Guide to insulin initiation and titration. Home Clinical resources Clinical guidelines Key RACGP guidelines View all RACGP guidelines Management of type 2 diabetes: A handbook for general practice Appendix 2.

Types of insulin available Appendix 2. Guide to insulin initiation and titration Appendix 3. Detailed information on glycaemic emergencies Appendix 4. Practice summary: Diabetes in pregnancy Appendix 5. Medication-related care plan considerations for aged care residents with type 2 diabetes.

Supplementary material Disclaimer Acknowledgements Acronyms and initialisms About the RACGP About Diabetes Australia Updates in this edition Explanation and source of recommendations Summary of recommendations Figures and tables Provided under licence Search.

Download chapter Last revised: 17 Sep Starting and adjusting basal insulin 1—3. Figure 1. Starting and adjusting basal insulin. Pharmacotherapy: A Pathophysiologic Approach. New York: McGraw Hill; Mandal TK. Inhaled insulin for diabetes mellitus. Am J Health Syst Pharm. Davidson MB, Mehta AE, Siraj ES.

Inhaled human insulin: an inspiration for patients with diabetes mellitus? Cleve Clin J Med. Riddle MC. Glycemic management of type 2 diabetes: an emerging strategy with oral agents, insulins, and combinations. Endocrinol Metab Clin North Am. Hermansen K, Davies M, et al. A week, randomized, parallel, treat-to-target trial comparing insulin detemir with NPH insulin as add-on therapy to oral glucose-lowering drugs in insulin-naive people with type 2 diabetes.

Hirsch IB. Insulin analogues. N Engl J Med. Brunton S, Carmichael B, et al. Type 2 diabetes: the role of insulin. J Fam Pract. Rave K, Bott S, et al. Time-action profile of inhaled insulin in comparison with subcutaneously injected insulin lispro and regular human insulin.

Riddle MC, Rosenstock J, Gerich J. The treat-to-target trial: randomized addition of glargine or human NPH insulin to oral therapy of type 2 diabetic patients. Siebenhofer A, Plank J, et al. Short acting insulin analogues versus regular human insulin in patients with diabetes mellitus.

Cochrane Database Syst Rev. Jellinger PS, Davidson JA, Blonde L, et al. Texas Department of State Health Services. Insulin algorithm for type 2 diabetes mellitus in children and adults.

Publication Raskin P. Can glycemic targets be achieved--in particular with two daily injections of a mix of intermediate- and short-acting insulin?

Yki-Jarvinen H. Combination therapies with insulin in type 2 diabetes. Rosenstock J, Davies M, et al. Insulin detemir added to oral anti-diabetic drugs in type 2 diabetes provides glycemic control comparable to insulin glargine with less weight gain.

Malone JK, Bai S, et al. Twice-daily pre-mixed insulin rather than basal insulin therapy alone results in better overall glycaemic control in patients with type 2 diabetes.

Diabet Med. Raskin P, Allen E, et al. Initiating insulin therapy in type 2 diabetes: a comparison of biphasic and basal insulin analogs. Garber AJ, Wahlen J, et al. Diabetes Obes Metab.

Liebl A, Prager R, et al. Biphasic insulin aspart 30 BIAsp30 , insulin detemir IDet and insulin aspart IAsp allow patients with type 2 diabetes to reach A1C target: the PREFER study. Asakura T, Seino H. Assessment of dose selection attributes with audible notification in insulin pen devices.

Diabetes Technol Ther. Paparella S. Avoiding errors with insulin therapy. J Emerg Nurs. Triplitt C, Wright A, Chiquette E. Incretin mimetics and dipeptidyl peptidase-IV inhibitors: potential new therapies for type 2 diabetes mellitus.

To comment on this article, contact editor uspharmacist. Pharmacy Practice Affordable Medicines Biosimilars Compliance Compounding Drug Approvals. COVID Dermatology Diabetes Gastroenterology Hematology.

Long-acting insulin can still be titrated and adapted using SMBG values from a glucose meter. SMBG is taken in fasting conditions, usually before the breakfast meal, and used to adjust the basal insulin dose. In the following, two existing strategies to adapt the basal insulin dose using SMBG records will be presented and compared against the proposed algorithm.

