Construction of a predictive model for relapse of primary autoimmune hemolytic anemia: a retrospective cohort study.

OBJECTIVES

To develop a machine learning-based model to predict the relapse risk of Primary Autoimmune Haemolytic Anaemia (AIHA) after the last remission.

METHODS

A retrospective study was conducted on primary AIHA cases who visited the Affiliated Hospital of Southwest Medical University and Xuyong County People's Hospital from May 2017 to May 2022. Cases were categorized as relapsed or non-relapsed based on the 1-year outcomes. Twenty-two features were analyzed to identify relapse risk factors. The least absolute shrinkage and selection operator (LASSO) regression model and multivariate logistic regression analysis were used to establish a predictive model. The -index, Calibration curves, ROC, and Decision curve analysis (DCA) were used to evaluate the discriminatory, corrective, accurate, and clinical effectiveness of the predictive model.

RESULTS

A total of 232 cases of primary AIHA were included, and five potential variables including 'DAT results', 'Hb', 'Multiline therapy', 'Complicating ITP', and 'Complicating infection', have been screened for constructing a 1-year relapse risk prediction nomogram for primary AIHA. The nomogram has a -index of 0.852 (95% CI: 0.797-0.907), confirmed by bootstrapping validation as 0.829. The area under the ROC was 0.846. The DCA shows that when the threshold probability is in the range of 1 ∼ 91%.

CONCLUSIONS

By following the current diagnostic and treatment criteria for AIHA in China, we retrospectively collect a multitude of medical records and analyze several relevant variables of AIHA, construct a predictive model by machine learning. Using this 1-year relapse risk nomogram can effectively predict the risk of relapse within 1 year after remission of primary AIHA.