Patient Empowerment Through Predictive Personalised Decision Support(PEPPER)-Validation Study. (PEPPER)

Imperial College London, Oxford Brookes University, Universitat de Girona, Cellnovo Limited, Romsoft SRL

Patient Empowerment through Predictive PERsonalised Decision Support (PEPPER) is an European Union (EU) funded research project to develop a personalised clinical decision support system for Type 1 Diabetes Mellitus (T1DM) self-management. The tool provides insulin bolus dose advice, tailored to the needs of individuals. The system uses Case-Based Reasoning (CBR), an artificial intelligence methodology that adapts to new situations according to past experience. The PEPPER system also incorporates a safety module that promotes safety by providing glucose alarms, low-glucose insulin suspension, carbohydrate recommendations and fault detection. The principal research objective is to demonstrate safety, feasibility and usability of the PEPPER system compared to a standard bolus calculator.

Decision support algorithm: The decision support algorithm is based on case-based reasoning (CBR). CBR is an artificial intelligence technique that tries to solve newly encountered problems by applying the solutions learned from solved problems encountered in the past. A combination of parameters makes up the case problem presented to the CBR algorithm. Using this information the algorithm will find a similar case scenario from the pool of previously encountered cases and recommend an improved solution (insulin dose) to achieve the best possible outcome (postprandial blood glucose). Each new case includes information about the problem (e.g. capillary blood glucose, meal information etc), solution (recommended insulin dose) and outcome (postprandial blood glucose). PEPPER offers a dual architecture for both Multiple Daily Injections (MDI) and Continuous Subcutaneous Insulin Infusion (CSII) therapy. In both cases, the user periodically wears a continuous glucose monitor (CGM) and an activity monitor. Handset CSII version: The handset is a portable touch-screen device, which communicates directly with the insulin pump with PEPPER application running.Its primary function in PEPPER is to allow the user to precisely manage insulin therapy by accepting or rejecting bolus insulin dose recommendations, calculated by the CBR based decision support algorithm, via a graphical interface. In addition, it automatically records glucose levels and allows logging of food intake and other parameters. MDI version: This handset is a commercially available Smartphone with the insulin recommendation application running locally on a standard operating system such as iPhone Operating System (iOS) or Android. It has the same functionality as Handset in CSII version, except that is not connected to insulin pump and patients have to administrate insulin by insulin pens. Secure web server revision: The handset wirelessly reports the user's case history to the secure portal, a web site which allows the clinician to add new patients, and review the cases to decide which ones should be kept. A case comprises multiple parameters such as carbohydrate intake, BG reading, meal composition, physical activity and hormone cycle. Personal Health Record on the secure portal allows patients and clinicians to view and update selected components of the detailed history. Insulin delivery system: MDI: it will be provide insulin pens with 0,5 IU of insulin CSII: will use Cellnovo insulin pump Continuous glucose monitoring system The glucose sensor that will be used throughout the clinical studies is the Dexcom sensor (CE marked, manufactured by Dexcom). This current is proportional to the glucose concentration in interstitial fluid and is calibrated against blood glucose a minimum of 12-hourly. The Dexcom CGM data is automatically transmitted to a secure web-based server and the secure PEPPER web-portal. Participants will be able to see their CGM data at all times and this will be used continuously throughout the studies. Safety features: Low and high glucose alarms will be incorporated to alert the user when hypo- and hyperglycaemia is detected to enable the user to act accordingly to bring the glucose levels back to target range. Glucose prediction algorithm for hypoglycaemia prediction. The hypoglycaemia prediction algorithm will enable the system to automatically activate the low glucose suspension feature in pump participants (suspension of insulin delivery until glucose levels are within the target range) and/or trigger an adaptive carbohydrate adviser, which will recommend a personalised carbohydrate snack. Insulin safety constraints. Personalised maximum insulin dose thresholds will be incorporated to prevent overdosing on insulin. Fault detection. Insulin pumps and CGMs are well-established technologies, but faults in these devices (e.g. pump occlusion, loss of sensor sensitivity) may occur. A fault detection system will identify such faults and alert the user to recommend a corresponding action to revert to the normal state. Clinical study Recruitment. This