Method Description
The duration of inpatient treatment is one of the most objective and unambiguously interpreted indicators that can be used to indirectly assess the severity of the patient's condition. To predict the duration of inpatient treatment of a patient using the method «COVID-19 Predict», the purchase of expensive equipment or reagent kits is not required, the results of a routine clinical examination of the patient with an assessment of such indicators as age, height, weight, body mass index, number of days from the onset of the disease are sufficient , body temperature, pulse rate, respiratory rate, diastolic and systolic blood pressure, and SpO₂ level. All these clinical indicators can be assessed at the prehospital stage, which makes it possible to use the method at the stage of primary triage to support medical decisions on further tactics of patient management, including the decision on the need for hospitalization. Most of the known analogous methods require laboratory or instrumental studies, which in real clinical practice may not be available at the prehospital stage.
Method precision
The accuracy of the method for predicting the duration of inpatient treatment of patients with COVID-19 «more than 10 days» is 83.75% (sensitivity 82.50%, specificity 85.00%)
Interpretation of results
The method «COVID-19 Predict» is a gradient boosting machine learning model. The result of the «COVID-19 Predict» study is a calculated index, which is a number in the range from 0 to 1. An index value of more than 0.5 corresponds to a positive test result (duration of inpatient treatment «more than 10 days»). An index value less than 0.5 corresponds to a negative test result (the duration of inpatient treatment is «less than 7 days»). The probability of getting the correct test answer depends on the index value. So the greater than 0.5 or less than 0.5 the index value, the higher the likelihood of getting correct positive or correct negative test results, respectively.
Publications
Predicting the duration of inpatient treatment for COVID-19 patients
Authors: Tsvetkov V.V., Tokin I.I., Lioznov D.A., Venev E.V., Kulikov A.N.
Source: Meditsinskiy sovet = Medical Council. 2020;(17):82-90. (In Russ.)
https://doi.org/10.21518/2079-701X-2020-17-82-90
More details
Introduction. In the context of a high load on all links in the structure of providing medical care to patients with COVID-19, solving the issue of effective triage of patients seems to be extremely urgent. The duration of inpatient treatment is one of the most objective and unambiguously interpreted indicators that can be used to indirectly assess the severity of the patient’s condition.
Objective. Develop a machine learning model to predict the duration of inpatient care for patients with COVID-19 based on routine clinical indicators assessed at the prehospital stage.
Materials and methods. A total of 564 patients were examined with diagnoses: U07.1 COVID-19, virus identified (n = 367) and U07.2 COVID-19, virus not identified (n = 197). The study included 270 patients, of whom in 50.37% of patients the duration of inpatient treatment did not exceed 7 days, in 49.63% of patients the duration of inpatient treatment was more than 10 days. Eleven clinical parameters were chosen as the most important predictors for predicting the duration of inpatient treatment: age, height and weight of the patient, SpO2 level, body temperature, body mass index, pulse rate, number of days from the onset of illness, respiratory rate, systolic and diastolic arterial pressure.
Results. The accuracy of our machine learning model for predicting the duration of inpatient treatment more than 10 days was 83.75% (95% CI: 73.82–91.05%), sensitivity — 82.50%, specificity — 85.00%. AUC = 0.86.
Conclusion. The method developed by us based on machine learning is characterized by high accuracy in predicting the duration of inpatient treatment of patients with COVID-19, which makes it possible to consider it as a promising new tool to support medical decisions on further tactics of patient management and to resolve the issue of the need for hospitalization.
Rules
The results of the «COVID-19 Predict» study should be assessed by a specialist doctor, taking into account other results of the patient's clinical and laboratory examination.