Correlates of Taxane-Induced Neuropathy, an Electronic Health Record Based Observational Study

Simple Summary Chemotherapy-induced peripheral neuropathy (CIPN) is a common adverse effect of taxane chemotherapy. We investigated the associations of demographic and treatment variables with the risk of developing CIPN using electronic health record data. We found higher doses, more taxane cycles, female sex, overweight and obesity, or a history of diabetes were associated with developing CIPN in patients treated with taxane, while concurrent chemotherapy or concurrent radiotherapy were related to a reduced risk. Further studies are needed to uncover the underlying reasons for the observed associations. Abstract Background: Chemotherapy-induced peripheral neuropathy (CIPN) is a common therapeutic complication affecting cancer patients’ quality-of-life. We evaluated clinical characteristics, demographics, and lifestyle factors in association with CIPN following taxane treatment. Methods: Data were extracted from the electronic health record of 3387 patients diagnosed with a primary cancer and receiving taxane (i.e., paclitaxel or docetaxel) at Vanderbilt University Medical Center. Neuropathy was assessed via a validated computer algorithm. Univariate and multivariate regression models were applied to evaluate odds ratios (ORs) and 95% confidence intervals (CIs) of CIPN-associated factors. Results: Female sex (OR = 1.28, 95% CI = 1.01–1.62), high body-mass index (BMI) (OR = 1.31, 95% CI = 1.06–1.61 for overweight, and OR = 1.49, 95% CI = 1.21–1.83 for obesity), diabetes (OR = 1.66, 95% CI = 1.34–2.06), high mean taxane dose (OR = 1.05, 95% CI = 1.03–1.08 per 10 mg/m2), and more treatment cycles (1.12, 95% CI = 1.10–1.14) were positively associated with CIPN. Concurrent chemotherapy (OR = 0.74, 95% CI = 0.58–0.94) and concurrent radiotherapy (OR = 0.77, 95% CI = 0.59–1.00) were inversely associated with CIPN. Obesity and diabetes both had a stronger association with docetaxel CIPN compared to paclitaxel, although interaction was only significant for diabetes and taxane (p = 0.019). Increased BMI was associated with CIPN only among non-diabetic patients (OR:1.34 for overweight and 1.68 for obesity), while diabetes increased CIPN risk across all BMI strata (ORs were 2.65, 2.41, and 2.15 for normal weight, overweight, and obese, respectively) compared to normal-weight non-diabetic patients (p for interaction = 0.039). Conclusions: Female sex, obesity, and diabetes are significantly associated with taxine-induced CIPN. Further research is needed to identify clinical and pharmacologic strategies to prevent and mitigate CIPN in at-risk patient populations.


Study Variables and Covariate Assessment
The study variables and covariate assessment were defined in a previous study with the same study population [18]. Briefly, taxane-treatment information was extracted from medication administration records, including overall taxane-treatment start date, mean dosage, mean treatment duration (months), and number of treatment cycles. We focused on only the first taxane treatment course for each individual. Cancer type was determined from the most recent diagnosis code prior to cancer treatment and grouped by anatomical site. Cancer-stage data were retrieved from the VUMC North American Association Central Cancer Registry.
Demographic information was retrieved from EHRs and included: self-reported race, age at first taxane treatment, sex, most recent body mass index (BMI, weight in kilograms/height in centimeter 2 ) measurement prior to first taxane treatment, ever smoking status, and history of diabetes. Age at first taxane treatment was categorized into <55, 55 to <65, 65 to <75, or ≥75. BMI was categorized into underweight (BMI < 18.5), healthy weight (BMI ≥ 18.5 to < 25), overweight (BMI ≥ 25 to <30), or obese (BMI > 30). Due to the small sample size for underweight patients (n = 102), they were excluded from the current study. Additional covariates retrieved from the EHRs included prior and concurrent chemotherapy (Table S1). Prior chemotherapy was defined as any chemotherapy regimen prior to the first taxane treatment period. Concurrent chemotherapy was defined as any chemotherapy regimen during the first taxane treatment period prior to the first incidence of CIPN.

