Circadian rhythm of glycemic control in patients with mellitus diabetes

(1)

Universidad

Pública de Navarra

�afarroako

Unibertsita:e Publikoa

Universidad Pública de Navarra, Departamento de Cie!:.cias de la Salud

CIRCADIAN RHYTHM OF GLYCEMIC CONTROL IN PATIENTS WITH MELLITUS DIABETES.

TESIS DOCTORAL

Manuel Antonio Vasq uez Muñoz Septiembre, 2022

Tutores

1lfikel Izquierdo Redin,PhD David Andrade Andrade,PhD

upna

Universidad Pública 00 Navarra Nofrurooko UmbortSltote Pubhkoo

(2)

Listado de Tablas Listado de Figi;ras Abreviaciones Declaración Resumen

Listado de Pub,icac:ones Agradecimientos

C:tpítulo I

Tabla de Contenidos

Descripción General, Conceptos Básicos, hipótesis y objetivos Capitulo lill

Osci!latory pattem of giycemic control in patients with diabetes mel'.itus

Capítulo IIJ

Pá

gJ

na

1

ii iii iv

V

X

xi

8

21

Dynamíc circadian f1uctuations in glycemia in patients with type 2 32 diabetes mellitus

Czpítulo IV

Discusión general Capítulo V

Conclusiones y aplicaciones ;:,ract:cas Capítulo VI

Artículos

44

49

52

upna

Universidad Pública de Navarra NalorrookoUnbertsitateP\.bllkoo

(3)

Listado de Tablas

Capitulo I

Table l. Baselíne characteristics of subjec,s groups, with subgroups of temales

Capitulo U

and control

Table 1. Baseline characteristics controls

índividuals with T2D and

•

37

upna

Universidad Pública de Navarro Noforrooko Uniberts1tate Pl.bhkoo

wíth TlD and males.

of

Página

19

(4)

Listado de J;"iguiras

Capitulo llI Página

\Figure 1. Prevalence of time spent in different glycemic states 21 (hyperglycemic, hypoglycemic and euglycemic period).

Figure 2. Glycosylated hemcglobin (HbA 1 e), glucose variability and 22 prevalence of stationary and non-stationary glycemic oscíllatory

: pattern in females and males with TID and control pasticipants

! Figure 3. Circadian rhythm and oscillatory patterr. of glycemia in 24

• patients TlD and control participant

Figure 4. Circadian rhythrn of glycemia and magnitude of 26 hyperglycemia and euglycemia events during day a.'"ld night phases in

females TlD and control partici_pan!.

·'Figure 5� Circadian rhyth,ü of glycemia and magnitude of 27 hyperglycemia a.11d euglyce;nia ever,ts during day ai:d night phases in

males Ti D and control participant

Figure 6. Predíctive model ofmaxirnum and mínimum glycemia and 29 hour of maximum ar'd mínimum glycemia status in Tl D patients

Capítulo III Página

Figure l. Glycemic status in patients wíth type 2 diabetes and control 44 p;,rticipants

Figure 2. Glycemic variabiíity in T2D ;:,atíents 46 Figure 3. The predictive model c,f maximim and mínimum glycemia 48 oscillation in T:2D patients

upna

Universidad Pública de Navarra Nafarroako Unibertsitate Publikoa

(5)

Abreviaturas

MCG o 'to. :m m nzac1on con mua f d

!glucosa

TlD: diabetes mellitus tipo 1 HbAlc: hemoglobina glicosilada

NON-STATIONARY: random oscillation

FFT: Fast Fourier Transform

PSD: Power spectral density

e 1a um ⁾ u ^{.., ill} ^{'· lS}

T2D: diabetes mellitus tipo 2

!STA TIONARY: Specific pattem

!varia'.JEítv

' ,

j:3MI: body mass index

iFFT: Inversc Fast Fourier Transforrn

jCV: Coefficient ofvariation

m

Universiclnd Pública de Navarra Nafarrooko Unllertsitate Publikoa

elltL

(6)

Declsirndó,n

Yo, :fi:fanuei Vasquez fviuño.l,. dedaro que ia investigación presentada en esta tesis se basa en 2 artículos ( capitule II y de los cuales el prime:·o se encuentra publicado en :.ina revista imen:acional y el seizurido está aceptado. Declaro que los artículos presentados en este documento de tesis corresponden a versioaes fieles del publicado y aceptado.

Tan,bíén declaro que todos los procedimientos llevados a cabo las personas que participaron como voluntarios contaron con la aprobación de' Comité Ético Científico de la Clínica Santa Masía de Chile, comité acreditado por la SEREMI de Salud Metropolitana bajo la resolución exenta N'334990. El acta de aprobación para las investigaciones corresponde a la n^ei 4.

Por últimc, declaro que mi participación en las investigaciones presentadas en esta tesis incluyó, la elaboración en eí éiseño del estc1dio, e! anáíisis de datos e interpretación de resultados, y la redacción de los trabajos incluidos en la presente tesis

upna

lI),

iv

(7)

ResnmeRll

La presente tesis doctoral está basada en la publicación ele 2 estudios c¡ue lienen como objetivo determinar si la variabilidad ele la glucemia, evaluada con monitorización continua de la glucosa (MCG), presenta ritmo circadiano en pacientes con diabetes mellitus (DM) tipo I y II.

Estudio I

La variabilidad diaria de la glucosa es mayor en pacientes diabéticos mellitus (DM), lo c¡ue se ha relacionado con la gravedad de !a enfermedad. Sin embargo, no está claro si la variabilidad glucémica muestra un patrón de oscilación específico o si es completamente aleatorio. Por lo tanto, para determinar el patrón de variabilidad glucémica, medimos y analizamos los datos ele monitoreo continuo de glucosa (CGM), en sujetos control y pacientes con DM tipo 1 (TID). Los datos de MCG se evaluaron durante 6 días (día: 08:00-20:00 h; y noche: 20:00-08:00 h). Los participantes (n⁼l 72; edad⁼l 8-80 años) fueron asignados a grupos de DMl (n⁼l44, mujeres⁼65) y control ( es decir, sanos; n⁼28, mujeres⁼22). Se determinó antropometría, tratamientos farmacológicos, hemoglobina glicosilada (HbAlc) y años de evolución. Las mujeres con DTI mostraron una glucemia más alta entre las 10:00 y las 14:00 h en comparación con los hombres con DTl y las mujeres de control. Los pacientes con DM presentan principalmente oscilaciones estacionarias (deterministas), con características de ritmo circadiano. La glucemia osciló entre 2 y 6 días. El modelo predictivo de glucemia mostró c¡ue es posible predecir hiper e hipoglucemia (R2⁼0,94 y 0,98, respectivamente) en pacientes con DM independientemente de su etiología. Nuestros datos mostraron c¡ue la variabilidad glucémica tenía un patrón de oscilación específico con características circadianas, con episodios de hipoglucemia e hiperglucemia en las fases diurnas, lo c¡ue podría ayudar a la acción terapéutica para esta población.

