Introduction

Esophageal cancer is the 8th most common cancer in the world with a global incidence of 6.7 per 100,000 and a generally poor prognosis [1, 2]. There are two major subtypes, squamous cell carcinoma (ESCC) and adenocarcinoma (EAC). Overall 5-year survival rates for populations developing these cancers are reported to be 17–20% [3]. The majority of patients present with vague symptoms including difficulty in swallowing and weight loss, with diagnosis obtained using endoscopy and biopsy. Treatment for EAC depends on TNM staging (T = tumour growth through tissue, N = nodal involvement and M = metastatic involvement). Uni-modality treatment involves surgery or endoscopic resection, dual modality involves chemotherapy and radiation while tri-modality involves surgery, chemotherapy and radiation. Uni-modality treatment is preferred for the earliest stages of cancer. Surgery for EAC or esophagectomy is an invasive operation involving access to thorax and abdomen. Parts of the esophagus and stomach are resected along with adjacent lymph nodes followed by anastomosis of the reshaped stomach to the remaining esophagus [4]. Overall, treatment for esophageal cancer is expensive, resource intensive, and carries a high morbidity and quality of life detriments [5, 6]. It is important to understand the QoL experienced by patients undergoing treatment to guide tailoring of interventions and patient information but also to inform health funding decisions.

A plethora of instruments exist for the assessment of quality-of-life outcomes, including both generic and condition-specific measures, which can be preference based, also refered to as mulit-attribute utility instruments (MAUIs) or non-preference based. Preference-based instruments facilitate the calculation of quality-adjusted life years (QALYs) and are therefore suitable for application in cost-utility analysis (CUA). Decision making bodies including the National Institute for Health and Care Excellence (NICE) in the United Kingdom, and the Pharmaceutical Benefits Advisory Committee (PBAC) and Medical Services Advisory Committee (MSAC) in Australia, require the use of generic MAUIs such as the EuroQol 5-dimensions EQ-5D (EQ-5D-3L and EQ-5D-5L) in submissions of economic evaluations of cost effectiveness evidence [7, 8]. However, it has been argued that generic measures are not as sensitive to changes in quality of life, particularly in cancer, and thus may not adequately reflect quality of life detriments or gains in these conditions which would ultimately affect any QALY estimations undertaken [9, 10]. This has led to the development of cancer-specific MAUIs such as the European Organization for Research and Treatment of Cancer Eight dimensions (EORTC-8D) and EORTC Quality of Life Utility Measure-Core 10 dimensions (QLU-C10D), which are both derived from the non-preference-based EORTC Core Quality of Life Questionnaire (EORTC QLQ-C30) [11, 12].

The EORTC-8D has eight dimensions obtained from 10 of the 30 items of QLQ-C30. It has four functional scales including physical functioning, role functioning, emotional functioning, and social functioning, and four symptom scales of pain, fatigue and sleep disturbances, nausea and lastly constipation and diarrhoea. This instrument generates a total of 81,920 health states [12]. Health states were valued by a general population sample in the United Kingdom using the time trade-off valuation method [13]. The newer MAUI, QLU-C10D, is an extension of the work done in developing the EORTC-8D [11, 14]. It has four functional scales and six symptom scales obtained from 13 out of the 30 items of QLQ-C30 [11]. The functional scales are similar but the symptom scales also include appetite and bowel problems instead of constipation and diarrhoea. QLU-C10D generates 1,048,576 health states and a value set generated from an Australian general population sample using discrete choice experiments methodology [14]. Although both measures are obtained from QLQ-C30, they are different in some respects. Emotional functioning, physical functioning, fatigue, nausea and bowel problems are obtained from the same QLQ-C30 items for both measures; however, physical functioning has four levels in QLU-C10D and five levels in EORTC-8D. Role functioning, social functioning and pain are obtained from different items for both measures and QLU-C10D has two more dimensions, sleep and appetite which are not present in the EORTC-8D [11].

Given both preference-based generic and cancer-specific MAUIs are available to measure and value health related quality of life (HRQoL) in cancer patient populations, it is important to understand the relative psychometric performance of these instruments in clinical trials. Two studies that compared the generic EQ-5D-3L and EORTC-8D found that although both measures had similar discriminatory power, the calculated QALYs based on EQ-5D-3L were significantly lower than those obtained using EORTC-8D [9, 15]. This inconsistency could lead to substantial opposite conclusions as to whether an intervention should be regarded as cost-effective or not, and this is critical for resource intensive and high morbidity interventions like esophagectomy.

