Proxy ratings of psychological well-being in patients with primary brain tumors: A systematic review

Abstract

Objective:

This systematic review examined the agreement of proxy ratings of depression and anxiety in neuro-oncology patients.

Methods:

Searches were conducted across 4 databases (MEDLINE, Embase, PsycINFO, CINAHL, and Web of Science) to identify studies that compared proxy ratings (non-health care providers) of anxiety and depression in patients with brain cancer. Methodological quality and potential risk of bias were evaluated using the Joanna Briggs Institute Critical Appraisal Checklist.

Results:

Out of the 936 studies that were screened for inclusion, 6 were included for review. The findings were mixed in terms of whether patient and proxy ratings were accurate (e.g., deemed equivalent), with many of the selected studies suggesting moderate level of agreement for several of the selected studies and, when both depression and anxiety were included, depression ratings from proxy raters were more accurate than for anxiety. We identified important limitations across the selected articles, such as low sample size, clarity on defining proxy raters and the different instructions that proxy raters are given when asked to assess patients’ mood symptoms.

Conclusions:

Our findings suggest that proxy ratings of depression and anxiety should be interpreted with caution. While there is some agreement in proxy and patients with brain cancer ratings of depression and anxiety (greater agreement for depression), future work should recruit larger samples, while also remaining mindful of defining proxy raters and the instructions given in collecting these ratings.

1 |. INTRODUCTION

Patients with cancer are particularly susceptible to suffering from depression and anxiety.¹ Prevalence rates vary widely, between 5% and 50% for depression² and 19%–42% for anxiety.^3,4 This mood disturbance can negatively impact adherence to treatments, patients’ perception of other symptoms, and quality of life. There is also suggestion that depression and anxiety contribute to increased cancer incidence, and both cancer-specific and all-cause mortality,⁵ While screening for more general stress remains a particular focus in psycho-oncology,⁶ screening and measuring depression and anxiety, specifically, remains critically important as effective supportive interventions exist to treat these conditions.⁷ Despite the knowledge of this gap,⁸ depression and anxiety remain widely undetected in much of oncology practice.^9,10

Neuro-oncology presents inherent challenges to assessing depression and anxiety. With expressive communication frequently impacted, neuro-oncology patients may be challenged in completion of questionnaires inquiring about these domains. In addition to the neural location of the tumor, treatment often includes surgical resection, radiation to the brain and potential systemic therapy, all of which can also impact patients’ ability to communicate, as well as their mood and well-being.^11,12 These challenges extend to patients enrolled in clinical trials, as patients have more motor dysfunction and worsening clinical status, they are less likely to complete questionnaires and more likely to drop out of studies.¹³ Despite these challenges, many have advocated for the incorporation of questionnaires into the assessment of clinical endpoints, termed RANO (Response Assessment in Neuro-oncology),¹⁴ given the importance to patients’ quality of life and ability to distinguish between comparable treatment regimens. These inherent challenges in depression/anxiety measurement in neuro-oncology have led many investigators to focus on capturing proxy ratings of patients’ quality of life and well-being.

The challenges in assessing another’s well-being are multifold. Measuring other’s perceived ability to function or subjective experience can be challenging, and researchers rarely apply appropriate methodology to instruct proxies in measuring patient behavior¹⁵ nor apply appropriate analytic techniques to compare the accuracy of proxy ratings¹⁶ leading to disagreement in ratings. When clinical care providers attempt to rate quality of life in their patients, agreement is poor^17,18 compared to close relatives or spouses in which the agreement is much higher.¹⁹ There is more agreement in concrete domains (such as functioning) than in subjective domains such as mood. For instance, when dementia patients rate their own quality of life, they rate this as higher than both their relatives and staff.²⁰ Parents of children with cancer may underestimate their children’s emotional concerns overall, but overestimate the anxiety they have over the course of treatment.²¹ Collectively, the challenge of consistent agreement between patients’ self-assessment and proxy’s assessment is referred to as the “interrater gap” and acknowledges that imagining another’s experience should be an intentional shift in perspective.¹⁵

