Recommendations for Insulin Dose Calculator Risk Management

Abstract

Several studies have shown the usefulness of an automated insulin dose bolus advisor (BA) in achieving improved glycemic control for insulin-using diabetes patients. Although regulatory agencies have approved several BAs over the past decades, these devices are not standardized in their approach to dosage calculation and include many features that may introduce risk to patients. Moreover, there is no single standard of care for diabetes worldwide and no guidance documents for BAs, specifically. Given the emerging and more stringent regulations on software used in medical devices, the approval process is becoming more difficult for manufacturers to navigate, with some manufacturers opting to remove BAs from their products altogether. A comprehensive literature search was performed, including publications discussing: diabetes BA use and benefit, infusion pump safety and regulation, regulatory submissions, novel BAs, and recommendations for regulation and risk management of BAs. Also included were country-specific and international guidance documents for medical device, infusion pump, medical software, and mobile medical application risk management and regulation. No definitive worldwide guidance exists regarding risk management requirements for BAs, specifically. However, local and international guidance documents for medical devices, infusion pumps, and medical device software offer guidance that can be applied to this technology. In addition, risk management exercises that are algorithm-specific can help prepare manufacturers for regulatory submissions. This article discusses key issues relevant to BA use and safety, and recommends risk management activities incorporating current research and guidance.

Automated insulin dose bolus advisors (BAs) are used by insulin-treated diabetic patients to calculate insulin dosages for correction of out-of-target blood glucose and/or intake of carbohydrates. BAs provide a useful and simplified alternative to manual calculations of insulin dosing. To calculate appropriate bolus doses, a BA requires accurate settings for the insulin-to-carbohydrate ratio, glucose-correction factor, duration-of-insulin action, and correction target.¹ The BA then utilizes a proprietary algorithm to calculate the dose of insulin recommended for the user based on their inputs, and in some insulin dose advisors, historical data input by the user. Current literature has shown that integration of BAs into insulin pump software may help individuals more accurately meet prandial insulin dosage requirements, improve postprandial glycemic excursions, and achieve optimal glycemic control.^2-5 However, evidence is sparse regarding the specifics of risk management for algorithms such as those used in BAs. In addition, there is no worldwide standard of care for diabetes, which complicates the development of software and/or devices intended to optimize glycemic control. Not surprisingly, BAs currently on the market around the world differ widely in their approaches to dosage calculation. More specifically, inputs required for calculation, handling of insulin-on-board calculations, and outputs shown to the user all vary widely from manufacturer to manufacturer.

Medical device safety is a risk management process that must encompass the life span of medical devices from inception to disposal.⁶ These activities should be conducted in a consistent fashion, globally, to ensure the health and safety of diabetic patients in all countries. There are many local and global organizations with the charter of ensuring adequate regulation of medical devices and publishing and communicating regulations and standards to their audience. This article discusses key issues relevant to BA use and safety and provides recommendations for risk management activities incorporating current research and guidance. The purpose of this article is to present the current landscape and provide prompt further discussion on the topic of BA risk management.

Current Guidance for Development of BA Algorithms

Literature Review

Published articles, regulatory submissions, and books, primarily published within the past 10 years, with the majority being published within the last 3 years, were reviewed. International and local guidance documents were reviewed irrespective of year published, only with regard to topic and latest revision. A total of 142 sources were reviewed (see the appendix), with the following breakdown: insulin BA use/benefit (28%), standards of care/guidance documents (18%), insulin BA regulation/safety (25%), novel insulin dose algorithm (25%), and other (2%). Sources were included if they contained relevant information on drug BAs or insulin BAs, medical device regulation, and medical product risk management. Literature was excluded from the review if it did not include discussion regarding risks of drug BA use, regulation of medical devices or software, BA standards of care, development of insulin dose algorithms, or insulin BA use.

Current Guidance Documents

While numerous studies have shown that use an insulin BA is beneficial to most diabetic patients on continuous subcutaneous insulin infusion (CSII) therapy,^2-5 the standards of care published by medical organizations vary in their recommendations for therapy goals, glycemic targets, and starting parameters for insulin BAs and CSII.^7,8 For example, the 2011 American Association of Clinical Endocrinologists (AACE) Diabetes Care Plan Guidelines recommend fasting plasma glucose levels of less than 110 mg/dL and 2-hour postprandial glucose concentrations of less than 140 mg/dL. However, the 2012 American Diabetes Association (ADA) Standards of Diabetes Care recommends preprandial capillary plasma glucose of 70-130 mg/dL and peak postprandial capillary plasma glucose of less than 180 mg/dL. Neither guideline describes use or setting up of BAs for diabetes care.^7,8 Moreover, many of the publications regarding medical device risk management are case studies and implementation guides, which can make interpretation and utilization of this information difficult for manufacturers.

