Full Carbohydrate Consumption with Hybrid Closed-Loop Control of Insulin Delivery for Type 1 Diabetes - A Case Report Robert S. Silvers, S.M. and Christine Tsien Silvers, M.D., Ph.D. Hosted by https://www.loopcases.com/blog AbstractUsing five variations of hybrid closed-loop insulin pumpcontrollersplusaconventionalpumpduringafour-monthperiod,a48-yearoldmalewithType1diabeteswasableto lower his hemoglobin A1c (A1C) from 8.7% to 6.0%,both without increasing his rate of hypoglycemia, andwhile consuming an average for each systemconfigurationofupto259gramsofcarbohydrateperday.The percentage of time that the patient’s blood glucoselevel was in the 60-180 mg/dL range and the averageestimatedA1CfromDexcomcontinuousglucosemonitordata1were75%/6.3%withnormalMedtronic530Ginsulinpumpuse,88%/6.1%withaBetaBionics iLetusingbothinsulin and glucagon, 86%/6.5%with a Beta Bionics iLetconfigured to use insulin only, 89%/5.6% for a systemcalledOpenAPS version oref0, 92%/5.5% usingOpenAPSversion oref1, and 91%/5.6%using a system called Loopversion 1.4. This seemingly similar performance greatlydivergeswhenadjustedforcarbohydrateconsumptionbydividingaveragedailycarbohydratebytheestimatedA1C.Using thismetric of average carbohydrates per A1C, thehighest-performingclosed-loopsystemofthefivesystemsdescribedwas Loop version 1.4,which outperformed by89%thelowest-performingsystem,BetaBionicsiLetusinginsulinonly.Overall,thecasereportdemonstratesthatapatient with Type 1 diabetes using a closed-loop insulinpump system not only can safely lower his A1C, butfurthermorecanalsoconsumemoderatetohighamountsof carbohydrate while maintaining tight blood glucosecontrol.BackgroundA 48-year-oldmalewas diagnosedwith Type 1 diabetesmellitusin1988attheageof19.Forthefirst14years,hisdiabeteswascontrolledusingmultipledaily injectionsofregular insulinandNPH.Approximately15years ago,hebeganusingaMedtronic insulinpumpwithNovologandHumalog.Overtheyears,hehasusedvariouscontinuousglucose monitoring systems, including the GlucoWatch,Freestyle Navigator, Dexcom G4, Medtronic Enlite,Dexcom G5, and Senseonics Eversense. His body massindex (BMI) lifetime high was 24.6. He became much

more physically active in 2011, participating in regularfitness activities including running and cycling. His BMIduringthemostrecentfiveyearshasbeenapproximately21.5.Closed-Loop“ArtificialPancreas”SystemsTheclosed-loop“artificialpancreas”hasbeenamilestonein the treatment of type 1 diabetes,withmany types ofsystems being developed and being worked on towardcommercial availabilty.2 TheMedtronic 670G is the onlycurrently commercially-available, FDA-approved, closed-loop insulin pump. A 2016 pivotal trial3 established thatamong124patients,A1Cwentfromanaverageof7.4%to6.9%whileusingthissystem.Timewithintherangeof70-180mg/dLbloodglucose levels increased from66.7% to72.2%. Carbohydrate consumption was not establishedforthattrial.The patient has used three types of hybrid closed-loopinsulin delivery systems in which the user enterscarbohydrate intake information.One systemwasapre-production unit planned for future commercial releasecalled the Beta Bionics iLet, which was used while thepatientparticipatedinanFDA-sponsoredtrial.Thesecondwas a homemade system based on the OpenAPS open-sourceprojectthatisnotcurrentlyplannedtobecomeanFDA-approved medical device. The third was an open-source iPhone app called Loop. The iLet and OpenAPSsystemswere each tried in two different configurations.This case report presents results from use of those fivedifferent hybrid closed-loop insulin delivery systems aswell as fromuse of an open-loop system consisting of acommercially-available insulin pump and a separatecommercially-availablecontinuousglucosemonitor.DescriptionoftheBetaBionicsiLetTheBetaBionicsiLetisahybridclosed-loopsystemunderdevelopment for potential commercial distribution withthegoalofbeingsimple tousewhilebeingable toworkwith either single or dual hormones.4 The iLet userinterface (Figure 1) provides a history of blood glucose

levels.Menuoptionsallowtheusertoentermealsize indescriptiveterms,suchas“typical,”or“lessthantypical.”Overtime,theiLetsystemlearnswhatthosetermsmeanfor a particular user. There are additional controls forspecifying an elevated blood glucose target to helppreventhypoglycemiaduringmomentsofactivity.Settingup the iLet system requires only that the care providerenter the user’s weight in kilograms. Everything else islearnedbythesystemasthesystemisused.Thesystemlearns in an ongoing manner, with significant learningover the first 18 hours; after several more days, thesystem stabilizes. The patient found that one advantageof a self-learning system is that the chances for a setuperrorareminimized.

