Abstract: Evictions are a major social and public health concern in the United States. The development of Natural Language Processing (NLP) technologies allows for analysis of medical record notes to identify eviction cases in healthcare systems. The current study uses medical records data from the largest integrated healthcare system in the United States to develop a surveillance system to estimate incidence rates of NLP-identified evictions (NIEs) and associated patient characteristics. Data on over 8.5 million unique patients in the Veterans Affairs (VA) healthcare system from March 2018-March 2020 were analyzed and NLP was used to identify incidences of eviction. The 2-year incidence rate for NIEs was 2.38% (95% CI = 2.37-2.39%), with an annualized rate of 1.37% (95% CI = 1.36-1.38%). Logistic regression analyses found greater risk for NIEs among patients who were 45-64 years old, were male, non-Hispanic Black, were unmarried, had a high school education or less, had annual household income equal to or below $45,000, lived in an urban area, lived in a high area deprivation index, lived in the West region of the country, and had a history of military sexual trauma. Patients with a history of homelessness (aOR = 6.45; 95% CI = 6.36-6.54), and diagnoses of drug use disorder (aOR = 2.53; 95% CI = 2.49-2.57) or schizophrenia (aOR = 1.88; 95% CI = 1.83-1.93) were also at greater risk for NIEs. These findings suggest evictions are a rare, but important event among veterans, and may inform homeless prevention efforts by identifying veterans from certain backgrounds at greater risk. This study helps demonstrate the utility of using NLP for a surveillance system to identify evictions and track changes over time.