A Comparative Pulse Accuracy Study of Two Commercially Available Patch Insulin Infusion Pumps

Background: Patch pumps are a relatively new method of Insulin delivery. This study explores the accuracy of patch-pumps by reporting on comparative pulse-accuracy study of two patch pumps. Methods: The accuracy of two patch pumps (Cellnovo, [Cellnovo Ltd., Swansea, UK] and OmniPod® [Ypsomed Ltd, Escrick, UK]) was evaluated micro-gravimetrically. Pulse accuracy was analysed by comparing single and time-averaged pulses for each device. Results: Single-pulses outside accuracy thresholds ±5%, ±10%, ±15%, ±20%, ±25% and ±30% were: Cellnovo; 79.6%, 55.6%, 35.0%, 19.9%, 9.7% and 4.3%; OmniPod; 86.2%, 71.6%, 57.4%, 45.5%, 35.2% and 25.4%. For 10, 20 and 40 pulse-windows mean values outside ±15% accuracy level were: Cellnovo; 7.3%, 1.5% and 0.4%, OmniPod; 37.6%, 31.8% and 25.9. Conclusions: This study showed that not all patch pumps are the same. The pumping mechanisms employed in these pumps play a significant role in the accuracy and precision of such devices.

Insulin pumps provide a convenient way of delivering a continuous subcutaneous insulin infusion (CSII). Devices are highly flexible to individual patient needs, allowing personalised 24-hour basal infusions and on-demand bolus delivery for acute periods of elevated blood glucose (for example, carbohydrate consumption during meals). Insulin pumps offer patients certain advantages over multiple daily injection (MDI) regimens including fewer injections, more flexibility and the ability to better fine-tune regimens to their personal lifestyle/needs. 1 Clinically, they have been shown in randomised, controlled trials to provide improved glycaemic control (lower glycated haemoglobin [HbA 1c ]) 2,3 reduce the frequency of hypoglycaemic episodes, 3,4 and enhance quality of life versus specific MDI regimens in type 1 diabetes mellitus. 2 For the paediatric population, quality of life gains, extend beyond pump users to their families and carers. 5 Several varieties of insulin pump are commercially available, with the two main products being durable pumps and patch pumps. Durable pumps are the most common and include an infusion set that connects the subcutaneous cannula to the pump device via an infusion line (~30-100 cm). Examples include the Animas Vibe ® and the Animas Ping ® (Animas, West Chester, Pennsylvania, US), the Accu-Chek ® Combo (Roche, Basel, Switzerland), the MiniMed Paradigm ® Veo™ (Medtronic, Dublin, Republic of Ireland) and the DANA Diabecare R ® (Advanced Therapeutics, Sooil, Seoul, Korea).
Unlike durable pumps, patch pumps are free of infusion sets as the cannula and delivery system are built into the device. They are worn directly on the body and controlled by a wireless device making them more discrete than the traditional durable pumps. Patch pumps aim to increase patient compliance by providing freedom from long-tubing, increased flexibility, easier technical operation and a smaller, lightweight device capable of being manipulated discretely. 6 Examples include the OmniPod ® (Insulet Corporation, Billerica, Massachusetts, US) and the Cellnovo system (Cellnovo Ltd., Swansea, UK).
Whilst patch pumps offer clear aesthetic advantages, a recent study found the dosing accuracy of the OmniPod patch pump to be unfavourable when compared to several durable pumps. 7  To date no groups have sought to compare dosing accuracy between commercially available patch pumps. We report a comparative pulse accuracy study comparing OmniPod and Cellnovo patch pump devices.
The Cellnovo system utilises a post-reservoir, wax-powered micro-pump and micro-fluidics that dispenses insulin on demand in a pulsatile fashion.
It has a closed-loop feedback mechanism that continually interrogates a reservoir position sensor residing within the durable element of the pump. It has the ability to make drop-by-drop alterations to the delivered volume to ensure continuous, accurate delivery. By contrast, the OmniPod system employs a more traditional disposable syringe pump mechanism to actuate movement of its reservoir plunger, relying on a shape-memory alloy to drive its motor, utilising a paired tick-tock action.
The Cellnovo system is approved for use within the EU through its CE Mark, the OmniPod system is available for use within the US as well as the EU and other countries.
The Methodology employed in this study was a slightly modified version of that described by Jahn et al. 7 Specific adjustments were made to the setup to account for comments made by Zisser 12 regarding the potential for mechanical oscillations to create measurement spikes when housing the patch pumps within the measurement equipment. Pumps were compared by evaluating single and average-pulse accuracy over clinically relevant periods of pump use. It is important to note that this was not intended to be a clinical study, merely one that evaluates relative technical performance of two patch pump systems.

