Abstract: OBJECTIVE To develop the radiomics integrated with multimodal prediction model of clinical indicators and validate its efficacy in precise assessment of the mortality risk of the human immunodeficiency virus (HIV) infection patients complicated with Talaromyces marneffei pneumonia (TMP) during hospital stay. METHODS Totally 216 patients with HIV-TMP who were treated in Nanning Fourth People's Hospital from Jan. 1, 2020 to Aug. 31, 2025 were enrolled in the study and were randomly divided into the training set with 151 cases and the validation set with 65 cases in a 7∶3 ratio. The chest CT imaging findings and clinical data were collected from the patients within 48 hours after the admission. The radiomic features were obtained through PyRadiomics platform, the stability features were screened out by more than 0.8 of interclass correlation coefficient (ICC). The clinical model, radiomics model and the integrated model were established and validated, respectively. RESULTS The indicators of the clinical model, including blood platelet counts, C-reactive protein and albumin, were associated with the mortality risk (P＜0.05). The area under the curve (AUC) of the clinical model that was established based on features was 0.894 in the training set, 0.785 in the validation set, with the specificity reaching 0.930 in the validation set, the sensitivity only 0.591. The radiomics features were screened through LASSO regression combined with 10-fold cross-validation, 24 features were finally selected for the establishment of the radiomics model, and the AUC of the validation set was 0.722, with the sensitivity up to 0.955, the specificity down to 0.488. The integrated model exhibited the best performance in both the training set and the validation set, with the AUC 0.821 in the validation set (95%CI:0.702 to 0.941), the specificity up to 0.977. Hosmer-Lemeshow test indicated that the integrated model had good calibration. Decision curve analysis (DCA) demonstrated that the model yielded higher net benefit in the validation set with the threshold probability ranging between 0.2 and 0.6 than the strategy of ' complete intervention' or 'complete no intervention'. CONCLUSION The radiomics integrated with prediction model of clinical features can remarkably intensify the capability of discriminating the mortality risk of the HIV-TMP patients during the hospital stay, boost the clinical adaptability, facilitate the achievement of risk stratification in early stage, and optimize the allocation of medical resources.