is a multicentre study and recruiting for the clinical study will be undertaken in the diabetes clinics at the Institut d'Investigació Biomédica de Girona (IdIBGi) (Spain) and the Imperial College London (ICL) (UK) from registered research databases and from interested participants who contact us. Objective: To demonstrate safety, feasibility and usability of the PEPPER system compared to a standard bolus calculator. The standard bolus calculator system will include the standard RT-CGM low and high glucose alarms. Methodology: Randomised open-label cross-over study Population: 50 adults with T1DM (25 on MDI and 25 on CSII) Randomisation: Participants who fit the inclusion criteria will be randomized to PEPPER/Control or Control/PEPPER in a 1:1 ratio. The groups will be stratified by pump or MDI. Both groups will be using CGM. After 3 months each group will revert to their standard therapy for 4 weeks (wash-out period) and then crossover to the other group. Timescale: Each participant will be in the study for 7 months. It is anticipated that it will take 12 months to complete this phase. 2.1 Usability assessment during clinical trial Task-driven testing with users using standard metrics for effectiveness and satisfaction. Mixed methods will be used to collect usability and satisfaction data including: A. Structured validated and non-validated treatment satisfaction and acceptability questionnaires B. Semi-structured interviews C. Automatic collection of quantitative data from the PEPPER system on time taken to get insulin bolus advice for users and time taken to perform revisions by the clinical investigators. D. Audio recording E. Video recording -training session where the participant will be asked to perform tasks in PEPPER system. This process will be observed in order to determine common problems and misunderstandings as well as aspects that have worked successfully. The handset will be filmed during the training to see which aspect is being described. PEPPER case base revision. The PEPPER case base will be revised every 2 weeks throughout the intervention period (PEPPER with CBR) and this will be a semi-automatic process done remotely via the secure web-server. Any new cases will be approved by the study team prior to inclusion in the case base. This provides remote supervision and prevents any potential system faults despite safety measures. The participants will not need to attend the clinical research unit for the revisions. Usual care will be maintained for diabetes throughout the study. Support will be offered to any participants who have concerns about their diabetes management. Participants in both the intervention group and control group will have the opportunity to call a physician for medical support and an engineer for technical support 24 hours a day throughout the study. Statistics This is a 7-month randomised controlled cross-over pilot study to evaluate the safety and usability of the PEPPER system versus a standard bolus calculator. 50 participants (25 participants on MDI and 25 participants on CSII) will be included in the study. To allow for a 10% drop-out rate we will screen 55 participants. The primary outcome is percentage time in target and secondary outcomes as outlined in section Outcomes Measures. With 50 participants a 0.57 SD difference can be demonstrated as significant with α of 0.05 and 80% power (two-tailed). Based on a pilot study population mean (SD) % time in target (3.9-10mmol/l) of 61.6 (18.8) a 10.7 (=0.57x18.8) difference in % time in target can be demonstrated as significant between the intervention and control in this study. The sample size is comparable to other technology transfer studies, is a realistic number for recruitment and provides robust safety date of a new technology. The primary analysis in the randomized crossover study will follow the intention-to-treat principle. The change in the primary outcome between the two treatment arms will be compared using an Analysis of Variance (ANOVA). The ANOVA model will include period sequence and subject within sequence as fixed effects. Missing data will be imputed using Rubin's method if feasible. For all CGM outcomes, a multilevel model of repeated measures (MMRM) will be used. Similar analyses will be performed separately for daytime and nighttime. The CGM data will be pooled giving equal weight to each hour of the day. Safety analyses will include severe hypoglycemia, diabetic ketoacidosis, and all reported adverse events. Event rates will be computed per 100 person-years. The numbers of events will be compared between the two treatment groups using robust Poisson regression and the percentage of subjects with at least one event will be compared using Fisher's exact test. Treatment group comparisons for total daily insulin and weight, will be made using analysis of covariance models, adjusted for the corresponding baseline value. For each questionnaire, mean ± SD values or percentiles appropriate to the distribution will be given by randomization group for the total score and each subscale. Treatment group