Decision Support Systems and Closed‐Loop
2022; Mary Ann Liebert, Inc.; Volume: 24; Issue: S1 Linguagem: Inglês
10.1089/dia.2022.2504
ISSN1557-8593
AutoresRevital Nimri, Moshe Phillip, Boris Kovatchev,
Tópico(s)Diabetes and associated disorders
ResumoDiabetes Technology & TherapeuticsVol. 24, No. S1 Original ArticlesFree AccessDecision Support Systems and Closed‐LoopRevital Nimri, Moshe Phillip, and Boris KovatchevRevital NimriDiabetes Technology Center, Jesse Z and Sara Lea Shafer Institute for Endocrinology and Diabetes, Schneider Children's Medical Center of Israel, Petah Tikva, IsraelSackler Faculty of Medicine, Tel Aviv University, Tel Aviv, IsraelSearch for more papers by this author, Moshe PhillipDiabetes Technology Center, Jesse Z and Sara Lea Shafer Institute for Endocrinology and Diabetes, Schneider Children's Medical Center of Israel, Petah Tikva, IsraelSackler Faculty of Medicine, Tel Aviv University, Tel Aviv, IsraelSearch for more papers by this author, and Boris KovatchevUniversity of Virginia Center for Diabetes Technology, University of Virginia School of Medicine, Charlottesville, VASearch for more papers by this authorPublished Online:25 Apr 2022https://doi.org/10.1089/dia.2022.2504AboutSectionsPDF/EPUB Permissions & CitationsPermissionsDownload CitationsTrack CitationsAdd to favorites Back To Publication ShareShare onFacebookTwitterLinked InRedditEmail IntroductionThe introduction of continuous glucose monitoring (CGM) more than two decades ago prompted a gradual but significant paradigm shift in the treatment of diabetes (1,2). A number of studies documented the benefits of CGM (3–6) and predicted its future as a component of decision support (DS) and closed‐loop (CL) systems (7–10). The advantage of CGM is the frequent (e.g., every 5 minutes) estimation of glucose levels, which generates large amounts of data and allows accurate reconstruction of the time course of glycemic fluctuations of a person. The “curse” of CGM is the frequent estimation of glucose levels, which generates large amounts of data that are difficult to interpret without clever DS or CL systems (11,12).Regularly published reviews explain how DS systems digest in a meaningful way the vast amounts of CGM and insulin delivery information prior to serving it to patients and their healthcare providers, and provide evidence for their efficacy (13–15). While the classic CGM data interpretation tool is the Ambulatory Glucose Profile (AGP) report incorporated in the data reports of several CGM devices (16), DS are beginning to include new data science methods, such as CGM pattern classification and data‐driven methods for prediction of blood glucose dynamics (17,18), classification of therapeutic strategies (19), as well as artificial intelligence (AI) and advanced machine‐learning applications (20,21). A multitude of Smartphone apps connect to glucose monitoring devices and offer actionable data interpretation, as well as advice regarding diet, exercise, bolus calculation, or insulin dosing. The choices are many and the DS field continues to grow rapidly.The first hybrid CL system approved in 2016 for clinical use was the Medtronic 670G, which regulated automatically the pump's basal rate but did not administer insulin boluses. The state of CL progress prior to 2019 was documented by systematic reviews and meta‐analyses by high‐impact journals, including Lancet Diabetes Endocrinology (22), Nature Reviews Endocrinology (23), British Medical Journal (24), and Metabolism (25), which confirmed the efficacy of the CL for diabetes control (22,24,25). Thereafter, several longer‐term randomized trials were reported, assessing the efficacy of CL systems over 3 months, or more: (i) a 3‐month trial of CL implemented on a mobile phone and