Psychometric Properties of the 37-item Functional Assessment of Cancer Therapy-Cognitive Function (FACT-Cog) scale in Cancer Patients.

Background: The Functional Assessment of Cancer Therapy-Cognitive Function (FACT-Cog) scale is a self-assessment scale validated in routine clinical practice to assess cognitive function in cancer patients. This study aimed to validate the 37-item version of FACT-Cog exploring particularly the psychometric properties of four items related to multitasking that were not previously included in the scoring algorithm and assess its correlates in Lebanese cancer patients. Methods: A cross-sectional study was carried out including 261 patients with breast, colorectal and lung cancers undergoing chemotherapy (Ethics: CEHDF1016). Validity was conrmed using a factor analyses using the principal component analysis technique with a varimax rotation. Analyses of internal consistency, “test-retest" reliability, and convergent validity were also performed. Finally, a multiple linear regression was conducted, using the total cognition scale as a dependent variable. Results: The scale had an appropriate construct validity, and items loaded on subscales with adequate sample adequacy to factor analyses outcomes. The test-retest reliability was appropriate for the total cognition score/all sub-scores except for the FACT-Cog QOL. Moreover, a weak but signicant and inverse correlation between the FACT-Cog scores and patient’s pain, fatigue, anxiety and depression. Finally, better cognition functioning was noted with age and in working patients, whereas lower functioning was observed in previous smokers and in patients with ovary/brain metastasis. Conclusions: The 37-item tool is valid and reliable. Questions related to multitasking could be included in the scoring system.

impact of cancer and its treatment on patients' quality of life. This hypothesis was supported by neuroimaging studies that con rmed the correlation between the self-reported cognitive decline and alteration in the central nervous system. Also, self-reported measures were more sensitive in assessing the association between anxiety and depression than neuropsychological tests.
The FACT-Cog scale is a self-assessment scale validated in routine clinical practice to assess cognitive function and quality of life in various cancer populations [5]. It is of particular interest since it focuses on the noticeability and functional interference of multiple speci c domains associated with perceived cognitive functioning [6]. The initial scale consisted of 33 questions evaluating four different components of the perceived cognitive function, i.e., impairments, abilities, comments from others, and the impact on quality of life. In 2016, FACT-Cog scoring directions were updated to include four items related to multitasking (MT), not previously included in the scoring algorithm: "I have trouble keeping track of what I am doing if I am interrupted"; "I have trouble shifting back and forth between different activities that require thinking"; "I am able to shift back and forth between two activities that require thinking"; and "I am able to keep track of what I am doing, even if I am interrupted". However, the internal consistency and correlation coe cients between individual items and the total score were not calculated. Moreover, studies have shown that ethnicity and cultural preferences can affect patients' perception of their cognitive function [7,8]. In the absence of a validated version in Lebanon, it was deemed essential to validate the French version of FACT-Cog, in a country whose second mother language is French, to use it in future epidemiological and clinical studies.
Therefore, this study aimed to validate the 37-item French version of FACT-Cog and assess its correlates in Lebanese cancer patients.

Study Design
A prospective clinical study was conducted between November 2017 until December 2019 at Hôtel-Dieu de France (HDF) Hospital, including 261 cancer patients. Patients had to be over 18 with a primary diagnosis of breast, colorectal, or lung cancer (all stages for all three types) and be treated with chemotherapy to be eligible. Patients were recruited during their outpatient chemotherapy at the daycare hospital.
Non-inclusion criteria consisted of patients with relapse/other types of cancer, who have had neurosurgeries or suffer from disorders of the central nervous system (dementia, multiple sclerosis, epilepsy, Parkinson's disease, and mental retardation) that may affect cognitive evaluation. Patients who received adjuvant hormone therapy (especially for breast cancer patients) were also excluded.
The nal sample was divided into three groups: patients receiving their rst chemotherapy ever, those who have already had several sessions, and those undergoing palliative chemotherapy (for patients requiring more than 10 sessions of chemotherapy). None of the participants received any nancial incentive.

