Prediction of Survival for Advanced Cancer Patients by Recursive Partitioning Analysis: Role of Karnofsky Performance Status, Quality of Life, and Symptom Distress

Abstract

We performed an exploratory recursive partitioning analysis (RPA) in 429 metastatic cancer patients who had completed a Functional Assessment of Cancer Therapy–General (FACT-G) and a Memorial Symptom Assessment Scale–Short Form (MSAS-SF) to define survival prognostic groups. The Cox model analysis also was performed. Both RPA and Cox models included Karnofsky performance status (KPS), age, FACT-G subscales, and MSAS-SF subscales as survival predictors. Of 429 patients, 348 patients (81.1%) had expired at time of analysis. The median age was 67 years (27–89), with median length of survival of 147 days. The RPA identified four distinct survival groups (p < .0001) with three variables: KPS, physical well-being, and physical symptom distress. The most significant split was KPS of 50%, followed by physical well-being score of 25 and physical symptom distress score of 0.6. The median survival time was 29 days for patients with KPS < 50%; 146 days for patients with KPS ≥ 50% and physical well-being < 25; 292 days for patients with KPS > 50%, physical well-being ≥ 25, and physical symptom distress score > 0.6; and 610 days for patients with KPS ≥ 50%, physical well-being ≥ 25, and physical symptom distress score ≤ 0.6. The Cox model found, in addition to KPS (p < .0001) and physical well-being (p = .08), different predictors: psychological symptom distress (p = .0007), global distress index (p = .02), and age (p < .0001). We concluded that the KPS, quality of life, and symptom distress scores can be combined to define prognostic groups. Such models may be helpful for clinical decision making.

• Recursive partitioning analysis
• Survival prediction
• Karnofsky performance status (KPS)
• Quality of life
• Symptom distress
• Advanced cancer
• Cox proportional hazard model