Statement of Research Significance
Research Question(s) or Topic(s): Among patients with breast cancer who are undergoing chemotherapy, does a real-time, task-specific self-report measure of cognitive workload (National Aeronautics and Space Administration Task Load Index) associate with neurocognitive test performance? Main Findings: Greater cognitive workload, as measured by the National Aeronautics and Space Administration Task Load Index, was robustly associated with worse neurocognitive test performance with strongest relationships in the executive function domain. Study Contributions: Among patients with cancer and other populations in which greater cognitive workload is a central feature, integration of the National Aeronautics and Space Administration Task Load Index into neuropsychological testing environments may enhance characterization of cognitive difficulties. Future research should explore the extent to which relationships between the National Aeronautics and Space Administration Task Load Index and neurocognitive performance may differ between cancer patients and controls and the extent to which these differences may vary pre- to post-treatment.
Introduction
Cancer-related cognitive impairment (CRCI) refers to decline in memory, attention, executive function, and other aspects of thinking attributed to cancer and its treatments. The reported prevalence of CRCI varies from 15% to 75% (Dijkshoorn et al., Reference Dijkshoorn, van Stralen, Sloots, Schagen, Visser-Meily and Schepers2021; Mayo et al., Reference Mayo, Edelstein, Atenafu, Ajaj, Li and Bernstein2024). While there are many patient and cancer-related factors contributing to the wide prevalence estimate, a major one is heterogeneity in measurement approaches. In 2011, to improve standardization in CRCI measurement, the International Cognition and Cancer Task Force (ICCTF) recommended a specific battery of neuropsychological assessment measures (Wefel et al., Reference Wefel, Vardy, Ahles and Schagen2011). More recently, guidelines have been established for CRCI self-report measures (e.g., FACT-Cog PCI and PROMIS Cognitive Function Short Form 8a) (Henneghan et al., Reference Henneghan, Van Dyk, Kaufmann, Harrison, Gibbons, Heijnen and Kesler2021).
Notably, there is a well-established discrepancy between self-reported and objectively tested cognitive function across a wide range of patient populations, including cancer (Costa & Fardell, Reference Costa and Fardell2019; Onyedibe et al., Reference Onyedibe, Schmidt, Bizer, Zimmer and Steindorf2025; Schneider et al., Reference Schneider, Lee, Hernandez, Junghaenel, Stone, Meijer, Jin, Kapteyn, Orriens and Zelinski2024; Van Patten et al., Reference Van Patten, Mulhauser, Austin, Bellone, Cotton, Chan, Twamley, Sawyer and LaFrance2025). This manifests as a lack of correlation between objective and self-report measures of CRCI, as well as considerably lower prevalence of cognitive impairment when objective tests are used, as compared to self-report. Explanations for this discrepancy include lack of ecological validity and sensitivity limitations of neuropsychological assessment. Also, because neuropsychological assessment measures were originally developed for non-cancer populations (e.g., neurodegenerative disorders, traumatic brain injury), they may not fully capture the type of cognitive deficits most experienced by cancer patients. Accordingly, some thought leaders have suggested that rather than relying on overall test performance, additional indices should be included in CRCI evaluation, such as intra-individual variability in performance within and between neuropsychological tests (Bernstein et al., Reference Bernstein, Catton and Tannock2014; Vance et al., Reference Vance, Collette, Frank, Billings, Deaver, Del Bene, Fazeli, Bail, Li and Triebel2023). Others have proposed computerized cognitive neuroscience tasks as a potentially more sensitive way to objectively quantify CRCI (Gaynor et al., Reference Gaynor, Ahsan, Jung, Schofield, Li, Ryan, Ahles and Root2024; Horowitz et al., Reference Horowitz, Suls and Treviño2018).