Examination of proxy raters’ accuracy suggests some agreement between patients and caregivers’ perception of the patients’ quality of life,²² much less attention has been paid to domains of mood.²³ We know very little about assessing depression and anxiety in neuro-oncology patients with proxy raters, yet up to 48% of patients and 40% of caregivers experience significant depression or anxiety following diagnosis of a brain tumor.²⁴ Several demographic and treatment-related factors, such as being female, having a lower WHO tumor grade classification, lower education level, and a history of psychiatric illness are significantly related to greater likelihood of clinically significant anxiety or depression.²⁵ To our knowledge, there has not been a systematic review of the agreement of proxy ratings of depression or anxiety in neuro-oncology, despite their high prevalence rates and inherent challenges in assessing mood in this population.

With this background, the current aims of the study were as follows:

1. What is the level of agreement between proxy ratings of patients’ level of anxiety and depression and patient reported measures of these domains in primary brain cancer?

2. If both patient and caregiver proxy measures of anxiety/depression are collected and reported, what is the strength of this relationship and are they deemed equivalent?

2 |. METHODS

We followed the statement of Preferred Reporting Items for Systematic Review and Meta-Analyses outlined by Moher and colleagues.²⁶ The systematic review was registered with PROSPERO registration number CRD42021247201.

2.1 |. Search strategy

We conducted a search of the following electronic bibliographic databases: MEDLINE (Ovid), Embase (Elsevier), Web of Science Core Collection (Clarivate), PsycINFO (EBSCO), and CINAHL (EBSCO). The search strategy was designed to capture publications reporting any outcome using a proxy measure in brain cancer patient-caregiver dyads (Appendix 1). “Proxy ratings” were defined by ratings of patients’ well-being from someone outside of the treating medical team. This included any type of informal/unpaid caregiver, involved in the patients’ care; however, to ensure that comparisons were appropriate across the selected article, we extracted the definition of each manuscript’s proxy rater. We relied on a previously applied search strategy to define primary brain tumors.²⁷ For including proxy ratings, we applied the definition of ratings of patients’ well-being from someone outside of the treating medical team. Controlled vocabulary terms were included in the search when available; no date limit was applied. The search was last run on 28 April 2022. Covidence software²⁸ was used to organize all abstracts.

2.2 |. Inclusion criteria

Studies were included if (1) they recruited individuals 18-years-old or older with primary brain tumors diagnosed at or after 18-year of age; (2) examined psychological factors (depression and anxiety) assessed by a caregiver proxy who was not part of the healthcare team; (3) contained depression and anxiety rated by the patient; (4) were published in the English language. Studies were excluded if (1) published as abstract only; (2) review paper only; (3) lacked relevance (e.g., basic science, animal studies, etc); (4) captured other proxy ratings (e.g., from a clinical professional, for instance); (5) psychological factors not quantitatively measured; (6) No comparison of proxy ratings of patients’ anxiety/depression and patient rating of anxiety/depression. There were no limitations on study setting or country of conduct. After employing this search strategy, however, and identifying only five papers meeting inclusion criteria, we opted to also include any paper captured by our search criteria that compared patient and caregiver ratings of anxiety and depression, even if these constructs were part of a larger battery of measuring quality of life, in line with evidence that anxiety and depression significantly overlap with quality of life.^29–31 We also opted to include studies that measured proxy ratings of depression or anxiety, not restricting our search to having both. In our screening of studies, it also became apparent that several studies did not describe how measures of anxiety and depression were adapted to obtain proxy ratings or, more specifically, the instructions that proxy raters were given which we felt particularly important given the interrater gap referenced above. Thus, we extracted how each study described their approach to administering their specific measures to proxies, or if these details were not provided. Based on these revised criteria, six studies were included in this systematic review (see Figure 1 for reporting according to PRISMA guidelines).²⁶

FIGURE 1.

PRISMA describing screening and selection of studies.

