Lagged Relationships Among Chemotherapy-Induced Peripheral Neuropathy, Sleep Quality, and Physical Activity During and After Chemotherapy

Abstract

Background

Chemotherapy-induced peripheral neuropathy (CIPN) is well-documented and can become chronic for up to a third of patients. CIPN management is hampered by limited pharmacological options. Thus, identifying modifiable behaviors that influence CIPN may help inform future interventions.

Purpose

The purpose of the current study was to examine bidirectional relationships between sleep quality, physical activity, and CIPN during and after chemotherapy.

Methods

Participants were 138 women with gynecologic cancer (M age = 61, 94% white, 96% non-Hispanic), collected as part of an ongoing study. Assessments occurred at postcycle 1, postcycle 6, and 6- and 12-month postchemotherapy. CIPN (EORTC-CIPN20), sleep quality (PSQI), and physical activity (IPAQ) were assessed via self-report. Objective physical activity was assessed via wrist actigraphy. Latent change score models were used to examine lagged relationships between CIPN, sleep quality, and physical activity pairs.

Results

Over the study period, sleep quality was found to contribute to CIPN (p = .001), but not the reverse (p > .05). Bidirectional relationships were observed between CIPN and both objective and subjective walking (ps ≤ .001). Illustrations of these relationships showed that patients with less CIPN early in treatment demonstrate more substantial increases in walking over time, while those with higher CIPN demonstrate more consistent levels of walking during and after treatment.

Conclusions

These findings suggest that worse sleep quality and lower walking levels may contribute to the course and maintenance of CIPN. Future investigation should evaluate the impact of early interventions aimed at improving sleep quality and encouraging physical activity in patients treated with chemotherapy.

Chemotherapy-induced peripheral neuropathy during treatment and in the year post-chemotherapy is worse among patients with poor sleep quality and low walking levels

Introduction

Chemotherapy-induced peripheral neuropathy (CIPN) is a common, intrusive side effect of cancer treatment that is characterized by pain, numbness, and tingling in the hands and feet. For up to a third of patients, CIPN can become a chronic condition [1]. Current understanding of CIPN pathophysiology is limited, there are no known prevention strategies for CIPN, and effective pharmacological management strategies are limited [2]. CIPN negatively impacts quality of life and has significant clinical implications, as uncontrolled CIPN can necessitate modification or discontinuation of potentially life-saving chemotherapy regimens [3, 4]. Thus, modifiable treatment targets that ease the impact of CIPN are needed.

One way to conceptualize future intervention targets for CIPN is to consider modifiable behaviors that contribute to the course and maintenance of CIPN over time. Theoretically, cancer patients may either engage in behaviors at the beginning of therapy that make them more susceptible to developing CIPN, or they may adopt unhelpful behaviors to manage their CIPN symptoms that ultimately perpetuate CIPN into a chronic phase. For example, chronic CIPN has been associated with poor sleep quality and low physical activity. However, literature describing these behaviors in CIPN is sparse, and what literature does exist is primarily cross-sectional [5]. Consequently, it is difficult to establish the directionality of relationships among CIPN and modifiable behaviors.

Because data evaluating the directionality of modifiable behaviors is scarce in CIPN, and because CIPN has a significant pain component, we examined prior studies in chronic pain to inform hypotheses in this study. Literature suggests that bidirectional relationships may exist between sleep quality, physical activity, and pain. For example, subjective patient reports in clinic often attribute poor sleep and lack of exercise to their pain, rather than the reverse. Explanations like the fear-avoidance model [6] support this idea, suggesting that chronic pain patients may avoid activities that they fear exacerbate their pain, which results in a vicious cycle of increased pain and inactivity. However, it is also possible to conceptualize that worse sleep may increase pain sensitivity. Though limited, there is literature to indicate that the influence of sleep on pain symptoms may be stronger than the influence of pain on sleep disruption in cancer patients [7]. Similarly, physical inactivity leads to muscle deconditioning, which may contribute to increased pain when attempting to use weakened muscles. Consistently, in CIPN specifically, recent studies assessing activity habits and exercise-based interventions in cancer patients report that CIPN improves in patients who maintain recommended levels of physical activity [5, 8, 9]. Cross-sectional evidence suggests that meeting physical activity guidelines is inversely associated with peripheral neuropathy symptoms in both ovarian [5] and colorectal cancer survivors [8]. Additionally, a recent evaluation of a 6-week progressive walking and resistance exercise program in 355 cancer patients undergoing chemotherapy indicated that exercise reduced CIPN symptoms in the hands and feet, particularly for older patients, male patients, and patients with breast cancer [9]. Taken together, this research suggests that there may also be bidirectional relationships between CIPN, sleep quality, and physical activity.

