Evaluation of cold pain tolerance in patients with fibromyalgia and opioid use by survival analysis

2.1. Participants

All patient and control participants were female and participated in one of two separate studies during which data were collected. The two independent data collection sources were conducted by different research teams at different times and locations. CPT data from 35 patients and 17 controls, recruited from the regions surrounding Palo Alto, CA, were collected between 2015 and 2018 for Study 1 which was conducted at Stanford University. Results from analyses of other non-CPT-related data collected from Study 1 have been reported previously.^27–29 Data from 50 patients and 30 controls, recruited from the regions surrounding Durham, NC, were collected between 2019 and 2021 for Study 2 which was conducted at Duke University.

In Study 1, all fibromyalgia patients met the modified American College of Rheumatology (ACR) 2011 criteria for fibromyalgia [widespread pain index (WPI) ≥ 7 + symptom severity (SS) ≥ 5, or WPI 3–6 + SS ≥ 9; symptoms present at a similar level for at least 3 months; no disorder that would otherwise explain the pain].³⁰ In Study 2, fibromyalgia patients met the revised ACR 2016 criteria which includes a slight change in WPI criteria (range 4–6).³¹ All fibromyalgia patients were required to have pain in all four quadrants of the body, have an average pain score over the past month of at least 2 on a 0-10 verbal scale, have no MRI contraindications, and not be pregnant or nursing. Healthy individuals were required to have no history of chronic pain, not be pregnant or nursing, have no MRI contraindications, not be taking pain- or mood-altering medications at the time of the study, and have no depression or anxiety disorder. In total, 85 fibromyalgia patients and 47 healthy participants signed written informed consent indicating their willingness to participate in the study, understanding of all study procedures, and acknowledgement that they could withdraw from the study at any time. All procedures conducted were in accordance with the Declaration of Helsinki and were approved by the Stanford University Institutional Review Board for Study 1 and the Duke University Institutional Review Board for Study 2. A data use agreement was established allowing the data collected at Stanford to be analyzed by the research team at Duke University (K.T.M. was the PI for both Study 1 and Study 2). Pre-registered data analysis plans including hypotheses and exploratory analyses were published on the Open Science Framework (OSF) website at: DOI: 10.17605/OSF.IO/H2KGW (https://archive.org/details/osf-registrations-h2kgw-v1).

2.2. Medication usage

A total of 42 patients with fibromyalgia (18 from Study 1; 24 from Study 2) were enrolled based on their use of prescribed opioid medications as part of their ongoing pain treatment. These 42 participants were required to have been taking opioid medications for at least 3 months prior to and at the time of their participation in the study. The remaining 43 fibromyalgia participants had never taken opioids for a period greater than 30 days, had not taken any opioids within the 90 days before study participation, and were not taking opioid medications as part of their pain treatment at the time of their study participation. Opioid use and additional information about medication dosage and duration of use were recorded, and all participants were allowed to continue their medication use as normal during their participation in the study (see Table 2 and Table 5).

2.3. Study procedures

2.4. Statistical analyses

Questionnaire data were stored in and downloaded from secure REDCap databases. CPT data that were documented on case report forms were entered manually (by double data confirmatory entry) into Excel spreadsheets prior to analysis. All analyses were pre-registered in the OSF registry (DOI: 10.17605/OSF.IO/H2KGW) and were conducted in R version 3.6.0.

Because the CPT produces time-to-event data with right-censoring, the data were fit with Cox proportional-hazards models predicting participants’ hazard in the CPT from their group status (patient or control), demographic information, behavioral and psychological measures, and medication information. Prior to analysis, variables were assessed for greater than 5% missingness to determine whether data imputation would be required. Missing values were not imputed for variables with less than 5% missingness. All models used in this analysis contained an adjustment variable for starting temperature and a binary variable indicating Study (Study 1 or Study 2). Nested Cox models were compared using the likelihood ratio test (LRT), and non-nested Cox models were compared using the plrtest function in R (created by Thomas Hielscher) which implements a partial LRT suited for non-nested Cox models.⁴¹