comparisons will be made using similar linear models as described above for the primary outcomes. The following tabulations will be performed according to treatment group without statistical testing: baseline demographics and clinical characteristics, flow chart accounting for all subjects for all visits, protocol deviations, device malfunctions and other reported device issues, bolus recommendations accepted, screen views, and App using time per day. Data During the course of the study visits some data will be stored on laptop computers, not connected to the Internet, for later statistical analysis. These data will be coded and non- identifiable. Laptop computers may be used during the visits for portability and convenience. At the end of each visit the anonymised data will be transferred immediately to a secure web-server (details below) and will be deleted from the laptop. Any identifiable participant data will be stored in a locked filing cabinet in a secure room in each investigation centre. Only clinical research team will have access to this participant identifiable data. 5.1 Confidentiality of data collected during interviews To ensure security, data obtained during the course of the interviews will be encrypted and stored securely, with access limited solely to the researchers. Data will be de-identified such that only the researchers will be able to link the data to the participant involved using reversible codes. This is done purely for the purpose of comparison and evaluation across the separate interviews. Any resulting publications using the data will not identify the participants, and any quotes will kept anonymous should participants consent to this. All data will be kept securely for a period of 10 years following completion of the project. All audio or video recordings will avoid the use of names or other easily identifiable statements. Any recordings that violate this will be edited to omit the statement and the original destroyed. Audio and video recordings will be kept for 10 years. All data will be stored in an anonymised form by using study numbers for identification of participants. A data monitoring committee will be convened to assess the data collected throughout the study. This committee will comprise of the Chief Investigator at each site, a lay member, an independent diabetologist and an independent scientist. 5.2 Electronic data storage on secure web-server Data security and privacy will be a priority whilst dealing with medical data such as that held in the PEPPER system. During the clinical trail anonymous clinical data will be entered and stored on a secure web-server. Anonymous data collected by the PEPPER handset (such as glucose, meal information, physical activity, alcohol, exercise) and the Dexcom CGM system will be automatically transmitted to the secure web-server. For this purpose, EU regulatory procedures (Directives 95/46/EC and 2002/58/EC) will be observed. Medical data will be stored and protected against non-authorised access; transmission of data will be secured; only authorised users will have access to services and stored data. Authentication will be required for application use and data synchronisation. PEPPER will operate according to standard interoperability guidelines (e.g. HL7), so that information can be exchanged seamlessly between the various components. Authorised users will include study team members from the PEPPER collaborators. Collaborators will only be able to view anonymous PEPPER handset data. The data generated by the study will be analysed by the collaborative PEPPER research team at their respective sites. The analysis will be on anonymised data which will be aggregated during joint meetings on either clinical site. Missing, unused, and spurious data will be assessed on an individual basis and may be ignored, withdrawn or the visit may be removed from the analysis with appropriate justification adjudicated by the Principal Investigator. Adverse Events (AEs) Reporting Procedures. All adverse events will be reported. Depending on the nature of the event the reporting procedures below will be followed. Any questions concerning adverse event reporting will be directed to the Chief Investigator in the first instance. Non serious AEs: All such events will be recorded. Serious Adverse Events (SAEs): An SAE form will be completed and faxed to the Chief Investigator within 24 hours. However, hospitalisations for elective treatment of a pre-existing condition do not need reporting as SAEs. Reports of related and unexpected SAEs will be submitted within 15 days of the Chief Investigator becoming aware of the event. The Chief Investigator will also notify the Sponsor of all SAEs, where in the opinion of the Chief Investigator, the event is: 'related', i.e. resulted from the administration of any of the research procedures; and 'unexpected', i.e. an event that is not listed in the protocol as an expected occurrence Local investigators will report any SAEs as required by their Local Research Ethics Committee, Sponsor and/or Research & Development Office.

