Predictors of duloxetine response in patients with oxaliplatin-induced painful chemotherapy-induced peripheral neuropathy (CIPN): a secondary analysis of randomised controlled trial – CALGB/alliance 170601

Abstract

Duloxetine is an effective treatment for oxaliplatin-induced painful chemotherapy-induced peripheral neuropathy (CIPN). However, predictors of duloxetine response have not been adequately explored. The objective of this secondary and exploratory analysis was to identify predictors of duloxetine response in patients with painful oxaliplatin-induced CIPN. Patients (N = 106) with oxaliplatin-induced painful CIPN were randomised to receive duloxetine or placebo. Eligible patients had chronic CIPN pain and an average neuropathic pain score ≥4/10. Duloxetine/placebo dose was 30 mg/day for 7 days, then 60 mg/day for 4 weeks. The Brief Pain Inventory-Short Form and the EORTC QLQ-C30 were used to assess pain and quality of life, respectively. Univariate and multiple logistic regression analyses were performed to identify demographic, physiologic and psychological predictors of duloxetine response. Higher baseline emotional functioning predicted duloxetine response (≥30% reduction in pain; OR 4.036; 95% CI 0.999–16.308; p = 0.050). Based on the results from a multiple logistic regression using patient data from both the duloxetine and placebo treatment arms, duloxetine-treated patients with high emotional functioning are more likely to experience pain reduction (p = 0.026). In patients with painful, oxaliplatin-induced CIPN, emotional functioning may also predict duloxetine response. ClinicalTrials.gov, Identifier NCT00489411

INTRODUCTION

Chemotherapy is a mainstay of cancer treatment that is received by millions of cancer survivors. Chemotherapy-induced peripheral neuropathy (CIPN) is a common side effect of neurotoxic chemotherapeutic agents such as oxaliplatin, and many others (Argyriou et al. 2010; Beijers et al. 2014). Numbness and tingling in the hands and feet are the most common symptoms and CIPN becomes chronically painful in approximately 20–42% of cases (Dworkin 2002; Cavenagh et al. 2006; Hausheer et al. 2006; Taylor 2006; Argyriou et al. 2008a, Windebank & Grisold 2008; Smith et al. 2010; Sonneveld & Jongen 2010; Kautio et al. 2011; Geber et al. 2013). Consequently, CIPN can evolve into a chronic pain syndrome that impairs function and quality of life (QOL) (Calhoun et al. 2003; Cella et al. 2003; Almadrones et al. 2004; Bakitas 2007; Bruner et al. 2007; Kiser et al. 2010; Tofthagen 2010; Plotti et al. 2011; Mols et al. 2014).

Unfortunately, evidence-based, effective interventions for painful CIPN are rare. Duloxetine, a serotonin–norepinephrine reuptake inhibitor, is the only recommended treatment for painful oxaliplatin-induced CIPN (Hershman et al. 2014). Duloxetine works by increasing the amount of key pain-inhibiting neurotransmitters, serotonin and norepinephrine, within the central nervous system (Bymaster et al. 2003). Although the original randomised, placebo-controlled trial conducted by the Cancer and Leukemia Group B (CALGB) provides strong evidence of duloxetine efficacy, just 33% of the duloxetine-treated patients experienced a moderate pain reduction (≥30%), and even fewer (21%) experienced a substantial decrease in pain (≥50%) (Smith et al. 2013). Moreover, duloxetine was ineffective in 41% of the study participants (Smith et al. 2013). Since duloxetine was not completely effective, nor did it work for everyone, identifying predictors of duloxetine response is a priority area for future research. More specifically, if we know why duloxetine works, for whom and in what circumstances, a personalised approach can be used to prescribe duloxetine to those most likely to benefit.

