[Skip to Navigation]
[Skip to Navigation Landing]
Views 160
Citations 0
Comment & Response
December 13, 2018

Assessing the Prognostic Value of the Automated Bone Scan Index for Prostate Cancer

Author Affiliations
  • 1Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, Massachusetts
  • 2Department of Medicine, Icahn School of Medicine at Mount Sinai, New York, New York
JAMA Oncol. 2019;5(2):270. doi:10.1001/jamaoncol.2018.5857

To the Editor Armstrong and colleagues1 conducted an important study to assess the use of the automated Bone Scan Index (aBSI) as a tool in predicting, among other outcomes, overall survival (OS) in patients with metastatic prostate cancer. As a continuous prediction score, the aBSI’s hazard ratio (HR) for OS was statistically significant, with a lower aBSI associated with better survival. However, evaluations based on HRs are difficult to interpret clinically.2-4 Moreover, the concordance index for the discriminative ability of the aBSI was only 0.63, which suggests that the continuous aBSI score may not be an effective prediction tool at the individual patient level. In practice, one may use the aBSI to stratify patients into several ordered categories. When stratifying by quartiles of the aBSI score, Armstrong and colleagues1 reported observed median OS times (lowest to highest aBSI quartiles) of 34.7, 27.3, 21.7, and 13.3 months, respectively. These values appear to demonstrate the clinically interpretable discriminative ability of the stratified aBSI. Unfortunately, statistical inference for median times across strata was not provided. To further investigate, we generated 95% CIs of median OS times for patients in quartile 1 to quartile 4 using reconstructed data from the Kaplan-Meier curves in Figure 2A.1 Some CIs (quartile 2 and quartile 3) overlapped. That is, the true median OS times of patients in quartile 2 and quartile 3 may be identical, suggesting that the stratified aBSI may lack discriminative capability. The wide, overlapping CIs likely resulted from unstable median time estimates.