Clinical trial design and analysis for comparing three treatments with intra-individual right- and left-hand data

Abstract

Background:

Chemotherapy-induced peripheral neuropathy can occur in the right and left hand. Studies on prevention treatments for chemotherapy-induced peripheral neuropathy have largely adopted either self-controlled designs or parallel designs to compare two preventive treatments. When three treatment options (two experimental treatments and a control treatment) are available, both designs can be extended. However, no clinical trials have adopted a self-controlled design to compare three prevention treatments for chemotherapy-induced peripheral neuropathy. The incomplete block crossover design for more than two treatments can be extended to compare three treatments in the self-controlled design. In simple extension, some of the participants receive two experimental treatments in both hands; however, it may be difficult to administer different experimental treatments in both hands for practical reasons, such as a concern for the different types of unexpected adverse events. This study proposes a design and analysis method appropriate for the situation where only one experimental treatment is provided to each participant.

Methods:

We assume clinical trials to compare each of the two experimental treatments (E₁ and E₂) with the control treatment (C) and between two experimental treatments only when both experimental treatments are superior to the control treatment. We propose a self-controlled design, which equally randomizes to four arms to adjust for the dominant hand effect: Arm 1: E₁ for right hand, C for left hand; Arm 2: C for right hand, E₁ for left hand; Arm 3: E₂ for right hand, C for left hand; and Arm 4: C for right hand, E₂ for left hand. We compare operating characteristics of the proposed design with the three-arm parallel design in which the same treatment is performed in both hands by participants. We also assess three proposed analysis methods for comparisons between experimental treatments in the self-controlled design under several conditions of correlations between right and left hands using simulation studies.

Results:

The simulation studies showed that the proposed design was more powerful than the three-arm parallel design when correlation was 0.3 or higher. For comparisons between experimental treatments, the methods based on the regression model, including the outcome of hands with C as a covariate, had the highest power under modest to high correlation among the analysis methods in the self-controlled design.

Conclusion:

The proposed design can improve the power for comparing between two experimental treatments and the control treatment. Our design is useful in situations where it is undesirable for participants to receive different experimental treatments in both hands for practical reasons.

Keywords

Self-controlled design correlated data three treatments supportive care

Introduction

Chemotherapy-induced peripheral neuropathy (CIPN), which can occur in the right and left hand, is a common dose-limiting adverse event in patients receiving cancer drugs such as taxanes, platinum-based drugs, vinca alkaloids, thalidomide, and bortezomib.^1,2 CIPN can also have a negative influence on the quality of life (QOL) of cancer patients.³ Several studies on prevention treatments for CIPN have been reported.^4–10 These trials have predominantly adopted either self-controlled designs^4–7 or parallel designs,^8–10 to assess the efficacy of a preventive treatment (e.g. cryotherapy vs placebo) or compare treatments (e.g. cryotherapy and compression therapy). In the self-controlled design, participants received different treatments to each hand, which can mitigate confounding by individual differences in sensory detection and drug metabolism.⁴ In the parallel design, participants were assigned to either of two treatment arms and received the same treatment in both hands. When three treatment options are available, both of two designs can be extended. For example, the CONTRoL trial¹¹ is a randomized, controlled, clinical trial with parallel design to select the best intervention from three treatments: cryotherapy, compression therapy, and loose glove/sock (control treatment). Participants are randomly assigned to one of three treatments and receive the same treatment in both hands. However, to the best of our knowledge, no clinical trials have adopted a self-controlled design to compare three prevention treatments for CIPN.

The incomplete block crossover design for more than two treatments¹² can be extended to a self-controlled design that assigns each hand to treatments, that is, the period effect corresponds to the dominant hand effect. In this extension, participants receive two out of three treatments in the right and left hands (six combinations) when three treatments (e.g. two experimental treatments and the control treatment) are compared; participants are randomly assigned to one of six arms and receive different treatments in each hand. However, if the standard treatment is established or the placebo is included as the control treatment, it may be difficult to provide different experimental treatments in both hands for practical reasons, as the effectiveness and safety of these treatments have not been confirmed. In this study, we focus on the situation in which it is undesirable for participants to receive different experimental treatments in both hands.