Thus, the purpose of this longitudinal study was to evaluate directional relationships between three variable pairs in gynecological cancer patients treated with chemotherapy. CIPN was paired with (i) self-reported sleep quality, (ii) self-reported walking, and (iii) average step counts measured by actigraphy. Lagged behavior changes were examined using latent change score (LCS) models [10], an analytic strategy used to describe reciprocal influences between pairs of variables. We conducted a dual coupling bivariate LCS analysis for each variable pair with the hypothesis that for each pair, a form of dynamic coupling exists (e.g., CIPN and the behavior influence subsequent changes in one another over time.)

Methods

Participants

Participants were recruited as part of an ongoing, larger IRB-approved study examining side effects of chemotherapy in patients with gynecologic cancer (IRB #5797, Sickness Behaviors in Gynecologic Cancer Patients Treated with Chemotherapy, PI: Jim). Patients were recruited prior to the start of chemotherapy (i.e., first line or later). To be eligible, participants were required to: (i) be 18–89 years of age; (ii) be diagnosed with gynecologic cancer (i.e., ovarian, endometrial, peritoneal, fallopian, uterine, or vaginal); (iii) be scheduled to undergo intravenous or intraperitoneal chemotherapy at Moffitt Cancer Center (MCC); (iv) have not undergone chemotherapy or radiation in the month prior to enrollment in the current study; (v) not have psychiatric or neurological disorders that would interfere with study participation (e.g., dementia, psychosis) documented in the medical record or observable during study enrollment; (vi) have no reported or documented diagnosis of immune-related disease (e.g., HIV, systemic lupus erythematosus, rheumatoid arthritis); (vii) not be pregnant; (viii) be able to speak and read English; and (ix) be able to provide informed consent. Participants were recruited between August 2013 and July 2018.

Procedures

Potential participants were identified by their treating physicians at MCC and were contacted via phone or in person by a trained research assistant in order to determine initial eligibility and interest in the study. Potential patient participants who met eligibility requirements were recruited in person during a regularly scheduled outpatient appointment with the medical oncologist, at which time written informed consent was obtained. Data analyzed in the current paper were collected at four timepoints: (i) following the first chemotherapy cycle (postcycle 1), (ii) following the sixth chemotherapy cycle (postcycle 6), (3) 6 months postchemotherapy (6 mo postchemo), and (iv) 12 months postchemotherapy (12 mo postchemo). These timepoints were chosen in order to capture changes in CIPN and behavior throughout treatment and into the year following the conclusion of chemotherapy at approximately 6-month intervals. CIPN was not assessed at recruitment since participants starting first-line therapy were chemotherapy-naïve at that time.

Measures

Demographic and Clinical Data

Demographic characteristics were assessed via patient report. These characteristics included age, race/ethnicity, marital status, education level, and income. Clinical characteristics were obtained by a medical record review, which included cancer type, stage, recurrence status, and prior chemotherapy.

Chemotherapy-Induced Peripheral Neuropathy

The European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire, Chemotherapy-Induced Peripheral Neuropathy-20 (EORTC QLQ-CIPN20) is a 20-item quality of life questionnaire developed to assess symptoms and functional limitations related to CIPN [11]. The overall score is comprised of sensory, motor, and autonomic symptoms. Participants rated their symptoms over the past week on a 1 (not at all) to 4 (very much) scale. Studies assessing the QLQ-CIPN20 have revealed strong evidence that the measure distinguishes between contrasting groups (e.g., those who have received neurotoxic chemotherapy vs. those who have not), acceptable validity (alpha coefficients of .88, .88, and .78 for the sensory, motor, and autonomic subscales, respectively), and responsiveness to change over time [12]. For the purposes of this study, the total QLQ-CIPN20 score was used. Alpha coefficients for the overall score ranged from .78 to .91. Higher scores indicate worse CIPN.