In order to keep a fair comparison between the algorithms, the same update rule, with the same saturation and dead-zone, reported in equation 1 is used but with the newly calculated B C opt. Long-acting insulin dose can be titrated following a control-to-range heuristic rule.

Rules are composed of a chosen range e. Most titration algorithms used in clinical practice can be considered of this sort Here, an algorithm similar to the one proposed by Visentin et al. to titrate insulin glargine was used 37 , which defines the optimal insulin dose as:.

This baseline algorithm is taken from recent work from Cescon et al. where an iterative learning control ILC is proposed to optimize the basal insulin dose This ILC algorithm finds the new optimal dose B C opt to be administered such that the SMBG values are driven as close as possible to the desired reference trajectory G target :.

Similarly to 16 , the DC gain of the filter F q is individualized to each subject as 3 BW. The basal insulin dosing adaptation algorithm was tested in a day simulation study including virtual adult subjects with T1D using MDI therapy.

Daily basal insulin doses of glargine were administered either in the morning or before bedtime and were usually taken around mealtime, when possible i. The virtual subjects could treat hypoglycemia events, but a hypoglycemia unawareness algorithm was implemented where there is a chance that the hypoglycemia event is not treated for a period e.

Additionally, hypoglycemia events between 0 am at 6 am were not treated to reinforce night hypoglycemia. Each virtual subject was characterized by a basal glucose and the basal dose needed to achieve this basal glucose, U basal ss. As reported in 27 , a model for glucose meter was used to generate SMBG values from blood glucose levels, and a model for a CGM sensor was used to generate glucose readings Consumed carbohydrate amount was predetermined in the simulator but unknown to the adaptation algorithms.

Five treatment arms were simulated: a a control arm where altered insulin dosing parameters were kept the same throughout the experiment CTR ; b a baseline arm where the control to range SMBG algorithm is employed SMBG-Rule ; c a baseline arm where the control to reference SMBG algorithm is employed SMBG-ILC ; d an experimental arm where the proposed algorithm is employed CGM-Opt ; e another experimental arm similar to d but where counted carbohydrate amount was given to the optimization algorithm instead of being reconstructed CGM-Opt-Carb.

Similar to clinical practice 39 , the insulin dose was titrated every three days to provide enough time for insulin dose change effects on fasting glucose to stabilize. In the SMBG-Rule and SMBG-ILC arms, three pre-breakfast SMBG values were used. In the CGM-opt and CGM-Opt-Carb arm, the CGM data up to the time of the next basal dose recommendations around three days was used.

The simulation experiment was not interrupted during the days. The same simulation was repeated for two different scenarios: nominal and variance , described below.

In this scenario, the virtual subjects consumed three similar meals each day at the same time. Meals were taken at 7AM, 1PM, 7PM and the amount of carbohydrates per meal was 50g, 75g and 75g. Behavioral variability consisted of consuming three main meals and up to three unannounced and unbolused snacks.

Meal sizes were variable, but the total carbohydrates consumed over the day were between g and g, main meals were bigger than 30g, and snacks were smaller than 40g e. The insulin bolus could be delayed by up to 1 hour after consuming a meal.

Metabolic variability was implemented by varying the insulin sensitivity during the day and between days. Metrics were calculated for each arm every 15 days.

Before starting the days experiment, an initial 15 days was simulated and used as a baseline. All results are reported as mean ± standard deviation across subjects for a day duration. Changes of a certain day period as compared to the baseline period are reported as average and confidence interval CI.

At the same time, TBR was decreased using the CGM-Opt in both scenarios These results are summarized in Figure 1 for the nominal scenario and Figure 2 for the variance scenario.

An exhaustive comparison between the four arms CTR, SMBG-Rule, SMBG-ILC, and CGM-Opt is provided in Table 1. Figure 1 Summary of glycemic outcomes in every 15 days period for the nominal scenario of the in-silico experiment.

Values are shown as mean and standard deviation. Figure 2 Summary of glycemic outcomes in every 15 days period for the variance scenario of the in-silico experiment.

The average basal dose in the last days was compared to the original U basal ss. In Table 2 , results of this comparison are shown separately for virtual subjects that started with a higher basal dose and with a lower basal dose.

In Figure 3 , the percentage basal dose changes in both scenarios are shown. Table 2 Summary of changes in basal dose from theoretical steady-state optimal value. Figure 3 Summary of long-acting dose changes in titration days every 3 days. Values are shown as median and interquartile range.