V

upna

Universidad Públk:a de Navarra Nalarroako lhber1silate P\.bli<o8

(8)

IEsrudio 11

31 objetivo del estudio 2 fue detenninar las características de las oscilaciones glucémícas en T2D y verificar sí se pueden predecir con el tiempo en los pacientes.

Se informa que la variabilidad diaria de la glucosa es mayor en pacientes con diabetes mellitus (DM) que en la població;i general. Se sabe que existen patrones oscilatorios específicos de control glucémico en pacientes con DM tipo 1 (TI D), pero no está claro si lo mismo es cierto pa.ra pacientes con DM tipo 2 (T2D). Aquí, buscamos deter;ninar los patrones de variabilidad glucémica en pacientes con T2D mediante el monitoreo continuo de glucosa. Los datos se evaluaron durante 6 días continuos ( día:

08:00-20:00 h; y noche: 20:00-08:00 h). '..,os par::icipan,es fueron asignados a grupos de T2D (n⁼24, mujeres⁼l0) y de contrnl (es decir, sanos; n⁼28, mujeres⁼22). Los

�esultados mostraron que la hemoglobina glicosilada, la glucemia y el índice de masa corporal fueron más altos en pacientes con DT2 que en los controles (todos p<0,05).

Además, e! tiempo en hiperglucemia y euglucemia fue marcadamente mayor y menor, respectivamente, en el g:::upo de DT2 (p<0,05), sin diferencias significativas para el tiempo en hipoglucemia. Los datos sobre la variabilidad glucémica revelaron que los valores de la desviación estándar, el coeficiente de variación y el poder total de la variabilidad g;ucémica füeron significativamente más altos en el grupo de DT2 que en el grupo de control (p<0,05). Además, los patrones oscilatorios fueron significativamente diferentes entre los grupos (p⁼0 ,032): el grupo de control se distribuyó principalmente a los 2 -3 y >6 días, mientras que el grupo de T2D mostró una distribución más homogénea en 2-3 a >6 ciias. El modelo predictivo de glucemia, utilizado anterionnente en DM l, c.emostró que es posible predecir con precisión eventos de híper e hipoglucemia (R2⁼0,97 y 0,98, respectivamente). Por lo tanto, similar a lo que se observa en la DTJ, los pacientes con DT2 exhiben patrones oscilatorios específicos de ccrtrol gíucémico, que son posibles de predecir. Estos haliazgos pueden ay:1dar a mejorar el tratamiento de la DM al considerar los patrones oscilatorios individuales de los pacientes.

upna

U111V8fSldad Pública de Navarra Nafmooko Unibertsrtate F\JJlikoo

vi

(9)

Abstrnd

This doctoral thesis is based on the publication of 2 studies that aim to determine if the variability of glycemia, evaluated with continuous glucose monitoring (CGM), presents a circadian rhythm in patients with diabetes mellitus (DM) type I and II.

Study I

Daily glucose variability is higher in diabetic mellitus (DM) patients which has been related to the severity of the disease. However, it is unclear whether glycemic variability displays a specific pattem osciHation or if it is completeiy random. Thus, to determine glycemic variability pattern, we measured and analyzed continuous glucose monitoring (CGM) data, in control subjects and patients with DM type-1 (TlD). CGM data was assessed for 6 days (day: 08:00-20:00-h; and night: 20:00- 08:00-h). Participants (n⁼l 72; age⁼l 8-80 years) were assigned to TlD (n⁼144, females⁼65) and Control (i.e., healthy; n⁼28, females⁼22) groups. Anthropometry, pharmacologic treatments, glycosylated hemoglobin (HbAlc) and years of evolution were determined. TlD females displayed a higher glycemia at 10:00-14:00-h vs.

TlD males and Control females. DM patients displays mainly stationary oscillations ( deterministic ), with circadian rhythm characteristics. The glycemia oscillated between 2 and 6 days. The predictive model of glycemia showed that it is possible to predict h_yper and h_ypoglycemia (R2⁼0.94 and 0.98, respectively) in DM patients independent of their etiology. Our data showed that glycemic variability had a specific oscillation pattern with circadian characteristics, with episodes of hypoglycemia and h_yperglycemia at day phases, which could help therapeutic action for this population.

vii

upna

(10)

Stm:liyll

The cbjective of study 2 was to determine the characteristics of glycemic oscillations in T2D and to ver:fy if they can be predic:ed o ver time in patients. Daily glucose variabílity is repcrted to be higher in patients v:ith diabetes mellitus (DM) than in the general populat:on. Specífic oscillatory patterns of glycemic control are knov.'11 to exist ín patien!s with type l DM (Tí;:)), but it is unclear whetherthe same is true for patients with type 2 DM (T2D). Here, we seek to determine patterns of glycemíc variability in T2D patients usi:lg cont;nuous glucose monitoríng. Data were evaluated for 6 continuous days (day: 08:00-20:00; and níght: 20:00---08:00).

Participants were assigned to T2D (n⁼24, wom.en⁼J O) and control (ie, healthy; n⁼28, women⁼22) groups. The results showed that glycosylated hemoglobin, blood glucose a,,.d body mass index were higher in T2D patients than in controls (ali p<0.05).

Furthermore, the time in hyperglycemia and eug!ycemia was markedly longer and shorter, respectively, in the T2D group (p<0.05\ with no signíficant difference for time in hypog:ycemia. The áata on g:ycem:c variability revealed that the values of the standard deviatíon, the coeffieient of variation and the total power of glycemic varíabílity were sign.ificantly higher in the T2D group than in the control group (p<0.05). In addition, oseillatory pattems were significantly different between groups (p=0.032): the control group was mainly distr:buted at 2-3 and >6 days, while the T2D group showed a more homogeneous c'..istributíon at 2-3 days. >6 days. The glycemia predictive model, prevíous!y used in DM:, demonstrated that it is possible to accurately predict hyperglycemíc and hypoglycemie events (R2=0.97 and 0.98, respectively). Therefore, like what is observed in TI D, patients with T2D exh.ibít specific oscíliatory pattems of g!ycemíc control, which are possible to predíct. These findings may help improve the treatment of DM hy consideiing the individual oscil!atory patterns of patients.

vrn

upna

UniversidEld Púbica de Navarra NnforrookoUnibel1srtatePublikoo

(11)

Listado de Publkacio1I1esº

Esta Tesis Doctoral se presenta como un compendio de dos artículos:

L- Vasquez-Muñoz, Mº, Arce-Alvarez, Aºº von Igel, M., Veliz, C., Ruiz

Esquide, G., Ramirez-Campillo, R., & Andrade, D. C. (2021). Oscillatory pattem of glycemic control in patients with diabetes mellitus. Sdentific reports, 11 (] ), 1- 12.

