Baseline Cognition Assessment Among Patients With Oropharyngeal Cancer Using PROMIS and NIH Toolbox

Key Points

Question

What is the baseline cognitive function in patients with oropharyngeal cancer (OPC)?

Findings

In this cohort study of 56 patients with oropharyngeal cancer, use of standardized Patient-Reported Outcomes Measurement Information System (PROMIS) and National Institutes of Health (NIH) Toolbox Cognitive Battery instruments identified self-reported impairment in 6 patients, and objective intelligence-stratified impairment in 12 patients. Self-reported cognition correlated with anxiety, depression, fatigue, and pain but not with objective cognition; objective impairment was more common in the presence of male sex, p16-negative OPC, moderate to severe comorbidity, and hearing loss.

Meaning

PROMIS and the NIH toolbox allowed immediate scoring of demographically adjusted cognition and may be useful in clinical practice to identify impaired patients at baseline for early institution of interventions to minimize posttreatment cognitive decline.

Abstract

Importance

Cognitive dysfunction (CD) is recognized by the American Cancer Society as a treatment effect in head and neck cancer, but the extent of this problem at baseline in oropharyngeal cancer (OPC), the most common subsite in current practice, to our knowledge has never been studied.

Objective

To assess the baseline cognition of patients with OPC using National Institutes of Health (NIH)-sponsored instruments of Patient-Reported Outcomes Measurement Information System (PROMIS) and NIH Toolbox Cognitive Battery (NIHTB-CB).

Design, Setting, and Participants

This was a prospective cohort study conducted at a tertiary academic center. Of 83 consecutive patients, newly diagnosed as having OPC from September 2016 to May 2017, 16 were ineligible, 8 refused to participate, and 3 were lost to follow-up after screening, resulting in 56 study participants.

Main Outcomes and Measures

Self-perceived and objective cognition with PROMIS and NIHTB-CB standardized T scores, respectively, were main outcomes. Impairment was defined as (1) T scores less than 0.5 SD for PROMIS; (2) T score less than 1.5 SD in at least 1 cognitive domain or less than 1 SD in 2 or more domains for NIHTB-CB total cognition; and (3) T score per previously published criteria for NIHTB-CB intelligence-stratified cognition.

Results

Of the 56 study participants (52 men, 4 women; median age, 59 years [range, 42-77 years]), 19 (34%) had a college degree, and 20 (36%) had a professional or technical occupation. Thirty (about 53%) were never-smokers, 26 (46%) were never-drinkers, 29 (52%) were obese, 13 (23%) had a moderate to severe comorbidity, 3 (5%) used antidepressants, and 25 (52%) had hearing loss. Impaired self-reported, NIHTB-CB total, and intelligence-stratified cognition scores were observed in 6 (11%), 18 (32%), and 12 (21%), respectively. Among all variables, objective impairment was more common in men (23% vs 0%) and those with p16-negative OPC (33% vs 20%), moderate to severe comorbidity (31% vs 18%), and hearing loss (31% vs 12%).

Conclusions and Relevance

Impaired objective cognition was more common at baseline than self-reported, and was more frequent in men, participants with p16-negative OPC, moderate to severe comorbidity, and hearing loss. NIHTB-CB allowed immediate scoring of demographically adjusted cognitive function. In clinical practice, these scores can be used to identify patients with impaired cognition at baseline who may be susceptible to developing further impairment after treatment. Identification of impairment at baseline will help to institute early cognitive interventions, which may lead to an improved posttreatment quality of life.

This cohort study assesses the baseline cognition of patients with oropharyngeal cancer using the Patient-Reported Outcomes Measurement Information System and National Institutes of Health (NIH) Toolbox Cognitive Battery instrument.

Introduction

Cancer-related cognitive impairment (CRCI) can have a negative effect on quality of life (QOL) in survivors, impeding return to their personal and professional pursuits.^{1,2,3,4,5,6,7,8,9,10,11} Most studies on CRCI, which is mainly attributed to chemotherapy (“chemobrain”), have been performed in breast cancer, but it is an understudied problem in head and neck cancer (HNC). The epidemiology of HNC in the United States is changing, with a rapidly increasing incidence of oropharynx cancer (OPC) driven by human papillomavirus (HPV).^12,13,14 Population-based data show an annual trend toward increasing incidence of pharynx cancer with decreasing mortality.^{13,15,16,17,18,19} The number of oral and pharynx cancer survivors is predicted to increase by nearly 20%, from 229 880 in 2016 to 293 290 in 2026.²⁰ Owing to the increasing HNC survivorship, particularly for OPC, a greater number of patients are at risk of experiencing CRCI.

