Introduction
Health-related quality of life (HRQOL) has become an important secondary outcome measure in clinical trials of glioma patients [
1,
2] with all European Organization for Research and Treatment of Cancer (EORTC) brain tumor clinical trials and most trials by other cancer groups now incorporating HRQOL. HRQOL is an important complement to conventional outcome parameters such as time to tumor progression, overall and progression-free survival, and radiological response might be inadequate or less relevant for meaningful evaluation of this type of treatment [
3]. HRQOL in glioma patients is influenced by both tumor- and treatment-related factors [
4,
5] with seizures and antiepileptic drugs [
6], fatigue [
7], anxiety and depression [
8], and neurocognitive deficits [
9,
10] affecting HRQOL in particular. Apart from negatively affecting HRQOL, neurocognitive deficits may also hamper adequate patient self-reports, as patients’ neurocognitive deficits may render HRQOL patient-reported outcomes through questionnaires unreliable [
11]. Exclusion of these patients at the lower end of the neurocognitive spectrum from analyses obviously introduces undesirable bias in the evaluation of patients’ HRQOL during experimental treatments. Moreover, cognitively impaired patients may be less compliant regarding questionnaire-based HRQOL monitoring, thereby introducing another source of bias.
The incorporation of HRQOL estimates of the partner or another close relative or friend (denominated as ‘proxy’) might solve this problem to a large extent. Previous reports indicate that high-grade glioma patient- and proxy-reported HRQOL have a high level of concordance as long as the patient shows no signs of decline in neurocognitive function [
12,
13], but differences, particularly in mood-related issues, increase when neurocognitive functioning decreases [
13,
14]. While HRQOL is by definition subjective, and it is typically measured with self-reports, it has been suggested that substituting proxy ratings when a patient’s self-report is absent or unreliable should be considered [
15]. When differences between patient- and proxy-reported HRQOL ratings develop in the course of the disease (presumably at the time, decline in neurocognitive function becomes an issue), proxy-reported instead of patient-reported HRQOL ratings might be regarded as the most reliable source of information on patients’ HRQOL.
Previous studies reported that low observed correlations between patient- and patient-by-proxy-reported outcomes might be explained by methodological weaknesses such as small sample size, suboptimal reliability, and score variability [
13,
14]. This was supported by Bland and Altman, who stated that a single measure such as a correlation coefficient may not be sufficient to summarize agreement adequately [
16,
17]. In the present study, by using a wide range of statistical measures of agreement, we investigated patient–proxy HRQOL agreement in a large sample of low-grade glioma (LGG) patients, both with intact and with impaired neurocognitive functioning. The cohort used in this analysis is unique because of the extensive neurocognitive test battery incorporated.
The aim of the present study is to investigate the agreement between patient and patient-by-proxy ratings of HRQOL and to investigate whether the level of neurocognitive functioning influences the level of patient–proxy concordance. We hypothesized that (1) concordance levels are relatively high on mental and physical functioning in cognitively intact patients, and (2) there is a decrease in mental functioning in cognitively impaired patients, with proxies being more negative on patients’ HRQOL.
Discussion
Measuring neurocognitive functioning is essential in brain tumor patients, because this may strongly influence their HRQOL and also patient–proxy concordance levels. Although the patient is the primary source of information when measuring HRQOL, the information collected from patients with glioma may be unreliable, especially in those patients who are experiencing significant neurocognitive deterioration [
14]. It is recommended to obtain proxy (i.e., caregivers) HRQOL ratings alongside a patient’s own self-report and to consider substituting patient-by-proxy ratings when a patient’s self-report is absent [
43]. The original analysis of this study showed that glioma patients reported lower levels of self-reported neurocognitive functioning as measured by the MOS scales than did the healthy controls: 47.80 versus 82.40, respectively [
18].
In the present study, we found that there was overall a high agreement between the patient and patient-by-proxy rating of LGG patients HRQOL in most subscales of the SF-36 and QLQ-BN20. The only statistically significant differences were observed in SF-36 physical functioning and general health and QLQ-BN20 visual disorder, communication deficit, and itchy skin. Noticeable mean differences were observed in the cognitively impaired group especially on the QLQ-BN20 scores regarding visual disorder, headaches, itchy skin, and bladder control. A noticeable difference in the cognitively intact patient group was only observed in the QLQ-BN20 headaches score. The difference between patient and patient-by-proxy ratings found in the whole group of patients could be due to the cognitively impaired patients. Although all of the differences were statistically significant, they were less than the 10 points generally accepted as clinically meaningful. However, some scores may have represented small, potentially noticeable changes in the range of 5–10 points [
44] which could be important to the individual patient and warrant clinical attention.