Figure1:iLetclosed-loopsysteminterface

DescriptionofOpenAPSOpenAPS is one of several “do-it-yourself” closed-loopsolutionsforinsulindelivery.Usingopen-sourcetools,thepatient self-assembled a non-FDA-approved wirelessinsulin-pump controller as pictured in Figure 2. Theuserinterface is primarily provided by an iPhone runningsoftware called Nightscout, which is an open-sourceproject for the purpose of remote monitoring of bloodglucoselevelsandtreatmentinformation.Nightscoutrunsonacloud-computingplatform,suchasHerokuorAzure,andcanbeusedtomonitordependents,suchaschildren,whentheyareawayfromtheirparents’immediatecare.OpenAPS can be built using several different types ofhardware.5 The patient used an Intel Edison wearableLinux computer, aHamshieldbrandExplorerboard (thatpowers theEdisonandprovidesa915MHz radio link totheMedtronicpump),andaself-designedand3D-printedenclosure. Because Medtronic disabled certain wirelessfeatures in its newer insulin pumps, an older pump thathas firmware version 2.4a or lower had to be used. Thehardware is powered by a 2500-mAh lithium battery,which provides approximately 16 hours of battery life.Thewirelessrangeistypicallylessthan30feet.

Figure 2: OpenAPS hardware in enclosure designed by thepatient

One typically uses OpenAPS by manually enteringcarbohydrate content using the standard Medtronicpump “Bolus Wizard,” which then triggers an insulinbolus.Thesystemusestheradiolinktoquerythepump’shistory,torecognizethataboluswasgiven,andtoknowthe pre-programmed basal rates within the pump.Software running on the Edison Linux computer usesthese pump settings as a starting point, but can alterthem up or down by a user-configurable maximumchange. The algorithms employed by OpenAPS haveevolvedovertime.Anearlierversioniscalledoref0,andalaterversioniscalledoref1;theprimarydifferenceisthatoref0 only alters basal rates temporarily, whereas oref1canmorequicklyprovide insulin throughmicro-boluses.6Over time, theOpenAPS system adapts andwill suggestnewdefault pump settings for basal rates, carbohydrateratios,andsensitivityfactors.Forthistoworkwellandtowork in a safe manner, the user has a responsibility toentercorrectcarbohydrateinformation.Thisfeatureisofgreatbenefit totheskilleduser,butmostpeoplecannotcount carbohydrates accurately,7 which can lead topotential problems. For example, Dr. Howard Wolpert,M.D., a former senior physician at the Joslin DiabetesCenterinBostonandauthorofSmartPumping(AmericanDiabetesAssociation, 2002), has found that for an applewith 30 grams of carbohydrate, carbohydrate-countingpatientshaveguessed theapple tohaveanywhere from10to60gramsofcarbohydrate.Forthisreason,itisbestfor the user toweigh his/her food, or enter informationdirectlyfromthenutritionallabelonpre-packagedfood.

DescriptionofLoopThefinalsystemtestediscalledLoop,aniPhoneappthatcancontrolsomeolderMedtronicinsulinpumpsthroughasmallBluetooth-basedhardwareinterface.8Becausetheapp is not approved by Apple or the FDA, it is onlydistributed in source-code form; in order to use Loop,then, one must build the app using a Macintoshcomputer, as well as purchase an Apple iOS developerlicense to install theapponan iPhone. Loopversion1.4wasusedforthistesting.ThecontrollerthatLoopusesforpumpcommunicationiscalled RileyLink, and, as seen in Figure 3, is small andunobtrusive. It uses an 850-mAh lithium-ion battery toprovide for about 30 hours of battery life. The rangebetweentheRileyLinkandthepumpvaries,butgenerallythe system works well if the two are within about tenfeet.Forsoftware, theLoopmainuser interface,asseenin Figure 4, is complete and does not rely on additionalsoftware such as NightScout for input or visualization.LoopwilluploadtreatmentdatatoNightScoutifdesired,whichcanbeusedforremotemonitoringofapatient,orforstatisticsandanalysisofpastresults.Loopcanreceivebloodglucosedatafromseveralsources–Medtronic Enlite,G4 Share,G5, orAppleHealth. Loopaccepts carbohydrate input either from a meal-entrybuttonpress (notpictured),or itwillautomatically“see”food input fromAppleHealth as providedby a separatediet app. When food information is entered, Loop willsuggestaninsulinbolus;ifthebolusisaccepted,Loopwillinstruct thepumptodose it. It isnotnecessary tobolusfromthepumpitself,which ishelpful forpeoplewhodonotwanttheirpumpseenbyothers.