A clue regarding one possible predictor of duloxetine response can be found in the original report of the CALGB study (Smith et al. 2013). More specifically, the results of an exploratory responder analysis suggest that patients with oxaliplatin-induced painful CIPN are more likely to experience a benefit from duloxetine than patients with paclitaxel-induced CIPN (Smith et al. 2013). This finding suggests that duloxetine’s mechanism of action may be uniquely tied to very specific mechanisms of chemotherapy-induced neurotoxicity. Although the precise mechanism of chronic oxaliplatin-induced CIPN is still unknown, studies have shown that oxaliplatin accumulates in the dorsal root ganglion where it causes nerve cell apoptosis (Cavaletti et al. 2001; Renn et al. 2011). In addition, an oxaliplatin metabolite, oxalate, chelates calcium and impairs calcium-sensitive sodium-dependent ion channel function, leading to peripheral nerve hyper-excitability (Grolleau et al. 2001; Wilson et al. 2002; Krishnan et al. 2005; Benoit et al. 2006; Park et al. 2009; Beijers et al. 2014). In contrast, taxanes disrupt microtubules, causing CIPN by subsequent demyelination and impairment of axonal transport (Persohn et al. 2005; Argyriou et al. 2008b; Park et al. 2011). Other taxane-induced nerve injury mechanisms include macrophage activation in peripheral nerves and dorsal root ganglia, microglial activation and down-regulation of glutamate transporters in the spinal cord, and damaged mitochondria in A- and C-fibres (Cata et al. 2006; Flatters & Bennett 2006; Peters et al. 2007; Argyriou et al. 2008b, 2012; Jin et al. 2008). We found no other published studies exploring a differential response to duloxetine based on the causative chemotherapeutic agent; more research in this area is needed.

The chronic pain literature provides additional clues about other possible predictors of duloxetine response. Widespread body pain, emotional distress (e.g., anxiety, depression), fatigue, impaired cognition, and sleep disturbance are centrally mediated symptoms that co-occur in a variety of chronic pain conditions and predict pain severity (Clauw & Chrousos 1997; Fukuda et al. 1997, 1998; Clark et al. 2000; Bair et al. 2003; Giesecke et al. 2003; Gore et al. 2005, 2006; Heitkemper & Jarrett 2005; Postma et al. 2005; Castillo et al. 2006; Zelman et al. 2006; Bakitas 2007; Geisser et al. 2007, 2008a,b; Allen et al. 2008; Clemens et al. 2008; Fishbain et al. 2008; Roy-Byrne et al. 2008; Zhang & Jordan 2008; Warren et al. 2009; Desaulniers 2011; Tofthagen 2011; Geber et al. 2013). Individuals with painful CIPN also experience similar centrally mediated symptoms (Nail 2011; Tofthagen et al. 2013). Since co-occurring symptoms can make chronic pain worse, patients with painful CIPN who also experience co-occurring symptoms may be less responsive to analgesic interventions like duloxetine. Accordingly, this paper reports the results of secondary and exploratory analyses (using data from the CALGB trial) that were performed to determine whether the European Organisation for the Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-C30) subscale scores for emotional and cognitive functioning, fatigue and insomnia would predict duloxetine response in the CALGB patient cohort that experienced the best effect – those with oxaliplatin-induced CIPN. The primary hypothesis is that baseline severity of co-occurring symptoms common to chronic pain disorders – emotional distress, impaired cognition, fatigue, and insomnia – will predict duloxetine efficacy in patients with chronic, painful oxaliplatin-induced CIPN.

METHODS

Sample and setting

Between April 2008 and March 2011, CALGB 170601 enrolled 231 participants ≥25 years of age from 105 academic and community sites throughout the United States. All patients provided signed Institutional Review Board-approved informed consent. Patient eligibility has been previously described (Smith et al. 2013). Briefly, eligible participants reported sensory neuropathy >grade 1 using the National Institutes of Health Common Terminology Criteria for Adverse Events (CTCAE) v. 3.0 grading scale, an average CIPN-related neuropathic pain score ≥4 on a 0–10 scale using the Brief Pain Inventory-Short Form (BPI-SF) item 5, and persistent pain at least 3 months after completion of paclitaxel or oxaliplatin treatment. Patients could not have received other types of neurotoxic chemotherapy drugs (e.g. vinca alkaloids, bortezomib, thalidomide), and those with neuropathy due to other comorbid conditions were not eligible. Concurrent use of other antidepressants, anticonvulsants, high-dose vitamin supplements or drugs known to influence serotonin levels (e.g. tramadol) was not allowed.