2.- . Manuel Vásquez-Muñoz; Alexis Arce-Álvarez; Cristian Álvarez; Rodrigo Ramirez-Campillo; Femando A. Crespo; Dayana Arias; Camila Salazar-Ardiles;

Mikel Izquierdo; David Cristobal Andrade. (2022). Dynarrüc Circadian Fluctuations Of Glycemia In Patients With Type 2 Diabetes Mellitus. (Aceptado) Biological Research.

ix

upna

(12)

Agradecimientos.

Agradezco muy profundamente a mis directores de Tesis, al Dr.

Dr David Anc'rade, por su confianza y apoyo desinteresado.

Izquierdo_,.y

Dedico esta a mí amada estJo,:a Johanna que, con su amor y paciencia infinita, me ha apoyado y dado animo .. .

A hijos Agustin y ::vlaite ... por su mrne;1s0 amor y alegría, y por todas las horas tes adeudo ... A mis padres y hi•1cm,mn

upna

Universidad Públlca de Navarra Naforrooko UnlbertMate F\bhkoo

mis que

Mikel

tesis

)(

(13)

upna

(14)

apítulo I

D•escripción general, conceptos básicos, h!pótesis y objetivos

upna

e

8

(15)

INTRODUCCIÓN

La diabetes mellitus (DM) es uno de los mayores problemas de salud pública a nivel mundial, lo cual queda evidenciado en los 415 millones de adultos que padecen la patología y los 3 1 8 millones de prediabéticos que la padecerán. Esta información demográfica es en base a la Intemational Diabetes Federation (IDF), lo cual es una proyección para el año 2020 (Bloomgarden 2016; !DF, 2015)

Tomando en consideración este problema de salud pública, es de vital importancia el monitoreo, control y estabilización de los índices de glicemia en esta población.

Consecuentemente, de tal relevancia es la vigilancia del control glucémico, que esto a ayudado a mejorar la incidencia de mortalidad en esta población. Mas aún, se ha descrito que el seguimiento a largo plazo de la glucemia y de la hemoglobina glucosilada (HbAI c) están estrechamente relacionados con la incidencia de complicaciones microvasculares y aumento de la mortalidad (Athappa_n and Khan 2010; UKPDS 1998). Sin embargo, a pesar de que la HbAlc es un indicador con valor diagnóstico, este es un marcador estacionario, el cual no refleja los eventos glucémicos más rápidos (i.e. hiperglucemia postprandiales ). Por lo tanto, se hace necesario el control de la glucemia con implementos y análisis no-estacionarios, como, por ejemplo, el monitoreo continuo de glucemia (Danne et al. 2017; Wallia et al. 2017; Battelino et al. 2019).

Los dispositivos de MCG son una alternativa que permiten a los profesionales de la salud evaluar el control glucémico, detectar hipoglucemias, especialmente nocturnas, y proveen información adicional, incluyendo h variabilidad glucémica. El desarrollo de terapias integradas posibilita el cumplimiento de metas de control glucémico estrictas con una disminución en los episodios de hipoglucemia severa en una población de alto riesgo (Danne et al. 2017; Wallia et al. 2017; Battelino et al. 2019).

Tomando en consideración que la glucemia puede variar durante el día y noche ( curvas circadianas del control glucémico ). es importante destacar que no se ha descrito si es que existe alguna diferencia entre damas y varones, con diabetes tipo I y II, sobre las curvas circadianas del control glucémico.

9

upna

(16)

F..EFERENCIAS

L Atha¡:¡pan, Ganesh, AND Faisaí M. Khan. 20 l O. "Effect Of lntensíve Blood

Glucose Control Compared To Conver:tional Treatment On Cardiovascular Endpoínts In Patients With Type 2 Diabetes." Jm.1rrud of the American College of Cardiology. DOLORGil 0.1 O l 6/S0735-l 097(! 0)61276-4.

2. Battelino, '::'adej, Thomas Danne, R:chard :Vl. Bergenstaí, Stephanie A. Amiel, Roy Beck, Torben Biester, Emanuele Bosi, ET AL 201 9. "Clinicai Targets FOR Continuous Glucose Monitoring Data ln:e::-pretation: Recommendations From THE lntemational Consensus ON Time IN Rru'l!ge." Diabetes Care, June.

DOLORG/10.2337/DCI! 9-0028.

3. Bioomgarden, Zachary. 201 6. "Queslioning Glucose Measurements U sed IN THE International Diebetes Federation (1df) Atlas.'' Journal of Diabetes.

DOJ.ORG/I0.l l l l/l 753-0407J2453.

4. Danne, Thomas, Revital Nimrí, Tadej Battdino, Richard M. Bergenstal, Kelly L.

Cicse, J. Hans Devries, Satish Garg, ET AL 2017. "International Consensus ON Cse OF Continuous Glucose '.'v1onitoring." Diabetes Care 40 (12): 1 631-40.

5. Intemational Diabetes Federation. 201 5. IdfDiabetes Atlas.

6. Ukpds. 1 998. "Intensive Blood-Glu�ose Control W!TH Sulphonylureas OR Insulin Compared WITH Conver:tionai T ,eatment AND Risk OF Complications IN Patíents WITH Type 2 Diabetes (Ukpds 33)." The Lancet.

DOLORG/10.1016/S0 l 40-6736(98)0701 9-6.

7. Wallia, Amisha, Guillenno E. Umpierrez, Robert J. Rushakoff, David C. Klonoff, Dariiel J. Rubín, Sherita Hill Gol den, Curtiss B. Cook, Bithika Thompson, AND THE Dts Continuous Glucose Monitoring � THE Hospital Panel. 2017. "Consensus Statement ON Inpatient Cse OF Continuous Glucose Monítoring." journal of dhibetes science ar.d tecllmofogy. DOLORG/1 0. l 1.77/19322968 1 7706 1 5 1 .

upna

to

(17)

OBJETIVOS E HIPÓTESIS DE LA TESIS DOCTORAL

Hipótesis

El control de la glucemia durante las fases diurna y nocturna logra un patrón estacionario, con características de oscilación del ritmo circadiano.

Objetivo general

Determinar si la variabilidad de la glucemia, evaluada con monitorización continua de la glucosa (MCG), presenta ritmo circadiano en pacientes con diabetes mellitus (DM) tipo I y 11.