In parallel, there has been increasing interest in self-report measures of CRCI, given that they capture the patient experience (Lycke et al., Reference Lycke, Lefebvre, Pottel, Pottel, Ketelaars, Stellamans, Van Eygen, Vergauwe, Werbrouck and Cool2019), align with a broader movement towards patient-reported outcomes for the monitoring of cancer and cancer treatment-related toxicities (Di Maio et al., Reference Di Maio, Basch, Denis, Fallowfield, Ganz, Howell, Kowalski, Perrone, Stover and Sundaresan2022), and associate with underlying brain changes (Schroyen et al., Reference Schroyen, Vissers, Smeets, Gillebert, Lemiere, Sunaert, Deprez and Sleurs2022). Self-report measures may also present the advantage of scalability, which is particularly important in oncology as most cancer patients receive treatment in community settings with limited access to trained neuropsychologists. Recommended CRCI self-report measures typically inquire how often patients have noticed issues with various aspects of cognition (processing speed, task shifting, concentration) recalled over a 7-day period (Valentine et al., Reference Valentine, Weiss, Jones and Andersen2019; Wagner et al., Reference Wagner, Sweet, Butt, Lai and Cella2009). However, these recall-based approaches are biased towards most recent and/or most severe symptoms (Hill et al., Reference Hill, Mogle, Whitaker, Gilmore-Bykovskyi, Bhargava, Bhang, Sweeder, Tiwari and Van Haitsma2019) and appear to be influenced by psychological and physical symptom burden (e.g., depression, anxiety, fatigue, pain) (Onyedibe et al., Reference Onyedibe, Schmidt, Bizer, Zimmer and Steindorf2025). In light of these limitations, methods that assess self-reported function in real-time and specific to cognitively demanding tasks may represent an underutilized approach to CRCI characterization. For example, the NASA Task Load Index (NASA-TLX) is a widely used, multidimensional instrument outside of neuropsychology that assesses perceived workload (e.g., mental demand, effort, performance, frustration) across a broad range of tasks and environments, ranging from occupation-specific (e.g., manning an aircraft cockpit) to activities of daily living and which is beginning to be explored as a complement to neuropsychological testing (Hardy & Hinkin, Reference Hardy and Hinkin2022; Hardy & Wright, Reference Hardy and Wright2018; Hart & Staveland, Reference Hart and Staveland1988).
To better understand reasons for the divergence between objectively tested and self-reported CRCI and, by extension, inform future patient-centered measurement approaches that accurately reflect cognitive abilities, we tested associations between objective neurocognitive performance and two different self-report paradigms: (1) a standard, recall-based approach (PROMIS Cognitive Function) and (2) real-time, task-specific self-evaluation (NASA-TLX) among 56 patients with stage I–III breast cancer evaluated prior to (pre-) and 4 weeks after completion of (post-) chemotherapy. We hypothesized that compared to recall-based appraisal of cognitive difficulties, patients’ real-time self-evaluation of their cognitive workload would be more strongly associated with their actual performance. We additionally explored the degree to which each of these self-report measures were predictive of cognitive decline from pre- to post-chemotherapy.
Method
Study design
This is a secondary analysis from a phase II, single-arm trial of memantine to prevent cognitive decline in 56 adult patients receiving chemotherapy for stage I–III breast cancer (NCT04033419). The main findings from the clinical trial have been previously published (Nakamura et al., Reference Nakamura, Deal, Park, Stanton, Lopez, Quillen, Kelly, Heiling, H., Nyrop, Ray, Dees, Reeder-Hayes, Jolly, Carey, Abdou, Olajide, Rauch and Joseph2023). Participants were recruited from three sites in the University of North Carolina (UNC) hospital system between September 2019 and August 2021. Participants were excluded for the following reasons: (1) another cancer diagnosis with estimated survival <5 years; (2) prior chemotherapy; (3) severe cognitive impairment, defined according to a Blessed Orientation Memory Concentration Test score ≥11; and (4) current alcohol or drug abuse. Participants with severe cognitive impairment were excluded given the trial’s objective to prevent cognitive decline and anticipated barriers to preventing and/or reliably measuring subtle cognitive changes due to cancer treatment in patients with severe pre-existing impairment. Participants completed neurocognitive assessments and self-report measures pre- and post-chemotherapy. Memantine was administered by mouth twice daily (BID) between pre- and post-assessments. The study was approved by the UNC Institutional Review Board and UNC Lineberger Protocol Review Committee, and all participants provided written informed consent. This study was conducted in accordance with the ethical principles of the Helsinki Declaration and its later amendments.
Measures
Demographics and cancer characteristics
Participants self-reported age, race, and years of education. Participants also self-reported Karnofsky Performance Status, which ranges from 30 (“severely disabled, require continuous nursing care”) to 100 (“normal, no complaints, no symptoms of disease”) (Loprinzi et al., Reference Loprinzi, Laurie, Wieand, Krook, Novotny, Kugler, Bartel, Law, Bateman and Klatt1994). The Weschler Test of Adult Reading (WTAR) was used to estimate premorbid cognitive reserve (Wechsler, Reference Wechsler2001). Tumor stage, HER2 and hormone receptor positivity, history of surgery and radiation, and chemotherapy timing (adjuvant vs. neoadjuvant) and regimen were abstracted from the electronic medical record.
Neurocognitive test battery
Clinical neuropsychological measures with well-established track records for evaluation of CRCI were used, including the Hopkins Verbal Learning Test-Revised (HVLT-R; verbal learning and memory), Digit Span (attention and working memory), Controlled Oral Word Association (COWA; phonemic fluency), and Animal Fluency (semantic fluency).
