Circadian rhythm disruption (CRD) represents a major contributor to tumor proliferation. Nonetheless, the role of CRD in the clinical prediction of cancer outcomes has not been well studied. In this study, we developed a computational algorithm, which was implemented in an open-source R package CRDscore, to define the intratumoral status of circadian disruption in three representative single-cell RNA-seq data sets of lung adenocarcinoma. We found that the malignant cells with high CRDscore were characterized by activation of glycolysis and epithelial-mesenchymal transition pathways. Furthermore, cell communication analysis indicated that CRD played a pivotal role in T cell exhaustion, which may be responsible for the poor prognosis of the malignancy. We then validated the findings with public bulk transcriptome datasets involving 22 cancer types. Cox regression analysis revealed that the CRDscore was a valuable prognostic biomarker. A model containing 23 circadian-related genes performed well in predicting immunotherapeutic outcomes in 14 independent cohorts. Importantly, decreased CRDscore was detect by RNA sequencing on H1299 cells with melatonin treatment. Meanwhile, the cells downregulated the expression level of SNAIL and TWIST, which contributed to an invasive phenotype. In conclusion, this study provides a novel computational framework for characterizing CRD status using single-cell transcriptomic data and further confirmed the molecular mechanisms underlying metabolic reprogramming and T cell exhaustion under CRD. The better understanding of the mechanisms may provide new possibilities for incorporating "anticancer approaches based on circadian clocks" into the treatment protocols of precision medicine.