Accuracy of a New Patch Pump Based on a Microelectromechanical System (MEMS) Compared to Other Commercially Available Insulin Pumps

Abstract

The JewelPUMP™ (JP) is a new patch pump based on a microelectromechanical system that operates without any plunger. The study aimed to evaluate the infusion accuracy of the JP in vitro and in vivo. For the in vitro studies, commercially available pumps meeting the ISO standard were compared to the JP: the MiniMed® Paradigm® 712 (MP), Accu-Chek® Combo (AC), OmniPod® (OP), Animas® Vibe™ (AN). Pump accuracy was measured over 24 hours using a continuous microweighing method, at 0.1 and 1 IU/h basal rates. The occlusion alarm threshold was measured after a catheter occlusion. The JP, filled with physiological serum, was then tested in 13 patients with type 1 diabetes simultaneously with their own pump for 2 days. The weight difference was used to calculate the infused insulin volume. The JP showed reduced absolute median error rate in vitro over a 15-minute observation window compared to other pumps (1 IU/h): ±1.02% (JP) vs ±1.60% (AN), ±1.66% (AC), ±2.22% (MP), and ±4.63% (OP), P < .0001. But there was no difference over 24 hours. At 0.5 IU/h, the JP was able to detect an occlusion earlier than other pumps: 21 (19; 25) minutes vs 90 (85; 95), 58 (42; 74), and 143 (132; 218) minutes (AN, AC, MP), P < .05 vs AN and MP. In patients, the 24-hour flow error was not significantly different between the JP and usual pumps (–2.2 ± 5.6% vs –0.37 ± 4.0%, P = .25). The JP was found to be easier to wear than conventional pumps. The JP is more precise over a short time period, more sensitive to catheter occlusion, well accepted by patients, and consequently, of potential interest for a closed-loop insulin delivery system.

Use of continuous subcutaneous insulin infusion (CSII) has been shown to reduce HbA1c levels in patients with type 1 diabetes (T1D) while reducing the incidence of hypoglycemic events, compared to multiple daily injections.^1-3 Even though the latest meta-analyses comparing CSII with rapid insulin analogs showed only slight benefit in terms of HbA1c and hypoglycemic events,^4-7 an improvement in quality of life was observed in most studies.⁸ Over the past decade, insulin pumps have received broader acceptance thanks to the development of functional insulin therapy (live insulin dose modification, bolus assistant), data recording, analysis and transfer capacity, and their ability to be coupled with continuous glucose monitoring. Consequently, CSII is currently considered the most efficient therapeutic option for patients with T1D. However, some issues remain. Traditional insulin pumps are connected to the patient via flexible plastic tubing (from 30-110 cm) that may kink, catch or become occluded. Air bubbles can arise, and a siphon effect may be observed when the pump is moved in relation to the infusion site.⁹ Furthermore, traditional insulin pumps must be worn on a belt, or inside a specific pocket, potentially constituting a psychosocial barrier, especially with regard to the private life of patients.^10,11 In this context, there is an increasing interest in patch pumps, which readily adhere to the body without any external tubing.^12,13

The accuracy of external pumps is of great concern, especially in children or adults with low infusion rates, where they can limit glycemic variability and unexplained hypo- or hyperglycemic events. All commercially available insulin pumps comply with international infusion pump standard IEC 60601-2-24, and insulin is injected through the tubing via a plunger system. While such infusion systems are accurate over a 24-hour period, they are far less accurate over shorter periods,¹⁴ probably due to the elasticity of the plunger, which progresses by fits and starts (known as the “stick-slip” effect). In a proposed closed-loop delivery system, the insulin pump must be accurate and capable of delivering the exact dose of insulin calculated and programmed, and it must also be highly reactive for rapid detection of infusion failures such as a catheter obstruction.

The JewelPUMP™ (JP) is a new patch pump developed by Debiotech (Lausanne, Switzerland), whose infusion system is based on a microelectromechanical system (MEMS). This system relies on direct insulin stroke delivery via the pumping mechanism, while conventional pumps are based on a syringe driver, which may be affected by the compliance of the piston, situated between the motor and the insulin. MEMS pumps provide flow-error rate below 5%, even for a single stroke, such as for low volumes and for short time periods. The presence of a highly sensitive strain gauge detector in the MEMS pump enables rapid detection of catheter obstruction thanks to very sensitive pressure microsensors. Each shot infuses 200 nl of insulin (0.02 insulin units) at a frequency of between 0 and 3 Hz. For example, at a basal rate of 1 IU/h, the JP delivers 0.02 IU every 1.2 minutes, whereas other pumps deliver 0.05 IU/3 min.

