REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

RNPS 2487 • ISSN 2708-3411
Vol. 4 • Nro. 3 • julio-septiembre 2023 •e225
eXe215

ARTÍCULO ORIGINAL

ACTUALIZACION SOBRE MODELOS DE INTELIGENCIA ARTIFICIAL,
PARA PREDICCION DE EVOLUCION EN PACIENTES DE COVID-19 EN CUBA

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Eduardo Garea Llano
eduardo.garea@cneuro.edu.cu • https://orcid.org/ https://orcid.org/0000-0001-9101-1589

Evelio Gonzalez Dalmau
evelio.gonzalez@cneuro.edu.cu • https://orcid.org/ 0000-0003-4569-2103

Centro de Neurociencias de Cuba, Cuba

Recibido: 2023-07-03 • Aceptado: 2023-11-06

RESUMEN
En este artículo presentamos una revisión de los modelos basados en Inteligencia
Artificial (IA), orientados a la predicción de manifestaciones graves provocadas
por el virus SARS-COV2. El objetivo de esta revisión fue evaluar los hallazgos
más relevantes publicados entre 2020 y 2023, que puedan servir de base al
desarrollo de un modelo propio ajustado a las condiciones de nuestro país,
caracterizadas por la ausencia de un modelo propio ajustado a la ausencia de una
base de datos de datos clínicos bien estructurada y con la presencia de estudios
radiológicos apoyados por imágenes de rayos X de tórax (CXR) y tomografía
computarizada (CT). Realizamos una revisión sistemática para resumir y evaluar
críticamente los estudios disponibles que han desarrollado modelos de
pronóstico de COVID-19 basados en IA, que predicen resultados de salud, sobre
todo en modelos que utilizan como base las imágenes CXR y CT. Se realizaron
búsquedas en tres bases de datos bibliográficas, para identificar artículos
publicados sobre modelos de pronóstico que predijeran resultados adversos en
pacientes adultos con COVID- 19, incluido el ingreso a la unidad de cuidados
intensivos, la necesidad de ventilación mecánica y la mortalidad.
El estudio demostró que los modelos basados en aprendizaje profundo, que
utilizan imágenes CXR o CT y su combinación con datos clínicos no complejos,
pueden alcanzar un rendimiento significativo en la predicción. Por ello, aquí
proponemos una estrategia para abordar este desafío, según las condiciones de
nuestro país, combinando la clasificación del grado de gravedad de la afectación
pulmonar en imágenes de CXR, datos clínicos de comorbilidades y datos
biográficos.
Palabras clave: COVID-19, modelos predictivos, Inteligencia Artificial, datos
clínicos, datos radiológicos.

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

ABSTRACT
In this work we present a review of models based on Artificial Intelligence (AI)
aimed at predicting serious manifestations caused by the SARS-COV2 virus.
The objective of this review was to evaluate the most relevant findings published
between 2020 and 2023 that can serve as a basis for the development of our own
model adjusted to the conditions of our country characterized by the absence of
a well-structured clinical database with the presence of radiological studies
based on chest x-ray (CXR) and computed tomography (CT) images. We
conducted a systematic review to summarize and critically evaluate available
studies that have developed AI-based COVID-19 prognostic models that predict
health outcomes, especially models based on CXR and CT images. Three
bibliographic databases were searched to identify published articles on
prognostic models predicting adverse outcomes in adult patients with COVID-
19, including intensive care unit admission, need for mechanical ventilation,
and mortality.
The study demonstrated that Deep Learning-based models using CXR or CT
images and their combination with non-complex clinical data can achieve
significant prediction performance. On this basis, in the work we propose a
strategy to address this challenge under the conditions of our country by
combining the classification of the degree of severity of lung involvement in
CXR images, clinical data on comorbidities and biographical data.
Keywords: COVID-19; Predictive models; Artificial intelligence; Evolution;
Clinical and radiological data.

INTRODUCTION

With the decrease in COVID-19 cases in our country due to the development of our own
vaccines and their massive application, as well as the health measures adopted, this disease
still remains a national concern. The Cuban Neuroscience Center (CNEURO), as part of a
project leaded by it and within the multidisciplinary scientific research program
"Biotechnology, Pharmaceut ica l Industry and Medical Technology" (put in place by the
Cuban government to respond to the CONVID19 pandemic), is developing a model based
on AI for the prognosis of patient evolutio n affected by the disease. CNEURO is also
improving the use of available CXR images in the diagnosis, prognosis and monitoring of

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

Covid-19 patients. This effort integrates specialists from the Cuban Society of Imaging,
MINSAP and two Cuban universities (Technological University of Havana and Central
University of Las Villas). The project includes the validation of artificial intelligence
protocols and algorithms for predicting the trajectory of COVID-19 disease by combining
patient health records (clinical data) and chest images (CXR, CT). Additionally, the timely
translation of AI-related research into clinically validated and appropriately regulated
systems that can be used in a real-life hospital environment.

As previous results obtained in this research, there are several works published by
researchers of this project that have contributed to obtaining tools for the diagnosis and
evaluation of the disease. In Portal et al (2022), a model was presented to automatically
classify COVID-19 from CXR images. For this, the architecture of a resnet34 CNN was
used with input images of size 512 × 512 pixels. Furthermore, training of the networks was
limited to the internal regions of the lungs, using a segmented image and patch partitioning.
This model demonstrated together with human specialists its usefulness in the identification
of COVID-19 in clinical settings.

In Garea et al (2021a, 2021b, 2022, 2023), a model is presented and refined to estimate the
degree of lung involvement and its severity in CXR images in patients diagnosed with
COVID-19 in an advanced stage of the disease. The index is obtained from a method that
combines image quality evaluation, digital image processing and deep learning for lung
region segmentation.

As another important result of this project, in February 2023, the health registration was
obtained by the Cuban regulatory authority (CECMED) of the Rx-COVID 19 system.
Automated and weighted multiplatform software to record chest X-ray and CT and facilitate
the diagnosis of COVID. -19. In this system, a significant amount of radiological data of
patients affected by COVID-19 in our country has been documented.

