg/dL)	Dexcom G7 - abdomen									Dexcom G7 - arm
	n	% of CGM data within ±15/20/40 mg/dL of SMBG			n	n	Median value		P value (95% CI one-side)	n	% of CGM data within ±15/20/40 mg/dL of SMBG			n	n	Median value		P value (95% CI one-side)
		±15	±20	±40	<70	>180	CGM	SMBG			±15	±20	±40	<70	>180	CGM	SMBG
<70	4	50 (Goal >85%)		50 (Goal >98%)		0 (Goal =0)	54	45	0.25	11	18 (Goal >85%)		64 (Goal >98%)		0 (Goal =0)	59	73	0.0205
70–180	162	66 (Goal >70%)		89 (Goal >99%)			143	137	0.4672	175	75 (Goal >70%)		95 (Goal >99%)			132	140	0.0091
>180	216	60 (Goal >80%)		84 (Goal >99%)	0 (Goal=0)		231	226	<0.0001	197	79 (Goal >80%)		99.5 (Goal >99%)	0 (Goal=0)		241	232	<0.0001
All data	382		76 (Goal >87%)				194	187	<0.0001	383		80 (Goal >87%)				184	187	0.0097

CGMs, continuous interstitial glucose sensors; FDA, Food and Drug Administration; iCGMs, integrated continuous glucose monitoring; SMBG, self-monitoring of blood glucose.

We conducted a clinical accuracy analysis using the SEG methodology, which categorises discrepancies between the tested device and the gold standard into risk zones. To validate accuracy, the SEG requires at least 97% of data to fall within the green ‘no-risk’ zones, corresponding to <5% of data pairs deviating beyond the 15 mg/dL (0.83 mmol/L)/15% standard limits. Our analysis revealed that 99.7% of the data for the DG7-ar and 97.5% for the DG7-ab fell within the green A+B ‘no-risk’ zones, leaving 0.3% and 2.5% of data in the remaining risk zones, respectively (figure 2).

Figure 2. Surveillance Error Grid for Dexcom G7 placed on abdomen versus arm.

The ability of the two sensors to correctly identify hypoglycaemia before/after the dives was assessed using contingency tables. As shown in table 3, in a context of very low prevalence of hypoglycaemia, it appears that DG7-ar sensitivity is higher than that of DG7-ab, driven by level 2 hypoglycaemia.

Table 3. Results of contingency tables on hypoglycaemia identification.

	Dexcom G7 - abdomen	Dexcom G7 - arm
Overall
Prevalence of hypoglycaemia	0.018	0.016
Sensitivity	0.57	0.83
Specificity	1	1
Positive predictive value	1	0.83
Negative predictive value	0.992	1
Positive likelihood ratio	NA	314.17
Negative likelihood ratio	0.429	0.17
Diagnostic OR	NA	1880
Level 1
Prevalence of hypoglycaemia	0.005	0.003
Sensitivity	0	0
Specificity	0.995	0.982
Positive predictive value	0	0
Negative predictive value	0.995	0.997
Positive likelihood ratio	0	0
Negative likelihood ratio	1.005	1.019
Diagnostic OR	0	0
Level 2
Prevalence of hypoglycaemia	0.013	0.013
Sensitivity	0.4	0.8
Specificity	1	1
Positive predictive value	1	1
Negative predictive value	0.992	0.997
Positive likelihood ratio	NA	NA
Negative likelihood ratio	0.6	0.2
Diagnostic OR	NA	NA
Level 3
Prevalence of hypoglycaemia	0.00	0.00
Sensitivity	NA	NA
Specificity	NA	NA
Positive predictive value	NA	NA
Negative predictive value	NA	NA
Positive likelihood ratio	NA	NA
Negative likelihood ratio	NA	NA
Diagnostic OR	NA	NA
Prevalence of hypoglycaemia	NA	NA

Discussion

The rationale behind this study stems from a real-world observation: in a logistically complex context such as dive preparation, many people with T1DM attempt to simplify procedures and rely on CGM data for convenience, thereby reducing (or even skipping) capillary testing. Although understandable, this practice introduces risks, especially if sensor accuracy is compromised, as decisions based on incorrect data can lead to serious or even life-threatening consequences. Thus, a very strong need emerges to provide useful information, leading to the empowerment of people with diabetes who dive. The ability to safely participate in such activities depends not only on the device’s performance, but also on therapeutic education that promotes responsible and personalised use of the technology. In line with the basic principles of precision care, it is crucial to tailor the use of CGM, including calibration strategies, to each person’s specific context and needs, as highlighted by this study.

