Previously published multivariate analyses have not considered date of entry as a possible factor associated with length of stay (LOS), nor has the pattern of decreasing LOS been analyzed. We set out to assess mean LOS over time and to determine the factors, including date of death, which are independently associated with LOS. Cancer patients admitted to our hospice from 1996 through 2001 were assessed for dates of admission/discharge/death, age, gender, race, specific cancer diagnosis, referring physician characteristics, place of death, and heath insurance type. Statistical process control (SPC) charts and proportional hazard models were constructed for patients prioritized by date of admission, with active or discharged patients censored. A total of 2126 patients were analyzed. An abrupt and significant drop in mean LOS was seen for all cancer patients and for most cancer sites from April to December 1998 (temporally associated with a “Special Fraud Alert” issued by the Office of Inspector General) and again in the second and third quarter of 2000. A proportional hazards model revealed that LOS was associated with cancer site (p < 0.0001), quarter in which patient was admitted (p = 0.0020), and sex (women surviving longer, p=0.013), age (older patients surviving longer, p 0.0149), and insurance (p = 0.071). Mean LOS is associated with date of admission to hospice independent of other associated factors. LOS decreases do not occur in a gradual, continuous fashion but suddenly and intermittently, and they are not associated with changes in referral numbers or read-missions. SPC charts proved to be an effective method of tracking and evaluating hospice LOS on an ongoing basis.