There was no clinically significant change in the glycemic metrics between the CGM-Opt arm and the CGM-Opt-Carb arm. Similarly, as it can be seen in Table 3 , there was no differences between calculated optimal basal dose B C opt for each day in the two arms CGM-Opt vs CGM-Opt-Carb.

Table 3 Summary of changes between the optimization procedure when the carbohydrate input is reconstructed or counted by virtual subjects. People with T1D live with the life-long burden of making important decisions about their daily insulin doses. Technological advances in diabetes treatment can help in easing this burden.

Specifically, the new generation of SIP and the affordability of CGM are facilitating the development of a decision support system designed for people using MDI therapy. The proposed algorithm will enable such decision support systems by automatically suggesting adaptation of the basal insulin dose after analyzing SIP and CGM records.

Our algorithm is based on a metabolic model that describes the complex glucose traces by separating the effects of the basal insulin dose from other system inputs, i. This approach is inspired by the clinical practice where patients are usually asked to skip meals in order to optimize their basal insulin Once the model is able to describe the data, we can mathematically eliminate the effect of meals and boluses on the glucose trace, thus isolating the effect of basal dose on the theoretical fasting glucose and allowing for its optimal tuning.

This method follows a similar insulin basal rate optimization approach described by Fabris et al. Estimating the residual metabolic signal is key to our approach since it detects changes in the glucose curve that are independent of delivered insulin boluses and consumed meals but needs to be controlled through the basal dose.

The original idea of a model-based residual metabolic signal estimated for insulin titration was introduced by Patek et al.

and refined in other works 29 , 33 , Another similar model-based method was proposed previously by El Fathi et al. In this work, a recent subcutaneous absorption model of basal dose was employed Another difference is that the basal dose is optimized independently from other model parameters, giving the possibility to mold the cost function to enforce a desired outcome e.

To put the performance of this algorithm into perspective, we compared it with a control-to-range algorithm inspired by the current clinical practice and a control-to-reference algorithm that was recently proposed. Both algorithms use the current standard of titrating the long-acting insulin dose from the pre-breakfast SMBG measurement.

In general, results have shown that the proposed algorithm outperforms the other methods at night and can achieve comparable results overall.

This can be explained by multiple factors i with CGM, we can observe the full glucose profile, thus clearly detecting degradations in night control; ii we explicitly biased the optimization equation in 9 to reduce hypoglycemia events during the night; iii once the night period is optimized, we did not aim to optimize glycemic metrics in the day period by optimizing insulin boluses.

Our algorithm also reduced glycemic variability as measured by the glucose standard deviation in both scenarios Table 1. This is aligned with our observation in Table 2 that this algorithm can recover the theoretical steady-state basal dose, which in theory is the one that will cause the least variations in the glucose curve.

Furthermore, one can argue that reducing overall glycemic variability will facilitate optimizing parameters used to compute insulin boluses in the day period.

Our simulations have shown that the control-to-range algorithm used in the clinical practice SMBG-Rule is effective in titrating the long-acting insulin dose by reducing both hypoglycemia and hyperglycemia.

Unexpectedly, the control-to-reference algorithm SMBG-ILC did not perform similarly in the two scenarios. In spite of showing good performance in the nominal scenario, the ILC-based algorithm was not able to reduce hypoglycemia in the variance scenario. In Table 1 , we can see that the mean fasting SMBG values were driven close to as expected by the algorithm, but this was achieved at the cost of a higher hypoglycemia exposure.

This suggests that individualizing the target of the ILC algorithm or stopping titration early for each subject may be necessary in clinical practice. This simulation also hints that the ILC algorithm may benefit from the use of CGM data instead of SMBG. We have also shown that our algorithm is robust to carbohydrate information, as seen in Table 3.

This is a result of keeping meal parameters free to describe the glucose curve with the least a-priori knowledge during the model individualization. Therefore, the proposed meal reconstruction approach using the simple equation in 6 is shown to be sufficient for titration purposes.

Not relying on carbohydrate counting will facilitate the use of this algorithm by T1D patients. In Figure 3 , we can see that the algorithm converges in about 30 days after ~10 cycles.

However, in the variance scenario, the algorithm continued to make small changes. This can be attributed to the metabolic and behavioral variability in this scenario. In this scenario, the algorithm might be undesirably chasing noise, which suggests that the dead zone in Eq.

