Factors Associated With Perceived Cognitive Dysfunction in Patients With Primary Brain Tumors: The Utility of the Neuro‐QoL Cognition Function Scale

ABSTRACT

Background

Primary brain tumor (PBT) patients commonly experience cognitive dysfunction. This study aims to describe the prevalence and factors associated with perceived cognitive function (PCF) in PBT patients enrolled in a natural history study (NCT02851706: Principal Investigator T. S. Armstrong).

Methods

The Neuro‐QoL Cognition Function was used to measure PCF with T‐scores < 40 categorized as cognitively impaired. Clinical and demographic characteristics, patient‐reported outcomes (PROs: symptom burden, depression, anxiety), and KPS were collected and characterized using descriptive statistics. Univariate tests on PCF with p‐value < 0.10 were performed and included in a subsequent multivariate logistic regression with backward selection. Clinical variables and PROs were evaluated separately, then combined for a final multivariate model with p‐value < 0.05.

Results

The cohort included 327 patients (median age: 47 years [range: 18–85], primarily white [78%] males [55%] with high‐grade tumors [73%]). Perceived cognitive impairment occurred in 29% of patients. Patients presenting with cognitive dysfunction at diagnosis, or who had undergone multiple surgeries, were more likely to have poor PCF, as were those currently on corticosteroids and reporting cognitive symptoms and symptom interference. Evaluated together, a multivariate model found that only higher scores on symptom interference and the cognitive symptom factor increased the likelihood of cognitive impairment (symptom interference odds ratio [OR] = 1.25, 95% confidence interval [CI] [1.09, 1.44], p < 0.001; cognitive symptom factor OR = 1.81, 95% CI [1.50, 2.19], p < 0.001) with an 84% classification rate.

Conclusions

Self‐reported cognitive symptoms and interference were the strongest patient‐reported correlates of PCF and should be interpreted as concurrent associations among related constructs. Further psychometric and longitudinal work is needed to clarify the incremental clinical value of the Neuro‐QoL Cognition Function Short Form in PBT care and research.

1. Introduction

An estimated 1 million Americans are living with a primary brain tumor (PBT) diagnosis [1]. This diagnosis often follows the development of neurologic and cognitive symptoms, many of which persist or may worsen throughout the disease course [2, 3]. Cognitive dysfunction is common in the PBT population and is associated with heterogeneity in both the course and severity and is impacted by the tumor, antitumor therapies, and concomitant medications [4]. Cognitive dysfunction can impact patients' achievement of social milestones, educational and career goals, ability to care for themselves and others, and participation in daily activities [5, 6]. Understanding the impact and prevalence of cognitive symptom burden and dysfunction in PBT patients is important for improving quality of life and creating targeted treatment methods and supportive care interventions [7, 8].

Neuropsychological assessment is a laborious procedure often completed by a trained neuropsychologist, but computer‐based screening has been used to show significant deficits in various neurocognitive domains prior to the initiation of treatment in brain tumor patients [9]. Patients also often observe cognitive symptoms before they are detected by performance‐based tests [10]. Considering this, clinicians and researchers are interested in brief, reliable, effective, and precise self‐report measures of cognitive dysfunction [11]. Patient‐reported outcomes (PRO) are valuable tools to evaluate experience, symptoms, and quality of life in PBT patients [10]. One such measure is the Quality of Life in Neurological Disorders (Neuro‐QoL) Item Bank v2.0‐Cognition Function‐Short Form, which is a tool designed to measure a patient's subjective report of cognitive dysfunction. Neuro‐QoL is a system of PRO measures designed to evaluate and monitor the physical, mental, and social effects experienced by people living with neurological conditions [12, 13, 14]. The psychometric properties and usefulness of the Neuro‐QoL Cognition Function measure have been evaluated in a variety of clinical populations, including those with multiple sclerosis [15, 16], epilepsy [17], Huntington's disease [18, 19], and Parkinson's disease [20], but it has not yet been validated in the PBT population, where other neuropsychological measures assessing various domains are often used [21].

Research on the Neuro‐QoL among PBT patients has been limited, with its use in only a few studies as a secondary end point or supporting measure. One study used the Neuro‐QoL as a supporting measure for studying health‐related quality of life in adult low‐ and high‐grade glioma patients [22], while another used it to describe cognitive dysfunction in long‐term survivors of central nervous system (CNS) tumors [23]. Research has shown that PBT patients may experience deficits in memory, processing speed, executive function, and learning [24]. Conceptually, the Neuro‐QoL Cognition Function tool provides an assessment of these specific factors, as well as decision making, attention, and the application of such abilities to everyday tasks (e.g., planning, organizing, calculating, remembering, learning). Studies support that, in both PBT patients and other cancer populations, cognitive dysfunction may be more common or pronounced in older patients [25, 26], those with fewer years of education [26], those with a tumor located in the frontal/parietal lobe [24, 27], and those with more treatments [4, 9, 28, 29, 30, 31, 32, 33], specifically more radiation treatments [34, 35, 36, 37]. In addition to cognitive dysfunction, patients often report other symptoms caused by their disease or treatment, such as mood disturbances, sleep disturbances, seizures, fatigue, gastrointestinal symptoms, other treatment and disease‐related symptoms, and interference of these symptoms with daily life [38]. We expected that those with a greater symptom burden and interference would report more cognitive dysfunction. Further, this increased symptom burden is often associated with poorer performance status, indicating that Karnofsky Performance Status (KPS) could be a factor associated with more severe cognitive dysfunction [39]. Additionally, we predicted that those who report more moderate‐to‐severe mood disturbance symptoms, would make up a larger proportion of the patients reporting impaired cognitive function [5, 40, 41, 42].