On this basis, the project proposes to move to a higher phase by developing a computational
tool for the prognosis of evolution towards adverse outcomes in adult patients with COVID-
19. The need to continue developing new technological capabilities for the evaluation of
patient follow - up has been identified in the face of resource limitations in the current Cuban
context to respond effectively to similar challenges in the future. This development aims at
a more integrative and interdisciplinary approach to prognostic evaluation protocols so that
patients admitted with COVID-19 have a better chance of survival and improve their health
status by optimizing protocols and treatments.

In this work we will carry out a systematic literature review (SLR) with the following
objectives:

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

(1) Systematically review the models that have been developed to predict COVID-19
outcomes based on the use of RX, CT images and clinical data; (2) Analyze the most relevant
predictive models that are currently available; (3) Synthesize and extract useful results and
conclusions about the prediction models of COVID-19 outcomes that can serve as a basis
for the development of our own predictive model adjusted to the conditions of our country
(characterized by the absence of a well-known structured clinical database and with the
presence of radiological studies based on CXR and CT images given their low cost and the
portability of the existing equipment in healthcare centers).

The work is structured as follows. The "Methodology" Section provides a description of the
method of searching and selecting papers for this review and present various classifications
of the selected papers. In the "Results and Discussion" section present a detailed analysis of
the 6 significant selected works. On the basis of the analysis of the performance of the
studied methods we carried out a general discussion. Finally we offer the conclusions of the
work.

METHODOLOGY
To carry out this review study we used the SLR methodology. SLR is a methodology used
to evaluate studies through primary analyzes on published articles based on specific search
criteria. An SLR is performed based on previous similar studies through a systematic review
in national and international publication databases. The ultimate purpose of the SLR is to
summarize the studies conducted and identify advantages and limitations between previous
studies and current studies.

According to Okoli (2015), SLR is “a systematic, explicit, detailed and repeatable approach
to identifying, evaluating and analyzing the body of existing work carried out by researchers,
academics and practitioners”. On the other hand, Tranfield et al. (2003), considers SLR a
“fundamental scientific activity”. In Moher et al. (2009) a checklist of Preferred Reporting
Items for Systematic Reviews and Meta-Analyses (PRISMA) was presented.

The research period for this study was from December 2020 to May 2023. This study was
conducted in 4 phases: (1) the development of literature search strategies, (2) the formulation
of inclusion and exclusion criteria, (3) quality evaluation, and (4) analysis and conclusion.

Research Questions

For the development of this documentary research, we stated a research question (RC) to
achieve a clear definition of the main topic. The motivation and frequently asked questions
of this study were as follows:

Motivation: Identify models based on artificial intelligence for the prediction of COVID-19

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

outcomes and that use RX images, CT and clinical data or their combination.

RQ1: What factors support the prediction of COVID-19 outcomes based on artificial
intelligence techniques?

RQ2: What methods and techniques do the models identify to predict COVID-19 outcomes?

Inclusion and exclusion criteria

Current search engines provide a high level of retrieval, which leads to a large number of
irrelevant resources being retrieved. Therefore, to obtain effective results, we followed a
systematic search strategy. This stage of SLR examines the literature to find relevant
literature based on particular criteria. In this study, 3 inclusion and exclusion criteria were
adopted to identify relevant content and restrict irrelevant content. The first inclusion
criterion was the type of document: only published documents were included, excluding
manuscripts under review and unpublished ones. Domain (i.e., the topic area identified for
the study) was the second selection criterion; we then included papers containing predictive
models developed or used in the COVID-19 domain, while other papers were excluded. The
last selection criterion was the language in which the document was published. To avoid
confusion and complexity related to translation, only documents available in English were
included, while documents in other languages were excluded (Table 1).

Table 1. Inclusion exclusion criteria

Criteria Inclusion Exclusion
Document type Published documents Under review, unpublished or upcoming

documents

Domain Predictive AI models of
COVID-19 outcomes

Other predictive models of COVID-19
Other predictive models of COVID-19
outcomes not based on AI

Language English Other than English

Databases and search strategies
The terms were searched in several databases (Google Scholar, Scopus, Publish or Perish
and Web of Science [WoS]). The search terms are as follows: predictive AI models, COVID-
19 outcomes, Coronavirus, SARS-CoV, SARS-CoV-2, healthcare, healthcare system,
survival model, healthcare. Various combinations of search terms were used to retrieve
resources in particular databases. Some of the search strings used are the following:
“Predictive AI models” and “COVID-
19 outcomes”; “COVID-19 RX and CT Datasets” and “Prediction Modeling”; “Predictive

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

analysis” and “COVID-19 clinical data”.
After applying the inclusion and exclusion criteria, 574 documents of original articles and
16 relevant reviews were recovered. Therefore, a total of 590 documents went to the second
stage of scrutiny and quality evaluation (Table 2). The percentages of articles from various
databases in the initial, selection and acceptance stages of the document selection process
are illustrated.

Table 2. Document selection

In the initial phase, of the total number (i.e., 590 documents) of retrieved documents, Google
Scholar accounted for 62%, Scopus for 23%, and WoS for 15% (Figure 1A). After initial
screening, 35 papers were included for further consideration. During the selection phase,
48% of the initially included documents were retrieved from Google Scholar. Of the
remaining 52%, Scopus and WoS had a share of 26% each (Figure 1B). Of the total accepted
documents, 44% were retrieved from Google Scholar, 39% from Scopus, and 17% from
WoS (Figure 1C).

Figura. 1. Database's percentage (May 2023). (A) Document selection (initial). (B)
Document selection (screened). (C) Document selection (included). Document selection was
carried out based on selection criteria.