Unlike the EORTC-8D, there is a lack of longitudinal evidence on the comparisons of psychometric properties between the newly developed QLU-C10D and EQ-5D-3L or any generic MAUIs. This paper aimed to contribute to the literature by assessing the responsiveness and convergent validity of the newly developed cancer-specific QLU-C10D relative to the generic EQ-5D-3L in the context of short-term quality of life/utilities after esophagectomy for esophageal cancer.

Methods

Sample

This analysis was undertaken from a pooled sample of patients participating in a randomised control trial, the details of which are reported elsewhere [16]. Briefly, patients were randomised into four groups to receive an immunonutrition supplement or standard nutrition without added immunonutrients; (1) before but not after; (2) after but not before; (3) both before and after; and (4) none before or after esophagectomy. Quality of life was assessed seven days before and 42 days after esophagectomy. As briefly stated in an earlier publication there were no significant differences in quality of life or clinical outcomes between the groups before and after esophagectomy [16]. In this paper data was analysed as a pooled sample including all patients with both baseline (pre-operative) and follow-up (post-operative) quality of life scores on the EORTC QLQ-C30 and EQ-5D-3L measures. Patients were excluded from this analysis if they had any missing value of these two quality of life measures at any timepoint.

Quality of life assessment

Quality of life was assessed using the self-administered and widely validated generic utility-based EQ-5D-3L [17] and the cancer specific EORTC-QLQ-C30 [18]. The EQ-5D-3L descriptive system comprises five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression with 3 levels for each dimension: no problems, some problems, and extreme problems. The EQ-5D-3L was scored using an Australian-specific tariff [19]. The QLQ-C30 has one global HRQoL scale, five functional scales (physical, role, emotional, cognitive, social), three symptom scales (fatigue, nausea or vomiting, pain) and six single items (sleeping disorders, appetite loss, dyspnoea, diarrhoea, constipation, and financial problems). Each item has four alternative responses (1- not at all; 2- a little; 3-quite a bit; 4-very much). Responses to the QLQ-C30 were mapped onto the cancer specific preference based QLU-C10D to generate utility scores using an Australian-specific scoring algorithm developed by Norman et al. [14] for this purpose. Research has showed that country-specific value sets, where available, are preferable when evaluating interventions in that country/region [20, 21]. As such, although both EORTC-8D and QLU-C10D utilities can be derived from the QLQ-C30, QLU-C10D was preferable in this context because it’s health states have been valued by an Australian general population sample.

Data analysis

Data was analysed using Stata (StataCorp, College Station, TX, USA) [22]. Normally distributed data was analysed with one-way analysis of variance (ANOVA), and Kruskal–Wallis H test was used for analysis of non-normally distributed data. A p-value < 0.05 was considered statistically significant.

Mean difference

Utility scores for EQ-5D-3L [19] and QLU-C10D [14] were generated based on Australian general population scoring algorithms pertaining to each instrument. Basic descriptive statistics including means, medians and ranges were compared for each instrument at baseline and follow-up. The clinically important mean difference (MCID) for EQ-5D-3L when used in populations with cancer varies from 0.07—0.12 [23]. A change of > 10 with the EORTC QLQ-C30 is considered clinically relevant and > 20 as strongly relevant [24].

Responsiveness

To assess responsiveness (i.e. the ability of an instrument to detect changes in response to esophagectomy in this study), two statistical tests were applied, including the Standardized Response Mean (SRM) and Effect Size (ES) [25]. They are calculated as ES = ratio of the mean change to the standard deviation of scores at baseline statistic and the SRM = ratio of the mean change to the standard deviation of that change. For ES scores, the recommended minimum effect size = 0.41, moderate effect = 1.15 and strong effect = 2.70 [26]. SRM scores of < 0.20 = trivial effect, 0.20– < 0.50 = small effect, 0.50–0.80 = moderate effect, > 0.80 = large effect [27].

Ceiling effects

Ceiling effect is a measure of how accurately an instrument measures the intended domain by considering the proportion of respondents who achieve the highest level of the domains or the highest score of the instrument [28]. Ceiling effects are present ‘if more than 15–20% of respondents achieved the best possible score’ [29, 30]. Ceiling effects were calculated at both baseline and follow-up. The ceiling effect for EQ-5D-3L was calculated as the proportion of ‘no problem’ responses on each dimension and the proportion of ‘no problem’ in all dimensions. Similarly, for the QLU-C10D the ceiling effect was calculated as the proportion of level 1 (no trouble/limitation) on each dimension as well as on all dimensions. Ceiling effects were further explored by selecting those reporting full health in one instrument to see what they report in the other instrument. Lower ceiling effects suggest greater discriminant ability.

