Role functioning before start of adjuvant treatment was an independent prognostic factor for survival and time to failure. A report from the Nordic adjuvant interferon trial for patients with high-risk melanoma

Abstract

Purpose. To investigate the role of health-related quality of life (HRQoL) at randomization as independent prognostic factors for survival and time to failure, and to explore associations between HRQoL and treatment effects. Material and methods. In the Nordic adjuvant interferon trial, a randomized trial evaluating if adjuvant therapy with intermediate-dose IFN had the same beneficial effects on overall and disease-free survival in high-risk melanoma as high-dose IFN, 855 patients in Denmark, Finland, Norway, and Sweden were included. The EORTC QLQ-C30 questionnaire was used to assess HRQoL before randomization. Results. A total of 785 (92%) agreed to participate in the HRQoL-study and provided baseline HRQoL data. Prognostic variables included in the multivariate model were age, sex, performance status, tumor thickness, stage, and number of positive lymph nodes. Univariate analyses revealed an association between prolonged survival and age, stage/ number of metastatic lymph nodes and the HRQoL variable role functioning (p ≤ 0.01). After controlling for other prognostic factors, these variables remained independently statistically significant for survival. The univariate analyses of time to failure showed significant associations with the clinical variable stage/nodes and with the HRQoL variables physical functioning and role functioning. Adjusted multivariate analyses including the same clinical conditions as above showed statistically significant relationships between time to failure and global quality of life, physical functioning, role functioning, social functioning and fatigue (p ≤ 0.01). No interactions between HRQoL variables and treatment were found, with the exception for cognitive functioning. Conclusion. Role functioning was found to be an independent prognostic factor for time to failure and survival in patients with high-risk melanoma. Thus, also in this early stage of melanoma, HRQoL variables might be useful as important prognostic factors for time to failure and overall survival.

Health-related quality of life (HRQoL) has been investigated as an independent prognostic factor in cancer [1]. A recently published study of a large sample of cancer patients after radiotherapy revealed that among several psychosocial factors, HRQoL was the most important predictor for survival [2]. A review [3] including publications between 1982 and 2008 showed that early studies did not demonstrate associations between survival and psychosocial factors in melanoma patients [4,5]. Later studies, however, have indicated a relationship between HRQoL and survival. Low levels of global quality of life, bad mood and loss of appetite were predictive of shorter survival in a study of 152 patients with metastatic malignant melanoma [6]. In an Italian study of 140 patients with metastatic malignant melanoma, low levels of global quality of life and physical symptom distress were independent predictive variables for shorter survival [7]. A meta-analysis of HRQoL studies including a number of cancer diagnoses in connection with clinical trials within the European Organization for Research and Treatment of Cancer (EORTC) evaluated associations between HRQoL variables and survival [8]. The analysis included 2446 patients with malignant melanoma and showed that low levels of physical functioning, pain, and loss of appetite provided independent prognostic information. In summary, there are data suggesting independent prognostic value of some HRQoL variables in stage IV metastatic melanoma. A probable reason for the association between HRQoL and survival in metastatic cancer lies in the ability of patients to judge their own health status with better accuracy than some conventional prognostic factors [9–11]. Few studies in patients with localized cancer have shown associations between HRQoL and survival [12,13]. It has also been suggested that for early disease, clinical parameters override patients’ self-reported HRQoL in predicting survival [14].

No previous study has, to our knowledge, explored the associations between HRQoL and survival and time to failure in patients with stage IIB-C high-risk melanoma. These patients have a high risk of recurrence and death due to melanoma, although their melanoma is not clinically generalized at the time of diagnosis. There are currently a number of clinical trials investigating adjuvant treatment for these patients. It has been suggested that, if associations are found between HRQoL and survival and time to failure, these variables could be considered for stratification in clinical trials [1]. Thus, it was of interest to investigate whether HRQoL is associated with survival and time to failure in this patient population. The hypothesis, based on previous studies, was that higher levels of HRQoL would be associated with longer survival and time to failure.

