Testing the sequential model of pain processing in irritable bowel syndrome: a structural equation modeling analysis

European Journal of Pain 9 (2005) 207–218 www.EuropeanJournalPain.com

Testing the sequential model of pain processing in irritable bowel syndrome: a structural equation modeling analysis Jeffrey M. Lackner a

a,*

, James Jaccard b, Edward B. Blanchard

b

Behavioral Medicine Clinic, Department of Medicine, University at Buffalo School of Medicine and Biomedical Sciences, State University of New York, ECMC, 462 Grider Street, Buffalo, NY 14215, USA b Department of Psychology, University at Albany, State University of New York, Albany, NY 14222, USA Received 25 February 2004; accepted 2 June 2004 Available online 25 June 2004

Abstract Pain, the cardinal feature of irritable bowel syndrome (IBS), is a multidimensional phenomenon with sensory and affective dimensions. Price’s [Psychological Mechanisms of Pain and Analgesia, 1999] pain processing model was used to delineate four a priori sequentially related stages (pain sensation intensity, immediate pain unpleasantness, long-term suffering, and pain-related behavior). Although prior research with both healthy individuals and somatic pain patients supports the model in general, its applicability to IBS is unclear. Our goal was to extend the scope of the sequential model and test its fundamental tenets using structural equation modeling (SEM) with data obtained from 168 Rome II diagnosed IBS patients (19% male, 81% female). A secondary goal was to assess the relationship between a set of contextual factors associated with IBS (age, gender, trait anxiety) and the four pain stages. Results were consistent with a successive order of pain processing such that the pain sensation directly impacts pain unpleasantness, which, in turn, leads to suffering and illness behaviors. However, contrary to a model with strictly successive stages, pain sensation had independent effects on illness behaviors over and above pain affect. The effect of anxiety on illness behavior was mediated by suffering, while psychopathology directly influenced pain sensation and pain unpleasantness but not later stages. Age was related to pain sensation and illness behaviors but not pain affect. Gender tended to be more strongly associated with more distal pain stages (e.g., pain affect) vis-a-vis its effects on pain sensation. These data are generally supportive of a four-stage pain processing model. Ó 2004 European Federation of Chapters of the International Association for the Study of Pain. Published by Elsevier Ltd. All rights reserved. Keywords: Chronic pain; Age; Gender; Irritable bowel syndrome; Anxiety; Psychopathology

1. Introduction Irritable bowel syndrome (IBS) is a persistent, painful functional gastrointestinal disorder whose global prevalence of
20% (Camilleri and Choi, 1997) makes it one of the most common chronic pain disorders. Although the primary clinical features of IBS include both abdominal pain and altered bowel habits (constipation, diarrhea, or an alternating pattern), pain/discomfort is its cardinal feature (Drossman et al., 2000.), the most bothersome symptom to patients (Society for Women’s Health Re*

Corresponding author. Tel.: +1-716-898-5671; fax: +1-716-8983040. E-mail address: lackner@buffalo.edu (J.M. Lackner).

search, 2002), and the symptom that best predicts health care seeking behavior (Talley et al., 1997). Like other forms of pain, abdominal pain is a multidimensional experience involving both sensory and affective components (Melzack and Casey, 1968). Although the medical literature has tended to focus on the sensory aspect (Price, 1999), the affective dimension of pain has received growing attention partly because it has been identified as playing a greater role in shaping the course of clinical pain problems (Chapman and Stillman, 1996). A growing body of research indicates that the affective dimension of pain is not a unitary construct but composed of two subcomponents: immediate and secondary pain affect (Price, 2000). Immediate pain affect refers to the immediate unpleasantness of pain experi-

1090-3801/$30 Ó 2004 European Federation of Chapters of the International Association for the Study of Pain. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.ejpain.2004.06.002