Interestingly, as seen in Figure 2 , glycemic outcomes are kept stable even after these changes. We should recognize that our results are limited by the type of scenarios we chose and the capabilities of the simulation platform, mainly the amount and frequency of the metabolic and behavioral variabilities.

These results are also limited by the chosen cost function in Eq. This algorithm may be combined with an insulin bolus optimization algorithm that optimize glycemic outcomes during the daytime period.

This paper introduces a novel algorithm to titrate long-acting insulin doses in individuals with T1D following MDI therapy and using CGM and SIP. With the quick rise of CGM use and the arrival of SIP, the need for such algorithms is warranted. Our proposed method did not require carbohydrate information, and a proof-of-concept in - silico study demonstrated that the method performs well in simulation, increasing time spent in the target range, while reducing exposure to hypoglycemia, hyperglycemia, and glycemic variability.

This algorithm will be evaluated as part of a decision support system in an upcoming clinical trial with people with T1D using MDI therapy NCT The raw data supporting the conclusions of this article will be made available by the corresponding author upon reasonable request, without undue reservation.

AE developed the algorithm, performed the in-silico experiments and drafted the manuscript. CF and MBD contributed to the theoretical development of the algorithm and reviewed the manuscript. All authors contributed to the article and approved the submitted version. MBD receives research support from Tandem Diabetes, Dexcom, Novo Nordisk, and Arecor paid to his institution.

MDB serves as a consultant for Tandem, Dexcom, Adocia, Air Liquide, and Roche. MBD received speaker fees from Tandem and Arecor. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

American Diabetes Association. Diagnosis and Classification of Diabetes Mellitus. Diabetes Care 37, no. Supplement 1:S81—

People injecting insulin are advised to alter their insulin doses depending on a variety of factors such as meal time carbohydrate load, pre injection blood glucose, planned or completed exercise and so on. These factors can be quite complex and the knowledge on how to do this safely and effectively is covered in courses run by the Diabetes Team BITES for people with Type 1 Diabetes and Insulin Skills workshops for those with Type 2 Diabetes.

If you wish to enrol in any of these courses please discuss this with a member of the Diabetes Team, The Diabetes Centre, York Hospital - For people starting insulin treatment, often the initial dose is a relatively small one and increased over the course of the ensuing few days and weeks.

Starting and adjusting basal insulin 1—3. Figure 1. Starting and adjusting basal insulin. Download View full size. Figure 2. Starting and adjusting pre-mixed biphasic. Figure 3. Guide to basal plus insulin intensification schedules.

Lau ANC, Tang T, Halapy H, Thorpe K, Yu CHY. Initiating insulin in patients with type 2 diabetes. CMAJ ; 7 — Wong J, Tabet E. This strategy keeps mealtime and basal insulin in balance, thus avoiding over-insulinization with either mealtime or basal insulin, either of which can lead to hypoglycemia.

For example, if patients are physically active, too much basal insulin onboard can lead to hypoglycemia during the day 39 , whereas too much mealtime insulin, particularly at the evening meal, increases the risk for overnight hypoglycemia Moreover, this algorithm means that patients do not have to wait for their clinician to make needed dose adjustments, thereby addressing the issue of therapeutic inertia and facilitating more timely achievement of optimal glycemic control.

However, it is important that clinicians and patients review the diary together to make sure patients understand what to do. Instructing patients in the use of the algorithm also creates opportunities for clinicians to learn more about initiating and adjusting mealtime insulin.

It should be noted that participants in our study based their insulin adjustments on glucose values obtained from traditional fingerstick BGM.

However, the algorithm can be easily applied to individuals who use CGM. In addition to eliminating the need for multiple daily fingersticks, CGM provides an additional level of safety by providing information about glucose trends, direction and velocity of changing glucose, and alerts that warn patients of current and impending adverse glycemic events.

Although BGM is the most common glucose testing method currently in use by people with type 2 diabetes, with growing positive clinical trial and real-world study results of CGM in individuals with insulin-treated type 2 diabetes, use of CGM within this population continues to expand.

