GLOW-ENV: A Dual-Data IoE-Based Approach for Integrating Glucose

and Environmental Data into a Diabetes Recommendation System

Juan F. Gait

an-Guerrero

1 a

, Carmen Mart

ınez-Cruz

2 b

, Jos

e L. L

opez

1 c

, Francisco Mata-Mata

1 d

Luis G. P

erez-Cord

1 e

, Jos

e-Mar

ıa Serrano

1 f

, Juana M. Morcillo-Mart

ınez

3 g

Angeles Verdejo-Espinosa

4 h

, Juan C. Cuevas-Mart

ınez

5 i

, Raquel Viciana-Abad

5 j

Pedro J. Reche-L

opez

5 k

, Jos

e M. P

erez-Lorenzo

5 l

, David D

ıaz-Jim

enez

1 m

and

Macarena Espinilla

1 n

Department of Computer Science, University of Ja

en, 23071, Ja

en, Spain

Department of Languages and Computer Systems, University of Granada, 18071, Granada, Spain

Psycology Department, Faculty of Social Work, University of Ja

en, 23071 Ja

en, Spain

Electrical Engineering Department, University of Ja

en, 23071 Ja

en, Spain

Telecommunication Engineering Department, University of Ja

en, 23071 Ja

en, Spain

Keywords:

Internet of Everything, Diabetes, Continuous Glucose Monitoring, Environmental Factors, Artiﬁcial

Intelligence Models, Recommendation Systems, Mobile Health Applications, eHealth, Healthcare

Information Systems.

Abstract:

This paper introduces GLOW-ENV, an intelligent Internet of Everything (IoE)-driven mobile application de-

signed with the objective of integrating real-time glucose monitoring data and environmental metrics to en-

hance diabetes care and management. The proposed IoE ecosystem integrates a continuous glucose monitoring

with a personalized Artiﬁcial Intelligence model designed to predict glycemic ﬂuctuations in a near-future. Ad-

ditionally, GLOW-ENV integrates a rule-based recommendation system to dynamically adapt its suggestions

based on contextual glucose and environmental data. This framework advances personalized diabetes care,

contributing to their progression and well-being offering valuable insights and improving decision-making.

1 INTRODUCTION

Diabetes diagnosis has emerged in recent decades,

positioning it as a global concern within societies in

https://orcid.org/0009-0007-6872-1401

https://orcid.org/0000-0002-8117-0647

https://orcid.org/0000-0003-2583-8638

https://orcid.org/0000-0001-6099-0016

https://orcid.org/0000-0002-0753-6460

https://orcid.org/0000-0001-5046-0724

https://orcid.org/0000-0002-5271-6145

https://orcid.org/0000-0002-7998-553X

https://orcid.org/0000-0003-3749-5986

https://orcid.org/0000-0003-2545-7229

https://orcid.org/0000-0002-5417-3551

https://orcid.org/0000-0002-5286-8026

https://orcid.org/0000-0003-1791-4258

https://orcid.org/0000-0003-1118-7782

general, due to the profound risk factors that are de-

rived from it. According to the International Diabetes

Federation (IDF), this disease is medically deﬁned

as a chronic condition that appears by the time the

pancreas is no longer capable of producing insulin or

when the body is unable to manage it in a proper man-

ner (International Diabetes Federation, 2024). Essen-

tially, there are three main different ways in which

diabetes can be classiﬁed. First, type 1 diabetes is the

condition of the disease that might be developed at

any time while requiring insulin supply to guarantee

patients’ survival. Type 2 diabetes is the one that is

more often diagnosed in adults, accounting for nearly

90% of all diabetes cases. Last but not least, ges-

tational diabetes appears by the time high glycemia

values are registered during pregnancy, whose conse-

quences might be reﬂected on both, the mother and

child. In any form, statistics are critical; the IDF has

reported that there are around 537 million adults that

Gaitán-Guerrero, J. F., Martínez-Cruz, C., López, J. L., Mata-Mata, F., Pérez-Cordón, L. G., Serrano, J.-M., Morcillo-Martínez, J. M., Verdejo-Espinosa, Á., Cuevas-Martínez, J. C.,

Viciana-Abad, R., Reche-López, P. J., Pérez-Lorenzo, J. M., Díaz-Jiménez, D. and Espinilla, M.

GLOW-ENV: A Dual-Data IoE-Based Approach for Integrating Glucose and Environmental Data into a Diabetes Recommendation System.