Objetivos específicos

1. Evaluar CGM en pacientes con DM I y 11

2. Comparar la variabilidad glucémica en pacientes con DM I y 11 3. Comparar el ritmo circadiano entre DMI y DMII

11

upna

Universidad Pública ele Navarra Nafmoako lnber1Sl1ate P\Jblikoe

(18)

apítulo 11

OSCILLATORYP ATTERN OF GLYCEMIC CONTROL IN PATIEJVTS WITH DIABETES lvfELLITUS

12

upna

e

(19)

lntroduction.

Diabetes mellitus (DM) is a significant public heaith problem, affecting 415 million adults, in addition to 3 1 8 million prediabetics ^{1 •} Importantly, public spending associated with DM reached US$237 billion in 201 7 in the US ^{2 .}Moreover, DM is a significant cause of blindness, kiclney failure, heart attacks, strokes, ancl lower limb amputation ^34.Consiclering the prevalence, poor prognosis and hazarcls associatedl with DM, it is essential to monitor, control, and stabilize the glycemia in this population. Along with this, it has been proposed that glycosylated hemoglobin (HbAl c) could predict the risk of long-term diabetes complications ^3-4-5-5_However, HbAl c allows only long-term metrics, limiting personalizedl therapy, particuiarly in DM type I patients (TlD). Compared to HbAlc, continuous glucose monitoring (CGM) allows real-time measurements of glucose, although it is feasible to assess hypoglycemia/hyperglycemia episodes and consequently glycemia variability over severa! days, weeks or even months ⁷-s However, whether the dynamic of glycemic fluctuation is exclusively represented by a specific pattem variability (stationary) or random oscillation (non-stationary) has not been completely explored. In fact, Kovatchev et al. (2016; 201 7)^9-10indicated that the dynamic of glycemic fluctuation is more related to stationary signals rather !han to random oscillation; however, it is no clear if a different pattem of oscillation occurs between patients, which could be relevant to different therapies applied in DM population.

Along to the, glucose variability is of such relevance due that vascular complications occurrences in DM patients have been attributed to hyperglycemias and dysglycemias (higher and lower levels of glycemia during the day and night phases) events 9-10-1 1-12 Additionally, despite that robust information exists regarding CGM data ^7-3, today there are no predictive models to extrapolate the moment and magnitude of hyper and hypoglycemia in DM patients. This is extremely relevant, considering that severe cardiovascular comp!ications are related to dysglycemic events ^{1 1-12•}Thus, glycemia variability and its pattems may be considered a health issue; however, i) whether a given specific pattem wavering represents glycemia oscillation and ii) if it is possible, to predict severe glycemic events, is yet to be determined. Therefore, we hypothesized Lh.at g!ycemia control during the day and night phases accomplish a stationary pattem vvith characteristics of circadian rhythm oscillation, which is possible to modulate by a mathematical function.

13

upna

(20)

Materials and metll:wds

Patient Po]!mlation and efüical approveó

Particípants between 18 and 80 years of age were recruited either with diagnosis of TI D or healthy partícipants (Con:rol). Those with DM had regular medica! check

ups. and their treaLT.:ents were fully described in clinical files. Patients were recruited from an Endocrinology and Diabetology se;-vice from Clínica Santa :Viaria, Santiago, Chíie, between January 20 l 5 and June 20 l 9 {Supplememary Fignre S l ). Partícipants were excíuded whefaer they had i) gestatíonal diabetes; íi) <6 days of continuous blood glucose monitoring; íii) <18 years-oid; md ív) female participants should not be in the first seven days of the follicular phase v) no clínica! records. Patients were dívided into tw::i groups: T l D (n⁼l 44, females⁼65) and Control (healthy; n⁼28, females⁼22) group.

Ali methods ami experimental ptotoeols were ca.nied out in accordance váth the American Diabetes Association and ín accordance with the Declaration of Helsinki (2013). In addition, ali methods and experimental protocols were reviewed and approved by the Ethical Cornmittee from Clúlica Santa María, Santiago, Chile (approved #14 ), \Vrirten ínfonned consent was obtained from each participant according to CIOMS Guideline # 4. Ali suójects were o ver l 8 years o!d; therefore, the ínformed essent signed by a legal g-\Jardían was not necessary,

CGM Metl1wd and (hntcomes

During six consecutive days, a retrospective CGtvl system (M edtronic Inc., Northridge, CA) for subcutaneous ínterscitial glucose monitoring was used. This electronic device was inserted subcutaneously in the non-dominant arm, and removed ru."ter six days, under sterile conditions. Ali CGM recordíngs were performed blind with the M edtronic M iniMed iPro2 (iPro2 digital recorder) wíth an Enlite sensor (M edtroníc Inc., Nonhridge, CA). Data was manually registered to calibrate the sensor data. The sample rate of the sensor was 0.003 Hz ( one poim every 5 minutes, 288 daily measurements). When a sensor failed, the missed data from recording was not replaced by interpolation or mean calculation. However, for analyses purposes, measurernents obtzined after each hour were averaged, generating 24 data points every day, for six days. In additiorr, to fast Fouríer transform analysís, the missed data WciS replaced by füe mínimum errergy oscíllation reconstruction, as previously described³⁵. The repo;-ts were analyzed índividually to find calibraúon

upna

14

(21)

errors. Finally, the data was exported in CSV format. The CGM system detennines interstitial glucose using a sensor with glucose oxidase, a..n enzyme that catalyzes the electrochemical reaction between glucose and oxygen, obtaining an electric current in nanoamps, wirelessly transmitted to the CGM receiver 36_ A calibration algorithm was used for the CGM system according to the manufacturer instructions. From continuous recordings, the HbA 1 c was determined according to füe following formula: HbAl c %= (Avg glucose + 46.7)/28.7; after HbAlc mmol/mol= (10.93 x HbAlc %) - 23.5 36. Where Avg glucose is lhe arilhmetic mean calculated from ali data points from CGM. Ali these calculations were performed with CareLink iPro software (version 2.2.005, Northridge CA, USA). Glycemia was reportee! as mg/dL.

After six days of CGM, glucose variabi!ity para,-neters were calculated. SD and CV were used as glucose variability outcomes ^37.To determine CV, the SD (from 6 days) was divided by the arithmetic mean (from 6 days) of the corresponcling glucose reading. Written instructions regarding food consumption hours were provided. Foocl consumption times included breakfast (07:00-10:00 h), lunch (12:30-15 :00 h), dinner (19:00- 22:00 h) and post-dinner (22:30 -00:00 h). The suggested ca!oric intake for patients with DM was prescribed according to the American Association of Clínica!

Endocrinologists and the American Diabetes Association recommenclations.

