A global analysis of multitrial data investigating quality of life and symptoms as prognostic factors for survival in different tumor sites


  • We thank the European Organization for Research and Treatment of Cancer (EORTC) Headquarters, EORTC Clinical Groups, and all of the principal investigators (W. Albrecht, J. Becker, R. E. Coleman, T. Conroy, R. de Wit, A. Eggermont, S. Fossa, G. Giaccone, J. C. Horiot, F. Keuppens, C. H. Koehne, J. L. Lefebvre, F. Levi, G. O. N. Oosterhof, M. Piccart, P. Postmus, H. J. Schmoll, E. F. Smit, T. A. W. Splinter, M. J. Taphoorn, P. Therasse, J. Van Meerbeeck, and H. Van Poppel) who helped us better understand the needs of cancer patients, which will ultimately lead to better patient care. We also extend a warm thanks to Cheryl Whittaker for proofreading the article. We give very special thanks to all of the patients who participated in these trials.

  • The sponsors had no role in the design of the study; the collection, analysis, or interpretation of the data; the writing of the article; or the decision to submit the article for publication.



The objective of this study was to examine the prognostic value of baseline health-related quality of life (HRQOL) for survival with regard to different cancer sites using 1 standardized and validated patient self-assessment tool.


In total, 11 different cancer sites pooled from 30 European Organization for Research and Treatment of Cancer (EORTC) randomized controlled trials were selected for this study. For each cancer site, univariate and multivariate Cox proportional hazards modeling was used to assess the prognostic value (P < .05) of 15 HRQOL parameters using the EORTC Core Quality of Life Questionnaire (QLQ-C30). Models were adjusted for age, sex, and World Health Organization performance status and were stratified by distant metastasis.


In total, 7417 patients completed the EORTC QLQ-C30 before randomization. In brain cancer, cognitive functioning was predictive for survival; in breast cancer, physical functioning, emotional functioning, global health status, and nausea and vomiting were predictive for survival; in colorectal cancer, physical functioning, nausea and vomiting, pain, and appetite loss were predictive for survival; in esophageal cancer, physical functioning and social functioning were predictive for survival; in head and neck cancer, emotional functioning, nausea and vomiting, and dyspnea were predictive for survival; in lung cancer, physical functioning and pain were predictive for survival; in melanoma, physical functioning was predictive for survival; in ovarian cancer, nausea and vomiting were predictive for survival; in pancreatic cancer, global health status was predictive for survival; in prostate cancer, role functioning and appetite loss were predictive for survival; and, in testis cancer, role functioning was predictive for survival.


The current results demonstrated that, for each cancer site, at least 1 HRQOL domain provided prognostic information that was additive over and above clinical and sociodemographic variables. Cancer 2014;120:302–311. © 2013 American Cancer Society.


Health-related quality-of-life (HRQOL)[1] domains have become a routine endpoint in clinical trials, although they have been valued mainly when survival gain in clinical trials remains unanswered or when 2 or more interventions yield equivalent survival.[2] Sprangers[3] addresses the limited use HRQOL data received for clinical decision making, despite the relevant information they provide on cancer symptom burden,[4, 5] treatment effectiveness,[6, 7] and the adaptation of the patient to their disease and treatment over time.[8]

Cancer patients themselves provide a unique and needed perspective on their own symptom burden and quality of life, and such ratings may provide clinicians with additional data on patients' chances of survival. Therefore, research has focused lately on whether HRQOL and other patient-reported outcomes (PROs) are prognostic for cancer survival, either alone or alongside biomedical and sociodemographic data. “Patient-reported outcomes” is a broader term that was introduced by the US Food and Drug Administration (FDA) and encompasses not only HRQOL but also information on any aspect of a patient's health status that comes directly from the patient.[9]

Several studies[10-12] previously demonstrated the significant relation between pretreatment HRQOL and survival. This finding has been documented both in a variety of cancer-site-specific studies and in large mixed-malignancy cohorts.[13] A systematic literature review by Gotay et al[14] confirmed the significant relation between patient survival and baseline HRQOL. In 36 of the 39 studies they evaluated, at least 1 HRQOL domain was associated significantly with survival.