Procedure

The research methods used in CALGB 170601 have been previously described (Smith et al. 2013). Eligible participants were randomised to receive either duloxetine 60 mg or placebo. Stratified, random assignment to treatment groups was determined by the CALGB/Alliance Statistics and Data Center based on the neurotoxic drug received (taxane versus platinum) and CIPN risk [high risk (those with diabetes mellitus) versus low risk]. All patients and personnel were blinded to treatment assignment.

Duloxetine/placebo was started at 30 mg daily for the first week. Beginning on day 8, 4 weeks of full dose (60 mg) duloxetine/placebo treatment began. Starting at week 6, participants underwent a 2-week washout period and then crossed over to the other treatment arm.

A clinical research associate telephoned each patient weekly to ask them to rate CIPN pain severity using the BPI-SF. The BPI-SF is a well-validated 15-item instrument that includes items quantifying average, worst, least and immediate pain severity using a 0–10 numeric rating scale (Cleeland et al. 1994; Cleeland 2009). The BPI-SF has been tested in culturally and linguistically diverse populations with various types of painful disorders, providing evidence of internal consistency and test–retest reliability, and construct, structural, concurrent and discriminant validity (Cleeland 2009). BPI-SF item #5, which quantifies average pain severity using a 0–10 scale, was used to assess duloxetine response.

Participants also completed the EORTC QLQ-C30 on day 1 of weeks 1, 6, 8 and 13. The EORTC QLQ-C30 is comprised of several core components applicable to all cancer patients (Aaronson et al. 1993). Its 30 items are grouped into subscales assessing global health status and QOL (2); physical (5), role (2), emotional (4), cognitive (2) and social (2) functioning; fatigue (3); nausea and vomiting (2); and pain (2). The questionnaire assesses six additional items: dyspnoea, insomnia, appetite loss, constipation, diarrhoea and financial difficulties. Respondents rate their global health status and QOL from 1 (very poor) to 7 (excellent). The degree to which respondents are experiencing other problems is rated on a 4-point Likert scale, ranging from ‘Not at All’ to ‘Very Much’. For the current analysis, we focused on specific subscales known to be associated with chronic pain: emotional and cognitive functioning, fatigue, pain and insomnia. The emotional subscale items quantify whether the respondent worries, or feels tense, irritable or depressed. Items in the cognitive function subscale assess concentration and memory. The fatigue subscale asks about weakness, the need for rest and feeling tired. Two pain subscale items ask if respondents have had pain and about pain’s influence on performance of daily activities. The insomnia question quantifies whether the respondent has had trouble sleeping. Cronbach’s alpha coefficients for the global health status, emotional and cognitive functioning, fatigue and pain subscales range from 0.73 to 0.89 (Aaronson et al. 1993). Test–retest reliability correlations range from 0.70 to 0.90 for all subscales (Hjermstad et al. 1995). Satisfactory construct validity has been previously demonstrated (Aaronson et al. 1993).

Analyses

Given the exploratory nature of the analyses, an a priori power analysis was not conducted. However, to minimise the risk of false discovery, all analyses were conducted using data obtained only from oxaliplatin-treated patients in the initial treatment period (weeks 1–5), because the subgroup analysis for the paclitaxel cohort did not yield significant differences in pain reduction. A post hoc power analysis revealed that there would be 60% power (two-sided inflated alpha of 0.10) to detect predictors of duloxetine response if 25% of patients responded in the below-median group versus 53% of those in the above-median group. Statistical analyses were performed by the Alliance Statistics and Data Center, using SAS 9.3 (Cary, NC) on a database locked in April 2012.

We focused on exploring whether EORTC QLQ-30 subscale scores quantifying known predictors of chronic pain severity (emotional function, impaired cognition, fatigue and insomnia) might predict duloxetine response, defined as a ≥ 30% improvement (Dworkin et al. 2009) in pain severity during the initial treatment period based on the pain score obtained from BPI-SF item # 5. Differences in demographic variables between duloxetine and placebo arms were compared using t-tests for continuous variables and chi-squared tests or Fisher’s exact tests for categorical variables (Fleiss 1981; Altman 1991). Descriptive statistics [medians, frequencies, 95% confidence intervals (CI)] were calculated to describe the incidence of duloxetine response by chemotherapy agent. Young/old age and low/high EORTC QLQ-30 subscale scores were defined based on medians of the distributions. Changes in EORTC subscale scores were summarised using means and standard deviations (SDs), and compared by general linear modelling adjusted for baseline score and neuropathy risk (presence/absence of diabetes). High/low subscale scores were defined as being either above or below the medians.