We assume the clinical trials to compare each of the two experimental treatments with the control treatment on symptoms that can appear in both hands. In this comparison, both experimental treatments with different risk-benefit balances may be an option in the future, regardless of the result of the comparison between experimental treatments. For example, cryotherapy and compression therapy have different risk-benefit balances regarding the incidence of adverse events and costs.⁷ In addition, we are interested in determining which experimental treatment is more effective. Thus, the comparison between two experimental treatments is also performed if both experimental treatments are superior to the control treatment. In this study, we propose a self-controlled design and statistical analysis methods appropriate to this situation. Then, we compare operating characteristics of the proposed design with the three-arm parallel design in several situations of correlations between right and left hands by using a simulation study.

Methods

We consider a clinical trial comparing the effects of each of the two experimental treatments (E₁ and E₂) to the control treatment (C) and between experimental treatments on symptoms that can appear in both hands. The purposes of this trial are to determine (1) whether each of the two experimental treatments is superior to the control treatment and (2) which experimental treatment is more effective if both experimental treatments meet the minimum effect size. In the “Study designs” section, we present the proposed design (self-controlled design) for these purposes and the design used in the previous three-arm parallel trial. In the following section, we describe statistical analysis methods for continuous and binary outcomes in each design.

Study designs

In this study, we distinguish between right and left hands, acknowledging that the baselines of the dominant and non-dominant hands are different (dominant hand effect). For simplicity, all participants are assumed to be right-handed; thus, it is necessary that the right hand read as the dominant hand and balancing between dominant and non-dominant hands is important in each comparison. We propose a self-controlled design, which equally randomizes to four arms: Arm 1: E₁ for right hand, C for left hand; Arm 2: C for right hand, E₁ for left hand; Arm 3: E₂ for right hand, C for left hand; and Arm 4: C for right hand, E₂ for left hand (Figure 1: top figure).

A previous trial¹¹ adopted a parallel design, which equally randomizes to three arms, where the same treatment is performed in both hands of participants: Arm 1: C in both hands, Arm 2: E₁ in both hands, and Arm 3: E₂ in both hands (Figure 1: bottom figure).

Statistical analysis methods

To compare each experimental treatment with the control treatment, the paired tests were used in both outcomes, that is, the paired t test for continuous outcome and McNemar’s test for binary outcome. To compare between experimental treatments, three methods are considered in each outcome type.

In the following, we explain the settings; then, we describe proposed analysis methods for the self-controlled design in continuous and binary outcomes. Next, we describe the analysis methods for the parallel design and perform all comparisons in each outcome using the generalized estimating equations. Finally, we describe the method for multiplicity adjustment.

Settings

Let the total sample size be $n$ . The continuous outcome or binary outcome variables $y_{ij}$ are obtained in each hand $j$ (1: right hand, 2: left hand) by participant $i$ $(i = 1 to n)$ . The outcome variables ${Y_{i 1}, Y_{i 2}}$ follow the bivariate normal distribution with a mean of $(τ_{ij} + λ_{j})$ and standard derivation of $σ$ for continuous outcome, or they follow the bivariate Bernoulli distribution with parameter of $(τ_{ij} + λ_{j})$ for binary outcome, where $τ_{ij}$ is the baseline parameter ( $τ_{E_{1}}$ for E₁, $τ_{E_{2}}$ for E₂, and $τ_{C}$ for C), and $λ_{j}$ is the dominant hand effect. The $g_{i}$ is Arm $m$ of participant $i$ and $i : g_{i} = m$ indicates a subset of participants who are assigned to Arm $m$ , which corresponds to the numbers of each arm in Figure 1, that is, $m = 1 to 4$ in the self-controlled design, and $m = 1 to 3$ in the parallel design.

Analysis of self-controlled design for continuous outcome

To compare E₁ with C, we analyze the outcome data on Arms 1 and 2 with a paired t test without distinguishing which hand receives E₁. The mean difference $(τ_{E_{1}} - τ_{C})$ is estimated as follows

\frac{1}{n / 2} (\sum_{i : g_{i} = 1} (y_{i 1} - y_{i 2}) + \sum_{i : g_{i} = 2} (y_{i 2} - y_{i 1}))

As the allocation ratio of each arm is the same, the dominant hand effect is offset. The same method can be applied to the comparison between E₂ and C.

