Formulas developed with WEKA software with Linear Regression and J48
Decision Tree algorithms can be used to determine the status of
patients.
With these formulas, the presence and level of thoracic CT involvement
and the need for intensive care can be predicted by using the
laboratory data of COVID-19 patients.
INTRODUCTIONCoronavirus disease 2019 (COVID-19) is an alarming public health concern
worldwide. This disease may cause a variety of clinical conditions from
an asymptomatic disease to severe viral pneumonia that can result in
respiratory failure or death.1Presence of accurate and fast diagnostic tests is of clinical importance
in controlling the COVID-19 pneumonia. The gold standard diagnostic test
in the diagnosis of COVID-19 is the reverse transcriptase polymerase
chain reaction (RT-PCR). However, this test has several limitations such
as potential false negative results, high cost, difficulties in
collecting, storing and transport of sample materials. Serological tests
have also attracted attention as alternative or complementary to RT-PCR
and other nucleic acid tests in the diagnosis of acute
infections.2 However, they are not useful to diagnose
acute cases as the IgM antibody response can be detected 6-15 days after
the disease onset.3 Studies have reported that the
sensitivity of fast SARS-CoV-2 antigen diagnosis tests is between 45 and
97%.4 Lack of common standardization among
serological tests causes difficulties in the use of the test. As another
diagnostic method in this disease, lung imaging is quite valuable, but
having access to thoracic CT is not always possible. Moreover, early
interpretation of this imaging, which is valuable for the diagnosis of
COVID-19, requires an extra workload. This situation necessitates
methods that can predict lung involvement and provide high diagnostic
performance.5Another problem with this disease is that the patients experience
unpredictable progression, giving rise to a need for intensive care.
Although the intensive care need of COVID-19 patients varies by country
and institution, it ranges between approximately 5 and
32%.6 It is important in terms of disease morbidity
and mortality to predict patients’ intensive care needs and to transfer
such patients quickly from healthcare institutions having no intensive
care facilities.7There is a need for defining clinical and laboratory predictors enabling
early prediction of progression to serious clinical form and intensive
care need when combating this disease which puts severe stress on many
healthcare systems across the world. Defining such predictors will
enable risk classification so that patients having high risk of
developing severe disease can be identified and necessary interventions
initiated at an early stage. This will optimise human and technical
resources during this ongoing pandemic.8 It could be a
good alternative in support of physicians to define, with the help of
the deep learning technology, a system consisting of laboratory
parameters to be used to predict presence of lung involvement, the
extent of such involvement and intensive care need in COVID-19 patients.
In recent years, technologies such as deep learning and artificial
intelligence started to be used in many areas in medicine with a high
diagnostic accuracy.9,10 Although there are studies
using these methods in the thoracic CT interpretation of COVID-19, we
have not encountered any study on the use of these technologies for the
assessment and interpretation of laboratory parameters in COVID-19.
The present study aimed at defining a system of formulas that may
predict the presence of thoracic CT involvement, the extent of such
involvement and the need for intensive care stay on the basis of patient
laboratory data using the Waikato Environment for Knowledge Analysis
(WEKA) software, Linear Regression and J48 Decision Tree algorithms and
that may in this way contribute to the process of diagnosing and
treating the disease.