Ethical aspect
The study was approved by Hôtel-Dieu de France Hospital ethical committee (HDF, CEHDF1016, July 2017) and Medical Direction (Protocol N.DAM-2017/288, November 2017). All patients gave their written informed consent before enrollment.

Sample Size Calculation
Comrey and Lee suggested that a minimum of 10 observations per variable is necessary to perform an exploratory factor analysis [9]. Since the FACT-Cog (PCI subscale) is a 20-item questionnaire, a minimum of 200 patients was required for this study. Other subscales have fewer items, and thus necessitate smaller samples.

Sociodemographic information
Clinical and demographic data were collected, including age, gender, weight and height (to calculate the body mass index, BMI), Body Surface Area (BSA, calculated using the Mosteller formula) [10], ethnicity/nationality, marital status, education level, and the use of alcohol, tobacco, and medications.
Cancer-related clinical features were also recorded from patients' medical records. It included information on the type and stage of cancer, metastases, and the number of chemotherapy cycles.

FACT-Cog validation
David Cella, PhD, who holds the copyright of the scale, approved the use of the French and English versions of the Functional Assessment of Cancer Therapy -Cognitive Function (FACT-Cog, version 3) to evaluate cognitive function (Licensing agreement granted on November 2, 2017). The FACT-Cog scale was reliable and valid in assessing the cognitive function before, during, and after chemotherapy, in different cancer populations, including breast, colorectal, and lung cancer (The Functional Assessment of Chronic Illness Therapy system of Quality of Life questionnaires and all related subscales, translations, and adaptations ("FACIT System")).
The questionnaire was administered twice (noted Test and Retest) in 108 patients three weeks apart, corresponding to the time between two sessions of chemotherapy. A trained research assistant performed data collection and made sure that all questions were answered.
FACT-Cog scale scoring This instrument assesses patients' memory, attention, concentration, language, and thinking skills and the impact of cognition disturbances on their quality of life. It consists of 37 questions exploring four different subscales of the cognitive function: perceived cognitive impairments (CogPCI: 20 items); perceived cognitive abilities (CogPCA: 9 items); comments from others (CogOth: 4 items); and the impact of perceived cognitive impairments on quality of life (CogQOL: 4 items). The patient must answer the questions by referring to the last seven days, expressing how many times a given situation has occurred during this period.
The total FACT-Cog score is the sum of the four subscales and ranges from 0-148. The higher the total score, the better the cognitive function, and the lower the impact on patients' quality of life. The detailed FACIT's recommended scoring method is presented in Supplementary le 1.

Other assessment measures
Pain was assessed using the visual analogue scale (VAS) ranging from 0 (no pain) to 10 (maximum pain). The self-report Hospital Anxiety and Depression Scale was used to evaluate anxiety and depression (HADS-A and HADS-D, respectively). Symptoms of the previous week were reported on a scale from 0 (not at all) to 3 (most of the time). Finally, the level of fatigue was measured following the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC-QLQ C30 scale).