208

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

ence and reflects the perceived threat to one’s physical integrity associated with actual or potential tissue damage (Chapman and Gavrin, 1999). Under laboratory conditions, the intensity of immediate pain affect correlates, albeit imperfectly, with the intensity of pain sensation triggered by a noxious stimulus (Price et al., 1987). Psychological factors have not been found to correlate strongly with pain sensation and only modestly with immediate pain affect (Harkins et al., 1989; Wade et al., 1992). If persistent, pain can compromise quality of life, heighten attentional focus to bodily sensations and other sources of internal experience (e.g., worry), and tax adjustment (Chapman and Gavrin, 1999). With these changes, psychological distress spreads to and damages other aspects of one’s self-concept (e.g., selfevaluative concerns relating to one’s self-identity, selfesteem, and role status, Chapman and Gavrin, 1999). These changes make up the long term suffering aspect of pain, which Price (1999) terms secondary pain affect. Its associated emotions reflect not only the physical threat value of pain stimuli but also the discrepancy between one’s idealized expectations and aspirations, and actual performance (Cassell, 1991; Chapman and Gavrin, 1999; Price, 1988) which widens as pain and its effects persist. Secondary pain affect presumably involves more sustained and deeper cognitive processing and is characterized by negative emotional states (e.g., frustration, depression, anxiety) and greater distortions in information processing than the more pre-conscious process of immediate pain (Price, 1988). 1.1. Dimensions of pain in irritable bowel syndrome A detailed analysis of the primary dimensions of pain and their interrelationships is critical to understanding the nature of IBS pain but has been complicated by the absence of: (1) an empirically validated theoretical model for understanding the dimensions of pain in IBS, their subcomponents, and interrelationships; (2) psychometrically sound assessment instruments that differentiate immediate pain affect from secondary pain affect; (3) the application of theoretically appropriate statistical procedures for analyzing causal models that describe patterns of relationships among pain dimensions in clinical populations. Price and associates’ (Price, 1999; Wade et al., 1996; Wade et al., 1992) pain processing model provides a conceptual framework for understanding aspects of pain and their relationship to one another. Their model features four pain processing stages (pain sensation, immediate pain affect, secondary pain affect and pain-related behavior) that supposedly impact on each other in a sequential manner (Price, 2000). Wade et al. (1996) tested this model in chronic somatic pain patients using structural equation modeling (SEM) and reported results that they interpreted as

generally consistent with a multistage formulation. Of interest is whether the formulation extends to the problem of IBS. One purpose of this study was to test causal models of the relationships between intensity of pain sensation, immediate pain affect, long term suffering and illness behaviors in more severely affected IBS patients. The theoretical structure proposed by Price (1999, 2000) and associates (Wade et al., 1996) was used as a starting point for model evaluation. SEM based methods were used to elaborate this causal structure and elucidate psychological mechanisms not previously observed by Wade et al. and that appear to be unique to IBS. 1.2. Individual difference variables across pain processing stages Subsequent elaborations of the sequential model have explored the extent to which psychosocial factors exert differential effects on various pain processing stages. Particular emphasis has been on gender, personality, and age factors. We consider each in turn. 1.2.1. Gender Most research on gender differences has focused on the sensory dimension of pain (Price, 1999). In general, the literature suggests that while women tend to show lower pain thresholds and lower pain tolerances than men (Dubreuil and Kohn, 1980; Ellermeier and Westphal, 1995; Leon, 1974; Woodrow et al., 1972), these differences are relatively modest and are most reliably observed in response to acute somatic pain stimuli applied under controlled laboratory conditions. There is reason to doubt that this pattern of results necessarily extends to the problem of visceral pain problems like IBS (Cook et al., 1987; Whitehead et al., 1990). In one of the few studies that explored gender differences of visceral pain, Naliboff et al. (2003) found that female IBS patients had lower perceptual thresholds than male IBS patients for visceral stimuli. What little research has investigated gender differences on+ affective dimensions of pain has found inconsistent findings (Bush et al., 1993; Riley et al., 2001; Unruh, 1996) Gender research has yielded no more conclusive findings in IBS patients (Chang and Heitkemper, 2002). We sought to clarify our understanding of gender effects by embedding their analyses in a broader nomological network characterized by the four-stage model of pain processing. Such multivariate evaluations permit us to identify mediators of gender effects at the different stages of pain experience and yield greater insights into the mechanisms by which gender impacts distal outcomes. 1.2.2. Age In contrast to the influence of gender, age appears to affect later stages of pain processing, with older indi-