DOI: 10.5220/0013307100003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 797-804

ISBN: 978-989-758-731-3; ISSN: 2184-4305

797

are currently suffering from diabetes –3 in 4 adults

live in low- and middle-income countries–, a number

that is expected to rise in the upcoming years.

In parallel, the society has witnessed an impres-

sive ﬂow of information generated through the uti-

lization of technologies such as the Internet of Things

(IoT), as characterized by the capacity of characteriz-

ing different phenomena that can be monitored. Ad-

ditionally, the reduction in the cost of devices and

storage components has paved the way to a social

panorama of massive interconnection, with a growing

use of the IoT technology that often goes unnoticed

in the environment. Many different and diverse ﬁelds

have taken advantage of this paradigm. Speciﬁcally,

in the healthcare realm, sensors are acquiring a no-

table importance in human beings’ safeguarding and

care. The information retrieved by this sort of sys-

tems is of high value and risk, not only for the need to

ensure data privacy and safety, but also for the thor-

ough treatment of the information, with the objective

of not comprising the health state or the well-being of

the patients involved. Different applications are being

developed and some are cited next: hospital bed oc-

cupancy for accelerating the process of taking patients

from the emergency room to an inpatient unit (Afﬂeck

et al., 2013), vital signs monitoring to prevent future

complications (Chakraborty et al., 2020), or activity

recognition approaches like (Schmitter-Edgecombe

et al., 2024), (Lupi

on et al., 2020) and (L

opez Ruiz

et al., 2024). Beyond this, the intersection of people,

processes, data and things is named after the Internet

of Everything (IoE), materializing by the time real-

time data coming from different and diverse sensors

is processed to aid “automated and people-based pro-

cesses” (Bradley et al., 2013).

In the domain of diabetes, commonly used de-

vices refer to traditional glucometers. This instru-

ments consists of a the device, a lancet with a needle,

and a test strip. Therefore, the objective is to mea-

sure capillary blood glucose carrying out a tap over

the patient’s ﬁnger, extracting a drop of blood, plac-

ing it on the test strip to measure the level of glucose

using the device. This process is characterized to be

harmful, while necessitating the repetition of it many

different times to be able to fully observed glycemic

dynamics. As a consequence, technological advance-

ments have paved the way for developing continuous

glucose monitoring sensors. These devices are capa-

ble of solving the problem of the traditional ones, by

their application over the patient skin to sample the

glucose level from the interstitial ﬂuid. This measure-

ment is performed through the utilization of a ﬁlament

from which an enzymatic reaction takes place to gen-

erate the glucose value. An example of this sensor

is the Freestyle Libre 3 device (Abbott, 2022), which

has been observed to be used widely in the diabetic

population, or other similar versions.

Beyond this, glucose monitoring has allowed re-

searchers to further investigate in glucose ﬂuctua-

tions, in a attempt to better understand patients’ dy-

namics, rather than assuming only those factors that

are directly associated to a patient’s daily life (e.g.:

physical effort, insulin administration or food intake).

In these sense, different approaches evolve the study

evaluation of the correlation between weather and en-

vironmental factors on people suffering from, or in

the risk of, type 1 diabetes, type 2 diabetes or gesta-

tional diabetes. The main purpose of these proposals

is to potentially determine how external factors varies

blood glucose concentration, or even how the body

performs when being exposed to certain contextual

circumstances. According to type 1 diabetes, (Chi-

acchiaretta et al., 2024) determine that there is neg-

ative correlation between temperature and blood glu-

cose levels, being even more pronounced in extreme

heat conditions; a seasonal variation is observed de-

riving in stating that glycemia is more likely to be

stable in hotter months with an increased risk of hy-

poglycemia, while showing higher variability during

colder months, being the latter correlated to a higher

risk of hyperglycemia. These ﬁndings are also at-

tested by (Vodrahalli et al., 2023) and (Richardson

et al., 2020) in the case of type 2 diabetes patients.

The ﬁndings sustain that these patients are also af-

fected by environmental factors like temperature and

air quality, showing that high temperatures enhance

glucose absorption while air pollution might worse

glucose metabolism under certain conditions, by in-

creasing insulin resistance (Vallianou et al., 2021).