The aim of this study was to evaluate the accuracy of insulin delivery by the JP in vitro and in vivo compared with commercially available durable or patch pumps.

Materials and Methods

Laboratory Accuracy Study

A continuous microweighing method was used to measure water delivery (ISO class III water, density = 0.998 g/ml) by different pumps, as previously described.¹⁴ Briefly, a Sartorius MC5 microbalance (load limit = 5.1 g, resolution = 1 µg) was placed on a vibration-isolating table and enclosed under transparent cover. A glass vial of water was placed on the weighing pan inside the weighing chamber of the microbalance, and the distal tip of a glass capillary tube was immersed in water. A 1-mm film of paraffin oil was placed on top of the water to limit evaporation. The capillary tube was connected to the cannula of the pump through a draft shield in the chamber lid. Unlike Jahn et al, we used the same settings to test the patch pumps (see Supplementary Material S1). The patch pumps were placed outside the chamber, over the lid. The pump being tested was loaded, programmed (0.1 or 1 IU/h), and primed in accordance with the manufacturer’s instructions, using the recommended catheter (Quick-set® 23” for the MiniMed® Paradigm® 712 [MP], Ultra-Flex® 24” for the Accu-Chek® Combo [AC], and Inset® 24” for the Animas® Vibe™ [AN] or the built-in cannula for the OmniPod® [OP]). Unlike the OP, the JP is connected to the commercially available Accu-Chek FlexLink cannula (used without the tubing) via the luer-lock system, and can be disconnected at will. Durable pumps were placed on a platform near the microbalance at the level of the paraffin layer. Since the patch pumps did not have external tubing, they were placed on top of the internal weighing chamber lid.

Each test lasted 24 hours and data were collected using Infuscale software 7.0 (TüvSüd Product Services, Wallisellen, Switzerland) without any stabilization period.

Flow rates were calculated every 15, 30, and 60 minutes over 24 hours using the following formula,

$$MeasuredFlowrate\left(ml/min\right)=\frac{\text{Δ}W(mg)}{0.998(gH20/ml)}\times \frac{1}{1000(mg)}\times \frac{1}{OW(min)}$$

where ΔW is the weight difference for water over the observation window (OW).

The measured flow rate was then expressed in IU/h, considering that 100 IU = 1 mL.

The percentage error in flow rate over the OW was then calculated as follows:

$$\%errorinflow=\frac{Measuredflowrate(IU/h)-1(IU/h)}{1(IU/h)}\times 100$$

The absolute value of flow error was used to compare pump accuracy.

Different types of pump were tested at basal rates of 0.1 and 1 IU/h: JP (n = 3, provided by Debiotech), MP (Medtronic France SAS, Boulogne-Billancourt, France, n = 3), AC (Roche Diagnostic, Meylan, France, n = 3), AN (Novalab France, Malakoff, France, n = 3), and OP (Insulet, Bedford, MA, USA, n = 3).

As the JP delivers 0.02 IU per dose, the 15-minute OW was reduced to 14.4 minutes for this pump (12 strokes), and the results were compared the 15-minute results obtained with the other pumps.

Occlusion Study

To determine pump reactivity to a catheter occlusion, a real full catheter occlusion was created by clamping the cannula using a surgical clamp. A bolus was then programmed and the number of noninfused insulin units was recorded when the alarm sounded. The alarm threshold was then tested at basal rates of 0.5 IU/h and 2 IU/h by measuring the time between occlusion and alarm. The experiment was repeated 6 times. After the alarm sounded, we removed the clamp and determined the amount of insulin still flowing from the cannula, that is, the “bolus released after occlusion.”

Clinical Study

Patients were recruited at 3 university hospitals in France (Corbeil-Essonne, Grenoble, and Besançon). Patients with T1D aged over 18 years and on insulin pump therapy for at least 6 months with HbA1c < 9% were included in this prospective interventional study.