Additionally, computerized cognitive neuroscience tasks from the Cambridge Neuropsychological Test Automated Battery (CANTAB) were used, including the Delayed Matching to Sample (DMS; visual working memory), Rapid Visual Processing (RVP; processing speed), and One Touch Stockings of Cambridge (OTS; executive function) (Fray & Robbins, Reference Fray and Robbins1996). These three CANTAB tasks were chosen based on their previous use in the CRCI field (Chan et al., Reference Chan, Yeo, Shwe, Tan, Foo, Chu, Khor and Ho2019; Janelsins et al., Reference Janelsins, Heckler, Peppone, Ahles, Mohile, Mustian, Palesh, O’Mara, Minasian, Williams, Magnuson, Geer, Dakhil, Hopkins and Morrow2018; Magnuson et al., Reference Magnuson, Lei, Gilmore, Kleckner, Lin, Ferguson, Hurria, Wittink, Esparaz, Giguere, Misleh, Bautista, Mohile and Janelsins2019; Vardy et al., Reference Vardy, Dhillon, Pond, Rourke, Xu, Dodd, Renton, Park, Bekele, Ringash, Zhang, Burkes, Clarke and Tannock2014).
Based on ICCTF recommendations, cognitive impairment was defined as at least one neurocognitive test with a demographically corrected z-score < −2 or at least two tests with demographically corrected z-scores < −1.5 (Wefel et al., Reference Wefel, Vardy, Ahles and Schagen2011). Characterization of potential false positive rates of impairment was accomplished using a previously published tool available within R 4.5.3 MASS package (Agelink van Rentergem & Schagen, Reference Agelink van Rentergem and Schagen2025).
PROMIS Cognitive Function
The PROMIS Short Form Cognitive Function 8a was used to evaluate self-reported cognition (Valentine et al., Reference Valentine, Weiss, Jones and Andersen2019). Scores are reported as a T-score with a mean of 50 and standard deviation (SD) of 10. Higher scores represent better cognitive function. Based on established norms, T-scores < 45 were considered significant for at least mild cognitive difficulties (Rothrock et al., Reference Rothrock, Cook, O’Connor, Cella, Smith and Yount2019).
NASA Task Load Index (NASA-TLX)
The NASA-TLX is a multidimensional self-report measure for assessing task-specific, perceived workload in real-time across six domains: mental demand, performance, effort, frustration, physical demand, and temporal demand (Cao et al., Reference Cao, Chintamani, Pandya and Ellis2009; Hart & Staveland, Reference Hart and Staveland1988). The NASA-TLX has robust psychometric properties and has been used to evaluate cognitive and physical workload in occupational settings and everyday activities in both laboratory and real-world environments (Hertzum, Reference Hertzum2021). Each NASA-TLX domain is presented as a single-item, visual analog scale ranging from 0 (least demand/best performance) to 100 (highest demand/worst performance). For example, for mental demand, shortly after completing a specific task, respondents are asked, “How mentally demanding was this task?” and to indicate their response by circling 1 of 21 gradations along a line, ranging from “very low” on the far left to “very high” on the far right. The NASA-TLX is commonly analyzed both as a global unweighted average of the domains and with individual domain scores presented independently. In the present study, four of the NASA-TLX domain items – mental demand, performance, effort, and frustration – were completed by participants about and immediately following completion of each neurocognitive subtest. Participants were not asked to complete the temporal and physical demand items, as these were not relevant to many of the neurocognitive tests.
Power and sample size
As previously published, the primary efficacy outcome in the clinical trial was change in the CANTAB Delayed Matching to Sample, % correct at 12 s delay (Nakamura et al., Reference Nakamura, Deal, Park, Stanton, Lopez, Quillen, Kelly, Heiling, H., Nyrop, Ray, Dees, Reeder-Hayes, Jolly, Carey, Abdou, Olajide, Rauch and Joseph2023). We determined that a sample size of 45 participants, using a paired t-test with 0.05 one-sided significance level, had 80% power to detect an effect size of 0.38 (3% mean increase in the Delayed Matching to Sample score with SD of 8%) compared to the null hypothesized change from a historical control of patients with early-stage breast cancer receiving chemotherapy who did not receive any intervention for CRCI (Janelsins et al., Reference Janelsins, Heckler, Peppone, Ahles, Mohile, Mustian, Palesh, O’Mara, Minasian, Williams, Magnuson, Geer, Dakhil, Hopkins and Morrow2018). All available data from the clinical trial was used in the present analysis.