Gestational diabetes and environmental risk factors

are addressed in (Preston et al., 2020) or (Elshahidi,

2019), where it is highlighted how elevated ambient

temperatures are associated with increased odds of

this condition. Taken together, these works acclaim

the need for developing a decision support system that

continuously inform diabetic patients about the ambi-

ent exposure they might experienced, together with

the possible actions to take to prevent potential con-

sequences.

Alongside these advancements, healthcare in-

formation technology systems are essential to pro-

vide personalized care and attention over the mon-

itored patients Consequently, different approaches

have been developed to continuously monitor glu-

cose. (Murakami et al., 2006) proposes the develop-

ment of a system capable of being implemented in

cardiac patients in the intensive care unit, establish-

ing a client-server architecture. (Helal et al., 2009)

HEALTHINF 2025 - 18th International Conference on Health Informatics

798

propose the design of a smart home-based platform to

monitor, analyze and even alter patient behaviors, im-

proving healthcare efﬁciency through IoT devices. To

the best of our knowledge, there is no technological

platform that supply diabetic populations with a mon-

itoring tool that suggest actionable behavior based on

glucose and ambient data. Nonetheless, these valu-

able digital tools are the premises of the main contri-

butions of this work:

1. Deﬁnition of an IoE-based real-time monitoring

architecture for information data retrieval of pa-

tients’ glycemia and environmental conditions.

2. Construction of a simpliﬁed artiﬁcial neural net-

work model to predict glucose values in a near-

future, being tailored to a concrete patient.

3. Establishment of a predeﬁned rule-based recom-

mendation system integrating IoT devices data to

enhanced diabetic patients’ quality of life.

The rest of this paper is organized as follows. Sec-

tion 2 describes the designed architecture, together

with the procedures associated to it, a step-by-step de-

scription of the system developed. While Section 3

brieﬂy evolves the interaction between the end-user

and the graphical interface of the monitoring frame-

work in the form of a mobile application, Section 4

contemplates the limitations of it and the future direc-

tions that may apply. Finally, Section 5 includes the

concluding remarks of this proposal.

2 IoE ECOSYSTEM

ARCHITECTURE DESIGN

This section is dedicated to the speciﬁcation of all in-

terconnected components of the IoE architecture de-

veloped. All of comprised elements come to the res-

cue after reviewing the different approaches and ﬁnd-

ings found in the literature. Therefore, the proposed

system aims to promote the development of a recom-

mendation system for diabetic patients based on en-

vironmental and continuous glucose monitoring data,

both in real-time, providing a clear and straightfor-

ward response based on the observed phenomena.

In this regard, this work ﬁrst contemplates the def-

inition of a connection schema between a proprietary

server and the sensor for continuously monitor glu-

cose values. This process derives in the implemen-

tation of a RESTful API service that allows different

operations for further treatment of the collected data

(Section 2.1). With the purpose of not only provid-

ing the diabetic patient with an application capable

of graphing the observed glucose values, our system

Listing 1: Example of JSON formatted message from the

glucose sensor.

{

” i d ” : ” 6739 c e e 3978132 d f c 8 bca 3 0 a ” ,

” t i m e s t a m p ” : ” 1 7 3 1 8 4 1 7 6 3 7 4 1 ” ,

” s g v ” : 75 ,

” t y p e ” : ” sgv ” ,

” u t c O f f s e t ” : 6 0

}

opts for the utilization of these measures to generate

an Artiﬁcial Neural Network capable of predicting ex-

pected measures in a close future (Section 2.2). In

parallel, due to the different relationships established

in scientiﬁc research between glycemia dynamics and

environmental conditions, another connection is es-

tablished with an external API for this data retrieval

(Section 2.3). Ultimately, the developed application

gather all the knowledge obtained by processing the

raw data for establishing the recommendation that

should be communicated to the patient (Section 2.4).

The whole architecture is illustrated in Figure 1.

2.1 Device Layer and Operational

Workﬂow of the RESTful API

This layer concerns the glucose values sampling from

an IoT glucose sensor, acquiring a signiﬁcant role in

providing a real-time solution in the domain handled

in this work. Recently commercialized sensors con-

sist of a ﬁlament or electrode for being placed subcu-

taneously, typically, on the upper arm of the patient.

These devices measure interstitial glucose, which dif-

fers from capillary glucose only in terms of a slight

delay during glycemia ﬂuctuations; the system’s ar-

chitecture of this work is tested with the utilization of

the Freestyle Libre 3 sensor (Abbott, 2022). To per-

form the sampling, Bluetooth Low Energy (BLE) and

Near Field Communication technologies are needed,

together with a third-party application named xDrip+

and a smartphone.