These studies often used a wide variety of HRQOL measurement tools, hampering cross-study comparisons. Therefore, it is not surprising that the results are contradictory across studies. Moreover, studies may differ in design and type of statistical analyses, thus further limiting the possibility of meta-analyses.[15]

A previous article[16] by our group made use of an HRQOL data set of 7417 individuals with cancer pooled from 30 European Organization for Research and Treatment of Cancer (EORTC) randomized clinical trials (RCTs). Those results demonstrated the prognostic value of HRQOL for survival in the pooled EORTC data set. Physical functioning, pain, and appetite loss were identified as variables with prognostic value for survival. An important additional strength of the pooled data set is that all HRQOL data were collected on a single, standardized, and validated measure, the EORTC Core Quality-of-Life Questionnaire (QLQ-C30). Gotay et al[14] reported the advantages of data like these, including well characterized samples, the use of consistent treatment protocols not based on PRO ratings, the availability of mature data for adequately powered analyses, and rigorous quality control.

To our knowledge, the relative contribution of different HRQOL domains as prognostic for each separate cancer type has never been examined in a single cohort. Patients with cancer express different frailty profiles, depending on primary tumor site, with each site exhibiting different biologic characteristics that, ultimately, may define different associations of HRQOL with mortality. Our previous work led us to ask, “Does the value of different HRQOL domains for predicting survival vary among different cancer sites?”

Therefore, in this study, we examine the same data, not at an aggregated level as we did previously[16] but per cancer type. The empirical finding that certain HRQOL domains are prognostic for specific tumor sites, alongside biomedical and sociodemographic parameters, may improve clinical decision making, thereby stimulating the use of patient-reported HRQOL information by medical researchers and clinicians.


Study Selection and Data Analysis

In total, 30 RCTs were selected for this study according to the following eligibility criteria: 1) closed EORTC randomized phase 2 or 3 RCTs; 2) overall survival as the primary endpoint and HRQOL as a secondary endpoint; 3) use of a valid baseline EORTC QLQ-C30 questionnaire (version 1, 2, or 3), in which the baseline was set as the date of randomization (a time window of 14 days after the date of randomization was allowed, provided the questionnaire was completed before treatment and the patients were not aware of the treatment received); and 4) statistical analysis was performed using a priori defined EORTC standard operating procedures.

The EORTC QLQ-C30 was designed to measure the HRQOL of patients with cancer. It incorporates 5 functioning scales (physical, role, cognitive, emotional, and social functioning); 3 symptom scales (fatigue, pain, and nausea and vomiting); and a global health status scale. The remaining single items assess additional symptoms commonly reported by cancer patients, including dyspnea, appetite loss, sleep disturbance, constipation, and diarrhea, as well as the perceived financial impact of the disease and treatment.[17] All scale and item scores are linearly transformed to a scale that ranges from 0 to 100. For the 5 functional scales and the global health status scale, a higher score represents a better level of functioning. For the symptom-oriented scales and items, a higher score corresponds to a higher level of symptoms.[18] The EORTC QLQ-C30 versions differ only with regard to scoring of the role functioning scale, the physical functioning scale, and the global health status scale.[19, 20]

Additional data that were evaluated included age, sex, World Health Organization (WHO) performance status, and distant metastases, all factors that have been recognized previously as providing prognostic information and that were collected routinely across the studied cancer sites.[21, 22] Age was dichotomized into ≤60 years versus >60 years based on a consensus of the investigators; and WHO performance status was dichotomized into WHO scores of 0 or 1 (good) versus scores of 2 or 3 (poor), as indicated by current clinical practice.

Statistical Analysis

A Kaplan-Meier plot was generated to graphically explore the survival curves of the different cancer groups. Differences in survival between each cancer group were examined using the chi-square test, which tests whether heterogeneity is present, and the I2 statistic, which quantifies the degree of heterogeneity in a meta-analysis by describing the proportion of total variation that is because of heterogeneity.[24]

Cox proportional hazard (PH) regression models were used to model the prognostic value of HRQOL parameters for each cancer site.[25] Patients were censored when their survival at the end of the study was unknown or if the patients dropped out before the study was completed or were lost to follow-up. The Cox PH model assumptions for both the univariate and multivariate analyses were assessed using a graphic method. The PH requirement, assuming that the hazard ratio (HR) was constant over time, was visually checked using log-log plots, and violation of the requirement was assumed when the lines were not parallel.

