PCN110 STATISTICAL METHODOLOGY IS CRUCIAL IN PROGNOSTIC FACTOR ANALYSIS OF HEALTH-RELATED QUALITY OF LIFE

Abstract

We read with great interest the recent publication by Park et al. [1] entitled Self-reported health-related quality of life predicts survival for patients with advanced gastric cancer treated with first-line chemotherapy. We congratulate the authors on their important work. This topic is critically relevant to patients, clinicians, researchers, and policy makers. Further research in this field and detailed reporting of the results of prognostic factor analysis of health-related quality of life (HRQOL) data in cancer patients are surely needed. Unfortunately, the question still remains: do baseline patient HRQOL data provide additional prognostic information to clinical factors in cancer patients? After nearly two decades of research, from one of the early pivotal publications by Kaasa et al. [2] dealing with psychosocial well-being as a prognostic factor for survival in patients with inoperable non-small-cell lung cancer to the present, this remains a debatable area. Recently, Gotay et al. [3] published an extensive and rigorous review of publications pertaining to the prognostic value of HRQOL in predicting cancer patient survival. Of the 37 studies reviewed, 34 found that, in the multivariate analysis controlled for major clinical predictors, at least one HRQOL score was significantly associated with survival, supporting an independent association between HRQOL and survival. However, the considerable diversity in patient groups, types of HRQOL measures used, and analytical strategies complicate cross-study comparisons to such an extent that current evidence is still inconclusive. The results from the Park et al. study are a further important contribution to this field of research, illustrating how vital social functioning appears to be in predicting survival. Recently we looked into this area critically, undertaking a systematic review focusing on the statistical methodology used in recent HRQOL prognostic factor studies, in order to identify key issues for HRQOL prognostic factor analyses and to propose recommendations for future outcome research [4]. The prognostic factor analysis of HRQOL data is complex as is the statistical methodology. We concluded that the results published lacked adequate validation. Validation is the only way to avoid overfitting (i.e., developing models too dependent on the dataset at hand so that its value on new data is dubious), which is a common problem, particularly when many factors are analyzed. Furthermore, most publications lacked an assessment of the added prognostic value of HRQOL factors compared to clinical factors alone (the burden of collecting HRQOL data for that purpose should be balanced by a substantial increase in the predictive accuracy of the prognostic models for survival). In the current paper by Park et al., social functioning was identified as an independent prognostic factor. Social functioning was also identified as an important independent prognostic factor for survival in metastatic colorectal cancer patients [5] and validated as such in an independent setting [6]. This replication improves the credibility of the prognostic value of social functioning in metastatic colorectal cancer patients beyond the hypothesis of a mere chance finding. The prognostic value of social functioning in the Park et al. study appears impressive (HR = 0.36, 95% CI of 0.21–0.62 in the multivariate model) and the survival curves by baseline social functioning scores (\67 and C67) demonstrate a significant difference in survival rates at 12 months (45.3% versus 18.3%). However, the modeling M. E. Mauer (&) A. Bottomley C. Coens Quality of Life Department, European Organization for Research and Treatment of Cancer, Brussels, Belgium e-mail: Murielle.Mauer@eortc.be