Statistical analysis
Statistical analyses were performed using SPSS software version 25.0. Descriptive statistics were calculated for all the variables. The Kolmogorov-Smirnov test veri ed the normality of the variables within each group: all the variables were not normally distributed, except for the total scale at the rst cycle.
Thus, the Spearman's correlation test was used to examine the association between scales and subscales. A p ≤ 0.05 was considered statistically signi cant.
The validity of the subscales' construct in this sample was con rmed, by launching four factor analyses for the items of the subscales. Using the principal component analysis technique, a varimax rotation was applied when extracted factors were not signi cantly correlated, and a promax rotation was applied when factors were correlated.
The Kaiser-Meyer-Olkin (KMO) measurement and the Bartlett sphericity test were performed to ensure the adequacy of the sampling. The number of factors retained corresponded to Eigenvalues greater than one.
Cronbach's alpha was recorded for reliability analysis of the total scale and subscales: α ≥ 0.7 and ≥ 0.8 were considered as acceptable and excellent internal consistency values, respectively [11]. The "testretest" reliability was evaluated by the intra-class correlation coe cient (ICC, mean measurement) for the scores of the scales. Values less than 0.5, between 0.5 and 0.75, between 0.75 and 0.9, and greater than 0.90 were indicative of poor, moderate, good, and excellent reliability, respectively [12].
Finally, studies have shown that self-reported questionnaires, such as FACT-Cog tend not to be associated with neuropsychological performance but rather depression and anxiety [5,6,13]. We therefore performed a convergent validity analysis to explore this hypothesis using Spearman correlations; this measure allowed examining to what extent the FACT-Cog scale/subscales correlated not only with depression and anxiety but also with pain and fatigue. An absolute correlation coe cient value (IRI) of 0.70 and above indicates a strong correlation, a moderate correlation between 0.40 and 0.70, and a weak correlation for values below 0.4 [14].
Multiple linear regression was conducted to answer the secondary objective, using the total cognition scale as a dependent variable; a backward LR method was applied to choose the most parsimonious model and decrease confounding. Assumptions of the model adequacy, linearity, normality, and homoscedasticity were assessed before adopting the nal presented model. Bootstrapping was conducted to improve the stability of coe cients' con dence intervals.  A factor analysis was carried out over the whole sample, using the Varimax rotation since the factors were not highly correlated. A KMO measure of sampling adequacy of 0.870 was found, with a signi cant Bartlett's test of sphericity (p < 0.001). The sample was adequate, and communalities were all higher than 0.3. None of the FACT-CogPCI subscale items were removed; items converged over a solution of ve factors that had an Eigenvalue > 1, explaining a total of 64.39% of the variance (Table 2). Items loaded on ve factors: mental acuity, memory and attention, verbal uency, functional interference, and spatial orientation. Table 2 shows the mapping with the objective cognitive domains of the neuropsychological tests. The Quartimax rotation gave similar results, with slight variation in the loading of one item, the CogM9 that is related to the "memory" domain (See Supplementary le 2).
As for the reliability analysis, results were in the acceptable range according to the number of included items. The only item that showed a low correlation coe cient of 0.272 was the item CogC33c "I have had to use written lists more often than usual so I would not forget things". The Cronbach alpha values of the two newly evaluated items CogMT1 and CogMT2, were 0.522 and 0.420, respectively. (Table 3) (Table 3). As for the reliability analysis, results were in the acceptable range according to the number of included items. The correlation coe cient values of the two newly evaluated items CogPMT1 and CogPMT2, were 0.650 and 0.533, respectively. Table 3 presents the comparison with the objective cognitive domains of the neuropsychological tests: items related to the same domain loaded into the same factor. (Table 4)  were all higher than 0.3. Items loaded on one factor. None of the FACT-Cog OTH subscale items were removed; items converged over a solution of one factor that had an Eigenvalue > 1, explaining a total of 70.28% of the variance ( Table 4). As for the reliability analysis, results were in the acceptable range according to the number of included items. (Table 5) (Table 5).

FACT-CogQOL Test results
As for the reliability analysis, results were in the acceptable range according to the number of included items.

Test-retest analysis of total scale and subscales
High test-retest correlation was found between Cycle 1 and Cycle 2, for the scale and all subscales, except for the QOL subscale that had a relatively lower value (Table 6). A borderline internal consistency was found between subscales (p < 0.700). All correlations were signi cant (p < 0.001) and were moderate to high between the total scale and subscales; however, correlations between subscales were of lower magnitude (Table 7). The convergent validity with pain, fatigue, anxiety, and depression was also evaluated: inverse, weak but signi cant correlations between the FACT-Cog total score/subscales scores were found (IrI = 0.206-0.351; p-values less than 0.05). For the subscales, PCI correlated with depression, OTH with anxiety, and depression, and PCA with fatigue, anxiety, and depression. A detailed description is presented in Table 8.