Convergent validity

Convergent validity was explored between dimensions measuring similar constructs on both measures such as mobility and physical function, pain/discomfort and pain [31, 32]. Correlations were classified as very weak (r = 0–0.2), weak (r = 0.2–0.4), moderate (r = 0.4–0.7), strong (r = 0.7–0.9) or very strong (r = 0.9–1.0) [27].

Agreement

The limits of agreement between the instruments were explored using Bland–Altman plots [33, 34]. Good agreement was demonstrated by less than 5% of points being outside of the limits of agreement (LOA).

Results

Demographics

164 of the original cohort of 276 patients completed both quality of life questionnaires before and after surgery and were included in the analysis presented in this study. 112 patients were excluded as they only had one set of quality of life data. There were no significant differences in demographic characteristics between patients included (n = 164) in this analysis vs. those excluded (n = 112) for all but two variables. More patients in the excluded group had TNM stage IIIc/4 (14% compared to 4%) and underwent a thoracoscopic esophagectomy (66% compared to 54%)—see Additional file 1: Table A1. However, these differences did not translate into differences in HRQoL outcomes, see Additional file 1: Table A2.

Table 1 Demographic and clinical characteristics
Full size table
Table 2 Descriptive statistics including mean difference and effect size
Full size table

Table 1 summarises the demographic characteristics of the study sample. Participants were predominantly male (81%), mean age was 63 years, with a history of alcohol consumption (66%). The commonest surgical technique (54%) was a “hybrid” esophagectomy entailing and open abdominal phase, thoracoscopic chest phase and anastomosis in the left neck. Most patients had preoperative chemotherapy (79%), and length of the hospital stay was more than 10 days (81%) for most patients.

HRQoL mean difference and effect size

HRQoL reduced between baseline (7 days before) and follow-up (42 days) after esophagectomy on all measures, 0.85–0.69 on EQ-5D-3L, 0.81–0.52 with QLU-C10D, and 84.3 to 62.5 with QLQ-C30 and these changes were statistically significant (Table 2). The mean score differences for EQ-5D-3L and QLQ-C30 exceeded their respective MCID at 0.16 and 21.6 respectively. The mean difference for QLU-C10D was 0.29 but the MCID for this measure has not yet been established. As such, the MCID for QLQ-C30 was reported and used as a reference. Moderate effect was detected for all measures using ES (> 1.15 but < 2.70) and strong effect using SRM (> 0.80). QLU-C10D was a more responsive measure compared to EQ-5D-3L with a larger ES (1.53 compared to 1.08) and SRM (1.37 compared to 0.84).

Ceiling effects

Distribution of the scores was similar for both measures at baseline and follow-up. However, a clustering of EQ-5D-3L indices at the upper level with a gap between 1 and the lower levels was observed at baseline. Ceiling effects (> 15%) were observed on the EQ-5D-3L for all dimensions at baseline and similarly after esophagectomy except for the usual activities dimension (Fig. 1). The self-care dimension showed the greatest ceiling effect at both baseline (99%) and follow-up (86%). Usual activities and pain showed the greatest (> 50%) reduction in ceiling effects between baseline and follow-up.

Fig 1:
figure 1

EQ-5D-3L ceiling effect at baseline and follow-up

Full size image

For QLU-C10D ceiling effects were observed for all dimensions pre-operatively but only with emotional function (53%), physical function (16%), nausea (35%), sleep (31%), bowel problems (21%) and pain (20%) following esophagectomy (Fig. 2). Nausea (70%) and emotional function (53%) had the highest ceiling effects at baseline and follow-up respectively. Role function and appetite showed the greatest (> 50%) reduction in ceiling effect between baseline and follow-up.

Fig 2:
figure 2

QLU-C10D ceiling effects at baseline and follow-up

Full size image

At baseline 13 (8%) respondents reported full health on both measures, while 55 (34%) reported full health on EQ-5D-3L but not QLU-C10D. Nine respondents (5.5%) reported full health on EQ-5D-3L but none reported full health on QLU-C10D at follow-up. Table 3 summarises QLU-C10D responses for the respondents who reported full health on EQ-5D-3L but not QLU-C10D at baseline (55) and at follow-up (9). Patients reporting full health with EQ-5D-3L still had problems when the disease specific measure was used, particularly with social function (51%) and fatigue (67%) at baseline and fatigue (78%) at follow-up where majority had less than the highest domain score.