Most of the cognition studies in HNC have been performed in patients with cancer of the nasopharynx or skull base^{21,22,23,24,25,26,27,28,29,30,31}; only a few^{32,33,34,35,36,37} include a small number of patients with cancer of other head and neck subsites. The risk of patients developing cognitive changes from cranial irradiation is high in those with nasopharynx or skull base cancer.^30,31,38 Moreover, most of the studies are retrospective^{22,23,24,25,27,28,31,33,39} and cross-sectional,^21,34,35,36 with no baseline information. Baseline information is important to correctly quantify the magnitude of posttreatment impairment. Thus, the available literature does not inform about the extent of CRCI in patients with OPC prior to treatment. Therefore, we designed a prospective study with a uniform cohort comprised of patients with newly diagnosed OPC, the subsite head and neck oncologists are most commonly treating in their current practice.

In addition to the heterogeneity in the subsites and treatment protocols, the cognitive assessment of CRCI in clinical HNC practice is made more challenging by the lack of well-defined criteria for a clinically meaningful impairment at baseline. There is also heterogeneity in the cognitive function instruments and the measured cognitive domains. A variety of instruments are used to assess self-reported and objective CRCI. The self-reported measures involve questionnaires to assess patients’ perceptions of difficulties with general cognitive abilities in their daily life. The objective measures are performance-based and mainly assess the cognitive domains of attention, processing speed, executive function, memory, and language. The interpretation of objective measures can vary, and the score for each domain is separately considered by some researchers, while total score across all domains is considered by others.⁴⁰ Another approach is to group tests that assess more than 1 specific domain but share certain characteristics across the lifespan. This approach has its basis in the 2-component model of intellectual development consisting of fluid and crystallized abilities.^41,42 Fluid cognitive abilities are used to solve problems, think and act quickly, and build new episodic memories, and are presumed to be influenced by biological processes across the lifespan or potential brain insults,⁴⁰ whereas crystallized abilities, such as language, are dependent on an individual’s learning experiences and are presumed to be relatively consistent across the life span.⁴⁰ Fluid and crystallized abilities are summated to produce a total score representing an individual’s global cognition. To capture the self-reported and objective cognition, we used the standardized instruments, Patient-Reported Outcomes Measurement Information System (PROMIS) and National Institutes of Health (NIH) Toolbox Cognitive Battery (NIHTB-CB), respectively. Developed under NIH initiatives, PROMIS and NIHTB-CB have been normed on large US nationally representative samples.^43,44 These measures are considered to have greater precision than traditional neuropsychological measures.^43,44,45,46 PROMIS and NIHTB-CB were validated with the item response theory method that results in fewer items, reduces the assessment burden, and generates immediate results. NIHTB-CB assesses objective cognitive function in accordance with the most influential 2-component model of intellectual development and generates 3 summary scores of fluid, crystallized, and total cognition. Thus, the specific goal of this study was to assess the baseline self-reported and objective cognitive function of patients with OPC using PROMIS and NIHTB-CB, respectively. Additional objectives were to assess the correlation between self-reported and objective cognitive function, and to identify factors that may be associated with an impaired cognitive function at baseline.

Methods

An institutional review board–approved observational, prospective cohort study of patients newly diagnosed as having OPC was conducted. Participants provided written informed consent. They were not compensated. Potential participants were identified from the head and neck multidisciplinary tumor board at the time of their first presentation. The inclusion criteria were adults with (1) newly diagnosed OPC, (2) unknown primary at presentation if the primary site was confirmed as the oropharynx during operative biopsy or the results of p16 immunohistochemical (IHC) analysis were positive, and (3) planned curative treatment. Curative treatment with definitive surgery or chemoradiation was planned by the head and neck oncology multidisciplinary team independent of the study. The exclusion criteria were (1) distant metastasis, (2) true unknown primary with negative results from p16 IHC, (3) history of chemotherapy or radiotherapy, (4) any acute or chronic neurological conditions with residual deficits. Pertinent demographic, tumor, pathologic, and treatment-related data were collected. Data on addiction, comorbidity (Adult Comorbidity Evaluation-27),⁴⁷ body mass index, antidepressant use, and hearing status (self-reported and pure-tone audiogram [PTA]) were collected. HPV-relatedness was recorded by the p16 IHC.