Local and international regulatory bodies currently provide no guidance documents published specifically pertaining to insulin BAs. However, representatives from the regulatory agencies have shared expectations regarding BA functionality and safety at various meetings and conferences in recent years. As well, both the UK Medicines and Healthcare Products Regulatory Agency (MHRA) in the United Kingdom and the US Food and Drug Administration (FDA) have recently released documents pertaining to infusion pumps specifically, which often incorporate BA technology. Both of these documents describe basic infusion pump safety considerations for their respective countries.

The MHRA document (Device Bulletin, Infusion Systems) was a device bulletin released in 2010 and is intended for use by physicians and medical staff using infusion pumps.⁹ It provides general infusion pump safety considerations for use and training of the devices.

The FDA document (Guidance for Industry and FDA Staff—Total Product Life Cycle: Infusion Pump—Premarket Notification) was released in 2010 as draft guidance and has not been finalized as of the date of this research.¹⁰ This document provides recommendations for infusion pump regulatory submission as well as specific information on insulin dosing algorithms; the MHRA document does not provide these recommendations. The document also states that for infusion pumps that contain algorithms intended to provide dosing recommendations, it is recommended that manufacturers include the following information in their premarket notification: each dosing algorithm used, each algorithm in symbolic form, all parameters in each algorithm with identification of which parameters can be modified by the user, clinical data or other justification for why it is believed that the algorithm is appropriate for the target population, proof of verification and validation activities showing that calculations made by the BA are correct, and data demonstrating that the algorithm calculates insulin doses correctly.¹⁰

In presentations at various meetings and conferences, the FDA has identified decision-making software such as insulin BAs as medical devices and actively regulates these devices.¹¹ Additional guidance released by the FDA’s Center for Devices and Radiological Health (CDRH) in 2005 (Guidance for Industry and FDA Staff: Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices) describes the expectations regarding software development, level of concern, and scope of the guidance.¹² The FDA considers insulin BAs to fall under this guidance¹¹ and recognizes 4 primary review areas for insulin BAs: (1) intended use, (2) algorithm, (3) clinical evidence, and (4) documentation.

Risk management activities are also described in documents from the American National Standards Institute (ANSI), Association for the Advancement of Medical Instrumentation (AAMI), and the International Organization for Standardization (ISO). Medical Devices—Application of Risk Management to Medical Devices ¹³ and Medical Device Software Risk Management ¹⁴ recommend the following software-specific risk management activities: (1) analysis of the clinical or other intended use(s) of the device to identify clinical hazards and related software functionality providing clinical information or control commands related to safety and effectiveness; (2) identification of potential software causes that could contribute to a hazardous situation; (3) identification of software components that implement or could affect functionality related to safety or the implementation of risk control measures and identification of hardware, software, and/or labeling risk control measures to prevent software errors from resulting in a hazardous situation or to reduce the probability of such errors; (4) evaluation of residual risk after risk control measures are identified and determination of whether the risk is acceptable given the intended use and target population of the device; and (5) verification of risk control measures and software-related software components.¹⁴

Use of an In Vitro Simulator

While some rules regarding BA algorithm assessment are supported by current literature and standards of care, other rules require additional support to demonstrate safety. Exercising the algorithm with an in vitro simulator can provide medical evidence of safety.

In vitro simulation allows the full functionality of the algorithm to be tested, especially the least likely scenarios that may not occur in a clinical trial. However, for product risk management activities, it is important that the simulator used be standardized and approved by regulatory agencies.

In 2011, the FDA approved the University of Virginia (UV) simulator for in silico testing of artificial pancreas control systems in the United States. This simulator is a computer-based diabetes simulation tool developed by UV and has been accepted by the FDA as an alternative to animal testing of type 1 diabetes control strategies.¹⁵ The simulator uses a software algorithm to model the human metabolic system. There are now hundreds of theoretical, FDA-accepted patient models that researchers can use to test control algorithms.¹⁶

Based on patient data from 300 children, adolescents, and adults with type 1 diabetes, the UV algorithm uses 26 different parameters to mimic human metabolism at the individual level, through several distinct patient profiles. Within these individual profiles, variables such as diet, exercise behavior, and insulin intake can be manipulated to test the accuracy or effectiveness of a new product under varying conditions—or to compare it to existing products.¹⁵ This simulator, or a similar approach, would be ideal for exercising the capabilities of insulin dose algorithms to demonstrate safety.