Database Initial Screened Accepted

Google Scholar 366 17 8

Scopus 136 9 7

Web of Sciences 88 9 3

Total selected 590 35 18

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

A total of 16 relevant review articles published in the period 2020 to 2023 were detected
(Albahri, O. S. et al., 2020; Bansal, A. et al., 2020; Jalaber, C. et al., 2020; Kumar, A. et al.,
2020; Naudé,W., 2020; Shi, F. et al., 2020; Swapnarekha, H. et al., 2020; Gallo Marin B. et
al., 2021; Mbunge, E. et al., 2021; Rahimi, I. et al., 2021; Rasheed, J. et al., 2021; Hassan,
A. et al., 2022; Jamshidi, M. B. et al., 2022; Almotairi, K. H. et al., 2023; Buttia, C. et al.,
2023; Heidari, A. et al, 2023).
For the primary review of the 16 identified surveys we found 368 relevant papers based on
title and abstract. From them after an additional review, 166 were excluded en the initial
phase because they were not directly related to predictive models of disease evolution, but
rather to predictive models of evaluation of the development of the epidemic in different
countries or to the direct diagnosis of the disease, 202 paper were chosen for a more in-depth
analysis in the second phase, of which we finally chose 6 reports.
The rest of the initial selected 590 papers (388) 12 were selected in the second phase. All
selected papers are based on the use of chest clinical data alone, X-ray images (CXR) and/or
Tomography (TC) alone, or their combination with clinical data for the prediction of severe
manifestations and mortality from COVID-19.

Quality Assessment and Coding
Quality evaluation of a phenomenon is conducted as a systematic way to avoid biases and
errors. 590 documents were chosen. Based on their titles, these documents were further
analyzed and 35 documents were screened. The content was scrutinized on the basis of the
title, abstract, introduction, and conclusion and 18 studies were finally selected for the
review. The distribute ion by year of publication of the selected papers is shown in Table 3.

Table 3. Distribution by year of publication of the elected report

Year References Outcomes Used data Model
2020 Bae, J. et al.,

2020 (1)
Mechanic

ventilation (MV),
mortality

CXR Machine learning
(ML), Deep

learning (DL)
Liang W. et

al., 2020 (2)
MV,

mortality, ICU.
CXR, clinical and
demographic data

Machine Learning
(ML)

Zhang B. et al.,
2020 (3)

Severe Outcome CT, clinical data ML

2021 Heo J et al.,
2021 (4)

ICU Clinical data ML

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

Lieveld A. et
al, 2021 (5)

ICU, Death,
Hospitalization

CT ML

Quiroz J. C. et
al., 2021 (6)

Severity CT, clinical data ML

Tanboga I. et
al., 2021 (7)

Mortality Clinical data ML

Ho T. et al.,
2021 (8)

Respiratory failure CT, clinical data DL

2022 Ortiz A. et al.,
2022 (9)

Death CT, clinical data DL

Duanmu, H. et
al., 2022 (10)

Death CXR, clinical data DL

Gourdeau, D.
et al., 2022

(11)

MV CXR DL

Matsumoto, T.
et al., 2022

(12)

Mortality and time
to death at
admission

CXR, clinical data DL

2023 Olowolayemo
A. et al., (13)

Infection severity CXR DL

Asteris et at.
(14)

ICU Clinical data DL

Wendland. et
al. (15)

mortality, ICU and
MV

Clinical data ML, DL

Verzellesi et al. .(16) Mortality CT, clinical data ML, DL
Panagiotis et

al. (17)
ICU Clinical data DL

Fu et al. (18) Severe Outcomes CT, clinical data DL

As a first approximation to a taxonomic classification of the selected models, we simply took
the year of its publication (Table 3), then on this basis we carried out the analyzes of various
aspects that allowed us to reach conclusions about the nature of the models, the type of data
from which they are fed to make the predictions and the types of predictions they are capable
of making along with their effectiveness. In this sense we decided to make four large
groupings of the reviewed methods. In the following subsection we present the proposed
taxonomic divisions. The numbers in parentheses correspond to the index assigned to each
report in Table 3.

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

Taxonomic classifications of the studied models
Taking into account the type of outcome prediction:
• Models that estimate the need for mechanical ventilation: (1), (8), (11), (15)
• Models that estimate the need for patient admission to the ICU: (2), (4), (5), (14), (15),
(17)
• Models that estimate the possibility of death of the patient: (5), (7), (9), (10), (12), (15),
(16)
• Models that estimate the possibility of patient evolution towards more severe forms of
the disease: (3), (6), (13).

By the number of predictive outputs:
• Simple models that predict only one condition: (1), (3), (4), (6), (7), (9), (10), (11), (13),
(16), (17)
• Combined models that predict two or more conditions: (2), (5), (8), (12), (14), (15)

By the type of data used in the prediction process:
• Models using only CXR images: (1), (11), (13)
• Models using only CT images: (5)
• Models using only clinical data: (4), (7), (14), (15), (17)
• Mixed models using CXR images and clinical data: (2), (10), (12)
• Mixed models using CT images and clinical data: (3), (6), (8), (9), (16)

By the nature of the algorithms used in the predictive process:
• Models based on classic machine learning algorithms: (2), (3), (4), (5), (6), (7)
• Models based on deep learning algorithms: (8), (9), (10), (11), (12), (13), (14)
• Mixed models that combine classic machine learning and deep learning algorithms: (1),
(15), (16).

RESULTS

For the development of this report, due the limited number of pages, we made a selection
of the 6 more representative papers for a deeper analysis in this work. Papers were chosen
under the criteria of the biggest impact factor of the publication. In the following section
we carry out a detailed analysis of the selected papers in terms of scope and results. The
order of the analyzed reports responds to the classification based on the nature of the

