Insulin Metrics: Need for Development of Consensus Standards for Reporting of Insulin Dosing Data

Abstract

The advent of connected insulin pens will generate an avalanche of digital insulin data, especially in the context of prandial- and multiple daily injection-insulin regimens. There is a need for the diabetes community to develop standards for such data, analogous to what has been achieved using the ambulatory glucose profile and associated metrics for glucose, permitting harmonization of data reporting for multiple devices and facilitating integration of glucose, insulin, food intake, and physical activity data. Several studies have estimated the timing of meals by analyses of glucose excursions but using diverse criteria. There is need for uniform criteria for multiple types of insulin boluses, including premeal, perimeal, delayed, missed, and correction boluses to facilitate research studies and patient care. This article contains a first preliminary proposal for standards regarding reporting of insulin dosing data. Clinical usage of these reports will require sensitive communication between health care providers and patients.

Introduction

The widespread use of continuous glucose monitoring (CGM) and flash glucose monitoring has been the catalyst for the development of consensus guidelines for reporting of glucose data and metrics, as well as treatment goals such as time in range. ^{1 –3} Until recently, insulin dosing data could only be collected from downloadable insulin pumps. In the forthcoming years, connected insulin pens will provide dosing data from a much broader population of individuals with both type 1 and type 2 diabetes. Availability of these insulin data can potentially drive improvements in glucose control. ⁴ However, insulin data will need to be distilled into clinically meaningful metrics and actionable insights to drive clinical intervention and ensure that data overload does not overwhelm the busy clinician.

Insulin metrics generated by automated data analysis could be used to identify insulin dosing practices that contribute to suboptimal glucose control. ^{5 –11} This includes several aspects of insulin dosing behavior such as missed food boluses, delayed boluses, suboptimal meal boluses, dosing to correct hyperglycemia, and insulin dosing practices contributing to hypoglycemia. Connected insulin pens also can provide insights about needle priming practices. Insights derived from use of connected insulin pens and pumps about dosing practices contributing to suboptimal glucose control could provide the foundation for the development of age-appropriate and individual-specific “insulin metric” goals.

To set the stage for this new era, we believe that it will be important for the professional community to establish a defined process to reach consensus regarding the analysis and reporting of insulin dosing data. In this article, we propose a set of “placeholder” insulin metrics that will need to be reviewed, critiqued, refined, and vetted by the diabetes community, including health care professionals and people with diabetes as well as representatives of the medical device industry. In addition, we present some foundational behavioral principles to guide the clinician in communication with the person with diabetes about insulin data.

Metrics for Insulin and Glucose

Table 1 gives an initial proposal from the present authors, intended to serve as a placeholder and prototype. The criteria to identify these different dosing practices and related metrics—based on both CGM and intermittent self-blood glucose monitoring—will need to be defined. Table 2 gives the areas in clinical practice where these metrics are expected to be useful. Again, this is intended to serve as a placeholder and needs to be critically refined by a large number of stakeholders. Data sets from a broad spectrum of persons with diabetes will need to be collected to determine the relationship between different insulin dosing metrics and glycemic control and to establish the clinical utility of these metrics.

Table 1.

Insulin and Glucose Metrics Obtainable Using Connected Insulin Pens Combined with Continuous Glucose Monitoring and Food Intake Data

Insulin metrics	Parameters/considerations
Total daily insulin dose (Units, U/kg)	Basal, meal, and correction insulin (if available from bolus calculator)
	Basal–bolus ratio
Distribution of insulin administration by time of day for any 1 day and for data pooled over multiple days (e.g., 2 weeks) (Units, U/kg)	No. of injections
	Average dose of injections at any specified time (or time segment) of the day
	No. of units administered at any specified time (or time segment) of the day (calculated as average no. of insulin injections during a specified 1-h time period multiplied by the average dose administered during the same time interval)
	Insulin profiles by time of day—for 1 day, or the average value for data collected on multiple days
	i. Timing of bolus administration
	ii. Insulin on board
	iii. Plasma insulin levels (calculated using pharmacokinetic modeling)
	iv. Insulin action/activity profiles (calculated using pharmacodynamic modeling)
No. of missed boluses per day	Overall and by type of meal
No. of delayed boluses per day (average)	Overall and by type of meal
No. of early boluses per day (average)	Overall and by type of meal
No. of episodes of bolus stacking, average per day
Identification of correction boluses	Presumptive if glucose is elevated and no carbohydrate consumption has been entered
Distribution of episodes of hypoglycemia throughout the day	Graphical displays of %Time Below Range (TBR) by time of day, or of onset, duration, and end of hypoglycemic episodes
Distribution of episodes of hyperglycemia throughout the day	Graphical displays of %Time Above Range (TAR) by time of day, or of onset, duration, and end of hyperglycemic episodes
Distribution of mealtime and snacks throughout the day	Graphical displays of timing of meals and snacks by time of day, with option to display size and duration of meal
Distribution of duration of time intervals between insulin bolus administration and onset of food ingestion for each type of meal	Graphical displays by time of day using pooled data collected over multiple days (e.g., 2 weeks), or summary statistics (median, interquartile range)
Analyses by day of the week	Displays and comparisons of insulin dosing patterns, glucose distribution and patterns, dietary intake, and physical activity for each day of the week, for example, comparisons of weekdays versus weekends

Table 2.