2.3 |. Data extraction

Study authors (Hermioni L. Amonoo, Elizabeth G. Broden, Miryam Yusufov and William F. Pirl) extracted the following from each of the selected articles: lead author, publication year, study design, sample size, cancer characteristics of study population, and measures employed (Table 1). Other study authors (Damien M. Miran and Darcy E. Burgers) also extracted statistical values related to the level of agreement of proxy ratings and how studies defined caregivers/proxy raters, further described below. Study authors (Timothy S. Sannes and Miryam Yusufov) assessed methodological quality and potential risk of bias using the Joanna Briggs Institute Critical Appraisal Checklist for Studies Reporting Prevalence Data³² (Table 2). Questions taken from these criteria³² assessed by our team included, “Was the sample frame appropriate to address the target population?” “Were study participants sampled in an appropriate way?” “Were the study subjects and the setting described in detail?” “Was the data analysis conducted with sufficient coverage of the identified sample?” “Was the condition measured in a standard, reliable way for all participants?” “Was there appropriate statistical analysis?” For the question, “Was the sample size adequate?” we consulted criteria for other dyadic studies.³³

TABLE 1.

Extracted papers, associated measures and statistical comparisons

Study authors/year	Participant sample (N)	Instrument	Statistical comparison	Numerical level of agreement: Group comparison VS. continuous scale of level of agreement	p value
Moreale et al. (2017)	Low grade glioma patients who underwent a neurosurgical procedure at least 1 year prior (n = 46)	Beck depression inventory scalea State- trait anxiety inventorya	Paired t-test and chi square tests using the pearson index.	Depression: - Patient: 7.76 versus caregiver: 6.23	p = 0.102
				Anxiety: - State: Patients: 44.26 versus caregiver: 40.28	p = 0.031
				- Trait: Patient: 37.95 versus caregiver: 35.17	p = 0.047
Rooney et al. (2013)	New glioma diagnosis; receiving active treatment (N = 41)	Patient health Questionnaire-9 (PHQ-9; for internal validity)a	Patient-proxy agreement = PHQ-9 within two points:	PHQ-9 agreement: 36.6% (15/41) of cases 6.80 versus proxy ratings of 8.40	p = 0.016
		Structured clinical interview for DSM-IV (SCID; for external validity)b	Paired sample wilcoxon rank sum test
			Intra-class correlation coefficient	ICC: 0.69
			Cohen’s kappa (K; proxy reports on questionnaire compared to diagnostic interview for patients’ depression)	Proxy K = 0.58 Patient K = 0.47
Milbury et al. (2019)	Patients with WHO grade I-IV glioma (n = 20)	Center for epidemiological studies-depression (CES-D)a	Paired t-tests between patient-proxy ratings	CES-D: patient (mean = 16.49) versus caregiver (mean = 16.25) t = 0.13	p = 0.90
		MD Anderson symptom inventory-brain tumor module (MDASI-BT): subscalesa	Intra-class correlation coefficient	MDASI-BT: No difference in overall scores, subscales: ICC 0.51–0.82	P = 0.007–0.0001
Papers that compared agreement of emotional well-being without specific measures of anxiety/depression
Armstrong et al. (2012)	Primary brain tumor patients receiving neurocognitive testing as part of routir care (n = 115)	MD Anderson symptom inventory-brain tumor (MDASI-BT)a	Comparison of SD/CI (group differences with >90%)	Distress: 0.92 Sadness: 0.62	p = 0.001 p = 0.032
			Mean difference	Irritability: 0.26 Mood: 0.47	p = 0.28 p = 0.074
			Pearson correlation	Distress: 0.09 Sadness: 0.14 Irritability: 0.15 Mood: 0.11	p = 0.32 p = 0.14 p = 0.11 p = 0.25
Brown et al. (2008)	Patients with grade 3 astrocytomas in 3 treatment protocols (n = 197)	Profiles of mood states short form (POMS-SF)a	- Paired signed rank test	% w/in 10 pts of one another POMS: 61%	p = 0.30
	Baseline: 118 dyads	The symptom distress scale (SDS)a		SDS: 62%,	p = 0.11
			- Spearman’s correlation	POMS: 0.65 SDS: 0.66	N/A
			- Intraclass correlation (ICC)	POMS: 0.61 SDS: 0.64
Jacobs et al. (2014)	Grade III or IV malignant glioma (n = 45)	FACT-Br subscale: Emotional well-being (EWB)a	Intraclass correlation coefficient (ICC)	EWB: 0.58	N/A