On day 0, the catheter for the JP (Accu-Chek FlexLink) was inserted by a trained operator whereas patients inserted their normal pump catheters themselves. The JP, filled with physiological serum, and the patient’s normal pump, filled with insulin, were weighed on a precision balance (Sartorius ED325S-320 g-1 mg, Sartorius France SAS, Aubagne, France) immediately before connection to the catheter. A 2-unit bolus was injected with the JP to check the proper functioning of the system. Before returning home, patients were shown how to change the basal rates and to enter boluses with the JP system, at the same time and in the same way as for their own pump. Outpatient visits were scheduled at days 1 and 2. The 2 pumps were disconnected and weighed. The usual pumps were reconnected, and a new JP was weighed and connected to the catheter. Because the JP battery absorbs oxygen, it gained weight over time (on average, 21 mg over the duration of a session), potentially leading to underestimation of the real insulin infusion, which ranges from 0 to 3 insulin units per day. The battery was therefore removed and weighed in isolation before and after each 24-hour session.

The daily injected volume with the JP was determined using the following formula:

Measured injected volume (MIV) (mL) = (weight difference between day 0 and day +1 (g) + battery weight gain (g))/physiological serum density (1.009 g/mL).

The following formula was used for the patients’ own pumps:

MIV (mL) = weight difference between day 0 and day +1/insulin density (1.005 g/mL)

MIV was converted to measured insulin dose (MID) on the basis of 1 mL = 100 insulin units. The MID of each pump was compared to the recorded insulin dose (RID) downloaded from each pump and recorded manually by the patient. The 24-hour error ratio, defined as (MID – RID)/RID, was compared between the 2 pumps.

At inclusion, patients were asked to complete a visual analog scale (VAS) ranging from 0 to 10 indicating whether treatment with their usual pump was extremely unpleasant (0) or extremely pleasant (10), and whether their usual pump was extremely difficult (0) or easy (10) to use and to wear. After completing the study, the patients were asked to complete the same VAS for the JP as well as a satisfaction questionnaire (see Supplementary Material S2).

Statistical Analyses

Statistical analyses were performed using SAS for Windows. For experimental studies, since Gaussian distribution could not be assumed (D’Agostino and Pearson normality test), the data were expressed in terms of medians, interquartiles, and range. The median absolute errors were compared using the Kruskal–Wallis test with posttest analyses. Chi-square tests were used to compare percentage of values outside the thresholds.

Quantitative variables in the clinical data are expressed as mean ± SD (range) and compared using a paired Student t test. Correlation between RID and MID was studied using linear regression analysis and compared between each pair of pumps (by excluding the 2 IU test bolus delivered with the JP). Statistical difference was assumed for P < .05.

Results

Laboratory Accuracy Study

Accuracy flow rate results at 1 IU/h are summarized in Table 1. The 5 models of pump tested were accurate over 24 hours without any significant difference. Over an OW of 15 minutes, however, the JP showed a significantly lower absolute median error rate than the other pumps: ±1.02% (JP) vs ±1.60% (AN), ±1.66% (AC), ±2.22% (MP), and ±4.63% (OP), P < .0001 (Figure 1a). Over this 15-minute OW, the JP exhibited significantly fewer values outside the 5% threshold than the other pumps (Table 1). Over the 30-minute and 60-minute OWs, the absolute median error value for the JP was lower than for the MP and OP, but not for the AN and AC. Figure 1b shows the trumpet curves of all devices over 8 hours.

Table 1.

Description of Percentage Error in Flow Rate (Median, Range, and Interquartiles) and Percentage of Doses Outside the Threshold (±5%, ±10%, ±15%) Over Different Observation Windows (15 Minutes, 30 Minutes, 60 Minutes, and 24 Hours), at 1 IU/h.