Data analysis
All analyses were conducted using either SAS 9.4 or R 4.5.3. To minimize the number of statistical tests, we combined the demographically corrected z-scores, using United States normative data, according to age, sex, and education (COWA, Animal Fluency, CANTAB) (Cambridge Cognition, 2019; Heaton et al., Reference Heaton, Ryan and Grant2009) or age only (HVLT-R, Digit Span) (Duff, Reference Duff2016; Lichtenberger & Kaufman, Reference Lichtenberger and Kaufman2012), from all neurocognitive tests of interest in a single criterion. We applied a parallel analytic approach to characterize changes in NASA-TLX domains, which were aggregated into a single score due to their high degree of inter-correlation (r > 0.55) within our dataset. Within this data reduction procedure, we first combined all NASA-TLX indices for individual domains (mental demand, effort, frustration, and performance) and then took an average across all neurocognitive subtests to generate an overall NASA-TLX score.
In order to examine the associations between NASA-TLX, PROMIS Cognitive Function, and objective neurocognitive tests, we utilized simple regression analyses at pre-chemotherapy. Within these analyses we controlled for age, cancer stage, and estimated premorbid function, as indexed by the WTAR, given that these factors are consistently implicated in their associations with CRCI. To reduce the total number of analyses and mitigate type-I error, we aggregated cognitive outcomes into a single criterion using a unit-weighted composite or “gatekeeper” approach (Koch & Gansky, Reference Koch and Gansky1996; Sun et al., Reference Sun, Binkowitz and Koch2017). Within this approach, all cognitive outcomes are standardized into within-sample z-scores and then combined, such that higher scores suggest better performance. To characterize the magnitude of association between specific objective neurocognitive tests and the NASA-TLX we also conducted explanatory analyses in which we examined the association between NASA-TLX and each neurocognitive subtest individually. All predictors were scaled using their interquartile range (IQR) to facilitate interpretation and comparison of the relative magnitude of effect between predictors given differences in their units/scales (Babyak, Reference Babyak2004, Reference Babyak2009). The IQR presents the additional advantage of representing meaningful change (i.e., moving from the 25th to 75th percentile) of a given predictor, rather than a 1-unit change, which, depending on the predictor, may be trivial, not readily interpretable, or of varying clinical significance.
To characterize cognitive changes from pre- to post-chemotherapy, we examined temporal changes in cognition using a repeated measures, linear model, as well as reliable change indices (RCI). For these supplementary analyses, reductions in performance were indicated by an RCI z-score change of <−2.0 on at least one subtest or <−1.5 on two or more subtests. We then characterized these changes using an integrated model in which RCI analyses were combined with pre- and post-treatment levels using a two-step approach that provided the following four-category classification (Boon et al., Reference Boon, de Boer, van Dorp and Nijssen2019): recovered (impaired at baseline + RCI evidence of improvement), improved (normal at baseline + RCI evidence of improvement), unchanged (normal at baseline and post-treatment, without evidence of RCI change), and deteriorated (normal at baseline and either impaired at post-treatment or with evidence of RCI decline), in which impairment was defined using the previously described ICCTF classification.
We explored univariate associations between a wide range of demographic factors, cancer and cancer treatment variables, and each of the self-reported cognition measures with post-treatment neurocognitive performance, adjusting for baseline neurocognitive performance. We then conducted a multivariable model examining the association between baseline NASA-TLX and post-treatment neurocognitive performance, adjusting for the same variables as in our cross-sectional pre-chemotherapy models – age, cancer stage, and WTAR – in addition to baseline neurocognitive performance.
We conducted two exploratory analyses to better delineate relative changes in neurocognitive performance and NASA-TLX scores. First, neurocognitive performance and NASA-TLX were transformed to produce ratios (e.g., NASA-TLX / performance), such that higher scores suggested greater subjective workload for a given level of cognitive performance at pre- and post-treatment. We also explored the possibility, for hypothesis generation only, that baseline NASA-TLX scores might moderate the association between baseline neurocognitive function and changes in performance by examining their interaction in a post-hoc analysis.
In terms of missing data, 53 individuals provided baseline cognitive data, among whom 11 (21%) did not complete post-treatment assessments. To minimize risk of model misspecification and inaccurate inferences from limiting models to existing data only, which is particularly relevant in clinical trials in which missing data may be related to treatment (e.g., missingness in participants who tolerated memantine less well), Monte Carlo simulation, using 200 imputations, was used to address missingness in the analyses involving cognitive change from pre- to post-chemotherapy.