PEPPER study is a randomized cross-over clinical trial with 50 participants. After 4 weeks of ruin-in period all subjects will be randomized to either control group or intervention group in 1:1 ratio. The groups will be stratified by pump or MDI. After 3 months, each group will revert to their standard therapy for 4 weeks (wash-out period) and then crossover to the other group. All participants will be using CGM during whole study.

Group with intervention applied (PEPPER system) during the first three months and then, after wash -out period, swap to control group (using standard bolus calculator) for the next 3 months.

Group without intervention applied (using standard bolus calculator) during the first three months and then, after wash -out period, swap to intervention group (using PEPPER system) for the next 3 months.

The PEPPER system offers insulin dosing advice that is highly adaptive to the insulin needs of individuals by using Case-Base Reasoning (CBR) approach.

Change in the percentage of time in glucose target range (glucose levels 3.9-10 mmol/l or 70 - 180 mg/dl)

Change in the number of episodes of post-prandial hypoglycaemia (glucose level below 3,9 mmol/L or 70 mg/dl) within 4 hours

Change in the number of episodes of post-prandial hypoglycaemia (glucose level below 3,9 mmol/L or 70 mg/dl) within 6 hours

Change in the post-prandial area under the curve (AUC) (< 3,9 mmol/l or <70 mg/dl) of glucose level at 4 hours

Change in the post-prandial area under the curve (AUC) (< 3,9 mmol/l or <70 mg/dl) of glucose level at 6 hours

Low blood glucose index (LBGI) is a parameter that quantifies the risk of glycaemic excursions in non-negative numbers

High blood glucose index (HBGI) is a parameter that quantifies the risk of glycaemic excursions in non-negative numbers

Change in glycaemic variability measured with continuous overall net glycemic action at 2 hours (CONGA-2)

Measurement of satisfaction of diabetes treatment using Diabetes Treatment Satisfaction Questionnaire (DTSQ)

The DTSQ s (status version) and DTSQ c (change version) contain eight items each, six of them (questions 1 and 4-8) measure the Treatment Satisfaction and questions 2 and 3, concerning Perceived Frequency of Hyperglycaemia ('Perceived Hyperglycaemia')/Perceived Frequency of Hypoglycaemia ('Perceived Hypoglycaemia') respectively, are treated separately from the satisfaction items and from each other. DTSQs scores range from 6 = very satisfied to 0 = very dissatisfied and DTSQc scores from +3 = much more satisfied now to -3 = much less satisfied now, with 0 (midpoint), representing no change.

Measurement of quality of life in patients with diabetes mellitus using Diabetes Quality of Life (DQOL) questionnaire

The instrument provides an overall scale score, as well as four subscale scores for satisfaction with treatment (with a range of 1-very satisfied to 5 -nothing satisfied) (overall results ranging from 15 to 75) impact of treatment (with a range of 1-never to 5 -always; minimum score 17, maximum 85) worry about social/vocational issues (with a range of 1-never to 5 -always; minimum score 7, maximum 35) worry about the future effects of diabetes (with a range of 1-never to 5 -always; minimum score 4, maximum 20)

The Problem in Areas In Diabetes (PAID) is a questionnaire with 20 items in which responders are asked to indicate the degree to which each of the items is currently a problem for them, from 0 (not a problem) to 4 (a serious problem). The points of the scores for each item are summed , then multiplied by 1.25 to generate a total score out of 100. A total score of 40 and above indicates severe diabetes distress. A minimum score of 0 indicates no diabetes-related distress. Individual items score 3 or 4: moderate to severe distress; to be discussed during the consultation following completion of the questionnaire.

The descriptive system comprises five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. The EQ VAS records the patient's self-rated health on a vertical visual analogue scale. This can be used as a quantitative measure of health outcome that reflects the patient's own judgement. The scores on these five dimensions can be presented as a health profile or can be converted to a single summary index number (utility) reflecting preferability compared to other health profiles.

Participants are asked to solve 12 tasks by using PEPPER application. The difficulty of the task is evaluated from 0 till 3: 0 = User completed task with zero difficulty. (Zero Frustration) = User completed task with only minor problem(s). (Little Frustration) = User completed task, but it required more effort/time/dead-ends than the user expected. (Medium/High Frustration) = User did not complete task. (Point of Failure)