We consider three analysis methods to compare E₁ with E₂. The first method (hereinafter called “the t-test method”) only uses outcome data on hands with the experimental treatments: right hands in Arms 1 and 3, and left hands in Arms 2 and 4. The means are calculated as follows

\frac{1}{n / 2} (\sum_{i : g_{i} = 1} y_{i 1} + \sum_{i : g_{i} = 2} y_{i 2}) for hands with E_{1}

$and$

\frac{1}{n / 2} (\sum_{i : g_{i} = 3} y_{i 1} + \sum_{i : g_{i} = 4} y_{i 2}) for hands with E_{2}

These values are then compared using Student’s t test. The mean difference $(τ_{E_{1}} - τ_{E_{2}})$ is estimated as follows

\frac{1}{n / 2} (\sum_{i : g_{i} = 1} y_{i 1} + \sum_{i : g_{i} = 2} y_{i 2}) - \frac{1}{n / 2} (\sum_{i : g_{i} = 3} y_{i 1} + \sum_{i : g_{i} = 4} y_{i 2})

The second method (hereinafter called “the delta method”) compares improvements of E₁ and E₂ from C. Let $Δ_{i}$ be the difference of outcome data between hands with an experimental treatment (E₁ or E₂) and C in each participant $i$ ; $Δ_{i} = y_{i 1} - y_{i 2}$ for $i : g_{i} = 1 or 3$ and $Δ_{i} = y_{i 2} - y_{i 1}$ for $i : g_{i} = 2 or 4$ . The mean differences for $Δ_{i}$ are calculated as

\frac{1}{n / 2} \sum_{i : g_{i} = 1, 2} Δ_{i} for E_{1} and \frac{1}{n / 2} \sum_{i : g_{i} = 3, 4} Δ_{i} for E_{2}

These values are then compared using Student’s t test. The mean difference $(τ_{E_{1}} - τ_{E_{2}})$ is estimated as follows

\frac{1}{n / 2} \sum_{i : g_{i} = 1, 2} Δ_{i} - \frac{1}{n / 2} \sum_{i : g_{i} = 3, 4} Δ_{i}

The third method (hereinafter called “the regression method”) uses the general linear model, including outcome data of the hand with C as a covariate. Let $Y_{i}'$ be the outcome data of the hand with E₁ or E₂; $Y_{i}' = Y_{i 1}$ for $i : g_{i} = 1 or 3$ and $Y_{i}' = Y_{i 2}$ for $i : g_{i} = 2 or 4$ . The $x_{i 1}$ indicates which experimental treatment is received; E₁: $x_{i 1} = 0$ for $i : g_{i} = 1 or 2$ and E₂: $x_{i 1} = 1$ for $i : g_{i} = 3 or 4$ . The $x_{i 2}$ is the outcome of the hand with C; $x_{i 2} = y_{i 2}$ for $i : g_{i} = 1 or 3$ and $x_{i 2} = y_{i 1}$ for $i : g_{i} = 2 or 4$ . The general linear model is represented as $E [Y_{i}'] = β_{0} + β_{1} x_{i 1} + β_{2} x_{i 2}$ for balanced data (equal randomization). For comparison between experimental treatments, the Wald test is used to test the hypothesis $H_{0} : β_{1} = 0$ versus $H_{1} : β_{1} \neq 0$ . The mean difference between experimental treatments is represented by $β_{1}$ .

Analysis of self-controlled design for binary outcome

To compare E₁ with C, we analyze the outcome data on Arms 1 and 2 with the McNemar test without distinguishing which hand receives E₁. The same method can be applied to the comparison between E₂ and C. The details of the equations are omitted in this section, as the same formulations used for continuous outcome can also be used for binary outcome.

To compare E₁ with E₂, we consider three analysis methods. The first method (hereinafter called “the chi-square test method”) only uses outcome data on hands with experimental treatments. The proportion of occurring events in each experimental treatment is calculated using the same equation as for continuous outcome. These values are then compared with a chi-square test.

The second method (hereinafter called “the delta method”) compares improvements in E₁ and E₂ from C. The $Δ_{i}$ , which takes three values of ${- 1, 0, 1}$ in the binary outcome, are compared between experimental treatments with the Wilcoxon sum rank test.