Carbohydrates, protein, and fat representecl ·:ietween 45%-65%, 1 5-20% ancl <30%

of total daily energy intake, respectively. Fiber intake between 25-50 g/day was included. This recommendation was given to each patient advised by a certified nutritionist ^38•To determine the possible effects ofpharmacological treatments in all experimental groups, we assessed the prevalence of the main medícaticns that each patient received, taking into consíderation that patients follow a scheme delivered by the diabetes unit. The DM patients administrated standard insulin before breakfast and the ultrafast insulin before every food intake. Patient education was conducted by qualified diabetes nurse educators and nutritionists from füe Diabeto logia Clínica Santa Maria team.

The treatment was divided into three groups: rapid-acting insulin and long-acting insulin, represented with "n" and % in each experimental group the patients did not receive any other medications that could affect glycemic control (See Table l and Supplementary Table S 1 ). The insulin was administrated by a calculated bolus (relative to g of consumed carbohydrates and body mass) before each mea!. The dosage is showed in Table 1 and Supplementary Table l . The subjects were strongly

15

upna

(22)

advised to not engage or perform any physical exercise, other than their regular daily work-home actívities; however, tl:e daily actívitíes were not restrícted.

A!l.thropometric variables anél medical histrnry

Every patíent, who was adherent to the present research, underwent a physical examination that bclude determbation of age (years), body mass (kg), height (m) and body mass index (BM I; kg·m-2). After that, their medica! history was assessed, Including treatment type and years wíth diabetes. Toe prevalence of euglycemía, hyperglycemía and hypoglycemia were d.etermined in ali experimental groups, tlmmgh CG M .

Stationary/non-statio11acy CGM ama!ysis

To determine the stationary and non-statione.ry pattem of glucose variability from the time series provit.!ed by the CGI\.J syster.1, the Dickey-Fuller test was used as prevíousíy described 20_ The Díckey-Fuller test, which is a nonlinear estimation, assumes thal the data has interdependency with the prevíous data point ( delay time).

To detem1Íne the stationary a,--¡d non-statim,ary pattern, the following formula was used: yt= pyt-1 + i:t. \lilhere y is glucose data, t is the time, and p is a constant coefficient related to autoregressive analysis 20. A statíonary variable was defined as a variable without a significant (p>0.05) change in variance across all time.

Contrarily, a non-sta'Íonary variable was defined as a variable vvith a significant (p<C.05) change in variance across time. The analysís was perfonned using R Core Te2m (2020) 39.

Glycemk oscillato:ry pattem

The circadian rhythm was detemüned 10 all patients through the Fast Fourier Transform (FFT) algorithm, according to prevíous research 40• For the current data structure, every glyce;níc signa! corresponded to l ,440 data points, and missing data in the signa! was replaced by the mínimum energy oscillation reconstruction 35. To obtain different :fi-equencies of the cycles, the FFT algoríthm was applied ⁴1, using the functions available from the FFT module in the NUM PY package (Python Anaconda 3.6.6, 64 bits version) 42. From dif:erent frequencies, function pondered, and the power spectrni densities of each pondered were obtaíned. Afterwards, the inverse FFT (iF!'T) pondered was used to verify the quality ofthe estimation. Then, fae frequencies with the highest power were se!ected (over 1 5,0 00 a.u.) according to energy weight signa! 35 . The total power of the signa! arad the frequency at maximum

16

upna

(23)

power spectral density (PSD) were plotted. To determine the prevalence of different oscillatory pattern in ali patients, the data was divided on >2 to <3; >3 to <4; >4 to

<5; >5 to <6; and >6 days to maximum oscillation. This analysis was performed using Python Anaconda 3.6.6, 64 bits version (Python Software Foundation, Amsterdam, Netherlands) ^42.

The predictive model of glycemia

Our predictive model was the convergence of severa! successive steps to find the best linear models which display a higher and robust adjustment ^{4243 .}Thus, our algorithm was applied as follow: i) natural logarithm transformed was applied to the time ofthe mínimum and maximum glycemia; ii) conelations were made between the variables of mínimum and maximum glycemia and mínimum ancl maximum glycemic time with ali FFT weights (see: Glycemic oscillatory pattem section). Where the first 140 data, for adjustment, are chosen at random. The operation was repeated at least 40 times per model to identify points which could generate problems in more than one adjustment model. Thirty-three data points that had difficulties in their adjustment behavior were taken from the total data, probably due to missing data points. With the remaining data, the final model was calculated. The Akaike criterion was applied to select the best model. Then the non-significant variables were eliminated, one by one, leaving only those variables that were significa..n.t for a level <0.05 and with the adjusted models we proceeded to see in one case. This analysis was performed using the Python software version 3.8.1 (Python Software Foundation, Amsterdam, Netherlands).

Statistical Analyses

Data is expressed as mean ± SD or 95% confidence interval (glycemic oscillatory pattern data). Ali data was subjected to nom1ality (Shapiro-Wilk) and homoscedasticity (Levene) testing. Data was evaluated using a 2 (control and TID) x 2 (female-male) analysis ofvariance (ANOVA two way), followed by Holm-Sidak posthoc analysis according to the data structure. Non-parametric variables were evaluated using Kruskal-Wallis analysis followed by Dunn' s posthoc test. To determine the magnitude of hyperglycemia (dinical cut-off > 1 80 mg/dL) ³¹ and euglycemia between female and male, during the day and night phases, the Mann

Whitney test was used. P < 0.05 was considered statistically significant. Ali analyses were performed with GraphPad Prism 9.0. l (La folla, CA, USA) and R Core Team

17

upna

(24)

( 2 0 2 0) ^{3 9 .}

upna

(2020) ³⁹^.

18

(25)

Results

Baseline characteristics of TID and Control participants.

Baseline characteristics are described in T 2.blc í. and Suppíementary Tabie S 1 . Age, HbAlc, BMI, period of euglycemia, period of b.ype:-glyce□ia episodes, a.n.d standard deviation (SD) of glucose were significanüy differem be:vveen Control compared to TlD patients (Supplementary Table SI. a] :'.)<0.01). Females with T1D showed a significant decrease in body weight ar:d �1eight compared to their Control group (Table 1 ). Euglycemia and hyperglycemis :¡Je:-iod i:1 females and males with TlD were significantly lower (p<0.05) and higher :p<0.05), respectively, compared to their sex-matched Control grocp (Fig. 1 ;.

�ne

average daily units of ult::-a-rapid insulin were calculated for each group and foere was no difference between groups (Male vs. female; from TlD gro:.ip; Tab:e1 ).

19

upna

Universidad Publica de Navarro Naf~UllibertsltateF\A:Jlikoo.

(26)

Table 1. Basclinc characteristics ofsubjects with TI D and control groups, with subgroups offernales and males.

