Risk factors for chemotherapy‐induced peripheral neuropathy in patients receiving taxane‐ and platinum‐based chemotherapy

Abstract Background Chemotherapy‐induced peripheral neuropathy (CIPN) is a significant and difficult to manage side effect of neurotoxic chemotherapies. Several risk factors for CIPN have been identified to date, but inconsistencies and methodological limitations exist in past research. Also, a limited number of potential risk factors has been investigated in the past. Aim The objective of this study was to assess the relative contribution of a wider range of risk factors in the development of CIPN. Methods This analysis used the 6‐month data after starting chemotherapy from a larger prospective observational study on CIPN risk, prevalence, and quality of life. Patients were assessed at recruitment for possible CIPN risk factors, including prior history of neuropathies, current/past infectious diseases; neurotoxic medication history; personal and treatment characteristics; smoking history, alcohol use, and vegetable/fruit intake. Neuropathy was assessed at 6‐months after starting chemotherapy with the neuropathy (motor/sensory) items of the NCI‐CTCAE scale and the WHO criterion for neuropathy. Data on symptom burden were also collected. Results Data were available from 255 patients from three cancer centers in Hong Kong, Singapore, and UK. The use of different scales did not always identify the same predictor variables. Key risk factors in multivariate regression models included older age (highest OR = 1.08, p < 0.01 with the WHO scale), chemotherapy (platinum‐based chemotherapy had OR = 0.20–0.27 in developing CIPN compared to taxane‐based chemotherapy), history of neuropathy (for motor CIPN only, OR = 8.36, p < 0.01), symptom burden (OR = 1.06, p < 0.05), number of chemotherapy cycles received (OR = 1.19–1.24, p < 0.01), and alcohol intake (OR = 0.32, p < 0.05). In univariate analysis, the use of statins was implicated with CIPN (p = 0.03–0.04 with different assessments) and diabetes showed a trend (p = 0.09) in the development of CIPN. Conclusion This study confirmed the CIPN risk related to certain variables and identified new ones. This knowledge can assist with treatment decisions and patient education.

CIPN may be challenging for clinicians to diagnose, assess and manage, especially in patients with co-existing or preexisting conditions or disorders that involve the peripheral nervous system (Hausheer, Schilsky, Bain, Berghorn, & Lieberman, 2006). A general predisposition for developing CIPN is observed in nerves previously damaged by diabetes mellitus, alcohol, or inherited neuropathy (Quasthoff & Hartung, 2002). Thyroid dysfunction, metabolic and infectious diseases (i.e., hepatitis B or C, poliomyelitis, HIV), vitamin deficiencies (i.e., B12, B1, B6), and monoclonal gammopathy have also been implicated in the pathogenesis of CIPN (Armstrong, Almadrones, & Gilbert, 2005;Kaley & DeAngelis, 2009). Many medications that are commonly used in cancer patients, such as metronidazole, misonidazole, sulfasalazine, or phenytoin, are all documented to be associated with some degree of peripheral neurotoxicity (Hausheer et al., 2006).
Research around risk factors for CIPN has increased over recent years, although at times findings are inconsistent or a limited pool of potential factors is assessed. In a large study (n = 3,106), worse neurotoxicity was observed in colorectal cancer patients, those with longer duration of cancer, on current therapy, older patients, and in African Americans (Lewis et al., 2015). Being obese and having more insomnia severity, anxiety, and depression were all associated with CIPN in other studies (Bao et al., 2016;Simon, Danso, Alberico, Basch, & Bennett, 2017). Older age, lower income, higher BMI, comorbidities, being born prematurely, higher cumulative dose of chemotherapy, and poorer functional status were also predictive of CIPN (Miaskowski et al., 2017). Diabetes was also shown to be predictor of CIPN (Ottaiano et al., 2016) although other studies have found no such link (Pereira et al., 2016;Simon et al., 2017). However, many of the potential predictors of CIPN have not been fully investigated to date. Hence, the aim of this study was to assess the relative contribution of a wider range of risk factors in the development of CIPN, providing a stronger explanatory model, and further explore the potential link between CIPN and other symptoms.

| Design
This analysis used data from the 6-month CIPN assessment after starting chemotherapy from a larger prospective observational study on CIPN prevalence and quality of life , focusing on one of the primary objectives of the study.

| Sample and settings
The sample included patients receiving platinum-based chemotherapy (primarily cisplatin) and taxane-based chemotherapy

| PROCEDURE S
Eligible patients were identified and approached at hospital outpatients clinics. Those who agreed to participate and provided informed signed consent completed all the baseline measurements including personal characteristics and presence of potential risk factors as identified in the literature. Clinical data were obtained from the medical records as well as information on medication used and past medical history. Participants in the larger project underwent a neuropathy assessment repeated at each cycle of chemotherapy (up to six cycles), 6 months, 9 months, and 12 months postchemotherapy. For the current analysis, data from the 6-month assessment were used as it had the highest number of patients across all Conclusion: This study confirmed the CIPN risk related to certain variables and identified new ones. This knowledge can assist with treatment decisions and patient education.