A recent systematic review [43] acknowledged that while the majority of studies have focused on clinical factors, minimal research has evaluated the role of demographic factors with perceived cognitive function (PCF) in large, diverse samples of PBT patients or that predispose patients to experience cognitive symptoms. This will also be one of the first studies using the Neuro‐QoL measure in this population and evaluating the relationship between demographic and clinical factors, and other clinical outcomes with cognitive impairment. This study aims to assess the prevalence of PCF in PBT patients enrolled in a natural history study, using the Neuro‐QoL subjective measure of cognition. Finally, taking all these factors into account, we developed association models to characterize factors related to perceived cognitive impairment rather than to identify causal predictors.

2. Materials and Methods

2.1. Patient Sample

This retrospective study consisted of a cohort of 327 PBT patients who completed the Neuro‐QoL at their study entry visit for the Neuro‐Oncology Branch's Natural History Study (NHS; NCT02851706: Principal Investigator T.S. Armstrong) at the National Institutes of Health in Bethesda, Maryland between February 28th, 2019 and January 27th, 2023. The NHS is an IRB‐approved observational study designed to longitudinally follow patients diagnosed with a primary CNS tumor from the time of study entry throughout their disease course, with PROs completed at each clinical visit. Written informed consent was obtained from all participants enrolled in the study. Patients were eligible if they were > 18 years of age, diagnosed with a PBT, and had an in person or telehealth study entry consult with completed PROs. A detailed description of broader study aims and inclusion criteria are reported in clinicaltrials.gov.

2.2. Sample Characteristics, Perceived Cognitive Function, and Other Patient‐Reported Outcome Measures

Demographic, clinical, and treatment information were collected directly by the clinical provider and extracted from medical records and study intake forms. KPS was part of the clinical evaluation performed by trained clinical staff during a patient's clinic visit as part of the physical examination. PROs, including measures of symptom burden and interference and emotional distress, were collected via an electronic data capture system prior to the patient being seen in the clinic and were considered valid if they were completed up to a week prior to a patient's clinical evaluation.

Perceived cognitive function was assessed using the Quality of Life in Neurological Disorders (Neuro‐QoL) Item Bank v2.0‐Cognition Function‐Short Form. This eight‐item, self‐report tool was developed specifically for those with neurologic illness and includes four questions related to specific cognitive symptoms and four questions involving a patient's perceived ability to complete daily tasks over the last 7 days [14, 44]. Respondents use a five‐point Likert scale (1–5; higher scores indicate better cognition; options: 5‐never, 4‐rarely [once], 3‐sometimes [2–3 times], 2‐often [once a day], 1‐very often [several times a day]) to rate how often they had to read something several times to understand it, their thinking was slow, they had to work hard to pay attention or they would make a mistake, and they had trouble concentrating. A similar Likert scale (1–5; higher scores indicate less difficulty; options: 5‐none, 4‐a little, 3‐somewhat, 2‐a lot, 1‐cannot do) was used to assess ability to perform daily tasks, with respondents rating current level of difficulty while reading and following complex instructions, planning for and keeping appointments, managing their time, and learning new tasks or instructions. Individual responses were summed and converted to a T‐score metric with a mean of 50 and standard deviation (SD) of 10. Scores one SD below the general population mean (T‐scores < 40) indicate perceived cognitive impairment. A comprehensive overview of the Neuro‐QoL item bank development and initial validation can be found in the National Institute of Neurological Disorders and Stroke User Manual for the Quality of Life in Neurological Disorders Measures, Version 2.0, March 2015 [44] or on the Health Measures website [14].

The MD Anderson Symptom Inventory‐Brain Tumor (MDASI‐BT) module is a validated, self‐report instrument measuring symptom burden and inference caused by symptoms (activity‐related and mood‐related interference) using a numeric scale (0–10; higher score indicates worse severity) and assessing severity within the past 24 h [45]. This 29‐item scale (23 symptoms and 6 interference items) measures six symptom factor subgroupings; affective, cognitive, neurologic, treatment‐related, general disease, and gastrointestinal. Symptoms rated ≥ 5 or an interference score of ≥ 2 are considered moderate‐to‐severe, with a 1‐point change indicating a minimum clinically important difference. For this study, overall symptom burden, cognitive symptom factor, and overall interference scores were selected a priori from the various MDASI‐BT symptom factor subgroupings and were used in the analysis. Because the MDASI‐BT cognitive factor and interference scale include cognition‐related or functionally proximal content, these variables were interpreted as patient‐reported correlates of Neuro‐QoL cognitive impairment rather than as independent etiologic predictors. Although there is an affective factor in the MDASI‐BT, we chose to use only the Patient‐Reported Outcomes Measurement Information System PROMIS Anxiety and Depression measures to control Type 1 error rates. While highly correlated with the MDASI‐BT affective subscale, the PROMIS Anxiety and Depression scales were specifically developed and focused for these constructs.

The Patient‐Reported Outcomes Measurement Information System (PROMIS) self‐reported item bank v1.0‐Emotional Distress‐Anxiety Short Form 8a and the v1.0‐Emotional Distress‐Depression Short Form 8a are validated, eight‐item instruments used to assess symptoms of anxiety and depression, respectively. These measures use a Likert scale (1–5; higher scores indicate worse severity) with values converted to T‐scores, with scores one standard deviation above the general population mean (T‐scores > 60) considered moderate‐to‐severe [46]. The percentage of patients reporting moderate‐to‐severe symptoms was reported [47].

2.3. Statistical Analysis

All statistical analyses were performed using IBM SPSS Version 25.0.41 [48].