Study Instruments

PROMIS

Study participants completed the 8-item Applied Cognition-Abilities ^48,49 short-form questionnaire (version 1.0) to capture the self-reported cognitive function. A sample of the questions from PROMIS Applied Cognition form is provided (eQuestionnaire in the Supplement). Depression, anxiety, fatigue, and pain interference were secondary outcomes and were captured using relevant instruments from the PROMIS Mental,^48,49 and Physical Health item banks.⁵⁰ Study participants rated their responses using a 5-point scale. The raw score was translated into a standardized T score with a mean (SD) of 50 (10). A higher PROMIS T score represents more of the concept being measured.⁴⁵ For negative constructs, such as anxiety, T scores above the mean are worse, while for positive constructs, such as cognitive abilities, T scores below the mean are worse.

NIHTB-CB

The NIHTB-CB,⁵¹ version 1.0, was administered for objective assessment of cognitive function. The NIHTB-CB uses 7 tests to assess the core cognitive domains (eFigure in the Supplement) of processing speed, episodic memory, working memory, executive function, and attention as well as language (vocabulary and reading). The first 5 domains generate a fluid cognition score. The sixth domain of language generates a crystallized cognition score. The fluid and crystallized together give a total cognition score (eFigure in the Supplement). NIHTB-CB reports these scores as fully demographically adjusted scores, with a mean (SD) of 50 (10), that compare a participant’s score with a normative sample while adjusting for age, sex, race/ethnicity, and education.⁴⁶ One of us (P.S.) received training to conduct the NIHTB-CB and administered the test to all patients in the study cohort.

Statistical Analysis

The sample size of our study was feasibility driven. Self-reported and objective cognition with PROMIS self-cognition and NIHTB-CB scores, respectively, were main outcomes. Secondary outcomes were PROMIS fatigue, pain, anxiety, and depression scores. Descriptive statistics were used to describe the study variables, and the PROMIS and NIHTB-CB scores. For PROMIS self-cognition, we considered T scores 0.5 SD below the mean as denoting impaired self-reported cognition based on previous references.^52,53 For NIHTB-CB, clinical impairment was explored using 2 methods. The first method included a modified psychometric criterion for cognitive impairment based on the Diagnostic and Statistical Manual of Mental Disorders (5th edition).^54,55 Under this criterion (referred to as total cognition), impairment was considered to be present if the demographically adjusted T score for at least 1 cognitive domain was 1.5 SD below the mean or 2 or more domains had T scores 1.0 SD below the mean.^54,55 The second method explored clinical impairment based on the approach of Holdnack and colleagues^54,55 in which the crystallized score serves as an estimate of premorbid intelligence. Under this method (referred to as intelligence-stratified cognition), for individuals with crystallized T scores of 58 or higher, the cutoff for clinical impairment for fluid tests was a T score of less than 44; for crystallized T scores of 50 to 57, the cutoff was a T score of less than 41; for crystallized T scores of 43 and 49, the cutoff was a T score of less than 38; and for crystallized T scores of less than 43, the cutoff was a T score of less than 35.^54,55 Independent t test was performed for comparison of continuous variables between groups. Correlations were performed for PROMIS self-cognition and NIHTB-CB domains. Effect size metrics were defined with 95% CIs. All statistical tests were 2-sided. Statistical analysis was performed using SPSS software (IBM SPSS Statistics, release 25.0.0) and SAS software (version 9.4; SAS Institute Inc).

Results

Patient Characteristics

A total of 83 patients newly diagnosed as having OPC presented to our institution from September 2016 to May 2017, of whom 16 were ineligible, 8 refused to participate, and 3 were lost to follow-up after screening. Thus, the final study included 56 participants. The causes for ineligibility were palliative therapy (6 patients), transfer of care (3), dementia (2), true unknown primary (2), treatment refusal (1), stroke with deficits (1), and previous head and neck radiation (1). The detailed demographic, patient, and tumor characteristics are presented in Table 1.

Table 1. Demographic, Patient, and Tumor Characteristics of the Study Cohort.