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

algorithms used in the predictive process.
Analysis and discussion of the selected methods
Models based on classic machine learning algorithms
In Zhang, B et al (2020), a study for the early detection of COVID-19 patients is carried
out, for which they developed and validated a nomogram to predict a poor outcome at 30
days in patients with COVID-19. The prediction model was developed on a primary cohort
consisting of 233 patients with laboratory-confirmed COVID-19, and data was collected
from January 3 to March 20, 2020. Significant prognostic factors for poor outcome were
identified and integrated at 30 days to build the nomogram. The model underwent internal
validation and external validation with two separate cohorts of 110 and 118 cases,
respectively. The performance of the nomogram was evaluated with respect to its predictive
accuracy, discriminatory ability, and clinical utility. The model was externally validated in
two cohorts achieving an AUC of 0,946 and 0,878, sensitivity of 100 and 79 %, and
specificity of 76,5 and 83,8 %, respectively.
This study has some limitations. The retrospective nature of the sample may introduce
potential risks of bias in the data. The CT score is subjective with large intra- and inter-
observer variability obtained from the initial CT examination; CT-based radiomics or deep
learning and follow-up CT may provide further prognostic information. This model is not
applicable for patients with critical illness on admission, which may result in inclusion bias.
COVID-19 triage could lead to less severe cases having delayed testing, which would skew
the inclusion set to a more critical status; therefore, the performance of the model may be
overestimated. Finally, the identification of predictors depends on the features available,
the feature selection method used, and the sample size of the studies.
Models based on deep learning algorithms
In Ho. T et al. (2021), the authors developed deep learning models to rapidly identify high-
risk COVID-19 patients based on CT images and clinical data. To do this, data from 297
patients with COVID-19 from five hospitals in Daegu, South Korea were analyzed. A deep
learning model combining an artificial neural network for clinical data and a convolutional
neural network for 3D CT imaging data, was developed to classify these cases as high risk
of severe progression (i.e. , event) or low risk (i.e. no events). A total of 19 clinical features
from were concatenated with a 64- dimensional feature vector from the CT image. The
developed model achieved high classification performance using novel images of
coronavirus pneumonia lesions (ie, 93,9 % accuracy, 80,8 % sensitivity, 96,9 % specificity,
and 0,916 AUC) and lung segmentation images (ie, 94,3 % accuracy, 74,7 % sensitivity,
95,9 % specificity, and AUC score of 0,928) for event versus non-event groups. However,
the study has several limitations. First, it was a retrospectively designed study, where the

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

size of the data set (ie the number of patients) was small. Furthermore, the number of
patients who progressed to the severe stage was relatively small. Therefore, the precision
and sensitivity of the CNN model were based on CT images that can be affected by variation
in unbalanced data sets. Second, only COVID-19 data was included in this study. However,
a true diagnostic model must contain the features to distinguish COVID-19 from other types
of pneumonia. Third, the authors compared their model with three typical 2D models that
were developed in a 3D domain. As these models were designed for 2D imaging, this
comparison did not present the alternatives for the same application domain. Therefore, 3D
context is important to differentiate between evented and eventless COVID-19 structures,
which requires the development of pretrained 3D models.
In Ortiz A. et al (2022), the authors evaluated the value of aggregated chest CT data for
COVID- 19 prognosis compared to clinical metadata. They developed a patient-level
algorithm to aggregate chest CT volume into a 2D representation that can be integrated with
clinical metadata to distinguish COVID-19 pneumonia from chest CT volumes of healthy
participants and participate ants with other viral pneumonia.
Furthermore, the authors present a multitask model for the joint segmentation of different
classes of lung lesions present in lungs infected with COVID-19 with the aim of extracting
highly relevant local features. From this they create prognostic models using the extracted
characteristics along with patient demographics, comparing the performance of the models
using different combinat ions of relevant data to predicting mortality. The overall objective
of this work was to enable automated extraction of relevant features from chest CT volumes
that can be incorporated with clinical data for risk stratification of COVID-19 patients. They
compared this multitask segmentation approach with combining feature-independent
volumetric CT classification feature maps with clinical metadata to predict mortality. For
the prediction task, the authors compared the performance of five different machine
learning models for predicting mortality in patients with COVID-19.
Experimental results showed that the combination of features derived from chest CT
volumes improves the AUC performance to 0.80 from the 0.52 obtained using only clinical
patient data. Olowolayemo. A et al (2023) presented a model to determine the mortality of
an infected patient upon arrival at health facilities to determine whether or not admission to
intensive care is necessary. A CNN model based on the ResNet-18 architecture was trained
on CXRs of COVID-19 patients to estimate their mortality risk. Training of the proposed
model was performed using the Stony Brook University (SBU) dataset. The first stage of
the study used the IEEE8023 dataset, obtained from a public GitHub repository. The model
using the original unimputed data with class weight adjustment achieved the best
performance among the six compared models. This model was used in the presented study

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

using the SBU data set. Subsequently, the first model, that is, the best model from the first
training phase, and the second model, that is, the model that was fitted on the SBU data set,
were compared using a test set created from a part of the SBU data set.
The result showed that the second model trained on the SBU dataset performed much better
in terms of accuracy, significantly outperforming the first model trained only on the
IEEE8023 dataset, with an accuracy of 86 %. This was despite the fact that chest
radiographs were not discriminated based on the duration of infection of the patients at the
time of the radiograph. The model also achieved a high recovery value of 81,5 % with a
false negative rate of 18,4 % on the test set.
The main limitation of this work is the number of CXR images used. The limited amount
of data, especially CXRs of non-surviving cases, forced them to use oversampling methods,
such as image augmentation, to increase the number of images in the minority class.
Mixed models that combine classic machine learning and deep learning algorithms
Bae J et al., (2020). The aim of this study was to predict mechanical ventilation requirement
and mortality using computational modeling of CXRs for COVID-19 patients. A total of
530 unidentified CXRs from 515 COVID-19 patients treated between March and August
2020 were analyzed. They trained and evaluated random forest (RF) machine learning
classifiers to predict mechanical ventilation requirement and mortality using radiomic
features extracted from patients` CXRs and explored deep learning (DL) approaches to the
clinic.
The authors studied 143 radiomic features as those derived from the energy calculation. In
this study, the authors performed an exploratory analysis of these radiomic features to
determine their relative value in predicting clinical outcomes for patients with COVID-19.
For each radiomic feature, statistics including measures of median, skewness, standard
deviation, and kurtosis were calculated. These statistics and clinical factors, including
expert scores and patient age/sex, were used for the construction of the classifier.
The authors demonstrated that radiomic features can provide information about which
features of a patient's CXR are significant for making clinical predictions and may be more
informative to a clinician than DL-only approaches. From the results achieved, it can be
seen that radiomic characteristics play an interesting role in predicting results for COVID-
19.
This work has some limitations. First, the study used reference CXRs that are likely to be
inconsistent in the interval between COVID-19 infection and imaging. While this is
representative of the clinical reality that patients receive baseline chest x-rays at different
time points in the course of their illness, future studies could build better time-to-event
prediction models using data with a distribution more uniform temporary, although in the