Between 1996 and 2004, high-risk melanoma patients in Denmark, Finland, Norway, and Sweden were included in “The Nordic IFN trial” (Clinical Trials gov. #NCT01259934), a randomized trial evaluating whether adjuvant therapy with intermediate-dose IFN have effects on overall (OS) and relapse-free survival (RFS) in high-risk melanoma, and to evaluate if two years of treatment was more efficient than one year [15]. Intermediate-dose IFN in high-risk melanoma patients significantly improved RFS without significantly increasing overall survival, but prolonged maintenance of treatment from one to two years did not seem to improve the outcome [15]. HRQoL was prospectively investigated, demonstrating a negative impact of IFN treatment [16]. However, the impact was reversible when treatment was stopped.

The aim of the present study was to explore HRQoL variables at randomization as prognostic factors for survival and time to failure, independent of other prognostic variables, including: sex, age, performance status, tumor thickness, stage, and number of positive lymph nodes, in patients with high-risk melanoma included in the “The Nordic IFN trial”.

In addition, interactions between HRQoL variables and treatment were explored. The study was approved by the appropriate local research ethics committees and undertaken according to the principles of the Declaration of Helsinki.

Material and methods

Patients

In this trial, 35 centers in the Nordic countries participated: two in Denmark, five in Finland, six in Norway, and 22 in Sweden [15]. Patient inclusion criteria were: histologically verified resected cutaneous melanoma, AJCC stage IIB–IIC (T4N0M0), or stage III (TxN1–3M0); age 18 years or older; ECOG performance status 0–1; normal bone marrow function; adequate liver chemistry and renal function. Patients were staged by clinical examination, blood chemistry, chest x-ray and ultrasound of liver, according to state-of-the-art at the time of study design. Excluded from the trial were patients with non-cutaneous melanoma, with melanoma with unknown primary site, with incompletely resected melanoma, with a second cancer diagnosis (except for basal cell and squamous cell skin cancer or in situ cervical carcinoma), with ongoing systemic corticosteroid medication, female patients who were pregnant or lactating and patients with a history of or ongoing depressive disease.

The patients were randomized to one of three arms. Randomization was performed after written informed consent at the Clinical Trial Unit, Department of Oncology, Karolinska University Hospital, Stockholm, Sweden. The patients were stratified according to 1) country and 2) tumor stage at randomization. Arm A consisted of observation only. Patients in Arm B received one year of treatment: Induction: IFN, 10 MU (flat dose), sc, 5 days/week, 4 weeks, maintenance: IFN, 10 MU (flat dose), sc, 3 days/week for 12 months (n = 285). Arm C same induction regimen as Arm B however maintenance therapy included IFN, 10 MU (flat dose), sc, 3 days/week for 24 months (n = 286). Adjuvant intermediate-dose IFN was delivered with moderate and expected toxicities in both treatment arms [15].

Point of assessment

The baseline questionnaire was handed to the patient after informed consent, to be completed before information about the result of randomization was given, and thus before IFN-treatment started, but after surgery was completed. The questionnaire was returned to the HRQoL study coordinator in each country in a prepaid envelope by the patient. The follow-up questionnaire was sent one month after randomization together with a prepaid return envelope from the Clinical Trial Unit, Department of Oncology, Karolinska University Hospital, Stockholm.

Instruments

The European Organization for Research and Treatment of Cancer Core Questionnaire (EORTC QLQ-C30) Version 2, developed by the EORTC Quality of Life Group for assessment of HRQoL in cancer trials, was used [17]. It consists of 30 items that comprise five functional subscales (physical, role, emotional, cognitive, and social functioning), three multi-item symptom scales (nausea/vomiting, fatigue, pain), and six single items assessing symptoms (insomnia, dyspnea, loss of appetite, constipation, diarrhea, and financial difficulties) [18]. Most items are scored in four categories from “Not at all” to “Very much”. In EORTC QLQ-C30 Version 2, however, items included in the physical and fatigue subscales are responded to in categories “Yes” and “No”. The global quality-of-life subscale consists of two items, each scored in seven categories. The following HRQoL variables were chosen for the analysis: physical-, role-, emotional-, cognitive-, and social functioning, global quality of life, fatigue, pain, appetite loss, dyspnea and sleep disturbances. These were found to be the most important variables for predicting survival times in cancer patients after adjusting for other prognostic variables in the meta-analysis of HRQoL data in EORTC trials [8].