Observation window			JewelPUMP (n = 3)	Animas Vibe (n = 3)	Accu-Chek Combo (n = 3)	MiniMed Paradigm 712 (n = 3)	OmniPod (n = 3)	P
15 minutes		n	226	284	284	284	284
	% flow error (±)	Median	−0.28	0.47	0.78	−1.20	0.14
		Range	−5.71; 26.06	−10.96; 12.60	−32.80; 13.16	−55.60; 27.79	−38.41; 34.52
		Interquartiles	−1.41; 0.85	−1.21; 1.86	−0.99; 2.19	−3.26; 1.16	−4.38; 4.92
	% absolute flow error	Median	1.02	1.60*	1.66*	2.22*	4.63*	<.0001
		Interquartiles	0.57; 1.69	0.84; 2.83	0.88; 2.88	1.18; 3.79	1.99; 8.75
	% values outside threshold	±5%	2.6	9.1*	9.5*	18.8*	48.3*	<.0001
		±10%	1.5	1.1	2.1	4.9*	22.2*	<.0001
		±15%	1.5	0	1.1	3.5	10.1*	<.0001
30 minutes		n	132	144	144	144	144
	% flow error (±)	Median	−0.43	0.29	0.46	−0.98	0.11
		Range	−3.91; 20.66	−7.23; 7.01	−18.50; 7.70	−23.22; 12.22	−32.04; 19.31
		Interquartiles	−1.24; 0.51	−0.49; 1.26	−0.26; 1.75	−3.32; 0.67	−4.02; 4.15
	% absolute flow error	Median	0.91	0.98	1.18	2.30*	4.09*	<.0001
		Interquartiles	0.46; 1.58	0.37; 1.77	0.36; 1.99	0.80; 3.75	1.35; 6.86
	% values outside threshold	±5%	3.0	2.1†	5.7	13.2*	41.0*	<.0001
		±10%	1.5	0†	0.7	3.5	9.7*	.004
		±15%	0.8	0	0.7	1.4	4.2	ns
60 minutes		n	66	72	72	72	72
	% flow error (±)	Median	−0.47	0.60	0.71	−1.33	−0.45
		Range	−2.86; 10.43	−4.51; 4.07	−21.14; 3.06	−9.15; 4.85	−17.47; 8.96
		Interquartiles	−1.36; 0.55	−0.42; 1.09	−0.17; 1.28	−3.57; 0.63	−2.27; 3.18
	% absolute flow error	Median	0.87	0.80	0.95	1.98*	2.67*	<.0001
		Interquartiles	0.49; 1.58	0.55; 1.36	0.56; 1.63	1.12; 3.62	0.92; 4.25
	% values outside threshold	±5%	3.0	0	4.3	11.1*	22.2*	.003
		±10%	1.5	0	1.4	0	5.6	ns
		±15%	0	0	1.4	0	5.6	ns
24 hours	% flow error (±)	n	3	3	3	3	3
		Median	0.49	−0.06	0.37	−1.44	−0.64
		Range	−1.71; 0.64	−0.56; 0.34	−1.04; 0.85	−2.24; –0.90	−1.31; 0.31

^* P < .05 compared to JewelPUMP.

Figure 1.

Flow rate accuracy. Percentage of flow error with absolute values (a, c), trumpet curves (b), according to pump type and observation window, at 1 IU/h (a, b) or 0.1 IU/h (c). Data are expressed as median and interquartiles. *P < .05 compared to the JP.

At a low basal rate of 0.1 IU/h, the JP infused 1 stroke of 0.02 IU every 12 minutes whereas other pumps infused 0.05 IU every 30 minutes. Over a 12-minute OW (1 stroke), the JP showed an absolute error rate of 3% (1.5; 6) (Figure 1c). This parameter could not be assessed for the other pumps (1 stroke every 30 minutes). Over a 30-minute OW, the JP had a significantly lower error rate than the other pumps tested. The OP has not been included in this experiment.

Occlusion

After a full real catheter occlusion, the amount of uninjected insulin before the alarm sounded in bolus mode was less than 0.1 IU for the JP, 1.15 IU (1.10; 1.25) for the MP, 1.59 IU (1.59; 1.59) for the AN, 3.0 IU (2.9; 3.2) for the AC, and 4.7 IU (n = 1) for the OP (Figure 2a). At a basal rate of 0.5 IU/h, the JP alarm sounded much earlier than with the other pumps: 21 (19; 25) minutes vs 143 (132; 218) for the MP, 58 (42; 74) for the AC and 90 (85; 95) for the AN (Figure 2b). The difference was also observed at a basal rate of 2 IU/h (Figure 2b). After removal of the occlusion, the insulin bolus released with a basal rate of 0.5 IU/h was 0.04 (0.04; 0.08) for the JP, 0.25 (0.07; 0.42) for the AC, and 0.68 (0.61; 0.89) for the MP (P < .05 with the MP).

Figure 2.

Occlusion detection. Uninjected insulin dose before alarm in bolus mode (a). Time before alarm in basal mode at different rates (b).