Agc (years) Weighl (l(g) Height (m) BMI (kg/111²⁾ Duration of diabetes (Years) Treatment ^(%) Units ofinsulin per clay

(U/mi)

Aspart Lispro

Lanlus Tresiba Toujeo Levcmir

Control Fernales

(n�22;78.57%) 40. I H 1 8.12

71 .04±7.38 169.09±8.95

24.86±2.00 ---

Males (11�6;21.43%)

47.33±19.58 68.50± 7.01 1 70.50±8.57 24.23± 1 .42

---

Fernales (n�65;45. 14%)

34.68± 15.42 63.49±6.7 l 'i'

157±22-r 24.24±2.48 29.48± 12.60

5.0 ±1.51

Rapid Acting lnsulin 11�21 (33.82%) IF 31 (47.06%) Lo11g-Acti11g l11suli11

IF47 (73.53%) IF3 (4.41%) 11�1 ( 1 .47%) 11�1 1 (16.18%)

TID

Males (IF79;54.86%)

30.72±14.4 1 * 69.54± 13.29

170±14 23.74±3.72 23.43± 1 1 . 1 O

5.2± 1 .9

IF24 (30.30%) IF 34(42.42%) IF55 (69.70%) IF4 (4.55%) 11�1 (1 .52%) 11�25 (3 1.82%)

P-value

<0.0001 0. 1785 0.2723

<0,0001 0.0267 0.36

Distribution F (Dl'n. Dl'd) (2, 1 8 1 ) � 62.33 (2, 181) � 1 .740 (2, 18]) � 1.310 (2, 1 8 1 ) � 21.70 (1, 154) � 5.004

Data are showed as mean ± Standard deviation (SO). SO of glucose: Standard deviation of glucose. %CV for glucose: Percentage Coefficient of variation for glucose. Two ways ANO VA, following Holrn-Sidak post hoc. Units of insulin per day (mean ± SO).

*, P<0.005, Males TlD vs Males Control; t, P<0.005, Females TID vs Fernales Control.

upna

(27)

Glycemia Control

1 00

� 80

� �

60 40 20

o

Female Male

B

Glycemia T1 D 100]

:;- 80

o' ,

�

fl

_e ⁶⁰

15 40 (l)

&í 0::

²⁰

o

Female Male

• Hyperglycemia iilii Hypoglycernia

• Euglycemia

Figure l. Prevalence of time spent in different glycemic sfates (hyperglycemic, hypoglycemic ami euglycemic period). h; females ^2!11drr:.ales with diagnosis efTlD and Control group. One-way ANOV A, following Holm-Sidak posthoc.

Glycemic status and stationary í non-statimiary glycemia variability pattern ⁱⁱ¹ females and males with TlD ana Conrt:rols

Toe HbAlc, calculated from CGM data, ,evealed that femaies and males with TID displayed a significant increase ofHbAíc compared to their respective sex-matched Control group (Fig. 2A, ali p<0.0 l). Regardíng gíucose variability data, the SD of glucose was significantly higher in :irales and females ,vi'.h TE), cornpared to the:r respective sex-matched Control group (Fíg. 2B, all p<0.05). Females and males wíth TlD showed a sígníficant íncrease ín the coefficient of variatíon (CV) of glucose compared to theír respective sex-matched Control group (Fig. 2C, p<0.J,).

To determine if glycemíc variability accomplishes a deterministic or random oscillation, we evaluated ifthe sígnal displayec a stationary (detenninistic) and non

statiomny (random) behavíor tlrrough the Dkkey-Fuller test. Our analysis reveals that 91.7% and 86.1% of femaies and maies wíth TlD, respectiveíy, displayed a stationary glucose oscíllator¡ pattem (Fig. 2D). In addition, 77 .3% a;-id SC.0% of Control females and males, respectively, showed. a stationary pattern (Pig. 2D). These data suggests that partícíp¡uns accomp!ísh for :he most pert stationary glycemia oscillatory pattern and not a random osciHat:on onc.

21

upna

Universidad Pública de Navarra f-.wfmooko Unberls1tote PubliKOO

A

(28)

A Control 15 :21 T1D

� 1 0 _o

* *

i

_:e ₅ ^T

o ^l

Female Male Stationary signa!

/ A\

^Variance

/\�v:Mean

� Vanance

Non-staliona¡y signa!

��• Var.aneé

Mean Variance

iOO 80

O>

fil E u, 60 2

8

40

O)

20

o D

* *

7

Female M2le Control

e

100

� 80 o 8 60 o, 40

o 6

²⁰

o

T1D

Female Mala

Figure 2. Glycosylated hemogiobin glueose varia ^t1ty and prevalence of stationary :imd non-stationary glyeemic oscilfatory pattern in females and males with TH) 2111d control partkipants. A: HbAic is increased in TlD patients índependent of sex. lndeed, femaies and male pa!ients showed a sígnificant

¡ncrease HbA1c compared to her/hís matched Control group (Two ways ANOV A, follo"ing Holm-Sidak posthoc; F (DFn: 1 8 1 ) = 32.43; p<0.000 1 ). B: TlD females and males patients display an inc,ease of standard deviation (SD) of glycemia compared to herfüis makhed Com�ol group (Two ways A.c"JOV A, follcváng Holm-Sidak posthoc; F (DFn: 2, DFd: 1 80) = 35.30; p<0.0001 ). C:

compared to her matched Control group, TI D female subjeets showed a significant 'ncrease of coefficicnt ofvariation (CV) of giycemia (Two ways ANO VA, following Ho:m-Sidak pos1:hoc; F (DFn: 2, DFd: 1 = 1 5.89; p<0.0001). Male with TlD showed a significant difference compirred to his control group (Two ways Ac'sfOV A, following Holm-Si dak posthoc; F (DFn: 2, 1 80) 15.89; p<0.0001 ). D: left panel showed an example of stationary &,d non-stationary oscillatory signa!.

Stationary signa! is characterized by a cor:stant v2riance, and contrarily, non

stationary signa! did not display a constan! varia.nce. Right panel showed the prevalence of stationary and no:1-statíonary glycemic oscillatory pattem for 6 days of anaiysís. Statíonary and non-statíonary signals were analyzed by R Core Terun (2020) ³⁹

22

upna

Universidad Pública de Navarra Nafnrrooko Uni:Jertsrtate F\.t:Jlikoo

of

\

/

B

o

(/)

Fema!e

T1D

(HbA1c),

2, DFd:

80)

DFd:

=

(29)

Glycemic oscilfatory frequency in females aml males with TUJi

Females and males control group did not show significant differences during day and night phases (Fig. 3A). The CGM revealed that between l 0:00-14:00 h, the female TID patients displayed an increase of glycemia compared to TID male patients (Fig.

3B, ali p<0.05).