In detail, the process is initiated by the time the

sensor is paired with a mobile device via NFC con-

nection, allowing for the interstitial glucose values

collection once the BLE connection becomes active.

Next, data transmission for its storage and process-

ing is performed by the xDrip+ application. The

latter allows for the implementation of a proprietary

RESTful API Service, enabling request-level proce-

dures without establishing a dependency on other

more restricted services that do not provide multi-

patient monitoring.

The data is provided in JSON-formatted messages

(see Listing 1), mainly containing the timestamp and

GLOW-ENV: A Dual-Data IoE-Based Approach for Integrating Glucose and Environmental Data into a Diabetes Recommendation System

799

API REST

send/request

return

Database

Data collection

running

+ NFC

response

request

response

request

GLOW-ENV APP

GRU Model

Rules

Proprietary server

Open-Meteo API

Figure 1: Description of the proposed system architecture.

the glucose value in mg/dL. The API service directly

communicates with the smartphone for data sampling

(GET) and the database (POST, GET) for persistent

storage and data consultation, being the latter mainly

used for its transfer to the application layer.

2.2 Artiﬁcial Neural Network for the

Prediction of Glucose Values

The storage of the samples in the database have al-

lowed for the obtaining of a large dataset compre-

hending glucose values for a single patient in a time

interval of 405 days

. At ﬁrst, the validation of the

dataset in terms of validity and diversity is performed

through the computation of linguistic protoforms; dif-

ferent linguistic labels are deﬁned to clearly observe

how different patterns and features appear along the

dataset. As a consequence, distinct membership func-

tions are deﬁned for the linguistic labels associated

to temporal day intervals (see Table 1), glucose val-

ues (see Table 2; values are speciﬁed according to the

WHO (World Health Organization, 2024)), and quan-

tiﬁers (see Table 3). This procedure takes advantage

of the Fuzzy Logic (Zadeh and Aliev, 2018) theory.

Table 1: Speciﬁcation of the different linguistic labels for

temporal day intervals (in hours).

Day interval labels Membership Function

At night z-shape(−∞, −∞, 6, 8)

In the morning trapmf(6, 8, 12, 2)

In the afternoon trapmf(12, 2, 8, 10)

At the end of day s-shape(8, 10, ∞, ∞)

During the daytime s-shape(7, 9, ∞, ∞)

The different linguistic labels permit the creation

of type 2 protoforms in the form R (day interval la-

bel) Q (quantiﬁer) glucose values are S (glucose la-

bel). The activation of the protoform within the ana-

lyzed time series comes from Equation 1 as speciﬁed

This dataset is available in

https://zenodo.org/records/10713570

Table 2: Speciﬁcation of the different linguistic labels for

glucose registered values.

Glucose label Membership Function

Low z-shape(−∞, −∞, 75, 80)

Medium trapmf(75, 80, 125, 140)

High s-shape(125, 140, ∞, ∞)

Table 3: Speciﬁcation of linguistic labels for quantiﬁers.

Quantiﬁer label Membership Function

Few s-shape(10, 30, ∞, ∞)

Many s-shape(40, 60, ∞, ∞)

Most s-shape(60, 80, ∞, ∞)

Almost all s-shape(80, 100, ∞, ∞)

in (Zadeh and Aliev, 2018) (note that µ represent the

distinct membership functions).

θ(R , Q , A , S) = max



∑

(µ

∗ µ

)(c

)

∑

)



, (1)

where c

corresponds to the evaluated instance of the

dataset and ∗ denotes minimum.

As a result, it has been observed that nearly the

92% of all possible protoforms are activated, ensuring

the diversity of the individuals in the dataset. At this

point, it must be highlighted that a 15% of the time

series are excluded from the dataset in order to test the

ﬁnal generated outputs within the system, also based

on protoform activation and pattern identiﬁcation as

attested by (Martinez-Cruz et al., 2021) or (Pel

aez-

Aguilera et al., 2019).