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

context of the COVID-19 pandemic this is not easy to achieve given the urgency and the
large number of cases to be evaluated. However, for the study of other respiratory
infections, it could be taken into account.
It is also important to note that the two clinical outcomes studied in this paper are not
independent or mutually exclusive; in general, a patient who requires mechanical
ventilation is more likely to succumb to his illness than one who does not. On the other
hand, the study took into account a limited number of clinical characteristics, so the
inclusion of data on comorbidities such as a history of cancer, chronic obstructive
pulmonary disease, hypertension, etc., could improve the performance of the proposed
models. Finally, further validation would be needed to demonstrate the robustness of the
classification models in the broader context of treating COVID-19 and other respiratory
conditions at other hospitals and locations.
In Wendland. P. et al. (2023), the study was aimed to improve early risk stratification of
hospitalized Covid-19 patients by predicting in-hospital mortality, transfer to intensive care
unit (ICU) and mechanical ventilation from electronic health record data of the first 24 h
after admission.
For model development they conducted an observational retrospective cohort study using
electronic health record data from a hospital of medium level of care located in the federal
state of Rhineland-Palatinate in the west of Germany initially collected for billing purposes.
In the study were included 520 patients with a positive RT-PCR for SARS-CoV- 2
identified admitted to the hospital from March 2020 until December 2021.
For the model development and prior to any preprocessing steps authors performed a
random train- test split using 80 % of the data as training set and 20 % as test set. They
defined three Covid-19 associated endpoints:
1. Death during hospital stay, short “in-hospital mortality” 2. Admission to intensive care
unit (ICU), short “transfer to the ICU” 3. Necessity for mechanical ventilation (al OPS
beginning with “8–71“), short “mechanical ventilation”
In the experiment authors compared three supervised classifiers: Logistic regression (LR),
Random forest (RF) and XGBoost. To select predictive features for each of these three
model classes they employed 5-fold cross validation. For LR they performed forward-
backward selection. For the random forest classifier and the XGBoost classifier they used
the mean feature importance as a criterion for feature selection and in addition also trained
these tree based classifiers using the same features as identified for the LR models. Further,
for RF and XGBoost they performed a hyperparameter optimization on the training set. The
model (including selected features) with the highest receiver operator characteristics area
under the curve (ROC-AUC) averaged over the cross validation folds from the training data

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

set was selected as the final model for the respective endpoint.
As result, the machine learning model predicts in-hospital mortality (AUC = 0,918),
transfer to ICU (AUC = 0,821) and the need for mechanical ventilation (AUC = 0,654)
from a few laboratory data of the first 24 h after admission. Models based on dichotomous
features indicating whether a laboratory value exceeds or falls below a threshold perform
nearly as good as models based on numerical features. Authors devise completely data-
driven and interpretable machine-learning models for the prediction of in-hospital
mortality, transfer to ICU and mechanical ventilation for hospitalized Covid-19 patients
within 24 h after admission. Numerical values of CRP and blood sugar and dichotomous
indicators for increased partial thromboplastin time (PTT) and gluta mic oxaloacetic
transaminase (GOT) were amongst the best predictors.
The study only included patients admitted to hospitals from the beginning of the pandemic
until the end of 2021. Due to the rapidly changing epidemiological circumstances of the
pandemic the authors were not able to test the generalizability of the proposed models to a
population, where the Omicron mutation is the dominating virus mutation. From 2020 until
December 2021 the Wildtype, Alpha, Beta and Delta mutations were the dominating Covid-
19 variants in Germany. The study do not included vital parameters, pre-existing
comorbidities and vaccination. The cohort of 520 patients is relatively small. Further, the
studied data was imbalanced, because only 50 to 90 patients with poor outcomes were
observed for each respective endpoint.
Performance of the studied models
Given the conditions in which the COVID-19 pandemic develops, the studied models have
been tested on a wide variety of conditions in terms of number of patients, availability,
completeness, diversity and quality of images and clinical data, character of the study,
ethnic and age diversity of the patients participating in the studies, balance of the classes
taken into account for the training of the predictive models. This great diversity does not
allow for a comparison based on equal performance conditions. However, we have made a
summary of the results that they offer in terms of the most common metrics used in the
studied reports. Table 4 offers a summary of these results for each of the studied models.
From the table it is possible to observe that the models that achieve the best performance in
their respective evaluations do not depend on the type of outcome they predict, however
they have a high dependence on the type of data with which they work. The models with
the best results are those that use clinical data and CT images and their combination. On
the other hand, models that combine CXR images and clinical data have shown acceptable
performance.
Regarding the type of model, there is no clear differentiation between models based on

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

classical methods of machine learning and those based on deep learning, however
everything seems to indicate that in the case of machine learning algorithms, those based
on regression have offered more accurate results.

Table 4. Summary of performance results of the studied models

Ref.
in

table3

Used data Model type AUC Sensiblity. Specifity. Accuracy. (%) c- score

(1) CXR ML +DL 0.794 0.722 0.732 - -
(2) CXR, clinical ML 0.880 - - - -
(3) CT, clinical ML 0.912 0.895 0.8015 - -
(4) Clinical ML 0.895 - - - -
(5) CT ML 0.916 0.808 0.969 - -
(6) CT, clinical ML 0.910 0.894 0.872 - -
(7) Clinical ML 0.950 0.764 0.919 - -
(8) CT, clinical DL 0.942 - - - -
(9) CT, clinical DL 0.800 - - - -
(10) CXR, clinical DL 0.870 - - - -
(11) CXR DL 0.743 - - - -
(12) CXR, clinical DL - - - - 0.75
(13) CXR DL - - - 86 -
(14) Clinical DL - 0.935 - 95.97 -
(15) Clinical ML+ DL 0.797 - - - -
(16) CT, clinical ML+ DL 0.864 - - - -
(17) Clinical DL - - 93.55 95.97 -
(18) CT, clinical DL 0.880 - - - -

Regarding the limitations of the models studied, in the vast majority of them they are related
to the size of the training, validation and test sets, the imbalance of the samples of each
class used in the training and the origin of the samples taken in one or a limited number of
hospital institutions. Another limitation is related to the time of the pandemic in which the
samples were taken, where most of them were considered in the first 2 years (2020, 2021)
when the highest peaks of patients were recorded and more lethal strains of the disease
(Alpha, Beta and Delta mutations).
None of the analyzed studies takes into account the post-vaccination scenario and its
influence on the decrease in lethality, as well as the predominance of the more transmissible
but less lethal variants of the Omicron strain.