Note: For comparing the accuracy between two groups, paired samples t-tests or, for data that is not normally distributed, paired signed rank test are common. Across both tests, a significant value is interpreted as the two samples being nonequivalent. For assessing the level of agreement between two measurements of a continuous variable common tests include Pearson correlation and ICCs. Correlation coefficients and ICCs provide a single value for the level of agreement and, if statistically significant, then the two ratings are comparable. The scale for the ICC is the same as for other correlations—0 represents no relationship, while 1 represents a perfect linear relationship. Generally, as outlined by Koo and Li³⁴ an ICC of less than 0.50 is poor, agreement between 0.50 and 0.75 is moderate, 0.75 and 0.90 is good and above 0.90 is considered excellent agreement. Cohen's kappa coefficient (Κ) is used to measure the level of agreement for a categorical outcome and has a similar scale: ≤ 0 as indicating no agreement and 0.01–0.20 as none to slight, 0.21–0.40 as fair, 0.41–0.60 as moderate, 0.61–0.80 as substantial, and 0.81–1.00 as almost perfect agreement.⁵¹

^aSame scale administered to both patient and caregiver.

^bProxies “encouraged to participate [in interview] as they saw fit.”

TABLE 2.

Methodological quality, risk of bias, and quality assessment for the 6 included empiric studies

	1. Sample	2. Sampling	3. Sample size	4. Description	5. Data analysis	6. Methods	7. Measures	8. Statistical analysis	9. Response rate
Moreale et al. (2017)	+	+	−	+	+	+	+	+	+
Rooney et al. (2013)	+	+	−	+	+	+	+	+	+
Milbury et al. (2019)	+	+	−	+	+	+	+	+	+
Armstrong et al. (2012)	+	+	+	+	+	+	+	+	+
Brown et al. (2008)	+	+	+	+	+	+	+	+	+
Jacobs et al. (2014)	+	+		+	+	+	+	+	+

Note: +yes; −no; ?unclear; N/A = not applicable. Contents for this table were guided by the “Critical Appraisal Checklist for Studies Reporting Prevalence Data” from Munn, Z., Moola, S., Lisy, K., Riitano, D., Tufanaru, C. (2015). Methodological guidance for systematic reviews of observational epidemiological studies reporting prevalence and incidence data. International Journal of Evidence Based Healthcare, 13(3)147–153.

2.4 |. Extracted statistics from selected publications

We extracted all reported statistics related to within dyad comparisons; included paired samples t-test, paired signed rank tests, correlation coefficient, intraclass coefficients (ICCs) and Cohen’s Kappa coefficients (K). Specifically, we extracted the numerical level of agreement and associated significance level documented in each selected publication. For assessing the level of agreement between two measurements of a continuous variable (e.g., Pearson correlation and ICCs), the correlation coefficients and ICCs were extracted. In this case, this single correlation coefficient, if statistically significant, suggests that there is a linear relationship between the two measurements (patient vs. caregiver). If the extracted coefficient was not significant, we assigned a label of not related (“NR”). Generally, as outlined by Koo and Li³⁴ an ICC of less than 0.50 is poor, agreement between 0.50 and 0.75 is moderate, 0.75 and 0.90 is good and above 0.90 is considered excellent agreement. Cohen’s kappa coefficient (Κ) is used to measure the level of agreement for a categorical outcome.

3 |. RESULTS

3.1 |. Study characteristics

Electronic database searching yielded 930 unique records and 39 full-text articles were read and examined for inclusion. Six studies met our inclusion criteria (Figure 1). Regarding the search criteria within the six selected studies, only two met the original inclusion criteria, 4 met the adjusted inclusion criteria such that these articles compared broader emotional well-being within a larger battery of quality of life (e.g., MD Anderson Symptom Inventory) and not specifically measures of anxiety and depression. Two^35,36 of the six studies were longitudinal, whereas the rest were cross-sectional study designs. Finally, we excluded three papers^37–39 that were captured within the established search criteria of proxy ratings but only reported on comparisons between patients and caregivers’ ratings of their own symptoms.