The selected model to predict glucose values in

a close future is a recurrent neural network, particu-

larly, a Gated Recurrent Unit (GRU), due to its capac-

ity to combine an update and a reset gate to control the

ﬂow of information. This kind of neural networks ne-

cessitate fewer parameters, being simpler and faster

to train than other models. Additionally, as glucose

time series data may exhibit rapid ﬂuctuations accord-

ing to different activities (e.g.: food ingestion, insulin

supply, weather conditions and/or physical activity), a

GRU architecture can effectively capture without the

HEALTHINF 2025 - 18th International Conference on Health Informatics

800

need of a more complex one. Furthermore, the con-

sideration of the ﬁnal model to be intended for mobile

devices, derives in efﬁciency as a key feature to look

up a this stage. Then, the proposed recurrent neural

network is established in Algorithm 1.

Algorithm 1: GRU Model for Glucose Time Series Fore-

casting.

Data: X = {x

, x

, . . . , x

}: Time series

Result: ˆy

: Predicted glucose values

Parameters: Window size w, GRU units

, u

, dropout rate p, learning rate α;

= {x

, x

t+1

, . . . , x

t+w−1

}, y

= x

t+w

train

, X

val

, X

test

with ratios r, s, t

= GRU(X

, u

), h

∈ R

′

= GRU(h

, u

), h

′

∈ R

′′

= Dropout(h

′

, p)

ˆy

= Dense(h

′′

)

optimizer(α)

Loss function: L =

∑

i=1

− ˆy

)

Minimize: min

train

)

[L ]

Validation loss: L

val

= E

val

)

[L ]

Test loss: L

test

= E

test

)

[L ]

Firstly, the hyperparameters are deﬁned in terms

of the number of samples used for prediction (win-

dow size), the dropout rate for reducing overﬁtting

by randomly deactivating a fraction of neurons dur-

ing training, and the learning rate to control the con-

vergence while avoiding local optima; the GRU units

correspond to the number of neurons in the ﬁrst, the

second and successive layers within the network. Af-

ter that, the dataset is split into three subsets, which

refer to the training (70%), validation (15%) and test

(15%) datasets, respectively. The model then starts

to be constructed by the deﬁnition of the ﬁrst GRU

layer, used to transform the input data to capture tem-

poral dependencies within it; the second one performs

a further processing with a reduced dimensionality, to

extract deeper temporal dynamics in data. Finally, a

dropout layer is employed to prevent overﬁtting by

deactivating some neurons prior to the ﬁnal prediction

of the glucose values. In this process, the Amam Op-

timizer is employed alongside the loss function, the

minimization of the loss function, and the validation

loss function to optimize the model parameters during

training.

As a result, the designed GRU model is expected

to estimate glucose values in a close future, by quanti-

fying and labeling them according to the membership

functions provided in Tables 1–2–3, with the objec-

tive of providing a clear message to the patient show-

ing the expected values to happen.

An experimental process have been carried out to

determine the speciﬁcations of the ﬁnal model. Con-

cretely, this work implements the GRU model incor-

porating two layers with 64 and 32 units, respectively,

followed by a dropout layer with a rate of 40% to mit-

igate overﬁtting. The Adam optimizer is applied with

an initial learning rate of 0.001, and the mean squared

error (MSE) is selected as the loss function. The train-

ing process runs for a maximum of 35 epochs with

early stopping set to a patience of 10 epochs and a

learning rate reduction factor of 0.5 when improve-

ments are no longer seen.

The ﬁnal model setup results of training demon-

strate a substantial reduction in both the training and

validation losses over the 35 epochs, highlighting the

GRU model’s ability to effectively learn temporal de-

pendencies and patterns in the dataset. The train-

ing loss began at 8384.49 and steadily decreased to

366.72 by the ﬁnal epoch, while the validation loss

showed a similar trend, reducing from 8303.06 to

239.60. Both metrics are stable after approximately

20 epochs. An output example is graphically shown

in Figure 2 for the prediction of one hour values, as

each sample is expected every 5 minutes.

Figure 2: GRU model performance in predicting glucose

values.

The results reveal how the model can effectively

capture the general trend of the future values. Even

though notable deviations may occur, it must be clari-

ﬁed that further improvement of the model is not con-

sidered necessary as the expected and real values are

established within the same label (Table 2), thereby

not necessitating higher complexity.

2.3 Environmental Data Collection

In this section, the environmental data corresponding

to the temperature and the air quality is collected. As

the nature of these phenomena reveals little variation

in their recordings in a near future, the data utilized is

the one that is present in the moment when the request

GLOW-ENV: A Dual-Data IoE-Based Approach for Integrating Glucose and Environmental Data into a Diabetes Recommendation System

801

Listing 2: Example of JSON Response from Open-Meteo.

