An Administrative Data Algorithm to Predict Inpatient Death at One Tertiary Cancer Center
MetadataShow full item record
CitationLamont, Elizabeth. 2018. An Administrative Data Algorithm to Predict Inpatient Death at One Tertiary Cancer Center. Master's thesis, Harvard Medical School.
AbstractImportance – Despite almost universal agreement that the end of life is a time during which medical care should be directed at alleviation of symptoms and provided at home if possible, one quarter to one half of age-eligible Medicare beneficiaries still die in acute care hospitals.
Objective – In this single institution study, we evaluated a decedent cohort of patients with histories of solid tumors to determine patient attributes associated with risk of subsequent inpatient death at our institution.
Design – Using retrospective data, we divided the decedent cohort (N=11,614) into training (n=5,807) and testing patient cohorts (n=5,807). Using logistic regression to model inpatient death with the training cohort, we then applied it to the testing cohort.
Setting – A large tertiary care hospital in Boston, MA.
Participants – An inception cohort of all 11,614 patients who had died between 1999-2013 with histories of breast, lung, pancreatic, or prostate cancer in the hospital tumor registry.
Main Outcome and Measure – Inpatient death at our hospital vs. death in any other setting.
Results – Overall, 12% (1,461/11,614) of the decedent cohort died as inpatients at our hospital. In the training cohort, the attributes that portended for terminal hospitalization were primarily prior admissions for management of symptoms stemming from progressive cancer. Admission for hypercalcemia, embolism, ascites, supportive care, and blood transfusions were among the predictors. The model was robust when evaluated using the test cohort. Further the interval between the discharge date from the penultimate admission and the admission date for the terminal had a median value of 48 days and an interquartile range of 13 to 211 days.
Conclusions - We found prior admissions for symptoms of progression of advanced cancer were strong predictors of subsequent terminal hospitalizations, with half of terminal admissions occurring within the seven weeks following discharge from the preceding admission. Most of the predictive conditions are expected sequelae of advanced cancer and manageable at home via palliative care providers. For patients at high risk for terminal readmission, the ambulatory post-admission follow-up visit may be the last moment to determine prefer place of death if nothing else.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37366135