Bolus Calculator Safety Mandates a Need for Standards

The first bolus calculator (BC) in an insulin pump was introduced in 2002 in the Deltec Cozmo pump to simplify the mathematical calculations needed to estimate prandial and corrective insulin doses through an algorithm. BCs improve dosing accuracy and convenience for many people, especially in the complex calculations required to track bolus insulin on board (BOB or IOB), a function needed to minimize stacking of residual insulin activity from recent bolus doses.^1-5 Accuracy of BC calculations depends on personalized BC settings with the assumption that basal rates or long-acting insulin doses have previously been selected and tested for accuracy to minimize their effect on bolus accuracy. Calculation and tracking of bolus doses by a BC is needed because as one pump expert stated in testimony before the FDA it is “not realistic to expect most patients or clinicians to understand complexities and nuances of IOB calculations.”⁶

Discussions with numerous clinicians involved in diabetes care indicate a limited understanding of the algorithms that are used in insulin pump BCs, the potential sources for error in these algorithms, and the best methods available to select physiologically appropriate BC settings for their patients. Some ISO standards, IEC 82304-1 regarding Health Software and ISO EN ISO 14971:2012 regarding Risk Management, provide guidance for health software under development, but these standards do not mandate or regulate BC safety. Most BC algorithms remain unpublished, and this lack of clarity raises safety concerns. Hypoglycemia remains a significant risk with current insulin pump BCs due to inaccurate pump settings and inappropriate BOB tracking and handling. It is essential to verify safety, and ensure the consistency of dose recommendations and clinical outcomes to determine how different BCs behave in routine and outlier situations.

A BC’s basic function is to recommend how many units of insulin or grams of carb are needed once a blood glucose (BG) measurement is entered from a blood glucose meter (BGM) or in the near future from a continuous glucose monitoring (CGM) system. The goal of a BC is to increase time in target and reduce hypoglycemia. Today’s insulin pump BCs have remained virtually unchanged over the last 15 years. Even though they collect the data needed to test and verify basal and bolus doses, BCs have not integrated the pattern recognition and dose adjustment guidance needed to substantially improve glucose outcomes. A fully functional BC would recommend bolus doses that improve glucose control and insulin stacking, while also recommending the carb intake needed to prevent or treat hypoglycemia when BOB is excessive. The current BCs found in insulin pumps, BGMs and smartphone BC apps are not adequately configured to fulfill even these basic missions.

In the rapidly expanding world of BCs, only eight BCs have received FDA approval in the United States, with five in pumps, two in BGMs, and one in a phone app connected via Bluetooth to an insulin pen.⁷ The first phone app received CE approval in the European market in 2013. Many more BCs are available as apps or in BGMs around the world: A 2014 review by Huckvale et al identified 46 BC apps that were available at that time.⁸ By August 2016, the FDA was aware of 84 unique BC apps in insulin pumps, BGMs, and smartphones.⁹ The proliferation of BC apps remains largely unregulated. In the 2014 review mentioned, only one of the 46 apps fulfilled all of the study authors’ desired criteria, and fewer than a third of BC developers made available the bolus calculations that were used to generate bolus dose recommendations.⁸ Only 1 provided automatic transfer of glucose readings from a BGM, while 38 adjusted for the current glucose but only 9 tracked BOB to prevent insulin stacking. The terminology used in BCs can also be an issue. With 22 of 46 phone apps containing ambiguous terminology, even experienced physicians and CDEs may encounter difficulty when one BC says “active insulin” and another “insulin on board” or “bolus on board” for residual bolus insulin activity. In their analysis, the study authors concluded that “the majority of insulin dose calculator apps provide no protection against, and may actively contribute to, incorrect or inappropriate recommendations that put current users at risk of both catastrophic overdose and more subtle harms resulting from suboptimal glucose control.”

Similar to US consumers who are exposed to unregulated vitamins and minerals in the health food market, people with diabetes are placing their trust in largely untested and unregulated insulin dose advisors. BC users and clinicians remain woefully unaware of when or how often a BC’s advice may prove harmful. Attention in this area is essential as more and more people on MDI therapy adopt BCs in apps and BGMs. The need to standardize and systematically test BCs is obvious and long overdue.

Clinical research regarding BCs remains limited in scope. Most BC research studies have demonstrated modest glycemic results: Mild reductions have been found in bolus dose errors,^10-12 postprandial glucose levels,^12-16 glucose variability,^12,17 hypoglycemia frequency and HbA1c levels,^7-9,12,18-20 with small increases in time in target,¹⁹ BGM testing frequency,¹² and user satisfaction.^{10,11,13,18-25} To date, no independent or systematic evaluations have used virtual clinical cases under comparable head-to-head circumstances to evaluate the safety of the bolus recommendations offered by dozens of different BCs. In addition, only a handful of studies have provided information on the ease and ongoing BC usage. BC algorithms that have been documented to substantially improve glucose control and increase safety in clinical studies while also displaying ongoing use should receive acclaim.

BCs continue to use static rules to generate bolus dose recommendations and lack methods to improve glucose control or prevent hypoglycemia, so one explanation for the modest glycemic benefits in these studies may arise from the BCs themselves. A BC that takes too much time, requires too many steps, is difficult to understand, or gives inaccurate recommendations provides little benefit. Some BCs provide post-facto glucose pattern analysis only after the device is downloaded, yet no guidance is provided within the device to improve existing BC settings or make immediate dose adjustments. Common sources for BC bolus recommendation errors are mediocre algorithm logic, the use of too short DIA time settings, and the use of other physiologically inappropriate BC settings. Research is critically needed to verify BC safety. Outlier or corner cases are ideal to compare variations in bolus recommendations, detect failure points, and uncover safety issues.