{

” l a t i t u d e ” : 40 . 4 3 7 5 ,

” l o n g i t u d e ” : −3 . 6 8 7 5 ,

” t i m e z o n e ” : ”GMT” ,

” h o u r l y u n i t s ” : {

” t e m p e r a t u r e 2m” : ” degC” ,

”pm10 ” : ”mu−g /mˆ 3 ” ,

”pm2 5 ” : ”mu−g /mˆ 3 ”

} ,

” h o u r l y ” : {

” t i m e ” : [

” 2024−11−03T00 : 0 0 ”

] ,

” t e m p e r a t u r e 2m” : [

16 . 3

]

}

is made, in contrary to what happens with glycemia,

due to its fast ﬂuctuation according to both, internal

and external factors of the patient.

To achieve this, the Open-Meteo API (Open-

Meteo, 2024a) has been selected for the real-time

environmental data retrieval. This API provides ac-

cess to different environmental phenomena, including

weather and air quality, providing the users the ca-

pability to request both, historical and real-time data,

making it an ideal choice for this study, as being pro-

vided for non-commercial use, including the utiliza-

tion of the service for public research (Open-Meteo,

2024b). For a better understanding of outdoor condi-

tions on behalf of diabetic patients, as speciﬁed in the

literature review, Open-Meteo API incorporates the

metrics of temperature, PM10 and PM2.5 (the latter

ones corresponding to air-suspended particles), which

inﬂuence in patient health, if being exposed for a con-

siderable amount of time. The data collection pro-

cess involves constructing requests with different pa-

rameters. In our case, the utilized ones are related to

the geographic coordinates of the patient (latitude and

longitude) the date and the metrics of interest. These

requests return structured JSON responses, detailing

hourly values for the chosen variables. A sample re-

sponse is presented below:

From this output, the metrics data is ﬁltered to fo-

cus on the recordings that allows a real-time response.

For air quality, PM10 and PM2.5 values are analyzed

to determine the category (i.e.: “good”, “fair”, “un-

health” or “extreme”) based on established thresh-

olds (Sloss and Smith, 2000). On the other hand,

temperature is represented as low, normal or high. At

last, all metrics together, i.e. glucose levels, air par-

ticles and temperature, derives in the generation of a

rule-based recommendation system.

2.4 Application Layer

This layer may be consider the core of the system,

responsible for performing the execution of the neu-

ral network model and the rule-based recommenda-

tion system. In detail, the application located in the

patient’s smartphone is capable of connecting to the

RESTful API service developed for glucose values

retrieval, serving as an input for the prediction model

that is stored locally. This last decision has been made

considering that each patient, due to the simple de-

signed of the network, could be provided with a tai-

lored model, reducing server dependency and satura-

tion in a context of multi-patient monitoring. Along-

side this connection, the application is also able to re-

quest environmental data from the proposed API ser-

vice. Through the combination of these components,

the application layer dynamically adapts its glucose

predictions and personalized recommendations, aim-

ing to ensure context-aware guidance for users.

According to the application view (see Figure 3),

it can be observed how the patient is capable of ob-

serving the collected glucose values in real-time, be-

ing able to interact with his/her glycemic dynamics.

Followed by the graphical representation of the raw

data, a message is provided to the patient, based on

the results thrown by the GRU model for a near-

future prediction of 1 hour, which are subsequently

processed through fuzzy labeling and quantiﬁcation.

Next, the air quality and temperature data is shown,

both in a quantitative and a qualitative manner, to

make it easier and faster to understand. Based on all

the information gathered, a ﬁnal message is given to

the diabetic patient as a recommendation, according

to the literature review and expert knowledge.

3 USER WORKFLOW

In this section, the interaction of the user with the

whole system’s architecture is addressed.

On the one hand, the patient must request access

to the API service for glucose sampling. Therefore,

he/she can be provided with an URL that is neces-

sary for sending the data via the mobile application

(xDrip+). Right after the patient has performed the

connection between the mobile device and the glu-

cose sensor, all the data collected is sent to the server

automatically, allowing for data consultation and re-

trieval through the proprietary API.