3.2 |. Description of study participants

There was a wide range of sample sizes across the selected studies. The sample sizes of these studies ranged from 20 patient-caregiver dyads³⁶ to 118 dyads.³⁵ Across the six selected studies, there was a total of 385 patient-caregiver dyads included.

3.3 |. Clinical characteristics

Characteristics of patients’ diagnoses ranged widely, all reporting on patients with a general diagnosis of glioma and were recruited across a wide range of points along the cancer trajectory (e.g., diagnosis, treatment, post-treatment etc).

3.4 |. Proxy rater definitions

Our search criteria specified that proxy raters would be, “ratings of patients’ well-being from someone outside of the treating medical team;” however, during our search, it became apparent that there was wide heterogeneity by which investigators defined proxy raters. As such, we chose to pull the exact wording each investigator used to define proxy raters for their study (see Table 2). The majority of studies relied on patients to identify the proxy rater who could report their symptoms. Definitions of proxy raters were as simple as cohabitating partner,⁴⁰ patient-identified caregiver,³⁵ whereas other studies further specified that family caregivers included spouses, family members or adult children³⁶ or could include these roles as well as friends of the patient.⁴¹ Other selected studies further defined the roles of family caregivers such as providing care are home,⁴² or, more specifically, home care for physical, emotional and social needs.⁴³ See Table 3 for exact wording of each study’s definition of a proxy rater.

TABLE 3.

Extracted definition of caregivers and instructions given to proxy raters for the 6 selected studies

Study authors/year	N	Definition of proxy	Instructions for proxy raters
Moreale et al. (2017)	46	“provided care at home for physical, emotional and social needs”	“…collected from the caregivers by asking them to describe their relative’s depression state (e.g. “Do you think that your relative doesn’t feel sad, or feels sad, or feels sad all the time …”), coping strategies (e.g. “Do you think that your relative tries to look at the problem objectively and see all sides?”) and anxiety (e.g. “Do you think that he/she feels nervous and restless?).”
Rooney et al. (2013)	41	Co-habiting partner	“…their proxy also completed it ‘‘for the patient’’, based on their observations at home.”
Milbury K. et al. (2019)	20	Family caregiver (eg, spouse, sibling, adult child)	“Caregivers completed the MDASI-BT as they perceived patients’ symptom severity and interference.”
Armstrong et al. (2012)	115	Caregiver “who are primarily involved in their care in the home setting (biological, legal, or functional relationship)”	MD Anderson Symptom Inventory-Brain Tumor (MDASI-BT; completed by caregiver on behalf of patient) Prompt for answering questions adapted from Lobchuk et al.15
Brown et al. (2008)	118	Patient-identified caregiver	Unclear whether it was on behalf of the patient
Jacobs et al. (2014)	45	Patient identified nonprofessional caregiver (spouse, family member, or friend)	“…caregivers were asked to complete the same questionnaire as they perceived the patient would answer the questions. Caregivers could not fill out the questionnaire for patients.”

3.5 |. Instructions for proxy ratings

There was also a wide range in the reporting of instructions that proxy raters were given in their estimation of their patients’ emotional well-being. One study³⁵ did not specify instructions given to proxy raters, despite the authors forming their study rationale around the importance of proxy raters. Other studies referenced that proxy raters completed measures based on “observation”⁴⁰ and two studies suggested that proxies were instructed to complete measures based on what they “perceived” the patient was experiencing^36,41; however, for one of these studies it was unclear whether similar instructions were provided for the scales extracted in our search methods (e.g., CESD). Only two studies^42,43 outlined a detailed script provided to proxy raters. The first asked proxy raters, “Do you think that your relative doesn’t feel sad, or feels sad, or feels sad all the time?” for rating depression and, “Do you think that he/she feels nervous and restless?” for assessment of anxiety.⁴³ The second study⁴² instructed proxy raters, “In your mind’s eye put yourself in the patient’s shoes. Forget yourself,” borrowed from a palliative care study by Lobchuk.¹⁵ See Table 3 for exact wording that participants were provided in providing proxy ratings.