Study Endpoint
The primary study endpoint was any occurrence of CIPN during and up to 6 months after completion of an individual's first taxane treatment course. We restricted the study period to six months to minimize the influence of loss to follow-up. We defined CIPN as any grade 2 or 3 CIPN, which is defined by the U.S. National Cancer Institute's Common Terminology Criteria for Adverse Events, v5.0 as "moderate to severe symptoms that limit instrumental activities of daily living". CIPN is typically documented in the free text of clinical notes but differentiation between mild CIPN vs. grade 2 or 3 is often not clearly indicated. We trained an L1-regularized logistic regression model to predict CIPN status using a subset of 242 individuals with a manual review of CIPN assessment by a clinician. Variables used to train the model included diagnosis codes, key terms extracted from clinical notes with natural language processing, and changes in taxane dosage, which may indicate dosage adjustments for patients that experience CIPN. We evaluated the area under the receiver operating characteristic curve (AUROC) with five-fold cross-validation. The five-fold cross-validation of the neuropathy status prediction model had a mean AUROC of 0.861 (range: 0.715 to 0.962). Additional details on the model can be found in the Appendix A. CIPN can take months to resolve, and patients may not report symptoms until after the end of treatment regimens [4,19]. Therefore, we applied the CIPN prediction model to all 3387 patients with data from the start of taxane treatment up to six months after the end of taxane treatment.

Statistical Analysis
Pearson's χ 2 tests for categorical variables and Student's t tests for continuous variables were performed to compare characteristics of patients who received taxane treatment by incidence of CIPN. Logistic regression models with different adjustments were applied to evaluate odds ratios (ORs) and 95% confidence intervals (CIs) of CIPN associated with categorical age, sex, self-reported race, categorical BMI, history of diabetes, ever smoking status, co-chemotherapy, and co-radiotherapy. All categorical variables were treated as dummy variables in the analysis. Additional multivariate analyses were performed to evaluate associations of CIPN with categorical BMI and history of diabetes with further stratification by taxane drug type (paclitaxel vs. docetaxel) and sex (male vs. female) to evaluate effect modification and joint effects of these variables. Lastly, we evaluated the joint effect and interaction between categorical BMI and history of diabetes on CIPN. p-value for interactions was derived from calculating the p-value of the F-test comparing the reduced model without interaction term and the full model with the interaction term. All statistical tests were based on two-tailed probability and a significance level set at alpha (α) less than 0.05. Statistical analyses were performed in R v.3.6.1.