On the other hand, once the GLOW-ENV applica-

tion is also running on the patient’s device, as an user,

the person will be able to check the collected glucose

values and, the summarization of the expected one in

HEALTHINF 2025 - 18th International Conference on Health Informatics

802

Patient is logged in

Glucose data is retrieved

Patient has chosen a location

Figure 3: Dashboard of the designed mobile application.

the future. At this point, the patient could type man-

ually his/her current location for environmental data

consultation, being provided with a recommendation

message at last (see Figure 3).

4 LIMITATIONS AND FUTURE

RESEARCH

Since this research aims to enhance the care of di-

abetic patients by promoting personalized, context-

based guidelines through API requests for data re-

trieval, fuzzy labeling and tailored Artiﬁcial Intelli-

gence models, it is evident that addressing the limita-

tions and future research in this topics is essential.

Regarding the API petitions, the real-time recom-

mendation system could incorporate location and date

parameters speciﬁcation, thereby enabling patients to

perform a historical data analysis for a better under-

standing of past events. Additionally, the automation

of the latter mentioned parameters could improve the

system by incorporating an automatic alert function-

ality. Ending up with this point, it is considered inter-

esting to perform a data mining procedure to discover

new relationships between glycemia and environmen-

tal conditions.

Noticeably, the architecture heavily relies on

third-party API services for glucose data sampling

and environmental conditions retrieval, creating a de-

pendency that may introduce potential challenges due

to restrictions imposed by the external providers; ex-

ploring alternatives to reduce dependency on them is

a must in the future.

Lastly, future research could focus on developing

a federated system that integrates a more complex

model on the server side to improve glucose value

predictions, since the current system already incor-

porates local personalized models. Therefore, this

betterment could improve the accuracy of the predic-

tions, which may be required by the time the patient

monitoring integrates personal activities recordings

(f.i.: food ingestion, physical activity or medicine in-

take).

5 CONCLUSIONS

In this proposal, a new intelligent IoE-based mobile

application designed for glucose and environmental

data fusion is presented, with the objective of enhanc-

ing diabetes care, management and prevention.

The system’s architecture holds a sophisticated in-

terconnection of different devices and servers, facili-

tating the collection, processing, and analysis of dis-

tinct data streams. Glucose sampling does not require

direct patient interaction as being performed automat-

ically, while the environmental data will not be avail-

able prior to the speciﬁcation of the parameters on

patients’ behalf. However, the combination of con-

tinuous glucose monitoring data with environmental

metrics, both in real-time, provides a context-aware

approach throughout the utilization of tailored predic-

tions and actionable insights.

GLOW-ENV: A Dual-Data IoE-Based Approach for Integrating Glucose and Environmental Data into a Diabetes Recommendation System

803

ACKNOWLEDGMENTS

This result has been partially supported

by grant PID2021-127275OB-I00, grant

PID2021-126363NB-I00 funded by MICI-

U/AEI/10.13039/501100011033 and by “ERDF

A way of making Europe”, grant PDC2023-145863-

I00 funded by MICIU/AEI/10.13039/501100011033

and by “European Union NextGenerationEU/PRTR”,

and grant M.2 PDC 000756 funded by Consejer

ıa

de Universidad, Investigaci

on e Innovaci

on and by

ERDF Andalusia Program 2021-2027.

REFERENCES

Abbott (2022). Sistema freestyle libre 3.

https://www.freestyle.abbott/es-es/productos/

freestylelibre-3.html. (Accessed on 11/10/2024).

Afﬂeck, A., Parks, P., Drummond, A., Rowe, B. H., and

Ovens, H. J. (2013). Emergency department over-

crowding and access block. Canadian Journal of

Emergency Medicine, 15(6):359–370.

Bradley, J., Reberger, C., Dixit, A., Gupta, V., and

Macaulay, J. (2013). Internet of everything (ioe): top

10 insights from cisco’s ioe value at stake analysis for

the public sector. Economic Analysis, 1.

Chakraborty, S., Bhatt, V., and Chakravorty, T. (2020). Big-

data, iot wearable and mhealth cloud platform integra-

tion triads-a logical way to patient-health monitoring.

International Journal of Engineering and Advanced

Technology, 9(3):388–394.

Chiacchiaretta, P., Tumini, S., Mascitelli, A., Sacrini, L.,

Saltarelli, M. A., Carabotta, M., Osmelli, J., Di Carlo,

P., and Aruffo, E. (2024). The impact of atmospheric

temperature variations on glycaemic patterns in chil-

dren and young adults with type 1 diabetes. Climate,

12(8):121.

