Although a helpful review on this topic was published recently [
115], the present review, to the author's best knowledge, is the first comprehensive study examining the prognostic value of quality of life data for survival time in cancer patients. The review contained 104 studies and with only a few exceptions, the results in most instances indicated that health-related quality of life data or some quality of life measures were significant predictors of survival duration.
The early studies reported here used
ad hoc instruments, while more recent studies used well-validated cancer-specific quality of life questionnaires. The most recent studies supplemented their assessments with site-specific questionnaires. Overall, 59 different instruments have been used to measure quality of life in cancer patients [Additional file
1]. The EORTC QLQ-C30 was found to be the most widely used cancer-specific instrument, and as the tables in this review show, the questionnaire often gave fairly consistent and reliable results. In addition, the supplementary EORTC quality of life modules, such as QLQ-BR23, QLQ-LC13 and QLQ-BN20, proved very useful instruments for analysing prognostic indicators, provided that other methodological requisites were ensured. Such instruments could even capture information important to the patients and thus provide better prognostic profiles, enabling clinicians to manage cancer patients more effectively. However, with regard to instruments listed in the tables, one should note that some of them were used for a tailor-specific study, treatment or trial such as the Daily Diary Card (DDC) and the Auckland Quality of Life Questionnaire. Evidently some instruments were well-known generic measures, such as the SF-36, a psychological instrument such the Hospital Anxiety and Depression Scale (HADS), and the General Health questionnaire (GHQ), and/or symptom measures such the Brief Pain Inventory (BPI), and the Symptom Distress Scale (SDS). Therefore the information given in the tables was simply to reflect the variance that existed in the instruments used and neither to convey their psychometric validity nor indicate that they were cancer-specific. As such, the results from studies that used
ad hoc instruments, a study-specific questionnaire or only general measures should be interpreted with caution.
Many studies reported that the global or the overall quality of life was a significant independent predictor of survival. Global quality of life is a straightforward measure, asking people to evaluate their own health status or quality of life individually (or in combination). It is argued that measures such as global quality of life are patient-rated and thus have the potential to reflect the patient's well-being better than a physician's observed indicators. However, it has (for instance) been recommended that since the global quality of life scale of the EORTC QLQ-C30 is highly correlated with other scales, it should not be included in prognostic indicator analyses when other variables from the EORTC QLQ-C30 are used, in order to achieve model stability [
116]. This might explain why a recent review on the association of psychosocial factors with survival in head and neck cancer found that the baseline overall quality of life and depression were not predictors of survival [
85]. In addition, when quality of life is included in prognostic indicator analyses, pre-treatment (baseline) and follow-up assessments should be distinguished. Furthermore, the relationship between baseline health-related quality of life data and survival refers to disease-specific characteristics, while follow-up health-related quality of life data and survival in addition refer to treatment-specific characteristics. Indeed baseline data are more often reported to be prognostic because they are more straightforward to assess. However, collecting follow-up data is a major challenge and should be encouraged, since pre-treatment quality of life data were not prognostic for survival times in some cancers, while changes in quality of life scores or follow-up data were usually prognostic in these occasions. More importantly, tumour type and stage of disease are essential for drawing conclusions from such findings. In many studies, quality of life data were prognostic indicators of survival duration in patients with solid tumours and advanced diseases, but not in those with soft tumours and early-stage diseases.
Several measures, such as physical functioning, showed particularly significant associations with survival duration in cancer patients. It is argued that physical functioning might be a surrogate marker for an unrecognized biological prognostic indicator, so a causal association between physical functioning and survival time should not be inferred [
65]. In addition, it is argued that since performance status and physical functioning are significantly associated with each other, in many instances when one includes both physical functioning and performance status in the regression models, the likelihood of finding inconsistent results can be expected. In other words, in such circumstances in some studies physical functioning would emerge as an independent prognostic factor and in some others performance status or even in certain cases both might be found prognostic factors for survival duration. Thus, as indicated earlier, the role of physical functioning and performance status in prognostic studies need to be evaluated with caution. A recent meta-analysis of the relationship between baseline quality of life data from the EORTC clinical trials and survival indicated that physical functioning was a significant independent prognostic factor but performance status (as measured by the World Health Organisation performance status) was not [
5], whereas a study in metastatic kidney cancer patients reported that both physical functioning and performance status were correlated with a longer progression-free survival [
117].
Among symptoms, appetite loss, pain and fatigue at baseline were the most important or strongest independent predictors of survival in many of the studies on different cancer populations. One possible explanation is that these symptoms are very sensitive markers of patient well-being. In addition, as explained by Efficace et al. [
58], such findings might arise because quality of life measures in effect mask each other in multivariate analyses, so making variables such as appetite loss or pain or fatigue appear to be the most important or strongest predictors of survival time. Another possible explanation is that such symptoms might reflect, for instance, weight loss, which itself is an important prognostic indicator.
As suggested by Gotay et al. [
115], there are several explanations for the association between health-related quality of life data and survival duration in cancer outcome studies. They summarized four possible explanations: (i) quality of life measures include different items and thus provide more sensitive information than traditional performance status and toxicity measures; (ii) quality of life data especially those collected at baseline before disease progression could pick up relevant information earlier than established clinical prognostic factors; (iii) quality of life data are markers of patients' behaviour because they relate to diagnosis, treatment and subsequent outcomes of the disease; and (iv) quality of life data are markers of individual characteristics such as personality style and adapting coping strategies, which affect the disease process and outcomes in cancer patients.