Results
Participant characteristics
Demographic and clinical characteristics of the sample have been previously reported (Nakamura et al., Reference Nakamura, Deal, Park, Stanton, Lopez, Quillen, Kelly, Heiling, H., Nyrop, Ray, Dees, Reeder-Hayes, Jolly, Carey, Abdou, Olajide, Rauch and Joseph2023) and are reproduced again in Table 1. As shown, participants tended to be middle-aged or older, highly educated, and White. At pre-chemotherapy baseline, participants had relatively normal cognitive function on average, though 15 (28%) showed evidence of impairment based on ICCTF criteria and 15 (28%) reported at least mildly severe cognitive difficulties according to the PROMIS Cognitive Function. Individuals who dropped out between pre- and post-assessments had higher premorbid estimated intellectual functioning (p = .029) and had a slightly lower level of education (p = .030), but otherwise did not differ from completers.
Baseline participant characteristics (N = 56)
Table 1 Long description
The table presents demographic and clinical characteristics of 56 participants in a study. It includes 14 rows and 2 columns. The first column lists variables such as age, education, sex, race, cancer stage, hormone receptor status, chemotherapy timing, chemotherapy regimen, prior radiation, Karnofsky performance status score, mild or greater self-reported cognitive difficulty, and objective cognitive impairment. The second column provides mean values, standard deviations, or percentages for each variable. For example, the mean age is 56.2 years with a standard deviation of 12.8 years. The majority of participants are female (98%), White (77%), and have stage 1 cancer (57%). Most participants received adjuvant chemotherapy (64%) and an anthracycline-based regimen (43%). The Karnofsky performance status score averages 88.7 with a standard deviation of 10.0. Notably, 28% of participants reported mild or greater self-reported cognitive difficulty, and 28% showed objective cognitive impairment.
Note: self-report and objective cognitive measure data were available for 53 of the 56 enrolled participants.
* Objective cognitive impairment is defined as at least one neurocognitive test with a demographically corrected z-score < −2, or at least two tests with demographically corrected z-scores < −1.5.
Pre-treatment NASA-TLX and neurocognitive performance associations
Examination of NASA-TLX scores and neurocognitive test performance revealed a robust association pre-chemotherapy (β = −0.36 [−0.53, −0.19], p < .001), such that higher NASA-TLX scores (greater perceived workload) were associated with worse objective performance (Table 2; Figure 1a). Explanatory analyses were conducted to better understand the pattern of findings (Supplemental Table 1). Examination of individual cognitive outcomes revealed a strong relationship between NASA-TLX scores and neurocognitive performance across subtests, with a particularly strong relationship for executive function measures: One Touch Stocking of Cambridge (IQR = 31): −0.70 (−1.0, −0.3), p < .001; Animal Fluency (IQR = 35): −3.3 (−5.8, −0.8), p = .011; COWA (IQR = 36) = −6.7 (−12.8, −0.7), p = .031. For example, a 35-point higher NASA-TLX (greater workload) was associated with 3.3 points lower animal fluency and a 36-point higher NASA-TLX corresponded to a nearly 7-point lower performance on the COWA.
Pre-chemotherapy relationships of a.) NASA-TLX and b.) PROMIS Cognitive Function with neurocognitive performance among patients with stage I–III breast cancer.
Regression model examining pretreatment associations between NASA-TLX, PROMIS Cognitive Function, and neurocognitive performance (unit-weighted, within-sample z-score), controlling for age, cancer stage, and WTAR score. Values are parameter estimates standardized using the interquartile range of the predictor
Table 2 Long description
A table showing regression model results for associations between NASA-TLX, PROMIS Cognitive Function, and neurocognitive performance. The table has 5 rows and 4 columns. The columns are labeled Variable (Interquartile range), β, and p. The rows are labeled Age (35), Cancer stage, WTAR (15), PROMIS Cognitive Function (9), and NASA-TLX (20). The values in the table are as follows: Row 1: Age (35), -0.28 (-0.44, -0.13), <.001. Row 2: Cancer stage, -0.11 (-0.28, 0.08), .259. Row 3: WTAR (15), 0.52 (0.31, 0.73), <.001. Row 4: PROMIS Cognitive Function (9), 0.05 (-0.15, 0.24), .618. Row 5: NASA-TLX (20), -0.36 (-0.53, -0.19), <.001.
Pre-treatment PROMIS and neurocognitive performance associations
There was no significant association between PROMIS self-reported cognitive function and objective performance, overall (β = 0.05 [−0.15, 0.24], p = .618) and across individual neurocognitive subtests (data not shown), at pre-chemotherapy baseline (Table 2; Figure 1b).