The Fast Fourier Transform (FFT) approximation to determine the glycemic oscillatory pattern in DM patients showed that the total power of glycemia was significantly increased in TID patients cornpared to the Control condition, regardless of sex (Fig. 3D; all p<0.01 ). Moreover, the maximum giycemia oscilhtion varíes from 2 to > 6 days (Fig. 3C and 3E). Female TID patients displayed a prevalence for glycemia oscillation of 43.0%, 1 8 . 1 %, 6.9%, 5.6%, and 26.4% on >2 to <3, >3 to <4,

>4 to <5, >5 to <6, and >6 days, respectively (Fig. 3F). Male TlD patients displayed a prevalence of36. l %, 1 9.4%, 6.9%, 1 1 . 1 %, and 26.4% on >2 to <3, >3 to <4, >4 to

<5, >5 to <6, and >6 days of glycemia oscillation, respectively (Fig. 3E).

23

upna

(30)

A

e

o

₆₀₀

o

2ay 2

Conircl ilm E

Tí D ]:] Control

T1D Maie

Cay 3

13

Glycemia - [320 iFFTG!ycemía••· 160 � íCi

a�

o :;;;;:�-

Düy 4 F::maie

Dsy 5 :=>2-"''.: tays \f\°' :3. 59.' '/, ) . ':'.'►�-"� '""'l'' '.•"' o.o o�c;

.•><!--<$ 0;1.y1i•"' O; 0.1%)

•"i . .;2 d;,_ys íll• 2: ) 1'�) l'IP$c.:a"� (11"' 1',)1.��;\

�:>-<J .. ,,.: (�"' ''; � � :-.¡

::;;.:,_,u o,;¡,; (,,"' •. �-IS't,¡

::::� ��;:;: �/t;1t

¹

m,o-\ cbys :;=-:;1: •)<,,-}

Day6

[345 0

2 1 0

iI

^{� w-}

75

_ , 1,.1:."::; �•►·• ('"'"'e: ao.7'1;.J . O>J...<A C•11.(n"' 1; 11.1'�) .,,.� ... e d,1., ¡ ... o: o.o••¡

mp$-,;Q �•"(""'o: c.c•.:.1 ' m:>G díl)'S (rr⁼1: 11.1�(-:

Figure 3. Circllldfan rhythm end oscmator¡ pllittern of glycemia in patients TlD and eontiroi participl!int. A: fema.les and males Control participant did not show significant diffem.-ice on circadian rhythm of glycemía. B: during day phase, females with TlD showed a signific3"'1t difference compared to males with TlD patients between 1 0:00 te 15:00, wíthout differences during night phases (Two ways ANOVA, following Fisher posthoc; F (DFn: l, Dfd: 1 33) = 3.51; p=0.049). C:

representative recotstructior: from inverse Fast Fourier Transform (iFFT) of maximum energy glycemic oscillation and real oscilíation of glycamia at 2, 3, 4 ,5 and 6 days. D: Total power is hcreased in TlD patients independent of sex. Indeed, TlD females and males patients showed a signifieant increase of total power wmpared to her/his matched Controi group (Two ways ANOVA, fol!owing Ho!m

Sídú. posthoc; F (DFn: 2, DFd: 1 81) = 32.43; ;:,<0.0001). E: prevalence ofmaximum energy oscillaíion of glycemia in all experimental conditions. Note that in TlD patients (femaíes and males) the maximurn oscillation mainly is related to 2 to 3 days.

24

upna

Universidad Pública de Navarra Nafarroako UntiertMate Pu:Jl1koo

~

'F

⁴⁰⁰

º "E

n.. e

'"u

²⁰⁰

~é

Time

~- ---r----•~----,,----~ ^r-"

Fema!e

Time

"

DayS

Oey5

Day6

¡

³⁰⁰¹⁸⁰₆₀

fv1a!e

(31)

Between 10:00-14:00 h, the female TlD patients displayed an increase of glycemia compared to the female Control group (4A, alí p<0.05). Males with TID between 02:00-10:00 h, showed an increase of glycemia leve! compared to Control male participants (Fig. 5A; ali p<0.05).

The interindividual variability of glycemia (hyperglycemia and euglycemia, from cut off: 1 80 mg/dL, accordingly to Miranda-Massari et al. 2016)¹³during the day and night phases, revealed no significant differences between females Control and TlD participants (Fig. 4B and C). Regarding to males, T!D patients showed a significant difference in hyperglycemia magnitude between day and night phases (Fig. 5C;

p<0.05). Control participants didn't show significant differences in interindividual variability during the whole experiment (Fig. 5B).

25

upna

(32)

:e

·O· Control

· @

D

250

� 1 50

(.;)

1 00J

Control

08:00-20:00 Day

Night 20:00-08:00

� 1

SOl n""1 (4.5'%}

.:S ^50j

, ...

-� -50�

a., J

� _150 n-""21 (95 . .a.%) Above cut off: X₂, p= N/A Below cut off: X₂, p;:: 0.618

i-ernale

T1 D

08:00-20:00 Day

Night 20:00-08:00

n=42 (62.7%) Above cut off: X2, p= 0.854 Below cut off: x₂• p= 0.737

Fig1me 4. Ciireadiim �hythm of glycemfa and magnit11.de of byperglycemia and euglycemia events during day and nightr pil:iases fema.les TlD and control participant. A: during day phase, females TlD showed a sígnificant difference compared to !heir matched control group between l 0:00 to 1 5 :00. (Two ways AKOVA. fol!owing Fisher posthoc; F (DFn: 2, DFd: 1 1 9) = 3.91; p=0.022). B and C: magnitude of hyperglycemías and euglycemia's during day and níght phases, in Control and TlD femaie patients, respectively There are no significant differences between all groups,

26

upna

Universidad Publlca de Navarra Nafarrooko Unibensitata Pl.bhkoe

A

B

- T1

E°' 150

.2

^~

l

^.^n.,1]''"•: ⁽⁴^.5%)

s

^ro ⁵

º j '

^....

^· ¹¹

¹^!

o ~ '

-~ -50 • . ,

l.