Sleep Quality

The Pittsburgh Sleep Quality Index (PSQI) [13] is a 19-item questionnaire assessing sleep quality and patterns in adults. The measure is comprised of seven components: subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disruption, use of sleep medications, and daytime dysfunction over the past month. Each component is scored from 0 to 3, in which 0 indicates “no difficulty” and 3 indicates “severe difficulty” with that aspect of sleep quality. The components then sum to provide a global assessment of sleep quality. For the purposes of this study, the global PSQI score was used. Continuous global scores were used in this analysis and range from 0 to 21, with higher scores indicating worse sleep. The PSQI has been demonstrated to have excellent sensitivity and specificity in distinguishing good and poor sleepers [13] and has been validated for use in cancer patients [14], with a Cronbach’s alpha of .77–.81. In this study, alpha coefficients for the global score ranged from .57 to .71.

Subjective Physical Activity: Walking

The International Physical Activity Questionnaire – Short Form (IPAQ-SF) [15] measures health-related physical activity. The IPAQ-SF assesses sedentary behavior, walking, moderate-intensity exercise, and vigorous-intensity exercise. Participants are instructed to indicate whether they engaged in each type of physical activity over the past week, and if so, to estimate the amount of time typically spent engaging in that type of physical activity in hours or minutes. Results are then computed continuously by weighting each type of activity by its energy requirements (defined in metabolic equivalents, or METs).

Because walking is the primary physical activity among cancer patients in this sample, the walking subscale was used in this study. In the walking subscale, participants are instructed to indicate whether they engaged in walking for at least 10 min at a time over the past week (including for work and at home, to travel from place to place, and any other walking for recreation, sport, leisure, and exercise). Higher scores represent more self-reported walking.

Objective Physical Activity: Average Steps/Minute (Actigraphy)

Wrist actigraphy (ActiGraph, Pensacola, FL) was used to assess physical activity levels [16]. Participants wore the actigraph continuously on the nondominant wrist for a week at each timepoint. The actigraph uses an accelerometer to monitor and store the degree and intensity of activity in 60-s epochs. Each participant’s actigraph was visually inspected for completeness, and the Troiano 2007 algorithm available via the ActiLife Wear Time Validation tool used to identify and exclude any invalid data from further analysis. Actigraph data were included in this study if the participant wore the monitor for at least 72 consecutive hours. Wear times (periods in which a subject was wearing the actigraph) were validated for each file, and the physical activity data was scored to yield step counts per minute. Because rates of moderate to vigorous physical activity were low, step counts were used as the primary measure of objective physical activity.

Statistical Analysis

The statistical analyses were conducted in two phases. First, we examined longitudinal changes in each of the variables using univariate LCS models. In these models, a symptom or behavior score is considered to be the sum of the prior assessment of that variable plus a latent change score. Next, we evaluated bivariate LCS models that estimate longitudinal changes in each pair of variables, but also evaluate the lead-lagged relationships among CIPN/behavior pairs [10]. For example, relationships between sleep quality and subsequent CIPN may be examined at multiple timepoints with LCS modeling, and these models will take into account changes between each of the timepoints. The univariate and bivariate LCS models use all available data for each participant, making them advantageous for longitudinal studies. Finally, all analyses were conducted using MPlus [17] with syntax generated by Zhang [18].

Results

Sample Descriptives

A study flow diagram is shown in Fig. 1. Sociodemographics and clinical characteristics are reported in Table 1. The sample included in analyses (total N = 138) had a mean age of 61.0 years (range 38–83 years). Most participants were non-Hispanic (96%), white (94%), and married (71%). Over a third of participants held a college degree (35%), and 68% reported household incomes of $40,000/year or more. Ovarian cancer was the most common cancer diagnosis (52%), followed by endometrial cancer (24%), and most patients had advanced (stages III–IV) cancer (70%). No associations were observed between baseline CIPN and cancer stage (p > .05). Approximately 40% of patients were being treated for a recurrent cancer diagnosis, with the majority of recurrence patients having undergone one previous line of chemotherapy (51%). No differences were observed in an ANOVA comparing baseline CIPN within recurrence patients as a function of the number of previous chemotherapy regimens (p > .05). All participants completed at least 3 days of actigraphy, and as such, no participants were excluded on the basis of incomplete actigraphy data.

Fig. 1.

Study flow diagram.