Correlates of total cognition: multivariable analysis
A multivariable analysis, taking the total cognition score as a dependent variable, showed that higher cognition scores were signi cantly associated with older age (Beta = 0.252) and in those who work compared to those who do not (Beta = 8.415), whereas lower scores were noted in previous smokers versus non-smoker (Beta=-13.484), in patients having ovary metastasis (Beta=-21.285), and brain metastasis (Beta=-8.283) versus those without metastasis (Table 9). To the best of our knowledge, this is the rst study to validate the French version of the 37-item FACT-Cog scale in a population of Lebanese cancer patients, using exploratory factor analysis for the four subscales (FACT-Cog PCI, PCA, OTH, and QOL domains) as per scoring recommendations. Previous validations of the original English, the French, and the Korean version of FACT-Cog were performed on the 33-item scale without the multitasking components nor factor analysis, except for the Korean study [5,15,16]. Thus, comparing our results to theirs was not possible.
Only two studies investigated the 37-items scales [6,17]. The rst examined the psychometric properties and measurement equivalence of the English and Chinese FATC-Cog based on the cognitive domains drawing items from the four subscales [6]. The second performed con rmatory structure analysis of the 37-item scale in three populations: 158 cancer patients, community older adults, and undergraduate students using the scoring recommendation in one of the models [17]. Hence, the need to validate the French 37-item Fact-Cog scale on a large sample of cancer patients, and evaluate the internal consistency and the correlation between individual items and the total score, following the scoring recommendation [5]. In our study, none of the items were removed from factor analyses of the four subscales, consistent with research that con rmed the traditional four-factor structure of the 37-item FACT-Cog [17].
When mapping the loading of items to the objective cognitive domains of neuropsychological tests, items related to the same domain loaded to the same factor for FACT-Cog PCA, OTH, and QOL, but not for the FACT-PCI subscale: questions related to mental acuity and concentration loaded together on the same factor, but questions related to memory and multitasking ability loaded over several factors, and items CogF25 and CogC33a did not load with their related questions over the "functional interference" or "verbal acuity" domains, respectively. Our results are similar to those of the original validation article that identi ed borderline properties for the memory items [5] and to those of the English and Chinese versions of the FACT-Cog, where authors failed at identifying unidimensionality for the memory domain [6]. A possible explanation for such results could be cross-cultural differences and perceptions of these questions.
Furthermore, of all studied items, the only that presented a poor correlation was the CogC33c, related to the memory domain: "I have had to use written lists more often than usual, so I would not forget things". This is not surprising since this item has been previously revised to capture if the patients noted an increase in the use of such methods to help them remember things, not only implying a simple organizational style [5]. However, even after revision, this item might still capture, to some extent, a personality characteristic rather than a cognitive function. Moreover, as stated in both Chinese [6] and Korean [16] validation studies, this item and the CogM12 "I have had trouble remembering new information, like phone numbers or simple instructions" might not be suitable for today's context where technological advances reduce our need to recall information or use to-do lists. Therefore, memory items deserve to be revised again [5].
Also, items related to multi-tasking t their respective subscales (CogMT1 and CogMT2 for FACT-CogPCI; CogPMT1 and CogPMT2 for FACT-CogPCA) with acceptable Cronbach's alpha values for all items except for the CogMT2 that had a value less than 0.5. Furthermore, there was a high correlation between each item of the subscales and the total cognition score.
The scale had excellent internal consistency values: each of the subscales had excellent Cronbach's alpha values (over 0.8), supporting the appropriate reliability of this version. The values for the FACT-CogOTH and FACT-CogQOL were even higher than those reported in the rst validated French version (0.847 versus 0.7, and 0.954 versus 0.85, respectively) [15], and the Korean validation [16]. The test-retest reliability was also appropriate; the ICC between the test and retest was good for the total cognition score and all sub-scores, except for the FACT-Cog-QOL that had poor reliability, showing that the impact of cognition on QOL may differ across patients between chemotherapy cycles [2].

Convergent validity with pain, depression, anxiety, fatigue
Our results demonstrated a weak but signi cant and inverse correlation between the FACT-Cog scores and patients' pain, fatigue, anxiety, and depression, similar to previously reported weak to moderate correlations [5,6] in the 33-item scale, likely due to the multifactorial nature of cognitive decline and the possible interaction between psychological, psychosocial, and demographic factors in the chemo brain [6,18].
However, regardless of the version used, the language, or the number of items assessed in previous studies, our results are overall consistent with previous research, highlighting su cient reliability and validity for FACT-Cog. These results further con rm that the English, French, Chinese, and Korean versions of the questionnaire are effective tools to assess cognitive function in cancer patients at any stage of their treatment [5,6,15,16].