Univariate and multiple logistic regression analyses were performed to identify predictors for duloxetine response. To test for an interaction effect of emotional functioning and treatment arm, we constructed a multiple logistic regression model (controlling for baseline CTCAE neuropathy grade) using patient data from the duloxetine and placebo treatment arms. Odds ratios for the selected baseline EORTC QLQ-30 subscales scores were calculated using the scores between 0 and 100 and divided by 33⅓. This rescales the scores back to the original range for ease of interpretation. The odds ratio for global health status was calculated based on the actual scores. The proportion of missing data was ≤4% for the primary outcome; therefore, we took a complete case analysis approach. To test for a treatment group effect on the change in EORTC QLQ-C30 subscale scores during the initial treatment period, we used analysis of covariance, each stratified by neurotoxic agent and comorbid risk (presence/absence of diabetes), and including the baseline measure of the corresponding subscale scores.

RESULTS

Patient characteristics

Patients’ demographic characteristics (Table 1) were derived using data obtained only from the patients with oxaliplatin-associated painful CIPN (n = 106; duloxetine n = 49; placebo n = 57). The oxaliplatin sample was primarily men (64.2%) and Caucasian (85.9%). Most patients had good performance status (85.9%) and had undergone chemotherapy treatment for a stage I–III (79.2%) gastrointestinal malignancy (98.1%). The mean age was 59.7 years, and the mean baseline pain score was 5.8 out of 10. With the exception of the baseline neuropathy grade (duloxetine group mean grade = 2.35, SD 0.52); placebo group mean grade = 2.26, SD 0.41), no statistically significant differences in demographic characteristics between the duloxetine- and placebo-treated groups were found.

Table 1.

Demographics characteristics

Characteristics	No. of participants (%)			p
	Duloxetine (n = 49)	Placebo (n = 57)	Total (n = 106)
Age (years)
30–39	0 (0)	1 (1.8)	1 (0.9)	0.882*
40–49	6 (12.2)	11 (19.3)	17 (16.0)
50–59	22 (44.9)	20 (35.1)	42 (39.6)
60–69	16 (32.7)	15 (26.3)	31 (29.3)
≥70	5 (10.2)	10 (17.5)	15 (14.2)
Mean (SD)	59.86 (9.52)	59.56 (10.97)	59.70 (10.28)
Sex
Men	28 (57.1)	40 (70.2)	68 (64.2)	0.163
Women	21 (42.9)	17 (29.8)	38 (35.8)
Race
White	43 (87.8)	48 (84.2)	91 (85.9)	0.547
Black	5 (10.2)	4 (7.0)	9 (8.5)
Other	1 (2.0)	3 (5.3)	4 (3.8)
Not reported	0 (0)	2 (3.5)	2 (1.9)
High risk of chemotherapy-induced peripheral neuropathy
No	22 (44.9)	26 (45.6)	48 (45.3)	0.941
Yes	27 (55.1)	31 (54.4)	58 (54.7)
Primary disease
GI	48 (98.0)	56 (98.2)	104 (98.1)	1.000
Other	1 (2.0)	1 (1.8)	2 (1.9)
Disease stage
Early I–II	10 (20.4)	15 (26.3)	25 (23.6)	0.604
III	27 (55.1)	32 (56.1)	59 (55.7)
Metastatic	12 (24.5)	10 (17.5)	22 (20.8)
Performance status
0	26 (53.1)	29 (50.9)	55 (51.9)	1.000
1	22 (44.9)	26 (45.6)	48 (45.3)
2+	1 (2.0)	2 (3.5)	3 (2.8)
Sensory neuropathy grade†
Mean (SD)	2.35 (0.52)	2.26 (0.41)	2.25 (0.47)	0.040*
Baseline pain score
<4	1 (2.0)	0 (0)	1 (0.9)	0.189*
4–5	19 (38.8)	32 (56.1)	51 (48.1)
6–7	19 (38.8)	18 (31.6)	37 (34.9)
8–10	10 (20.4)	7 (12.3)	17 (16.0)
Mean (SD)	6.00 (1.70)	5.58 (1.58)	5.77 (1.64)

^*Tested as a continuous variable.

