OBJECTIVE To describe the prevalence of, timing of, and factors associated with decisions not to hospitalize nursing home residents with advanced dementia who were dying. METHODS Retrospective cohort study. METHODS Six hundred seventy five-bed nursing facility in Boston. METHODS Two hundred forty residents in a teaching nursing home who died between January 2001 and December 2003 with advanced dementia. METHODS The prevalence and timing of do-not-hospitalize (DNH) orders were determined from the medical record. Data describing demographic characteristics, health conditions, advance care planning, sentinel events, and health services usage during the last 6 months of life were examined. Factors associated with having a DNH order were identified. RESULTS At the time of death, 83.8% of subjects had a DNH order. The prevalence of DNH orders was 50.0% and 34.4%, 30 and 180 days before death, respectively. Hospital transfers were common during the last 6 months of life (24.6%). Factors independently associated with having a DNH order before death included surrogate decision-maker was not the subject's child (adjusted odds ratio (AOR)=4.39, 95% confidence interval (CI)=1.52-12.66), eating problems (AOR=4.17, 95% CI=1.52-11.47), aged 92 and older (AOR=2.78, 95% CI=1.29-5.96), and length of stay 2 years or longer (AOR=2.34, 95% CI=1.11-4.93). CONCLUSIONS For most institutionalized persons with advanced dementia, a decision to forgo hospitalization is not made until death is imminent. Thus, hospital transfers are common near the end of life. The finding that DNH orders are associated with patient and surrogate factors can help clinicians identify cases in which decisions to forgo hospitalizations may be facilitated.