The data described in a previous publication highlighted poor accuracy of the Dexcom G7 sensor when used during repeated diving sessions in people with T1DM. A potential determinant of this suboptimal performance was identified as the lack of calibration. In that study, which was a head-to-head comparison with the FreeStyle Libre 3 sensor, both devices were factory-calibrated and additional DG7 calibration was not introduced to avoid introducing bias in favour of that model. In the present study, focusing solely on the DG7, we observed a marked improvement in accuracy when a calibration was performed (arbitrarily carried out, in the absence of specific instructions from the manufacturer, after the initialisation phase and subsequently once daily in the morning, during the stability phase). The overall MARD was significantly improved compared with the MARD reported for the same uncalibrated sensor model (MARD 13.7% for calibrated sensors vs MARD 31% for uncalibrated ones¹⁰).

In the previous study, where patients could choose the application site for the DG7, significantly better accuracy was observed for sensors applied to the abdomen. Among 13 patients, 4 placed the sensor on the abdomen, yielding an MARD of 22.4% and an MeARD of 15.7%, while 9 placed it on the arm, with an MARD of 41.6% and an MeARD of 24.1% (p<0.0001). In the present study, where each patient simultaneously applied one sensor to the abdomen and another to the arm (in a kind of head-to-head set-up) and both were calibrated, the difference between the two sites was reversed in favour of arm placement (DG7-ar MARD 10.9% vs DG7-ab MARD 15.9%, p<0.0001; DG7-ar MeARD 8.3% vs DG7-ab MeARD 10.7%, p=0.0003). This difference in favour of arm placement was even more pronounced and significant within the normoglycaemic range (70–180 mg/dL; 3.9–10 mmol/L), while it appeared to be markedly reduced in the days following the end of the diving cycle.

Out of 68 dives, 18 were performed using air, while the remaining 50 were conducted using nitrox (with an oxygen content ranging from 23% to 33%, average 29±3%). Remarkably, the data analysis based on the type of breathing mixture (air vs nitrox) highlighted that using nitrox leads to a potentially clinically significant improvement in the accuracy of sensors applied to the abdomen (MARD DG7-ab air vs nitrox=17.5% vs 13.8%). In comparison, it had a modest impact on the accuracy of sensors applied to the arm (MARD DG7-ar air vs nitrox=8.7% vs 10%). However, from a statistical perspective, those differences based on the breathed gas were not significant (p=0.5125 for DG7-ar and p=0.3560 for DG7-ab).

Evaluation using the SEG analysis revealed that in both cases, the system met the required conditions (>97% in zones A+B). However, despite this achievement and the significant improvement in MARD/MeARD due to regular calibration, neither case met the FDA criteria for non-adjunctive use of the system.

Trend analysis highlighted distinct patterns in glycaemic variation detected by SMBG and the two sensors at different intervals within the protocol (−60 vs −10 min and −10 min vs post-dive), potentially reflecting differences in sensitivity or lag times among the devices. Notably, the absolute difference between the trends identified by capillary measurements and sensor data was <10 mg/dL (0.56 mmol/L), which is clinically insignificant and unlikely to lead to inappropriate therapeutic decisions. Finally, the arrow trends provided by the sensors were consistent with capillary data in over 70% of cases for both sensor application sites, confirming the potential utility for identifying glycaemic changes and, particularly, for preventing hypoglycaemic episodes.

Specifically concerning hypoglycaemia and the sensor’s ability to detect it before and after dives, it was unfortunately not possible to calculate and directly compare the two diagnostic ORs. This limitation prevented an assessment of the sensor’s performance independent of the low prevalence of hypoglycaemia.

Based on the above findings, the new data further clarifies certain aspects of the DG7 sensor’s use in real-life scenarios involving repeated diving in individuals with type 1 diabetes.