Variable	No. (%)
Age, median (IQR) [range], y	59 (53-65) [42-77.5]
Sex
Male	52 (93)
Female	4 (7)
Ethnicity
Not Hispanic or Latino	56 (100)
Race
White	54 (96)
African American	2 (4)
Education
10th Grade to high school graduate	20 (36)
Some college/associate’s degree	17 (30)
Bachelor’s degree	10 (18)
Postgraduate	9 (16)
Education, median (IQR) [range], y	14 (12-16) [10-24]
Occupationa
Partly skilled/unskilled	3 (5)
Skilled occupation—manual and nonmanual	33 (59)
Professional, managerial, and technical	20 (36)
Smoking status
Never	26 (46)
Current	13 (23)
Former	17 (30)
Pack-years, median (IQR) [range]	25 (9.4-40) [0.5-60.0]
Drinking status
Never	30 (54)
Current	19 (34)
Former	7 (12)
Drinks per wk, median (IQR), No.	8 (4-13) [2-48]
Illicit drugs
Never	54 (96)
Current	1 (2)
Former	1 (2)
Comorbidity (ACE-27)
None to mild	43 (77)
Moderate to severe	13 (23)
ECOG performance score
0	50 (89)
1	5 (9)
2	1 (2)
BMI
Normal (18.50-24.99)	9 (16)
Overweight (25.00-29.99)	18 (32)
Obese (≥30.00)	29 (52)
Use of antidepressants
Yes	3 (5)
No	53 (95)
Self-reported hearing loss
None	17 (30)
Yes	39 (70)
Hearing loss (PTA)b
No	23 (48)
Yes	25 (52)
AJCC 8th edition, clinical T category
T0	10 (18)
T1	7 (13)
T2	26 (46)
T3	5 (9)
T4	8 (15)
AJCC 8th edition, clinical N category
N0	1 (2)
N1	38 (68)
N2	15 (27)
N3	2 (4)
AJCC 8th edition, stage
I	33 (59)
II	13 (23)
III	10 (18)
p16 status
Positive	50 (89)
Negative	6 (11)

Abbreviations: ACE, adult comorbidity evaluation; AJCC, American Joint Committee on Cancer; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); ECOG, Eastern Cooperative Oncology Group; IQR, interquartile range; PTA, pure-tone audiometry.

^aCategorized by National Statistics Socioeconomic Classification.

^bAvailable for 48 patients.

Baseline PROMIS and NIHTB-CB Scores

PROMIS Scores

Excellent reliability of PROMIS was noted for self-reported cognition (Cronbach α = 0.97) as well as other domains of anxiety (α = 0.92), depression (α = 0.94), fatigue (α = 0.972), and pain interference (α = 0.98). Impaired self-reported PROMIS cognition scores were observed in 6 patients (11%). Impaired scores for secondary outcomes were noted in 20 (36%) for the PROMIS domains of anxiety, 17 (30%) for pain interference, 13 (23%) for depression, and 11 (20%) for fatigue. The distribution of PROMIS T scores for different predefined intervals measured by SD is presented in Table 2.

Table 2. Distribution of Overall and Domain-Specific PROMIS and NIHTB-CB (Fully Adjusted) T Scores.

Test	Score, Mean (SD) [range]	Patients With T Scores Within Each Interval, No. (%)a
		0.5 to <1.0 SD	1.0 to <1.5 SD	1.5 to <2.0 SD	>2.0 SD
PROMIS
Self-reported cognition	54.0 (8.5) [33.0-64.8]	3 (5)	1 (2)	2 (4)	0
Anxiety	50.3 (8.2) [37.0-65.6]	14 (25)	5 (9)	1 (2)	0
Depression	47.2 (8.4) [38.0-70.7]	8 (14)	4 (7)	0	1 (2)
Fatigue	48.4 (10.5) [33.0 -77.8]	5 (9)	2 (4)	2 (4)	2 (4)
Pain interference	48.6 (9.6) [40.7-77.0]	7 (12.5)	6 (11)	2 (4)	1 (2)
NIHTB-CB
Fluid cognition composite	50.8 (9.5) [28.0-72.0]	9 (16)	3 (5)	2 (4)	1 (2)
Flanker inhibitory control and attention	48.0 (7.9) [31.0-74.0]	13 (23)	6 (11)	2 (4)	0
Dimensional change card sort (executive function)	55.2 (11.0) [35.0-81.0]	9 (16)	3 (5)	1 (2)	0
List sorting working memory	51.9 (8.8) [33.0-73.0]	12 (21)	1 (2)	1 (2)	0
Pattern comparison processing speed	47.2 (13.3) [19.0-68.0]	10 (18)	7 (13)	7 (13)	4 (7)
Picture sequence episodic memory	50.9 (9.0) [30.0-68.0]	13 (23)	3 (5)	1 (2)	1 (2)
Crystallized cognition composite	55.0 (8.7) [38.0-77.0]	4 (7)	3 (5)	0	0
Picture vocabulary	49.0 (6.5) [37.0-65.0]	12 (21)	4 (7)	0	0
Oral reading recognition	60.3 (12.4) [33.0-86.0]	3 (5)	2 (4)	1 (2)	0
Cognition total composite: composite FCC and CCC	53.5 (8.9) [40.0-71.0]	13 (23)	1 (2)	0	0