Clinical data

Clinical data were collected from the study database at the Clinical Trial Unit, Karolinska University Hospital, Stockholm.

Statistical methods

The EORTC QLQ-C30 items were scored and scaled according to scoring manual [18]. Raw scores were transformed to a linear scale ranging from 0 to 100. Higher scores indicate better functioning on the functioning subscales and global quality-of-life scale, and more symptoms on the symptom scales and single items. No substitution for missing items was made.

Survival time was calculated from the date of randomization to the date of death or the end of the follow-up period. Time to failure was calculated from the date of randomization to the earliest of the date of recurrence or date of death. Patients were censored at the end of follow-up. In these analyses, both recurrence and death are considered events. The median follow-up time was estimated using the reverse Kaplan-Meier method [19].

The effect of HRQoL variables on time to event was studied using proportional hazards regression. Results from these models are presented as hazard ratios (HR) together with 99% confidence intervals. Analyses of interactions between HRQoL variables and treatment were performed by including product terms into the regression models.

A stepwise variable selection procedure using backward elimination was used to identify the strongest independent variables for survival and time to failure. Only clinical or HRQoL variables with significance levels ≤ 0.01 remained in the final model. A bootstrap re-sampling procedure was used to assess the replication stability of the stepwise model [20]. This technique generates a number of samples (1000), each of the same size as the original data, by random sampling with replacement from the original dataset. The frequencies of inclusion are indicative of the prognostic importance of the variables.

Correlations between the HRQoL variables were calculated by Spearman's correlation coefficient.

In all the statistical modeling, the HRQoL scales were included as continuous variables and the reported effect on survival or time to failure corresponded to a 10-unit scale shift. A shift of this magnitude is considered to be of moderate clinical importance [21].

Due to multiple testing, the level of statistical significance was set to 0.01 to avoid type I errors.

Results

A total of 855 patients were included between 1996 and 2004 [15]. Patient characteristics have been described in detail previously [16]. Of the 855 randomized patients, 785 (92%) agreed to participate in the quality-of-life study and provided baseline HRQoL data. During the follow-up period, 400 deaths and 479 relapse-related events (including deaths as first events) were observed among the patients who participated in the HRQoL study. Clinical characteristics among patients with information at baseline for any of the studied HRQoL variables are listed in Table I. Follow-up and number of events among patients included in the HRQoL study are presented in Table II. Table III shows observed baseline mean values and standard deviations for the HRQoL variables, and the expected mean values.

Table I. Clinical characteristics of 785 patients with information at baseline on any of the studied HRQoL variables.

Patient characteristics	N (%)
Allocated study arm:
A (control)	251 (32)
B (1 year treatment)	264 (34)
C (2 years treatment)	270 (34)
Age at randomization, mean (SD):	52 (18–77)
Sex:
Males	483 (62)
Females	302 (38)
AJJC Stage and positive nodes:
II	149 (19%)
IIIN1	343 (44)
IIIN2–3	187 (24)
IIIN4+	106 (14)
Tumor thickness^1, mean (SD):	3.6 (3.9)

¹For 45 patients, data on tumor thickness is not available.

Table II. Follow-up and number of events among patients included in the HRQoL study.

Characteristics	N (%)
The interferon trial
Recruitment to trial:	1996-11-11 to 2004-09-01
Total number of patients recruited:	855
Date for end of follow-up in the present analyses^1:	2009-10-15
Median follow-up time^2:	7.2 years
The HRQoL studya
Patients included in the study	785
Total number of deaths during follow-up:	400
Total number of events during follow-up:	479

^aNumber of patients with information on any of the studied HRQoL scales at baseline; ¹Last date for visit among patients still alive; ²Calculated using the reversed Kaplan-Meier technique.

Table III. Mean scores and standard deviations (SD) measured at baseline and expected values based on sex- and age-specific values from the normal Swedish population.