Cold pain tolerance was first compared between patient (i.e., combined opioid and non-opioid patients) and healthy control groups using a Cox model predicting CPT hazard from fibromyalgia group status. To investigate the effect of demographic and psychological/clinical measures on the relationship between fibromyalgia and CPT hazard, a multivariate Cox model was built by selecting up to 4 additional covariates to add to the model in addition to group status (Figure 1A). The 4 additional covariates were first selected based on preliminary significance with CPT hazard (described below) from an array of a priori identified possible covariates, including age, race (White, Asian, Black, other), BDI score, BAS reward responsiveness, BAS drive, BAS fun-seeking, BIS score, PANAS positive affect, PANAS negative affect, POMS total mood disturbance, STAI state anxiety, and STAI trait anxiety. The CPT hazard function given by each covariate was first modeled individually (including an adjustment variable for starting temperature and a binary variable for Study), and then up to 4 of the most significant variables with coefficient p < 0.2 were selected. These variables were tested for high correlation (continuous variables: Pearson r > 0.7, categorical variables: χ² test p < 0.01), and only the most significant variable of any group of highly correlated variables was included in the model.

• Download figure
• Open in new tab

Figure 1: Diagram of statistical models.

The primary hypotheses were first tested (A) across all participants (both healthy control [HC] and fibromyalgia [FM] groups) and (B) within the fibromyalgia group. Exploratory tests were then conducted (B) within the fibromyalgia group as well as (C) within the opioid-taking subgroup. All tests were conducted using Cox proportional hazards models. Abbreviations: BDI, Beck Depression Inventory; BIS/BAS, Behavioral Inhibition System/Behavioral Activation System; PANAS, Positive and Negative Affect Schedule; POMS, Profile of Mood States; STAI, State-Trait Anxiety Inventory; BPI, Brief Pain Inventory; PROMIS, Patient-Reported Outcome Measurement Information System.

State anxiety, negative affect, number of pain areas in the body, and opioid medication use were hypothesized to decrease pain tolerance in patients with fibromyalgia. To test these hypotheses within the patient group, the CPT hazard function was modeled by each including STAI state anxiety, PANAS negative affect, FAS pain areas, and a binary variable indicating opioid medication use (Figure 1B).

Due to non-normality of the data, all group comparisons of the psychological/clinical variables were conducted using the Mann-Whitney Wilcoxon Test. The cox.zph function in R was used to confirm proportional hazards for all models. Observations with deviance residuals greater than 2.5 or less than -2.5 were identified as possible outliers.

2.5. Exploratory analyses

To validate the multivariate model and address the possibility of correlated groups of covariates in the model, Lasso regression was also used to identify variables predictive of CPT hazard. The Lasso approach tends to select one predictor out of a group of correlated predictors and discard the rest, and it can be more accurate than stepwise selection.⁴²

In addition to state anxiety, negative affect, number of body areas with pain, and opioid medication use, other factors potentially predictive of CPT survival within the patient group were also explored with Cox models. These exploratory variables included the covariates tested in the primary analysis as well as patients’ number of years with pain (BPI), average pain intensity (BPI), average pain interference (BPI), and PROMIS Fatigue score (Figure 1B). Because the study effect on CPT survival was significant, Study 1 and Study 2 were also separately analyzed for these relationships. Additionally, a stronger relationship between opioid medication use and CPT survival was observed in Study 2 compared to Study 1; to determine whether opioid dosage, duration of use, or timing of last intake (the opioid half-life range during which the CPT test was performed, calculated based on the time of last opioid dose and the drug-specific half-life) might account for this between-study difference observed in the opioid-taking population, a sub-analysis of the opioid-taking fibromyalgia group was conducted by testing for study differences in these three measures and separately testing them as predictors of CPT survival (Figure 1C). Lastly, a post-hoc sensitivity analysis of opioid medication effects was conducted by re-evaluating the primary results on the dataset including only non-opioid-taking patients and healthy controls.

3.1. Participants

Data from a total of 85 patients and 47 controls (see Table 1 for demographic information) were analyzed. Forty-three out of the 85 fibromyalgia participants were not taking opioid medications and were taking over-the-counter pain medications. The remaining 49% (n = 42) of the fibromyalgia patients were taking opioid pain medications. This sub-group had a median daily morphine equivalent dose (MED) of 20 mg morphine and a median opioid-taking duration of 5 years. Of the participants with fibromyalgia, 59% (n = 50) were taking anti-depressants, 26% (n = 22) were taking benzodiazepines or benzodiazepine-like medications, and 56% (n = 48) were taking muscle relaxants or anti-seizure medications (Table 2).