^†Based on National Cancer Institute Common Terminology for Adverse Events (NCI-CTCAE) Sensory Neuropathy Grade.

Incidence of duloxetine response

Table 2 illustrates the incidence of duloxetine responders in the oxaliplatin-treated cohort. To avoid the risk of false discovery due to multiple testing, statistical tests were not performed using these data. Nevertheless, the findings support the hypothesis that patients with more severe symptoms are less likely to benefit from duloxetine. The biggest differences in duloxetine response were seen when examining the percentage of responders experiencing high versus low emotional functioning, fatigue and pain. Approximately 28% more patients with high versus low emotional functioning responded to duloxetine. Furthermore, when compared to patients with higher levels of fatigue and pain, more patients with low scores, 25.5% and 23.8% respectively, responded to duloxetine.

Table 2.

Incidence of responders^*

Variable	Median	Duloxetine (n = 49)		Placebo (n = 57)
		# Responders/Total n	% Responders (95% CI)	# Responders/Total n	% Responders (95% CI)
Age
Younger	59.2	8/23	34.8 (16.4–57.3)	4/30	13.3 (3.8–30.7)
Older		13/26	50.0 (29.9–70.1)	4/27	14.8 (4.2–33.7)
Gender
Male		11/28	39.3 (21.5–59.4)	6/40	15.0 (5.7–29.8)
Female		10/21	47.6 (25.7–70.2)	2/17	11.8 (1.5–36.4)
Global health status
Low	66.6	9/26	34.6 (17.2–55.7)	4/23	17.4 (5.0–38.8)
High		12/22	54.6 (32.2–75.6)	4/34	11.8 (3.3–27.5)
Emotional functioning
Low	75.0	4/16	25.0 (7.2–52.4)	5/25	20.0 (6.8–40.7)
High		17/32	53.1 (34.7–70.9)	3/32	9.4 (2.0–25.0)
Cognitive functioning
Low	83.3	9/21	42.9 (21.8–66.0)	4/23	17.4 (5.0–38.8)
High		12/27	44.4 (25.5–64.7)	4/34	11.8 (3.3–27.5)
Fatigue
Low	33.3	10/17	58.8 (32.9–81.6)	2/21	9.5 (1.2–30.4)
High		10/30	33.3 (17.3–52.8)	6/36	16.7 (6.4–32.8)
Pain
Low	50.0	12/21	57.1 (34.0–78.2)	2/19	10.5 (1.3–33.1)
High		9/27	33.3 (16.5–54.0)	6/38	15.8 (6.0–31.3)
Insomnia
Low	33.3	4/11	36.4 (10.9–69.2)	2/14	14.3 (1.8–42.8)
High		17/37	46.0 (29.5–63.1)	6/43	14.0 (5.3–27.9)

^*Responders = ≥30% improvement in pain score during initial treatment period based on the pain score from BPI-SF item # 5. EORTC QLQ-30 subscale scores are baseline measurements obtained at week 1 of the initial treatment period. Young/old age and low/high scores are defined based on medians of the distributions.

EORTC QLQ-C30 subscale scores

Table 3 presents the change from baseline to 6-week EORTC QLQ-C30 subscale scores in duloxetine- and placebo-treated patients. After adjusting for the baseline pain score and CIPN risk (presence/absence of diabetes), duloxetine-treated patients reported a greater improvement in global health (p = 0.005), cognitive function (p = 0.021) and pain (p = 0.020) than did placebo-treated patients.

Table 3.

Change from baseline to 6-week EORTC QLQ-C30 subscale scores

Variable	Change in mean score following duloxetine treatment			Change in mean score following placebo treatment			p
	Mean	SD	N	Mean	SD	N
Global health status	8.1	23.1	44	−3.8	18.0	50	0.005*
Emotional functioning	6.8	16.2	44	3.7	17.3	50	0.074
Cognitive functioning	7.6	18.1	44	0.67	16.8	50	0.021*
Fatigue	−3.6	23.7	43	−3.3	16.5	50	0.725
Pain	−9.8	24.2	44	−4.1	16.5	49	0.020*
Insomnia	−3.0	21.3	44	0.0	26.1	50	0.547

Change scores were calculated by subtracting baseline from week 6 subscale scores obtained only in the sample of patients who provided a week 6 score.