K E Y W O R D S
cancer, chemotherapy-induced peripheral neuropathy, platinum chemotherapy, risk factors, taxanes assessments and the highest incidence of CIPN. Ethical approval was obtained from each site before commencing the study.

| Neurotoxicity assessment
1. The National Cancer Institute -Common Terminology Criteria for Adverse Events (NCI-CTCAE) version 4.03 is a physicianrated grading system that includes criteria and definitions for quantifying and grading CIPN. This grading scale comprises two items, with a sensory and a motor assessment and utilizes a 5-point scale ranging from grade 1 to grade 5.
2. The WHO criterion is also a physician-rated CIPN item, and includes paresthesia, reflex decreases and extend of motor loss as parameters (WHO, 1979).
These two assessments were completed using both a checklist of neuropathy-related indications and physical/neurological examination to aid in the diagnosis. Also, a new composite variable (combined scale, supported by the combined scale's Cronbach alpha of 0.74, intraclass correlation of 0.74 and item-to-item correlations of 0.41-0.61, p < 0.01) was also created with a combination of the above three items, in order to have maximum variation in the data, as the two scales were identifying varying prevalence of CIPN at different patients (the highest prevalence rate with the WHO criterion item). This combined outcome variable was flagged as having CIPN when at least one of the three items used in the assessment of CIPN indicated so, and it was used in the risk factor analysis.

| Symptom burden
This variable responds to a secondary objective of the study to explore any links between CIPN and other symptoms. In order to estimate symptom burden, we used data from the single-item symptom measures (items 8, 9, 11-25 of the European Organization for Research and Treatment (EORTC) QLQ-C30. It incorporates nine multi-item scales to assess quality of life: five functional scales (physical, role, cognitive, emotional, and social); three symptom scales (fatigue, pain, and nausea and vomiting); and a global health and quality-of-life scale (Aaronson et al., 1993). In order to estimate symptom burden, we used data from the single-item symptom measures (items 8,9, 11-25 of the scale, including pain, tiredness, appetite loss, breathlessness, nausea, vomiting, constipation, diarrhea, cognitive impairment, psychological symptoms) after transforming them to 0-100 scores, thus creating a new predictor variable of "symptom burden". This scale has been validated in China (Wan et al., 2008) and Singapore (Tan et al., 2014). Its Cronbach's alpha in our sample was 0.90.

| DATA ANALYS IS
Descriptive statistics were used to summarize the data. Chi-square analysis assessed differences in categorical variables while Student's t tests were used for comparing continuous variables. Logistic regression models were used for the main risk factor analysis. The relevant covariates for initial model inclusion were identified using a multivariate analysis, with rules (p-values < 0.20) for retaining variables in the model. This was followed by the final model which only included significant (defined above) variables. This is a recommended approach for removing unimportant covariates so that a more manageable set of variables can be used with more complex multivariate statistical techniques (Lee, 2014).
A multilevel logistic regression analyses took place taking account of center effect and time since last cycle of chemotherapy, to develop the predictive model for CIPN. Data were analyzed using SPSS v.21.

| Sample characteristics
Data from 255 participants were available for analysis at the 6month assessment of CIPN (chosen as this point had the highest TA B L E 1 Chemotherapy-induced peripheral neuropathy and its risk factors in categorical variables (n = 255)  Note: Each regression model is presented in one column, the dependent variable is shown in the first row of the table and all variables were put in the regression as independent variables *p < 0.05. **p < 0.01. chemotherapy; those with history of neuropathy had higher risk for (motor) CIPN, as well as older patients. Symptom burden had some contribution to (primarily to sensory) CIPN. Number of chemotherapy cycles received was also a strong predictor of CIPN. One unit of alcohol use decreased the risk of CIPN by 68% (only in the combined scale).