3.6 |. Measure of psychological well-being

Selected studies also used a variety of measures to capture depressive and anxious symptoms from the patient and their respective proxies. Measures included the Beck Depression Inventory and the State-Trait Anxiety Inventory,⁴³ the Personal Health Questionnaire-9 and Structured Clinical Interview for the DSM-IV,⁴⁰ and the Center for Epidemiological Studies-Depression scale.³⁶ Within the expanded criteria that included studies using quality of life measures, two studies^36,37 used the MD Anderson Symptom Inventory-Brain Tumor Module (from which information on the subscales pertaining to emotional well-being were extracted) with another study⁴⁰ using the Functional Assessment of Cancer Therapy—Brain. Also in the category of selected papers that used measures not specific to depression or anxiety, one study³⁵ used the Profile of Mood States and the Symptom Distress Scale. All selected studies administered the same measures (in full; unique to each study) across patients and proxies, save for one study⁴⁰ that administered a clinical interview, which was only given to patients but for which proxies were, “encouraged to participate as they saw fit.”¹

3.7 |. Quantitative data on level of agreement between patient and proxy ratings

We extracted the numerical level of proxy rater agreement with patient reports across two types of comparisons: group comparisons (t-test, chi-squares etc.) versus comparison of continuous ratings (Pearson correlation coefficient, ICC’s etc). These are described in Table 1. As referenced above in our expansion of search criteria during our search, very few selected papers reported on proxy ratings of both anxiety and depression, but in the one paper that included both,⁴³ ratings of depressive symptoms were statistically equivalent to patients, whereas anxiety ratings were not. Further, within the subscales of a commonly used measure of quality of life (MD Anderson Symptom Inventory-Brain Tumor), distress was one of two subscales that was not equivalent between patients and caregivers.⁴² While not the focus of the current review, this manuscript did have the interesting finding that when patients’ performance status was considered, caregiver and patient reports were equivalent when performance status was high but, when performance status was low, caregivers reported significantly worse symptoms than patients.⁴²

Perhaps the most comprehensive proxy assessment of depression uncovered in our search administered a self-report survey (PHQ-9) to both patients and proxy raters, in addition to a follow-up diagnostic interview to determine whether patients met criteria for major depressive disorder.⁴⁰ The authors opted not to compare patients and proxies on their raw scores of the PHQ-9, rather classified patients as meeting the threshold for depression if they scored above the cutoff¹⁰ and included endorsement of one of the cardinal symptoms (anhedonia or depression). With these criteria, proxy ratings agreed with patient ratings 36.6% of cases with proxy raters generally reporting more depressive symptoms in patients than patients reported. When compared to the diagnostic interview, proxy ratings were slightly more accurate (Κ= 0.58) than patients’ PHQ-9 reports of their own depressive symptoms (Κ= 0.47). Interestingly, within depressive symptom categories rated by proxies, appetite change, and guilt had the highest ICC’s from proxy ratings (0.69 and 0.66, respectively), whereas poor concentration had the lowest level of agreement (ICC = 0.23). The authors concluded that proxy raters were more accurate than patients’ own report in objective symptoms (sleep change, appetite change, fatigue and psychomotor change) compared to subjective symptoms (low mood, anhedonia, guilt).⁴⁰ This study did not assess anxiety.

Studies that compared continuous measures of proxy and patient ratings reported moderate levels of agreement between patient and proxy ratings. More specifically, there was moderate agreement across measures of depression,^40,43 mood and distress,³⁵ and emotional well-being,⁴¹ One of these studies³⁶ did not list specific ICCs associated with each subscale (only a range was provided) within the larger measure of quality of life, thus we were unable to extract which scales corresponded to the range of ICCs reported.

3.8 |. Risk of bias and assessment of quality

Based on established criteria,³² most studies met criteria suggesting low risk of bias and high methodologic quality. For the question, “Was the sample size adequate?” criteria for other dyadic studies,³³ suggested that sample sizes under 50 may impact differing levels of agreement and thus only two of the selected studies^35,42 met this criteria for adequate sample size. See Table 2.