compared to sensor-augmented pump (SAP) resulted in 4.8% improvement in the time within 70–180 mg/dL and 1.7% improvement in the time < 70 mg/dL (26); (ii) a 12‐week trial reported a 10.8% improvement in the time within 70–180 mg/dL and 0.8% improvement in the time < 70 mg/dL (27); (iii) another 12‐week trial reported 9.2% and 2.4% improvement, respectively, but with five episodes of severe hypoglycemia in the CLC group (28), and (iv) the 6‐month multicenter randomized Protocol 3 of the International Diabetes Closed‐Loop (iDCL) Trial, which compared Control‐IQ to SAP and reported 11% increase in the time in the target range of 70–180 mg/dL and simultaneous reduction in the time < 70 mg/dL by 0.9% (29), without occurrence of severe hypoglycemia. This study led to the FDA approval of Control‐IQ for clinical use in 2019 (30).Today, after years of development and testing of system components and algorithms, closed‐loop control of diabetes, known as the “artificial pancreas,” is a clinical reality. Two CL systems—Medtronic's 670G/770G and Tandem's Control‐IQ—have FDA clearance for clinical use in the United States and CE mark for clinical use in Europe. Another two systems—Medtronic 780G and CamAPS FX—received CE mark for use in European countries. These systems are at different stages of their clinical implementation: while 670G/770G and Control‐IQ already have hundreds of thousands of users around the world, 780G and CamAPS FX are making their first strides in real‐life application. Several other systems have passed extensive testing and are along their ways to regulatory approval, including Omnipod 5, Diabeloop, Tidepool Loop, and iLet (in two configurations—insulin only and insulin plus glucagon, the latter still in pilot‐feasibility stages). With the increasing clinical use of CL systems, real‐life data began to emerge, specifically from the use of the first CL—the MiniMed 670G (31–36), revealing improved glycemic control and quality of life (33,34), but also frequent discontinuation of system use due to suboptimal user experience (35–36).In this article, we will cover advanced DS systems, including some that are based on cutting‐edge AI, large‐scale CL pivotal trials and real‐life data, as well as small studies testing faster‐insulin or multi‐hormone CL systems.Key Articles Reviewed for the ArticleInsulin Dose Optimization Using an Automated Artificial Intelligence‐Based Decision Support System in Youths with Type 1 DiabetesNimri R, Battelino T, Laffel LM, Slover RH, Schatz D, Weinzimer SA, Dovc K, Danne T, Phillip M, NextDREAM ConsortiumNat Med 2020;26: 1380–1384An Artificial Intelligence Decision Support System for the Management of Type 1 DiabetesTyler NS, Mosquera‐Lopez CM, Wilson LM, Dodier RH, Branigan D, Gabo VB, Guillot FH, Hilts WW, El Youssef J, Castle JR, Jacobs PGNat Metab 2020;2: 612–619DIABEO System Combining a Mobile App Software with and Without Telemonitoring Versus Standard Care: A Randomized Controlled Trial in Diabetes Patients Poorly Controlled with a Basal‐Bolus Insulin RegimenFranc S, Hanaire H, Benhamou PY, Schaepelynck P, Catargi B, Farret A, Fontaine P, Guerci B, Reznik Y, Jeandidier N, Penfornis A, Borot S, Chaillous L, Serusclat P, Kherbachi Y, d'Orsay G, Detournay B, Simon P, Charpentier GDiabetes Technol Ther 2020;22: 904–911Validation of a Diagnostic Support System for Diabetic Retinopathy Based on Clinical ParametersRomero‐Aroca P, Verges‐Pujol R, Santos‐Blanco E, Maarof N, Valls A, Mundet X, Moreno A, Galindo L, Baget‐Bernaldiz MTrans Vis Sci Tech 2021;10: 17A Randomized Trial of Closed‐Loop Control in Children with Type 1 DiabetesBreton MD, Kanapka LG, Beck RW, Ekhlaspour L, Forlenza GP, Cengiz E, Schoelwer M, Ruedy KJ, Jost E, Carria L, Emory E, Hsu LJ, Oliveri M, Kollman