General discussion
Based on the results of the study carried out, we have been able to determine that predictive
models based on clinical data and their combination with CT or CXR images can achieve
significant performance in the evolutionary prediction of the state of severity of the disease,
death or the need for specialized treatments such as mechanical ventilation, hospital

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

admission or ICU. The trend that is observed is the development of models based on Deep
learning, however from the analysis developed we see that the methods based on regression
have had a good performance.
The development of a predictive model under the conditions of Cuba would be in
accordance with the availability of CXR images and non-standardized clinical data. The
development of an investigation in this sense could be aimed at obtaining a prediction
model for severe manifestations from COVID-19 by combining the classification of the
degree of severity of lung involvement on CXR images, clinical data on comorbidities, and
biographical data. The model would be developed based on the use of deep neural networks
and taking as a starting point the developments achieved so far by our research team in the
evaluation of the level of severity of lung involvement in CXR images by means of
classification (Garea et al., 2021a, 2021b, 2022, 2023). The inclusion of radiomic features
for the evaluation of the index of involvement of the pulmonary region could lead to an
increase in the accuracy of the prediction of the degree of severity caused by the disease.
As a baseline for comparison, we could then first take the work presented in Bae J. et al.
(2020), where it would be convenient to implement the extraction of radiomic features, and
their combination in the last three experimental schemes presented by the authors (P3, P4
and P5) with CNN deep learning models. Another method that can be taken as a baseline
for comparison would be those presented in Olowolayemo, A. et al. (2023) based on deep
learning.

ACKNOWLEDGEMENT

The VLIR-UOS has partially financed this research under the South Initiative: Toward
Precision Medicine for the Prediction of Treatment Response to Covid-19 in Cuba
(COVID-19 PROMPT).

CONCLUSIONS

In this work, a review study was presented with the objective of evaluating the most relevant
findings published between 2020 and 2023 that could serve as a basis for the development
of an own model adjusted to the conditions of Cuba.
As part of the methodology applied to the study, 18 relevant articles were identified, whose
analysis demonstrated that predictive models based on clinical data and their combination
with CT or CXR images can achieve significant performance in the evolutionary prediction
of COVID-19 outcomes as death or the need for specialized treatments such as mechanical

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

ventilation, hospital admission or ICU.
The development of an investigation in this sense could be aimed at obtaining a prediction
model for severe manifestations from COVID-19 by combining the classification of the
degree of severity of lung involvement on CXR images, clinical data on comorbidities, and
biographical data.
As the main insufficiencies of the reviewed articles, it was found that the models that
achieve the best performance in their respective evaluations do not depend on the type of
outcome they predict, however they have a high dependence on the type of data with which
they work.
The development of an investigation in this sense could be aimed at obtaining a prediction
model for severe manifestations from COVID-19 by combining the classification of the
degree of severity of lung involvement on CXR images, clinical data on comorbidities, and
biographical data.
The model would be developed based on the use of deep neural networks and taking as a
starting point the developments achieved in the evaluation of the level of severity of lung
involvement in CXR images by means of classification.

REFERENCES

Albahri, O.S.; Zaidan, A.A.; Albahri, A.S.; Zaidan, B.B.; Abdulkareem, K.H.; Al-Qaysi,

Z.T.; Alamoodi, A.H.; Aleesa, A.M.; Chyad, M.A.; Alesa, R.M.; et al.(2020).
Systematic review of artificial intelligence techniques in the detection and
classification of COVID-19 medical images in terms of evaluation and
benchmarking: Taxonomy analysis, challenges, future solutions and
methodological aspects. J. Infect. Public Health 2020, 13, 1381–1396.

Almotairi, K.H.; Hussein, A.M.; Abualigah, L.; Abujayyab, S.K.M.; Mahmoud, E.H.;
Ghanem, B.O.; Gandomi, A.H.(2023). Impact of Artificial Intelligence on COVID-
19 Pandemic: A Survey of Image Processing, Tracking of Disease, Prediction of
Outcomes, and Computational Medicine. Big Data Cogn. Comput. 2023, 7, 11.
https://doi.org/10.3390/bdcc7010011

Asteris P. G., Kokoris S., Gavriilaki E. et al (2023). Early prediction of COVID-19
outcome using artificial intelligence techniques and only five laboratory indices.
Clinical Immunology 246 (2023) 109218.
https://doi.org/10.1016/j.clim.2022.109218.

Bae, J., Kapse, S., Singh, G., Gattu, R., Ali, S., Shah, N., Marshall, C., Pierce. J., Phatak
T., Gupta, A., Green, J., Madan, N., Prasanna, P.(2021). Predicting Mechanical
Ventilation and Mortality in COVID-19 Using Radiomics and Deep Learning on

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

Chest Radiographs: A Multi- Institutional Study. Diagnostics (Basel). 2021 Sep
30;11(10):1812. doi: 10.3390/diagnostics11101812. PMID: 34679510; PMCID:
PMC8535062.

Bansal, A.; Padappayil, R.P.; Garg, C.; Singal, A.; Gupta, M.; Klein, A. (2020). Utility of
artificial intelligence amidst the COVID-19 pandemic: A review. J. Med. Syst.
2020, 44, 1–6.