Variables	Number of patients	Mean (SD)	Expected
Global Quality of Life^1	771	74 (20)	77
Physical functioning^1	769	88 (20)	91
Role functioning^1	777	67 (32)	86
Emotional functioning^1	769	78 (20)	81
Cognitive functioning^1	771	91 (14)	89
Social functioning^1	765	83 (23)	90
Fatigue^2	768	21 (22)	21
Pain^2	765	10 (12)	19
Dyspnea^2	780	8 (17)	16
Appetite loss^2	783	6 (15)	4
Sleep disturbances^2	782	19 (26)	18

¹High scores indicate high levels of functioning and global quality of life; ²High scores indicate high levels of problems and symptoms.

Survival (Table IV)

In the univariate analyses, the clinical variables age and stage/number of metastatic nodes and the HRQoL variable role functioning were associated with prolonged survival (p ≤ 0.01). In the adjusted multivariate analyses, including the clinical variables age, gender, stage/nodal status (II, IIIN1, IIIN2-3, IIIN4+), and treatment arm (B+ C vs. A), role functioning remained statistically significant. The same pattern remained in the stepwise backward analysis, and was also confirmed in the bootstrap analysis.

Table IV. Univariate and multivariate prognostic factor analysis for survival.

	Unadjusted		Adjusted^a		Stepwise^b [backward, p = 0.01]		Bootstrap^c
Variable and coding	HR (99% CI)	p	HR (99% CI)	p	HR (99% CI)	p	Inclusion %
Clinical variables:
Age at randomization, years	1.15 (1.03–1.28)	0.002	1.15 (1.03–1.30)	0.002	1.20 (1.06–1.36)	< 0.001	83.9
Sex (Males vs. Females)	1.13 (0.87–1.48)	0.23	1.14 (0.86–1.50)	0.23			27.1
Stage/nodes (II, IIIN1, IIIN2–3, IIIN4+)	1.35 (1.17–1.55)	< 0.001	1.37 (1.19–1.57)	< 0.001	1.29 (1.12–1.50)	< 0.001	97.7
Thickness (Breslow), mm	1.17 (0.86–1.61)	0.19	1.30 (0.97–1.73)	0.020			19.0
Allocated treatment (B+ C vs. A)	0.86 (0.66–1.12)	0.16	0.85 (0.65–1.13)	0.14			9.5
HRQoL variables:
Global quality of life^d	0.97 (0.91–1.04)	0.27	0.96 (0.90–1.03)	0.14			4.0
Physical functioning^d	0.95 (0.90–1.01)	0.043	0.95 (0.89–1.02)	0.061			1.5
Role functioning^d	0.95 (0.91–0.99)	0.001	0.94 (0.90–0.98)	< 0.001	0.94 (0.90–0.99)	< 0.001	79.0
Emotional functioning^d	0.98 (0.92–1.05)	0.52	0.97 (0.91–1.04)	0.30			1.8
Cognitive functioning^d	0.98 (0.90–1.07)	0.62	0.98 (0.90–1.07)	0.55			4.1
Social functioning^d	0.96 (0.91–1.01)	0.039	0.95 (0.89–1.0)	0.016			2.5
Fatigue^e	1.03 (0.97–1.09)	0.18	1.05 (0.98–1.11)	0.071			5.9
Pain^e	1.08 (0.97–1.20)	0.064	1.08 (0.97–1.21)	0.060			2.2
Dyspnea^e	1.04 (0.97–1.11)	0.20	1.04 (0.97–1.12)	0.19			2.0
Appetite loss^e	1.03 (0.95–1.12)	0.37	1.06 (0.98–1.16)	0.069			3.3
Sleep disturbances^e	1.01 (0.96–1.06)	0.81	1.01 (0.96–1.06)	0.74			5.4