Elshahidi, M. H. (2019). Outdoor air pollution and ges-

tational diabetes mellitus: a systematic review and

meta-analysis. Iranian journal of public health,

48(1):9.

Helal, A., Cook, D. J., and Schmalz, M. (2009). Smart

home-based health platform for behavioral monitoring

and alteration of diabetes patients. Journal of diabetes

science and technology, 3(1):141–148.

International Diabetes Federation (2024). What is diabetes.

https://idf.org/about-diabetes/what-is-diabetes/. (Ac-

cessed on 11/09/2024).

opez Ruiz, J. L., Jim

enez Villar, J. M., Alb

ın Rodr

ıguez,

A. P., and Espinilla Estevez, M. (2024). Indoor area

location system using uwb technology and axis-linear

bounding boxes. In IEEE Congress on Evolutionary

Computation (IEEE CEC). IEEE Access.

Lupi

on, M., Redondo, J. L., Sanjuan, J. F., and Ortigosa,

P. M. (2020). Deployment of an iot platform for ac-

tivity recognition at the ual’s smart home. In Inter-

national Symposium on Ambient Intelligence, pages

82–92. Springer.

Martinez-Cruz, C., Rueda, A. J., Popescu, M., and Keller,

J. M. (2021). New linguistic description approach for

time series and its application to bed restlessness mon-

itoring for eldercare. IEEE Transactions on Fuzzy Sys-

tems, 30(4):1048–1059.

Murakami, A., Gutierrez, M., Lage, S., Rebelo, M.,

Guiraldelli, R., and Ramires, J. A. F. (2006). A con-

tinuous glucose monitoring system in critical cardiac

patients in the intensive care unit. In 2006 Computers

in Cardiology, pages 233–236. IEEE.

Open-Meteo (2024a). Free open-source weather api. https:

//open-meteo.com/. (Accessed on 10/31/2024).

Open-Meteo (2024b). Terms — open-meteo.com.

https://open-meteo.com/en/terms. (Accessed on

11/19/2024).

Pel

aez-Aguilera, M., Espinilla, M., Fern

andez, M., and

Medina, J. (2019). Fuzzy linguistic protoforms to

summarize heart rate streams of patients with is-

chemic heart disease. hindawi. Complexity.

Preston, E. V., Eberle, C., Brown, F. M., and James-Todd, T.

(2020). Climate factors and gestational diabetes mel-

litus risk–a systematic review. Environmental Health,

19:1–19.

Richardson, M. B., Chmielewski, C., Wu, C. Y., Evans,

M. B., McClure, L. A., Hosig, K. W., and Gohlke,

J. M. (2020). The effect of time spent outdoors during

summer on daily blood glucose and steps in women

with type 2 diabetes. Journal of behavioral medicine,

43:783–790.

Schmitter-Edgecombe, M., Luna, C., Dai, S., and Cook,

D. J. (2024). Predicting daily cognition and lifestyle

behaviors for older adults using smart home data and

ecological momentary assessment. The Clinical Neu-

ropsychologist, pages 1–25.

Sloss, L. L. and Smith, I. M. (2000). Pm10 and pm2. 5:

an international perspective. Fuel Processing Tech-

nology, 65:127–141.

Vallianou, N. G., Geladari, E. V., Kounatidis, D., Geladari,

C. V., Stratigou, T., Dourakis, S. P., Andreadis, E. A.,

and Dalamaga, M. (2021). Diabetes mellitus in the

era of climate change. Diabetes & Metabolism,

47(4):101205.

Vodrahalli, K., Lyng, G. D., Hill, B. L., Karkkainen, K.,

Hertzberg, J., Zou, J., and Halperin, E. (2023). Un-

derstanding and predicting the effect of environmental

factors on people with type 2 diabetes. In Conference

on Health, Inference, and Learning, pages 545–555.

PMLR.

World Health Organization (2024). Mean fast-

ing blood glucose. https://www.who.int/data/

gho/indicator-metadata-registry#:

∼

:text=Mean%

20fasting%20blood%20glucose. Accessed: 31

October 2024.

Zadeh, L. A. and Aliev, R. A. (2018). Fuzzy logic theory

and applications: part I and part II. World Scientiﬁc

Publishing.

HEALTHINF 2025 - 18th International Conference on Health Informatics

804