The third method (hereinafter called “the regression method”) uses a generalized linear model with the identity link function, including outcome data of the hand with C as a covariate. For comparison between experimental treatments, the Wald test is used to test the hypothesis $H_{0} : β_{1} = 0$ versus $H_{1} : β_{1} \neq 0$ . The risk difference between experimental treatments is represented by $β_{1}$ .

Analysis in parallel design for continuous and binary outcomes

The generalized estimating equations¹³ are used to compare each of two experimental treatments with the control treatment and between experimental treatments. The analysis methods are presented without distinguishing between the continuous and binary outcomes. The outcome data for each participant $i$ are redefined as $Y_{i} = {Y_{i 1}, Y_{i 2}}'$ with corresponding means $μ_{i} = {μ_{i 1}, μ_{i 2}}'$ . The mean $μ_{i}$ is related to the linear predictor through the link function $g (μ_{i}) = X_{i}^{'} β$ , where $X_{i}$ is the $3 \times 2$ design matrix representing which arm is assigned

X_{i} = {\begin{matrix} (\begin{matrix} 1 & 0 & 0 \\ 1 & 0 & 0 \end{matrix})' if g_{i} = 1 \\ (\begin{matrix} 1 & 1 & 0 \\ 1 & 1 & 0 \end{matrix})' if g_{i} = 2 \\ (\begin{matrix} 1 & 0 & 1 \\ 1 & 0 & 1 \end{matrix})' if g_{i} = 3 \end{matrix}

with the corresponding parameter vector $β$ consisting of $(β_{0}, β_{1}, β_{2})$ . The parameters can be obtained by solving the following equation

U = \sum_{i} {D'}_{i} V_{i}^{- 1} (y_{i} - μ_{i} (β)) = 0

where $D_{i} = \partial μ_{i} / \partial β$ , $V_{i}$ is the working covariance matrix of $Y_{i}$ . In this study, the Wald test with a sandwich variance is used to test the following hypothesis: $H_{0} : β_{1} \geq 0$ for E₁ versus C, $H_{0} : β_{2} \geq 0$ for E₂ versus C, and $H_{0} : β_{1} - β_{2} = 0$ for E₁ versus E₂, where the lower values of $β_{1}$ and $β_{2}$ indicate increased effectiveness. The mean differences are estimated as $β_{1}$ and $β_{2}$ , where the dominant hand effect is offset, since participants are equally randomized in each arm.

Multiplicity adjustment

In both designs, a maximum of three comparisons are performed: two comparisons of each experimental treatment with the control treatment, and a comparison between experimental treatments. To maintain the overall significance level of $α$ , the Bonferroni method is used for the comparisons of each experimental treatment with the control treatment, that is, testing with a significance level of $α / 2$ (one-sided) in each comparison. Since the comparison between experimental treatments is conducted only when both experimental treatments are superior to the control treatment, the significance level of $α$ (two-sided) is used based on a closed test procedure.

Results

Simulation settings

We assumed a clinical trial for comparing each of two experimental treatments (E₁ and E₂) with the control treatment (C), with a one-sided significance level of 0.05 / 2 (=0.025). To select the more effective experimental treatment when both experimental treatments were superior to C, we also compared between the experimental treatments, with a two-sided significance level of 0.05. E₁ was assumed to be more effective than E₂. The primary outcome was a continuous outcome or a binary outcome. Continuous outcome data were generated based on the multivariate normal distribution, where the lower value indicates increased effectiveness. Binary outcome data were generated based on the conditional linear family,¹⁴ where a lower probability indicates increased effectiveness (e.g. an adverse event). SAS program for implementing the conditional linear family was provided at http://www.bios.unc.edu/distrib/gee/clf/. The generalized estimating equations with the identity link function and the exchangeable working correlation structure were used in the parallel design.

