The statistical analysis found that the mortality rate of COVID-19 infection experienced a significant decline in the early stage of the epidemic. We suspect that the sharp deterioration of virus toxicity is related to point mutation and the deletion of the untranslated region of the virus genome. Through sequence analysis of mega-genome data, we found that the genome length of COVID-19 was deleted, which mainly occurred in the untranslated regions at both ends. Sequence similarity analysis further indicated that short UTR length strain emerged by deleting strain with long sequence length. This process is irreversible; the genome with a short sequence length could not restore to the long sequence length. By studying the relationship between genome length and mortality, we found a good correlation between them statistically, which demonstrated that the deletion of the untranslated region of the virus significantly affected the toxicity of the virus. We extracted the viral genome length of patients with different symptoms from the GISAID database for analysis to confirm this relationship. It discovered that the viral genome length of hospitalized patients was significantly more extensive than that of asymptomatic patients. In contrast, the viral genome length of asymptomatic patients was considerably longer than that of ordinary patients with symptoms. To further prove this idea, we performed a genome-level mutation scanning to systematically evaluate the influence of mutations at each position on virulence. After Pearson correlation analysis and chi-square investigation, UTR deletion was the primary driving force in alternating virus virulence. All those statistical evidence support the UTR deletion theory in SARS-COV-2. In the end, we proposed a mathematical model to explain why its UTR deletion was not continuous, which indicated humans could not eliminate SARS-COV-2 in a short time without robust intervention procedures.