3.2. Psychological and clinical measures

Less than 5% of the questionnaire responses were missing from the data. Among the patient group, participants’ reported number of body areas (FAS) with pain ranged from 3 (the minimum criteria) to 19 (maximum). The median pain severity was 5.8/10 and median pain interference was 6.4/10 (BPI). The patient vs. healthy control groups showed significant differences in CPT time, STAI trait anxiety, STAI state anxiety, PANAS positive affect, PANAS negative affect, BAS fun-seeking, BIS score, POMS total mood disturbance, and BDI score (Table 3).

3.3. Group comparison of CPT survival

The distributions of CPT time by group are shown in Figure 2 (Mann-Whitney p = 0.00080). Fibromyalgia patients had significantly lower CPT survival probability compared to controls (HR = 2.17, 95% CI: [1.42, 3.31], p = 0.00035), when including adjustment variables for Study 1 vs. Study 2 and CPT starting temperature (Figure 3, Table 6).

• Download figure
• Open in new tab

Figure 2: Distribution of CPT time by group.

Distributions of CPT times from the fibromyalgia (FM) and healthy control (HC) groups are shown. The control group contained a larger proportion of censored times (CPT cut-off at 120 seconds) compared to the fibromyalgia group.

• Download figure
• Open in new tab

Figure 3: Kaplan-Meier curves of CPT assessment results by group.

Kaplan-Meier curves showing CPT survival with 95% confidence intervals are shown for the fibromyalgia (FM) group and healthy control (HC) group. The p-value shown is for the group effect in the Cox regression model of CPT survival time with adjustments for starting temperature and study effect. Time-to-event data was censored at 120 seconds, and proportional hazards criteria were met for Cox regression.

Regarding study-specific effects, between the fibromyalgia participants in Study 1 and Study 2, there was a significant difference in CPT time (p = 0.00079). The patient group of Study 2 was characterized by a higher proportion of censored CPT times (16%) compared to Study 1 (3%). The healthy control group, however, showed no significant differences in CPT time between studies.

Results from the single-variable Cox models of CPT hazard, which each included a psychological/clinical variable, showed that none were significantly associated with CPT survival (p > 0.05) after accounting for starting temperature, study differences, and group status. However, both Study 1 and fibromyalgia group status were significantly associated with greater hazard in each of these models. The multivariate Cox model included the variable for patient group status (HR = 1.99, p = 0.0020), selected variables which included BAS fun-seeking (HR = 0.94, p = 0.28), BAS drive (HR = 0.97, p = 0.50), and “other” race (HR = 1.78, p = 0.15), as well as an indicator variable for Study 2 and an adjustment variable for starting temperature (exact, measured before the CPT assessment). The multivariate model did not improve the fit for the CPT data compared to the simpler model with no psychological predictors (LRT: p = 0.16).

Variable selection with Lasso regression revealed similar results: the binary variables for patient group status (HR = 1.71, p = 0.048) and Study 2 (HR = 0.61, p = 0.017) were significant covariates selected by Lasso (10-fold cross-validation using Harrell’s C-index, and coefficients were extracted at the value of λgiving the minimum mean cross-validation error). BAS fun-seeking, BIS score, “other” race, and STAI state anxiety were also selected but were not significant predictors of CPT survival. The Lasso-selected Cox model was neither a significant improvement upon the p-value-selected multivariate Cox model (partial LRT: p = 0.82) nor upon the original simple Cox model with no psychological/clinical predictors (LRT: p = 0.23).

3.4. Psychological and clinical predictors of CPT survival in fibromyalgia patients

Within the patient group, STAI state anxiety, PANAS negative affect, number of pain areas, and opioid medication use were all not significantly associated with CPT survival probability (p > 0.0125). Other exploratory variables, including BAS fun-seeking (HR = 0.86, p = 0.014) and BAS drive (HR = 0.89, p = 0.021), had stronger relationships (i.e., larger effect sizes) with CPT survival in fibromyalgia patients (Table 4) (but not in healthy controls). The CPT survival rate in Study 2 patients was greater compared to Study 1 patients (HR = 0.48, 95% CI: [0.30, 0.77], p = 0.0026) even when accounting for CPT starting temperature. (This study difference in CPT survival was absent in the healthy control group [HR = 0.83, p = 0.62]). No psychological or clinical variables were significant predictors of CPT survival in the patient group of either study after correction for multiple comparisons. However, use of opioid medication was associated with greater CPT survival in Study 2 (HR = 0.51, 95% CI: [0.27, 0.97], p = 0.041) while its effects were unclear in Study 1 (HR = 0.88, p = 0.71), although no interaction effect was detected between study and opioid medication use (p = 0.26).