2.3.1. Descriptive Data

Descriptive statistics characterized the patient sample (Table 1) and provided PRO summary scores. These variables were used in further analyses to investigate their association with patient perceived cognitive dysfunction as reported by the Neuro‐QoL. Tumor grade was classified as either high (Grade 3 and 4) or low (Grade 1 and 2). KPS was dichotomized as either “Good” (≥ 90) or “Poor” (< 90) based on previous work identifying that a KPS of ≤ 80 is associated with higher symptom burden, increased interference with daily activities, and disease recurrence [49]. PRO scores were described as moderate to severe based on the established cutoffs on how they interfere with daily functioning. The MDASI‐BT was completed, and mean scores were calculated for overall symptom burden, cognitive symptom factor, and overall symptom interference for this analysis (Table 2). Additionally, PROMIS and Neuro‐QoL mean scores and severity levels were reported in Tables 2 and 3.

TABLE 1.

Patient demographics and clinical characteristics (N = 327).

Characteristic	Results
Age at assessment	Median (range)	47 (18–85)
	Mean (SD)	47 (14)
Age at diagnosis	Median (range)	43 (5–79)
	Mean (SD)	44 (15)
		N (%)
Sex	Female	147 (45)
	Male	180 (55)
Race	Asian	29 (9)
	Black/African American	22 (6)
	White	254 (78)
	Native Hawaiian/Pacific Islander	2 (1)
	Other	8 (2)
	Unknown	12 (4)
Ethnicity	Hispanic or Latino	33 (10)
	Not Hispanic or Latino	286 (88)
	Unknown	8 (2)
Highest education level	Did not complete high school	8 (2)
	High school/GED	39 (12)
	Associates degree/Any college	58 (18)
	Bachelor's degree	113 (35)
	Advanced degree	109 (33)
Marital status	Married/Living with a partner	210 (64)
	Single	77 (24)
	Divorced/Separated	24 (7)
	Widowed	7 (2)
	Unknown	9 (3)
Employment status	Employed	157 (48)
	Retired	52 (16)
	On disability	1 (1)
	On medical leave/FMLA	30 (9)
	Student/Volunteer	6 (2)
	Unemployed prior to tumor diagnosis	18 (5)
	Unemployed due to tumor diagnosis	63 (19)
Income	Prefer not to answer	31 (9)
	< $25,000	29 (9)
	$25,000–$49,999	42 (13)
	$50,000–$149,999	128 (39)
	More than $150,000	84 (26)
	Missing	13 (4)
Current diagnosis	Astrocytoma	52 (16)
	Glioblastoma	118 (36)
	Oligodendroglioma	33 (10)
	No tissue diagnosis	12 (4)
	Other	112 (34)
Current tumor grade	Low grade (1/2)	65 (20)
	High grade (3/4)	238 (72)
	No tissue diagnosis	12 (4)
	No grade assigned	12 (4)
Tumor region	Supratentorial	282 (86)
	Infratentorial	40 (12)
	Both	5 (2)
Time with symptoms prior to diagnosis	No symptoms	24 (8)
	Less than 6 months	197 (60)
	6 months to 1 year	36 (11)
	More than 1 year	30 (9)
	Don't know	23 (7)
	Did not answer/Survey not submitted	17 (5)
KPS	≥ 90	212 (65)
	≤ 80 (40–80)	98 (30)
	Missing	17 (5)
Treatment phase at visit	Treatment	75 (23)
	Surveillance	252 (77)
Progression status at study entry	No progression	244 (75)
	Progression	83 (25)
Years from diagnosis	Median (range)	1 (0–31)
Corticosteroid use	Yes	82 (25)
Psychotropic medication use	Yes	92 (28)
Supplement use	Yes	131 (40)
Anticonvulsant use	Yes	202 (62)
Number of surgeries	0	11 (3)
	1	196 (60)
	2	79 (24)
	≥ 3	41 (11)
Number of radiation treatments	0	89 (27)
	1	192 (59)
	≥ 2	46 (14)
Number of treatments	0	166 (51)
	1	87 (27)
	2	40 (12)
	≥ 3	34 (10)
Number of recurrences	0	183 (56)
	1	72 (22)
	2	29 (9)
	≥ 3	43 (13)
IDH status	Mutant	62 (19)
	Wildtype	162 (50)
	Not recorded/Available after study entry	103 (31)
Original lobe side	Left	128 (39)
	Right	150 (46)
	Midline	19 (6)
	More than one area	14 (4)
	Missing	16 (5)
Original lobe	Frontal	97 (30)
	Parietal	44 (13)
	Temporal	71 (22)
	Cerebellum	16 (5)
	Occipital	6 (2)
	Brainstem	10 (3)
	Other	29 (9)
	Multiple lobes	47 (14)
	Missing	7 (2)

Note: All categorical variables were tested using chi‐square test. Age at assessment and age at diagnosis were tested using t‐test.

Abbreviations: FMLA, Family and Medical Leave Act; IDH, Isocitrate dehydrogenase; KPS, Karnofsky performance status.

TABLE 2.

Descriptive information for the PROMIS‐anxiety, PROMIS‐depression, and MDASI‐BT measures.

Patient‐reported outcome measure	T‐scores
	Median (Mean)	SD	Min	Max	Percentages none/mild % (N)	Percentages perceived cognitive impairment %(N)
Anxiety	52.28 (52.31)	9.40	37.10	75.70	79 (258)	21 (69)
Depression	50.80 (50.94)	9.14	38.20	74.90	84 (275)	16 (52)
MDASI‐BT overall symptom burden	1.50 (1.99)	1.68	0	8.40	93 (304)	7 (23)
MDASI‐BT overall interference factor	2.33 (3.03)	2.75	0	10	73 (239)	27 (88)
MDASI‐BT cognitive symptom factor	1.25 (2.23)	2.39	0	10	85 (278)	15 (49)

Abbreviations: MDASI‐BT, MD Anderson symptom inventory‐brain tumor; PROMIS, Patient‐Reported Outcomes Measurement Information System.

