BackgroundAmong the growing number of patients with hematologic neoplasms hospitalized in the intensive care unit (ICU), the largest proportion of these patients are diagnosed with lymphoma. However, less attention has been paid in the past to identifying critically ill patients and assessing the prognosis of patients in ICU. Traditional critical care-related scores have shown limitations and inaccuracy in predicting mortality risk.MethodsPatients diagnosed with diffuse large B-cell lymphoma (DLBCL) were searched for in the Marketplace for Information in Intensive Care Medicine III (MIMIC-III) database. We searched mortality within 28 days as the primary endpoint. Logistics regression was used to screen risk factors. A calibration curve was used for internal validation, and the ROC curve and AUC were used to compare the new model with traditional scores.Results405 patients with DLBCL are enrolled in the project. Multivariate analysis shows the patients with the level of lactate dehydrogenase (LDH) > 327 U/L had an increased risk of 28-day mortality in ICU than others (OR = 13.04, p<0.01). Notably, length of ICU stay, LDH, creatinine, white blood cell counts, and APS III score are independent prognostic factors for patients with DLBCL in the ICU. Then, all these independent prognostic factors are selected into our prediction model. The new model has good accuracy (C-index=0.863) and a calibration curve, which improves clinical status concerning established ratings such as IPI, NCCN-IPI score, SOFA, APS III, and LODS. The results of a multicenter external validation including 124 DLBCL patients also showed that the new model was more accurate than all other models.ConclusionsThe elevated level of LDH indicates a poor prognosis of patients with DLBCL in the ICU. Our risk score with crossed validation based on the level of LDH shows a significant prognostic value and may be a valuable tool for assessing the critically ill as well.