Abbreviations: CCC, crystallized cognition composite; FCC, fluid cognition composite; NIHTB-CB, National Institutes of Health Toolbox-Cognition battery; PROMIS, patient reported outcomes measurement system.

^aEach category represents the number of participants within a predefined interval measured by SD units below the mean (PROMIS self-reported cognition, NIHTB-CB) and above the mean (PROMIS anxiety, depression, fatigue, pain). The rates are calculated as the number of participants within each category divided by the total number of study participants (n = 56).

NIHTB-CB Scores

Impaired objective demographically adjusted NIHTB-CB scores were observed in 18 (32%) using the total cognition method, and in 12 (21%) using the intelligence-stratified fluid cognition method. Impaired intelligence-stratified fluid scores were most commonly noted in the domains of processing speed (18 [32%]) followed by attention (11 [20%]), episodic memory (7 [12%]), executive function (5 [9%]), and working memory (2 [4%]). The distribution of NIHTB-CB T-scores for predefined intervals by SD is described in Table 2. The numbers of participants showing impairment on different number of NIHTB-CB tests with various T-score cutoffs by SD are described in the eTable in the Supplement.

Correlation Between PROMIS and NIHTB-CB Scores

Medium to large effect size for correlation was observed between the PROMIS self-reported cognition and the PROMIS domains of anxiety, fatigue, pain interference, and depression.⁵⁶ There were no statistical correlations between PROMIS self-reported cognition or other subjective domains and the fully demographically adjusted NIHT-CB fluid, crystallized, or total cognition scores (Table 3).

Table 3. Correlation Matrix Between the PROMIS Domains and Fully Demographically Adjusted NIHTB-CB Cognition Scores (Pearson Correlation Coefficient [95% CIs]).

Domain	PROMIS					NIHTB-CB
	Anxiety	Depression	Self-reported Cognition	Fatigue	Pain	Fluid Cognition	Crystallized Cognition	Total Cognition
PROMIS
Anxiety	1	0.66 (0.48 to 0.79)a	−0.45 (−0.64 to −0.21)a	0.420 (0.18 to 0.61)a	0.22(−0.05 to 0.45)	0.23 (−0.04 to 0.46)	0.07 (−0.2 to 0.33)	0.18 (−0.09 to 0.42)
Depression	0.66 (0.48 to 0.79)a	1	−0.56 (−0.72 to −0.35)a	0.565 (0.36 to 0.72)a	0.50 (0.27 to 0.67)a	0.045 (−0.22 to 0.31)	−0.02 (−0.29 to 0.24)	0.01 (−0.25 to 0.27)
Self-reported cognition	−0.45 (−0.64 to −0.21)a	−0.56 (−0.72 to 0.35)a	1	−0.45 (−0.64 to −0.21)a	−0.48 (−0.66 to −0.25)a	0.04 (−0.22 to 0.30)	−0.13 (−0.38 to 0.13)	−0.04 (−0.30 to 0.22)
Fatigue	0.42 (0.18 to 0.61)a	0.56 (0.36 to 0.72)a	−0.45 (−0.64 to −0.21)a	1	0.59 (0.39 to 0.74)a	−0.02 (−0.28 to 0.25)	0.05 (−0.21 to 0.31)	0.03 (−0.24 to 0.29)
Pain	0.22 (−0.05 to 0.45)	0.50 (0.27 to 0.67)a	−0.48 (−0.66 to −0.25)a	0.594 (0.39 to 0.74)a	1	0.04 (−0.23 to 0.30)	0.17 (−0.10 to 0.41)	0.13 (−0.14 to 0.38)
NIHTB-CB
Fluid cognition	0.23 (−0.04 to 0.46)	0.05 (−0.22 to 0.31)	0.04 (−0.22 to 0.30)	−0.016 (−0.28 to 0.25)	0.04 (−0.23 to 0.30)	1	0.33 (0.08 to 0.55)	0.83 (0.73 to 0.90)a
Crystallized cognition	0.07 (−0.20 to 0.33)	−0.02 (−0.29 to 0.24)	−0.13 (−0.38 to 0.13)	0.05 (−0.21 to 0.31)	0.17 (−0.10 to 0.41)	0.33 (0.08 to 0.55)b	1	0.80 (0.68 to 0.88)a
Total cognition	0.19 (−0.09 to 0.42)	0.01 (−0.25 to 0.27)	−0.04 (−0.30 to 0.22)	0.03 (−0.24 to 0.29)	0.13 (−0.14 to 0.38)	0.83 (0.73 to 0.90)a	0.80 (0.68 to 0.88)a	1

