Patient Rated Weight Loss, Karnofsky Performance Status, and Prognosis: A Recursive Partitioning Algorithm (410-B)

Abstract

1.Describe the role of patient-rated outcomes in prognosis.2.Describe a recursive partitioning algorithm.3.Describe a recursive partitioning algorithm to predict survival based on Karnofsky Performance Status and patient-rated weight loss in palliative care patients. Background. Patient-rated outcomes have been shown to be prognostic for survival. Research objectives. We developed and tested a recursive partitioning algorithm (RPA) for prognosis in veterans based on the presence of symptoms and Karnofsky Performance Status (KPS). Methods. We used a database from the VA New Jersey Health Care System of 789 cancer patients (pts) with responses to symptoms in the Condensed Memorial Symptom Assessment Scale (CMSAS), KPS, and survival. These symptoms were lack of energy, lack of appetite, dry mouth, pain, weight loss, feeling drowsy, shortness of breath, constipation, difficulty sleeping, difficulty concentrating, nausea, worrying, feeling sad, and feeling nervous. Ten training sets and ten testing sets were used to develop the best RPA, which was then validated on a second dataset of 1,064 pts seen by VA-VISN 3 palliative care consultation teams—yes/no responses to CMSAS and KPS were available. Results. In the algorithm, patients with KPS > 90%, 80% without weight loss, 60–80% with weight loss, and less than 60% were identified with median survivals of 411, 255, 124, and 37 days, respectively. Patient distress from weight loss was closely correlated with documented weight loss (r = 0.48, p < 0.0001). In the validation set, median Kaplan Meier survivals were 173, 180, 102, and 50 days, (p < 0.0001). For 620 cancer pts, median survivals were 417, 102, 45 and 14 days, (p < 0.0001), and (blank), 276, 181 and 100 days for 163 non-cancer diagnosis pts (p non-significant). Conclusion. An RPA that incorporates KPS and patient-rated weight loss developed four prognostic groups in palliative care cancer patients. A larger sample is needed in pts with non-cancer diagnoses. Supported by VA HSRD, PDIA, and NJCCR. Implications for research, policy, or practice. This algorithm may be useful for research and practice in stratifying patients for survival. Physical Aspects of Care