3.5. Between-study variability: opioid-taking fibromyalgia patient subgroup

As noted above, the relationship between opioid medication use and CPT survival among fibromyalgia patients was stronger in Study 2 compared to Study 1 (Table 4). In the subgroup analysis of fibromyalgia patients who were taking opioids, patients had significantly different CPT times in Study 1 vs. Study 2 (p = 0.0028). In contrast, CPT times across studies did not differ significantly for the subgroup of patients with fibromyalgia who were not taking opioids (p = 0.11) nor for the healthy control group (p = 0.52) (Table 5).

Although CPT times in opioid-taking fibromyalgia patients differed between Study 1 and Study 2, neither duration of opioid medication use (p = 0.97), nor daily opioid dose (p = 0.96), nor distribution of intake timing (χ² test p = 0.091) significantly differed between studies (Table 5). Furthermore, none of these measures were significant predictors of CPT survival in the opioid-taking patient subgroup.

3.6. Post-hoc sensitivity analysis of opioid medication effects

Due to the additional between-study variation present in the subgroup of patients with fibromyalgia who were taking opioids compared to the rest of the sample (Figure 4), the opioid-taking subgroup was removed from the analysis to identify any changes in the group and study effects on CPT survival. After removing the opioid-taking subgroup, the Cox model indicated that CPT times were no longer different between studies (p = 0.10), the study effect was no longer significant (HR = 0.67, p = 0.11), and the group effect increased (HR = 2.70, p = 0.000054) (Table 6). When assessing study effects on CPT survival within each group/subgroup separately, Study 2 was associated with higher CPT survival probability in opioid-taking patients (HR = 0.40, p = 0.013), but the study effect was not significant in non-opioid-taking patients (HR = 0.60, p = 0.12) or healthy controls (HR = 0.83, p = 0.62).

• Download figure
• Open in new tab

Figure 4: Distributions of patient CPT times by study and opioid medication use.

CPT times of fibromyalgia patients in Study 1 and Study 2 are shown by subgroup (opioid-taking vs. non-opioid-taking). The distribution of CPT times for opioid-taking patients in Study 2 is shown to be more uniform compared to the right-skewed distributions of CPT times in all other fibromyalgia subgroups.

4.1. Reduced CPT survival in fibromyalgia and potential contributors

Confirming previous findings of how individuals with fibromyalgia demonstrate lower tolerance to painful stimuli,^8–10 in our dataset, individuals with fibromyalgia showed lower rates of survival in the CPT assessment. Under both p-value-based and Lasso regression variable selection methods, fibromyalgia status was the most significant predictor of CPT survival probability even when accounting for psychological measures including anxiety, depression, and negative affect. This suggests that fibromyalgia may impair cold pain tolerance in ways beyond such psychological processes alone, for example, in conjunction with underlying neurobiological mechanisms. In this study, neuropathic measures of chronic pain that may be potentially related to pain tolerance were not evaluated. However, pain sensitivity in fibromyalgia may be attributed to biological alterations spanning both peripheral and central nervous system.^43,44 Among different patients, fibromyalgia can be characterized by varying amounts of peripheral nociceptor pathologies^45,46 and central alterations such as altered descending control of pain and central sensitization.^47,48 The sensation of cold pain, specifically, is mediated by transient receptor potential melastatin 8 (TRPM8), and alterations to this receptor may account for cold hyperalgesia in a variety of chronic pain conditions, including fibromyalgia.^49,50 Nerve fiber diameter, collected via skin biopsy, has also been shown to be reduced in some fibromyalgia patients.⁵¹ In future studies of cold pain tolerance, measures of neuropathic pain should be collected in both patients and healthy controls, to complement questionnaire measures.

4.2. CPT survival, Behavioral Activation System, and reward response

Overall, the psychological and clinical questionnaire measures collected in the studies were not found to have clear relationships with CPT survival. However, some relationships became more distinct when survival analyses were performed separately for fibromyalgia or healthy control groups. For instance, higher scores in BAS drive and BAS fun-seeking were slightly predictive of higher CPT survival among fibromyalgia patients, but not among healthy individuals.