Figure3:Patient-createdcustomenclosure

Whencarbohydrate information isentered,onecanalsoselect the estimated number of hours required forcarbohydrate absorption. This is simplified withpictograms representing types of food, such as fruit orpizza. If one were to select a fast-acting source ofcarbohydrate, Loopwouldbemore likely to recommendtheentirebolusupfront.Incontrast,ifoneweretoselectpizza, for example, Loop would be more likely torecommend an initial partial bolus, using a dual-wavemethodtospreadouttheremainderofthedose.In either case, specifics are dependent on how muchinsulinandcarbohydrateLoopthinksthatthepatienthasingested, andwhat it predicts for the patient’s resultingblood glucose level. Furthermore, via clicking on thecarbohydrate portion of the interface, Loop will displaythe previously entered food entries and their actualestimatedabsorptiontimes.

Figure4:Loopuserinterface

ExtremeCarbohydrateTestingDuetotheneedforsafetyforawiderangeofusersofalldifferentexperience levels, the iLet isoptimizedtoavoidhypoglycemia at all costs. Thatmeans that it provides asmaller bolus than what is typically needed, and thencautiouslyaddsmoreinsulinastimegoeson.Becauseofthis difference between the partial bolus that the iLetgives, and the optimal amount required based on thecarbohydrate consumed, there is an ever-increasingdisparity as the size of a meal increases. The patientexperimentedwithveryhighcarbohydratemealswiththeiLet,butthesemealsconsistentlyresultedinahighbloodglucose level, even after a week of system training.Because OpenAPS and Loop allow the user to manuallybolus using the pump controls, and because thesesystems take over insulin dosing to help correct for anyerror,thereisnopracticallimittowhattheusercanhavefor meal size or type while still maintaining a normalrangeofbloodglucoselevels.

Figure5:Glucoselevelsbefore,during,andaftertwoicecreambars

Figure6:Glucoselevelsbefore,during,andafteraverylargemeal

Furthermore, when a closed-loop system maintainsoptimal basal rates, control is better and bolusingbecomes easier. For one, the overnight fasting glucoselevels tend to become perfected. Waking up at 90-100mg/dL became routine and expected during this testing.Suddenlyone-thirdof theday isatabloodglucose levelthatbringsdowntheaverageA1Cvalue.Also,well-tunedbasal insulin control makes eating a high-carbohydratemeal much more feasible. Figure 5 shows that it ispossiblewithOpenAPStoeattwoicecreambarswith52grams of carbohydrate and 39 grams of fat total duringthe day, with results the same or better as a personwithout diabetes. Figure 6 shows Loop processing 185grams of carbohydrate from two McDonald’s Big Macs,twoMcDonald’s Blueberry&Crèmepies, and a 6-ouncestrawberryyogurtsmoothie.

System DateStart DateEnd Carb(g) CaloriesCarb%ofDiet

Est.A1C StdDev

%<60mg/dL

%>180mg/dL

%inRange

Carb/StdDev

Carb/A1C

90daysbefore 1/1/17 3/31/17 N/A N/A N/A 7.4 66.1 3 39 58 N/A N/AOpen-loopnormalpump 4/13/17 4/26/17 221 1993 44.4 6.3 53.8 5 20 75 4.1 35.1iLet-insulin/glucagon 4/27/17 5/10/17 159 1974 32.2 6.1 45.9 1 11 88 3.5 26.1iLet-insulinonly 5/11/17 5/24/17 115 1856 24.8 6.5 41.5 0 14 86 2.8 17.7OpenAPS-oref0 5/25/17 6/7/17 218 2401 36.3 5.6 39.9 5 6 89 5.5 38.9OpenAPS-oref1 6/8/17 6/21/17 238 2450 38.9 5.5 36.8 3 5 92 6.5 43.3