4 |. DISCUSSION

This study aimed to systematically review studies in neuro-oncology that measured proxy-reports of patients’ well-being; specifically, anxiety and depression. Given that patients with brain cancer often have difficulty with expressive language or other communication challenges, there has been significant attention to whether proxy reports are accurate and reliable; however, very little of this work has focused on proxy ratings of patient mood, such as measures of anxiety or depression. Overall, in our systematic review, few studies targeted this direct question, nor did they consistently report quantitative outcomes across domains of depression and anxiety and, instead, gathered proxy ratings of quality of life more generally. As such, we broadened our search slightly to include studies that compared patient and proxy emotional well-being within a larger measure of quality of life. Even with this broader criterion, comparing quantitative outcomes as we originally intended was limited by the unclear definition of proxy ratings, heterogeneity in measures of depression or anxiety and wide range of statistical test reported. Taken together, the findings were mixed in terms of whether patient and proxy ratings were equivalent, and the level of agreement was also mixed, with the suggestion of a moderate level of agreement for several of the selected studies.^35,36,40,41

Overall, the findings from this review suggest that investigators in neuro-oncology should be cautious when considering using proxy raters. While few of the selected studies included measures of both depression and anxiety, in the one study that assessed equivalence, depression was deemed equivalent, whereas anxiety was not.⁴³ One possibility is that symptoms of depression are more observable from an outsider’s perspective than symptoms of anxiety. Upon closer inspection of the selected article by Rooney and colleagues, specific depressive symptoms were compared between proxy raters and patients, demonstrating that objective symptoms (sleep, appetite, fatigue and psychomotor changes) were more accurate than subjective symptoms (mood, anhedonia, guilt).⁴⁰ These findings are consistent with reviews examining the agreement of proxy ratings in pediatric oncology, in which proxy raters can more accurately rate children’s observable functioning than domains that are nonobservable, like emotional functioning,⁴⁴ as well as patients outside of oncology with expressive aphasia.⁴⁵ Given that mood is generally considered less observable than functional domains, this may be one explanation why there was a paucity of studies targeting anxiety and depression in our review.

An additional characteristic of studies that may impact the agreement between patient and proxy ratings, is whether the patient is cognitively impaired. While this was not a focus of the extracted metrics in our review, it was part of the rationale for examining proxy rater agreement in neuro-oncology.^11,12 Across the selected studies in our systematic review, the findings were mixed regarding the impact, or measuring, of neurocognitive impairment. While three of the studies in our review^36,41,43 did not measure cognitive impairment, one study³⁵ suggested that there was stronger agreement between proxy and patient ratings of mood when the patient was did not meet the study’s criteria for cognitive impairment. This parallels other work which suggests greater agreement in proxy and patient ratings of quality life in the absence of patient cognitive impairment.⁴⁶ Conversely, two other selected studies in our review did not detect any differences in agreement in symptom ratings when cognitive impairment was identified,³⁹ nor in measures of depression.⁴⁰ This is consistent with recent work demonstrating similar levels of agreement in patient and proxy measures of quality of life.⁴⁷ These mixed findings—both in our systematic review and across other comparisons of proxy and patient agreement that consider neurocognitive impairment—highlight the multitude of factors that impact how one can perceive the subjective experience of another individual. Administering a neuropsychological battery (or screener), in addition to employing validated measures of anxiety and depression from patients and their proxies is time consuming and thus not be feasible, yet future work should attempt to account for patients’ cognition given its potential impact on proxy and patient agreement.

One unanticipated result from our review was the challenge of clarifying who the proxy rater was across each study. We specified that proxy raters were not health care professionals, but this broad definition did not account for potential differences across whether proxies were a spouse, a friend, or a parent. Definitions in the selected studies ranged from cohabitating partner,⁴⁰ patient-identified caregiver,³⁵ spouses, family members or adult children,³⁶ friend⁴¹ or someone providing care at home,⁴² or, caring care for physical, emotional and social needs.⁴³ Of course, these distinct relationships may bias proxy ratings in different ways, particularly for subjective domains. Such heterogeneity in reporting makes studies difficult to reproduce.