TABLE 3.

Descriptive information for overall Neuro‐QoL Cognition Function measure and individual item bank questions (N = 327).

Measures	Median (Mean)	SD	Min	Max	Percentages (N) none/mild	Percentages (N) perceived cognitive impairment
Overall Neuro‐QoL T‐scores	45.67 (45.60)	10.05	20.20	64.20	71 (232)	29 (95)

Neuro‐QoL question	Mean score	Standard deviation
I have to read something several times to understand	3.49	1.221
My thinking was slow	3.42	1.255
I had to work really had to pay attention or I would make a mistake	3.50	1.241
I had trouble concentrating	3.38	1.278
Difficulty reading/following complex instructions	3.90	1.194
Difficulty planning/keeping appointments that are not part of your weekly routine	4.09	1.177
Difficulty managing time to do most of your daily activities	3.93	1.100
Difficulty learning new tasks/instructions	3.97	1.091

Abbreviation: Neuro‐QoL, Quality of Life in Neurological Disorders Item Bank v2.0‐Cognition Function‐Short Form scores.

2.3.2. Select Psychometric Properties of Neuro‐QoL

We calculated the Cronbach's alpha for the Neuro‐QoL to establish internal consistency reliability. We also presented the correlation coefficients between the Neuro‐QoL and the corresponding subscales on the MDASI‐BT as evidence of concurrent validity. Test–retest reliability and responsiveness were not evaluated in this cross‐sectional analysis.

2.3.3. Univariate Analysis

We investigated the association between PCF with various clinical and demographic factors (including age at visit, sex, race, marital status, education level, employment status, KPS, anticonvulsant use, corticosteroid use, psychotropic medication use, supplement use, current treatment phase, tumor lobe location, number of lobes involved, tumor side, tumor region, tumor grade, cognitive dysfunction as presenting symptom, time with symptoms prior to diagnosis, radiation treatment, number of treatments, number of surgeries, disease recurrence, progression status at study entry, and Isocitrate Dehydrogenase (IDH) mutation status) and PRO scores (MDASI‐BT overall symptom burden and interference, MDASI‐BT cognitive symptom factor, and PROMIS anxiety and depression T‐scores and severity level) using either a chi‐square test, correlations, or independent samples t‐test. For each univariate test, a significance level of 0.1 instead of the usual 0.05 was used to not inadvertently exclude important variables.

2.3.4. Model Building and Multivariate Analysis

Because of the large number of possible factors, the univariate approach was used as described above and clinical factors and PRO scores that were significantly associated with cognitive impairment at the p < 0.1 level were included in subsequent multivariate logistic regression analyses. To identify a parsimonious association model, a backwards elimination variable selection method was used with p < 0.05. The first model consists of clinical variables (age, cognitive dysfunction as presenting symptom, number of surgeries, corticosteroid use, anticonvulsant use, psychotropic medication use). The second model involves clinical outcome assessment (COA) variables that include both clinician‐rated measures (KPS) and PROs (overall symptom burden, overall symptom interference, cognitive symptom factor). Because the clinical outcome variables in the second model tend to be more conceptually proximal to the dependent variable (cognitive impairment) than clinical characteristic variables in the first model, we would like to ensure that we do not inadvertently eliminate clinical characteristic variables in the final model. The final model was composed of variables retained from the first two models and was interpreted as an association model, not a causal prediction model. Odds ratios and 95% confidence intervals were reported. We also tested for multicollinearity in the multivariate models.

With a sample size of 327 patients, we can detect a correlation of at least 0.2 with 95% power using a two‐tailed test at 5% significance level for each of our univariate tests, suggesting we have adequate statistical power.

3. Results

3.1. Sample Characteristics

Three hundred and twenty‐seven patients enrolled on the NHS had completed the Neuro‐QoL Cognition Function measure at study entry. These patients were predominantly white (78%), males (55%) ranging in age from 18 to 85 (mean = 47, SD = 14). This analysis looked at solely brain tumor patients and a majority had good KPS (65%), high‐grade tumors (73%), and underwent one surgery (57%). Approximately 73% of patients received radiation and 44% experienced a tumor recurrence prior to the study entry time‐point. Additionally, 26% had documented clinician‐reported cognitive dysfunction as a presenting symptom at study entry. Additional demographic and clinical characteristics of the sample are listed in Table 1.

3.2. Perceived Cognitive Dysfunction Prevalence and Severity

In this sample, 29% of the patients were classified as cognitively impaired, with an overall average Neuro‐QoL Cognition Function score of 45.8 (SD = 10.1; range: 20.2–64.2). Table 3 summarizes all descriptive results of PCF based on the use of the Neuro‐QoL in this cohort.

3.3. Select Psychometric Properties of Neuro‐QoL

Reliability analysis found a Cronbach's alpha of 0.94, which indicates excellent internal consistency reliability. The bivariate correlations of the Neuro‐QoL with each of the MDASI‐BT symptom factors were moderate, with the highest correlation with the cognitive symptom factor (r = −0.646). This inverse relationship indicates that as cognitive dysfunction gets worse on the Neuro‐QoL (lower values), the report of cognitive symptom burden gets worse on the MDASI‐BT (higher values). These results provide internal consistency and concurrent validity evidence in this sample; however, test–retest reliability and responsiveness were not evaluated in this cross‐sectional analysis. There will be a more comprehensive psychometric analysis of the Neuro‐QoL tool in a subsequent paper.