First, the data confirm that sensor use is feasible, with no issues of skin irritation or early detachment.

Second, they also confirm that although calibration significantly and markedly improves the accuracy of the DG7 in this specific setting, the FDA standards for non-adjunctive use are not met, underlining once again the essential role of capillary tests in preparation for diving.

Third and lastly, the data from the days following a diving cycle are insufficient to determine whether accuracy improves sufficiently to allow for safe non-adjunctive use of the sensors. The MARD and MeARD values during this phase were clinically inconsistent; however, the small sample size renders these analyses inconclusive. Therefore, we again recommend extreme caution when relying on sensor data for therapeutic decisions or in the context of hybrid closed-loop therapy, even in the days following a diving cycle.

The data presented in this paper complement the previous head-to-head study comparing the Dexcom G7 and FreeStyle Libre 3 systems in the context of repetitive dives in individuals with type 1 diabetes. Among the study’s strengths, we highlight the simultaneous, head-to-head use of the two sensor application sites. Additionally, the study setting accurately represents real-world conditions under which scuba diving typically occurs, making the results practical and informative for clinicians and individuals with diabetes who engage in this activity. Lastly, the decision to extend the use of the sensors and acquire capillary glucose readings beyond the end of the dive cycle allowed for a preliminary evaluation during this phase, aiming to identify potential changes in sensor performance.

The sample size of subjects is undoubtedly limited. On the other hand, the overall glucose data collected is substantial and sufficient to obtain meaningful information. However, in the context of the subanalyses conducted, the sample size decreases, representing a limitation of the present study. In particular, this aspect unfortunately led to inconsistent and/or non-significant results regarding the subanalyses of gender, dive versus no-dive days and air versus nitrox. Another limitation is the arbitrary choice of calibration frequency, due to the lack of specific guidance from the manufacturer. This raises the question of whether different calibration protocol might have produced similar or different results, and whether a minimum number of calibrations could further enhance sensor performance. The last but important limitation of this study is the lack of a cross-over design. Such a design is needed to clearly show if calibration improves sensor performance. Alternatively, two sensors could be applied per site, with only one calibrated. However, five out of eight subjects also took part in a previous study where DG7 was uncalibrated. Their clinical conditions, diabetes control and therapy remained unchanged. Therefore, our conclusions are based on comparing accuracy results from this study with those from the previous one.

The reversal of optimal sensor placement (arm outperforming abdomen in this study vs the opposite in prior) may reflect differences in calibration, hydrostatic pressure or wetsuit adhesion and new mechanistic studies are needed. We therefore advocate increasing the data collected, aiming to further explore the optimal sensor placement as well as the potential impact of gender and different breathing gas mixtures. While nitrox showed a non-significant trend toward improved accuracy for abdominal sensors, the impact of breathing gases on interstitial glucose dynamics warrants exploration in larger cohorts. Also, our arbitrary once-daily calibration improved accuracy but leaves open questions about optimal frequency or timing (eg, pre-dive/post-dive), worthy of further study on sensor performance under varying calibration protocols. Finally, clarifying the reliability of the sensors in the post-dive phase during the remaining days of use appears, in our opinion, to be a point of significant clinical importance that should be explored with further data collection.

Conclusions

This study demonstrates that calibrating Dexcom G7 sensors significantly improves their accuracy compared with non-calibrated sensors of the same model, in a repetitive diving context for patients with type 1 diabetes. However, the accuracy does not meet the standard for non-adjunctive use, and capillary checks remain a fundamental dive planning tool, as internationally recommended in specific protocols for the safe practice of diving in people with T1DM. The sensor placement site, breathing mixtures and calibration methods may represent both potential sources of inconsistency and targets for refinement, highlighting key areas for targeted investigation and future research development. Our findings contribute to the growing field of personalised medicine by highlighting the need to integrate glucose sensor use into individualised management strategies for people with T1DM who engage in high-risk physical activities, such as scuba diving. Promoting the conscious and contextualised use of technology, as part of a broader educational and therapeutic framework, may help optimise safety and autonomy, aligning diabetes care with the real-life needs and goals of each patient and empowering divers with T1DM to balance convenience and safety.

Data availability statement

Data are available upon reasonable request.