Clinical Study

Thirteen patients were included in this pilot study comprising 21 analyzable 24-hour time periods. The patient characteristics and described in Table 2. Patient insulin requirements (based on RID) were 42.7 ± 19.9 (15.7-79.5) IU/day. The mean weights of the JP and conventional pumps at day 0 and day +1 were, respectively, 26.04 ± 0.16 g and 25.60 ± 0.30 g, and 115.3 ± 11.33 g and 114.8 ± 11.25 g.

Table 2.

Patient Characteristics (n = 13).

Age (years)	39.2 (13.4)
Sex ratio (M/F, %)	4/9 (30.8/69.2)
BMI
18.5-<25 (%)	8 (61.5)
25-<30 (%)	3 (23.1)
≥30 (%)	2 (15.4)
T1D duration (years)	17.6 ± 8.8 (4-32)
HbA1c (%)	7.5 ± 0.8 (5.9-9.0)
Diabetes complications
Retinopathy	2 (15.4)
Nephropathy	0 (0)
Neuropathy	0 (0)
Macrovascular disease	0 (0)
Pump therapy duration (years)	4.3 ± 2.6 (1-10)
Pump
Animas	3 (23.1)
Accu-Chek	4 (30.7)
MiniMed Paradigm	6 (46.2)

Quantitative values are indicated as mean ± SD (range).

Figure 3a shows the correspondence between RID and PID for each pump. The linear regression analysis showed no difference between the 2 pumps. The 24-hour flow error was not significantly different: –2.2 ± 5.6% for the JP and –0.37 ± 4.0% for usual pumps (P = .25).

Figure 3.

Clinical study. Infused insulin dose according to programmed insulin dose in the 2 systems (a). Visual analog scales before and after the study (b).

Cutaneous tolerance was good for 10 patients (77%). Two patients experienced catheter patch detachment. One patient felt pain on connecting the pump to the catheter.

The VAS results are shown in Figure 3b. The same score was observed before and after the study when patients were asked if they wanted to be treated with the JP (0 = absolutely not, 10 = absolutely, 7.9 ± 2.1 vs 8.2 ± 2.3, P = .39). At inclusion, patients who found treatment with their usual pump rather unpleasant (VAS ≤ 5, n = 3) absolutely wanted to be treated with the JP at the end of the study (VAS = 10, 10, and 9.5). In both groups, patients found that the pumps were easy to use (8.6 ± 1.7 vs 8.8 ± 1.1 for JP and the usual pumps respectively; individual data not shown). However, the JP was found to be significantly easier to wear than the patients’ usual pumps (8.1 ± 1.7 vs 5.5 ± 2.1, P = .003).

The satisfaction questionnaire showed that respectively 46.2% (n = 6) and 53.8% (n = 7) of patients were completely or mostly satisfied with the JP. Eight patients (n = 61.5%) found the JP easier and more convenient than their usual pump. The lack of tubing was considered the main advantage of the JP, making it easy to wear but also more discreet.

Discussion

This is the first experimental and clinical study to evaluate the accuracy of the JP, a new patch pump. Experimental analyses suggest that the JP is more accurate over short time periods compared with both conventional pumps and an existing patch pump. This accuracy over short times is due to the MEMS technology used for insulin delivery, which allows accurate infusion control of 0.02 insulin units per delivery step. Consequently, during the 15-minute OW at 1 IU/h, 12 direct doses were effectively delivered by the JP, while only 5 motor steps were actuated by other pumps. These results are consistent with those of Jahn et al, who compared the accuracy of conventional pumps with that of the OP device¹⁴ using a technique similar to the 1 employed in this study, but different from international standard 60601-2-24. In this standard, pump accuracy is defined by statistical analysis over a period comprising 100 deliveries following a 24-hour stabilization period. Using different techniques based on digital microscopy and imaging software analysis for a graduated micropipette or spherical bolus, Zisser at al showed that the OP was very accurate, even with a single dose of 0.5 IU.¹⁵ This study did not compare the results with data obtained using other pumps. The difference observed in our study may be explained by the use of a different technique, or more probably by the fact that Zisser et al studied the OP in bolus mode and not in basal mode, thus minimizing the well-known stick-slip effect seen with syringe-driver systems. It is also important to indicate that, in our study, we performed the analysis on the pooled data of the 3 tested pumps in each group. However, the results of each pump separately were similar (data not shown).