Table 3 Distribution of QLU-C10D responses for participants reporting full health on EQ-5D but not on QLU-C10D
Full size table

Convergent validity

Results of the correlation between measures for both utility and dimensions scores are reported in Table 4. QLU-C10D and EQ-5D-3L utility score were strongly correlated (r = 0.71). Correlation between the EQ-5D-3L utility score and QLU-C10D functional domains were moderate (r > 0.4) but weak correlations (r < 0.4) were observed with the symptom domains. At the dimension level, anxiety/depression was strongly correlated (r = 0.71) with emotional function on the QLU-C10D while moderate correlations were observed for mobility and physical function (r = 0.6), usual activities with role function (r = 0.68), social function (r = 0.54), and fatigue (r = 0.41) as well as pain/discomfort with pain (r = 0.55). Very weak correlations were observed for pain/discomfort with role function (r = 0.18), nausea (r = 0.17) and bowel problems (r = 0.2), mobility with pain (r = 0.19) and nausea (r = 0.18), as well as usual activities with pain (r = 0.2) and personal care with fatigue (r = 0.18).

Table 4 Correlation between QLU-C10D and EQ-5D domains at follow-up
Full size table

Agreement

The Bland Altman plot (Fig. 3) showed a small mean difference and good agreement between QLU-C10D and EQ-5D-3L utility scores as only 3.7% observations were outside the limits of agreement.

Fig 3:
figure 3

Bland–Altman plot of QLU-C10D and EQ-5D-3L at follow-up

Full size image

Discussion

The clinical trial [16] underpinning the current study compared a homogenous group of patients with esophageal cancer (Table 1) who underwent surgical resection and received different regimens of nutritional support. As the outcomes from the trial showed no differences in clinical and quality of life outcomes between the different nutritional support regimens, the data was analysed as a pooled sample in this current paper. This analysis assessed the responsiveness and convergent validity of the cancer specific QLU-C10D and generic EQ-5D-3L for measurement of short-term HRQoL outcomes following esophagectomy.

As expected, both measures showed statistically significant reductions in HRQoL following surgery (Table 2). Utility decrements of 0.16 and 0.29 were observed with EQ-5D-3L and QLU-C10D respectively, and that of EQ-5D-3L was clinically significant [23]. As the QLU-C10D is a relatively new instrument, there has been no MCID reported in the literature yet, however, the reduction in QLQ-C30 score of 21.6 was clinically significant [24]. On comparing the EQ-5D-3L and QLU-C10D, the EQ-5D-3L mean utility score was greater than QLU-C10D by 0.04 at baseline and 0.17 at follow-up. The small difference at baseline but not at follow-up suggests a high degree of convergence for mild health states but not for severe states such as after esophagectomy, where the limitations of a generic measure become apparent, and a more sensitive disease specific measure is preferred.

Both instruments reported a large effect size with QLU-C10D being larger than the EQ-5D-3L. Most studies with EQ-5D-3L have reported low to moderate effect size [35]. However, studies with large expected changes in health status such as after surgery have reported large effect size [36,37,38]. Although both EQ-5D-3L and QLU-C10D are on the 0–1 QALY scale (where 0 is equivalent to being dead and 1 is full health), a much larger effect size was observed with QLU-C10D. This observation is similar to other studies comparing disease specific and generic measures where disease specific measures have much larger effect sizes or show greater responsiveness [39]. This is because the disease specific measures, unlike the generic, assess domains of quality of life that are of greatest importance to the condition.

For both EQ-5D-3L and QLU-C10D, ceiling effects were observed for all dimensions at both pre- and post-operative, with higher ceiling effects observed with EQ-5D-3L (Figs. 2 and 3). The highest ceiling effect for EQ-5D-3L was with self-care at both timepoints and the highest for QLU-C10D was nausea symptoms pre-operative and emotional function after esophagectomy. Ceiling effects of the EQ-5D-3L have been reported in several patient populations but the levels observed here were higher than reported in other studies among similar populations [40, 41]. In addition, the proportion reporting the best score for all dimensions (or full health) was much lower with the QLU-C10D at both timepoints. Yet patients reporting full health with EQ-5D-3L still reported problems when the disease-specific measure was used, particularly with social function and fatigue at baseline and fatigue at after esophagectomy (Table 4). These two dimensions are among four key bolt-on dimensions that been suggested to the current EQ-5D classification system [42]. Our findings are similar to Lorgelly et al. [9] who showed that EQ-5D-3L was less sensitive to fatigue impairment when compared to the disease specific EORTC-8D. Although EQ-5D-3L had a large effect size and demonstrated a clinically meaningful change in utility scores, the high ceiling effects suggest low discriminant ability. This means that the EQ-5D-3L is not as capable of identifying all or most of the change in HRQoL and distinguishing health states in this population. To address such ceiling effects generally experienced with EQ-5D-3L, a 5-level version of the instrument, EQ-5D-5L was developed [43], however, no studies are available comparing it to the QLU-C10D as yet. In addition, increasing the number of response levels alone may not be sufficient to improve the sensitivity of EQ-5D if the key dimensions are not covered. In their analysis investigating gaps in the EQ-5D descriptive system (using the EQ-5D-5L) Chen and Olsen argue for an extended health state classification system that includes four bolt-on dimensions namely vitality, sleep, social relationships and community connectedness [42].

