Trajectory of quality of life among patients with cancer at the end of life: a longitudinal survey study

Cancer is one of the leading causes of death and continues to impact the lives of millions worldwide profoundly. Historically, the primary focus of cancer treatment has been overall survival or progression-free survival [1]. However, health-related quality of life (HRQoL) is increasingly recognised as a pivotal outcome [2] particularly when cancer is beyond curative measures [3, 4].

Therefore, knowledge about HRQoL and its determinants in patients with cancer near the end of life (EoL) is important to health care providers and policymakers.

To measure HRQoL, various disease-specific and generic tools have been developed. Generic measures enable researchers to evaluate the impact of treatments on both health status and quality of life regardless of the disease type, treatment, or patient group [5]. One of the most widely used generic tools is the EQ-5D-5 L [6].

Condition-specific measures identify symptoms unique to a specific condition and the immediate effect of that condition on an individual’s quality of life [5]. The European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire (EORTC-QLQ-C30) is a widely used cancer disease-specific HRQoL instrument [7]. As the tool is not preference-based and does not yield utility scores, it cannot be used directly in economic evaluations [8].

Despite the importance of HRQoL, evidence remains sparse and with much of the existing literature relying on HRQoL data measured by various scales and, in most cases, from cross-sectional studies and clinical trials [9‐11] capturing one moment in time that may not be representative of the individual’s general HRQoL state [12].

Raijmakers et al. investigated a cohort of 892 diseased cancer patients who had replied to the EORTC-QLQ-C30 either 3, 3–6, 6–9, or 9–12 months before death. They found that patients who responded three months before death had a significantly lower HRQoL than individuals who answered earlier. Furthermore, they found that patients with cancer within their last year of life had a significantly lower HRQoL than the general population [13]. The study did not measure the development in HRQoL over time. Lee et al. [14] used the Functional Assessment of Cancer Therapy– General questionnaire to assess HRQoL in 345 patients with cancer every three months until death. Results showed a deterioration of HRQoL as death approached. Their conclusion emphasized the importance of systematic HRQoL monitoring, early identification, and referrals to palliative care to prevent the steep decline in HRQoL [14]. As with the study by Raijmakers et al., Lee et al. evaluated HRQoL using a disease-specific measurement tool. The individuals invited to participate in the study by Lee et al. were informed of their terminal state, which may in itself have influenced HRQoL [15]. Another cross-sectional study applied both EQ-5D-5 L and EORTC-QLQ-C30 to investigate HRQoL in patients with end-stage cancer. Both instruments were found to be applicable, but EQ-5D-5 L showed a pronounced ceiling effect with 13% reporting perfect health [11]. Strong ceiling (or floor) effects can make the instrument insensitive to change in the top and bottom of the scale [16, 17].

Randomized Controlled Trials (RCT) are another source of knowledge regarding HRQoL in cancer patients [9, 10]. The RCT population often comprises individuals with low-risk profiles who are younger and less co-morbid than the general cancer population [18]. Moreover, many RCTs only measure HRQoL during the intervention and not close to death [19]. Other studies have investigated HRQoL using real-world data focusing on specific cancer types only [20‐23], or using caregivers’ HRQoL assessment as a surrogate measure [24, 25]. While some studies have found generic HRQoL measurement tools to be less sensitive to changes in HRQoL in patients with cancer than the disease-specific tools, others have found the opposite, making results inconclusive, especially at EoL [26‐28].

In this study, our primary objective was to describe the development in HRQoL over time in patients with cancer near the end of life, using both the generic preference-based EQ-5D-5 L and the disease-specific EORTC-QLQ-C30 questionnaires. Our secondary objective was to show the trajectory of the subdomains of both HRQoL measures over time to show the drivers of the development in overall HRQoL.

Out of the invited individuals that died during or right after data collection, nearly half agreed to participate and responded to the first questionnaire as a minimum. Both HRQoL, as measured by the EQ-5D-5 L and by the EORTC-QLQ-C30, indicated a decline as the time of death approached. The deconstructed EQ-5D-5 L showed that the decrease in HRQoL was especially affected by loss of mobility and usual activities. The deconstructed EORTC-QLQ-C30 showed that the decline was especially driven by a loss in physical, role, and cognitive functioning as well as an increase in severity of the symptoms fatigue, dyspnoea, and loss of appetite.

A strength in our study was the ability to link data from different sources via the patients’ unique CPR numbers, allowing us to enrich our prospective data with retrospective registry data without asking for extra information from the patients. We were able to compare the respondents with non-respondents using hospital administrative data and Danish registry data. Our results showed that the respondents and non-respondents were comparable in many aspects, e.g. according to age, referral to SPC, and length of disease trajectory. This is important, as a major challenge in collecting survey data is that a select group tends to respond when invited to participate in surveys [35].