Materials and methods
The accuracy of the OmniPod and Cellnovo patch pump devices was investigated using a modified version of the time-stamped microgravimetric system reported by Jahn et al. 7 (Figure 1)  To ensure that the OmniPod was connected appropriately and running properly, discrete bolus pulses were delivered to prime the infusion line and to ensure, through careful observation, that the system was leak and obstruction free. A 60 cm length was chosen to ensure that the tubing was of a reasonable kinkfree length from the patch pumps to the weighing system. The Cellnovo device came fitted with an outlet valve connector that was used to connect the device to the infusion line. Within the balance 'cabinet' the infusion line was fitted with a needle (length, 3.5 cm) the end of which was positioned through the paraffin oil layer to project (~2 mm) into the underlying water. The needle was projected into the underlying water to reduce any evaporation effects and to ensure that each drop was fully assimilated into the underlying water in a way that the increased mass could be measured. To reduce the effects of siphoning, the patch pump devices were positioned on stages so that the device output valve/cannula was level with the tip of the needle in the collecting vial.
As in Jahn et al., 7 de-gassed deionised water was used as a surrogate neutral infusion fluid for fast-acting insulins giving comparable fluid properties. This complies with the international standard set out in EN 60601-2-24:2012 that with regard to the testing of essential performance of infusion pumps and controllers, requires the use of a liquid which can be expected to give similar test results to the liquid intended for use. 9 Both devices were loaded with deionised water primed and programmed as specified by the manufacturer's instructions. As both pumps deliver fixed 0.5 µl volumes, the rate of insulin delivery was controlled by varying the number of pulses per hour. 10 For both pumps and for all runs, the initial pulse rate was set to 0 µl/hour for 1 hour in order to gauge system stability and measure weight-loss due to evaporation.
All pumps were then run for 200 pulses per run. Evaporation and system stability was again, evaluated at a pulse rate of 0 µl/hour for one hour immediately post the 200 pulse run period ( Table 1). The total run time was therefore 22 hours with weights recorded every 90 seconds (two measurements for each delivery point) to ensure delivery was measured correctly at each point.
A total of 30 runs were completed for each device (n=30, two repetitions for each of 15 different pumps). Each pump was limited to two repeats due to the 72-hour expiry of the pumps after priming. Comparisons between pulse data of the devices were made for the 20-hour test basal rate period.

Data analysis
Data for each individual 0.5 µl pulse were isolated, and recorded. Pulse volume was derived directly from pulse weight (DW, simply the weight difference recorded by the balance between discrete pulse) using equation

Averaged pulse accuracy
Although investigating single-pulse accuracy is a valid metric to assess pump performance, averaged-pulse accuracy over sustained periods of pump delivery may be a more clinically relevant assessment since patch pumps are used continuously and the time to reach steady-state will vary. Averaged pulse accuracy was analysed by averaging discrete pulse errors over predetermined observation windows. For example, at a dosing rate of 1 unit per hour, 10, 20 and 40 consecutive pulse errors were averaged to calculate the averaged-pulse accuracy over a 0.5 units, 1 unit and 2 units respectively.

Typical patch pump performance
To further compare the two pumps, and specifically to gain insight into the underlying pumping mechanisms, a 'typical performer' was selected from the 15 individual devices tested for both OmniPod and Cellnovo. The typical pump was selected as the pump that had the median standard deviation in discrete pulse error over the 20-hour test basal rate period. 7

Statistical analysis
Unpaired t-tests were used for all comparisons between the devices.
The standard deviation of discrete dose percentage errors was calculated for each of the 30 experimental runs for both devices.