The level of significance for the HRQOL scales and clinical and sociodemographic variables was set at P = .05, with overall survival, measured from the date of randomization to the date of death, as the dependent variable. To account for possible heterogeneity within each cancer site derived from pooling the RCTs, each Cox PH model was stratified by trial.

The model selection was performed in several steps. In the univariate model, each sociodemographic, clinical, and HRQOL variable was assessed independently using a criterion of P < .05 to identify prognostic variables for overall survival and to derive a final list of independent predictors. Then, the final list of statistically significant HRQOL predictors from the univariate analysis was implemented in a multivariate model, which was adjusted for the established clinical prognostic factors (age, sex, and WHO performance status) and stratified for distant metastasis, if reported. With regard to the latter, we acknowledge that HRQOL information most likely differs according to the patient's disease stage. For example, in this report, in the article by Gotay et al, and in other studies,[26, 27] physical functioning was revealed as a prognostic parameter for patients with more advanced cancer.

The clinical and sociodemographic variables were forced into the model given their established prognostic value for survival. First, a backward selection procedure was applied to eliminate nonsignificant variables using a criterion of P > .05 in the multivariate framework. In addition, forward selection using a criterion of P < .05 was used to add the nonsignificant individual predictors from the univariate analysis to the final model obtained with backward selection. Finally, a stepwise selection procedure was used to ensure that a final model was obtained in which no further parameters could be either removed or included. The final multivariate model for each cancer group was refitted 1000 times using the bootstrap resampling technique to determine the precision of the estimates and to check for relatively small deviations from the final multivariate model.

Prognostic value was assessed using the HR, its 95% confidence interval, and the P value. The reported HR for the HRQOL scales takes into consideration the “clinically important difference,” which has been defined[28] as the smallest difference in a score in the domain of interest that a patient perceives as beneficial and that would mandate a change in patient management. Osoba et al[29] suggested a difference of 10 points to categorize patients as improved, stable, or worsened on the EORTC HRQOL scales/items. Therefore, the HRs were calculated not for a 1-point difference but for every 10-point difference in scores on the EORTC QLQ-C30 scales/items, which range between 0 and 100.

Finally, for each tumor site, we also performed a sensitivity analysis in which we omitted the clinical trials that had low (≤65%) baseline HRQOL compliance. Although completion of the EORTC QLQ-C30 is entirely voluntary, there may be an association between adherence and disease severity; ie, patients who experience a severe symptom burden at study entry may be reluctant to fill in the questionnaire, with the consequence that questionnaires completed by patients who have good performance status can drive the results. A study by Curran et al[30] demonstrated an inverse relation between HRQOL scores and the probability that a patient would drop out. In contrast, however, our previous study[16] indicated that sicker patients (defined by low performance status and distant metastases) achieved a significantly better level of baseline HRQOL compliance.


Thirty closed EORTC RCTs that were initiated between 1986 and 2004 were eligible for inclusion in this study. Although several studies were closed a few years ago, some of the trials were recently closed and published.[6, 7, 16] One or more RCTs were included for 11 cancer sites: lung (n = 7), melanoma (n = 4), prostate (n = 4), colorectal (n = 3), breast (n = 3), head and neck (n = 3), ovary (n = 2), brain (n = 2), testis (n = 1), pancreas (n = 1), and esophagus (n = 1). Baseline EORTC QLQ-C30 data were available for 7417 of 10,108 patients (73%) in the total data set. Table 1 provides information about adherence to baseline HRQOL for each cancer site. A significant association (P > .05) was noted between baseline HRQOL compliance and cancer site. Low compliance was reported for breast cancer (45%) and head and neck cancer (46%). This was mainly because of low baseline compliance in some specific studies within these disease groups.

Table 1. Compliance With Health-Related Quality-of-Life Assessment by Cancer Site
Cancer SiteNo. of Patients per Cancer SiteHRQOL Present: No (%) [n = 7417; 73%]
  1. Abbreviations: HRQOL, health-related quality of life.