CC, Dokken BB, Weinzimer SA, DeBoer MD, Buckingham BA, Cherñavvsky D, Wadwa RP for the iDCL Trial Research GroupN Engl J Med 2020;383: 836–845A Comparison of Two Hybrid Closed‐Loop Systems in Adolescents and Young Adults with Type 1 Diabetes (FLAIR): A Multicentre, Randomised, Crossover TrialBergenstal RM, Nimri R, Beck RW, Criego A, Laffel L, Schatz D, Battelino T, Danne T, Weinzimer SA, Sibayan J, Johnson ML, Bailey RJ, Calhoun P, Carlson A, Isganaitis E, Bello R, Albanese‐O'Neill A, Dovc K, Biester T, Weyman K, Hood K, Phillip M for the FLAIR Study GroupLancet 2021;397: 208–219One Year Real‐World Use of the Control‐IQ Advanced Hybrid Closed‐Loop TechnologyBreton MD, Kovatchev BPDiabetes Technol Ther 2021;23: 601–608Hybrid Closed‐Loop Glucose Control with Faster Insulin Aspart Compared with Standard Insulin Aspart in Adults with Type 1 Diabetes: A Double‐Blind, Multicentre, Multinational, Randomized, Crossover StudyBoughton CK, Hartnell S, Thabit H, Poettler T, Herzig D, Wilinska ME, Ashcroft NL, Sibayan J, Cohen N, Calhoun P, Bally L, Mader JK, Evans M, Leelarathna L, Hovorka RDiabetes Obes Metab 2021;23: 1389–1396Performance of the Insulin‐Only iLet Bionic Pancreas and the Bihormonal iLet Using Dasiglucagon in Adults with Type 1 Diabetes in a Home‐Use SettingCastellanos LE, Balliro CA, Sherwood JS, Jafri R, Hillard MA, Greaux E, Selagamsetty R, Zheng H, El‐Khatib FH, Damiano ER, Russell SJDiabetes Care 2021;44: e118–e120Reducing the Need for Carbohydrate Counting in Type 1 Diabetes Using Closed‐Loop Automated Insulin Delivery (Artificial Pancreas) and Empagliflozin: A Randomized, Controlled, Non‐Inferiority, Crossover Pilot TrialHaidar A, Yale JF, Lovblom LE, Cardinez N, Orszag A, Falappa CM, Gouchie‐Provencher N, Tsoukas MA, El Fathi A, Rene J, Eldelekli D, Lanctôt SO, Scarr D, Perkins BADiabetes Obes Metab 2021;23: 1272–1281Add‐On Therapy with Dapagliflozin Under Full Closed Loop Control Improves Time in Range in Adolescents and Young Adults with Type 1 Diabetes: The DAPAdream StudyBiester T, Muller I, von dem Berge T, Atlas E, Nimri R, Phillip M, Battelino T, Bratina N, Dovc K, Scheerer MF, Kordonouri O, Danne TDiabetes Obes Metab 2021;23: 599–608Fully Closed Loop Glucose Control with a Bihormonal Artificial Pancreas in Adults with Type 1 Diabetes: An Outpatient, Randomized, Crossover TrialBlauw H, Onvlee A, Klaassen M, van Bon AC, DeVries JHDiabetes Care 2021;44: 836–838Anticipation of Historical Exercise Patterns by a Novel Artificial Pancreas System Reduces Hypoglycemia During and After Moderate‐Intensity Physical Activity in People with Type 1 DiabetesGarcia‐Tirado J, Brown SA, Laichuthai N, Colmegna P, Koravi CLK, Ozaslan B, Corbett JP, Barnett CL, Pajewski M, Oliveri MC, Myers H, Breton MDDiabetes Technol Ther 2021;23: 277–285Dual‐Hormone Closed‐Loop System Using a Liquid Stable Glucagon Formulation Versus Insulin‐Only Closed‐Loop System Compared with a Predictive Low Glucose Suspend System: An Open‐Label, Outpatient, Single-Center, Crossover, Randomized Controlled TrialWilson LM, Jacobs PG, Ramsey KL, Resalat N, Reddy R, Branigan D, Leitschuh J, Gabo V, Guillot F, Senf B, El Youssef J, Steineck IIK, Tyler NS, Castle JRDiabetes Care 2020;43: 2721–2729Multicenter Trial of a Tubeless, On‐Body Automated Insulin Delivery System with Customizable Glycemic Targets in Pediatric and Adult Participants with Type 1 DiabetesBrown SA, Forlenza GP, Bode BW, Pinsker JE, Levy CJ, Criego AB, Hansen DW, Hirsch IB, Carlson AL, Bergenstal RM, Sherr JL, Mehta SN, Laffel LM, Shah VN, Bhargava A, Weinstock RS, MacLeish SA, DeSalvo DJ, Jones TC, Aleppo G, Buckingham BA, Ly TT for the Omnipod 5 Research GroupDiabetes Care 2021;44: 1630–1640Insulin Dose Optimization Using an Automated Artificial Intelligence‐Based Decision