Background and Aims: Hepatocellular carcinoma (HCC) is the most common primary liver cancers, and has a poor prognosis. Dysregulation of autophagy can promote the occurrence and development of HCC. This study was designated to investigate the potential prognostic roles of autophagy-related long non-coding RNAs (lncRNAs) in HCC patients and construct a risk prediction model based on autophagy-related lncRNAs.  
Methods: The transcriptomic and clinical data of 374 HCC samples and 50 normal control samples in TCGA database were analyzed using bioinformatics approaches, and the list of autophagy-related genes were obtained from HADb. The lncRNAs associated with autophagy genes were screened by Person’s correlation analysis. Three hundred and forty-two HCC samples obtained by selection were randomly assigned to train dataset and validation dataset with a ratio of 70%:30% using caret package in R. The autophagy-related lncRNAs with prognostic significance were identified by Kaplan-Meier method and univariate Cox regression analysis. Then, the autophagy-related lncRNAs with independent prognostic significance were determined by multivariate stepwise regression Cox analysis to construct a prognostic prediction model. After the risk scores were calculated using Cox regression coefficient, the patients were divided into low risk group and high risk group, the relationship between the risk score and clinicopathologic features as well as the overall survival (OS) was analyzed, and then was verified in the validation dataset. 
Results: A total of 347 lncRNAs were identified as autophagy-related lncRNAs (|R2|>0.3, P<0.001), including 26 lncRNAs with prognostic value for HCC patients. The risk model for predicting the prognosis of the patients was derived from the multivariate stepwise regression Cox analysis based on 12 autophagy-related lncRNAs (CYTOR, DANCR, LINC01138, LUCAT1, Mapkapk5-AS1, NRAV, NRSN2-AS1, LINC01871, LINC00864, LINC02362, TMEM220-AS1 and PSMB8-AS1). The risk scores of the 12- autophagy-related lncRNAs prognostic model was sufficiently associated with tumor grade, tumor stage and T stage (all P<0.05), but irrelevant to the age and sex of the patients (both P>0.05). In this model, the area under curve (AUC) of the time-dependent ROC for the 1, 3 and 5-year overall survival were 0.801, 0.819 and 0.787 in the train dataset, and the AUC of the time-dependent ROC for the 1, 3 and 5-year overall survival were 0.694, 0.733 and 0.746 in the validation dataset. 
Conclusion: The identified autophagy-related lncRNAs may play critical roles in the oncobioloy of HCC, and the 12- autophagy-related lncRNAs has certain predictive value for the prognosis of HCC.