| D ISCUSS I ON
This study assessed CIPN clinical risk factors using a prospective design and a wide range of potential predictors. Overall CIPN incidence was lower in this study than that reported in the literature, and this has to do probably with the scales used; past studies have used quality of life scales to estimate CIPN, which often include a range of general/broader items to indicate neuropathy.
Also, clinician-based assessments, such as the NCI-CTCAE tend to underestimate CIPN incidence (Dorsey et al., 2019). We have explained these reasons in more detail in the parent larger study . However, in a systematic review it was shown that CIPN incidence at 6 months was 30% (Seretny et al., 2014) and our incidence in the combined tools was 26%. Key risk factors identified include older age, history of neuropathy, symptom burden, alcohol intake (cautiously accepted as a risk factor in this study due to the small number of events needing further clarification in the future) and number of chemotherapy cycles used.
Patients receiving platinum-based chemotherapy had 17%-27% less chance of developing CIPN compared to those receiving taxane-based chemotheraopy. Risk factors were not always consistent across the scales used. This may reflect sensitivity or reliability issues with the various scales measuring CIPN. As the measurement tool(s) used in future risk factor research will be related with the identification of specific risk factors, it is important to use the most reliable and valid CIPN scale (Cavaletti et al., 2013;Dorsey et al., 2019) or a combination of scales to maximize the "pick up" rates of these tools that will include both patient-reported outcomes and objective CIPN indicators, such as with the Total Neuropathy Score clinical version (TNSc) (McCrary et al., 2017).
Older age somewhat contributed to CIPN (6% more chances), supporting findings from past research (Bandos et al., 2018;Hershman et al., 2016;Miaskowski et al., 2017). History of neuropathy was a potential risk factor for motor neuropathy and its ORs were high in the other CIPN scales (but did not reach statis-  (Gaist et al., 2002). However, more recent work from a case-control study showed that ever use of statins was not associated with a higher risk of polyneuropathy (Svendsen et al., 2017 has been discussed as a potential mechanism for behavioral toxicities (Vichaya et al., 2015). It will be interesting to explore these assumptions in the future more concretely and have a more in-depth understanding of the link between symptoms/symptom burden and CIPN. This finding is further supported by recent research showing that patients with CIPN had significantly poorer functional status (Miaskowski et al., 2017).
The role of (chronic) alcohol use in the development of CIPN is less clear, as contradictory findings have been presented in the literature, probably due to the inherent problems in measuring alcohol use accurately. Our findings suggest that no alcohol use had some protective effect in CIPN, but this was not consistent across all the scales used. Also, our sample had very few heavy drinkers and this may have impacted on the results. Alcohol use (as well as diabetes) may be associated with the development of neuropathy before the chemotherapy, and we have seen that preexisting neuropathy was a key CIPN risk factor.
The number of chemotherapy cycles received was a strong predictor both in univariate and multivariate analyses. This is not linked with cumulative dose (as the latter was not shown to be predictive of CIPN in our study). Hence, this finding may imply that "time" after starting chemotherapy may be strongly linked with the development of CIPN, suggesting that CIPN is time-dependent rather than dosedependent, although the link between cumulative dose and CIPN has been reported in past literature but not consistently (Seretny et al., 2014).
Two parameters in the final predictive model need some more consideration in the development of CIPN. Firstly, the role of hepatitis (possibly as a result of taking neurotoxic antiviral agents in the past or even as a result of disturbance in the pharmacokinetics of the chemotherapy drugs, i.e., decreased liver function and/or increased drug exposition). Secondly, current smoking with perhaps its connection with pain pathways. Both of them had very high odds ratios (3.93 and 1.18-2.54, respectively) but both these ORs were not statistically significant, highly likely as a result of the small number of patients reporting these two variables (n = 13 and 7, respectively).
Future research should provide more insight about the potential risk for CIPN for hepatitis and smoking status.
Some variables in the study had small frequency counts, and this may affect the interpretation and generalizability of the results and should be perceived as preliminary only. Identification of risk factors may assist the clinician to make chemotherapy treatment decisions accordingly in order to minimize not only the development of CIPN but also the morbidity and health care utilization linked with higher incidence of CIPN (while clinical effectiveness is not compromized).
However, the state of science in this area is not yet optimal for such clinical decisions, and more research in elucidating strong CIPN-related risk factors is needed, including the development of predictive models. Other consistent risk factors, such as higher BMI and obesity were not assessed in this study and these should be included in future models.
This study confirms the role of (older) age; number of chemotherapy cycles received, and type of chemotherapy as key CIPN

ACK N OWLED G M ENTS
We thank all the patients for their participation in the study. This study was supported by a Polytechnic University Direct grant (in HK) and an NCIS Seed Funding Grant, National Medical Research Council (Singapore).

CO N FLI C T O F I NTE R E S T
The authors hereby certify that we have all seen and approved this manuscript. We guarantee that the paper is the authors' original work and that it has not been the subject of prior publication and is not under consideration for publication elsewhere. On behalf of all the co-authors, the corresponding author bears full responsibility for the submission. There are no financial or other relationships that might pose a conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.