Cognitive changes from pre- to post-chemotherapy
Examination of changes in objectively assessed neurocognitive function revealed a trend towards higher (better) performance following chemotherapy compared to baseline (mean z-scores: PRE = −0.05 [−0.21, 0.11] vs. POST = 0.14 [−0.03, 0.32]), p = .06). The proportion of participants meeting ICCTF criteria for impairment was 28% (PRE) vs. 18% (POST). The prevalence of overall cognitive impairment was greater than the expected false positive rate at baseline but not at post-chemotherapy. Across the 45 participants who completed post-assessments, 38% (n = 17) improved (normal at baseline with RCI evidence of improvement), 27% (n = 12) deteriorated (normal at baseline and impaired at post-treatment or with evidence of RCI decline), 18% (n = 8) recovered (impaired at baseline + RCI evidence of improvement), and 18% (n = 8) remained unchanged (normal at baseline and post-treatment without RCI evidence of change). Examination of individual subtests revealed that the largest changes were observed on the HVLT-R Delayed Recall (−0.17 [−0.46, 0.12] vs. 0.22 [−0.09, 0.53]), HVLT-R Total Learning (−0.09 [−0.41, 0.24] vs. 0.24 [−0.11, 0.60]), and the RVP (−0.36 [−0.72, 0.0] vs. −0.07 [−0.45, 0.31]), with more modest changes on other indices.
Examination of changes in NASA-TLX scores over time revealed a non-significant pattern (44.9 [40.5, 49.3] vs. 46.0 [41.3, 50.8], p = .72). However, there was a trend suggesting that from pre- to post-treatment, participants reported higher NASA-TLX scores relative to cognitive performance: PRE = 1.06 [0.96, 1.15] vs. POST = 1.11 [1.02, 1.21], p = .06. No meaningful change was observed with self-reported cognition according to the PROMIS (32.0 [30.2, 33.8] vs. 30.5 [28.6, 32.4], p = .24). Raw scores for neurocognitive subtests and their associated NASA-TLX subscale scores at pre- and post-chemotherapy are shown in Supplemental Table 2.
Predictors of change in neurocognitive performance
In univariate regression, pre-treatment NASA-TLX was associated with post-chemotherapy neurocognitive performance (−0.35 [−0.67, −0.06], p = .021) (Table 3). However, in the final multivariable model adjusting for age, cognitive reserve, cancer stage, and baseline neurocognitive performance, pre-treatment NASA-TLX scores were no longer predictive of post-treatment performance. Of note, in exploratory analyses, we identified an interaction between baseline neurocognitive performance and NASA-TLX scores, such that participants with lower baseline neurocognitive performance and higher NASA-TLX scores (greater workload) were more likely to exhibit poorer post-treatment cognitive function (p = .048; Supplemental Figure 1). For example, among participants in the lowest tertile of baseline neurocognitive performance, those with NASA-TLX ≥ 50 exhibited greater cognitive decline than those with NASA-TLX < 50 (−0.67 [0.21, 1.13], p = .006). By comparison, no significant differences were observed according to NASA-TLX scores among those in the second (0.07 [−0.43, 0.56], p = .782) or highest tertiles (0.01 [−0.58, 0.60], p = .970) of baseline function. PROMIS Cognitive Function was not associated with post-treatment performance in either univariate or multivariable models.
Associations between demographic factors, cancer-related variables, and pre-treatment self-reported cognition with post-treatment neurocognitive performance
Table 3 Long description
A table with 12 rows and 6 columns. The columns are labeled Variable, Univariate Beta, Univariate p, Multivariable Beta, and Multivariable p. The table is divided into three main categories: Demographic, Cancer-related, and Cognitive. Each row lists a specific variable along with its corresponding beta and p-values for both univariate and multivariable analyses. The Demographic category includes Age, Race, and Education. The Cancer-related category includes Stage, HER2+, HR+, Chemotherapy timing, Chemotherapy regimen, and Karnofsky performance status. The Cognitive category includes WTAR, Baseline neurocognitive performance, NASA-TLX, and PROMIS Cognitive Function. Notable trends include significant p-values for WTAR and Baseline neurocognitive performance in both univariate and multivariable analyses, indicating their strong association with post-treatment neurocognitive performance.
Note: Given the very high degree of correlation between education and cognitive reserve (WTAR), WTAR, but not education, was included in the multivariable model.
* All cognitive variables listed are measured at pre-treatment baseline.
Discussion
In this study of 56 stage I–III breast cancer patients, who were enrolled in a clinical trial of memantine to mitigate cognitive decline during chemotherapy, we observed strong relationships pre-chemotherapy between real-time, task-specific self-evaluation of cognitive function (NASA-TLX) and objectively assessed neurocognitive test scores. These associations were strongest in the executive function domain. By comparison and as expected, there was no association between recall-based self-reported CRCI (PROMIS) and neurocognitive performance. Pre-treatment NASA-TLX scores were also associated with post-treatment cognitive performance in univariate analyses. While this relationship was not significant after adjusting for relevant covariates, including pre-treatment neurocognitive performance and cognitive reserve, exploratory analyses suggest a potential interaction such that individuals with lower baseline neurocognitive performance and higher NASA-TLX were at greater risk for post-treatment cognitive decline.