Brain941829 (88)
Colorectal14921141 (76)
Lung16191054 (65)
Esophageal7565 (87)
Ovarian317203 (64)
Prostate687557 (81)
Testicular318257 (81)
Breast587321 (55)
Head & neck841452 (54)
Melanoma31242446 (78)
Pancreas10792 (86)

The distribution of clinical and sociodemographic patient characteristics for each cancer site is displayed in Table 2. The compliance was 100% for both age and sex and was low for WHO performance status (range, 0%-11%) and distant metastases (range, 0%-6%). No information about metastasis was collected for the brain, ovarian, or testicular cancer groups, because these groups maintain different classification systems to define the disease stage. Most patients (61%) were aged >60 years, 40% were women, 88% had good performance status, and 48% had no distant metastases.

Table 2. Distribution in the Number and Percentage of Selected Clinical and Sociodemographic Patient Characteristics for Each of the Studied Cancer Sites in Those who Reported Valid Baseline Health-Related Quality of Life
CharacteristicCancer Site: No. of Patients (%)
BrainColorectalLungEsophagealOvarianProstateTesticularBreastHead & NeckMelanomaPancreas
  1. Abbreviations: WHO PS, World Health Organization performance status.

Age: ≤60 y639 (77)542 (47)623 (59)39 (60)111 (55)76 (13)257 (100)259 (81)307 (68)1938 (79)39 (42)
Sex: Men499 (60)694 (61)753 (71)59 (91)0 (0)557 (100)257 (100)0 (0)401 (89)1398 (57)48 (52)
0-1704 (85)1054 (92)855 (81)50 (77)174 (86)414 (74)0 (0)300 (93)447 (99)2435 (99)74 (80)
2-3122 (15)84 (7)87 (8)15 (23)29 (14)143 (26)0 (0)21 (7)5 (1)5 (<1)18 (20)
Unknown3 (<1)3 (<1)112 (11)0 (0)0 (0)0 (0)257 (100)0 (0)0 (0)6 (<1)0 (0)
Distant metastasis
No0 (0)210 (18)556 (53)0 (0)0 (0)557 (100)0 (0)194 (60)449 (99)2116 (87)21 (23)
Yes0 (0)886 (78)428 (41)65 (100)0 (0)0 (0)0 (0)127 (40)0 (0)330 (13)71 (77)
Unknown829 (100)45 (4)70 (6)0 (0)203 (100)0 (0)257 (100)0 (0)3 (1)0 (0)0 (0)

The Kaplan-Meier survival curves stratified by 11 cancer sites are depicted in Figure 1. The curves indicate that the length of survival varied considerably, depending on the disease site. The figure indicates that patients with testicular cancer survived longer than patients with other cancers, and patients with esophageal cancer had the lowest survival rate. The curve indicates that median survival (50%) was not reached for testicular cancer or melanoma.

Figure 1.

These are Kaplan-Meier survival curves stratified by 11 cancer sites.

The chi-square heterogeneity test demonstrated a significant variation (P = .01) in survival between cancer groups that could not be attributed to chance alone. The results of the I2 test indicate that 52% of the variability could be accounted for by variability between cancer groups, and the other 48% could be account for by variability within cancer groups. A percentage of approximately 50% indicates a medium heterogeneity.[24] Differences in survival and statistical heterogeneity between cancer sites suggest the appropriateness of looking for prognostic HRQOL variables separately within each cancer site.

Table 3 presents the prognostic value of baseline HRQOL and sociodemographic and clinical variables for the different cancer groups derived from the final multivariate model. Bootstrapping did not reveal any deviations from the final models. For each cancer site, the HRs, confidence intervals, and P values are reported for the clinical, sociodemographic, and EORTC QLQ-C30 dimensions. For example, an HR of 1.13 for breast cancer using the emotional functioning quality-of-life scale implies that, for every 10-point increase in the emotional functioning score, the chance of survival in patients who do not have breast cancer increases by 13% compared with those who do have breast cancer when the analysis is adjusted for the other covariates (age, sex, and WHO performance status). For some cancer sites, the prognostic value for survival of the sociodemographic and clinical variables is not calculated because of their unbalanced distribution across the categories. Because we stratified for distant metastasis, no results are reported for this variable.