Recommendations

The availability of published guidance documents can help alleviate some of the challenges that manufacturers face and can help ensure that products brought to market are safe and effective around the world. This is especially important when considering the safety concerns of algorithms, which do not specifically fall under the guidance of medical devices or medical software. However, if the guidance is not harmonized internationally, risk management activities and, ultimately, product safety may be critically inconsistent. My opinion, after performing the extensive research (appendix) and case study activities described in this report, is as follows: because of their unique functionality and risks posed to users, insulin dose calculating algorithms should have specifically dedicated risk management activities assigned. The following recommendations may be helpful in developing standardized, international guidance for software designers, manufacturers, and regulatory agencies (Table 1).

Table 1.

Recommendations for Insulin Dose Calculator Risk Management.

Recommendation 1	Require in vitro testing of insulin dosing algorithms utilizing an in vitro simulator, such as the previously described University of Virginia simulator.
Recommendation 2	Require risk management activities by global regulatory agencies in accordance with Medical Device Software Risk Management and Guidance for Industry and FDA Staff—Total Product Life Cycle: Infusion Pump—Premarket Notification; Draft Guidance, Guidance for Industry and FDA Staff: Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices or (similar updated documents) as an international standard.
Recommendation 3	Require standardized algorithm documentation for submission of stand-alone insulin dosing calculators, as well as those that are a constituent of a medical device.
Recommendation 4	Develop a worldwide database of registered products to improve regulatory transparency around the world, as well as to monitor and track device safety throughout the device life cycle.
Recommendation 5	Avoid requiring standardization of insulin dosing algorithms to preclude the stifling of scientific and technologic advances in the area.

• Recommendation 1: Require clinical study and/or in vitro testing of insulin dosing algorithms utilizing an in vitro simulator, such as the previously described UV simulator.

In vitro testing of insulin dosing algorithms with the educational version of the simulator would provide a standard platform for algorithm testing. If accepted as an international standard, all algorithms developed in other countries would be held to the same standard of proof. This would promote consistency and safety of insulin dosing algorithms, and would assist regulatory agencies in reviewing in vitro testing results.

• Recommendation 2: Require risk management activities by global regulatory agencies in accordance with Medical Device Software Risk Management and Guidance for Industry and FDA Staff—Total Product Life Cycle: Infusion Pump—Premarket Notification,¹⁰ Guidance for Industry and FDA Staff: Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices,¹² or similar updated documents as an international standard in addition to ISO 14971.

Risk management activities in accordance with these recommendations would ensure the consistency of risk management activities for all insulin dosing algorithms. Once provided as an international standard, this document would provide the much-needed guidance for manufacturers to follow and ensure the safety of their algorithms, as well as for submission of their algorithm-containing software or device to regulatory agencies. For the regulatory bodies, having an approved guidance would ensure clear directives for submissions to follow, thus making the approval process more straightforward with clear expectations.

• Recommendation 3: Require standardized algorithm documentation for submission of stand-alone insulin dosing calculators, as well as those that are a constituent of a medical device.

This type of exploration and documentation would be extremely helpful for manufacturers and regulatory agencies as it clearly demonstrates the medical evidence associated with each component of the insulin BA algorithm. In addition, appropriately modified information of this type may even be beneficial to consumers. The exploration of literature and justification of algorithm rules would identify gaps in the literature support of the algorithm. The areas of the algorithm not supported by current literature or standards of care could then be justified in another manner, such as in vitro simulation or clinical studies, to meet regulatory standards for safety. This would ensure that all aspects of the algorithm are thoroughly explored and supported to ensure safety.

• Recommendation 4: Develop a worldwide database of registered products to improve regulatory transparency around the world, as well as to monitor and track device safety throughout the device life cycle.

The benefit of these databases is well known; however, their benefit is limited to the geographic area(s) which they support. If a product is sold in an area of the world where it is not registered, then that country will have no record of the product or any postmarket reporting on that product. This is especially problematic in the EU, where registers are maintained by competent authorities and a manufacturer needs register with only one to market its product throughout the EU.

• Recommendation 5: Avoid requiring standardization of insulin dosing algorithms to preclude the stifling of scientific and technologic advances in the area.