Support System in Youths with Type 1 DiabetesNimri R1, Battelino T2, Laffel LM3, Slover RH4, Schatz D5, Weinzimer SA6, Dovc K2, Danne T7, Phillip M1,8, NextDREAM Consortium1The Jesse Z and Sara Lea Shafer Institute for Endocrinology and Diabetes, National Center for Childhood Diabetes, Schneider Children's Medical Center of Israel, Petah Tikva, Israel; 2Department of Endocrinology, Diabetes and Metabolic Diseases, UMC‐University Children's Hospital Ljubljana, and Faculty of Medicine, University of Ljubljana, Ljubljana, Slovenia; 3Joslin Diabetes Center, One Joslin Place, Harvard Medical School, Boston, MA; 4Barbara Davis Center for Childhood Diabetes, University of Colorado Anschutz Medical Campus, Aurora, CO; 5Department of Pediatrics, College of Medicine, University of Florida, Gainesville, FL; 6Pediatric Endocrinology & Diabetes, Yale School of Medicine, New Haven, CT; 7Diabetes Center for Children and Adolescents, Children's Hospital AUF DER BULT, Hannover, Germany; 8Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, IsraelNat Med 2020; 26: 1380–1384This manuscript is also discussed in article on Diabetes Technology and Therapy in the Pediatric Age Group, page XXAbstractDespite the increasing adoption of insulin pumps and continuous glucose monitoring devices, the majority of people with type 1 diabetes do not achieve their glycemic goals. It is possible that this is related to a lack of expertise or inadequate time for clinicians to analyze complex sensor‐augmented pump data. The authors tested whether frequent insulin dose adjustments guided by an automated artificial intelligence–based decision support system (AI‐DSS) is as effective and safe as those guided by physicians in controlling glucose levels. ADVICE4U was a 6‐month, multicenter, multinational, parallel, randomized controlled, noninferiority trial in 108 participants with type 1 diabetes, aged 10–21 years and using insulin pump therapy. Participants were randomized 1:1 to receive remote insulin dose adjustment every 3 weeks guided by either an AI‐DSS, (AI‐DSS arm, n=54) or by physicians (physician arm, n=54). The results for the primary efficacy measure, which was the percentage of time spent within the target glucose range (70–180 mg/dL [3.9–10.0 mmol/L]) in the AI‐DSS arm, were statistically noninferior to those in the physician arm (50.2±11.1% versus 51.6±11.3%, respectively, P<1×10–7). The percentage of readings below 54 mg/dL (<3.0 mmol/L) within the AI‐DSS arm was statistically noninferior to that in the physician arm (1.3±1.4% versus 1.0±0.9%, respectively, P<0.0001). Within the physician arm, three severe adverse events related to diabetes (two severe hypoglycemia, one diabetic ketoacidosis) were reported, while none were reported in the AI‐DSS arm.The authors concluded that the use of an automated decision support tool for optimizing insulin pump settings was noninferior to intensive insulin titration provided by physicians from specialized academic diabetes centers.CommentIn the present manuscript, the authors report the result of a 6‐month study designed to test an artificial intelligence decision support system (AI‐DSS) developed to help healthcare providers (HCPs) coping with the need to optimize insulin dosing of people with type 1 diabetes using pumps and continuous glucose sensor. The algorithm ADVICE4U was tested in a multicenter, multinational prospective randomized trial with four centers in the United States, two centers in Europe, and one center in Israel. A challenging age group of 108 youths aged 10–21 were enrolled to the study and were divided into two groups. Participants were randomized 1:1 to receive remote insulin dose adjustment every 3 weeks and were guided by either an AI‐DSS or by physicians. The primary