The accuracy of predicting survival in individual patients with cancer

Abstract

Object Estimating survival time in cancer patients is crucial for clinicians, patients, families, and payers. To provide appropriate and cost-effective care, various data sources are used to provide rational, reliable, and reproducible estimates. The accuracy of such estimates is unknown. Methods The authors prospectively estimated survival in 150 consecutive cancer patients (median age 62 years) with brain metastases undergoing radiosurgery. They recorded cancer type, number of brain metastases, neurological presentation, extracranial disease status, Karnofsky Performance Scale score, Recursive Partitioning Analysis class, prior whole-brain radiotherapy, and synchronous or metachronous presentation. Finally, the authors asked 18 medical, radiation, or surgical oncologists to predict survival from the time of treatment. Results The actual median patient survival was 10.3 months (95% CI 6.4–14). The median physician-predicted survival was 9.7 months (neurosurgeons = 11.8 months, radiation oncologists = 11.0 months, and medical oncologist = 7.2 months). For patients who died before 10 months, both neurosurgeons and radiation oncologists generally predicted survivals that were more optimistic and medical oncologists that were less so, although no group could accurately predict survivors alive at 14 months. All physicians had individual patient survival predictions that were incorrect by as much as 12–18 months, and 14 of 18 physicians had individual predictions that were in error by more than 18 months. Of the 2700 predictions, 1226 (45%) were off by more than 6 months and 488 (18%) were off by more than 12 months. Conclusions Although crucial, predicting the survival of cancer patients is difficult. In this study all physicians were unable to accurately predict longer-term survivors. Despite valuable clinical data and predictive scoring techniques, brain and systemic management often led to patient survivals well beyond estimated survivals.