There are many different insulin dosing algorithms currently available globally, and novel algorithms continue to be researched and developed. Many of these algorithms behave differently from one another in their calculation method and logic, but generate similar recommendations. Fine-tuning the algorithm, allowing varying inputs, creating “smart” algorithms, and integrating blood glucose (BG) and continuous glucose monitoring (CGM) data all enhance the functioning of insulin BAs over time. To avoid hindering innovation, it is essential that a standardized algorithm not be mandated by regulatory agencies. Acceptance of a policy such as this would stifle the evolution of these algorithms and prevent potential breakthroughs in diabetes care.

Conclusions

There is a general consensus that insulin BAs are useful and beneficial.^2-5 Current literature also shows that additional novel insulin dosing algorithms are being researched and developed.^17-20 After an extensive literature research (see the appendix) and case study exploration, my opinion on this topic is as follows: clearer guidance is needed to assist both the manufacturers creating these algorithms and the regulatory agencies that will ultimately review them.

Specific risk management activities for BA algorithms, including simulator testing, flowchart diagramming, and literature support search, have been recommended in this article. Risk management activities, as described in a few key guidance documents related to medical device software and infusion pumps, have also been identified.

Manufacturers and developers of insulin BAs would benefit from standardized guidance because risk management expectations would be clearly defined and well organized, thus facilitating greater adherence. It would also provide consistency for regulatory agencies reviewing submissions, to ensure the testing and results were consistent and more easily reviewed.

However, while required standardization of algorithm may ease regulatory burden regarding insulin BAs, this would likely stifle innovation in the field of diabetes. Mandating a standard method for calculating insulin dosage may hinder the development of algorithmic advances that could potentially provide breakthroughs in the field of diabetes care around the world.

In summary, international acceptance and adoption of the recommendations put forth in this article for insulin BA risk management and guidance may greatly benefit manufacturers, regulatory agencies, and, most important, people with diabetes worldwide. This article provides a short but solid survey of the problem and will hopefully prompt further attention and publication regarding this topic.

Appendix

Appendix.