Loop1.4 8/15/17 8/29/17 259 2403 43.1 5.6 37.1 3 6 91 7.0 46.3Figure7:Tableofresults

As meals get larger, the impact of fat on insulinrequirements becomesmore significant, especiallywhenamealcontainsoverapproximately40gramsoffat.Oneresearchstudyfoundthatforhigh-fatmeals,anadditional42% of insulin was needed for the same amount ofcarbohydrate.9 This additional insulin cannot be givenwith the initialmain bolus, as it would very likely causehypoglycemia.Thepatientfoundthatveryhigh-fatmealswould result inelevated glucose levels beginning 1.5 to2.5hoursafter thestartof themeal.WithOpenAPS, themethod used to combat that risewas to watch theglucoselevelsfromtheDexcom,andthenprovideapost-mealbolusofabout40%ofthecarbohydratevalue,ortotreat 80-90% of the fat grams as carbohydrate, using asimplified version of the Pańkowska et al. method offactoring fat andprotein into the dosingrequirements.10,11 With Loop, this additional fat-as-carbohydratequantity was entered at the time ofthemeal,butcouldbemanuallyspecifiedtobedeliveredtwohourslater. Loop would then add additional insulin asrequired. The next section demonstrates some high-carbohydrateexamples of what the patient found to bepossiblewithasystemthathasthistypeofflexibility.ExcerptsfromHigh-CarbohydrateTestingLogHereare some examples of well-controlled high-carbohydrate eating from the testingthat was doneduringthispatient-initiatedself-study.June 7th, 2017 (OpenAPS).Three slices of PapaGino’s large sausagepizza(99gramsofcarb,and54gramsoffat).ThehighestBGwas136from thepizza. Two cookies, Focaccia bread, Not Your AverageJoe’sBackyard Burger, Ice Cream. 279 grams of carb. 2738 calories.Meanglucose113.Standarddeviation26.9. EstimatedequivalentA1Cforday:5.6%.

June18th,2017 (OpenAPS).Orange juice, instantoatmeal,eightOreocookies, Pancakes, maple syrup,banana, blueberries, two Filet-o-Fish,10 pieces sushi/rice, cheesecake, greenbeans, eggplant. 406grams carb. 2879 calories. Stayed below 183 glucose. Meanbloodglucose 121.Standard deviation 31.3. Estimated equivalent A1C forday:5.8%.June19th.2017(OpenAPS).303gramsofcarband140gramsof fat-which included Life cereal, blueberries, two Filet-o-Fish,twoMcDonald’s Blueberry & Crème pies, two 12” flour tortillas,sixslicesofcheddarcheese.2872calories.Meanglucose108,standarddeviation34.1,estimatedequivalentA1C:5.4%Aug20th, 2017 (Loop). Two containers of Greek yogurt, two Filet-o-Fish,oneapplesauce,pastaandmeatballs, twoMcDonald’sBlueberry&Crème pies,Häagen Dazs ice creambar. 254 grams of carb, 104gramsoffat,2354calories.Meanglucose96,standarddeviation25.6,estimatedA1C:5.0%.ResultsThechartinFigure7showsdataandresultsfrom90daysprior to the patient’s period of self-study, the two-weekperiodof using an open-loop system (via a Medtronicpump and aDexcom continuous glucose monitor), twoweeksofusing the iLetwith insulinandglucagon (targetblood glucose level 100mg/dL), twoweeks of using theiLet with insulin only (target blood glucose level 110mg/dL),twoperiodsoftwoweekseachofusingOpenAPS(targetblood glucose level 90 mg/dL), and a two-weekperiod of using Loop (target blood glucose level 90-100mg/dL). The last column of the chartshowstherelationship between average carbohydratesconsumed per day and estimatedequivalent A1C fromusingeachsystem.A largernumber isindicativeofbeingable toeat a larger amount of carbohydrate whilemaintainingalowerA1C.

