Full Papers
A machine learning model for identifying systemic lupus erythematosus through laboratory information system and electronic medical record
J. Du1, H. Huang2, L. Pang3, N. Duan4, C. Huang5, C. Liu6, H. Li7
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China.
- Medical Records Statistics Office, Peking University First Hospital, Beijing, China. chenlong.002@163.com
- Department of Clinical Laboratory, Peking University First Hospital, Beijing, China. bdyylhx@126.com
CER16706
2024 Vol.42, N°3
PI 0702, PF 0712
Full Papers
Free to view
(click on article PDF icon to read the article)
PMID: 37976115 [PubMed]
Received: 28/03/2023
Accepted : 25/09/2023
In Press: 15/11/2023
Published: 27/03/2024
Abstract
OBJECTIVES:
Systemic lupus erythematosus (SLE) is a heterogeneous autoimmune disease. Its diagnosis poses significant challenges especially at early stages and in atypical cases. The aim of this study was to develop a machine learning model based on common laboratory tests that can aid SLE diagnosis.
METHODS:
A standard protocol was developed to collect data of SLE and control immune diseases. A 10-fold cross-validation was performed in the modeling dataset (n=862), and an external dataset (n=198) was used for model validation. Machine learning algorithms were applied to construct a diagnostic model. Performance was evaluated based on area under the curve (AUC) values, F1-score, negative predictive value, positive predictive value, accuracy, sensitivity, and specificity.
RESULTS:
The optimal model was based on a random forest algorithm with 10 clinical features. Thrombin time, prothrombin activity, and uric acid contributed most to the diagnostic model. The SLE diagnostic model showed sufficient predictive accuracy, with AUC values of 0.8286 in the validation dataset.
CONCLUSIONS:
Our diagnostic model based on 10 common laboratory tests identified the patients with SLE with high accuracy. An online version of the model can potentially be applied in clinical settings for the differential diagnosis of SLE.