Comprehensive Literature Review List

Group	First author	Subject	Year
Insulin dose algorithm/use
	Barnard	Use of an automated bolus calculator in MDI patients	2012
	Bernasko	Intensive insulin therapy	2007
	Blazik	The education of patients	2010
	Bode	Diabetes management in the new millennium	2002
	Cavan	Automated bolus advisor control and usability study (ABACUS): protocol	2012
	Clavel	Treatment by external insulin pump	2010
	Colin	Glucose meters with built-in automated bolus calculator	2013
	Colquitt	Clinical and cost-effectiveness	2004
	Cukiermann-Yaffe	Key elements for successful	2011
	Gross	A bolus calculator	2003
	Hirsch	Practical pearls	2010
	King	Basal bolus dosing	2005
	King	A prospective evaluation	2007
	Klupa	Clinical usefulness of a bolus calculator	2008
	Mastrototaro	The integrated minimed	2009
	Michon	Insulin pump and intensive	2002
	Murphy	Reducing hyperglycemia	2009
	Pollom	Optimizing inpatient glycemic control	2010
	Renard	Insulin pump use in Europe	2010
	Roemer	Tips and tools for insulin pump	2009
	Romatowska	The effect of using the insulin pump bolus calculator compared to standard insulin dosage calculations in patients with type 1 diabetes mellitus—systematic review	2013
	Scheiner	Insulin pump therapy	2009
	Schmidt	Use of an automated bolus calculator in MDI-treated type 1 diabetes: the BolusCal Study	2012
	Shalitin	Benefits of a bolus calculator	2010
	Shashaj	Benefits of a bolus calculator	2008
	Spellman	Management of diabetes	2003
	Tamborlane	Insulin pump therapy in childhood	2003
	Unger	Optimizing insulin therapy	2008
	Walsh	Guidelines for insulin dosing	2010
	Walsh	Guidelines for optimal bolus	2011
	Walsh	Pumping insulin	2003
	Walsh	Pumping insulin	2006
	Wei	Basal-bolus insulin protocols	2011
	Weissberg	Insulin pump therapy	2003
	White	Insulin pump therapy	2007
	White	When and how to implement	2009
	Wilson	Intensive insulin therapy	2007
	Wolpert	Hands-on demonstration	2005
	Zhang	The current situation review	2010
	Zhang	Second insulin pump safety	2010
	Zisser	Bolus calculator: a review	2008
	Zisser	Clinical performance of 3 bolus	2010
	Zisser	Clinical update on optimal prandial insulin	2009
Standards of care/guidance documents
	AACE	Diabetes care plan guidelines	2011
	AADE	Consensus insulin pump therapy	2008
	AAMI	Medical device software risk management	2005
	ADA	Practical insulin	2011
	ADA	Standards of medical care	2011
	EC	Directive 98/79/EC	1998
	EN	Biological evaluation	2009
	EN	Clinical investigation of medical devices	2009
	EN	Medical device software	2006
	EN	Medical devices—application of risk management	2009
	EN	Medical devices—application of usability	2008
	EN	Medical devices—quality management	2009
	EN	Medical electrical equipment part 1-10	2008
	EN	Medical electrical equipment part 2-24	1998
	FDA	Draft guidance	2011
	FDA	Guidance for industry	2010
	ISO	Guide to the development	2012
	ISO	Risk management	2009
	ISO	Safety aspects	1999
	Malerbi	Brazilian diabetes society	2006
	TIR	Technical information report	2009
	WHO	Medical device regulations	2003
		aami.org
		fda.gov
		iso.org
Medical device regulation/safety
	Babamir	Constructing a model-based	2010
	Bartoo	Risk management	2003
	Benet	A risk driven approach	2011
	Burton	Improving software risk management	2008
	Burton	A risk management capability model	2008
	Burton	A software risk management capability model	2008
	Coverity	Governance, risk, and compliance	2011
	Devices4Limited	Regulation of health apps	2012
	Donawa	Effective risk management	2004
	Goodman	Medical device error	2002
	Grimes	Medical device security	2004
	Hedge	Case study—risk management	2011
	Israelski	Human factors risk management	2004
	Jardine	Risk management frameworks	2003
	Jones	Risk management in the design	2002
	Klonoff	Insulin pump safety meeting	2009
	McCaffery	Improving software risk	2009
	McCaffery	Risk management capability model	2009
	Mingziao	A quantitative approach	2011
	Narvaez	Prospective risk management	2010
	Paul	A review of the security	2011
	Pietzsch	Review of US medical device	2007
	Rakitin	Coping with defective software	2006
	Reviewers	Guidance for industry and FDA	2000
	Robson	Risk management and biomedical	2005
	Schultz	Medical device safety	2007
	Shah	Current status of the regulation	2008
	Sherwin-Smith	Medical applications—the future of regulation	2012
	Small	Medical device-associated safety	2004
	Wang	Discussions on the regulatory	2009
	Wood	Software risk management	2008
	Xie	Risk management for medical devices	2007
	Xu	Discussion about the application	2006
	Zimlicki	Bolus calculator regulation	2011
	Zuckerman	Medical device recalls	2011
Novel algorithm
	Anderson	Multiple daily injections	2009
	Boiroux	Optimal insulin administration	2009
	Boiroux	Strategies for glucose control	2010
	Bondia	Coordinated basal-bolus	2009
	Campbell	Calculating insulin on board for extended bolus
	Charpentier	The Diabeo software	2011
	Cook	The intelligent dosing system	2005
	Davidson	Analysis of guidelines	2008
	Edelman	Insulin pump therapy	2010
	FDA	Endotool drug dose calculator	2006
	FDA	Glucommander plus	2006
	FDA	Glytec, LLC	2010
	FDA	Intelligent dosing system	2008
	Gonzalez-Rodriguez	Should the amount of fat and protein	2010
	Grandinetti	Web-based prediction	2011
	Hedbrandt	The Sarimner diabetes simulator	2007
	Juneja	Safety and effectiveness	2008
	King	How much do I give?	2010
	Kordonouri	Supplementary fat plus protein	2010
	Laha	Glucose control in critical illness	2008
	Lippaiova	Safety of new algorithms	2008
	Marling	Case-based decision support	2008
	Mathews	A new model to estimate bolus	2009
	Olinder	Postprandial glucose levels	2009
	Palerm	Prandial insulin dosing using	2007
	Pankowska	Bolus calculator with nutritional	2010
	Pankowska	A review of the “bolus guide”	2010
	Pelzer	Preliminary application of a new bolus	2011
	Revert	Calculation of the best basal-bolus therapy	2011
	Revert	Combining basal-bolus	2010
	Rossi	Diabetes interactive diary	2010
	Rossi	Interactive diary for diabetes	2009
	Sahota	Computerized clinical decision	2011
	Schwartz	Use of case-based reasoning	2008
	Shapira	Bolus guide: a novel insulin	2010
Other
	ADA	Workgroup on Hypoglycemia	2005
	Estabrook	Commercially available UV simulator	2011
	Kravitz	From insight to implementation	2005