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

viduals reporting worsening impact and secondary pain affect than their younger counterparts (Riley et al., 2000). These data (Gordon, 1979) are partly based on chronic pain syndromes (e.g., fibromyalgia, arthritis, herpes zoster, trigeminal neuralgia) whose prevalence increases with age. IBS however affects mostly women in their childbearing years and decreases with age (Drossman et al., 1990; Drossman et al., 1997). These data suggest the hypothesis of age related differences in pain perception patients such that older IBS patients may report lower pain intensities than their younger counterparts. To date, there is no known research that has systematically assessed gender and age in the context of a sequential pain processing model of IBS pain. The present study does so. 1.2.3. Psychopathology Interestingly, there is scant research exploring how psychopathology affects different stages of pain processing. This is a notable void in light of the established link between psychopathology and chronic pain in general (Dersh et al., 2002) and IBS in particular (Blanchard and Scharff, 2002). Gatchel (1996) has developed a multistage conceptual model for understanding the progression from acute pain to chronic pain that offers clues about the relationship between psychopathology and pain processing stages. Stage 1 is associated with psychological distress expressed primarily in the form of anxiety tied to the degree of physical threat evoked by pain stimuli. This resembles Price’s immediate pain affect process. With the persistence of pain beyond a normal healing time (2–4 months) comes stage 2 (descriptively similar to Price’s secondary pain affect stage). Whereas the primary emotional response in stage 1 involves anxiety, stage 2 is distinguished by an additional ‘‘layer of behavioral/psychological problems [depression, anxiety, anger, somatization] over the original nociception or pain experience itself’’ (Gatchel, 1996, p. 34), as the patient wrestles with the long-term implications of pain. This suggests that psychopathology is more likely to influence later stages of pain processing (e.g., secondary pain affect, behavioral functioning) not early stages (pain sensation, immediate unpleasantness). It also suggests a stronger impact of anxiety on earlier stages of pain than prior pain processing research has found (Harkins et al., 1989; Wade et al., 1992). The present study tested these possibilities.

209

described below). In addition, individual differences associated with gender, age, and psychopathology are explored in the context of the model, thereby isolating potential mediators of the effects of these variables on distal outcomes. Such analyses provide greater clarity on the mechanisms underlying these important individual differences. The literature to date suggests the following working hypotheses relative to the causal model we evaluated: Hypothesis 1. The four stages of pain processing will yield data patterns that are consistent with sequential causal influence, such that that pain sensation intensity impacts pain-related unpleasantness, which, in turn, impacts pain-related suffering which, in turn, impacts behavioral functioning. Hypothesis 2. Illness behavior measures will be predicted by measures of pain related suffering (PDS), but estimates of the effects of prior stages of pain processing on behavioral functioning will be mediated by pain related suffering. Hypothesis 3. Gender will exhibit data patterns consistent with a causal influence of gender on the intensity of pain sensation, with females reporting more intense pain sensation and males reporting more intense pain unpleasantness. Hypothesis 4. Age will exhibit data consistent patterns with causal effects on later stages of pain processing but not necessarily on earlier stages of pain processing. Hypothesis 5. Trait (chronic) anxiety will exhibit data patterns consistent with a causal influence on pain unpleasantness but may or may not do so for suffering. Hypothesis 6. Psychopathology will exhibit data patterns consistent with causal influence on pain-related suffering but may or may not do so for earlier stages of pain processing (pain sensation intensity and pain unpleasantness).