efficacy end point was the percentage of time spent within the target glucose range (70–180 mg/dL). The percent of time in the desired range in the ADVICE4U arm was not inferior to that of the physician arm. The authors also used the percent of time below 54 mg/dL as a safety parameter, and again the ADVIC4U arm was not inferior to the physician arm.The finding that intensive titration achieved by AI‐DSS is not inferior to that achieved by a physician from a specialized academic center might have important implications. Facing the reality of a shortage of expert endocrinologists around the globe and the constant increment in the number of people with diabetes, AI‐DSS algorithms might be able to help HCPs who are not in academic centers and primary care physicians to intensively treat people with diabetes and reach the level of care similar to that of physicians in specialized academic diabetes centers. Indeed, such studies should be conducted in primary care facilities with insulin‐treated patients who are also using multiple daily injection (MDI) and sensors, and MDI and self-monitoring of blood glucose (SMBG).An Artificial Intelligence Decision Support System for the Management of Type 1 DiabetesTyler NS1, Mosquera‐Lopez CM1, Wilson LM2, Dodier RH1, Branigan D2, Gabo VB2, Guillot FH2, Hilts WW1, El Youssef J2, Castle JR2, Jacobs PG11Artificial Intelligence for Medical Systems (AIMS) Lab, Oregon Health & Science University, Portland, OR; 2Harold Schnitzer Diabetes Health Center, Oregon Health & Science University, Portland, ORNat Metab 2020;2: 612–619AbstractType 1 diabetes (T1D) is characterized by pancreatic beta cell dysfunction and insulin depletion. Multiple injections of long‐acting basal and short‐acting bolus insulin, so‐called multiple daily injections (MDI), are used by over 40% of people with T1D to manage their glucose. Errors in dosing can lead to life‐threatening hypoglycemia events ( 180 mg/dL–1), increasing the risk of retinopathy, neuropathy, and nephropathy. Machine‐learning (artificial intelligence) approaches are being harnessed to incorporate decision support into many medical specialties. In this study, the authors report an algorithm that provides weekly insulin dosage recommendations to adults with T1D using MDI therapy. They employ a unique virtual platform to generate over 50,000 glucose observations to train a k‐nearest neighbors decision support system (KNN‐DSS) to identify causes of hyperglycemia or hypoglycemia and determine necessary insulin adjustments from a set of 12 potential recommendations. The KNN‐DSS algorithm achieves an overall agreement with board‐certified endocrinologists of 67.9% when validated on real‐world human data, and delivers safe recommendations, per endocrinologist review. A comparison of inter‐physician‐recommended adjustments to insulin pump therapy indicates full agreement of 41.2% among endocrinologists, which is consistent with previous measures of inter‐physician agreement (41%–45%). After 12 weeks of KNN‐DSS use, a substantial improvement in glycemic outcomes was indicated, with evaluation of artificial pancreas technologies performed with in silico benchmarking using a platform accepted by the U.S. Food and Drug Administration. The data indicate that the KNN‐DSS allows for early identification of dangerous insulin regimens and may be used to improve glycemic outcomes and prevent life‐threatening complications in people with T1D.CommentIn the present manuscript, Tyler and colleagues have described an artificial intelligence- (AI-) based decision support system (DSS) they have developed to treat adults with type 1 diabetes who use multiple daily injections and continuous glucose sensors. They developed an algorithm (k‐nearest neighbors decision