The Bland–Altman plot revealed a high agreement between the patient and patient-by-proxy rating of HRQOL, where about 95 % of the differences between the two measurements were within the 95 % limits of agreement (prediction interval) except for SF-36 physical functioning scale and the summary component score for physical functioning (PCS). However, the limit of agreement was lower in the cognitively impaired patient group. One of the reasons for this lower agreement may be that patients, who are aware of the fact that their cognitive functions are severely affected, regard their HRQOL as poor (which is also the case in AD patients). Proxies may not fully appreciate the emotions which accompany decline of intellectual functioning.
It is important that the extent of agreement across the range of measurement be stable between the patient and patient-by-proxy [
39]. Our findings showed that the agreement was poor for the central section of the scales (supplementary Figures). This was also shown by Giesinger et al. [
12] who likewise stated that the possible discrepancies (i.e., ‘bias’) between the patient and patient-by-proxy are reduced by the limited range scale. The current study found a moderate-to-strong correlation between patient and patient-by-proxy scores (CCCs >0.5 for 79 % of the measurements). It is thus much easier to demonstrate agreement when a patient is experiencing either very few or many symptoms, but as the number of symptoms moves closer to 50–50 distribution, patient–proxy agreement decreases.
Regarding the methodology, our results compare quite favorably with other studies that have examined proxy ratings for general cancer patients [
13,
45], brain cancer patients [
12,
13], epilepsy patients [
46], and stroke patients [
47]. For example, a previous study in HRQOL of brain cancer patients and their proxy raters showed that intra-class correlations (ICCs) were greater than 0.5 for 77 % multi-item measures and 38 % of single-item measures [
13]. However, in this previous study [
13], the authors did not implement an extensive testing of patients’ neurocognitive functioning, which we did in this study. Also, the use of sound statistical techniques such as the Bland–Altman limits of agreement [
16,
36‐
39], which are straightforward and easy to interpret, enabled us to investigate any possible relationship between the measurement error and the true value. Furthermore, our study has a large sample size and a homogeneous patient population.
Patient-by-proxy ratings may resolve compliance issues when assessing HRQOL in glioma patients with intact neurocognitive function. Probably more important in glioma patients than in any other cancer patient population, but comparable to other patients with neurological diseases associated with neurocognitive decline (e.g., Mild Cognitive Impairment and AD), patient-by-proxy ratings might also be helpful when patients cognitively deteriorate and lack the ability and insight to accurately interpret and understand the HRQOL measures. In the current study, although there was a good agreement between patient and patient-by-proxy ratings for the whole sample, there was less agreement between patient and patient-by-proxy ratings for those patients with impaired neurocognitive function compared to those patients with unimpaired neurocognitive function. While patient and patient-by-proxy ratings in such situations should not be regarded a priori as incorrect [
46], insight is needed into the sources of variation between patient and proxy ratings. In a small study that compared HRQOL ratings from proxies and patients with mild AD, mild cognitive impairment, and elderly controls, it was found that overall patient–proxy agreement did not differ significantly between groups despite evident differences in neurocognitive functioning [
48]. In a related study, patients with early AD generally reported a
higher HRQOL than their proxies, and discrepancies in patient–proxy ratings were associated with the presence of anosognosia [
49]. Although in the current study self-awareness was not evaluated, we found that LGG patients with cognitive deficits tended to report
more tumor- and treatment-related symptoms and thus a
lower HRQOL. This might indicate that potentially reduced self-awareness can be associated with both higher and lower patient HRQOL ratings relative to proxy ratings. There is currently no consensus on the best way to deal with inconsistent patient–proxy reports. While most methods rely on proxy report as a ‘gold standard’ with patient–proxy concordance taken as an indirect measure of patient (lack of) insight, the accuracy of proxy reports bares critical examination when the proxy is the caregiver. While the proxy-related factors affecting patient–proxy discrepancies are largely unknown in brain tumor patients, studies in patients with mild cognitive impairment as a prodromal phase of AD, for instance, have shown that caregivers’ cognitive skills and educational level are significant predictors of the discrepancies between caregiver ADL reports and directly assessed patient performance on ADL [
50]. Furthermore, caregivers’ age, financial situation and valuation of life as a whole [
51], the type of caregiver, the perspective used [
52], caregiver burden [
53], level of depression and anxiety [
54], and caregiver health may influence the accuracy of the caregiver report. As stated earlier, patient-related factors that might affect concordance between glioma patient and proxy ratings include compromised mood and decreased neurocognitive functioning [
13,
14]. Interestingly, a study that focused on screening for major depressive disorder in glioma patients [
55] did not find patient–proxy agreement to be associated with severity of patient cognitive impairment, although there was frequent disagreement between glioma patients and proxies reports of depressive symptoms. A study that focused on the effect of neurocognitive functioning and performance status (KPS) on patient–proxy concordance [
56] found patients and proxies to have highly congruent assessments of symptom severity regardless of patients’ neurocognitive functioning and performance status. Use of proxies as a substitute for the patient self-report of HRQOL should thus be treated with caution, always taking into consideration the possibility of potential bias.