Under total sample size and the means (probabilities) of E₁ and C were fixed, two scenarios in which only the mean (probability) of E₂ was different were considered in each outcome: Scenarios 1 and 2 for continuous outcome and Scenarios 3 and 4 for binary outcome. For continuous outcome, the effects of E₁, E₂, and C, ${τ_{E_{1}}, τ_{E_{2}}, τ_{C}}$ , were expected to be ${- 0.50, - 0.25, 0}$ in Scenario 1 and ${- 0.50, - 0.30, 0}$ in Scenario 2, assuming the dominant hand effect $λ$ of 0.10 (negative effect) and the standard derivation $σ$ of 1.0. Total sample size was 180 in Scenarios 1 and 2. For binary outcome, the effects of E₁, E₂, and C were expected to be $\{ 35{\rm%}, \,\ 45{\rm%}, \ \,50{\rm%}\}$ in Scenario 3 and $\{ 35{\rm%}, \ \,40{\rm%}, \,\ 50{\rm%}\}$ in Scenario 4, assuming a dominant hand effect $λ$ of 10% (negative effect). Total sample size was 480 in Scenarios 3 and 4. The correlation between right and left hands was changed from 0 to 0.9 by 0.1 for continuous outcome and 0 to 0.5 by 0.1 for binary outcome. The range of correlation of binary outcomes is limited by the marginal probabilities. In addition, we evaluated the type I error rates under the effects of E₁ and E₂ were the same as that of C, that is, $τ_{E_{1}}$ , $τ_{E_{2}}$ , and $τ_{C}$ were equally 0 for continuous outcome (Scenario 5) and 50% for binary outcome (Scenario 6). In Scenarios 5 and 6, other settings were not changed from Scenarios 1 and 3, respectively.

To compare the operating characteristics of the self-controlled design with the parallel design and to compare between the analysis methods in the self-controlled design, we calculated the four probabilities (power): (a) each experimental treatment was superior to C; (b) both experimental treatments were superior to C; (c) both experimental treatments were superior to C, and E₁ was superior to E₂; and (d) E₁ was finally selected as superior. Probability (d) can be calculated as the sum of the two probabilities: (1) only E₁ is superior to C and (2) E₁ is superior to E₂. In addition to probabilities (a) and (b), we calculated probability (e). This is the probability that at least one of E₁ and E₂ is superior to C when evaluating the type I error rates. The denominator of calculation for these probabilities was the number of simulations, and all simulations were repeated 100,000 times for each scenario. All simulations were performed using SAS version 9.4.

Power in various right–left correlations of continuous and binary outcomes

Figure 2 shows the power in each design for continuous outcome (Scenarios 1 and 2) and binary outcome (Scenarios 3 and 4). Probability (a) increased in the self-controlled design and decreased in the parallel design as the correlation increased (first column). Probability (b) was higher in the self-controlled design than in the parallel design under conditions of correlation of 0.3 or higher in scenarios of this study (the second column). Probability (c) increased in the self-controlled design and decreased in the parallel design as the correlation increased (third column). Probability (d) decreased once and then increased as the correlation increased in the self-controlled design with continuous outcome (fourth column). In contrast, in the parallel design, probability (d) did not change substantially depending on the correlation, since the probability that only E₁ was superior to C (i.e. E₂ was not superior to C) increased and probability (c) decreased as the correlation increased in comparison with each of the experimental treatments with C.

For comparisons between experimental treatments, the “regression” method had highest power under modest to high correlation condition among the analysis methods in the self-controlled design. There was no large difference between the methods using only hands with experimental treatments (the “t-test” or “chi-square test”) and the “regression” method under low correlation. The “delta” method had lowest power under low correlation; however, it had higher power than the “t-test” or “chi-square test” under high correlation.

Type I error rate in various right–left correlations of continuous and binary outcomes

Figure 3 shows the type I error rates for continuous outcome (Scenario 5) and binary outcome (Scenario 6). Probability (a) was approximately 2.7% for the parallel design and approximately 2.5% for the proposed design. Probability (b) was higher in the parallel design than in the proposed design since same data on C were used in the parallel design. Although probability (a) was slightly higher than 2.5% in the parallel design, probability (e) was lower than 5.0% in both scenarios. In both designs, it is important to maintain probability (e) at the overall significance level (5%), and no inflation of the type I error rate was observed.

Discussion

We proposed a self-controlled design to compare each experimental treatment to the control treatment and the experimental treatments to each other. In addition, we compared the power difference between the proposed design and the parallel design using a simulation study. In these scenarios, probabilities (a), (b), and (c) were higher in the proposed design than in the parallel design under conditions of correlation of 0.3 or higher. Of all the analysis methods in the proposed design, the method based on a regression model including outcome of hands with the control treatment as a covariate had highest power under modest to high correlation condition. Thus, the proposed design with the “regression” method is a good option when assuming correlation of 0.3 or higher. No inflation of the type I error rate was observed.