^*Greater improvement in duloxetine-treated patients than placebo-treated patients.

Predictors of duloxetine response

The results of a univariate logistic regression, presented in Table 4, suggest that patients with better baseline emotional functioning scores were four times more likely to respond to duloxetine treatment (OR 4.036; 95% CI 0.999–16.308; p = 0.050). Baseline cognition, fatigue and insomnia scores did not predict duloxetine response. Based on the results from a multiple logistic regression using patient data from both the duloxetine and placebo treatment arms, the interaction p-value of emotional functioning and treatment arm (duloxetine versus placebo) was 0.026, suggesting that duloxetine-treated patients with oxaliplatin-induced CIPN pain and high emotional functioning are more likely to obtain a ≥ 30% reduction in pain.

Table 4.

Predictors of duloxetine response

Variable	Baseline EORTC QLQ-30 score		Univariate logistic regression
	Mean	SD	Odds ratio*	95% CI	p
Duloxetine arm
Global health status	62.0	22.1	1.015	0.99–1.04	0.310
Emotional functioning	78.3	15.4	4.036	0.99–16.31	0.050ˆ
Cognitive functioning	80.6	17.0	1.286	0.41–4.01	0.665
Fatigue	35.7	23.6	0.526	0.22–1.29	0.159
Pain	45.8	28.2	0.749	0.38–1.50	0.412
Insomnia	38.2	29.2	1.398	0.71–2.74	0.328
Placebo arm
Global health status	62.6	20.1	0.980	0.94–1.02	0.266
Emotional	72.1	23.7	0.577	0.21–1.55	0.276
Cognitive	77.5	19.5	0.779	0.22–2.73	0.696
Fatigue	33.9	22.9	1.879	0.67–5.26	0.230
Pain	52.3	23.9	1.308	0.46–3.73	0.615
Insomnia	39.2	31.6	0.935	0.42–2.09	0.870

Duloxetine response = ≥30% improvement in pain score during initial treatment period based on the pain score from BPI-SF item #5.

^*Odds ratios for EORTC QLQ-30 function and symptom scores are calculated based on the scores divided by 33⅓ odds ratio for global health status is calculated based on the scores.

^ˆBased on the results from a multiple logistic regression using patient data from both treatment arms (duloxetine and placebo) and adjusting for baseline neuropathy severity, the interaction p-value of emotional functioning and treatment arm was 0.026.

DISCUSSION

Based on the results of a recently published systematic review of randomised controlled trials testing CIPN interventions, duloxetine is the only drug recommended for the treatment of chronic CIPN pain (Hershman et al. 2014). As expected, duloxetine is not universally effective (Smith et al. 2013), and the reasons for its selective efficacy are unknown. The results of these secondary and exploratory analyses suggest that patients with better baseline emotional health (feeling less worried, tense, irritated, depressed) are four times more likely to respond to duloxetine, suggesting that those whose pain is part of a larger symptom cluster may benefit the least. These findings are consistent with the results of many studies showing that anxiety predicts pain perception (Theunissen et al. 2012; Schreiber et al. 2013; Bruce et al. 2014; Miaskowski et al. 2014). One such study reported that patients who were more anxious were more likely to have painful versus non-painful CIPN (Geber et al. 2013). Their findings reinforce what is known about the relationship between anxiety and pain and suggest that the patients in our study with high emotional functioning scores responded better to duloxetine because they were less worried and tense – emotions/feelings which are similar to anxiety. The implication for clinical practice is that perhaps emotionally distressed/anxious patients should be offered anxiety-relieving interventions alongside duloxetine.