Although the algorithm can be used regardless of glucose monitoring method BGM or CGM , an important consideration relevant to persistent treatment adherence and quality of life is the method used for insulin delivery. Although participants in both of our study groups achieved equally significant A1C reductions, patient-reported outcomes revealed that overall satisfaction with the insulin delivery system and satisfaction with ease of use were notably higher with the patch than with the insulin pen.

Differences in quality-of-life measures such as ability to dose without attracting attention, painless mealtime insulin delivery, ease of administration, and lifestyle flexibility also favored patch use. There was a significantly higher preference for using the patch device than the pen among study participants who used the patch for the full 44 weeks.

A higher preference was also reported by pen users who crossed over to patch use for only 4 weeks at week 44 Moreover, study clinicians also reported favorable ratings for the patch for all measures of preference. Specifically, Given the large and growing proportion of patients with type 2 diabetes who are not meeting their glycemic targets 28 — 32 , innovative approaches to initiating and titrating basal-plus-mealtime insulin therapy are needed.

When used in conjunction with a simplified insulin delivery technology such as a mealtime insulin patch device, this insulin algorithm may facilitate more frequent intensification of therapy and result in significant improvements in medication adherence, treatment satisfaction, patient quality of life, and clinical outcomes.

The authors thank Christopher G. Parkin, MS, of CGParkin Communications, Inc. Her employer the nonprofit HealthPartners Institute contracts for her services, and no personal income goes to her. His employer the nonprofit HealthPartners Institute contracts for his services, and no personal income goes to him.

has received medical consulting services from CeQur, Nevro Corp. No other potential conflicts of intertest relevant to this article were reported.

All of the authors conceived the presented idea, contributed to the writing of the manuscript, and made extensive comments, criticism, and revisions to the manuscript. All reviewed and approved the final version. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Sign In or Create an Account. Search Dropdown Menu. header search search input Search input auto suggest. filter your search All Content All Journals Clinical Diabetes. Advanced Search. User Tools Dropdown. Sign In. Skip Nav Destination Close navigation menu Article navigation.

Volume 40, Issue 4. Previous Article Next Article. Basal and Mealtime Insulin Titration Algorithm. Considerations for Implementing the Algorithm. Article Information.

Article Navigation. Practical Pointers October 14 A Safe and Simple Algorithm for Adding and Adjusting Mealtime Insulin to Basal-Only Therapy Mary L.

Johnson ; Mary L. Corresponding author: Mary L. Johnson, Mary. Johnson ParkNicollet. This Site. Google Scholar. Richard M. Bergenstal Brian L. Levy ; Brian L. Darlene M. Dreon Darlene M. Clin Diabetes ;40 4 — Get Permissions. toolbar search Search Dropdown Menu. toolbar search search input Search input auto suggest.

FIGURE 1. View large Download slide. Example showing how to calculate starting doses for basal and mealtime insulin. U, units. TABLE 1 Mealtime Insulin Adjustments According to Premeal Glucose and Meal Size.

Adjustment leaflet. Insulin Titration Working out the right amount of insulin to take can often be quite a daunting task. If you wish to enrol in any of these courses please discuss this with a member of the Diabetes Team, The Diabetes Centre, York Hospital - For people starting insulin treatment, often the initial dose is a relatively small one and increased over the course of the ensuing few days and weeks.

Daily insulin adjustment leaflet Adjustment leaflet. For our purposes, a linearized subcutaneous oral glucose minimal model SOGMM is employed 29 , augmented with a basal dose channel see Appendix A.

In equation 3 , the initial state X 0 at time of day t 0 can be determined by assuming that basal insulin doses were given each day at the same time t B , considering the other model inputs to be zero. In other words, X 0 is the steady state resulting from a train of Dirac basal inputs given at t B in the absence of any meals and boluses.

A closed form for X 0 is derived in Appendix A. The SOGMM model takes the meal input stored in U meal in the form of the amount of carbohydrates in the consumed meals.

Since the amount of carbohydrates is unknown, this input is artificially reconstructed following insulin dosing rules 30 :. where TDI is the total daily dose of insulin and G b is basal glucose. According to 6 , the input U meal can be nonzero only if the input U bolus is nonzero, thus including only instances of bolused carbohydrate intakes.

To include in the input vector unbolused meal events, U meal is further augmented by including meal inputs identified through an unbolused-meal detection algorithm 31 , Additionally, carbohydrate absorption rate parameters f,k q1 ,k q2 ,k q12 are estimated for each meal to account for inter-meal absorption differences e.