Table 1:

Sociodemographics and clinical characteristics

Age: M (SD)	61.0 (10.3)
Race: N (% white)*	127 (94.1)
Ethnicity: N (% non-Hispanic)	129 (96.3)
Marital status: N (% married)	94 (70.7)
Education: N (% college graduate)	48 (35.3)
Household income: N (% >$40k)	71 (68.3)
Cancer type
Ovarian: N (%)	66 (52.4)
Endometrial: N (%)	30 (23.8)
Other: N (%)	30 (23.8
Cancer stage
Early (I & II): N (%)	35 (30.2)
Advanced (III & IV): N (%)	81 (69.8)
Recurrence: N (%)	51 (40.2)
Lines of previous chemotherapy
1: N (% of recurrence)	26 (51.0)
2: N (% of recurrence)	11 (21.6)
≥3: N (% of recurrence)	14 (27.5)

Note: Percentages based on patients with available data at time of analysis.

Univariate Models of CIPN and Behavior

To examine variability in each behavior and CIPN independently, univariate LCS models were created. Table 2 displays the parameter estimates for each of the univariate models. As a guide to this table, the intercept mean (μ₀), which represents the value of each measure at the time of the first assessment (postcycle 1); the slope mean (μ_s), which represents constant changes over time; proportion (β), which denotes changes that are dependent on prior levels of CIPN or behavior; the intercept variance (σ²₀), which indicates the extent of individual differences in individual CIPN and behavior; the slope variance (σ²_s), which represents constant change over time in these measures; and residual variance (σ²_ξ), which represents the amount of unexplained variance.

Table 2:

Univariate model results^a

Variable	Self-reported CIPN EORTC-CIPN20	Self-reported sleep quality PSQI	Self-reported walking IPAQ walk METs	Average steps/minute actigraphy
Measure
Intercept mean μ0	13.33 (1.08)*	8.63 (0.34)*	87.23 (10.66)*	5.61 (0.17)*
Slope mean μs	25.53 (2.41)*	0.58 (2.12)	18.95 (25.76)	0.14 (1.42)
Proportion β	−1.12 (0.09)*	−0.18 (0.27)	−0.11 (0.27)	0.07 (0.23)
Intercept variance σ20	60.49 (21.84)*	8.42 (1.91)*	7768.783 (2080.85)*	1.76 (0.48)*
Slope variance σ2s	285.73 (58.06)+	0.62 (0.73)	1577.28 (552.78)+	0.09 (0.66)
Residual variance σ2ξ	84.76 (10.62)*	6.34 (0.77)*	8571.33 (1006.35)*	1.83 (0.25)*

Note: Unstandardized estimates and standard errors (in parentheses) shown. ⁺p < .01; * p ≤ 0.001.

^aThe intercept mean (μ₀) represents the value of each measure at the time of the first assessment (postcycle 1). The slope mean (μ_s) represents constant changes over time. Proportion (β), denotes changes that are dependent on prior levels of CIPN or behavior. The intercept variance (σ²₀) indicates the extent of individual differences in individual CIPN and behavior. The slope variance (σ²_s) represents constant change over time in these measures. Residual variance (σ²_ξ) represents the amount of unexplained variance.

Results of the univariate LCS models indicated that CIPN changed significantly over time. From Fig. 2, scores increased from Visit 1 to Visit 6 and then appeared to plateau. For the behavior variables, despite a lack of statistically significant slope or proportional change components, significant intercept and residual variance were observed, suggesting that there was individual variability in the sleep quality and walking levels of the sample. Estimated means from the LCS models are shown in Fig. 2 (panels b–d).

Fig. 2.

Implied means from the univariate LCS models: (a) CIPN-20 score, (b) sleep quality, (c) self-reported walking, (d) objective walking.

Bivariate Models of CIPN/Behavior Pairs

Results of each of the dual coupling models are shown in Table 3. In this table, each behavior is represented as X, while CIPN is always represented as the Y variable. In this table, unstandardized estimates and standard errors are shown for each directional path. The X to Y path represents the influence of the behavior on subsequent CIPN, while the Y to X path represents the influence of CIPN on subsequent behavior. If both paths are found to be significant, a dual coupling model is accepted. If only one path is significant, a single coupling model in which there is a directional relationship from one variable to the other is accepted. If neither path is significant, there is no coupling supported by the models. For illustrative purposes, significant results of the dual coupling analyses are depicted in Fig. 3a–e. These figures depict longitudinal relationships based on the initial assessment.