Inclusion Criteria: Adults ≥18years of age Diagnosis of T1DM for > 1 year On MDI using a basal-bolus insulin regime or CSII (insulin pump) for at least 6 month Structured education done and good ability perform carbohydrates (CHO) counting HbA1c ≥ 48mmol/mol and ≤ 86mmol/mol Using insulin carbohydrates ratio (ICR) and insulin sensitivity factor (ISF) to calculate the mealtime bolus An understanding of and willingness to follow the protocol and sign the informed consent CBG measurements at least 2 times per day for calibration of the CGM Exclusion Criteria: Severe episode of hypoglycaemia (requiring 3rd party assistance) in the 6 months prior to enrolment Diabetic ketoacidosis in the last 6 months prior to enrolment Impaired awareness of hypoglycaemia (based on Gold score) Pregnancy, breastfeeding or intention of becoming pregnant over time of study procedures Enrolled in other clinical trials Have active malignancy or under investigation for malignancy Suspected or diagnosed endocrinopathy like adrenal insufficiency, unstable thyroidopathy, endocrine tumour Gastroparesis Autonomic neuropathy Macrovascular complications (acute coronary syndrome, transient ischaemic attack, cerebrovascular event within the last 12 months prior to enrolment in the study) Visual impairment including unstable proliferative retinopathy Reduced manual dexterity Inpatient psychiatric treatment Abnormal renal function test results (calculated Glomerular Filtration Rate (GFR) <40 mL/min/1.73m2) Liver cirrhosis Not tributary to optimization to insulin therapy Abuse of alcohol or recreational drugs Oral steroids Regular use of the acetaminophen, beta-blockers or any other medication that the investigator believes is a contraindication to the participant's participation. Participant withdrawal criteria: Loss of capacity to give informed consent The subject has a serious event related to study Cessation of MDI of insulin as usual care for T1DM Severe hypoglycaemia Diabetic ketoacidosis Positive pregnancy test Terminal illness Investigators initiated discontinuation of study due to participant or equipment concerns

Institut d'Investigació Biomèdica de Girona (IDIBGI) Unitat de Diabetis, Endocrinologia i Nutrició de Girona (UDENG) Hospital Universitari de Girona Dr Josep Trueta

Institut d'Investigació Biomèdica de Girona (IDIBGI) Unitat de Diabetis, Endocrinologia i Nutrició de Girona (UDENG) Hospital Universitari de Girona Dr Josep Trueta

Buckingham B, Wilson DM, Lecher T, Hanas R, Kaiserman K, Cameron F. Duration of nocturnal hypoglycemia before seizures. Diabetes Care. 2008 Nov;31(11):2110-2. doi: 10.2337/dc08-0863. Epub 2008 Aug 11.

Perez-Gandia C, Facchinetti A, Sparacino G, Cobelli C, Gomez EJ, Rigla M, de Leiva A, Hernando ME. Artificial neural network algorithm for online glucose prediction from continuous glucose monitoring. Diabetes Technol Ther. 2010 Jan;12(1):81-8. doi: 10.1089/dia.2009.0076.

Diabetes Control and Complications Trial Research Group; Nathan DM, Genuth S, Lachin J, Cleary P, Crofford O, Davis M, Rand L, Siebert C. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med. 1993 Sep 30;329(14):977-86. doi: 10.1056/NEJM199309303291401.

Dassau E, Cameron F, Lee H, Bequette BW, Zisser H, Jovanovic L, Chase HP, Wilson DM, Buckingham BA, Doyle FJ 3rd. Real-Time hypoglycemia prediction suite using continuous glucose monitoring: a safety net for the artificial pancreas. Diabetes Care. 2010 Jun;33(6):1249-54. doi: 10.2337/dc09-1487.

FDA Advisory Panel Votes to Recommend Non-Adjunctive Use of Dexcom G5 Mobile CGM. Diabetes Technol Ther. 2016 Aug;18(8):512-6. doi: 10.1089/dia.2016.07252.mr. Epub 2016 Jul 29. No abstract available.

Garg SK, Bookout TR, McFann KK, Kelly WC, Beatson C, Ellis SL, Gutin RS, Gottlieb PA. Improved glycemic control in intensively treated adult subjects with type 1 diabetes using insulin guidance software. Diabetes Technol Ther. 2008 Oct;10(5):369-75. doi: 10.1089/dia.2007.0303.

National Clinical Guideline Centre (UK). Type 1 Diabetes in Adults: Diagnosis and Management. London: National Institute for Health and Care Excellence (NICE); 2015 Aug. Available from http://www.ncbi.nlm.nih.gov/books/NBK315808/

Herrero P, Calm R, Vehi J, Armengol J, Georgiou P, Oliver N, Tomazou C. Robust fault detection system for insulin pump therapy using continuous glucose monitoring. J Diabetes Sci Technol. 2012 Sep 1;6(5):1131-41. doi: 10.1177/193229681200600518.

Herrero P, Pesl P, Reddy M, Oliver N, Georgiou P, Toumazou C. Advanced Insulin Bolus Advisor Based on Run-To-Run Control and Case-Based Reasoning. IEEE J Biomed Health Inform. 2015 May;19(3):1087-96. doi: 10.1109/JBHI.2014.2331896.