We found no other published studies designed to explore whether a patient profile predicts the likelihood of clinically meaningful duloxetine-induced pain relief in patients with CIPN. However, two studies of duloxetine response in patients with other chronic pain conditions (migraine headache, fibromyalgia) who received similar doses suggest that patients with more severe baseline anxiety and depression were more likely to respond (Taylor et al. 2007; Marangell et al. 2011). In the current study, those with better emotional functioning were more likely to achieve at least a 30% reduction in pain. Our findings may vary from those of other published chronic pain studies due to differences in underlying pain and stress mechanisms in patients with non-malignant pain versus cancer treatment-related pain. In addition, these discordant results may stem from variations in measurement approaches. For this secondary data analysis, emotional functioning was quantified using a four-item EORTC QLQ-30 subscale that assessed the degree to which the patient felt worried, tense, irritated or depressed. Because the EORTC QLQ-30 subscale quantifies a different emotional phenotype than instruments designed to diagnose mood disorders (anxiety and depression), the current findings are not directly comparable with other chronic pain studies.

An alternative explanation for the difference in duloxetine response rates may be related to differences in patients’ underlying pain mechanisms. For example, some patients may have pain caused by multiple/mixed mechanisms, both peripheral nociceptive and central neuropathic. This idea is supported by Geber et al. (2013), 60% of whose study subjects with painful CIPN also reported pain with musculoskeletal (nociceptive) characteristics. Perhaps non-responding CALGB 170601 participants experienced more nociceptive pain due to musculoskeletal symptoms, multiple surgeries, or radiation therapy. Since nociceptive/peripheral pain is less responsive to centrally acting drugs like duloxetine, a higher incidence of mixed pain in non-responding patients might partially explain duloxetine’s selective efficacy.

Given these considerations, the cause of duloxetine’s selective effect may lie within the central nervous system. Although not well understood, mechanisms involved in the development of chronic neuropathic pain include abnormal neuron receptors and ion channel function, increased production and release of pain-facilitating neurotransmitters, and faulty central nervous system-mediated pain excitatory and inhibitory systems (Baron et al. 2010). A study conducted in patients with painful diabetic neuropathy provides preliminary evidence that inefficient central nervous system-mediated pain inhibition – mediated by serotonin and norepinephrine – predicts better duloxetine response (Yarnitsky et al. 2012). This finding suggests that duloxetine may be less effective for patients with normally functioning pain inhibitory systems.

This study has several limitations. First, although the results suggest that better emotional health was the only hypothesised variable that predicts duloxetine response, the sample size of responders may have been too small to detect statistically significant associations between baseline cognitive functioning, fatigue, and insomnia subscale scores and duloxetine response. In addition, we hypothesised that the EORTC QLQ-C30 subscale scores for factors known to be associated with chronic pain severity in other populations – emotional and cognitive functioning, fatigue and insomnia – would be most closely associated with duloxetine response. However, other well-known predictors of chronic pain severity were not assessed, such as previous trauma exposure (e.g., sexual/physical abuse, physical trauma, deployment to war) (Campbell & Lewandowski 1997; Golding 1999; Coker et al. 2000; Baccini et al. 2003; Meltzer-Brody et al. 2007; Humphreys et al. 2010; Barry et al. 2011; Raphael & Widom 2011) and the tendency to catastrophise about pain (believing that pain is profoundly awful) (Sullivan et al. 2001; Edwards et al. 2006, 2011; Campbell & Edwards 2009). EORTC QLQ-C30 subscale scores, used to quantify the predictor variables, may be less sensitive than other validated measures of psychological and physical symptoms. Last, although patients in the duloxetine group were taking fewer concomitant analgesics than the placebo-treated patients at baseline and at study completion, analgesic dosage in the oxaliplatin group could have increased over the initial treatment period, accounting for improvements in pain. These possible changes in concomitant analgesic dosage were not quantified (Smith et al. 2013).

The findings of this secondary data analysis suggest that patients with better baseline emotional functioning may be more likely to benefit from duloxetine. The next step is to conduct adequately powered follow-up studies to confirm these findings, and to identify other predictors of duloxetine response that might be amenable to complementary interventions. Patients with greater emotional distress and other sources of pain (muscle/joint pain associated with endocrine therapy for breast cancer) may require additional pharmacologic and/or non-pharmacologic interventions combined with duloxetine in order to achieve clinically significant improvements in pain. Thus, uncovering a patient phenotype associated with duloxetine efficacy could help clinicians make more informed decisions about who should receive the drug, and which patients may benefit from multi-modality treatments.