Fear of hypoglycemia remains a significant impediment to improve BG levels for many patients. However, a number of developments are under way that will help to minimize hypoglycemia risk:

• Hypo-minimizer pumps enable larger bolus doses that are compensated by automated basal reductions.

• Hypo-manager software is available in some BC phone apps to suggest how many grams of carbohydrate are needed when BOB is greater than needed for the current glucose.

• Vigilant software helps to identify patients at risk and times when hypoglycemia poses a greater risk. Also, patients on conventional insulin therapy or nonadvanced pumps can use such approaches.

Could BCs provide additional features? Although not yet available, BCs can be combined with CGMs and other medical devices. Directional data from current CGMs can predict hypoglycemia up to 30 minutes ahead of time. When this glucose trend information is combined with BOB information from a pump or smart insulin pen and dosing data is automatically transferred to a cell phone or the cloud, a BC making use of all available data could recommend specific carb intakes or bolus reductions to prevent hypoglycemia.²⁶ Physical activity monitoring by an activity monitor built into pumps would also allow bolus dose reductions or free carb intake to improve glucose outcomes during and after exercise. Likewise, bolus and basal doses could be increased when activity is significantly reduced and glucose readings rise. Time in target could be increased through static algorithms combined with treat to target rules, or through more advanced and flexible approaches using artificial intelligence and other approaches. Advances in cell phone BCs or cloud connections could substantially benefit patients on MDI or nonadvanced pumps.

BCs have lagged behind other device advances and been viewed only as sources for immediate bolus dose recommendations. Even though progress has been slow, BCs collect the critical care data required to test and verify basal and bolus doses, and identify glycemic patterns. They could readily integrate problem-solving strategies to verify and guide the changes needed in BC settings to improve bolus dose recommendations. These more advanced approaches to BC insulin adjustment have unfortunately not been implemented, largely because of increased liability risks.

After a decade and a half of use, demand to establish standards for BC algorithms, terminology, and training is imperative. Rather than blocking innovation, BC standards should provide a steady foundation for innovation and reduce development errors in the rapid enhancements under way among BCs. Since clinicians and patients currently have to understand the function, usage, and handling of multiple BCs, a standard BC would also be easier to train and monitor. The implementation of BC standards could address many of the issues mentioned above and would provide a greater degree of safety. Standards would also help allay the liability issues inherent in providing dose guidance. Regulatory bodies may want to offer coverage for BCs that improve time in target while also demonstrating reasonable degrees of safety. Provision of liability insurance coverage could be provided through a set-aside from these payments. The health and economic benefits that ensue from reductions in complication rates would be expected to easily cover this financially smaller liability risk. Taking reasonable risk for significant gain seems reasonable given the current high costs in diabetes care.

Standardized training and clinical guidance are required for the basic and advanced skills required to operate a BC, select appropriate settings, check and change settings, adjust bolus recommendations when outside factors exist, operate adjunct features such as temporary basal rates and combo boluses in pumps, and download data. BC utilization rates in insulin pumps vary from 23% to 93% in different studies.^27,28 BC utilization rises when the user interface is easy to maneuver, when glucose readings are automatically entered into the BC, and when more training is provided. As a result of complexity, some BC wearers only utilize a BC for bolus recommendations to treat hyperglycemia and miss out on the protection BCs provide against insulin stacking. The absence of data sharing between diabetes devices can introduce clinical treatment errors when only one device, such as a BGM that is used primarily to verify hyperglycemic readings on a CGM is the only device downloaded at a physician’s office. Other users resort to frequent modifications to recommended bolus doses because these recommendations are derived from incorrect settings or generate aberrant glucose results.

Clear explanations are often lacking for the function of each BC setting and how to adjust these settings when erroneous dose recommendations or erratic glucose readings occur. It is surprising that this guidance is missing since many BCs have sufficient historical data regarding glucose readings, carb counts, and bolus doses to guide users to adjust both basal rates (or long-acting insulin doses) and bolus doses for better glucose outcomes. To promote better training, a financial incentive program might be offered for BCs that achieve specific HbA1c reductions and increase time in target while decreasing hypoglycemia. With the probability that significant financial gains would ensue from improved control and reductions in complication rates and hospitalizations, this approach merits consideration.

Unlike complex computer systems where computers, printers, operating systems, software, and internet connections are largely interchangeable, devices from one diabetes manufacturer rarely work with those of another manufacturer. BGMs, CGMs, insulin pumps, insulin pens, and other medical products typically do not communicate with each other despite the availability of common communication interfaces through Bluetooth Low Energy and health device communication standards like IEEE 11073. Even though software hacking remains a concern, medical products provide significant benefits to users and clinicians when manufacturers provide data access through protocols like OAuth2, JSON, and RESTful that allow devices to readily connect and safely share health data.

We believe that a systematic approach would increase patient safety and convenience, improve data access and device interoperability, and perhaps increase competition with fewer software patent infringements. Software patents related to BCs can also significantly block their development.²⁹

These are needed:

• Standardization of a core BC algorithm designed for safety, including how BOB is calculated and handled in dose recommendations.

• Systemization of how BC settings are chosen and verified.

• A systematic approach for sharing data between medical devices and software.

• Training for clinicians and patients to standardize BC use.

• A systematic evaluation of how each BC fulfills these standards.

• Insurance incentives for BCs that demonstrate glycemic improvements and that communicate with at least 3 other manufacturer’s devices.

With the upsurge in BC capability and usage now well under way, it would be helpful for a nonprofit organization like the Diabetes Technology Society to organize a Bolus Calculator Panel to discuss these important topics and make recommendations.