Abbreviations: NIHTB-CB, National Institutes of Health Toolbox-Cognition battery; PROMIS, Patient-Reported Outcomes Measurement System.

^aCorrelation is significant at the P = .01 level (2-tailed).

^bCorrelation is significant at the P = .05 level (2-tailed).

Distribution of PROMIS and NIHTB-CB Scores in Normal vs Impaired Objective Cognition Groups

PROMIS

The distribution of self-reported cognition score did not differ significantly when the normal and impaired groups were defined by the objective total cognition scores (mean difference, −1.8; 95% CI, −6.7 to 3.1). Nor was there any difference among the groups for the other 4 domains: anxiety (mean difference, 4.2; 95% CI, −0.34 to 8.8), depression (mean difference, 4.5; 95% CI, −0.2 to 9.2), fatigue (mean difference, −1.2; 95% CI, −7.2 to 4.9), and pain interference (mean difference, 0.9; 95% CI, −4.7 to 6.4). The distribution of all these domains did not differ significantly even when the normal and impaired groups were defined by the intelligence-stratified cognition scores.

NIHTB-CB

Statistically significant differences in the mean scores between participants with normal and impaired total cognition scores (Figure, A) were noted for the domains of attention (mean difference, 6.0; 95% CI, 1.6-10.0), executive function (mean difference, 8.5; 95% CI, 2.6-14.5), and processing speed (mean difference, 17.3; 95% CI, 11.0-23.4). The mean scores were not significantly different for the domains of working (mean difference, 2.1; 95% CI, −3.0 to 7.2) or episodic memory (mean difference, 4.2; 95% CI, −0.9 to 9.3). The same 3 domains of attention, executive function, and processing speed were also significantly different between participants with normal and impaired intelligence-stratified cognition (Figure, B).

Figure. Distribution of T Scores of the Core Domains.

Box and whisker plots showing the distribution of T scores of the core domains in the normal vs impaired cognition groups as determined by (A) total cognition scores and by (B) intelligence-stratified fluid cognition scores. A T score of 50 represents the mean. Of all domains, attention, executive function, and processing speed were significantly different between the 2 groups by both criteria.

Variables With Higher Frequency of Impaired Cognition Scores

Among all variables, impaired intelligence-stratified fluid cognition scores were more frequent in men (12 of 52 [23%] vs 0 of 4; ∆% = 23%; 95% CI, −27% to 36%), patients with p16-negative OPC (2 of 6 [33%] vs 10 of 50 [20%]; ∆% = 13%; 95% CI, −26% to 52%), moderate to severe comorbidity vs none to mild (4 of 13 [31%] vs 8 of 43 [18%]; ∆% = 13%; 95% CI, −14.6% to 40.6%) and PTA-identified hearing loss (7 of 25 [28%] vs 2 of 23 [9%]; ∆% = 19%; 95% CI, −1.7% to 40%]. Of these 4 variables, the 95% CI for the percentage difference were more precise for hearing loss and moderate to severe comorbidity than sex and p16-negative OPC. This enabled us to conclude with further exploratory analysis that the rate of impaired intelligence-stratified fluid cognition scores was 7% in the absence of moderate to severe comorbidity and hearing loss, 20% to 26% in the presence of either factor, and 40% in the presence of both.