support system [KMM‐DSS]) to identify causes of hyperglycemia or hypoglycemia and determine necessary insulin adjustments from a set of 12 potential recommendations on a weekly basis. They report that they achieved an overall agreement with a board‐certified endocrinologist of 67.9%, and delivered safe recommendations per endocrinologist review. In silico benchmarking indicates substantial improvement in glycemic outcome after 12 weeks of the use of their algorithm.With the lack of expert endocrinologists to cope with the constant increment in the number of people with diabetes around the globe, the need for AI‐based DSS that will serve both healthcare providers and people with type 1 and type 2 diabetes is obvious. Many apps with sophisticated algorithms have been produced over the last few years to try and fill in that “need.” However, most of them were never clinically tested.Despite the fact that the present study is important, it has several limitations. The KMM‐DSS was validated by only a single endocrinologist of a single center and was not yet clinically tested. The KMM‐DSS seems promising and indeed deserves to be tested in a prospective randomized clinical study of appropriate duration. The study should recruit people with diabetes (PwD) using MDI and sensors, and possibly also patients using MDI and SMBG, and should compare an arm of PwD who are getting their advice weekly from the algorithm to an arm of PwD who are getting their advice from an expert physician. Time in range and time below 54 mg/dL should be the end points and the study should be powered to be able to answer the questions: Can that specific algorithm be used safely? Is it effective? Many more questions related to patient outcome will also need to be answered.DIABEO System Combining a Mobile App Software with and without Telemonitoring Versus Standard Care: A Randomized Controlled Trial in Diabetes Patients Poorly Controlled with a Basal‐Bolus Insulin RegimenFranc S1, Hanaire H2, Benhamou PY3, Schaepelynck P4, Catargi B5, Farret A6, Fontaine P7, Guerci B8, Reznik Y9, Jeandidier N10, Penfornis A2,11, Borot S12, Chaillous L13, Serusclat P14, Kherbachi Y15, d'Orsay G16, Detournay B17, Simon P18, Charpentier G11Department of Diabetes, Sud‐Francilien Hospital, Corbeil‐Essonnes, and Centre d'étude et de Recherche pour l'Intensification du Traitement du Diabète (CERITD), Evry, France; 2Department of Diabetology, Metabolic Diseases and Nutrition, CHU Toulouse, University of Toulouse, Toulouse, France; 3Department of Diabetology, Pôle DigiDune, University Hospital, Grenoble, France; 4Department of Nutrition‐Endocrinology‐Metabolic Disorders, Marseille University Hospital, Sainte Marguerite Hospital, Marseille, France; 5Department of Endocrinology and Diabetes, University Hospital, Bordeaux, France; 6Department of Endocrinology, Diabetes and Nutrition, University Hospital, Montpellier, France; 7Department of Diabetology, University Hospital, Lille, France; 8Endocrinology‐Diabetes Care Unit, University of Lorraine, Vandoeuvre Lès Nancy, France; 9Department of Endocrinology, University of Caen Côte de Nacre Regional Hospital Center, Caen, France; 10Department of Endocrinology, Diabetes and Nutrition, CHU of Strasbourg, Strasbourg, France; 11University Paris‐Sud, University Paris‐Sud, Orsay, France; 12Centre Hospitalier Universitaire Jean Minjoz, Service d'Endocrinologie‐Métabolisme et Diabétologie‐Nutrition, Besançon, France; 13CHU de Nantes–Hospital Laennec, Saint‐Herblain, France; 14Endocrinology, Diabetology and Nutrition, Clinique Portes du Sud, Venissieux, France; 15Sanofi‐Diabetes, Gentilly, France; 16Voluntis, Suresnes, France; 17CEMKA‐EVAL, Bourg‐la‐Reine, France; 18National Association of Telemedicine, Evry, FranceDiabetes Technol