Note: HRs for age thickness and all HRQoL variables refers to a 10-unit increase. CI, confidence interval; HR, hazard ratio; HRQoL, health related quality of life. ^aEach of the clinical variables are adjusted for all the other clinical variables but ignoring the HRQoL variables. Each of the HRQoL variables are adjusted for all the clinical variables but ignoring the other HRQoL variables (i.e. the effect of the studied HRQoL variable on top of all the clinical variables); ^bStepwise model with backward-selection search. All the variables with a significance level < 0.01 remain in the final model. Estimation is based on observations with non-missing values for all specified variables; ^cTo assess the replication stability of the Stepwise model a bootstrap resampling procedure proposed by Sauerbrei et al. was used. This technique generates a number of samples (1000) each of the same size as the original data, by random sampling with replacement from the original data set. The frequency of inclusion is indicative for the prognostic importance of the factors; ^dHigh values indicate high level of functioning and global quality of life; ^eHigh values indicate high levels of problems and symptoms.

Time to failure (Table V)

The univariate analyses of time to failure showed significant associations with the clinical variable stage/nodes and with the HRQoL variables physical functioning and role functioning (p ≤ 0.01). Adjusted multivariate analyses including the same clinical conditions as above revealed statistically significant relationships between time to failure and global quality of life, physical-, role-, social functioning and fatigue (p ≤ 0.01). In the stepwise backward analysis, age, stage/nodes and the HRQoL variable role functioning were the strongest showing the strongest association with time to failure. Younger age and lower stage/nodes and higher role functioning were associated with longer time to failure.

Table V. Univariate and multivariate prognostic factor analysis for time to recurrence.

	Unadjusted		Adjusted^a		Stepwise^b [backward, p = 0.01]		Bootstrap^c
Variable and coding	HR (99% CI)	p	HR (99% CI)	p	HR (99% CI)	p	Inclusion %
Clinical variables:
Age at randomization, years	1.09 (0.98–1.21)	0.032	1.09 (0.98–1.21)	0.040	1.14 (1.02–1.28)	0.003	55.7
Sex (Males vs. Females)	1.10 (0.86–1.41)	0.31	1.10 (0.85–1.42)	0.33			33.3
Stage/nodes (II, IIIN1, IIIN2–3, IIIN4+)	1.27 (1.11–1.44)	< 0.001	1.28 (1.12–1.46)	< 0.001	1.23 (1.07–1.40)	< 0.001	90.2
Thickness (Breslow), mm	1.17 (0.89–1.53)	0.14	1.26 (0.98–1.62)	0.018			28.0
Allocated treatment (B+ C vs. A)	0.80 (0.62–1.03)	0.020	0.79 (0.61–1.02)	0.016			37.0
HRQoL variables:
Global Health^d	0.96 (0.90–1.01)	0.047	0.94 (0.88–1.00)	0.008			2.4
Physical functioning^d	0.94 (0.89–0.99)	0.003	0.93 (0.87–0.99)	0.002			7.8
Role functioning^d	0.94 (0.91–0.98)	< 0.001	0.94 (0.90–0.97)	< 0.001	0.94 (0.90–0.98)	< 0.001	84.9
Emotional functioning^d	0.97 (0.92–1.03)	0.19	0.96 (0.90–1.02)	0.068			2.6
Cognitive functioning^d	0.99 (0.92–1.08)	0.83	0.99 (0.91–1.08)	0.73			12.1
Social functioning^d	0.96 (0.91–1.00)	0.017	0.94 (0.89–0.99)	0.004			2.0
Fatigue^e	1.04 (0.99–1.10)	0.046	1.06 (1.00–1.12)	0.010			5.2
Pain^e	1.08 (0.99–1.19)	0.031	1.10 (1.00–1.22)	0.014			3.9
Dyspnea^e	1.06 (1.00–1.13)	0.017	1.06 (1.00–1.13)	0.016			7.8
Appetite loss^e	1.05 (0.97–1.13)	0.10	1.07 (1.00–1.16)	0.015			2.4
Sleep disturbances^e	1.02 (0.97–1.06)	0.38	1.02 (0.97–1.07)	0.27			1.8

