Keywords: Stroke - Benchmarking - Artificial Intelligence - Supervised Machine Learning - Emergency
Service - Hospital
Palavras-chave: Acidente Vascular Cerebral - Benchmarking - Inteligência Artificial - Aprendizado
de Máquina Supervisionado - Serviço Hospitalar de Emergência
INTRODUCTION
Stroke is the second leading cause of death and disability worldwide. It has a significant
socioeconomic impact on low- and middle-income countries, thereby affecting public
health status[1 ]–[3 ]. In the last 40 years, databank projects have been implemented to provide information
on the clinical courses and outcomes of stroke[4 ]–[8 ].
Implementation of a databank is important for planning and decision-making by healthcare
service managers, in order to improve processes and stroke care outcomes[9 ]. However, integration of these databanks is limited to large centers and clinical
trials, thus restricting access to low and middle-income countries[10 ]. In accordance with the principles of stroke care lines in developing countries,
it is necessary for all quality criteria during stroke hospitalization to be registered
in electronic medical records, with the aim of establishing new healthcare policies[11 ].
Based on these premises, implementation of a computerized database system containing
the main clinical indicators of stroke patients has become a reality in developing
countries[12 ],[13 ]. Over recent years, the introduction of artificial intelligence has helped in development
of databases that are integrated in networks, with the capacity to identify clinical
data with greater agility and security[14 ],[15 ].
Therefore, the aim of this study was to develop an interface for accessing stroke
patient data and the main benchmarks in acute stroke care, in order to create a semi-automated
databank for implementation in low and middle-income countries.
METHODS
Study design, setting and participants
This project with the aim of developing a semi-automated stroke databank was conducted
at the stroke unit of Faculdade de Medicina de Botucatu, Botucatu, Brazil. This stroke
unit has 10 beds, with an average of 30 patients/month, and belongs to a tertiary-level
hospital with 684 beds. The hospital has a hospital information system (HIS) that
includes electronic patient records containing clinical and treatment information.
All patients who were admitted to the stroke unit were included in this project. The
study was approved by the institution’s Ethics in Human Research Committee.
Procedures
To organize the information, the tool was designed to include initial data, a clinical
summary of the presentation, hospital course (including complications and infections),
investigation/complementary examinations (including computed tomography [CT], electrocardiography,
echocardiography, Holter data, carotid and vertebral duplex data, CT angiography,
magnetic resonance imaging [MRI] angiography, control CTs or MRIs and transcranial
Doppler) and discharge conditions (including etiological diagnosis, medications, discharge
with therapy plan, rehabilitation and destination of the patient). The rules for each
subgroup were planned so as to validate the completion of certain fields, all of which
were mandatory for the service to be completed.
The database interface was developed on the HIS platform, and clinical data were collected
based on the following benchmarks for stroke care: 1) prophylaxis for deep venous
thrombosis, starting no later than the second day; 2) hospital discharge with antiplatelet
therapy for patients with non-cardioembolic stroke; 3) hospital discharge with oral
anticoagulation for patients with atrial fibrillation or atrial flutter; 4) use of
antiplatelet agents when indicated, starting on the second day of hospitalization;
5) hospital discharge with statins for patients with atherothrombotic stroke; 6) hospital
discharge with prophylactic therapy and rehabilitation plan; 7) percentage of patients
with acute cerebrovascular disease; 8) length of hospital stay; 9) complications;
10) stroke type-specific ICD-10; 11) hospital mortality; 12) time to CT <25 min; and
13) door-to-needle time ≤60 min.
Execution, monitoring and control phase
To execute the development of the tool, an electronic editor of structured forms was
used, which was integrated with the electronic medical record platform used in the
institution. Parameterized fields and validation rules were created for essential
fields. The rules were created in the field editor. Radio button components were used
for information with only one choice and checkbox components were used when the information
consisted of multiple choices. In addition, the validation of fields was programmed
using a structured language, i.e. in structured query language (SQL)[16 ].
Mandatory fields were configured in the editor, and each field was parameterized.
All patients who were admitted to the stroke unit were included in the database, as
this was mandatory at the time of transfer to another ward or at the time of the patient’s
discharge or death.
Databank storage
To store the results from SQL, a virtual table was created and programmed to update
daily at 04:00 h. The interface between the data and user was developed using the
program Embarcadero Delphi, and the DevExpress component was used to generate the
information displayed on the screen.
Data quality
The stored data were analyzed to check the data quality in accordance with the pre-established
rules for creating the tool as mandatory fields with relationships between the fields.