Buttia, C., Llanaj, E., Raeisi-Dehkordi, H. et al. (2023). Prognostic models in COVID-19
infection that predict severity: a systematic review. Eur J Epidemiol 38, 355–372
(2023). https://doi.org/10.1007/s10654-023-00973-x

Duanmu, H., Ren, T., Li, H. et al. (2022). Deep learning of longitudinal chest X-ray and
clinical variables predicts duration on ventilator and mortality in COVID-19
patients. BioMed Eng OnLine 21, 77. https://doi.org/10.1186/s12938-022-01045-z

Fu. Y, Zeng L., Huang P., Liao M. et al (2023). Severity-onset prediction of COVID-19
via artificial-intelligence analysis of multivariate factors. Heliyon. V 9, I 8,
E18764, 2023. https://doi.org/10.1016/j.heliyon.2023.e18764

Gallo Marin, B., Aghagoli, G., Lavine, K., Yang, L., Siff, E.J., Chiang, S.S., Salazar-
Mather, T.P., Dumenco, L., Savaria, M.C., Aung, S.N., Flanigan, T., Michelow,
I.C. (2021).Predictors of COVID-19 severity: A literature review. Rev Med Virol.
2021 Jan; 31(1):1-10. doi: 10.1002/rmv.2146. Epub 2020 Jul 30. PMID: 32845042;
PMCID: PMC7855377.

Garea Llano, E., Castellanos Loaces, H. A., Martínez Montes, E., & González Dalmau, E.
(2021a). Estimación del Grado de Afectación Pulmonar por COVID-19 Mediante
la Clasificación Supervisada de la Imagen de Rayos X. Revista Cubana De
Transformación Digital, 2(3), 4–18. https://doi.org/10.5281/zenodo.5545908.

Garea-Llano, E., Castellanos-Loaces, H.A., Martinez-Montes, E., Gonzalez-Dalmau, E.
(2021b). A Machine Learning Based Approach for Estimation of the Lung
Affectation Degree in CXR Images of COVID-19 Patients. In: Hernández Heredia,
Y., Milián Núñez, V., Ruiz Shulcloper, J. (eds) Progress in Artificial Intelligence
and Pattern Recognition. IWAIPR 2021. Lecture Notes in Computer Science(), vol
13055. Springer, Cham. https://doi.org/10.1007/978-3-030-89691-1_2

Garea-Llano, E., Martinez-Montes, E., Gonzalez-Dalmaus, E. (2022). Affectation index
and severity degree by COVID-19 in Chest X-ray images using artificial
intelligence. Int Rob Auto J. 2022; 8(3):103‒107. DOI:
10.15406/iratj.2022.08.00252.

Garea-Llano, E., Diaz-Berenguer, A., Sahli, H., Gonzalez-Dalmau, E. (2023). Chest X-
Ray Imaging Severity Score of COVID-19 Pneumonia. In: Rodríguez-González,

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

A.Y., Pérez- Espinosa, H., Martínez-Trinidad, J.F., Carrasco-Ochoa, J.A., Olvera-
López, J.A. (eds) Pattern Recognition. MCPR 2023. Lecture Notes in Computer
Science, vol 13902. Springer, Cham. https://doi.org/10.1007/978-3-031-33783-
3_20.

Gourdeau, D., Potvin, O., Biem, J.H. et al. (2022). Deep learning of chest X-rays can
predict mechanical ventilation outcome in ICU-admitted COVID-19 patients. Sci
Rep 12, 6193 (2022). https://doi.org/10.1038/s41598-022-10136-9

Hassan, A.; Prasad, D.; Rani, S.; Alhassan, M. (2022). Gauging the Impact of Artificial
Intelligence and Mathematical Modeling in Response to the COVID-19 Pandemic:
A Systematic Review. BioMed Res. Int. 2022, 2022, 7731618.

Heidari, A., Jafari Navimipour, N., Unal, M. et al. (2022). Machine learning applications
for COVID-19 outbreak management. Neural Comput & Applic 34, 15313–15348
(2022). https://doi.org/10.1007/s00521-022-07424-w.

Heo, J., Han, D., Kim, H. J., Kim, D., Lee, Y. K., Lim, D. Seog, W. (2021). Prediction of
patients requiring intensive care for COVID-19: development and validation of an
integerbased score using data from Centers for Disease Control and Prevention of
South Korea. J Intensive Care, 9(1), 16. doi:10.1186/s40560-021-00527-x

Ho, T. T., Park, J., Kim, T., Park, B., Lee, J., Kim, J. Y., Choi, S. (2021). Deep Learning
Models for Predicting Severe Progression in COVID-19-Infected Patients:
Retrospective Study. JMIR Med Inform, 9(1), e24973. doi:10.2196/24973

Jalaber, C.; Lapotre, T.; Morcet-Delattre, T.; Ribet, F.; Jouneau, S.; Lederlin, M.(2020).
Chest CT in COVID-19 pneumonia: A review of current knowledge. Diagn. Interv.
Imaging 2020, 101, 431–437.

Jamshidi, M.B.; Roshani, S.; Talla, J.; Lalbakhsh, A.; Peroutka, Z.; Roshani, S.; Sabet,
A.; Dehghani, M.; Lotfi, S.; Hadjilooei, F.; et al.(2022). A Review on Potentials of
Artificial Intelligence Approaches to Forecasting COVID-19 Spreading. AI 2022,
3, 493–511.

Kumar, A.; Gupta, P.K.; Srivastava, A. (2020). A review of modern technologies for
tackling COVID-19 pandemic. Diabetes Metab. Syndr. Clin. Res. Rev. 2020, 14,
569–573.

Liang, W., Liang, H., Ou, L., Chen, B., Chen, A., Li, C. China Medical Treatment Expert
Group for, C. (2020). Development and Validation of a Clinical Risk Score to
Predict the Occurrence of Critical Illness in Hospitalized Patients With COVID-19.
JAMA Intern Med, 180(8), 1081-1089.
doi:10.1001/jamainternmed.2020.2033
(https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7218676/)

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

Lieveld, A. W. E., Azijli, K., Teunissen, B. P., van Haaften, R. M., Kootte, R. S., van den
Berk, I.