Note: HRs for age thickness and all HRQoL variables refers to a 10-unit increase. CI, confidence interval; HR, hazard ratio; HRQoL, health related quality of life. ^aEach of the clinical variables are adjusted for all the other clinical variables but ignoring the HRQoL variables. Each of the HRQoL variables are adjusted for all the clinical variables but ignoring the other HRQoL variables (i.e. the effect of the studied HRQoL variable on top of all the clinical variables). ^bStepwise model with backward-selection search. All the variables with a significance level < 0.01 remain in the final model. Estimation is based on observations with non-missing values for all specified variables; ^cTo assess the replication stability of the Stepwise model a bootstrap resampling procedure proposed by Sauerbrei et al. was used. This technique generates a number of samples (1000) each of the same size as the original data, by random sampling with replacement from the original data set. The frequency of inclusion is indicative for the prognostic importance of the factors; ^dHigh values indicate high level of functioning and global quality of life; ^eHigh values indicate high levels of problems and symptoms.

Unadjusted analysis: N = 740–785. Adjusted analysis: N = 719–740. Stepwise and Bootstap: N = 678.

Correlations between role functioning and the other HRQoL variables examined

Moderate correlations were found between role functioning and global quality of life (0.59), physical functioning (0.55), social functioning (0.59), fatigue (0.61) and pain (0.67).

In summary, the clinical variables age and stage/number of metastatic nodes together with the HRQoL variable role functioning showed the most consistent association with time to failure and survival.

Treatment-HRQoL interactions

Interactions between HRQoL variables and treatment were examined, showing no statistically significant effects, with one exception. Cognitive functioning (p < 0.01) appeared to be associated with larger effects among those with lower mean scores, indicating that the lower cognitive functioning at baseline, the larger benefit from treatment (Figure 1).

Figure 1. Interaction between treatment and cognitive functioning in the survival analysis.

Discussion

The aim of the present study was to investigate the role of HRQoL variables at randomization as independent prognostic factors for time to failure and survival in patients with high-risk melanoma. At randomization, the patients had undergone extended surgery and been informed about their high risk of recurrence. Role functioning appeared to be strongly related to prolonged survival and time to failure, independent of the clinical variables.

Role functioning has not been demonstrated as a prognostic factor in melanoma in earlier studies [6–8]. The role functioning subscale consists of two items: 1) “Were you limited in doing either your work or other daily activities?”; and 2) “Were you limited in pursuing your hobbies or other leisure time activities?” [18]. Gotay et al. [1] suggest four different explanations for the association between HRQoL variables and survival in cancer outcome studies. One explanation is that HRQoL measures include different items and therefore provide more sensitive information than known prognostic factors. Thus, HRQoL mirrors the general condition of the patient. Inability to pursue work, daily activities and leisure time activities might be indicative of the seriousness of the disease. The variables in the EORTC QLQ-C30 have been shown to correlate with each other, the strongest correlations observed between the role functioning and physical functioning and fatigue subscales (ranging from 0.54 to 0.63) [18]. Also in the present study, moderate correlations were found between role functioning and global quality of life, social functioning and pain (data not shown). Thus, one explanation for our findings might be that role functioning mirrors the extent of the disease, since metastatic screening only involved clinical examination, blood chemistry, chest x-ray and ultrasound of the abdomen. It is, thus, conceivable, that HRQoL data such as role functioning might give additional prognostic information related to disease status even in patients with EGOG 0-1 performance status. This would imply that patients are better judges of their disease than their doctors who rely on clinical laboratory and radiological staging. On the other hand, this interpretation is not entirely plausible, since minimal disease not detectable with the above methods, could significantly affect role functioning. A study in patients staged according to current stage of the art, including sentinel node biopsy and CT or CT/PET scan would be of interest, to confirm our findings, indicating whether role functioning is an independent prognostic factor even under more detailed screening procedures.

Another explanation is that HRQoL data, especially those collected at baseline, pick up relevant information earlier than established clinical prognostic factors. However, in our study, stage/nodal status and age also turned out as prognostic factors, in addition to role functioning.