Results
Selected characteristics of the 3387 study participants on taxane regimens stratified by presence of CIPN are shown in Table 1. Paclitaxel was more frequently used than docetaxel (75.6% vs. 24.4%) among our study population. The mean treatment dose of docetaxel was 70.9 mg/m 2 (standard deviation (SD) = 11.7), and the mean treatment dose of paclitaxel was 85.5 mg/m 2 (SD = 51.3). Among all study participants, 54.9% of patients received taxane treatment in combination with a platinum agent and 16.8% in combination with a nonplatinum agent. A greater proportion of patients received a platinum agent (70.2%) than non-platinum agent (23.8%) prior to taxane treatment. Approximately a quarter (23.6%) of participants received radiotherapy prior to taxane treatment. A similar proportion (21.2%) received radiotherapy concurrently with taxane treatment. Breast cancer (33.4%) was the most common cancer type among study participants, followed by head and neck cancer (21.7%). Approximately half of the participants ever smoked (45.0%), and 16.5% of patients had a history of diabetes. Compared to participants who did not experience CIPN, participants who experienced CIPN were, on average, younger (mean 58.5 vs. 59.5; p = 0.043) and had higher BMIs (mean 29.2 vs. 27.9; p < 0.001). In addition, greater proportions of participants with CIPN were female (68.5% vs. 55.0%; p < 0.001) and self-identified as Black (12.3% vs. 9.8%). Patients with CIPN were more likely to have a history of diabetes (19.8% vs. 14.8%; p < 0.001) and less likely to have ever smoked (41.8% vs. 46.7%; p < 0.001). Among cancer types, breast cancer was more heavily represented in the group of participants with CIPN (40.4% vs. 30.0%) but a higher proportion of participants without CIPN had head and neck (25.4% vs. 14.0%) or lung and other respiratory (19.4% vs. 15.6%) cancers. Participants with CIPN were more likely to have stage I cancer (13.3% vs. 10.3%) and less likely to have stage IV cancer (17.5% vs. 24.8%).
There was a non-significant difference in the type of taxane treatment received by participants with and without CIPN. Among paclitaxel recipients, those with CIPN received higher average doses of paclitaxel compared to those who did not experience CIPN (mean 94.2 vs. 81.1 mg/m 2 ; p < 0.001). In contrast, docetaxel recipients without CIPN had higher average doses than those with CIPN (mean 71.4 vs. 69.6 mg/m 2 ; p = 0.038). Interestingly, patients treated with prior (65.5% vs. 72.5%) or concurrent (49.5% vs. 57.5%) platinumbased chemotherapy regimens had a lower proportion of participants with CIPN. Likewise, there were lower proportions of participants with CIPN who had concurrent (13.5% vs. 25.0%) or prior (18.0% vs. 26.3%) radiation therapy.
When stratifying by paclitaxel vs. docetaxel therapy (Table 3), we observed that the association between BMI and CIPN was stronger among docetaxel recipients (OR = 1.79; 95% CI = 1.15 to 2.78, for overweight and OR = 1.77; 95% CI = 1.14 to 2.74 for obese patients) than paclitaxel recipients (OR = 1.43, 95% CI = 1.13 to 1.81 for overweight and OR = 1.43; 95% CI = 1.13 to 1.81 for obese patients). However, the interaction test was only significant between taxane type and diabetes (p = 0.019). The odds ratio for CIPN in patients with a history of diabetes who received docetaxel (OR = 2.63; 95% CI = 1.63 to 4.22) was greater than that for patients with history of diabetes who received paclitaxel (OR = 1.50; 95% CI = 1.17 to 1.92).
We also examined the associations between CIPN with BMI and history of diabetes stratified by sex ( to 2.44) than male patients with a history of diabetes (OR = 1.49; 95% CI = 1.07 to 2.08) when compared to their counterparts without a history of diabetes. However, there was no significant multiplicative interaction between sex and BMI (p = 0.931) or history of diabetes (p = 0.269).  Table 4 shows a joint association between categorical BMI and history of diabetes. Among patients with no history of diabetes, overweight and obese patients had higher odds of CIPN than normal-weight patients with ORs of 1.34 (95% CI = 1.07 to 1.68) and 1.68 (95% CI = 1.34 to 2.09), respectively. Normal-weight patients with a history of diabetes were at higher odds of CIPN than normal-weight patients without a history of diabetes (OR = 2.65; 95% CI = 1.62 to 4.35). Likewise, overweight and obese patients with a history of diabetes had higher odds of CIPN compared to normal-weight patients without a history of diabetes with ORs of 2.41 (95% CI = 1.65 to 3.52) and 2.15 (95% CI = 1.59 to 2.91), respectively. Notably, these ORs were not higher when compared to normal-weight patients with a history of diabetes across BMI strata. In other words, when comparing the OR of healthy weight individuals with the OR of overweight (2.41/2.65 = 0.91) or obese patients (2.15/2.65 = 0.81), high BMI was not associated with higher CIPN risk among patients with a history of diabetes. Tests for multiplicative interaction between BMI and history of diabetes were significant (p = 0.039). n: counts of patients in category with neuropathy/total number of patients in category; OR: adjusted for age, sex, race, ever smoking status, mean taxane dose, number of taxane cycles, co-chemotherapy, prior chemotherapy, co-radiotherapy, prior radiotherapy, cancer type, and cancer stage; p-value for interaction derived from calculating the p-value of the F-test comparing the reduced model without interaction term and the full model with the interaction term.

Demographic Correlates
We found several factors associated with increased prevalence of CIPN in univariate analysis among all patients within the cohort, such as cancer type and cancer stage at first taxane treatment. Among patients with CIPN, breast as well as head and neck cancers were the predominant cancer types. Additionally, patients with stage I cancer were more likely to have CIPN compared to patients with stage IV disease. This is likely the result of a combination of factors including cancer screening to detect early-stage breast cancers, early symptom development with head and neck cancers, and aggressive treatment with standard dose intensity for curative intent in all cancer types. We also found that 50% of patients with female genital and reproductive tract (i.e., gynecologic) cancers developed CIPN while only 30% of patients with non-gynecologic cancers developed CIPN. This is likely related to the high dose of paclitaxel used in adjuvant (175 mg/m 2 ) and maintenance (135 mg/m 2 ) therapy for ovarian cancers [20,21]. It is worth noting that none of these associations retained statistical significance in the multivariate analysis.
Our univariate analysis also showed a modest increase in the prevalence of CIPN based on patient demographic groups including Black patients compared with White patients and in female patients compared to their male counterparts. Previous studies have reported conflicting results regarding race as a risk factor for CIPN [22,23]. In our study, Black race was no longer a risk factor for CIPN after adjusting for age and sex. Of note, female sex remained significantly associated with higher risk of CIPN in the multivariate analysis, albeit an OR4 drop to 1.28 from the unadjusted OR1 of 1.78. The underlying mechanism for this association remains unknown. Treatment characteristics, including mean taxane dose and number of cycles, had a positive association with the development of CIPN, as reported elsewhere [7].