Nevertheless, the clinical relevance of such basal rate accuracy over short time periods remains to be evaluated. Regarding the pharmacokinetics of rapid-acting analogues after subcutaneous infusion, doubling the basal rate from 1 IU/h to 2 IU/h has an effect on glucose infusion rates in euglycemic clamp starting only after 30-60 minutes.¹⁶ Besides, in these adult male patients with normal insulin requirements (0.12 IU/kg), switching from 0.1 IU/h to 0.5 IU/h for 4 hours did not change the glucose infusion rate. However, the situation may be different in patients with very high insulin sensitivity or low insulin requirements (like children). Besides, ultrafast-acting insulin compounds are currently under development¹⁷ and may reduce the time between a change in basal rate and metabolic results, and might also show higher glucose variability in pumps with high flow-error rates over short time periods.

More significant is the time between a full occlusion and the alarm. Because of the rigidity of the system (silica chip, membrane), the JP showed a quick response to a full occlusion (after 3 or 4 strokes) compared to other pumps made of more readily deformable materials (plunger, plastic reservoir, catheter).¹⁸ Previous studies on cessation of rapid delivery of insulin analogues showed that glycemic changes began to occur 1 hour after cessation and were more consistent after 2 or 3 hours.^19-22 In the case of a 2-hour overnight insulin cessation, glucose levels at the end of the interruption period rose by 16 ± 34 mg/dL at 2 hours and by 63 ± 57 mg/dL at 4 hours.¹⁹ Another study with a 30-minute interruption reported that the rate of the rise in glucose was 1 mg/dL for each minute the insulin infusion was interrupted, but that the maximum increase was observed 3 hours afterward.²¹ In a study of prolonged insulin interruption, β hydroxybutyrate increased in parallel with blood glucose, reaching 1 mmol/L after 3 hours.²⁰ We may suppose that the JP is able to detect occlusion before glycemia rises, which may lead to a decrease in unexplained hyperglycemia potentially due to partial or complete catheter occlusion. Furthermore, the rapid occlusion detection of the JP may avoid insulin accumulation with rapid and uncontrolled release in the event of a temporary obstruction. This would be of great value, especially in subjects with a low basal rate, high insulin sensitivity or brittle type 1 diabetes, and in the context of a closed-loop system.

The clinical study showed that the in vivo 24-hour accuracy of the JP was not significantly different from that of conventional pumps, with a small flow-rate error. Clinical evaluation of pump accuracy in humans requires special considerations. The only way to measure the real infused dose is by establishing weight loss for the pump, which, in theory, corresponds to the weight of fluid infused. With regard to the JP, the fact that the battery gained weight throughout the study complicated the study design. Moreover, the weight of infused insulin was small compared to the weight of the system, especially for conventional pumps, and may have limited the precision of the weight difference measure. Further clinical investigations using pumps filled with insulin are warranted.

Like other patch pumps, the JP showed excellent cutaneous tolerability and excellent acceptance by patients. The VAS and satisfaction questionnaires showed that the tube-free aspect is of great interest to patients both for improved comfort and concealment, 2 of the issues frequently reported by patients regarding insulin pump therapy.^10,11 The results show that, even if patients declare to be comfortable with their usual pump before the study, most of them are even more satisfied by the patch pump at the end of the study. It would have been interesting to compare what they thought about their own pump before and after the study, but unfortunately, these data are not available. Compared to the OP, the JP is only partially disposable: the disposable unit is filled once and discarded entirely after use, while the controller unit (incl. the electronics) can be used for 2 years, making it more cost effective and environmental friendly than fully disposable pumps. Compared to the OP, the JP is detachable at will from the cannula patch with direct access bolus buttons and a discreet vibration and audio alarm on the patch-pump.

Several limitations to this study need to be acknowledged. Although the JP showed interesting short-term accuracy in vitro, we were not able to evaluate the infusion accuracy in vivo over short periods of time. The weight difference would have been too small compared to the measurement error, and the results not clinically relevant. This current pilot study in patients was designed to evaluate the 24-hour infusion accuracy, before pharmacological studies involving the patch pump filled with insulin in patients with type 1 diabetes.

This study shows for the first time that the JP is more accurate in vitro over short time periods than commercially available conventional or patch pumps, with the same 24-hour accuracy, and that it is more reactive to catheter occlusion. Consequently, further investigation in a closed-loop insulin delivery system is of great interest.