Table 3:

Bidirectional model results^a

Variable X	Variable Y	X to Y path	p	Y to X path	p
Sleep quality (PSQI)	CIPN	3.47 (1.00)	0.001	−0.04 (0.03)	0.22
Objective walking (Actigraphy)	CIPN	−4.64 (1.34)	0.001	0.06 (0.01)	<0.001
Self-reported walking (IPAQ)	CIPN	−0.15 (0.07)	0.024	3.50 (1.30)	0.007

Note: Unstandardized estimates and standard errors shown.

^aIn this table, each behavior is represented as X, while CIPN is always represented as the Y variable. Unstandardized estimates and standard errors are shown for each directional path. The X to Y path represents the influence of the behavior on subsequent CIPN, while the Y to X path represents the influence of CIPN on subsequent behavior. If both paths are found to be significant, a dual coupling model is accepted. If only one path is significant, a single coupling model in which there is a directional relationship from one variable to the other is accepted. If neither path is significant, there is no coupling supported by the models.

Fig. 3.

Graphs illustrating dynamic relationships between CIPN and behavior pairs. As a guide to interpreting these figures, for a given pair of symptoms, the y-axis indicates model-implied sample means of the lagging variable when initial levels of the lead variable are varied by half a standard deviation, and initial sample means for the lagged variable are kept constant. (a) Sleep quality and lagged CIPN, (b) objective walking and lagged CIPN, (c) CIPN and lagged objective walking, (d) self-reported walking and lagged CIPN, (e) CIPN and lagged self-reported walking.

In the CIPN/sleep quality pair, a unidirectional relationship was observed in which poor sleep quality was a leading indicator of subsequent changes in CIPN (p = .001), but not the reverse (p = .22). Higher scores on the PSQI, indicative of worse sleep quality, had a unique influence on subsequent CIPN scores and resulted in an increase in CIPN symptomatology. For illustrative purposes, Fig. 3a shows initial sample means for sleep quality varied by half a standard deviation and initial sample means for CIPN held constant (i.e., CIPN scores at average, higher, and lower levels of sleep quality).

When considering the relative influence of walking and CIPN, a dual coupling model was supported for both objective and subjective walking pairs (all p’s < .05). An inverse relationship was observed in the walking to lagged CIPN path, where more walking predicted lower CIPN ratings at subsequent timepoints (objective: p = .001, subjective: p < .05). When examining CIPN as a predictor of subsequent physical activity, a positive relationship emerged such that higher CIPN was associated with more lagged walking (objective: p < .001, subjective: p < .01). Illustrations of these relationships (Fig. 3b–e) showed that patients with less CIPN at the start of treatment demonstrate more substantial increases in walking over time (particularly after treatment conclusion), while those with higher CIPN may have more consistent levels of walking during and after treatment (i.e., CIPN scores at average, higher, and lower levels of walking, and Fig. 3c and e shows estimated walking levels at average, higher, and lower levels of CIPN).

Discussion

The goal of this study was to examine lagged relationships among CIPN and sleep quality, objective step counts, and subjective walking during and in the year after gynecologic cancer patients underwent chemotherapy. A directional relationship between sleep quality and CIPN emerged, such that worse sleep quality predicted higher subsequent CIPN scores but not vice versa. Our hypotheses were confirmed for objective and subjective measures of walking, in which a dual coupling relationship between low physical activity and CIPN was supported.

Taken together, these results suggest that there are specific, modifiable behavioral indicators of subsequent CIPN—worse sleep quality and low physical activity. Further research is needed to understand the mechanisms underlying observed relationships between CIPN, sleep, and physical activity levels, particularly in older cancer patients. For example, premorbid physiological and behavioral factors prior to the onset of chemotherapy (e.g., existing sleep impairments and/or reduced physical activity) may provide insight into mechanisms that initially place patients at risk for CIPN. Additionally, though sleep quality and measures of walking activity were not associated with each other in this study, prior research in cancer survivors indicates that physical inactivity and worse sleep quality may also be related [19]. Thus, an evaluation of potential shared underlying mechanisms is an avenue for continued research.