Ther 2020;22: 904–911This manuscript is also discussed in article on Using Digital Health Technology to Prevent and Treat Disease, page XXBackgroundThe DIABEO system (DS) is a telemedicine solution that combines a mobile app for patients with a web portal for healthcare providers. DS allows real‐time monitoring of basal‐bolus insulin therapy as well as therapeutic decision‐making, integrating both basal and bolus dose calculation. Real‐life studies have shown a very low rate of use of mobile health applications by patients. Therefore, we conducted a large randomized controlled trial study to investigate the efficacy of DS in conditions close to real life (TELESAGE study).MethodsTELESAGE was a multicenter, randomized, open study with three parallel arms: arm 1 (standard care), arm 2 (DIABEO alone), and arm 3 (DIABEO+telemonitoring by trained nurses). The primary outcome assessed the reduction in HbA1c levels after a 12‐month follow‐up.ResultsSix hundred sixty‐five patients were included in the study. Participants who used DIABEO once or more times a day (DIABEO users) showed a significant and meaningful reduction of HbA1c versus standard care after a 12‐month follow‐up: mean difference − 0.41% for arm 2–arm 1 (P=0.001) and −0.51% for arm 3–arm 1 (P≤0.001). DIABEO users included 25.1% of participants in arm 2 and 37.6% in arm 3. In the intention‐to‐treat population, HbA1c changes and incidence of hypoglycemia were comparable between arms.ConclusionsA clinical and statistically significant reduction in HbA1c levels was found in those patients who used DIABEO at least once a day.CommentIn the present study, the authors failed to show a meaningful benefit in HbA1c reduction using DIABEO, a mobile app for patients with a web portal for healthcare providers, by means of intention‐to‐treat analysis. They did show a significant reduction in HbA1c in the intervention arms compared to the arm of participants treated with the standard of care when only participants who actually used DIABEO once or twice daily were included in their analysis. The authors report that DIABEO was used only in 25.1% of participants in arm 2 (DIABEO alone) and in 37.6% in arm 3 (DIABEO + telemonitoring by trained nurses). The fact that so many of the participants failed to use the app to improve their metabolic control (mean HbA1c of 9.1%) calls for the need to develop better strategies to increase the use of the app and the participants' engagement. Decision support systems should be developed in a way that will minimize the burden of the person with diabetes. The insulin data and glucose information as well as carbohydrate consumption and sport activities should be transmitted by sophisticated sensors passively to be analyzed by the app with no diaries for the patient to document the needed data. The dashboards that both patients and healthcare providers use should be simple and self‐explanatory. I believe that these modern tools will improve patients' engagement even if they are poorly controlled, and these should be indeed tested in real‐life settings.Validation of a Diagnostic Support System for Diabetic Retinopathy Based on Clinical ParametersRomero‐Aroca P1, Verges‐Pujol R1, Santos‐Blanco E1, Maarof N2, Valls A2, Mundet X3, Moreno A2, Galindo L4, Baget‐Bernaldiz M11Ophthalmology Service, Hospital Universitat Sant Joan, Institut de Investigacio Sanitaria Pere Virgili, Universitat Rovira & Virgili, Reus, Spain; 2Department of Computer Engineering and Mathematics, Universitat Rovira & Virgili, Reus, Spain; 3Unitat de suport a la recerca Barcelona ciutat, Insititut Universitari d'Investigacio en Atencio Primaria Jordi Gol, Barcelona, Spain; 4Universidad Carlos III Madrid, Madrid, SpainTrans Vis Sci Tech 2021;10: 17PurposeThe authors aimed to