An additional finding from our study was the heterogeneity in reporting how instructions were given to proxy raters during each investigation. In general, much of the work on assessing the agreement of proxy ratings comes from quality of life and existing frameworks underscore that the “gap” between patient and proxy reports may be explained, in part, by the perspective from which the proxy was asked to rate the patients’ symptoms. This can include asking proxies to estimate how they think the patient would respond (termed proxy-patient perspective) or ask how they think the patient would respond (termed proxy-proxy perspective),⁴⁸ Prior work has suggested that providing a prompt for proxy raters to imagine themselves as the patient increases the agreement of proxy ratings.¹⁵ This may be particularly salient in emotional domains, like depression and anxiety, which can shift from day to day. Future research may benefit from incorporation of repeated assessments over time, such as ecological momentary assessment in which brief measures are administered repeatedly and which appears feasible in cancer patients.⁴⁹

4.1 |. Study limitations

There are several limitations from this systematic review worth noting. First, there were very few studies that met the original study criteria and only two additional studies met the expanded criteria that included proxy’s assessment of emotional well-being within a more general measurement of emotional well-being. As noted above, there was significant variability in what quantitative outcomes we were able to extract, further limiting comparisons within our group of selected articles. Thus, our original goal to compare proxy agreement across studies quantitatively may be premature. An additional weakness—also contributing to our inability to make quantitative comparisons among studies - is that we identified additional variability in the definition of a proxy rating in the selected articles. In hindsight, we had only specified that a non-healthcare worker (family, friend etc.) could serve as a proxy rater and this likely was not specific enough of a definition. Finally, there was a wide range of sample sizes in our selected studies (N’s from 20 to 118) and other reviews of proxy ratings of health related quality of life highlight that the agreement between patients and proxies is much lower in sample sizes less than 50.³³ This remains a significant limitation of our review, in which the majority of our selected studies (4 out of 6 of the studies) fell below this sample size cutoff.

4.2 |. Clinical implications

The results from this systematic review presented herein have many clinical implications. In a clinical setting, we will often turn to family members for collateral information by proxy reports about how patients are doing, sometimes delving into emotional domains such as anxiety and depression. As we’ve outlined, this is more likely in neuro-oncology in particular, where patients may have unique challenges to completing self-report measures. Our results suggest that while these are not always accurate, there is likely more agreement in domains of depression than anxiety. In our search, it also became apparent that there is a much more robust literature base on proxy measures of quality of life, which is not always as clinically actionable as anxiety or depression. If a clinician plans to use proxy ratings, they also should pay special attention to how they instruct the proxy to rate the patients’ well-being (e.g., one protocol suggests, “…put yourself in the patient’s shoes.” [15]). Further consideration of the relationship of the proxy rater to the patient may also be particularly important, as we found significant variability in how proxies were defined or the level of detail in reporting who the proxy was for each study. For patients, they may easily hand off distress screening or other assessments of emotional domains to their caregivers as proxy raters, but our results are a reminder that the most accurate data is from patients themselves. Future work should investigate how different proxy raters provide accurate or inaccurate assessments of mood, how proxy instructions provided and, eventually, develop and validate measurement tools specific to proxy raters in neuro-oncology.

The systematic review described herein uncovered important details about proxy ratings for depressive and anxious symptoms in neuro-oncology, suggesting that findings are mixed regarding the agreement of proxy ratings. Comparisons between studies proved difficult, given differences in measures used, clarity around instructions given to proxy raters and the statistics reported. Future studies should consult the extant literature to identify clinically relevant measures of anxiety or depression (the GAD-7 or PHQ-9, for example) or, alternatively, choose scales that have been promoted to increase consistency in research studies.⁵⁰ Accurately identifying how and when proxy ratings are appropriate will greatly contribute to our understanding of the emotional impact of brain cancer.

DATA AVAILABILITY STATEMENT

The data extracted during the course of this review is available upon reasonable request.