Population parameters are used for the remaining parameters. Parameters are estimated following a maximum-a-posteriori approach where the posterior probability of observing θ conditioned on the data 𝒟 is maximized.

Note that in this step the residual metabolic signal ω is not considered. The parameter vector is thus obtained as:.

The residual metabolic signal ω is estimated by regularized deconvolution via inversion of the individualized model outlined above. The deconvolution procedure is described by Patek et al.

The individualization procedure and the residual metabolic signal estimation are performed daily using an extended daily data a day padded by a 6-hour head and a 2- hour tail before and after the day. Using the identified model, the glucose trace in response to the basal dose, in the absence of meals and insulin boluses, can be predicted as:.

The basal dose can then be optimized to achieve a desired theoretical fasting glucose profile Y basal. The optimization problem can be written as follows:. If multiple days are analyzed simultaneously, B C opt is found for each day and then averaged over days. The final optimal dose is rounded to the nearest half unit, to accommodate SIP resolution.

This algorithm can be run routinely at night whenever the daily data is collected. Currently, most patients using MDI therapy do not own a SIP nor consistently use a CGM, even though these technologies are getting cheaper and more accessible 5 Long-acting insulin can still be titrated and adapted using SMBG values from a glucose meter.

SMBG is taken in fasting conditions, usually before the breakfast meal, and used to adjust the basal insulin dose. In the following, two existing strategies to adapt the basal insulin dose using SMBG records will be presented and compared against the proposed algorithm.

In order to keep a fair comparison between the algorithms, the same update rule, with the same saturation and dead-zone, reported in equation 1 is used but with the newly calculated B C opt.

Long-acting insulin dose can be titrated following a control-to-range heuristic rule. Rules are composed of a chosen range e.

Most titration algorithms used in clinical practice can be considered of this sort Here, an algorithm similar to the one proposed by Visentin et al. to titrate insulin glargine was used 37 , which defines the optimal insulin dose as:. This baseline algorithm is taken from recent work from Cescon et al.

where an iterative learning control ILC is proposed to optimize the basal insulin dose This ILC algorithm finds the new optimal dose B C opt to be administered such that the SMBG values are driven as close as possible to the desired reference trajectory G target :.

Similarly to 16 , the DC gain of the filter F q is individualized to each subject as 3 BW. The basal insulin dosing adaptation algorithm was tested in a day simulation study including virtual adult subjects with T1D using MDI therapy.

Daily basal insulin doses of glargine were administered either in the morning or before bedtime and were usually taken around mealtime, when possible i. The virtual subjects could treat hypoglycemia events, but a hypoglycemia unawareness algorithm was implemented where there is a chance that the hypoglycemia event is not treated for a period e.

Additionally, hypoglycemia events between 0 am at 6 am were not treated to reinforce night hypoglycemia. Each virtual subject was characterized by a basal glucose and the basal dose needed to achieve this basal glucose, U basal ss.

As reported in 27 , a model for glucose meter was used to generate SMBG values from blood glucose levels, and a model for a CGM sensor was used to generate glucose readings Consumed carbohydrate amount was predetermined in the simulator but unknown to the adaptation algorithms.

Five treatment arms were simulated: a a control arm where altered insulin dosing parameters were kept the same throughout the experiment CTR ; b a baseline arm where the control to range SMBG algorithm is employed SMBG-Rule ; c a baseline arm where the control to reference SMBG algorithm is employed SMBG-ILC ; d an experimental arm where the proposed algorithm is employed CGM-Opt ; e another experimental arm similar to d but where counted carbohydrate amount was given to the optimization algorithm instead of being reconstructed CGM-Opt-Carb.

Similar to clinical practice 39 , the insulin dose was titrated every three days to provide enough time for insulin dose change effects on fasting glucose to stabilize.

In the SMBG-Rule and SMBG-ILC arms, three pre-breakfast SMBG values were used. In the CGM-opt and CGM-Opt-Carb arm, the CGM data up to the time of the next basal dose recommendations around three days was used. The simulation experiment was not interrupted during the days. The same simulation was repeated for two different scenarios: nominal and variance , described below.

In this scenario, the virtual subjects consumed three similar meals each day at the same time. Meals were taken at 7AM, 1PM, 7PM and the amount of carbohydrates per meal was 50g, 75g and 75g.