Table 3. Hazard Ratios of Survival for Sociodemographic, Clinical and HRQOL Scales Across the 11 Cancer Sitesa
 Cancer Site
VariableBrainColorectalLungEsophagealOvarianProstateTesticularBreastHead & NeckMelanomaPancreas
  1. Abbreviations: CI, confidence interval; EORTC QLQ-C30, European Organization for Research and Treatment of Cancer Core Quality of Life Questionnaire; HR, hazard ratio; NA, not applicable; WHO PS, World Health Organization performance status.

  2. a

    HRs were, calculated based on a 10-point difference in scores on the EORTC QLQ-C30 scales/items.

  3. b

    Functional scales: PF, physical functioning; RF, role functioning; EF, emotional functioning; CF, cognitive functioning; SF, social functioning; global quality-of-life scales: QL, global health status; symptom scales/items: NV, nausea and vomiting; PA, pain; DY, dyspnea; AP, appetite loss.

Clinical and sociodemographic variables
Age: ≤60 y vs >60 y
HR1.8781.2911.0281.3261.1051.263 0.8771.2131.1341.359
95% CI1.553-2.2701.111-1.4990.871-1.2120.734-2.3960.748-1.6320.957-1.666NA0.568-1.3540.927-1.5880.949-1.3560.806-2.293
P< .0001.0008.7455.3498.6168.0996 .5544.1590.1672.2501
Sex: Men vs women
HR0.9130.9150.7821.477    0.3810.6090.602
95% CI0.770-1.0820.783-1.0700.653-0.9360.542-4.026NANANANA0.226-0.6440.519-0.7150.356-1.018
P.2930.2676.0074.4463    .0003< .0001.0583
WHO PS: Good vs poor
HR1.5541.2301.2211.1941.5131.564 2.141  1.397
95% CI1.242-1.9440.923-1.6380.923-1.6170.464-3.0710.892-2.5661.261-1.938NA1.170-3.916NANA0.740-2.636
P.0001.1569.1622.7130.1245< .0001 .0135  .3023
EORTC QLQ-C30 scalesb
HR 0.930.920.80   0.86 0.94 
95% CI 0.96-0.990.88-0.960.67-0.94   0.76-0.97 0.90-0.98 
P .0323< .0001.0041   .0119 .0055 
HR     0.960.80    
95% CI     0.94-0.980.64-0.96    
P     .0050.0144    
HR       1.131.08  
95% CI       1.05-1.221.02-1.15  
P       .0020.0100  
95% CI0.90-0.96          
P< .0001          
HR   1.09       
95% CI   1.04-1.34       
P   .0143       
HR       0.92  0.83
95% CI       0.86-0.99  0.70-0.96
P       .0171  .0098
HR 1.06  1.16  1.171.15  
95% CI 1.01-1.07  1.07-1.25  1.06-1.281.04-1.27  
P .0147  .0004  .0019.0091  
HR 1.041.07        
95% CI 1.01-1.071.04-1.10        
P .0039< .0001        
HR        1.08  
95% CI        1.02-1.13  
P        .0093  
HR 1.06   1.07     
95% CI 1.03-1.09   1.04-1.10     
P < .0001   < .0001     

In addition, we performed a sensitivity analysis in which we omitted the trials with low compliance to investigate their impact on the findings described above. Our sensitivity analysis excluded low-compliance trials (≤65%) in lung cancer (n = 2), ovarian cancer (n = 1), breast cancer (n = 1), and head and neck cancer (n = 2). With regard to the HRQOL parameters, for both ovarian and head and neck cancer, nausea and vomiting became insignificant. No changes were reported for the other cancer groups.


In this study, we investigated the significant prognostic relation between HRQOL and survival using a data set of 30 closed EORTC RCTs. On the basis of the results from heterogeneity tests and Kaplan-Meier curves, 11 different subsets were created, each representing 1 specific cancer site. Our results demonstrated that, for each cancer site, at least 1 HRQOL domain provided prognostic information that was additive over and above clinical and sociodemographic variables. However, the HRQOL parameters of greatest prognostic value differed across the cancer groups; and the effect size of each HRQOL parameter, indicated by the HR, depended on the tumor site.