2. Methods 2.1. Participants

1.3. Summary In sum, the present research evaluates a causal model of stages of pain processing but extends the model to the analysis of IBS patients. The research uses structural equation modeling to evaluate the model in a multivariate context and uses measures that have psychometric advantages to those used in previous studies (as

Participants were 168 consecutively evaluated IBS patients (81% female, 19% male) referred to an academic behavioral medicine clinic as part of an NIH-funded clinical trial of two psychological treatments for IBS. Inclusionary criteria required a Rome II diagnosis (Drossman et al., 2000) confirmed by a board certified gastroenterologist who conducted a medical examination

210

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

to exclude organic GI disease. The ethnic composition of the sample was 96% Caucasian, 2% African American, 1% Native American, and 1% Asian. The mean age of participants was 49 (SD ¼ 14) years. The average age at onset of IBS symptoms was 33 (SD ¼ 16) years. During pretreatment psychological testing, participants completed measures of pain sensation and affect, psychopathology, trait anxiety, secondary pain affect, and illness behaviors and these are the data analyzed here. The severity of IBS symptoms was assessed by the study gastroenterologist who rated the overall severity of IBS symptoms as mild, moderate, severe, or very severe (1 ¼ mild, 4 ¼ very severe) in a manner comparable with prior IBS research (Drossman et al., 2003). Three percent were classified as mild, 34.9% as moderate, 52.8% as severe, and 8.7% as very severe. Additional information (central tendency and variability) for the primary variables measured are presented in Section 3. 2.2. Measures The interpretability of past sequential pain stage research has been complicated by a reliance on single item Visual Analogue Scales (VAS) and composite indices of single item mood measures to assess painrelated affect. This approach requires individuals to rate the intensity of a pain-related mood state (e.g., fear, anger anxiety, depression) on a 10-point rating scale whose end points are anchored with adjectives of ‘‘none’’ and ‘‘the most severe imaginable’’. Notwithstanding its face validity, history, and convenience, the VAS approach to mood assessment suffers from acknowledged psychometric limitations (Riley et al., 2001; Wade et al., 1990). The practice of combining ratings of multiple emotions in an unweighted fashion to measure pain affect may be particularly problematic for IBS patients whose primary emotional experience involves anxiety and whose depressive symptoms are within clinically unremarkable levels (Blanchard, 2001). VAS mood measures also suffer from problems of content validity. Pain affect is, according to Price and others, a dimensional construct moderated by cognitive appraisal of threat (immediate pain affect) or long-term meaning of pain experience (secondary pain affect). By gauging the severity of pain affect on the basis of the magnitude of an affective response (e.g., general anxiety, fear, depression), VAS mood measures do not explicitly or directly assess the underlying cognitive content (e.g., perceptions of uncontrollability, unpredictability) which define pain affect (Buytendyck, 1961; Price, 1999). The present study sought to circumvent these limitations by using multi-item measures that have more established psychometric properties than measures used in previous research on the analysis of sequential pain processing. In addition, we applied analytic methods that examined the impact of varying