Single-pulse accuracy
Investigating the accuracy of discrete pulses is one way of assessing the performance of insulin infusion pumps. Figure 2 shows the percentage error in single-pulse volume for each pulse delivered over the 30 experimental runs for the Cellnovo (Figure 2A) and OmniPod ( Figure 2B) devices. Single-pulse accuracy ranged from -120.0% to 158.5% for the OmniPod pump and -51.6% to 61.8% for the Cellnovo pump. and 25.4%) ( Figure 3, Table 2). There is a significant difference between Cellnovo and OmniPod value for all of these thresholds (p<0.0001).

Averaged pulse accuracy
An alternative and perhaps more clinically relevant way to assess patch pump performance is to investigate averaged pulse accuracy over

Typical pump performance
The performance of a typical pump, selected as the pump that exhibited the median standard deviation in discrete pulse error, was investigated    Figure 5A and D), 10-pulse averaged accuracy ( Figure 5B and E) and 20-pulse averaged accuracy ( Figure 5C and F) of the device over the entire 20-hour test basal rate period. The Cellnovo pump exhibited markedly less variability in pulse accuracy ( Figure 5A-C) than the OmniPod pump ( Figure 5D-F). When dosing accuracy was averaged over a 2-hour time period the OmniPod profile still exhibited a highly variable dosing profile ( Figure 5F).  Figure 6). For both pumps the stabilisation period prior to the last 100 delivery assessment was 10 hours.

Discussion
Patch insulin infusion pumps are a relatively recent innovation aiming to increase CSII compliance for insulin-dependent diabetes patients. Whilst offering advantages in the form of discreteness, ease of use and overall patient satisfaction, a recent study found the dosing accuracy of a patch pump to be unfavourable compared to the traditional durable pumps; 7 it was significantly less accurate in terms of both single and averagedpulse measurements.
The current technical evaluation has demonstrated that the Cellnovo device displayed significantly better single-pulse accuracy than the OmniPod pump when assessed over predetermined accuracy thresholds (±5-30%), and was significantly more accurate when assessed over longer, more clinically relevant observation doses (0.5-2 units).
The markedly different performance may be explained by the differing pumping mechanisms integral to these devices. Any shortcomings found in the delivery performance of the OmniPod system can be attributed to the disposable nature of the device. The results reflect those of Borot

Limitations
There have been discussions in the literature on the optimal manner in which insulin pump delivery should be measured for precision and accuracy. 7,11,12,13 Jahn et al. 7 discuss a methodology that compares a reduced number of durable systems (n=6) in comparison to the patch pump (n=15) and this could be seen as an inconsistency, leading to greater variability being witnessed in the larger sample number.
However, in normal use, a patient would be wearing a single durable pump for four years and would change their patch pump every three days; this increases the need to understand the actions of disposable pumps both throughout their life and understanding the pump to pump (or pod to pod variability). It is believed that this information would have led to the increased n in the patch pump arm of the study, and also made fairer by the use of multiple durable pumps.
The Cellnovo system in this test consists of a durable element and a 3-day disposable element; it was however, tested by the same means as the disposable system.    However, this comes at the expense of introducing experimental design difference between the groups. Furthermore, we found that the short length of the cannula (6.5 mm for OmniPod, 5.0 mm for Cellnovo system) made it difficult to connect directly to the collecting vial whilst achieving a steady system setup. In this study both pumps were connected to the collecting vial by an infusion line ensuring both pumps were compared using identical experimental setups. Although this is a potential limitation of the study, as in practice patch pumps are free from infusion lines, it ensured a consistent and fair comparison between the pumps.
As mentioned previously there are alternative methods for measuring pulsed dose accuracy. 13 These methods are not easily employable for the measurement of large numbers of systems and care must be taken when measuring spheres with such small radii due to evaporation rates in varying environmental humilities.

Conclusions
Whilst a previous study showed that a patch pump performed poorly when compared to a number of durable pumps, this study showed that not all patch pumps are the same. The pumping mechanisms employed in these pumps play a significant role in the accuracy and precision of such devices, which in turn may impact on clinical outcome. q