Behavioral variability consisted of consuming three main meals and up to three unannounced and unbolused snacks. Meal sizes were variable, but the total carbohydrates consumed over the day were between g and g, main meals were bigger than 30g, and snacks were smaller than 40g e.

The insulin bolus could be delayed by up to 1 hour after consuming a meal. Metabolic variability was implemented by varying the insulin sensitivity during the day and between days. Metrics were calculated for each arm every 15 days.

Before starting the days experiment, an initial 15 days was simulated and used as a baseline. All results are reported as mean ± standard deviation across subjects for a day duration.

Changes of a certain day period as compared to the baseline period are reported as average and confidence interval CI.

At the same time, TBR was decreased using the CGM-Opt in both scenarios These results are summarized in Figure 1 for the nominal scenario and Figure 2 for the variance scenario. An exhaustive comparison between the four arms CTR, SMBG-Rule, SMBG-ILC, and CGM-Opt is provided in Table 1.

Figure 1 Summary of glycemic outcomes in every 15 days period for the nominal scenario of the in-silico experiment. Values are shown as mean and standard deviation. Figure 2 Summary of glycemic outcomes in every 15 days period for the variance scenario of the in-silico experiment.

The average basal dose in the last days was compared to the original U basal ss. In Table 2 , results of this comparison are shown separately for virtual subjects that started with a higher basal dose and with a lower basal dose. In Figure 3 , the percentage basal dose changes in both scenarios are shown.

Table 2 Summary of changes in basal dose from theoretical steady-state optimal value. Figure 3 Summary of long-acting dose changes in titration days every 3 days.

Values are shown as median and interquartile range. There was no clinically significant change in the glycemic metrics between the CGM-Opt arm and the CGM-Opt-Carb arm. Similarly, as it can be seen in Table 3 , there was no differences between calculated optimal basal dose B C opt for each day in the two arms CGM-Opt vs CGM-Opt-Carb.

Table 3 Summary of changes between the optimization procedure when the carbohydrate input is reconstructed or counted by virtual subjects.

People with T1D live with the life-long burden of making important decisions about their daily insulin doses.

Technological advances in diabetes treatment can help in easing this burden. Specifically, the new generation of SIP and the affordability of CGM are facilitating the development of a decision support system designed for people using MDI therapy.

The proposed algorithm will enable such decision support systems by automatically suggesting adaptation of the basal insulin dose after analyzing SIP and CGM records. Our algorithm is based on a metabolic model that describes the complex glucose traces by separating the effects of the basal insulin dose from other system inputs, i.

This approach is inspired by the clinical practice where patients are usually asked to skip meals in order to optimize their basal insulin Once the model is able to describe the data, we can mathematically eliminate the effect of meals and boluses on the glucose trace, thus isolating the effect of basal dose on the theoretical fasting glucose and allowing for its optimal tuning.

This method follows a similar insulin basal rate optimization approach described by Fabris et al. Estimating the residual metabolic signal is key to our approach since it detects changes in the glucose curve that are independent of delivered insulin boluses and consumed meals but needs to be controlled through the basal dose.

The original idea of a model-based residual metabolic signal estimated for insulin titration was introduced by Patek et al. and refined in other works 29 , 33 , Another similar model-based method was proposed previously by El Fathi et al. In this work, a recent subcutaneous absorption model of basal dose was employed Another difference is that the basal dose is optimized independently from other model parameters, giving the possibility to mold the cost function to enforce a desired outcome e.

To put the performance of this algorithm into perspective, we compared it with a control-to-range algorithm inspired by the current clinical practice and a control-to-reference algorithm that was recently proposed. Both algorithms use the current standard of titrating the long-acting insulin dose from the pre-breakfast SMBG measurement.

In general, results have shown that the proposed algorithm outperforms the other methods at night and can achieve comparable results overall.

This can be explained by multiple factors i with CGM, we can observe the full glucose profile, thus clearly detecting degradations in night control; ii we explicitly biased the optimization equation in 9 to reduce hypoglycemia events during the night; iii once the night period is optimized, we did not aim to optimize glycemic metrics in the day period by optimizing insulin boluses.

Our algorithm also reduced glycemic variability as measured by the glucose standard deviation in both scenarios Table 1. This is aligned with our observation in Table 2 that this algorithm can recover the theoretical steady-state basal dose, which in theory is the one that will cause the least variations in the glucose curve.