Our study indicated that the EORTC QLQ-C30 parameter physical functioning was linked to survival in 5 of the 11 cancer sites: colorectal, lung, esophageal, breast, and melanoma. This confirms the findings of Gotay et al,[14] demonstrating that physical functioning was significant in 15 of the 36 studies. Six of the 30 RCTs from our study also were reviewed in the article by Gotay et al. Both in a colorectal cancer study[31] and in several recently published lung cancer studies,[27, 32, 33] the prognostic value of baseline physical functioning has been demonstrated. In addition, studies on esophageal cancer[12, 34, 35] reported a significant association between survival and baseline physical functioning. Bergquist et al[36] reported that both physical functioning and fatigue were predictive of survival in patients with advanced esophageal cancer.

The EORTC QLQ-C30 parameter nausea and vomiting appeared to be significant in 4 of the studied cancer sites: colorectal, ovarian, breast, and head and neck. In our sensitivity analysis, nausea and vomiting appeared to be linked with individual ovarian and head and neck trials. In 1 of the 2 ovarian studies, nausea and vomiting had prognostic value, possibly because patients had already received chemotherapy treatment before entering that specific study. By removing that study from our analysis, the results indicated that different HRQOL information was prognostic in pretreated patients versus patients who had no history of previous treatment. In the head and neck cancer group, the prognostic value of nausea and vomiting was attributed to 1 specific study, which included only postoperative patients. Removing the study from our analysis because of low compliance demonstrated that the HRQOL variable nausea and vomiting may have been prognostic because of surgical recovery or complications or possibly as a result of previous chemotherapy. That nausea and vomiting provided prognostic value for survival in breast cancer is attributed to 1 study in the pooled breast data set, which included only patients who had bone metastases. Previously published articles by Efficace et al[37] and Coates et al[38] did not report any relation between HRQOL variables and overall survival in patients with metastatic and nonmetastatic breast cancer. It is possible that the presence of bone metastases, or their treatment, led to this result. When sorting out the “vomiting” from the “nausea and vomiting” parameter, we may expect that not many patients are vomiting in the week before their therapy, especially those patients with early stage disease. In our study, we observed that mainly nausea was a significant indicator for survival.

The results in the brain cancer group are in line with a previously published study[39] in which the EORTC QLQ-C30 parameter cognitive functioning was identified as predictive for survival. Another study in patients with high-grade glioma[40] in which a different HRQOL measure was used also reported a significant prognostic relation between cognitive functioning and survival.

It is noteworthy that we reported lower emotional functioning in the breast and head and neck groups and lower social functioning in the esophageal group, and these lower scores were significant for longer survival. When taking a closer look at these groups, we noticed that the findings were driven by the patients with metastatic cancer in these groups. A similar conclusion for the emotional functioning scale was reported previously by Dancey et al[41] in a cohort of mixed cancer patients that included breast cancer and head and neck cancer. Although those authors came up with several explanations for this finding, a more detailed look into psychosocial factors and survival is encouraged to explain this reversed pattern.

The sociodemographic and clinical variables used in this study (age, sex, WHO performance status, and metastatic status) are well established indicators for survival estimation in cancer patients and, thus, were forced into the multivariate model whether or not they were significant at the univariate level. However, in combination with other prognostic sociodemographic, clinical, and HRQOL variables, the prognostic relevance of some of the clinical and sociodemographic variables became insignificant compared with the univariate results. This is most likely because of the unavoidable overlap between variables (ie, between WHO performance status and the EORTC QLQ-C30 physical functioning scale [colorectal, lung] or the global health status scale [prostate]) and demonstrates the decreasing influence of biomedical variables as prognostic factors in a multivariate setting in which both clinical and HRQOL data are assessed. These findings have also been reported in previous studies[42] and in an aggregated analysis of the current data.[16]

When forcing clinical and sociodemographic variables into the model, a further investigation into possible interactions with HRQOL variables should be investigated. For example, for testicular cancer, the HRQOL parameter role functioning is a significant predictor for survival. This may be related to the finding that the disease mainly strikes at a very young age, intermitting the role functioning of this young male cancer population. To be complete, a more detailed examination might reveal prognostic factors other than HRQOL that are worth investigating; for example, the pretreatment history of the patient.