3.4. Patient‐Reported Outcomes

Using PROMIS measures, 16% of the sample reported moderate‐to‐severe depression and 21% reported moderate‐to‐severe anxiety (Table 2). A majority of the sample reported none‐to‐mild overall symptom burden on the MDASI‐BT (93%), 73% reported NMnone‐to‐mild symptom interference with daily living, and 15% of the sample reported moderate‐to‐severe cognitive symptoms on the cognitive symptom burden factor (including memory, concentration, difficulty reading, and speaking). Further information on MDASI‐BT, PROMIS‐Anxiety, and PROMIS‐Depression results can be found in Table 2.

3.5. Demographic and Clinical Characteristics and Perceived Cognitive Dysfunction

Considering associations with demographic and clinical factors, it was determined that age at diagnosis, employment status, cognitive dysfunction as presenting symptom, number of surgeries, corticosteroid use, psychotropic medication use, and anticonvulsant use were significant at the univariate level. Patients reporting cognitive impairment were on average 3.87 years older at diagnosis than those reporting none‐to‐mild impairment (95% CI [−7.280, −0.454]). Patient employment status affected their report of cognitive dysfunction, with employed patients less likely to be categorized as cognitively impaired (25% vs. 57%, p < 0.05), and those on medical leave (16% vs. 7%, p < 0.05) or unemployed due to tumor diagnosis (33% vs. 14%, p < 0.05) reporting more cognitive impairment. Forty‐six percent of patients reporting cognitive dysfunction at disease presentation were categorized as cognitively impaired, versus only 23% of those not reporting cognitive dysfunction at presentation (p < 0.001). There were also significant differences in the proportion of patients with cognitive impairment depending on their surgery history. Specifically, patients with no history of surgery are unlikely to be classified as cognitively impaired (0% vs. 5%, p < 0.05). These patients have not had any neurosurgical procedure (e.g., biopsy, resection) because the tumor is in the brainstem or eloquent brain areas where risk outweighs benefit. Eloquent brain areas refer to parts of the brain that control vital function such as speech and motor which are important in the care of these patients. Similarly, patients with two surgeries are more likely to have cognitive impairment (32% vs. 21%, p < 0.05). Taken together, perceived cognitive impairment was seen in 32% of patients with two surgeries compared to 0% of patients with no surgery. Patients taking corticosteroids are more likely to be categorized as cognitively impaired than those not on corticosteroids (34% vs. 24%; p = 0.036). Similarly, patients taking psychotropic medication (35% vs. 24%; p = 0.077) and anticonvulsants (32% vs. 20%; p = 0.027) were more likely to report cognitive impairment than those not taking the respective medications. See Figure 1.

FIGURE 1.

Significant demographic and clinical characteristics predictors of perceived cognitive dysfunction at the univariate level (A–G). Neuro‐QoL cognition function scores are shown in relation to anticonvulsant use (A), corticosteroid use (B), psychotropic medication use (C), cognitive dysfunction as presenting symptom (D), number of surgeries (E), employment (F), and age (G). Participants were divided into two groups according to their Neuro‐QoL cognition function scores: Normal cognitive (score ≥ 40) and perceived cognitive impairment (score < 40). Data represented as mean ± SD (A–F) by group for categorical variables. Data presented as scatterplot (G) for continuous variable with a point biserial R of 0.123. Neuro‐QoL, Quality of Life in Neurological Disorders Item Bank v2.0‐Cognition Function‐Short Form scores.

3.6. Clinical Outcome Assessments and Perceived Cognitive Dysfunction

Considering associations with PROs and clinician‐rated outcomes, it was determined that KPS, anxiety, depression, overall symptom burden, overall symptom interference, and cognitive symptom burden factor were significantly associated with reporting cognitive impairment. Fifty‐two percent of patients with a poor KPS had cognitive impairment versus only 17% of those with a good KPS (p < 0.001). Those classified as cognitively impaired also reported worse anxiety and depression. On average, patients reported an anxiety score six points higher (50.5 vs. 56.6, p < 0.001), and depression score eight points higher (45.7 vs. 56.5, p < 0.001), than those reporting none‐to‐mild deficits. Patients classified as cognitively impaired had an overall symptom burden score two points higher than those reporting none‐to‐mild (3.2 vs. 1.5, p < 0.001), an overall symptom interference three points higher than the none‐to‐mild (5.2 vs. 2.1, p < 0.001), and a cognitive symptom factor score three points higher than those reporting none‐to‐mild (4.5 vs. 1.3, p < 0.001). See Figure 2.

FIGURE 2.

Significant clinical outcomes assessment predictors of perceived cognitive dysfunction at the univariate level. Correlations between Neuro‐QoL cognitive function and MDASI‐BT cognitive symptom factor score (A), MDASI‐BT overall symptom factor score (B), MDASI‐BT interference score (C), KPS index (D), PROMIS self‐reported item bank v1.0‐Emotional Distress‐Depression Short Form 8a (E) and self‐reported item bank v1.0‐Emotional Distress‐Anxiety Short Form 8a (F) are shown. Data presented as scatterplot (A–F) labeled by perceived cognition. Abbreviations: Neuro‐QoL, Quality of Life in Neurological Disorders Item Bank v2.0‐Cognition Function‐Short Form scores; MDASI‐BT, MD Anderson Symptom Inventory‐Brain Tumor; KPS, Karnofsky Performance Status; PROMIS, Patient‐Reported Outcomes Measurement Information System.