Clinical Utility and Application of Insulin Metrics Integrated with Glucose, Meal, and Physical Activity Data

Clinical utility	Recommendations
Patient education	Reinforcement for effective behaviors.
	Identification of potential problem areas.
	Integrating the use of glucose and insulin data.
Physician and HCP education	Evaluating risks of hypo- and hyperglycemia.
	Understanding of the time course of insulin action.
	Distinguishing between timing of insulin administration, time course of insulin-on-board, plasma insulin levels, and profiles of insulin pharmacodynamic activity (glucose disposal rate).
Improved patient—HCP communication, including patient education	Review of recent glycemic control, patterns of glucose after meals, influences of dietary intake, insulin bolus dose, and timing on glycemic patterns; effects of missing, delayed, premature, and correction boluses, and of “bolus stacking.” Discuss needs for modification of timing, quantity or content of meals or of time intervals between insulin dosing and onset of meals. Periodic revision of parameters used for calculation of insulin boluses and possible use of an automated bolus calculator.
Assessment of patient lifestyle and adherence	Patient estimation of insulin dosage and timing of insulin administration (both clock time and in relation to meals).
	Assessment of missed boluses, delayed boluses, early boluses, correction boluses, and other types of insulin administration.
	Identification of frequency of insulin “bolus stacking” behaviors and relations to magnitude of the triggering hyperglycemia and to any subsequent hypoglycemic episodes.
	Assessment of self-adjustment of insulin doses in response to variations in carbohydrate content and carbohydrate load, and with respect to premeal glucose levels, and the combination or interaction of these two factors.
	Degree of standardization of lifestyle with regard to timing of insulin boluses, meals, and physical activity.
Assessment of appropriateness of parameters for calculation of insulin bolus doses	Carbohydrate-to-insulin ratio (CIR, CR, or ICR), correction factor, or insulin sensitivity factor, and duration of insulin action.
Assessment of possible need for restructuring the insulin dosage regimen, diet, meal timing, carbohydrate load, and other aspects of dietary intake (protein, saturated, and unsaturated fats, fiber), and single- versus multicourse meals
Assessment of possible need for modification of antihyperglycemic medications	Consider medications with major influence on postprandial glucose excursions (e.g., GLP-1 RAs, SGLT-2 inhibitors, DPP-4 inhibitors, alpha-glucosidase inhibitors, sulfonylureas, and glinides).

DPP-4, dipeptidyl peptidase-4; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HCP, health care professional; SGLT-2, sodium–glucose cotransporter-2.

The evaluation of insulin and CGM data can be complicated. One of the most common questions a clinician and patient may ask is whether insulin boluses have been administered at the time of onset of each meal. This should become an extremely simple and trivial exercise if one were to have a consistently reliable fully automated method for recording the time of onset of meals. However, most people do not record the exact time at onset of their meals. Even if available at all, these records are frequently incomplete and inaccurate. Hence, we usually must rely on available data—usually the glucose versus time profiles. Utilizing these glucose versus time profile together with a definitive, accurate, and complete record of time of insulin bolus administration, one would like to be able to identify the timing of meals (and hence meal-related insulin boluses) by assuming that they are consistently accompanied by a peak in glucose versus time. Several criteria have been used for identification of a significant upstroke or peak in the glucose, or what might be regarded as a putative glucose excursion after a meal. ^{12 –16} Consensus will need to be reached about the possible numerical parameters. We will also need to achieve a consensus on the best methods to report the observed characteristics for insulin dosing in terms of both statistics and graphical displays—both in individuals and in groups of subjects—for a single day, and for a timespan of multiple days or weeks. In most cases, it will be important to obtain data from multiple subjects in each of several categories including diabetes type, age group, geographical location, sociodemographics and ethnicity, quality of glycemic control, and modalities of therapy. Furthermore, we hope that the stakeholder community would develop a set of rules or algorithms for interpretation of those numerical values. In addition, development work will need to address how other inputs, including activity and food data, can be incorporated into insulin metrics.