Our study is not without limitations. Although we had a large sample for each cancer site, to achieve this, we needed to include a mixture of patients with differing clinical characteristics, such as first-line treatment[43] and disease stage. This variability set limits to the homogeneity within our subsets. This was demonstrated by our sensitivity analysis, in which some HRQOL parameters lost their prognostic value when trials with low compliance—representing mainly the trials with second-line treatment—were omitted from the analysis. Although heterogeneity of patient characteristics leads to less precise results with regard to the effect sizes, the large data set gives power and reliability to our findings. Currently, advanced statistical techniques allow us to account for this heterogeneity and to adjust for this variability in any meta-modeling.

In addition, focusing on specific studies with restrictive selection criteria might limit the generalizability of the findings and can lead to “over fitting.” However, more homogeneous groups make our findings more useful and clear for clinical decision making. A literature review recently performed by Lemieux et al[44] demonstrated that 7 out of 35 studies, looking only at advanced breast cancer HRQOL outcomes, were able to affect clinical decision making when only looking at a specific, homogeneous cancer population. Given this, we might question whether a clearer distinction between patients with early stage disease versus advanced disease could enforce the prognostic value of HRQOL parameters in 1 of the subgroups. Further research should be encouraged in this direction to determine whether HRQOL provides stronger prognostic value in certain clinical subgroups of the cancer population.

Relying solely on HRQOL information derived from clinical trial data may be too narrow, because trial participants in general are unrepresentative of the broader cancer population. For some of the cancer sites in our study, ie, melanoma and head and neck cancer, we had an under-representation of cancer patients with high WHO performance status. Ideally, large samples of similar patients derived in a standardized way from independent or observational studies should be used to validate and replicate our results. In addition, a comparison between different HRQOL measures could be analyzed to determine whether different HRQOL questionnaires applicable to oncology patients are able to identify the same prognostic HRQOL variables or a common set of prognostic variables. Also, the use of existing EORTC cancer-specific modules next to the core QLQ-C30 questionnaire might be investigated to increase the precision and overall picture of the symptom and functioning burden.

Because the prognostic value of clinical and sociodemographic variables are acknowledged and re-established in this study by their strong HRs, the HRQOL scales add proven, complementary prognostic value to standard clinical variables; they address different and additional aspects of a patient's well being, feelings, and functioning.[45] Sporadically,[46] HRQOL has even superseded the well established clinical variables. In addition, assessing HRQOL with these measures allows investigators to have a more comprehensive understanding of a cancer patient's burden with regard to the disease and the treatment envisaged. Patients who express a lower functioning level or higher symptom burden at study entry may benefit from a treatment regimen that is adapted to their status. Moreover, in our study, the observed effect of the prognostic parameter nausea and vomiting raises an interesting further question: do patients who are still suffering from the effects of their previous treatment respond differently to a proposed secondary treatment regimen than expected by their clinicians?

Although clinicians tend to view HRQOL as soft endpoint, there is a wealth of information that it can bring to clinicians; and, as such, HRQOL is gaining momentum. The finding that HRQOL is becoming a well established field, improving on its methodology[47] has been supported by governmental institutions like the National Cancer Institute,[48] most specifically in their recent PRO project.[49]


This work was sponsored by an unrestricted grant from the Pfizer Foundation, by the EORTC Charitable Trust, and by the National Cancer Institute. Divine E. Ediebah was funded by a Belgian Cancer Foundation Grant.


Martin Taphoorn has received compensation as a consultant to Hoffmann La Roche and has received grants from Jacobusstichting and the Devon Foundation. Kristin Bjordal has received reimbursement for travel/accommodations/meeting expenses from IGEA Clinical Biophysics for participation in the Second International User's Meeting for Electrochemotherapy (March 2013) and received a PhD fellowship from the Norwegian Cancer Society during the data-collection period of the current work.