levels of measurement error on the nature of conclusions made with respect to a model. 2.2.1. Pain sensation and pain-related affect Pain sensation and immediate pain affect were assessed using the short form of the McGill Pain Questionnaire (SF-MPQ, Melzack, 1987). The SF-MPQ consists of 15 words reflecting the most commonly used adjectives for describing the sensory (11 words) and affective (4 words) quality of pain experience during four weeks prior to assessment. Patients rate the intensity of these descriptors on a four point scale where 0 ¼ none, 1 ¼ mild, 2 ¼ moderate, 3 ¼ severe. Psychometric studies have found strong correlations between the major indices of the SF-MPQ and the original version (Melzack, 1975) which previous pain processing research has used to measure pain sensation and pain affect (Harkins et al., 1989). 2.2.2. Psychopathology Psychopathology was assessed using the 53 item Brief Symptom Inventory (Derogatis, 1993), Respondents indicate on a 5-point scale (0 ¼ not at all, 5 ¼ extremely) their level of distress for nine types of problems (e.g., anxiety, somatization, depression). The average intensity of all items yields a composite index of psychological distress (Global Severity Index) which has been used in prior pain research (e.g., Lackner and Gurtman, in press; Lee et al., 2001). The internal consistency, test– retest reliability, and validity of the BSI are well established (Derogatis, 1993). 2.2.3. Secondary pain affect Secondary pain affect (long term suffering) was measured using the Pain Discomfort Scale (PDS, Jensen et al., 1991). The PDS is a 10-item measure that requires the subjects to rate on a 5 point scale ranging from 0 (this is very untrue for me) to 4 (this is very true for me) the extent to which they agree with cognitive and affective responses associated with pain-related suffering (e.g., ‘‘The pain I experience is unbearable’’). The PDS has sound psychometric properties (internal consistently, test–retest reliability, construct validity, Jensen et al., 1991). 2.2.4. Trait anxiety The Trait subscale of the State-Trait Anxiety Inventory (Speilberger et al., 1970) was used to measure trait anxiety (i.e., general tendency to respond fearfully to stressors in general). In responding to the 20 items of the T -Anxiety scale, subjects indicate how they generally feel by rating the frequency of their feelings of anxiety on a four point scale ranging from 1 (almost never) to 4 (almost always). The trait scale of the STAI has sound psychometric properties (e.g., internal consistency, stability, validity) and is empirically separable from

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

psychopathology measures focusing on the intensity of emotional experience (Speilberger et al., 1970). 2.2.5. Illness behaviors Illness behaviors were assessed using the Physical Limitations (PF) and Role Physical Limitations (RP) subscales of the SF-36 Health Survey (Ware and Sherbourne, 1992). The PF subscale assesses self-reported limitations in performing physical activities due physical health problems. The RF subscale assesses perceived problems (e.g., difficulty performing or reducing the amount of time spent in work or other usual activities) due to physical health. The internal consistency, test– retest reliability, and validity of both the Physical Limitations and Role Physical Limitations subscales are well established (Ware et al., 2000).

3. Results 3.1. Preliminary analyses 3.1.1. Descriptive statistics Table 1 presents means and standard deviations for all of the continuous variables used in the models. The median values for each of the variables (not reported) were close to the mean values. Mean GSI scores correspond to T scores of 58 (female) and 61 (male) for adult nonpsychiatric norms (Derogatis, 1993). Mean scores for MPQ-SF sensory and affective indices were within typical levels (Melzack, 1987). The mean T-STAI value corresponded to a percentile rank of 78 (female) and 76 (male) for normal adults age 40–49 (Speilberger et al., 1970). 3.1.2. Outliers Both model based and non-model based outlier analyses were pursued. For the latter, a leverage score was calculated for each respondent based on their multivariate profile for the nine variables included in model analyses. The mean leverage score across respondents was 0.054 and an outlier was defined as anyone having a leverage score three times the value of the mean (Jaccard and Wan, 2003). No outliers were Table 1 Descriptive statistics Variable

Mean

SD

Skewness

Kurtosis

Pain sensation Pain affect Suffering Physical functioning Physical role limitation Age Psychopathology Trait anxiety