There are multiple behavioral interventions that may be useful to evaluate in future research. For example, cognitive-behavioral therapy for insomnia (CBT-I) has been shown to be useful in cancer patients, improving insomnia and sleep quality for up to 12 months [20, 21]. As an additional benefit, CBT-I also enhanced mood, general and physical fatigue, and global and cognitive dimensions of quality of life in those patients. Given the results of this study, we posit that early CBT-I for patients with poor sleep quality may help to mitigate subsequent CIPN, whether administered prior to chemotherapy or posttreatment for chronic CIPN. However, CBT-I has not been tested specifically in cancer patients with CIPN and as such, requires additional research.

As for physical activity, exercise-based interventions are also supported for managing CIPN symptoms in cancer patients who maintain recommended levels of activity [8]. Though it is unclear whether walking has a similar effect on CIPN as other, more vigorous types of exercise, this study suggests that simple, everyday levels of activity—e.g., maintaining higher average step counts—may be enough to influence subsequent CIPN symptomatology. Thus, in addition to exercise, specific skills to maintain good overall activity levels—for example, activity pacing to avoid overexertion, graded increases in activities to avoid under-exertion—may be useful to target in this group.

Finally, a recent pilot study provides preliminary evidence that a self-guided online resource for cognitive-behavioral strategies designed specifically for CIPN may be helpful in easing worst pain intensity in patients with chronic, painful CIPN [22]. This study included modules targeting physical activity and sleep hygiene strategies, among others. Thus, though literature is limited in this area, there is promising preliminary evidence that CBT strategies may have a positive impact on cancer patients with CIPN. However, future research is needed to evaluate CBT in a larger sample in order to generalize findings to patients with CIPN, as well as to establish ideal delivery (e.g., guided vs. self-directed), maximize impact on quality of life, and evaluate the specific impact of behavioral components.

Strengths of this study include a longitudinal design, allowing for CIPN symptom and behavioral assessment through the entire active treatment phase and into a year posttreatment, and advanced statistical techniques to model lagged changes in CIPN and behavior over time. Use of wearable sensors is another strength of this study, as objective evidence of step counts illuminated bidirectional relationships with CIPN symptoms without relying on patient recall. Limitations of the study include a relatively homogenous patient group comprised of primarily white, married, well-educated patients. These factors may limit the generalizability of findings. Actigraph-generated sleep quality variables were not available, limiting the sleep quality data in this study to self-report (unlike the walking data, for which objective and subjective measures were assessed). Additionally, although all participants received cytotoxic chemotherapy, there was heterogeneity in the chemotherapy regimens themselves, including dose and duration of treatment, as well as history of previous cancer and chemotherapy treatments. These factors may influence the rates of CIPN and health behaviors in this sample, despite the fact that there were no differences in baseline CIPN as related to their prior chemotherapy. We opted to use unadjusted models that do not control for these clinical variables. We believe that this decision enhances the generalizability of the LCS models to women undergoing cytotoxic chemotherapy for a variety of gynecological cancer diagnoses, often with complex prior histories and treatment courses. However, it is certainly possible that clinical variables influence CIPN severity and behavior change in ways not accounted for in these models, particularly in the unexpected direction of CIPN and lagged physical activity results. Thus, future investigations should explore these relationships further to ascertain particular clinical variables relating to CIPN and predisposing and/or perpetuating health behaviors.

In summary, the current study is the first to examine lagged relationships among CIPN and modifiable behaviors in gynecological cancer patients treated with chemotherapy. Findings suggest that poor sleep quality and low physical activity may contribute to the course and maintenance of CIPN during chemotherapy and throughout the first-year posttreatment. In light of these results, it is possible that early intervention on these behavioral targets may help to mitigate the impact of CIPN symptoms over time. Further research should replicate these findings in patients of other cancer and chemotherapy regimens in order to establish common and effective management strategies, which in turn will likely improve patient’s quality of life. Additionally, research should be conducted to evaluate the impact of early interventions aimed at mitigating CIPN during and after treatment (e.g., CBT-I, graded exercise).

Compliance with Ethical Standards

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards HSLJ: consultant for RedHill Biopharma, Janssen Scientific Affairs, Merck. Grant funding from Pfizer, Kite Pharma. The other authors declare that they have no conflict of interest.

Authors’ Contributions All authors participated in: 1) conception and design, acquisition of data, or analysis and interpretation of data; 2) drafting and/or critically revising the article for important intellectual content; and 3) final approval of the version to be published.

Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.