A third suggested explanation is that HRQoL data are markers of patient behavior, i.e. compliance with treatment. In the present study, however, only two thirds of the patients were allocated to IFN-treatment. As no differences in survival were found between the randomization arms including the control arm, this seems to be an unlikely explanation. Role functioning could be expected to affect time to failure in the other direction. If the patients did not adhere to medical follow-up visits due to lack of role functioning, their recorded date of recurrence might be delayed. The hypothesis, that compliance resulting in better treatment effect could be associated with high levels of HRQoL, was tested in analyses of interactions between HRQoL and treatment. It was hypothesized that higher levels of HRQoL would be associated with treatment effects, indicative of compliance. However, we did not find any association between role functioning and treatment, nor between the other HRQoL variables, with one exception, cognitive functioning. Those with the highest levels of cognitive functioning did not seem to benefit from treatment, whereas the lower the cognitive functioning the better treatment effect. This is in concordance with the findings in from the clinical study, where the major effect seemed to be in patients with the highest tumor burden before lymph-node dissection [15].

The fourth suggested explanation is that HRQoL data are markers of individual characteristics such as personality style and adapting coping strategies, which affect the disease process and outcomes in cancer patients. There is, however, only vague empirical evidence for such an explanation in melanoma [5,12,22–24]. Even if there were evidence of an association between personality and survival, it still would not explain why or reveal anything about the underlying process.

Since multiple analyses were performed in the present study, we chose to use the 1% level of statistical significance. A number of HRQoL variables were, however, associated with time to failure (global health quality of life, physical-, role-, social functioning, pain, dyspnea, appetite loss) on a 5% significant level in the adjusted multivariate analyses. Similarly, survival and social functioning showed an association (p = 0.02). Taking these findings into consideration, our results indicate that these HRQoL variables also might be important predictors of time to progression and survival.

In a study of 1588 breast Stage I cancer patients, emotional functioning predicted overall survival and fatigue was an independent predictor of recurrence-free survival [13]. In a Swedish study of the prognostic value of anxiety and depression at diagnosis in 437 Stage I melanoma patients, no associations with survival were found [25]. In the present study, no associations between emotional functioning and survival and time to progression were found. Looking at the baseline mean scores for emotional functioning, the figure for the patients is similar to the expected figure from age and gender adjusted normative data [16]. This indicates that most patients do not have major emotional problems at the time of inclusion in the study, which is surprising, considering the fact that they were informed about their high risk of recurrence at inclusion in the clinical trial. A Swedish study of 64 patients with Stage I melanoma, showed, however, higher levels of anxiety in patients with thin melanomas (< 0.8mm) as compared to those with thicker melanomas (≥ 0.8mm) [26]. A suggested explanation for this finding is that those who are more worried seek medical advice earlier, resulting in thinner tumors. Another explanation for the relatively high level of emotional functioning is that the patients in the study felt that they were well cared for and had confidence in their physicians. It should be kept in mind that the patients responded to the questionnaires before being informed about to which randomization arm they were allocated. Further studies are warranted in order to confirm our findings and also to explore why role functioning seems to be such an important prognostic factor.

The strengths of the present study are the large number of patients, that a validated questionnaire was used and the high response rate. In addition, the follow-up time was relatively long and included a large number of events. The data collection was performed within the frame of a clinical trial, which ensures control of follow-up data and clinical variables. This report also has some limitations. Data on ulceration was missing for 20% of the patients. Thus, this variable was not included in the analyses. It is unlikely, however, that ulceration of the primary tumor would affect role and social functioning after surgery. In addition, the study was designed as a randomized trial and the present analyses were not planned before the start of the study.

The results of the present study have clinical implications. Considering the finding that role functioning is an independent prognostic factor for time to failure and relapse, it might be important to identify patients who have low levels of role functioning at the time of diagnosis. They could be offered behavioral interventions designed to increase role functioning, e.g. behavioral activation.

Conclusion

Role functioning in high-risk melanoma was strongly related to prolonged survival and time to failure, independent of the clinical variables. This association has not been shown previously in this stage of the disease, and should be taken into account in the care of patients with high-risk melanoma. Further studies in patients staged according to current state-of-the-art are needed in order to confirm the results and to understand why role functioning seems important for survival and time to failure in this high-risk group of melanoma patients.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.