A rule was also created to identify patients who were admitted to the stroke unit
and then transferred to another unit, but for whom no discharge summary was filled
out at any time during hospitalization. In preparing SQL and business rules for both
data analysis and for out-of-base patients, the PL/SQL (Procedural Language for SQL)
tool was used via Embarcadero. The data were extracted from the fields of the form
and also from cross-referencing of other information from the computerized system,
such as patients who were admitted to the stroke unit.
RESULTS
Once extracted, the data were stored in a database, and any type of information could
be consulted at any time, thus allowing cross-referencing between the data of the
tool and the patients’ demographic profile. The indicators that were established according
to the area and database were available in almost real time, with daily updates at
04:00 h. The database began registering patients in August 2018 and has included data
from approximately 1,000 stroke patients to date.
Both qualitative and quantitative results can be obtained from the detailed data included
in the database. The results can be grouped to create filters that can be applied
to any column within the database. This provides an overview of the total number of
patients for each parameter, thus resulting in development of appropriate public health
policies.
In addition to the specific data characteristics of stroke patients, and owing to
the integration with the electronic medical record, important clinical data such as
hospital course, laboratory test results, imaging tests and surgeries, as well as
the main scales used in the stroke unit, can be displayed on the screen ([Figure 1 ]).
Figure 1 National Institutes of Health Stroke Scale displayed in the electronic medical record.
The quality indicators (benchmarks) were created in the database for the system to
track and perform decision-making in conjunction with healthcare service managers.
This resulted in improved processes and patient care after a stroke ([Figure 2 ]).
Figure 2 Safety and quality indicators.
An intelligent portal was created, in which all the information referring to patients’
care was available in a single location in a dynamic and objective manner. Development
of an interface with a specific database for a hospital area enabled creation of a
new concept in relation to the data that were already included in the computerized
system, thus allowing new ideas to emerge in order to improve and facilitate the interpretation
of large volumes of data ([Figure 3 ]).
Figure 3 Business intelligence applied to stroke databank.
DISCUSSION
This study reports the development of a tool integrated with electronic medical records
that is capable of recording clinical data on patients admitted to a stroke unit and
identifying the main indicators of quality of care, using artificial intelligence
to improve the quality of care and decision-making process.
Several studies using previous databases have been conducted in low and middle-income
countries to identify the quality of healthcare services[17 ]–[20 ]. Zetola et al. observed the registration of patients in a database with identification,
previous clinical history, family history, treatments, previous comorbidities, complementary
laboratory tests, cardiac tests and neuroimaging tests. Using a computerized system,
these data could be extracted dynamically, and their study suggested that a risk factor
control program might lead to a reduction in stroke incidence[17 ].
In 2013, a single database for stroke patients was established in Joinville, Brazil,
via a municipal law that required all public and private healthcare establishments
to forward monthly information on stroke patients to the Department of Health to create
an integrated database. In 2014, the Ministry of Health coordinated a simultaneous
study in five cities in Brazil to assess the incidence of, mortality due to and main
risk factors for stroke. The study included individuals with stroke and determined
the environmental context of stroke, using a national database of stroke patients
in Latin America[21 ],[22 ].
The Rede Brasil AVC (Brazil Stroke Network), created to improve overall care for stroke patients in Brazil,
is a non-governmental organization formed by professionals from various areas with
the aim of providing quality care to stroke patients, from prevention to rehabilitation[23 ]. However, population studies to monitor clinical data on stroke patients need to
be integrated into networks with shared access and periodic audits. The development
of the tool in this study will help facilitate exploration of clinical data and quality
indicators in low and middle-income countries.
Studies on the development of systems containing specific data on stroke patients
have not been previously reported. Data collection methods were not reported in previous
studies that included databases. Access to information has become increasingly faster
and easier through technological advances that enable implementation of structured
and standardized electronic databases that use artificial intelligence. Analysis on
large datasets of patient characteristics, outcomes from treatments and their costs
can help identify the most clinically effective and cost-efficient treatments for
a population. Models, analytics, visualization large amounts of data and use of artificial
intelligence come together to offer different perspectives on healthcare challenges
within the contexts of time and geography. They provide strategic solutions to assist
in management, decision-making and future clinical research[24 ],[25 ].
An integrated system significantly integrates separate sectors, and teamwork is recorded.
This enables top management to fully analyze healthcare processes. Development of
this database for stroke patients will optimize this search time and provide large
quantities of relevant information for the area available, which can be used to support
future studies[26 ],[27 ].
In conclusion, through development of integration of this tool and electronic medical
records, a dynamic and optimized stroke data bank was created. This database will
be useful for managing quality indicators, assisting in the planning of actions at
the stroke unit and supporting decision-making, and will serve as a basis for future
studies and generation of new knowledge.
