Patients' reports or clinicians' assessments: which are better for prognosticating?

Paddy Stone, Bridget Gwilliam, Vaughan Keeley, Chris Todd, Matthew Gittins, Laura Kelly, Stephen Barclay, Chris Roberts

The Prognosis in Palliative care Scale (PiPS) predicts survival in advanced cancer patients more accurately than a doctor's or a nurse's estimate. PiPS scores are derived using observer ratings of symptom severity and performance status. The purpose of this study was to determine whether patient-rated data would provide better prognostic estimates than clinician observer ratings. 1018 subjects with advanced cancer no longer undergoing tumour-directed therapy were recruited to a multi-centre study. Prognostic models were developed using observer ratings, patient ratings or a composite method that used patient ratings when available or else used observer ratings. The performance of the prognostic models was compared by determining the agreement between the models' predictions and the survival of study participants. All three approaches to model development resulted in prognostic scores that were able to differentiate between patients with a survival of 'days', 'weeks' or 'months+'. However, the observer-rated models were significantly (p<0.05) more accurate than the patient-rated models. A prognostic model derived using observer-rated data was more accurate at predicting survival than a similar model derived using patient self-report measures. This is clinically important because patient-rated data can be burdensome and difficult to obtain in patients with terminal illnesses.