A. H., Nanayakkara, P. W. B. (2021). Chest CT in COVID-19 at the ED: Validation of
the COVID-19 Reporting and Data System (CO-RADS) and CT Severity Score: A
Prospective, Multicenter,Observational Study. Chest, 159(3), 1126-1135.
doi:10.1016/j.chest.2020.11.026.

Matsumoto, T., Walston, S.L., Walston, M. et al. (2022). Deep Learning–Based Time-to-
Death Prediction Model for COVID-19 Patients Using Clinical Data and Chest
Radiographs. J Digit Imaging (2022). https://doi.org/10.1007/s10278-022-00691-y

Mbunge, E.; Akinnuwesi, B.; Fashoto, S.G.; Metfula, A.S.; Mashwama, P. (2021). A
critical review of emerging technologies for tackling COVID-19 pandemic. Hum.
Behav. Emerg. Technol. 2021, 3, 25–39.

Moher D, Liberati A, Tetzlaff J, et al. (2009). Reprint--preferred reporting items for
systematic reviews and meta-analyses: the PRISMA statement. Phys Ther
2009;89:873−80.

Naudé,W.(2020). Artificial Intelligence against COVID-19: An Early Review; Institute of
Labor Economics: Bonn, Germany, 2020.

Okoli C (2015). A guide to conducting a standalone systematic literature review.
Commun Assoc Inf Syst 2015;37:43.

Olowolayemo, A., Yasin, M., Raashid S. M. (2023). Predicting Mortality Risk of Covid-
19 Patients Using Chest X-Rays. International Journal on Perceptive and Cognitive
Computing, 9(1), 33–43.

Ortiz, A., Trivedi, A., Desbiens, J. et al. (2022). Effective deep learning approaches for
predicting COVID-19 outcomes from chest computed tomography volumes. Sci
Rep 12, 1716 (2022). https://doi.org/10.1038/s41598-022-05532-0

Panagiotis G. Asteris, Styliani Kokoris, Eleni Gavriilaki, et al (2023). Early prediction of
COVID- 19 outcome using artificial intelligence techniques and only five
laboratory indices, Clinical Immunology, Volume 246, 2023, 109218, ISSN 1521-
6616, https://doi.org/10.1016/j.clim.2022.109218.

Portal-Diaz, J.A., Lovelle-Enríquez, O., Perez-Diaz, M. et al. (2022). New patch-based
strategy for COVID-19 automatic identification using chest x-ray images. Health
Technol. 12, 1117–1132 (2022). https://doi.org/10.1007/s12553-022-00704-4

Quiroz, J. C., Feng, Y. Z., Cheng, Z. Y., Rezazadegan, D., Chen, P. K., Lin, Q. T., Cai, X.
R. (2021). Development and Validation of a Machine Learning Approach for
Automated Severity Assessment of COVID-19 Based on Clinical and Imaging
Data: Retrospective Study. JMIR Med Inform, 9(2), e24572. doi:10.2196/24572

REVISTA CUBANA
DE TRANSFORMACIÓN DIGITAL

UPDATE ON ARTIFICIAL INTELLIGENCE MODELS FOR
PREDICTING OUTCOME IN COVID-19 PATIENTS IN CUBA

Garea Llano, E. & Gonzalez Dalmau, E.

(https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7879715/).
Rasheed, J.; Jamil, A.; Hameed, A.A.; Al-Turjman, F.; Rasheed, A. (2021). COVID-19 in

the Age of Artificial Intelligence: A Comprehensive Review. Interdiscip. Sci.
Comput. Life Sci. 2021, 13, 153–175

Rahimi, I.; Chen, F.; Gandomi, A.H. (2021). A review on COVID-19 forecasting models.
Neural Comput. Appl. 2021, 1–11.

Shi, F.; Wang, J.; Shi, J.; Wu, Z.; Wang, Q.; Tang, Z.; He, K.; Shi, Y.; Shen, D. (2020).
Review of artificial intelligence techniques in imaging data acquisition,
segmentation, and diagnosis for COVID-19. IEEE Rev. Biomed. Eng. 2020, 14, 4–
15

Swapnarekha, H.; Behera, H.S.; Nayak, J.; Naik, B. (2020). Role of intelligent computing
in COVID-19 prognosis: A state-of-the-art review. Chaos Solitons Fractals 2020,
138, 109947.

Tanboga, I. H., Canpolat, U., Cetin, E. H. O., Kundi, H., Celik, O., Caglayan, M., . .
.Topaloglu, S. (2021). Development and validation of clinical prediction model to
estimate the probability of death in hospitalized patients with COVID-19: Insights
from a nationwide database. J Med Virol, 93(5), 3015-3022.
doi:10.1002/jmv.26844.

Tranfield D, Denyer D, Smart P. (2003). Towards a methodology for developing
evidence-informed management knowledge by means of systematic review. Br J
Manag 2003;14:207−22.

Verzellesi L, Botti A, Bertolini M, Trojani V, Carlini G, Nitrosi A, Monelli F, Besutti G,
Castellani G, Remondini D, et al. Machine and Deep Learning Algorithms for
COVID-19 Mortality Prediction Using Clinical and Radiomic Features.
Electronics. 2023; 12(18):3878. https://doi.org/10.3390/electronics12183878

Wendland. P., Schmitt V., Zimmermann J , Hager L., Gopel S., Schenkel-Hager C.,
Kschischo M (2023). Machine learning models for predicting severe COVID-19
outcomes in hospitals. Informatics in Medicine Unlocked 37 (2023) 101188.
https://doi.org/10.1016/j.imu.2023.101188

Zhang, B., Liu, Q., Zhang, X., Liu, S., Chen, W., You, J., Zhang, S. (2020). Clinical
Utility of a Nomogram for Predicting 30-Days Poor Outcome in Hospitalized
Patients With COVID- 19: Multicenter External Validation and Decision Curve
Analysis. Front Med (Lausanne), 7, 590460. doi:10.3389/fmed.2020.590460

Este obra está bajo una licencia de Creative Commons Atribución-No Comercial 4.0 Internacional