Strong correlations were observed between QLU-C10D and EQ-5D-3L utility scores. The EQ-5D utility score was moderately corelated with all functional domains but weakly correlated with the symptom domains of the QLU-C10D. This is not suprising as unlike QLU-C10D, the EQ-5D measures function and no cancer specific symptom. As such, QLU-C10D is a preferred measure for application in clinical trials and studies in populations with cancer. Similar to other studies that have assessed convergent validity [9], domains that assessed similar constructs showed a strong or moderate correlation such as emotional function and physical function on the QLU-C10D with anxiety/depression and mobility on the EQ-5D-3L respectively.

The limitation of our study is the reduced sample size which resulted from excluding nearly 41% of patients who participated in the original RCT. However, excluding these patients is unlikely to have impacted the results as there were no significant differences in HRQoL outcomes between patients included in this analysis and those that were excluded.

From this analysis, both QLU-C10D and EQ-5D-3L are suitable for economic evaluations assessing interventions following esophagectomy. However, because the QLU-C10D is more responsive to change, QALYs calculated based on QLU-C10D utility scores should be used when comparing interventions in cancer. The variation in mean utility differences from these 2 instruments (0.16 for EQ-5D-3L vs 0.29 for QLU-C10D) translates into a difference in the utility gained from each of them. Although such a difference in utility would normally be treated the same in modelling studies (and only reported in the sensitivity analysis), the choice of preference based instruments influences the CUA results [44]. The major implication is that a decision maker might potentially not fund a treatment strategy, based on a less sensitive tool, when in reality the treatment is worth funding. This is especially useful in clinical conditions like cancer, where the disease specific measure is more sensitive and provides additional differentiation between treatments [9, 11]. For these studies the QLQ-C30 and subsequently QLU-C10D can be applied as a complement to but also as a substitute to the generic measure in assessing HRQoL. Developers of this algorithm have argued that in cases where the QLU-C10D utility scores are correlated with and as sensitive and responsive as those of the generic MAUIs, then QLU-C10D can substitute the generic MAUI [11]. However, using a generic measure, unlike the disease-specific measure, allows for generalisability and comparison of interventions between disparate populations. The generic utilities provide a benchmark comparison for decision-makers, such as PBAC and NICE who are charged with making allocative decisions across a range of pharmaceuticals or diseases within health systems. Yet where only the condition specific utilities have been used, comparisons can still be made because they facilitate the calculation of QALYs which allows for these types of comparison to be made as outcomes are measured on a common QALY scale (although the content of the measures themselves may be quite different between generic and condition specific measures).

Therefore, cognisant of requirements by policy makers to apply generic utility measures in cost effectiveness studies, the disease-specific QLU-C10D could be used alongside the generic measures like EQ-5D-3L. Some studies have applied mapping algortihms where a disease specific measure is mapped onto a generic measures [45, 46]. However, mapping algorithms assume that the generic measure is sensitive to changes in the target population, an assumption that is rarely tested. They are also often of variable quality and can have poor prediction ability as a consequence [46, 47]. With the advent of condition specific utility algorithms, it is now possible and preferable to calculate QALYs directly from the condition specific measure. It is also notable that in Australia the EQ-5D is not mandated for use in economic evaluation studies (unlike NICE in the UK for example), so there are choices to be made and QALYs from condition specific measures can still be used.

Conclusion

The findings from this study suggest that the newer QLU-C10D is comparable to the more widely applied generic EQ-5D-3L. However, EQ-5D-3L may not be sufficient in isolation when assessing short-term HRQoL following esophagectomy and should be combined with a condition specific measure. Cognisant of requirements by policy makers such as NICE in the UK and PBAC and MSAC in Australia to apply generic utility instruments in cost effectiveness studies, we recommend the application of a condition specific utility instrument alongside the generic instrument. In this way the sensitivity of a cost-effectiveness assessment decision can be determined by considering utility estimates generated from both a condition specific and a generic measure.