Feuer EJ, Lee M, Mariotto AB, Cronin KA, Scoppa S, Penson DF, Hachey M, Cynkin L, Carter GA, Campbell D, Percy-Laurry A, Zou Z, Schrag D, and Hankey BF. The Cancer Survival Query System: Making survival estimates from the Surveillance, Epidemiology, and End Results Program more timely and relevant for recently diagnosed patients. Cancer. 2012;118:5652-62.
The authors wish to explain some specific corrections to their article, as follows.
We have continued our work on the development of the Cancer Survival Query System (CSQS) that was described in our manuscript. In our attempts to further refine the methods, particularly in regard to other cause survival, we discovered the need to correct how the validation statistics reported in the manuscript were calculated. Below are a revised Table 4 and Figure 2. The revisions to Table 4 are very minor with all but one of the 11 numbers that were corrected changing +/− .01, and one number changing -.03.
|Aged ≥66 Years With Comorbidity Scores|
|Aged <66 Years Without Comorbidity Scores|
In the revised Figure 4, the left hand panels showing the agreement between cumulative observed and model predicted death from cancer have changed very little from the original Figure 4. The upper right hand panel of the revised Figure 4 showing the agreement between cumulative observed and model predicted death from other causes for cases under age 66 without co-morbidity scores shows significant over-prediction for the model, while the lower right hand panel for cases aged 66+ with co-morbidity scores shows modest over-prediction for the model. It is not surprising that without a co-morbidity score the model cannot predict other cause mortality with very good accuracy, and this is reflected in the ROC statistics in Table 2. The modest over-prediction of other cause mortality using co-morbidity is probably due to a “healthy-screener” effect, since many of the men diagnosed with prostate cancer are screen detected, and may have better mortality from other causes associated with their health-seeking behavior even after adjusting for co-morbidity. We are working on ways to adjust for factors affecting other cause mortality for men under age 66, and to correct for the “healthy-screener” effect for prostate cancer patients.
The authors regret these errors.