~ · n:21 {SS.4%)

<..!) -150 '

· - ·

. -. ,

. _--_---_-------------. .

in with

(33)

A

-o- Control -e- T1 D

250

:!:: O) ²⁰⁰

.S

^{1 50}

.E 1 00 c..9 � 50

o

Male

_<::,<:::, _<:::,<::> .<:::,<::> .<::,<::> .<:::,<::> .<:::,c::o _,;:,<:::, _<::,<:> _e:,<::> _<:::,<:::, _<::,<:::, .<:::,<::>

�- �- �- �- �- �- �- �- �- �· o/· �·

B C

::g"

^{1 50}

E o 1 50

:5 _'-'

�

"'

50

·E

^-50

(9 - 1 50�

Control

08:00-20:00 Day

n=S (83.3%)

Night 20:00-08:00

... -�·

: --- : n=6 (100.0%)

1 111··1·

Above cut off: X2, p= N/A

Below cut off: X2, p= 0.981

me

T1 D

08:00-20:00 Day

n=46 (67.6%}

Night 20:00-08:00

n=23 (33.8%)

n=45 (66.2%)

Above cut off: X2, p= 0.048

Be!ow cut off: X2, p= 0.933

Figure S. Circadian rhythm of glycemia :md magnitude of hypeH"glycemia and euglycemia events during 1fay and night pitases in males TlD and contrnl participant. A: during day pitase, TI D mal e patients sltowed a significant difference compared to their matched control group between 09:00 to 10:00. During night phases, males with TI D patients showed a significant difference compared to their matched control group between 21 :00 to 08:00 (Two ways ANOVA, following Fisher posthoc; F (DFn: 2, DFd: 85) ⁼4.42; p=0.01 5). B and C: magnitude of hyperglycemias and euglycemia's during day and night phases, in Control and TlD males' patients, respectively. Note that TID male patients display a significant difference on h_yperglycemias during day and night phases (Wilcoxon rank test, p⁼0.048).

27

upna

Universidad Pública de Navarra Nafnrrooko Uni:Jertsrtate F\.t:Jlikoo

--

.E ro

(l.) c.J >..

n=1 (16.6%)

50- : ,

-50: □ ⁿ n n

' ... ---.. ---.

C9 -150-

Ti

í:2' 150 o

----

¹⁵⁰

~ o

"5

ü 50

s

ro

.E -so

© u

>..

CJ -150

n=22 (33.4%)

(34)

The predictive model ofbyperglycemia ,md hy¡poglycemia in DM patients.

The CGM elata was used to create a predictive model to estimate hyperglycemía and hypoglycemia and addit:onall y, to determine the exact time at which the hyperglycemia and hypoglycemia in DM patients was observed. Our data showed

thal il is possible to predict hyperglycemia's (R²= 0.92; ^F90.40= 4.557; p < 0.001) and hypoglycemia's DM patients (R²= 0.98; ^Fs6,84= 63.3 1; p < 0.001) (Fig. 6A), which was inde;,endent of age. Toe error of our predictive model accomplished a normal distribution to hyperglyce;:nia (W = 0.98; p = 0.35) and hypoglycemia (\V⁼ 0.98; p = 0.08) events (Fig. 6A). Along with these results, we found that it is possible to ¡:iredict the time in which the hyperglycemia (R²⁼0.99; Fs6.s⁴= 243.l; p < 0.001) and hypogiycemia (R²= 0.99; Fss.s2 = .58; p < 0.001) events occur (Fig. 6B).

Regarding to the error of the model, our revealed that the error to predict the lime at whích the events !iyperglycemia a.nd hypoglycemia occurred, showed a normal distribution (W = 0.99; p = 0.706; and W ⁼0.99; p ⁼0.1 9, for hyperglycemia and hypoglycemia events, respectively) (Fig. 6B).

To evaluate the accuracy of oi.:r modeL the tecording of one patient was assessed 6C). Toe data revealed füat úe real hyperglycemía was 395.9 mg/dL compared to the predictíon, which was 388.3 mg/dL 6C). Along v,::ith this, the real hypoglycemia was 70.16 mg/dL, the pre::icted l:ypoglycemia was 68.0 mg/dL (Fig. 6C). On the other hand, the real at which hyperglycemia and hypoglycemia events were 28:13 and 08:22 respectíve'y, comparod to the predicted time of

27:38 and 08:01 :iours, hyperglycemia and hypoglycemia, respectively (Fig. 6C).

28

upna

U111verstdad Publica de Navarra Nofmooko Unibertsitate F\.tJlikoo

(Fig.

m

of

61 data

and hour hours,

(Fig.

(35)

A Maximum predicted glycemia

Fllt',I,()ª 4.S57 p < 0.001

o+-�� O 100 200 300 "ºº 500 Glycemia (mg/dL) Mínimum predicled glycemia

o

R'�0.98 F,.:.,=63.31 p < 0.001 0-+---

0 100 200

Glycem1a (mgldl)

e

-·º

lb

� � 30 e: u

., -:) O 20

O" �

., .,

U: -D 10 _{- o • -}

� l

• W= 0.98 ^{p;; O l5}

-.1\

·100 -50 O 50 100 Error (mgldl)

.;;-

ªº

Lb

>, ., 60 ., g � � o "°CT � ., ., • .

U: -g 20 W•0.9S

� 11 p• o 08

- o •

-100 -50 O 50 100 Error (mg/dL)

B 1-lour of maximum oredicled glycemia

ri ·ao� _:=-_.,,

i 2-- o .

"' .,

'"O > _g ., • ecr ,· 1>= o 99

� °8 . F., 1,,1=243.1

5 ·;::

-

^" ² p <. 0.00�

o.. -9; o ' ' O :C A 6 € 10 Sampling (period every 5 min) Hour of mínimum oredicled glycemia

6 12 16 20 2> 32 36 'º « <8 52

_20Lk

>- en \5 u "' e u

�o ,o

� � 5 -�

- w�o.s, LL § .s. ^o-'l -2 O 2 < ^{1 ,.}fi p⁼07()15 Error (period every 5 min)

20

¡

^{W= O!l9}

f' o= 0.19 --4 -2 O 2 � Error (period every 5 min)

Figure 6. Predictive model of maximum and mínimum glycemia and hour of maximum and mínimum glycemia status in TlD patients. A: (left panel) Scatterplot between real maximum glycemia and predicted maximum glycemia and between real minimum glycernia and predicted minimum glycernia. (right panel) Distribution of error of the model to maximum and minimum glycemia. The error was normal to maximum and mínimum glycemia status. B: (left panel) Hour of maximum and minimum predicted glycemia. (right panel) Distribution of error of the model to hour of maximum and mínimum glycemia. The error was normal to both models. C: representative recoding of CGM and mean, mínimum and maximum predicted glycemia of one DM patient. Horizontai segmented line represents mínimum and maximum predicted glycemia. Venical segmented line showed time of mínimum (tmin) and maximum (t^max) real and predicted glycemia. Note that real and predicted glycemia are closely related.

29

upna

Universidad Públ!ca de Navarra NafUrrooko untiertMo.ta F\tll1koo

500~

E - 400

~i

³⁰⁰

~E 200

c.,- ,oo o

28

(36)

C(mdusion

Thís study shows that throug:1 CGM data. patients with Tl D ex:hibited mainly a glycemic variabífüy v.ith e specific oscillatory pattern that has specific circadian characteristícs for each patíent. Therefore, it is possible that the oscillatory pattem reveals sensitíve time range, which could help the curren, system based on machine learning and/or artificial intelligence, mc;iwJug new osciilatory frequencies, which tcrn could ccntribute to im,Jrove automated insuiin deliver system in this critical population.

30

upna

m the