Several barriers and limitations will need to be addressed to facilitate the adoption of insulin metrics into clinical practice. Integration of data sources from different insulin delivery and glucose monitoring devices will be essential. In addition, a set of insulin metrics based on intermittent glucose measurements will be required to address the needs of individuals who use blood glucose monitoring in their daily self-management.

Data Collection: General Considerations

Consideration also needs to be given to the conditions under which glucose and insulin data are collected. Having additional contextual information about glucose levels, food intake, and bolus dose calculation could enhance insulin data analysis. However, adherence behavior can be affected by the requirement for patients to record or log additional information. To limit this potential confounder, in a recent study directed at using connected pens to evaluate the frequency of missed insulin boluses, by design we did not ask subjects to document factors underlying dosing decision making. ¹⁴ Consideration also needs to be given to whether metrics generated from insulin data analysis are appropriate markers of desired self-care behavior. If bolus dose recommendations from the bolus calculator are known, any deviation between the recommended and administered insulin dose needs to be evaluated in the context of glucose control and other factors. For example, with exercise or with variation in food intake, deviations from dose recommendations may be necessary and, accordingly, compliance with the recommended dose alone (without consideration of glucose levels) could erroneously characterize appropriate and desirable self-care behaviors as “noncompliance.” ¹⁷ Once a set of metrics has been selected, one can conduct systematic evaluation of how much data and how long a period of data collection are needed to obtain reliable estimates. ^{18 –20} Furthermore, one can evaluate which metrics provide the greatest sensitivity to detect changes within an individual or to discriminate between two subjects or groups of subjects. ²¹

Behavioral Considerations

When insulin pen and pump data are summarized and discussed with the patient, whether in live conversation with one's clinician or in digital form, it is likely to be better received if it is framed in a collaborative manner that avoids blaming and shaming. This can be more readily accomplished if the clinician understands that suboptimal insulin bolusing behavior is common and usually not because someone is being foolish, lazy, or “noncompliant.” Patients are usually rational actors and, although they would prefer to make good use of their insulin devices to achieve desirable glycemic outcomes, they are responding in an understandable manner to a range of psychosocial influences. ²²

One recent study showed that elevated levels of diabetes-related emotional distress are associated with more missed boluses over a 9-month period. ²³ Insulin boluses may be frequently skipped or mismanaged in circumstances wherein the individual

• a. has become fearful of potential hypoglycemia ²⁴ ;

• b. avoids bolusing in social situations so that his/her diabetes and diabetes management remains private;

• c. becomes exhausted by the continuous cognitive and behavioral demands of diabetes self-management leading to the need for “breaks” from mealtime bolusing;

• d. believes that avoiding insulin can be an effective weight management strategy ²⁵ ; or

• e. has become skeptical that careful attention to bolusing recommendations is likely to lead to improved glycemic outcomes or contribute to long-term health. ^23,26

If insulin data are presented to the patient as a test of their “compliance” or as the basis for a judgment of the patient's performance, many people with diabetes may recoil from engaging with their health care professional and from attempting to understand and learn from the data. In many cases, appreciating the underlying rationality of patient behavior can be a first step toward developing an effective approach to summarizing and discussing insulin data from connected pens and pumps.

Since it is well established that language can shape perceptions and behavior, such conversations should be conceptualized as a nonjudgmental two-way conversation, and not as the delivery of a “compliance report.” The clinician might begin by pointing out where the patient has been successful, since the key to promoting behavior change is often to build upon previous successful behavior. Since the patient is the expert regarding his own behavior, it will be essential to make use of his/her expertise by asking about what may be making it difficult to administer insulin boluses as needed, and to encourage the patient to brainstorm regarding potential solutions in concert with the clinician. With a clearer understanding of the patient's perspective, the clinician can then focus on putting forward more effective recommendations by providing tools (e.g., bolus advisors) to reduce perceived burden and strategies to address fear of hypoglycemia. Clinicians can optimize the use of insulin data and encourage patient engagement by discussing the current findings in a relaxed conversation where the burdens and challenges of diabetes self-management are acknowledged and realistic expectations are presented. ²⁷

Conclusions

There is an urgent need for the diabetes community to work toward developing a consensus regarding the most effective methods to characterize and report insulin dosing data and to integrate insulin and glucose data reports. The availability of such a consensus—similar to what has been achieved for interpretation of glucose data—should facilitate the use of insulin data collected from connected pens and pumps in clinical care, and foster improvements in diabetes outcomes.

Consensus needs to be reached in the diabetes community about a set of metrics to characterize insulin dosing practices and thereby serve to provide actionable insights to the patients and their caregivers. As use of connected insulin pens becomes more widespread and insulin dosing data from multiple populations of people with diabetes become progressively more available, these metrics, methods, and constructs—including extensive forms of graphical, tabular, and textual display—must evolve and become further refined.