A limitation of our study is the cross-sectional nature of the data as opposed to longitudinal data or follow-up data and generally mild neurocognitive problems in LGG patients which did not allow the detection of small mean differences between patient and patient-by-proxy ratings. Follow-up data or assessment in high-grade glioma patients with probably more neurocognitive problems might have allowed the detection of small differences between patient and patient-by-proxy HRQOL ratings. The percentage of mean differences (equal or below 0, 5, and 10 points) could be impacted by the number of possible scores on a scale [
12]. A very low number of possible scores or a very large distance between two possible scores (i.e., >10 points) could distort the agreement accuracy. Also, since patients in this study had stable disease with mild neurocognitive impairment, stable LGG are not representative of the general brain tumor patient population. Investigating agreement on high-grade glioma (HGG) patients with severe neurocognitive impairment would provide additional information to assess agreement between patient and patient-by-proxy ratings. The present study specifically addressed HRQOL and did not include estimates of mood or depression. Theoretically, mood might have affected our outcomes to a certain extent as a study among patients with major depression showed that responses to self-report questionnaires are influenced by the presence of depression [
57].
One issue of potential concern is basing the analyses on 195 patients from the original sample of 281 patients might result in bias. This might have been the case if patients who did not participate in neurocognitive testing were excluded, for instance, because of poor neurological or physical status. We would argue, however, that it is unlikely that bias was introduced in our study. At the time that we conducted the original study, our expectation was that only approximately 100 LGG patients would be alive in the Netherlands and meet our eligibility criteria. In fact, we were ultimately successful in identifying 281 eligible patients. Because of limited financial and personnel resources, we were only able to assess neurocognitive functioning and HRQOL in 195 consecutive patients; we assessed HRQOL only in the remaining 86 patients. There was no evidence to suggest that those who underwent both neurocognitive testing and completed HRQOL assessment differed in any significant way from those who only completed the questionnaires.
In conclusion, our data demonstrate that there is an overall high level of agreement between patient and patient-by-proxy ratings of LGG patients’ HRQOL, although the agreement for some measures is weaker in those cases where patients have neurocognitive impairment. This implies that in general, patient-by-proxy-reported outcomes can be used to replace missing patient-reported outcomes to solve compliance issues in clinical trials in this patient population. This is not the case, however, for patients with cognitive deficits who are no longer able or willing to provide self-reported data. Specifically, regarding the lack of a ‘gold standard,’ discordant patient–proxy reports should currently be considered as a parallel source of information on patient functioning. Since it is not always possible to predict which patients will suffer from progressive neurocognitive deficits, or when, it is advisable to build proxy assessments into study designs from the start of brain tumor clinical trials as is currently the case in EORTC studies 26101 (NCT01290939) and 26091 (NCT01164189).
Acknowledgments
We thank the neurologists, neuro-oncologists, neurosurgeons, and radiation oncologists from VU University Medical Center, The Netherlands Cancer Institute/Antoni van Leeuwenhoek Hospital, Slotervaart Hospital, Academic Medical Center (Amsterdam), Leiden University Medical Center (Leiden); Medical Center Haaglanden, HAGA Hospital (The Hague); Erasmus University Medical Center (Rotterdam); St. Elisabeth Hospital (Tilburg); University Medical Center Maastricht (Maastricht); Hospital De Wever (Heerlen); University Medical Center Utrecht (Utrecht); Sint Antonius Hospital (Nieuwegein); University Medical Center Sint Radboud (Nijmegen); and Groningen University Medical Center, Martini Hospital (Groningen) for permission to include their patient–proxy dyads in the study. Furthermore, we would like to thank the Dutch Cancer Society, La Fondation Contre le Cancer, National Cancer Institute, EORTC Research Fund, and Pfizer Global Partnership Foundation for funding this study; and Cheryl Whittaker for editorial help.