These results suggest that the psychological contributors to pain tolerance may differ between fibromyalgia patients and healthy individuals. For example, pain tolerance in fibromyalgia patients, compared to healthy controls, may be more susceptible to individuals’ intrinsic drive to achieve goals (i.e., BAS drive), and/or willingness to engage in unknown but possibly rewarding activities (i.e., BAS fun-seeking). Previous research has proposed the BIS-BAS model of chronic pain, suggesting that pain responses can be predicted by individual perceptions of reward and punishment.⁵² Furthermore, prior neuroimaging research in patients with fibromyalgia has demonstrated that BAS drive is related to differences in regional brain response to anticipation of rewards²⁷ and avoidance of punishment.²⁹ Thus, alterations in recruited CNS activities ongoing during the CPT assessment may contribute to the observed differences in CPT survival between healthy controls and patients with fibromyalgia.

4.3. Effects of opioid medication use

CPT survival times were significantly longer in opioid-taking patients compared to non-opioid-taking patients in one cohort of patients (Study 2), but not both. Furthermore, there was no evidence to suggest that these differences could be explained by patients’ daily dose of opioid medication, duration of opioid medication use, or the timing of their last opioid intake. Meanwhile, in line with CPT survival relationships with BAS measures as described above, it is possible that changes in reward response in patients with fibromyalgia who take opioids²⁹ may relate to increased cold pain tolerance. Nonetheless, the observed inconsistencies across studies suggest that the relationship between opioid medication use and cold pain tolerance in fibromyalgia patients is tenuous, in that it may vary across patient cohorts and requires further investigation. Prior research has shown that individuals (with and without chronic pain) taking opioids demonstrate lower cold pain tolerance when measured prior to scheduled opioid dose.⁵³ In another study, compared to non-opioid patients, opioid-taking patients with chronic pain had similar cold pain tolerance when measured within several hours after their morning analgesic dose.⁵⁴ The timing of opioid dose prior to CPT was variable in the present analysis, therefore, timing effects of opioids are unclear, yet given such prior evidence, opioid timing may be an important variable in assessing cold pain tolerance. Additionally, other studies in individuals with opioid use disorder (i.e., opioid addicted individuals without noted presence of chronic pain) have shown decreased opioid tolerance (by survival analysis) when individuals were tested at a time when opioids (e.g., heroin, methadone) were present in their system.⁵⁵ Together, these prior research findings suggest potential differences in opioid effects on cold pain tolerance in patients with vs. without chronic pain.

4.4. Limitations

In this analysis, two separate studies were combined to maximize statistical power. However, differences between the two studies were found to significantly affect survival rates in the CPT. Study 2, in which the testing temperature was standardized to 5° C, was associated with lower CPT hazard compared to Study 1 (with individualized temperatures) when accounting for group status. Although the group effect was the most salient predictor of CPT survival in both studies (when each study was analyzed separately), the study effect in the combined analysis was nearly as notable as the group effect. Such significant differences in CPT survival between studies suggest limitations regarding generalizability of CPT findings across different geographic/demographic cohorts and indicate that slight differences in CPT procedures (i.e., water temperature) may influence generalizability of results across studies comparing fibromyalgia patients. Additionally, it is important to note that the within-group analyses for relationships of clinical/psychological factors (excluding opioid medication use, PANAS negative affect, number of pain areas, and STAI state anxiety) with CPT survival were exploratory, none of which were significant after correcting for multiple comparisons (which may be attributed to a lack of power). Lastly, as described above, timing of opioid intake prior to the CPT assessment in patients who take opioids may be an important contributor to evaluation of cold pain tolerance.

4.5. Conclusions

Through using survival analysis to analyze time-to-event CPT data more suitably, the results of this analysis confirm previous findings that, compared to healthy individuals without chronic pain, individuals living with fibromyalgia demonstrate lower tolerance to cold pain. To further understand the relationship between clinical/psychological factors and cold pain tolerance in individuals with fibromyalgia, future investigations should include measures to evaluate the presence of peripheral neuropathy and comparisons of CPT to other tests of CNS and pain processing function in the same patients. Furthermore, the results suggest a potential effect of opioid medication on increasing cold pain tolerance in some fibromyalgia patients. The effect of increased cold pain tolerance in a subset of patients with fibromyalgia who are taking opioids may be related to differences in reward response and motivation and regionally influenced variables such as prescribing practices, and it requires further investigation with timed opioid intake approaches.