Treatment Regimen Correlates
Unexpectedly, we found that taxane treatment concurrent with chemotherapy or radiotherapy was associated with a lower risk of CIPN and the associations persisted even after adjustment for dose and cycle of taxane treatment as well as several other clinical factors. Platinum agents have been previously reported to be associated with increased risk of CIPN [24]. However, patients treated with co-chemotherapy consisting of a taxane and platinum agent versus taxane alone had a lower risk developing CIPN despite having similar mean taxane dose and cumulative taxane dose. Differences in dosing schedules, treatment adherence, and medication contradictions, as well as clinical and patient characteristics (e.g., physical fitness) uncaptured by our study may have contributed to these correlative associations. Additional research, particularly in clinical-trial settings, is needed to draw a conclusion.

Correlates among Diabetic and Obese Patients
Our findings agree with several previous studies in supporting diabetes as an important risk factor for taxane-induced CIPN [2,13]. Adjusting for all other variables, including age, sex, race, BMI, lifetime smoking incidence, mean taxane dose, number of taxane cycles, concurrent chemotherapy, concurrent radiotherapy, or prior chemotherapy, we found that taxane-treated patients with a history of diabetes were more than 1.6 times as likely to develop CIPN when compared to non-diabetic patients. This association was consistently observed across BMI categories. Obesity was an independent risk factor for developing CIPN only among non-diabetic patients. Furthermore, the association of diabetes and taxane-induced CIPN was stronger among docetaxel than among paclitaxel treated patients.
Separating the effect of BMI from a history of diabetes is paramount due to the high correlation between these two factors. BMI is an established risk factor for developing type 2 DM (T2DM) [25], and the prevalence of adults with diabetes increases from 8% for normal-weight individuals to 43% for individuals with a BMI > 40 [26]. Of note, one quarter of patients with DM will develop diabetic sensorimotor polyneuropathy (DPN). As a result, patients with DM or pre-existing DPN of any grade are often excluded from clinical trials containing CIPN-inducing therapies to avoid its confounding effect [2,27]. Due to these complications, the association between diabetes and CIPN warrants further evaluation. In our study, we found that BMI was associated with CIPN only among non-diabetic patients, while diabetes was associated with CIPN across all BMI categories. Interestingly, our data showed that compared to normal-weight non-diabetic patients, normal-weight diabetic patients had a higher OR of CIPN than overweight or obese diabetic patients. Unfortunately, information on DPN was not available, which precluded us from separating DPN from CIPN. It is possible that some of the patients in our cohort with diabetes may have had DPN prior to receiving taxane. Furthermore, it is unclear whether adherence to pharmacologic diabetes management may have impacted CIPN. We were also unable to accurately classify types of DM in our study (e.g., type 1 DM from type 2 DM, insulin dependent vs. non-insulin dependent). Avoiding taxane regimens can lead to inferior progression-free survival and overall survival in specific patient populations [28], therefore there is a critical need to delineate further associations between DPN and CIPN while exploring prevention strategies for patients at high risk for CIPN. Future studies evaluating CIPN in DM patients should account for type of diabetes, time since diagnosis, type of diabetes management, medication adherence, HbA1c%, and duration of treatment, as these associations may provide clinical guidance and influence therapeutic decisions.

