The prognostic value of patient-reported outcomes (PROs) for survival outcomes in cancer patients: A systematic review.
Cottone F., Collins GS., Giesinger JM., Sommer K., Deliu N., Vignetti M., Anota A., Efficace F.
e18223 Background: Several studies have indicated that PROs may independently predict survival outcomes across various cancer populations. We aimed to synthesize current evidence in this research area and to investigate methodological quality of these studies. Methods: We conducted a systematic review of published studies (from January 2016 to December 2017) in oncology, which included at least one PRO domain in a multivariable prognostic model for survival outcomes. We used absolute and relative frequencies to assess the main characteristics of the studies, including tumor site, type of PRO questionnaire, type of PROs assessed (e.g., symptoms or functional scales) and their statistical significance for survival outcomes. As such studies pose important statistical challenges, we also investigated methodological issues based on a comprehensive set of predefined criteria. Results: Out of 385 candidate articles, 65 studies met eligibility criteria. Out of these, n = 56 (86%) reported at least one PRO as a statistically significant prognostic factor in a multivariable prognostic model and thirty-two studies (49%) were done with patients with metastatic/advanced cancer. Overall, the most used instrument to assess PROs was the EORTC QLQ-C30 questionnaire (n = 29, 45%) and the most frequent tumors were lung (n = 18, 28%), esophagogastric (EG) and gynaecological (GY) cancer (each n = 11, 17%). Among these, statistically significant prognostic PRO scales from the EORTC-QLQC30 included fatigue, pain and appetite loss for lung; nausea/vomiting, pain, dyspnea, appetite loss and constipation for EG and physical and role functioning for GY. With respect to methodological quality, we found important drawbacks. For example, only 24% assessed multicollinearity among PRO scales and only 29% assessed performance of the prognostic multivariable model (e.g., evaluating discrimination and calibration). Conclusions: In recent years, a remarkable number of studies has found that PROs provide prognostic information beyond well-established clinical and socio-demographic factors. However, some key methodological limitations have been found, which deserve special attention in future studies.