Furthermore, one can argue that reducing overall glycemic variability will facilitate optimizing parameters used to compute insulin boluses in the day period. Our simulations have shown that the control-to-range algorithm used in the clinical practice SMBG-Rule is effective in titrating the long-acting insulin dose by reducing both hypoglycemia and hyperglycemia.

Unexpectedly, the control-to-reference algorithm SMBG-ILC did not perform similarly in the two scenarios. In spite of showing good performance in the nominal scenario, the ILC-based algorithm was not able to reduce hypoglycemia in the variance scenario.

In Table 1 , we can see that the mean fasting SMBG values were driven close to as expected by the algorithm, but this was achieved at the cost of a higher hypoglycemia exposure. This suggests that individualizing the target of the ILC algorithm or stopping titration early for each subject may be necessary in clinical practice.

This simulation also hints that the ILC algorithm may benefit from the use of CGM data instead of SMBG. We have also shown that our algorithm is robust to carbohydrate information, as seen in Table 3.

This is a result of keeping meal parameters free to describe the glucose curve with the least a-priori knowledge during the model individualization. Therefore, the proposed meal reconstruction approach using the simple equation in 6 is shown to be sufficient for titration purposes. Not relying on carbohydrate counting will facilitate the use of this algorithm by T1D patients.

In Figure 3 , we can see that the algorithm converges in about 30 days after ~10 cycles. However, in the variance scenario, the algorithm continued to make small changes. The authors thank Christopher G. Parkin, MS, of CGParkin Communications, Inc.

Her employer the nonprofit HealthPartners Institute contracts for her services, and no personal income goes to her. His employer the nonprofit HealthPartners Institute contracts for his services, and no personal income goes to him.

has received medical consulting services from CeQur, Nevro Corp. No other potential conflicts of intertest relevant to this article were reported. All of the authors conceived the presented idea, contributed to the writing of the manuscript, and made extensive comments, criticism, and revisions to the manuscript.

All reviewed and approved the final version. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Sign In or Create an Account.

Search Dropdown Menu. header search search input Search input auto suggest. filter your search All Content All Journals Clinical Diabetes. Advanced Search. User Tools Dropdown. Sign In. Skip Nav Destination Close navigation menu Article navigation. Volume 40, Issue 4. Previous Article Next Article.

Basal and Mealtime Insulin Titration Algorithm. Considerations for Implementing the Algorithm. Article Information. Article Navigation. Practical Pointers October 14 A Safe and Simple Algorithm for Adding and Adjusting Mealtime Insulin to Basal-Only Therapy Mary L.

Johnson ; Mary L. Corresponding author: Mary L. Johnson, Mary. Johnson ParkNicollet. This Site. Google Scholar. Richard M. Bergenstal Brian L. Levy ; Brian L. Darlene M. Dreon Darlene M. Clin Diabetes ;40 4 — Get Permissions. toolbar search Search Dropdown Menu. toolbar search search input Search input auto suggest.

FIGURE 1. View large Download slide. Example showing how to calculate starting doses for basal and mealtime insulin. U, units. TABLE 1 Mealtime Insulin Adjustments According to Premeal Glucose and Meal Size. Adjustment for Premeal Glucose. View Large.

FIGURE 2. Example showing how to make daily mealtime dose adjustments. FIGURE 3. Example showing how to adjust basal insulin doses. TABLE 2 Basal Insulin Dose Adjustment Based on Morning Blood Glucose Pattern From Previous Week. Glucose Results Before Morning Meal or Upon Waking.

Bedtime Basal Insulin Dose Adjustment. TABLE 3 Weekly Starting Mealtime Dose Adjustments. Glucose Test Results Before Midday Meal. Morning Mealtime Dose Adjustment. FIGURE 4. Funding for the development of this manuscript was provided by CeQur. Induction of long-term glycemic control in newly diagnosed type 2 diabetic patients is associated with improvement of beta-cell function.

Search ADS. Induction of long-term normoglycemia without medication in Korean type 2 diabetes patients after continuous subcutaneous insulin infusion therapy. Effect of intensive insulin therapy on beta-cell function and glycaemic control in patients with newly diagnosed type 2 diabetes: a multicentre randomised parallel-group trial.