Discussion

On assessment of cognition in a cohort of patients with OPC prior to treatment, self-reported impairment was reported by 11%, whereas objective impairment was observed in 32% using total cognition scores and in 21% using intelligence-stratified scores. There was no correlation between self-reported and objective cognitive function. A higher frequency of objective impairment was noted among men, and the groups with p16-negative OPC, moderate to severe comorbidity, and hearing loss.

Self-reported cognitive function using PROMIS revealed impairment in 11% of the OPC cohort at baseline. Studies on cognition in HNC have focused mainly on the posttreatment objective function, and little is known about the self-reported function. Self-reported cognitive complaints at baseline were assessed by Bond et al³⁷ in an HNC cohort of 70 patients (39 OPC), using the Alertness Behavior Subscale, but the frequency of participants with impaired scores was not described. PROMIS was shown as a reasonable measure to determine QOL outcomes in patients with HNC by Stachler et al.⁵⁷ However, this study focused on evaluating how PROMIS was associated with other known QOL instruments, and the number of patients with impaired self-reported cognition was not reported.⁵⁷ We found PROMIS to have an excellent reliability for measuring self-reported cognition in the OPC cohort as well as the secondary outcomes of anxiety, depression, fatigue, and pain interference.

Objective testing with NIHTB-CB showed mild cognitive impairment in 32% of participants using demographically adjusted total cognition scores. Cognitive impairment prior to treatment initiation is postulated to result from alteration in the psychological status or coping strategies from the new diagnosis of cancer.⁵ The NIHTB-CB⁵¹ has not been used yet in populations with HNC, but we found it feasible for administration in the OPC cohort. The assessment could be completed within 30 minutes. Administration time of up to 90 minutes¹⁰ has been described for some of the traditional neuropsychological tests. When demographically adjusted scores were stratified by the pretreatment intelligence, the impairment rate reduced to 21%. The domains that were most commonly impaired in total as well as intelligence-stratified scores were attention, executive function, and processing speed. We also explored the various cutoffs for impairment from 0.5 to 2 SD, and observed that the number of participants and tests with impaired scores decreased as the cutoff was increased (Table 2; eTable in the Supplement). This observation suggests that defining impairment based on lower cutoffs or a single domain can falsely increase the rate of impairment. Bond et al³⁷ observed a higher rate of impaired objective cognition at baseline in 47% of their multisite HNC cohort, using a neurocognitive battery comprising several tests. In contrast with NIHTB-CB, norms for these tests did not uniformly adjust for all the demographic variables of age, sex, race/ethnicity, and education. Moreover, even though a reading test was used to estimate intelligence, the cognitive scores were not stratified by intelligence.³⁷ Compared with our study, Williams et al⁵⁸ also observed a higher rate of impaired objective cognition in 55% of a multisite cohort with HNC (n = 209, n for OPC unknown) using the Montreal Cognitive Assessment (MoCA). Lack of timed measures in MoCA limits its sensitivity to identify the common nonmemory domain impairment in CRCI, such as processing speed. Moreover, levels of education, sex, age, race/ethnicity, or premorbid intelligence were not taken into account.⁵⁸ Premorbid intelligence is a predictor of individual cognitive domains^55,59 and should be considered for accurate interpretation of the cognitive test performance.⁵⁵ Tests of reading and vocabulary (crystallized cognition) have been considered to serve as “proxy” measures for overall intellectual ability because a high correlation between intelligent quotient (IQ) scores and vocabulary has been observed.^60,61,62 Without participanting patients to additional IQ tests, NIHTB-CB generates cognition scores stratified by the intellectual ability of each individual using his or her crystallized scores. Thus, NIHTB-CB allowed us to estimate the rate of objective cognitive impairment while adjusting for both, the demographic variables and premorbid intelligence, in a simple and efficient manner.