Baseline factors affecting the cognitive function
The mean total FACT-Cog score in our sample was 106.48 ± 21.52, slightly lower than what was published in the English and Chinese 37-item versions (127 ± 19.6 and 126.6 ± 18, respectively) [19]. The lower cognitive function could be due to the difference in the studied population; our sample included patients with breast, colorectal, and lung cancer (17% having metastatic cancer) versus patients with only breast cancer and 7% metastasis. Better cognitive function was noted with younger patients and those who work compared to those who do not; lower capacity was observed in previous smokers versus nonsmokers and patients with ovarian and brain metastasis versus those without metastasis.
Surprisingly, higher cognition scores were signi cantly associated with older age, although aging is a known risk factor for cognitive impairment, especially in older adults with pre-existing cognitive decline [1]. One hypothesis that can explain our results is the exclusion of patients with major cognitive disorders such as dementia or other capacity-limiting disorders preventing patients from completing the questionnaire. Another explanation could be our sample: almost 80% were married, and more than half of the patients were diagnosed with breast cancer. In these women, particularly, studies have shown that psychological distress is higher than in other groups of cancer [20,21], which could impact both subjective and objective measures of cognitive impairment [22,23]. Indeed, patients who had a more altered body image (scarring, hair loss, and weight gain), lower less self-esteem, and lower self-e cacy (mothers with breast cancer, not being able to take care of their families/professional life), had higher levels of anxiety and depression [24][25][26]. In all cases, the FACT-Cog should be administered to a broader age group to examine the exact effect of age over a lifespan and determine the need for adjusting age scores [15].
To the best of our knowledge, no studies have explored the effect of smoking and working on the cognitive function of cancer patients, as evaluated by the FACT-Cog. However, research established the harmful effect of smoking on cognition, with ever-smokers having a reduced cognitive function compared to never-smokers [27][28][29],this risk persisting even after smoking cessation [27].
Recent guidelines have emphasized the importance of physical activity and social rehabilitation, both acquired in the workplace, to improve cognition in cancer patients [30][31][32]. The positive effect could be mediated by several biological mechanisms [33,34] but also by improved psychological factors such as anxiety and depression [32,35,36].
Finally, expectedly, lower FACT-Cog scores were seen in patients with brain metastases since the cognitive decline is among the most reported symptoms [37]. Additionally, cognitive dysfunction in patients presenting ovarian metastases might be consequent to their treatment (adjuvant endocrine therapy for metastatic breast cancer [38][39][40] or targeted therapy such as bevacizumab for metastatic colorectal cancer [41]).

Limitations and Strengths
We acknowledge some limitations related to the study design. In the absence of an Arabic version, we used the French version in a group of French-speaking patients, but some misunderstandings might have happened. Also, we did not include a control group of healthy individuals to explore the normative validation of the scale. Nevertheless, despite all these limitations, and to the best of our knowledge, this study is the largest to validate the FACT-Cog, enrolling a heterogeneous sample of patients with different cancer conditions, treatment statuses, and ages, which allows the generalizability of the results. Moreover, it includes essential components evaluating the desired constructs of the cognitive function [19]. Nevertheless, a study with a larger sample and broader patients' distribution is suggested to con rm our ndings; a con rmatory factor analysis would be interesting to assess the current structure suitability in the French-speaking Lebanese population.

Conclusions
Our study validated the 37-item FACT-Cog tool and con rmed its validity and reliability in a population of Lebanese cancer patients. The four new multitasking questions could be easily included in the new scoring system. In the absence of a validated Arabic version, the French self-reported scale can be easily used in clinical research and practice to optimize the diagnosis and management of cognitive impairment in cancer survivors which could facilitate the pooling of data from multinational studies into a single analytical framework in clinical trials or cognitive research [19].  Availability of data and materials: Any data or material required are available upon demand.
Competing interests: The authors have no con icts of interest to disclose.
Funding: This work was supported by the "Conseil de la recherche" of the Saint-Joseph University (FPH71).