Clinical and Genetic Predictors of Paclitaxel Neurotoxicity Based on Patient-Versus Clinician-Reported Incidence and Severity of Neurotoxicity in the ICON7 Trial

S. B. Park; J. B. Kwok; R. Asher; C. K. Lee; P. Beale; F. Selle; M. Friedlander


Ann Oncol. 2017;28(11):2733-2740. 

In This Article


In the present study, paclitaxel was associated with significant and long-lasting neuropathy, with considerable discordance identified between patient and clinician reports of neuropathy. Given the lack of substantive agreement between patient- and clinician-reported neuropathy in the present study, it is not surprising that there was often little overlap between clinical and genetic risk factors. These findings raise the important question of whether patient rather than clinician-reported toxicities are more appropriate measures of CIPN in studies investigating predictors such as genetic polymorphisms and risk of neuropathy.

While we identified significant discrepancy between CRN and PRN, the rates of CRN during treatment were consistent with previous studies.[1,19] Prior studies reported 1-year probability of residual CRN of 24%–33%,[19,20] similar to our finding of 27.3%. In contrast, PRN was evident in 90.6% of patients and did not resolve by 18-month follow-up in a significant proportion (54%). Likewise, a substantial proportion of paclitaxel-treated patients report ongoing PRN (50%–60%) in the 3 to 6 years after treatment.[4,21] Although a limitation of our study is that the cohort is a subset of the entire intention-to-treat ICON7 population, this suggests that the cohort is likely comparable to the wider population. Further, quality of life analyses in 890 ICON7 patients at 12-months follow-up indicated that PRN persisted in a broader ICON7 cohort, with average neuropathy corresponding to 'quite a bit' on the OV28 scale.[14]

These findings underscore the importance of including patient-reported adverse effects to provide comprehensive patient-centred assessment of CIPN. Direct comparison of PROs and clinician-based toxicity ratings have demonstrated poor to moderate agreement across a range of adverse effects, with patients often reporting more severe symptoms.[22] Although there is similarity between CIPN severity reported by clinicians and patients,[23,24] low concordance has been identified in previous reports,[25,26] similar to the current study. The reasons for this are multifactorial and include the detailed scope of PROs, improved sensitivity and responsiveness to change.[23,24] Clinician-based grading scales are limited by a lack of sensitivity and inter-observer reliability which may further impact on comparability to PROs.[5] Additionally, future studies that examine the use of tools such as balance and gait assessments to provide objective evidence of neuropathic dysfunction[27] will assist to improve replication and validation in risk factor analyses.

Risk factors associated with the development of neuropathy differed between CRN and PRN, suggesting that careful examination of CIPN outcome measures is required to accurately ascertain risk factors. Older age was associated with both CRN and PRN in dose-to-event analyses. Older age has previously been identified as a risk factor for CIPN in taxane-treated patients with 4% increased odds of neuropathy per year[28] or 13% increased risk of CIPN per decade,[6] respectively, identified in different cohorts. Older age has also been associated with neuropathy persistence.[29] CRN was also linked to bowel resection and PRN was associated with the volume of residual disease following debulking surgery. These factors have not previously been linked to neuropathy and may instead reflect association with other patient characteristics.

CRN was associated with bevacizumab treatment, which as a single agent is not associated with significant neuropathy.[30] Concomitantly with paclitaxel, bevacizumab was associated with greater neuropathy in ovarian cancer patients in the AURELIA trial although the discrepancy was attributed to longer chemotherapy exposure in bevacizumab-treated patients.[31] The addition of bevacizumab to carboplatin/paclitaxel treatment did not significantly alter the proportion of ovarian cancer patients with neuropathy in GOG218.[29] Bevacizumab has been associated with increased risk of neuropathy in conjunction with paclitaxel in metastatic breast cancer,[32] also related to higher response rates and longer duration of treatment. However, a retrospective study of metastatic breast cancer patients treated with bevacizumab and paclitaxel demonstrated increased risk of severe neuropathy with bevacizumab despite the same duration of treatment.[33] In contrast, a reduced risk of neuropathy with bevacizumab in addition to paclitaxel/cisplatin has been reported in cervical cancer.[34] Detailed examination of the potential role of bevacizumab in exacerbating paclitaxel-induced neuropathy is warranted. In addition, it should be noted that carboplatin is an independently neurotoxic chemotherapy[35] and may contribute to the present neurotoxicity profile.

PRN was linked to additive MAPT SNPs in dose-to-event analyses, while CRN was linked to additive GSK3B SNPs. The MAPT polymorphisms rs242557 and rs1052553 increase MAPT transcription and hence tau production.[16] Increased tau expression may promote paclitaxel binding to microtubules[36] and enhance stabilisation which could exacerbate the impact of paclitaxel on microtubule function and axonal transport. Likewise, GSK3B rs6438552 was associated with reduced risk of CRN in a prior study[10] and functions to increase tau phosphorylation, effectively reducing microtubule stability.[37] While these associations remained significant in multivariable analyses, there may also be additional functional polymorphisms relevant to patient phenotyping, which should be examined in further cohorts with appropriate outcome measures.

Prior studies have identified SNPs in other microtubule-associated genes linked to CIPN. TUBB2A SNPs (rs909964, rs909965) were protective against paclitaxel-induced neuropathy[9] but not in a separate cohort[38] or in docetaxel-treated patients.[39] In the present study, TUBB2A SNP rs909961 (highly correlated with rs909964; SNP Annotation and Proxy Search) did not demonstrate association with PRN or CRN. Similarly, CEP72 (rs924607) was previously linked to vincristine-induced CIPN[11] although not in all cohorts[40] and not in the present analysis. These findings underscore the difficulties in comparing neurotoxicity risk across different patient cohorts. Selection of phenotyping tools is a central issue in CIPN polymorphism studies. Predominantly clinician-based toxicity grading has been utilised to define neuropathy in prior studies, although exclusive use of CTCAE toxicity scales in pharmacogenomic studies has been identified as leading to poor reproducibility and reliability.[41] There have been difficulties in validation of SNPs as risk factors for CIPN across multiple cohorts,[6,42] potentially due to differential assessment methods and differences in case identification.

Our findings demonstrate that variability between CIPN grading between studies, particularly with regards to PRO selection, may contribute to the discordance between studies examining CIPN risk factors. Further, discordance between clinician and PRN may impact on paclitaxel dose reduction and delays in treatment, with unknown long-term consequences. The fundamental question remains as to whose opinion counts more - patients or clinicians? PROs are an essential component to future analyses of CIPN, particularly in studies of genetic risk factors. These results also underscore the importance of standardised assessment protocols for CIPN in order to enable identification of patients at risk of persistent neuropathy and to facilitate trials of interventions to prevent neuropathy.