S M Bentzen1. 1. Biostatistics in Oncology Group, Gray Laboratory Cancer Research Trust, Mount Vernon Hospital, Northwood, Middlesex, UK. bentzen@graylab.ac.uk
Abstract
OBJECTIVE: Prognostic factor studies are abundant in oncology. Nevertheless, most of them have very limited impact on clinical practice, in part because many of them have a low statistical power. The importance of statistical power is illustrated using bootstrap resampling of data from a series of osteosarcoma patients. METHODS AND MATERIALS: Osteosarcoma is a rare disease, the incidence being just a few cases per million person-years in the Western World. Very few Phase III studies have been conducted in the disease, and much knowledge on therapeutic progress has come from Phase II studies. This has caused controversy concerning the validity of historical controls, which again has stimulated interest in the identification of prognostic factors in this disease. A literature search in the National Library of Medicine MEDLINE database was performed to identify prognostic factor studies in osteosarcoma published between 1975 and 1998. Monte Carlo methods, so-called bootstrap resampling, are used to investigate the importance of sample size based on survival data for a previously published series of 158 osteosarcomas treated with surgery alone. RESULTS: Most published studies are too small to provide useful information on prognostic factors in osteosarcoma. Three-quarters of the papers reviewed included less than 100 patients with osteosarcoma. More than 20 different potential prognostic factors were included in these papers. Inherent differences between studies and poor reporting hamper a synthesis of information from various studies. The results from the bootstrap resampling illustrate how the majority of published studies would miss even quite significant prognostic factors. CONCLUSIONS: An effort is needed to improve the design, conduct, and reporting of prognostic studies in oncology.
OBJECTIVE: Prognostic factor studies are abundant in oncology. Nevertheless, most of them have very limited impact on clinical practice, in part because many of them have a low statistical power. The importance of statistical power is illustrated using bootstrap resampling of data from a series of osteosarcomapatients. METHODS AND MATERIALS: Osteosarcoma is a rare disease, the incidence being just a few cases per million person-years in the Western World. Very few Phase III studies have been conducted in the disease, and much knowledge on therapeutic progress has come from Phase II studies. This has caused controversy concerning the validity of historical controls, which again has stimulated interest in the identification of prognostic factors in this disease. A literature search in the National Library of Medicine MEDLINE database was performed to identify prognostic factor studies in osteosarcoma published between 1975 and 1998. Monte Carlo methods, so-called bootstrap resampling, are used to investigate the importance of sample size based on survival data for a previously published series of 158 osteosarcomas treated with surgery alone. RESULTS: Most published studies are too small to provide useful information on prognostic factors in osteosarcoma. Three-quarters of the papers reviewed included less than 100 patients with osteosarcoma. More than 20 different potential prognostic factors were included in these papers. Inherent differences between studies and poor reporting hamper a synthesis of information from various studies. The results from the bootstrap resampling illustrate how the majority of published studies would miss even quite significant prognostic factors. CONCLUSIONS: An effort is needed to improve the design, conduct, and reporting of prognostic studies in oncology.