Herrero P, Pesl P, Bondia J, Reddy M, Oliver N, Georgiou P, Toumazou C. Method for automatic adjustment of an insulin bolus calculator: in silico robustness evaluation under intra-day variability. Comput Methods Programs Biomed. 2015 Apr;119(1):1-8. doi: 10.1016/j.cmpb.2015.02.003. Epub 2015 Feb 16.

Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study Group; Bode B, Beck RW, Xing D, Gilliam L, Hirsch I, Kollman C, Laffel L, Ruedy KJ, Tamborlane WV, Weinzimer S, Wolpert H. Sustained benefit of continuous glucose monitoring on A1C, glucose profiles, and hypoglycemia in adults with type 1 diabetes. Diabetes Care. 2009 Nov;32(11):2047-9. doi: 10.2337/dc09-0846. Epub 2009 Aug 12.

Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study Group; Tamborlane WV, Beck RW, Bode BW, Buckingham B, Chase HP, Clemons R, Fiallo-Scharer R, Fox LA, Gilliam LK, Hirsch IB, Huang ES, Kollman C, Kowalski AJ, Laffel L, Lawrence JM, Lee J, Mauras N, O'Grady M, Ruedy KJ, Tansey M, Tsalikian E, Weinzimer S, Wilson DM, Wolpert H, Wysocki T, Xing D. Continuous glucose monitoring and intensive treatment of type 1 diabetes. N Engl J Med. 2008 Oct 2;359(14):1464-76. doi: 10.1056/NEJMoa0805017. Epub 2008 Sep 8.

Kovatchev BP, Breton M, Man CD, Cobelli C. In silico preclinical trials: a proof of concept in closed-loop control of type 1 diabetes. J Diabetes Sci Technol. 2009 Jan;3(1):44-55. doi: 10.1177/193229680900300106.

Lepore G, Dodesini AR, Nosari I, Scaranna C, Corsi A, Trevisan R. Bolus calculator improves long-term metabolic control and reduces glucose variability in pump-treated patients with Type 1 diabetes. Nutr Metab Cardiovasc Dis. 2012 Aug;22(8):e15-6. doi: 10.1016/j.numecd.2012.04.001. Epub 2012 Jun 4. No abstract available.

Reddy M, Pesl P, Xenou M, Toumazou C, Johnston D, Georgiou P, Herrero P, Oliver N. Clinical Safety and Feasibility of the Advanced Bolus Calculator for Type 1 Diabetes Based on Case-Based Reasoning: A 6-Week Nonrandomized Single-Arm Pilot Study. Diabetes Technol Ther. 2016 Aug;18(8):487-93. doi: 10.1089/dia.2015.0413. Epub 2016 May 19.

Reddy M, Rilstone S, Cooper P, Oliver NS. Type 1 diabetes in adults: supporting self management. BMJ. 2016 Mar 10;352:i998. doi: 10.1136/bmj.i998. No abstract available.

Vaddiraju S, Burgess DJ, Tomazos I, Jain FC, Papadimitrakopoulos F. Technologies for continuous glucose monitoring: current problems and future promises. J Diabetes Sci Technol. 2010 Nov 1;4(6):1540-62. doi: 10.1177/193229681000400632.

Sovik O, Thordarson H. Dead-in-bed syndrome in young diabetic patients. Diabetes Care. 1999 Mar;22 Suppl 2:B40-2. Erratum In: Diabetes Care 1999 Aug;22(8):1389.

Zisser H, Robinson L, Bevier W, Dassau E, Ellingsen C, Doyle FJ, Jovanovic L. Bolus calculator: a review of four "smart" insulin pumps. Diabetes Technol Ther. 2008 Dec;10(6):441-4. doi: 10.1089/dia.2007.0284.

Avari P, Leal Y, Herrero P, Wos M, Jugnee N, Arnoriaga-Rodriguez M, Thomas M, Liu C, Massana Q, Lopez B, Nita L, Martin C, Fernandez-Real JM, Oliver N, Fernandez-Balsells M, Reddy M. Safety and Feasibility of the PEPPER Adaptive Bolus Advisor and Safety System: A Randomized Control Study. Diabetes Technol Ther. 2021 Mar;23(3):175-186. doi: 10.1089/dia.2020.0301.