Figure8:Imagefromthepatient'sself-disclosedmedicalreport

Thepatientfoundthatitwasbesttoeatcarbohydrateinmoderation with the iLet; in doing so, the result wasanestimated equivalent A1Cwithin the “6.5% or below”most strict guidelines commonly recommended forpeoplewithType1diabetes12whilehavingaverylowrateofhypoglycemia.WithOpenAPS, thepatientwasable toeat a larger amount of carbohydrate while attaining anevenlowerestimatedA1C,butatahigherriskandrateofhypoglycemia. With Loop, the patient’s averagecarbohydrate-to-estimated-A1C ratio was even higher,representing the best overall results of the six methodsdescribedwhenadjustingforcarbohydrateintake.ConclusionThe iLethastheabilitytobe life-changingonceavailabledue to ease ofuse and the ability to allow a patient tothink much less about how tomanage his/her diabetes.Automatingthecorrectionofbloodglucoselevelswillalsomeanmore sleep, which is important for quality of life.TheiLetsystem, if usedasdirected, seems likely tohavethepotentialofenablinganyonetoattaina6.5%orlowerA1C,with a low risk of hypoglycemic episodes,providedthathe/shemoderateshis/hercarbohydrateintake.For those who are highly motivated, who would like touse a closed-loop system prior to the iLet’s becomingavailable, and who prefer muchmore control—althoughwith higher responsibility and higher risk ofhypoglycemia—onecanconstruct one’s own hybridclosed-loop insulin delivery system. While these

homemade systemshavenot beenproven to be safe bytheFDA,thepresentedpatient’sresultsdemonstratethateachsystem,onceproperlytuned,isabletofacilitatetheability to eat asmuch carbohydrate as a personwithoutdiabetesmight consume, while still having a reasonablechanceofachievinganA1Cbelow6%.In the described patient case, varioushybrid closed-loopinsulin delivery systems used over a 90-day periodloweredthe patient’s A1C by approximately 37%, from8.7% to6.0% (Figure 8). Moreover, the patient did notexperience a greater percentageof low glucosereadings—hence, experienced no greater riskofhypoglycemia—while averaging up to 259 grams ofcarbohydrate per day.This represents a normalamountof carbohydrate for someone without Type 1diabetesconsuming2000caloriesperday.13Thisloweringof A1C levels was possiblebecause the standarddeviation, or swingsin blood glucose throughout eachday,werereducedbyalmost60%.Inconclusion, it is possible for individuals withlongstanding Type 1 diabetes to eatfoods in anunrestricted manner (i.e., including high-carbohydratefoods) while having tight blood-glucose control,providedthey use the latest tools, not all of which arecurrentlycommerciallyavailable.Whilenotacure, thesedevices provide highly flexible treatment options thatmake good control of blood glucose and the resultinglower A1C values not only possible, but also possiblewithouthavingtoeatacarbohydrate-restricteddiet.

1 Nathan, David M., et al. "Translating the A1C assay into estimated average glucose values." The Journal of Clinical and Applies Research and Education Diabetes Care 31.8 (2008): 1473-1478. 2 https://bmcmedicine.biomedcentral.com/track/pdf/10.1186/s12916-017-0794-8?site=bmcmedicine.biomedcentral.com 3 Pivotal Trial of a Hybrid Closed-Loop Systemin Type 1 Diabetes (T1D). http://www.abstractsonline.com/pp8/#!/4008/presentation/44502 4 https://www.betabionics.org/faq 5 https://openaps.readthedocs.io/en/latest/ 6 https://diyps.org/2017/04/30/introducing-oref1-and-super-microboluses-smb-and-what-it-means-compared-to-oref0-the-original-openaps-algorithm/ 7 Smart Pumping: A Practical Approach to Mastering the Insulin Pump, Howard Wolpert, American Diabetes Association, 2002)

8 https://loopkit.github.io/loopdocs/ 9 Wolpert et al. Diabetes Care. April 2013; 36: 810-816. 10 Pańkowska et al.: Application of novel dual wave meal bolus and its impact on glycated hemoglobin A1C level in children with Type 1 diabetes. Pediatric Diabetes. 2009; 10(5): 298-308.

11 Pańkowska et al.: Bolus Calculator with Nutrition Database Software, a New Concept of Prandial Insulin Programming for Pump Users 12 American Diabetes Association Standard of Medical Care in Diabetes. The Journal of Clinical and Applies Research and Education Diabetes Care 40.1 (2017): 550 13 U.S Department of Agriculture and U.S. Department of Health and Human Services. Scientific Report of the 2015 Dietary Guideline Advisory Committee. 2015