In addition, the relationship between measures such as global quality of life or self-rated health and survival or mortality might be explained in the context of the body-mind relationship [
118-
120]. For instance, a recent publication on the topic concluded that self-reported health is a unique indicator of human health status; its origins lie in a process whereby information from the individual's body and mind is received, selected, reviewed and summarized and therefore it could predict the most absolute biological events, such as survival or death [
121].
The current review, however, suggests an additional explanation that might be helpful in interpreting the findings from studies of the relationship between quality of life data and survival duration. Quality of life data might be markers of the socio-economic status of cancer patients. Evidence for a relationship between socio-economic status and survival time for many cancers is being compiled [e.g. see [
122-
130]]. In this context, a cancer patient's socio-economic status predicts survival. For instance, cancer patients with higher social class would have a better quality of life [e.g. see [
131-
134]], and consequently those who report a better quality of life at baseline assessment may live longer. Thus it is not surprising that, in addition to clinical measures, quality of life data are predictive of survival duration. This hypothesis needs further assessment. In future studies on the relationship between quality of life data and survival duration, in addition to biomedical measures, adjustments should be made for patients' socioeconomic status. It would then remains to be seen whether health-related quality of life data still act as significant independent predictors of survival or not. However, the known clinical measures that most studies frequently entered into a multivariate model included age at diagnosis; gender (where necessary); stage (tumour characteristics); occurrence of metastases (or number of metastatic sites involved); weight loss; laboratory parameters (where necessary); performance status and type of treatment. It seems that co-morbidity, and measures of patients' socioeconomic status (for example income, education, occupation, living conditions or social class) are also important to be included in the final model when one considers assessing the relationship between quality of life data and survival duration.
Although this review has included studies that examined the relationship between quality of life data and survival, it excluded purely psychological studies. There are several useful studies on association between psychological data and survival and thus if one wishes to have a better understanding on the topic it is necessary to review these papers as well. For instance, a systematic review of the literature clearly documented the influence of psychological coping on survival and recurrence in cancer patients [
135]. The review concluded that there is little consistent evidence that psychological coping style is important in survival from or recurrence of cancer. Similarly, a systematic review of the effect of psychosocial factors on breast cancer outcome indicated that, although most studies on the topic have shown a significant relationship between psychosocial factors and survival, the relevant psychosocial variables were neither consistently measured across studies nor, in many cases, consistent in their findings [
136]. In contrast, a recent review on the relationship between stress-related psychosocial factors and survival in cancer patients indicated that stressful life experiences were related to poorer cancer survival and higher mortality. It also suggested that stress-prone personalities or unfavourable coping styles, and negative emotional responses or poor quality of life, were related to poorer cancer survival and higher mortality [
137]. However, some papers that belonged in principle to the discipline of psychology were inevitably included in the present review. These papers usually reported that a measure of quality of life had been incorporated in the study, but no well-known instruments were used for the measurements. Contrary to expectation, these papers found that, in multivariate analyses, conditions such as over-positive reporting of quality of life [
91] or having a better appetite were indicators of shorter survival [
54].
Finally, the inherent limitations and controversial issues related to studies of relationship between survival and quality of life data should not be neglected. For example, many studies reporting on a positive relationship between survival and quality of life data originate from previously conducted randomised clinical trials. Although this is the best-known methodology to evaluate treatments outcomes, it can also be argued that, since patients in randomised clinical trials have highly selected criteria (e.g. no associated co-morbidity), one might wonder whether this association also works in the real world [
10]. Perhaps only by testing this hypothesis in an observational setting would it be possible to actually verify whether health-related quality of life parameters have a prognostic value. In addition, since most evidence on positive relationship between quality of life data and survival comes from studies with different patients groups, or studies that used different instruments to measure quality of life, or studies that applied different statistical methodology (and sometimes even inappropriate statistical analysis), thus cross-study comparisons are impossible or very complicated, indicating that current evidence is still inconclusive [
138]. With regard to statistical analysis, it is argued that statistical methodology is crucial in prognostic factor analysis of health-related quality of life where different statistical strategies can lead to different findings. Mauer et al. suggest at least two recommendations to increase a substantial accuracy of the prognostic models for relationship between quality of life data and survival: validation strategy, and added prognostic value of health-related quality of life factors analysis. They refer to the former as the only way to avoid over-fitting logistic regression models. These are regression model that are too dependent on the data set at hand, making its value on new data doubtful. The latter strategy, however, refers to computing predictive accuracy of the final model (including health-related quality of life data and known clinical prognostic factors) and comparing it with the predictive accuracy of the model with known clinical prognostic factors only, using for instance, C-indexes [
138]. More technical details of Mauer et al. arguments and recommendations can be found elsewhere [
139].
This review included all major search engines in combination with a manual search. However, since the strategy was based on keywords in the titles of English language publications, there is a risk that some relevant papers were missed. Furthermore, individual reports were not examined in detail, and so the findings are not all-inclusive. Bottomley and Efficace have also remarked in their editorial comments that it seems necessary to stress that studies on the relationship between quality of life data and survival duration have yielded considerable evidence, but this is still a relatively novel area of research in oncology and has a long way to go. They suggested that more hypothesis-driven prospective studies are needed to provide robust evidence that health-related quality of life data and patient-reported outcomes independently predict survival duration [
140].
1,2
This article has been 