In a newly conducted study by Jensen et al. [36], the authors found an average EQ-5D-5 L score of 0.89 for the age group 60–69 in the Danish general population. This is similar to the mean EQ-5D-5 L score of 0.84 the 11th month before death found in this study. When approaching the time of death, especially in the last six months, HRQoL deteriorates, ending at a score of 0.44. This implies that when measured by the EQ-5D-5 L, HRQoL in patients with cancer is comparable to the general population until approximately the last six months of life. In the EORTC-QLQ-C30, the global health status/QoL measured at 12 months prior to death was 65, which was 8 points below the Danish population average as found by Juul et al. in 2014 [4]. While our results indicate a decline, they also highlight that HRQoL is not stationary, making continued measures over time important. A cross-sectional study would have concluded very different results depending on the timing of the study.

Several previous studies have found evidence that patients suffering from incurable cancer wish to prioritize the improvement or stabilization of their HRQoL [1, 14, 37]. This entails that the healthcare professionals know what factors are influenced by the disease and how to alleviate them. If the goal of care at EoL is improving HRQoL, continuous screening of HRQoL could help detect deterioration and thereby begin to remedy HRQoL sooner. Our results indicate that especially the ability to be mobile, perform self-care, and usual activities affect HRQoL. These are skills that are related to a feeling of autonomy and ‘being normal’ [38]. Houska and Loučka find that emphasis of care should be placed on supporting the patients in their daily activities and contribution to others [38]. This aligns with our finding that Role functioning in the EORTC-QLQ-C30 measurement was a driver of low HRQoL, indicating the importance of being able to fulfil one’s role, which may be given inadequate attention in a healthcare setting. Which type of role functioning the patient finds most important is individual and may vary according to age and sex, among other things [39]. Moreover, we found that physical functioning and loss of appetite worsened towards the time of death. Both of these have previously been identified as prognostic indicators in patients with cancer [40]. This suggests that applying these measurements continuously when caring for patients with end-stage cancer could help detect when patients are nearing the time of death, thereby securing timely initiation of appropriate EoL care. This could include termination of active cancer targeted treatment, referral to SPC, ensuring that the patient is able to spend more time at home (if possible), initiating extended pain management or other symptom relief. This may enable the patient to better fulfil a role or preform an activity that they find meaningful. More research is needed to investigate the relationship between HRQoL decline and time of death. As we only include cancer decedents and evaluate data retrospectively, we cannot causally claim that continuous measuring HRQoL would improve prospective accuracy of identifying EoL decline.

Knowledge about the functions and abilities that effect HRQoL at EoL may aid healthcare professionals in initiating correct intervention aimed at alleviating the steep decline at EoL. Of course, a complete alleviation is not realistic, as a decline in HRQoL at EoL is to be expected, despite care quality.

In this study, we applied both the EQ-5D-5 L and the EORTC-QLQ-C30 to the same population. This allowed us to evaluate HRQoL in a more nuanced manner compared to applying a single tool. The two tools are different in construction and designed to measure different elements of the disease, but as our results indicate, both tools capture the decline in HRQoL and the sum scores of both instruments follow a similar curve when patients approach the time of death. Our results do not indicate an insensitivity in the EQ-5D-5 L; on the contrary, it seems that the preference-based weighted categories capture change in HRQoL well. Ceiling and floor effects are recognized as a problem when using the EQ-5D-3 L [41], but this problem with the EQ-5D-5 L is not evident in our results, which is in line with previous studies that highlight ceiling effects as primarily being a problem in healthy populations [42].

Depending on the research question, a strategic approach might involve utilizing a combination of tools, an approach suggested by others [43, 44]. The EORTC-QLQ-C30 holds valuable data for customizing EoL care, specifically in addressing prevalent symptoms associated with cancer treatment. Meanwhile, the EQ-5D-5 L offers crucial insights that should be considered in inevitable priority discussions, as the EQ-5D-5 L outcome is comparable across different diseases and interventions. As cancer has a distinct EOL trajectory [45] and future research could focus on collecting equivalent data from patients at EoL suffering from diagnosis other than cancer.

Despite the population size being within range of previous literature [37, 46, 47], the number of participants prevents us from incorporating adequate independent variables in the models, thus risking residual confounding. Moreover, this study may be subject to selection bias as only responders are invited to participate in following questionnaires. This approach was chosen in an attempt to not pressure a vulnerable population excessively.

Future research could try to include a larger sample, to enable more independent variables or matching methods. Moreover, applications of the forthcoming Danish EORTC-QLQ-C30 utility weights could further enhance the comparison of the instruments. To compare EQ-5D-5 L and EORTC-QLQ-C30 more adequately, further studies including state-of-the-art mapping techniques are needed to verify the results of the current study. Mapping predicts health state utility values when no preference-based measure is included in a study. It estimates the relationship between non-preference-based and generic preference-based measures using statistical associations, requiring overlap in their descriptive systems and administration to the same population. Brazier et al. [48] reviewed various models used for this, such as OLS, generalized linear models, tobit, CLAD, 2-part models, and response mapping. Studies assess model performance using metrics like predicted mean and standard deviation, median, range, AIC, BIC, R², pseudo-R², RMSE, MSE, and MAE.