Self-reported cognitive function correlated with self-reported anxiety, depression, fatigue, and pain interference; however, objective cognition did not correlate with these symptoms. Our findings of medium to large correlation between subjective cognitive complaints and subjective constructs, such as depression, are consistent with those of previous research on CRCI, and so is our finding of the lack of correlation between the objective cognitive function and the subjective constructs.^5,32,37,63 We also did not find a correlation between the self-reported and objective cognitive functions, an observation similar to findings of previous published studies on CRCI in HNC^37,58 and breast cancer.^5,63 This lack of correlation may indicate that the 2 tests measure different cognition-related constructs.³⁷ Furthermore, self-reported cognition can explicitly capture the patient’s experiences over a period of time, whereas objective cognitive assessments are a “snapshot” in time.⁵ A higher frequency of objective vs self-reported impairment in the study cohort can also suggest that there may be more individuals with subtle limitations in their cognitive reserve at baseline that evade self-cognizance but manifest when challenged by certain tasks. In certain individuals, self-reported perception of cognitive abilities can also be affected by other latent factors, such as denial. For such patients, objective evaluation can assess the cognitive reserve, and when the objective measures suggest cognitive impairment, these patients may benefit from cognitive interventions. In addition, the value of objective testing lies in the fact that it can help identify the core cognitive domains that are impaired. For instance, the domains most frequently responsible for abnormal fluid cognition in our study were processing speed, attention, and executive function. Knowledge about the impaired domains will be useful in initiating specific, domain-targeted interventions even prior to treatment and continued during treatment in anticipation of mitigating the posttreatment decline. Moreover, in patients with self-reported impairment, objective testing can indicate the level of cognitive reserve as a predictor of CRCI outcomes after treatment. For instance, patients with both self-reported and objective impairment can have worse cognitive outcomes compared with the patients with self-reported impairment without objective impairment. We therefore believe that both self-reported and objective measures are needed to accurately understand the extent and nature of cognitive function at baseline. PROMIS and NIHTB-CB are 2 NIH outcome initiatives that promote measurement precision, and further research is needed to refine the use of these measures for optimal assessment of CRCI.

We observed a higher frequency of impairment at baseline in the groups with p16-negative OPC, moderate to severe comorbidity, and PTA-identified hearing loss, although the association was not statistically significant, possibly owing to small sample size. None of these factors have been evaluated in HNC studies on CRCI. Interestingly, impairment of both total and intelligence-stratified cognition scores was seen only among men in the study cohort. No definitive conclusions can be made owing to the small number of women (4), but the cohort was representative of the typical patients with HPV-driven OPC seen in current practice, predominantly middle-aged, white men. However, sex differences for cognitive abilities are well recognized.⁶⁴ In the future, we plan to follow the study patients after treatment and evaluate whether the factors identified as being associated with impairment at baseline are also associated with a greater magnitude of posttreatment decline. Knowledge about such factors, combined with the cognitive assessment, will facilitate identification of patients who are at higher risk of CRCI during and after treatment to initiate early cognitive rehabilitation interventions. Cognitive remediation measures have been associated with reduced cognitive dysfunction and improved QOL in cancer survivors with posttreatment cognitive impairment.^65,66,67 The association of cognition rehabilitation with CRCI was also studied at our institution in breast cancer survivors, and improved behavioral outcomes were noted.⁶⁸

Limitations

Our study did not have a control group of patients without HNC matched for age, sex, race/ethnicity, and education, but NIHTB-CB provided scores that were fully adjusted for these variables, thus reducing the relevance of such a control group. Second, we were not able to explore variability of the NIHTB-CB scores by repeating the test under different time or settings or even by comparing the scores with traditional neuropsychological instruments. This would have entailed adding to the patient burden of completing more tests or accruing a larger number of patients. The reliability and validity of NIHTB-CB has been confirmed in several studies, albeit in a population of patients without cancer.^40,54,60,69 Finally, we identified factors which were associated with a higher frequency of baseline impairment but cannot make any definitive conclusions owing to limited sample size. Further exploration of these factors in larger cohorts is needed.

Conclusions

In this study OPC cohort, the use of standardized PROMIS and NIHTB-CB instruments allowed immediate scoring of demographically adjusted self-reported and objective cognitive function. Impairment at baseline was more common for objective cognitive function than for self-reported impairment, and was associated with male sex, p16-negative OPC, moderate to severe comorbidity and hearing loss. Based on our findings, we believe it is important to use both self-reported and objective measures to identify individuals and domains with impaired cognitive function at baseline from the perspectives of cognitive interventions and outcome prediction. Early identification of such individuals and domains will help in timely institution of targeted cognitive interventions for an improved posttreatment QOL.

Supplement.

eQuestionnaire. PROMIS Applied Cognition sample

eTable. Frequency (%) of NIH Toolbox cognition battery (NIHTB-CB) tests with T-scores 0.5 SD, 1 SD, 1.5 SD, or 2 SD below the mean

eFigure. NIH Toolbox cognition battery