Strengths and Limitations
The strengths of our study include the large sample size, inclusion of multiple cancer types and all stages of disease, and consideration of multiple treatment modalities including concurrent combination chemotherapy and concurrent radiotherapy.
Our study had several limitations. First, our study was based on data from electric health records that were not specifically designed for research. Information misclassification, such as un-documented or mis-reported smoking behavior and comorbidities, may lead to biased estimation of true associations. Other uncaptured confounders, as mentioned above, may also result in false associations. As discussed, our cohort was limited by an inability to classify type of diabetes. However, given the high prevalence of obesity and diabetes in our geographic region, type 2 DM was suggested to account for 90-95% of diabetics in our patient population [25,29]. Absence of other DM-specific variables, including presence of DPN, time since diabetes diagnosis, glycemic control, medication adherence, HbA1c%, and duration of treatment are also limitations of our study. Additionally, CIPN was identified from EHR using a machine-learning model, which likely introduced some misclassifications. However, our model was validated by manual review and 32% of patients in our assessed cohort developed CIPN, which aligns with previous studies reporting 27% of patients reporting CIPN [2]. Lastly, we focused primarily on the first taxane exposure and did not include patients who may have received prior taxane chemotherapy.

Conclusions
We found that obesity, a history of diabetes, and female sex were significantly correlated with taxane-induced CIPN. Patients with both cancer and diabetes had more than a two-fold increased risk of developing CIPN after receiving taxane treatment compared to non-diabetic, normal-weight cancer patients. BMI was also associated with increased risk for CIPN among non-diabetic cancer patients. Race was not an independent risk factor for CIPN. Our study findings provide important information for clinical management of patients with cancer during and after taxane treatment. Studies focusing on investigating underlying biological mechanisms for obesity, diabetes, and sex associations are needed to develop preventive and therapeutic strategies to mitigate taxane-induced CIPN.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers15030754/s1, Table S1: List of chemotherapy agents evaluated.; Table S2: Mean and cumulative taxane dose when administered with co-chemotherapy. Informed Consent Statement: Patient consent was waived. This study exclusively utilized information extracted from medical records in the Vanderbilt University Medical Center "Synthetic Derivative" database (SD). The SD is a de-identified copy of the main hospital medical record databases created for research purposes. De-identified datasets are exported directly from the system into a statistical/analysis program. As no HIPAA identifiers are available in the SD database, and the proposed project does not plan to re-identify these records using the identified VUMC database, this study meets criteria for non-human subjects research.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author conditioning on compliance with the VUMC data sharing policy.

Conflicts of Interest:
The authors of this paper declare no conflict of interest that contributed to this original manuscript.

Appendix A.
Appendix A.1. Machine-Learning Model to Identify Chemotherapy-Induced Peripheral Neuropathy from Electronic Health Records We identified 3577 individuals who had undergone taxane treatments from electronic health records (EHRs) in Vanderbilt University Medical Center's Synthetic Derivative. Information about each individuals' taxane treatment was retrieved from the EHRs, such as type of taxane, length of treatment period, number of treatment cycles, dosage at each administration, and date of administration. A clinician reviewed a subset of 242 patients and identified grade 2 chemotherapy-induced peripheral neuropathy (CIPN).
Terms related to neuropathy were extracted from the individual's clinical notes during their taxane treatment period with the Clinical Language Annotation, Modeling, and Processing (CLAMP) toolkit [30]. CLAMP outputs a flag that indicates whether the extracted term was negated and will, where possible, include other modifiers in the extracted term, such as severity (e.g., mild neuropathy) or location (e.g., neuropathy in hands). We grouped the extracted neuropathy terms into four severity categories based on the term modifiers with case-insensitive regular expressions, as shown in Table A1. As our goal was to identify grade 2 and above CIPN, we grouped grade 2 and grade 3 CIPN in the "Severe" category. All terms not classified as "mild", "moderate", or "severe" a Neuropathy terms modifiers can include location, e.g., 'neuropathy in leg'.
We additionally separated the neuropathy terms into 'present' and 'absent' based on the CLAMP negation flag, giving eight term categories: mild-present, moderate-present, severe-present, other-present, mild-absent, moderate-absent, severe-absent, other-absent. For instance, a term such as 'worsening neuropathy' would be classified as moderate-present whereas a term such as 'no grade 2 neuropathy' would be classified as severe-absent ( Figure A1).

Persistent Ongoing
Leg a Painful a Neuropathy terms modifiers can include location, e.g., 'neuropathy in leg'.
We additionally separated the neuropathy terms into 'present' and 'absent' based on the CLAMP negation flag, giving eight term categories: mild-present, moderate-present, severe-present, other-present, mild-absent, moderate-absent, severe-absent, other-absent. For instance, a term such as 'worsening neuropathy' would be classified as moderatepresent whereas a term such as 'no grade 2 neuropathy' would be classified as severeabsent ( Figure A1).