3.7. Factors Associated With Self‐Reported Cognitive Impairment

Multivariate logistic regressions were performed using variables that were correlated with cognitive dysfunction based on the univariate analysis. In this approach, the model requires patients with non‐missing responses on all the variables included from the univariate analyses. Of the 327 patients, 57 were missing data for at least one of the variables used in the final model and were not included, reducing the sample to 270. Of note, further analysis showed no significant difference in population clinical and demographic characteristics when compared to the full sample. Based on the univariate approach results, age at diagnosis, cognitive dysfunction as a presenting symptom, number of surgeries, corticosteroid use, anticonvulsant use, and psychotropic medication use were entered into a multivariate logistic regression clinical model. This model identified cognitive dysfunction at presentation (OR = 3.15, 95% CI [1.70–5.84]), corticosteroid use (OR = 2.05, 95% CI [1.11–3.79]), and number of surgeries (OR = 2.26, 95% CI [1.17–4.36]) as variables significantly associated with cognitive impairment at p < 0.05.

In the multivariate logistic regression COA model, KPS, overall symptom interference, and cognitive symptom factor were entered into the model. This model found overall interference (OR = 1.33, 95% CI [1.17–1.52]) and the cognitive symptom factor (OR = 1.78, 95% CI [1.50–2.12]) as significant variables associated with cognitive impairment. According to this model, patients are 1.33 times as likely to have cognitive impairment if their overall symptom interference score is increasing by one point on a 0–10 scale. Furthermore, a one‐point increase on a 0–10 scale in a patient's cognitive symptom factor score is associated with patients being 1.78 times as likely to report cognitive impairment. Additional results are presented in Table 4.

TABLE 4.

Clinical and clinical outcome assessment factors of cognitive impairment (N = 270).

Model	Factor	Odds ratio	95% CI	p
Clinical model a	Cognitive dysfunction at presentation	3.15	1.70–5.84	< 0.001
	Surgeries	2.26	1.17–4.36	0.015
	Corticosteroid use	2.05	1.11–3.79	0.021
COA (PROs/Clinician‐rated outcomes) model b	Symptom interference	1.33	1.17–1.52	< 0.001
	Cognitive symptom factor	1.78	1.50–2.12	< 0.001
Combined model c	Symptom interference	1.25	1.09–1.44	0.001
	Cognitive symptom factor	1.81	1.50–2.19	< 0.001

Abbreviations: COA, clinical outcome assessment; PRO, patient‐reported outcomes.

^a Age at diagnosis, anticonvulsant use, and psychotropic medication use were included but were not significant.

^b KPS was included but was not significant.

^c Corticosteroid use, cognitive dysfunction at presentation, and number of surgeries were no longer significant.

The variables significant in each of the separate models were entered into a final model, resulting in the MDASI‐BT overall symptom interference (OR = 1.25, 95% CI [1.09–1.44]) and MDASI‐BT cognitive symptom factor (OR = 1.81, 95% CI [1.50–2.19]) being associated with cognitive impairment, with 84% correct classification (Table 4). Patients are 1.25 times more likely to be categorized as having cognitive impairment if the overall symptom interference score is increasing by one point on a 0–10 scale and 1.81 times more likely to have this report if their cognitive symptom factor score increases by one point on a 0–10 scale. Because these MDASI‐BT variables are conceptually proximal to the Neuro‐QoL Cognition Function outcome, they should be interpreted as concurrent patient‐reported correlates rather than independent etiologic predictors.

Multicollinearity among factor variables was tested for and was found between the MDASI‐BT overall symptom burden factor and cognitive symptom factor. After running the model separately with each variable, the cognitive symptom factor model was used as it correctly classified a higher percentage of cognitive impairment (61.4% vs. 54.5% and overall predicted 84.2% vs. 81.3%). Furthermore, despite the significant association of depression and anxiety T‐scores with cognitive dysfunction, these variables did not go into the model due to their strong correlation with MDASI‐BT symptom factor scores. Because of unstable results when both anxiety and depression are included in the multivariate logistic regression, psychotropic medication use was used as a proxy variable to represent anxiety and depression, as these medications are often used in this population to manage these symptoms. Employment status was significant at the univariate level, but also not included in the logistic regression as changes in employment may be viewed as a consequence of issues with cognition alone or in combination with them having a terminal disease or other symptoms [50]. Therefore, its' association with cognition could be described as an outcome of cognitive dysfunction rather than a factor. Finally, IDH status was not investigated because the sample included many diagnoses other than glioma and therefore was not relevant.

4. Discussion

The objective of this study was to describe the experience of cognitive dysfunction in PBT patients and examine factors associated with Neuro‐QoL Cognition Function Short Form scores. Nearly a third of the patients in this study were categorized by their scores as having cognitive impairment, highlighting the frequency and clinical relevance of perceived cognitive concerns in adults with PBT. It should be noted that our sample is a referral‐based convenience sample seen at the US National Cancer Institute. Referral patterns may differ by tumor rarity, disease stage, socioeconomic factors, neighborhood disadvantage, distance from NIH, length of time from diagnosis or symptom onset, and treatment access; these variables were not fully accounted for in the present cross‐sectional analysis. Recent work from the NIH Natural History Study cohort showed associations between neighborhood disadvantage, distance from NIH, and treatment‐access outcomes, especially among low‐grade PBT patients. High internal consistency and concurrent validity evidence in this sample support further evaluation of the Neuro‐QoL Cognition Function Short Form in clinical care and research. Additional psychometric validation, including evaluation of test–retest, sensitivity to change over time, and the relationship to performance‐based neurocognitive testing, is needed.