validate a clinical decision support system (CDSS) that estimates risk of diabetic retinopathy (DR) and to personalize screening protocols in type 2 diabetes mellitus (T2DM) patients.MethodsThe authors used a CDSS based on a fuzzy random forest, integrated by fuzzy decision trees with the following variables: current age, sex, arterial hypertension, diabetes duration and treatment, HbA1c, glomerular filtration rate, microalbuminuria, and body mass index. Validation was made using the electronic health records of a sample of 101,802 T2DM patients. Diagnosis was made by retinal photographs, according to EURODIAB guidelines and the International Diabetic Retinopathy Classification.ResultsDR was prevalent in 19,759 patients (19.98%). Of the participants who were evaluated, the authors found 16,593 (16.31%) true positives, 72,617 (71.33%) true negatives, 3,165 (3.1%) false positives, and 9,427 (9.26%) false negatives, with an accuracy of 0.876 (95% confidence interval [CI], 0.858–0.886), sensitivity of 84% (95% CI, 83.46–84.49), specificity of 88.5% (95% CI, 88.29–88.72), positive predictive value of 63.8% (95% CI, 63.18–64.35), negative predictive value of 95.8% (95% CI, 95.68–95.96), positive likelihood ratio of 7.30, and negative likelihood ratio of 0.18. The type 1 error was 0.115, and the type 2 error was 0.16.ConclusionsThe authors confirmed a good prediction rate for DR from a representative sample of T2DM in our population. The CDSS was also able to offer an individualized screening protocol for each patient according to the calculated risk confidence value.Translational relevanceResults from this study will be helpful in establishing a novel strategy for personalizing screening for DR according to patient risk factors.CommentIn the present manuscript, the authors describe a CDSS that estimates the risk of diabetic retinopathy in patients with type 2 diabetes. They have validated their algorithm using electronic healthcare records of 101,802 patients. The ability to better predict the population at risk of developing diabetic retinopathy or other diabetes complications among the population with type 2 diabetes is important and can tailor early screening to those who need it the most, which would increase the efficacy and cost productivity of screening programs. More AI‐based tools to screen for microvascular and macrovascular complications are needed and prospective studies are required to validate them.A Randomized Trial of Closed‐Loop Control in Children with Type 1 DiabetesBreton MD1, Kanapka LG2, Beck RW2, Ekhlaspour L3, Forlenza GP5, Cengiz E 6, Schoelwer M1, Ruedy KJ2, Jost E5, Carria L6, Emory E1, Hsu LJ3, Oliveri M1, Kollman CC2, Dokken BB4, Weinzimer SA6, DeBoer MD1, Buckingham BA3, Cherñavvsky D1, Wadwa RP 5 for the iDCL Trial Research Group1University of Virginia Center for Diabetes Technology, Charlottesville, VA; 2Jaeb Center for Health Research, Tampa, FL; 3Department of Pediatrics, Division of Pediatric Endocrinology and Diabetes, Stanford University School of Medicine, Stanford, CA; 4Tandem Diabetes Care, San Diego, CA; 5Barbara Davis Center for Diabetes, University of Colorado, Anschutz Medical Campus, Aurora, CO; 6Department of Pediatrics, Yale University School of Medicine, New Haven, CTN Engl J Med 2020;383: 836–845This manuscript is also discussed in article on Diabetes Technology and Therapy in the Pediatric Age Group, page XXBackgroundThe use of a closed‐loop system of insulin delivery (also called an artificial pancreas), which automates insulin delivery, may improve glycemic outcomes in children with type 1 diabetes.MethodsDuring a 16‐week, multicenter, randomized, open‐label, parallel‐group trial, the authors assigned children aged 6 to 13 who had type 1 diabetes, i
Referência(s)