The design can be selected based on probability (b), when conducting the clinical trial to have multiple treatment options with different risk-benefit balance, for example, experimental treatments, such as cryotherapy and compression therapy. Alternately, the design can be selected based on probabilities (c) and (d), when one experimental treatment must be selected, such as comparisons between strengths of compression therapy and between doses of ointment. Even in this situation, probability (b) is also important as comparison between experimental treatments is performed after both experimental treatments are demonstrated to be superior to the control treatment. Note that probability (d) also increases as the probability that either experimental treatment is not superior to the control treatment increases.

Our simulation results show that the correlation between right and left hands is an important factor for selecting the design. For example, the self-controlled non-randomized clinical trial conducted by Hanai et al.⁴ reported that the probability of incidence of CIPN based on the monofilament test, which was the binary outcome, was clinically and statistically significantly lower for the intervention side than the control side in 36 patients (hand: 27.8% vs 80.6%; foot: 25.0% vs 63.9%). Since 9 of 36 patients (25%) experienced CIPN on both sides of the hands and feet, the corresponding correlations were 0.15 for the hands and 0.43 for feet. In the secondary outcomes, the correlations for hands and feet were 0.23 and 0.35 in the warm-sense disturbance, and 0.42 and 0.30 in cold-sense disturbance, respectively. When assuming correlation of 0.42 in Scenario 3 (binary outcome), probability (a) of E₁ versus C and E₂ versus C and probability (b) were 89.8%, 19.3%, and 19.0% for parallel design and 99.1%, 29.3%, and 29.1% for the proposed design, respectively. Probabilities (c) and (d) were 5.4% and 76.1% for parallel design and 11.7% and 81.8% for the proposed design with the regression method, respectively. These results suggest that correlations between right and left hands and feet exist to some extent, and that the proposed design and analysis method can be an option.

This study assumed a clinical trial in which the primary outcome was an objective outcome for CIPN, such as outcomes based on Common Terminology Criteria for Adverse Events or the monofilament test.¹⁵ The proposed design is applicable to the clinical trials for other adverse events than CIPN that can occur in multiple sites in the same participant, such as paronychia^16,17 and hand-foot syndrome.^18,19 However, since the outcome for one hand may be affected by the treatment effect on the other hand, especially for subjective outcomes, it is difficult to appropriately obtain outcome data. In the proposed design, this affect may be smaller than in the self-controlled designs where participants can receive the different experimental treatments, since all participants receive the standard treatment in either hand. Despite this, we should pay close attention to that possibility when using subjective primary outcomes. In addition, it is necessary to guarantee that there is no contamination of treatments in other applications.

Several extensions can be considered. First, some inter-block information could be incorporated when comparing each of the experimental treatments with the control treatment in the self-controlled design. For example, right hands with E₁ in Arm 1 could be contrasted with right hands with C in Arm 4 to provide additional information regarding E₁ versus C. Second, when either experimental treatment is less toxic or more convenient in practice, the non-inferiority hypothesis may be appropriate for comparison between experimental treatments (e.g. the comparison between cryotherapy and compression therapy, and weak compression and strong compression therapy). The proposed design can be easily extended to consider these situations. Third, the proposed design can be generalized to trials with one control treatment and over two experimental treatments. In this case, study designs will become more complex as trial objectives, success criteria, decision rules, or the risk-benefit balance of treatments are diversified between trials. Therefore, it is necessary to re-assess the powers according to trial objectives. Finally, other measures of interest include the ordinal outcome (e.g. severity grade of CIPN) and the time-to-event outcome (e.g. the time from enrollment to incidence of CIPN); however, the optimal design and analysis methods required for these measures have yet to be developed.

Conclusion

We proposed a self-controlled design and associated analysis methods for clinical trials to compare three treatments on symptoms, such as CIPN. This design will be a beneficial option in situations where it is undesirable for participants to receive different experimental treatments in both hands for practical reasons.