Multiple prior studies within and outside of oncology have demonstrated that self-reported cognitive difficulties show weak or inconsistent associations with objective neurocognitive performance and may be subject to multiple sources of bias. For example, item wording of self-reported cognitive measures may be interpreted inconsistently between subjects, emotional and personality factors impact responses, and recall-based measures are biased by worst or most recent symptoms (Hill et al., Reference Hill, Mogle, Whitaker, Gilmore-Bykovskyi, Bhargava, Bhang, Sweeder, Tiwari and Van Haitsma2019; Huang & Maurer, Reference Huang and Maurer2019). Further, though less relevant to our study sample, patients with substantial cognitive impairment may have difficulties accurately comprehending item wording and may be prone to making response errors (Schneider et al., Reference Schneider, Lee, Hernandez, Junghaenel, Stone, Meijer, Jin, Kapteyn, Orriens and Zelinski2024).
In contrast to recall-based measures, the NASA-TLX evaluates a different construct – workload associated with a specific activity or task and in real-time. The largest body of literature of the NASA-TLX is in occupational settings, where it has been found to be moderately correlated with performance in a wide range of cognitively demanding occupations, including those in aviation and healthcare (Mansikka et al., Reference Mansikka, Virtanen and Harris2019; Said et al., Reference Said, Gozdzik, Roche, Braun, Rössler, Kaserer, Spahn, Nöthiger and Tscholl2020). However, it has been studied in diverse contexts with established reference values for a wide range of professions as well as educational, leisure, and daily living contexts (Hertzum, Reference Hertzum2021, Reference Hertzum2022). To date, the NASA-TLX has been minimally investigated in neuropsychological testing environments but has previously been shown to be highly associated with performance on the Tower of Hanoi (Hardy & Wright, Reference Hardy and Wright2018). Consistent with this prior research is our finding that the strongest associations were in the executive function domain, which is also clinically relevant since executive function tests tend to approximate real-world function. For example, in one study of 178 breast cancer survivors, return to work two years after diagnosis was predicted by performance on tests of executive function but not memory at diagnosis and one year (Lange et al., Reference Lange, Lequesne, Dumas, Clin, Vaz-Luis, Pistilli, Rigal, Lévy, Lerebours, Martin, Everhard, Menvielle and Joly2024).
It is worth noting that the self-appraisal of actual performance, as was captured in the current study with the NASA-TLX, has a high degree of overlap with the construct of metacognition, generally defined as the match between subjective performance appraisal and objectively observed behavior (i.e., self-monitoring of cognitive processes). In the metacognition literature, metacognitive impairments can lead to greater concerns about normal performance and both less accurate assessment of and increased focus on cognitive lapses (Bhome et al., Reference Bhome, McWilliams, Huntley, Fleming and Howard2019). In a recent CRCI review, authors specifically note it surprising that the well-established discrepancy between self-report and objective test results in CRCI has yet to be framed as metacognitive impairment (Spinetti & Orzelleca, Reference Spinetti and Orzelleca2025). The authors go on to highlight the need for questionnaires that assess metacognition, including self-awareness, its neurobiological basis, and to investigate the functional consequences of impairments in metacognition. Interestingly, in our study, though we too observed notable discrepancies in recall-based self-reported cognition and neurocognitive performance, the close association between NASA-TLX ratings and performance suggests that the self-report vs. objective performance discrepancy in CRCI may not be fully explained by metacognitive impairments.