In this study, we explored the association of cognitive dysfunction with clinical and demographic factors. Age at diagnosis, employment status, cognitive dysfunction at the time of initial diagnosis, number of surgeries, corticosteroid use, psychotropic medication use, and anticonvulsant use were associated with the occurrence at the univariate level, with logistic regression modeling identifying that having cognitive dysfunction at the time of initial diagnosis, corticosteroid use, and number of surgeries are associated with a greater than two‐fold risk of cognitive dysfunction. The association between cognitive symptoms at presentation and later patient‐reported cognitive dysfunction may reflect persistence of tumor‐ or treatment‐related symptoms, preexisting neurologic vulnerability, psychological distress, or other comorbidities; causal direction cannot be determined in this cross‐sectional analysis [3, 41]. Additionally, recent novel studies have demonstrated that the occurrence and degree of connectivity between neurons and glioma cells results in poorer survival and cognitive dysfunction, specifically related to performance in language tasks, supporting that remodeled neural circuits may underlie cognitive dysfunction occurring prior to and continuing during the course of the disease [51].

Corticosteroids and an increased number of surgeries are often used for patients who have significant edema, larger tumors, or other neurological deficits, which may partly explain the association with cognitive dysfunction [52]. Corticosteroids have also been independently associated with cognitive effects in diverse patient populations, including negatively affecting executive function and both short and long‐term memory [53]. Evaluation of cognition prior to, during, and after corticosteroid use would be needed to untangle these possible associations. The finding that the number of surgical resections is associated with cognitive dysfunction aligns with previous work that shows varying impact of surgical treatment on cognition, with executive functioning being the chief concern postoperatively [4]. Numerous studies have actually found a rate of improvement following surgery [9, 31], while some have found select groups of patients that still show decline [31], and others, such as a study assessing surgical effects in patients with insular tumors, show significant postoperative decline in multiple domains [33, 54]. Patients with resection of tumors in the temporal lobe also showed deficits after surgery, notably in areas of verbal memory and executive functioning [32]. The present study found that patients with multiple surgeries reported worse cognitive dysfunction than their counterparts with one or no surgery, which may be related to longer time since diagnosis, more malignant tumor biology, recurrent disease, and potential impact of the surgery itself on cognitive dysfunction. It should be noted that the diagnosis length of patients in this study is highly variable. Further analysis of our own data also suggests that patients with multiple surgeries have had multiple recurrences.

We recognize that some of the clinical factors that were significant at the univariate level did not remain in the model and may require further evaluation in longitudinal cohorts, in those with specific tumor types, or more aged populations. Future longitudinal analyses should examine age, tumor type, tumor location, seizure burden, and medication characteristics within more homogeneous cohorts. Notably, previous work indicates that both self‐report and performance‐based neurocognitive testing may be affected by age [55]. Although our sample includes individuals up to 85 years of age, overall, the sample is younger with a mean age of 47. Additionally, anticonvulsant medications are used to treat patients experiencing seizures and have been associated with cognitive side effects [56], but despite 62% reporting use in this study, there is variation in type, dose, and duration of therapy.

Based on previous literature, we expected associations between tumor grade, tumor location, education, and radiation therapy (RT) with worse cognitive dysfunction, but the present study showed no indication of such associations. Using intelligence and memory tests, Yamawaki et al. [57] found that patients with Grade 4 tumors had worse functional neurocognitive decline compared to their Grade 2 and 3 counterparts. Another study found that patients with higher grade tumors had lower processing speed, less cognitive flexibility, and delayed memory [27]. In this study, tumor grade was dichotomized to high (3/4) and low (1/2), with the majority of patients (72%) having high‐grade tumors, potentially removing differences between individual tumor grades. Research has also shown different cognitive deficits based on tumor location [32, 58]. One study found that lateralization in the frontal lobe affected symptom presentation, with impairments consistent with tumor location and associated brain functionality [59]. Another study found a similar trend in specific deficits based on tumor location in the left versus right hemisphere, but it is unclear if this is due to the tumor residing in that hemisphere or a result of surgery in that area [60]. The Neuro‐QoL tool measures functions of cognition that could be affected by deficits caused by both right and left lobe lesions, which could potentially explain the lack of significance in this study. There is also an established correlation between higher education and better cognition over the course of adulthood [61]. One study highlighted a similar relationship between education and performance‐based and patient‐reported measures of cognitive functioning in patients with gliomas [26]. The majority of this cohort (86%) had some form of higher education, which likely diminished differences between education levels. Lastly, numerous studies have also highlighted the deleterious effects of radiation therapy on neurocognitive outcomes [36]. One study assessed brain tumor patients treated with neurosurgery with or without RT and found that the patients who received RT scored lower in areas of processing speed, verbal fluency, and executive function [37]. The extent of radiation has also been shown to affect the severity of neurocognitive deficits, with one study finding that patients who received whole‐brain radiation therapy had significant decline in learning and memory function compared to patients who received stereotactic radiosurgery (SRS) [35]. Researchers have been working to preserve cognitive function by finding new treatment approaches and delivering precisely targeted radiation through SRS; however, this treatment is rarely used for malignant PBTs [62]. In the present cohort, 73% of patients had received RT, limiting contrast between exposed and unexposed groups. RT‐related cognitive effects are heterogeneous and vary by dose, fractionation, treatment field, tumor location, modality, and interval from treatment; the present study was not powered to disentangle these parameters. Future work should explore the relationship between cognitive dysfunction and each of these clinical variables in longitudinal cohorts.

