(i) The proposed method provides efficient hospital resources (ii) Simple and more generic features are used to encode the waveform dynamics in time and frequency domains (iii) Low-cost wearable sensors are used to collect data
(i) The manual encoding of features used to encode the waveform dynamics in time and frequency domains is time-consuming and may have errors (ii) The dataset is small
(i) The study shows an accessible, easy to use, flexible, ubiquitous, and cost-effective eHealth system for diagnosing infectious diseases from vital signs.
(i) The short period of sampling affects the classification results, and more accuracy is needed (ii) The small dataset affects the accuracy of the model
(i) The proposed model shows that a combination of symptoms assisted with the prediction of COVID-19 infection (ii) It helps improve the test strategy by prioritizing users for testing
(i) Some studies criticize the use of symptoms for classifying COVID-19 because of the existence of other respiratory coinfections and the nonspecific nature of some symptoms
(i) The study shows that voice-based screening for COVID-19 is possible (ii) Deep learning is useful in addressing the challenges of the long sequences in voice recordings, uncertain and presumably subtle vocal attributes of early COVID-19, and the lack of large labeled datasets
(i) The study uses a small set of vocal-input types that were self-recorded (ii) The results show a connection between COVID-19 voice symptoms and detection of it, whereas there are no reliable reports to assess this correlation
(i) The study is helpful in diagnosing latent infectious diseases in early stages without prior training data and in a short period.
(i) There is a need to improve the performance of the proposed model and to include accuracy measures when considering social media user information (e.g., age, gender, and posting frequency).
(i) It is more comprehensive compared to other existing works (ii) It establishes a reliable knowledge base for infectious diseases
(i) Symptoms are not weighed to distinguish syndromes and signs (ii) Important etiological factors such as a history of close exposure to other infected patients are not included
(i) The study determines the most useful routine blood parameters for COVID-19 diagnosis from a large (ii) Use of the ML model to diagnose COVID-19 from a routine blood test in the early symptomatic phase is effective when demands on the real-time reverse transcriptase-polymerase chain reaction (RT-PCR) test are enormous
(i) The proposed model might be inefficient at the stage where there are no systemic effects (ii) The study includes only features available from a single center (iii)the positive number of COVID-19 patients is limited in the study.
(i) The proposed ML model can be used as a decision support system tool (ii) A combination of routine laboratory results and ML models improves the COVID-19 diagnosis
(i) The study patients’ comorbidities are not available in the dataset (ii) Larger datasets are required (iii) Some important features are not included, such as medical imaging, vital signs, physical examination, symptoms, and increasing sample size (iv)the absence of data pertaining to vaccinations limits the study.
(i) The use of ML to predict Candidemia improves decision-making for appropriateness in antifungal and antibiotic therapies (ii) The use of ML reduces the delay in empirical treatment (iii) The study uses real-world data with a large number of features (as predictors)
(i) There is a need for a large number of patients (ii) It is a retrospective study, and the pooled data are anonymous
(i) The EHR-based model can be used as clinical decision support to predict complicated cases of CDI on the day of diagnosis (ii) Use of the ML approach is feasible for generating accurate and early risk predictions for complicated CDI
(i) The performance metrics that are used are not enough to evaluate the model (ii) The proposed model is not tested in a real-time manner (iii) The obtained results are based on a small dataset from one institution (iv) The study does not mention the applied ML algorithm
(i) Machine-learning classifiers perform better than expert constructed classifiers using the NLP extraction tool (ii) The large number of ED reports in training classifiers solves the imbalance problem in the dataset (iii) The applied method for dealing with missing values in the study shows improved performance
(i) The study focuses on the data of only one health system (ii) The number of selected features is considered small
(i) The study shows that the combination of clinical data and radiomic features, including all measures in the optimal model, has the height performance, and can effectively predict survival in COVID-19
(i) The used dataset is small, and there is a lack of an external validation dataset (ii) Clinical studies are required to verify obtained results. (iii) The study tests only one ML classifier and one feature selection method
(i) Diagnosing COVID-19 is faster and more accurate than other traditional methods applied to the same CT image dataset.
(i) None of the COVID-19 variants is included in the study (ii) The proposed model is trained on a tiny dataset (iii) The study uses only a COVID-19 patient dataset to train the model
(i) Various types of predictors are utilized (ii) Data quality is guaranteed in the study (iii) Standard statistical methods are used to validate the model’s performance
(i) The validation dataset is available for only 12 out of 25 infectious diseases (ii) The proposed model is limited to only 25 infectious diseases and cannot be generalized (iii) Real-time updates are missing