Footnotes

Acknowledgements

The authors would like to thank Dr Akiko Hanai for useful discussions. We also thank Ms Ikumi Iida, Ms Maki Moriya, and Ms Chikako Kondo. The clinical trial they planned motivated our work.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Ryunosuke Machida

References

Cavaletti

Marmiroli

Chemotherapy-induced peripheral neurotoxicity. Nat Rev Neurol 2010; 6: 657–666.

Staff

Grisold

, et al. Chemotherapy-induced peripheral neuropathy: a current review. Ann Neurol 2017; 81: 772–781.

Mols

Beijers

Vreugdenhil

, et al. Chemotherapy-induced peripheral neuropathy and its association with quality of life: a systematic review. Support Care Cancer 2014; 22(8): 2261–2269.

Hanai

Ishiguro

Sozu

, et al. Effects of cryotherapy on objective and subjective symptoms of paclitaxel-induced neuropathy: prospective self-controlled trial. J Natl Cancer Inst 2018; 110: 141–148.

Griffiths

Kwon

Beaumont

, et al. Cold therapy to prevent paclitaxel-induced peripheral neuropathy. Support Care Cancer 2018; 26(10): 3461–3469.

Tsuyuki

Senda

Kanng

, et al. Evaluation of the effect of compression therapy using surgical gloves on nanoparticle albumin-bound paclitaxel-induced peripheral neuropathy: a phase II multicenter study by the Kamigata Breast Cancer Study Group. Breast Cancer Res Treat 2016; 160(1): 61–67.

Kanbayashi

Sakaguchi

Ishikawa

, et al. Comparison of the efficacy of cryotherapy and compression therapy for preventing nanoparticle albumin-bound paclitaxel-induced peripheral neuropathy: a prospective self-controlled trial. Breast 2020; 49: 219–224.

Ohno

Mine

Yoshioka

, et al. Management of peripheral neuropathy induced by nab-paclitaxel treatment for breast cancer. Anticancer Res 2014; 34(8): 4213–4216.

Ruddy

Le-Rademacher

Lacouture

, et al. Randomized controlled trial of cryotherapy to prevent paclitaxel-induced peripheral neuropathy (RU221511I); an ACCRU trial. Breast 2019; 48: 89–97.

10.

Beijers

AJM

Bonhof

Mols

, et al. Multicenter randomized controlled trial to evaluate the efficacy and tolerability of frozen gloves for the prevention of chemotherapy-induced peripheral neuropathy. Ann Oncol 2020; 31(1): 131–136.

11.

ClinicalTrials.gov. The CONTRoL trial: cryotherapy vs. Compression neuropathy trial (CONTRoL), https://clinicaltrials.gov/ct2/show/study/NCT03873272 (accessed 17 November 2021).

12.

Senn

Cross-over trials in clinical research. 2nd ed. New York: John Wiley & Sons, 2002.

13.

Liang

Zeger

SL.

Longitudinal data analysis using generalized linear models. Biometrika 1986; 73: 13–22.

14.

Qaqish

BF.

A family of multivariate binary distributions for simulating correlated binary variables with specified marginal means and correlations. Biometrika 2003; 90: 455–463.

15.

Jayaprakash

Bhansali

, et al. Validation of bedside methods in evaluation of diabetic peripheral neuropathy. Indian J Med Res 2011; 133(6): 645–649.

16.

Kim

Seo

, et al. A prospective randomized controlled trial of hydrating nail solution for prevention or treatment of onycholysis in breast cancer patients who received neoadjuvant/adjuvant docetaxel chemotherapy. Breast Cancer Res Treat 2017; 164(3): 617–625.

17.

Morrison

Marshall-McKenna

McFadyen

, et al. A randomised controlled trial of interventions for taxane-induced nail toxicity in women with early breast cancer. Sci Rep 2022; 12: 11575.

18.

Xie

Wang

, et al. Effect of a novel soaking solution used in patients with hand-foot syndrome as a result of capecitabine treatment: a randomized and self-controlled trial. Clin Breast Cancer 2022; 22(5): e685–e690.

19.

Zenda

Ryu

Takashima

, et al. Hydrocolloid dressing as a prophylactic use for hand-foot skin reaction induced by multitargeted kinase inhibitors: protocol of a phase 3 randomised self-controlled study. BMJ Open 2020; 10: e038276.