Evaluation of the association of COA's with cognitive dysfunction demonstrated that KPS, anxiety, depression, overall symptom burden, overall symptom interference, and cognitive symptom burden factors were statistically significant on univariate testing. Only the cognitive symptom burden factor and overall interference remained significant in modeling, even when clinical factors were also included in the final model. The MDASI‐BT and Neuro‐QoL Cognition Function were developed for different measurement purposes: the MDASI‐BT assesses brain tumor symptom severity and symptom‐related interference, whereas the Neuro‐QoL Cognition Function captures patient‐reported cognitive functioning and difficulty with cognitively demanding daily tasks [12, 13, 14, 44, 45]. However, because both instruments include patient‐reported cognitive content, the current results should be interpreted as strong concurrent associations between conceptually related patient‐reported constructs. Prior work has described relationships among cognitive functioning, mood, quality of life, and health‐related quality of life in PBT or glioma cohorts [5, 22]. Previous literature further supports the importance of this finding, showing that there are core sets of symptoms that affect PBT patients and can be used to direct clinical care and assess treatment benefit [49]. As a potential limitation, because this study was cross‐sectional, we cannot infer causality or directionality in the relationships between the identified factors and perceived cognitive dysfunction.

5. Conclusions and Implications for Future Research

This study builds on previous work by assessing a broader range of putative factors affecting cognition as outlined by Kirkman et al. [43], including sociodemographic and clinical characteristics, and exploring association with other COAs. Overall, the resulting models demonstrate concurrent associations of MDASI‐BT cognitive symptom burden and interference with Neuro QOL report of cognitive dysfunction in PBT patients. These results should be interpreted as evidence of related patient‐reported experiences rather than causal direction. Understanding the impact and prevalence of cognitive symptoms, and how these are perceived, is essential for PBT patient care and identifying supportive care interventions. Additional psychometric validation, longitudinal assessment, performance‐based neurocognitive testing, and evaluation of incremental clinical value are needed before routine screening recommendations can be made. Patient‐reported tools are brief and can often be administered remotely, but the utility of repeated administration and longitudinal tracking in this population requires additional study. Their future use may facilitate understanding of response to interventions or need for referral for formal neuropsychological testing to quantify deficits, evaluate decision making capacity, and consider the utility of cognitive rehabilitation [63] throughout the survivorship period [8, 23]. With additional psychometric data, the Neuro‐QoL tool may prove to be useful in the clinical setting to evaluate the way PBT patients interpret how cognitive symptoms affect their functioning, and how this relates to their report of other symptoms and how they interfere with daily life.

In this report, perceived cognitive dysfunction was observed among patients with multiple tumor grades and tumor types, although prevalence did not differ significantly by grade or tumor type in this analysis. The diverse sample is useful for exploratory descriptive work, while tumor‐specific analyses in more homogeneous cohorts are needed to inform care. Future work should further explore factors of cognitive function in select populations and examine the association with biomarkers. Of particular interest is the ongoing investigation of the effect of IDH mutation status on cognitive symptom burden. IDH mutation status is an important diagnostic and prognostic biomarker in patients with gliomas [64]. Previous work has found that IDH‐wildtype patients exhibit a higher incidence of cognitive impairment [65, 66, 67]. One study showed that this relationship held true even with adjustment for various tumor and demographic characteristics [68]. Further exploring this relationship, as well as defining other significant markers and their relationship to clinical and demographic factors, can help define how these factors interact in select populations more at risk for cognitive dysfunction. This will help clinicians monitor patients and may allow for earlier recognition of cognitive deficits and timely interventions to improve outcomes.

Future analysis of cognitive function could include performance‐based neurocognitive testing and self‐report. Patient reports should be considered valid indicators of patient experience, while performance‐based neurocognitive testing can add mechanistic and domain‐specific information. This work is important to characterize, rather than adjudicate, the relationship between a patient's subjective experience of cognitive dysfunction and neurocognitive test performance [21, 69]. Cognitive dysfunction results in functional challenges for PBT patients and concurrent use of performance‐based testing with a self‐report measure of cognitive dysfunction will allow for a better understanding of the perceived impact of neurocognitive deficits and will help researchers develop ways to mitigate them.

Author Contributions

Morgan Johnson: conceptualization, writing – original draft, writing – review and editing, investigation, formal analysis, data curation, methodology. Kimberly Reinhart: writing – review and editing, investigation, formal analysis, data curation. Eric Burton: investigation, writing – review and editing. Michelle L. Wright: writing – review and editing, methodology. Tricia Kunst: writing – review and editing, investigation. Vivian A. Guedes: writing – review and editing, conceptualization, methodology. Marta Penas‐Prado: investigation, writing – review and editing. Elizabeth Vera: writing – original draft, writing – review and editing, data curation, formal analysis, investigation, methodology, conceptualization. Hope Miller: writing – review and editing, conceptualization, investigation, formal analysis, data curation, methodology. Jennifer Reyes: investigation, writing – review and editing. Jing Wu: investigation, writing – review and editing. Anna Choi: writing – review and editing, investigation. Terri S. Armstrong: writing – review and editing, writing – original draft, conceptualization, data curation, formal analysis, methodology, investigation, supervision, funding acquisition, project administration. Byram Ozer: investigation, writing – review and editing. Mark R. Gilbert: conceptualization, methodology, writing – review and editing, project administration, investigation. Alvina Acquaye‐Mallory: writing – review and editing, investigation, methodology, conceptualization. Tito Mendoza: conceptualization, methodology, data curation, formal analysis, supervision, writing – original draft, writing – review and editing.

Funding

This work was supported by the National Institutes of Health, 1ZIABC011786‐03.

Ethics Statement

Approval for this study was granted by the Institutional Review Board (IRB) of the National Institute of Health.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author (TM). The data are not publicly available due to restrictions (e.g., their containing information that could compromise the privacy of research participants).