NASA-TLX-captured workload may also provide additional information about the relationship between cognitive reserve – another well-described construct within and beyond the CRCI literature – and neurocognitive performance. Cognitive reserve is believed to buffer against the impacts of cancer and its treatment on cognitive decline, such that patients with greater cognitive reserve may be less likely to experience objectively measurable CRCI (Ahles et al., Reference Ahles, Saykin, McDonald, Li, Furstenberg, Hanscom, Mulrooney, Schwartz and Kaufman2010). Research beginning to explore relationships between cognitive reserve and workload in healthy populations suggest that there is an interplay between cognitive reserve and workload, such that people with higher reserve generally report less workload/demand in response to cognitively demanding tasks (Devos et al., Reference Devos, Gustafson, Liao, Ahmadnezhad, Kuhlmann, Estes, Martin, Mahnken, Brooks and Burns2023). Further, while workload increases linearly with increasing task difficulty, performance may not be impacted among individuals with high cognitive reserve. Interestingly, in pathological processes such as neurodegenerative disorders, the expected relationship between reserve and workload may be more variable (Gu et al., Reference Gu, Chen, Gao, Shu, Wang, Liu, Yan, Li and Zhang2018). Our findings that there was a trend towards increasing NASA-TLX scores (greater workload) to maintain equivalent neurocognitive test scores from pre- to post-chemotherapy and that there was an interaction between NASA-TLX and baseline neurocognitive performance such that participants with worse baseline performance and higher NASA-TLX were more likely to exhibit poorer post-treatment performance may suggest that measures of workload may be a more responsive marker to the cognitive impacts of cancer treatment than objective test scores alone and may also offer mechanistic insights into the impacts of cancer and its treatment (e.g., decreased neural efficiency). In line with this conceptualization, prior fMRI studies in cancer patients and healthy controls have revealed a pattern of increased activation to maintain performance in cancer patients (Ferguson et al., Reference Ferguson, McDonald, Saykin and Ahles2007; Saward et al., Reference Saward, Ellis, Cobden and Caeyenberghs2022). To this end, it would be interesting in future research to examine concordance between NASA-TLX and more direct markers of neural efficiency (e.g., fMRI).
The NASA-TLX could be integrated into neuropsychological testing environments, as in our study, to provide additional information. Given the heterogeneity in the clinical presentation of CRCI, the NASA-TLX may be useful in identifying task-specific problems for any given individual that could be tracked over time. Another valuable clinical insight that would be answerable in a controlled study utilizing the NASA-TLX would be if cancer survivors rate more tasks as more difficult compared to controls and, if so, if the discrepancy increases from pre- to post-treatment. We acknowledge that just as the discrepancy between recall-based self-report measures of cognition and neurocognitive performance is not unique to CRCI, it is very likely that the concordance between task-specific, real-time self-report measures like the NASA-TLX and neurocognitive performance would be observed across many other patient populations.
Though outside of the scope of this study, there is also research to suggest that the NASA-TLX could be deployed in an ecological momentary assessment (EMA) paradigm to capture patients’ experience in complex daily environments and account for variability in symptoms from one day to the next. Our team has previously shown that EMA improves the reliability and sensitivity of CRCI measurement (Scott et al., Reference Scott, Mogle, Sliwinski, Jim and Small2020; Small et al., Reference Small, Jim, Eisel, Jacobsen and Scott2019; Veal et al., Reference Veal, Scott, Jim and Small2023), and the NASA-TLX specifically has been used within EMA in other patient populations as a valid way to monitor routine and self-care activities (Hernandez et al., Reference Hernandez, Roll, Jin, Schneider and Pyatak2022). Given the growing interest of EMA in CRCI and the utility of the NASA-TLX in EMA studies outside of cancer, it would be interesting in future studies to use the NASA-TLX to evaluate workload experienced by cancer patients during everyday tasks.
Strengths and limitations
A major strength of this analysis is the richness of cognitive data, including traditional neuropsychological assessment, computerized cognitive tasks, recall-based self-report, and the NASA-TLX. We also acknowledge limitations related to our relatively small sample size and generalizability. Specifically, educational level was generally high, and patients were mostly White, but the proportion of Black patients was higher than in most published CRCI studies. It is also important to acknowledge that participants in our study had lower rates of cognitive decline from pre- to post-chemotherapy than what is typically reported in observational studies. Whether related to potential benefit from memantine within the open-label trial or other factors, the less than expected rate of decline may have impacted relationships between self-report and objective measures particularly as predictive markers of post-treatment performance.
Conclusions and future directions
In this study, we found that the real-time, task-specific NASA-TLX was a reliable proxy indicator of neurocognitive performance prior to chemotherapy. By comparison, and consistent with prior research, recall-based self-report was not associated with neurocognitive performance. We do not recommend that the NASA-TLX replace conventional, recall-based self-report but instead highlight the potential for the NASA-TLX to capture a different dimension of CRCI within the context of neuropsychological testing and possibly in naturalistic settings.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355617726102112.
Acknowledgements
None.
Funding statement
This research was supported by the National Institutes of Health grants (Z.M.N., grant numbers K12HD001441, K23AG086604; T.A.A., grant number P30 CA008748; and P.J.S, grant number R61AG080615), and the University of Cancer Research Fund. The content is solely the responsibility of the authors and does not represent the official views of the NIH.
Competing interests
No competing interests are relevant to this manuscript. Zev M. Nakamura receives industry research funding from Reunion Neuroscience and receives royalties for content contributed to Wolters Kluwer (UpToDate). These relationships are entirely independent of the current study and had no role in its design, conduct, or reporting.