10.54 3.74 17.01 76.23 44.94 49.11 0.52 40.28

6.74 3.37 8.26 24.64 39.03 14.08 0.44 9.91

0.82 0.87 )0.09 )0.98 0.17 )0.10 1.56 )0.04

0.19 )0.28 )0.25 )0.05 )1.49 )0.76 2.61 )0.76

211

evident using this criterion. Model based outliers were examined using limited information regression analyses for each of the linear equations dictated by the various path models tested (Bollen, 1996). We examined DfBeta values for each individual relative to each path coefficient to isolate unusually influential individuals in parameter estimation. An outlier was defined as individuals who had dfBetas three times larger than the standard error of a coefficient. No outliers were evident in these analyses. 3.1.3. Missing data There were small amounts of missing data amounting to no more than a few cases on any given variable. There was no coherent pattern to the missing data. For those individuals with missing data, values were imputed to conform to covariance estimates consistent with the application of the Expectation–Maximization (EM) approach to missing data (Schafer, 1997). We did not use full information maximum likelihood methods because of non-normality in the variables and our need to use bootstrapping to estimate standard errors and confidence intervals for the path coefficients (as described in the following section). Given the small number of instances of missing data, concerns surrounding estimation with missing information are moot. 3.1.4. Non-normality Traditional maximum likelihood methods of SEM assume that the continuous variables in the model are multivariately normally distributed. This was tested using the Mardia test for multivariate kurtosis, which yielded a statistically significant result (critical ratio ¼ 2.56, p < 0:05). This suggests the presence of nonnormality at the multivariate level. Skewness and kurtosis indices for each variable are presented in Table 1. Troublesome skewness and kurtosis values are evident for the measure of psychopathology. Given this, the decision was made to pursue parameter estimation under two scenarios, traditional maximum likelihood analysis and bootstrapping. For the bootstrap analyses, we performed 2000 bootstrap replications for purposes of estimating standard errors, p values, and confidence intervals. We used the bias corrected approach to interval estimation as implemented in the computer program AMOS (Arbuckle and Wothke, 1999). Estimation for the individual bootstrap samples was well-behaved and yielded convergence and meaningful solutions in all 2000 instances. The p value for overall fit of the tested models was calculated using the Bollen–Stine bootstrap approach in place of the traditional chi square statistic (Bollen and Stine, 1993). In general, conclusions were the same in both estimation approaches. All significance tests and confidence intervals reported are from the bootstrap analyses.

212

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

Psychopathology Trait Anxiety

(1.59).21 (3.51).23

.51

(.40).48

.57

.87

(.31).61

Immediate Unpleasantness

Pain Sensation

(.83).34

Suffering

(-1.38)-.19) (-2.99)-.18 (-1.05)-.22 (-.08)-.17

Gender

(-.88)-.24

(-1.70)-.29

(-.45)-.16

Age

Physical Role Li mitations (-.48)-.28

Physical Functioning

.82

.34 .79

Fig. 1. Final SEM Model for the Four Pain Processing Stages. Rectangles are observed (measured) variables, circles are standardized error variances, values in parentheses are unstandardized path coefficients, values not in parentheses are standardized path coefficients, straight lines with arrows are presumed causal paths, double headed curved lines are correlations (with the correlation value just above the line). All exogenous variables were assumed to be correlated, but these correlations are omitted from the diagram to reduce clutter.

3.2. Model tests The initial model tested can be described in conjunction with Fig. 1, which represents the final model that we settled upon. The central feature of the model was the sequential causal links between pain sensation, pain affect, long-term pain suffering and self reported behavior. In this case, there were two behavioral outcomes, physical role limitations and physical functioning. Although there is no causal link depicted in Fig. 1 between long term suffering and physical functioning, this path was included in the initial model. All other variables in Fig. 1 were included in the model and assumed the role of exogenous variables. Causal paths were defined from each exogenous variable to each of the five endogenous variables, creating a saturated model (with the exception of the patterning of causal links between the five endogenous variables). All resid-

ual variances (reflected by the circles in Fig. 1) were assumed to be uncorrelated and all exogenous variables were assumed to be correlated. The Bollen–Stine p value for this model was statistically significant ðp < 0:001Þ suggesting poor model fit. The more traditional indices of global fit yielded a mixed picture (v2 ¼ 26:76; df ¼ 4; p < 0:001; GFI ¼ 0.97; CFI ¼ 0.94, RMSEA ¼ 0.19; close fit test p value <0.001; standardized RMR ¼ 0.045). Inspection of model diagnostics revealed that the sole source of ill fit was the assumption of uncorrelated residuals for the two behavioral outcomes, physical functioning and physical role limitations. Two strategies were considered for dealing with the ill fit. One strategy was to model the two illness behavior measures as indicators of a common latent variable reflecting a more global construct of physical impairment. This strategy was rejected because diagnostics also made evident that

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

213

Table 2 95% Confidence intervals for unstandardized path coefficients Coefficient

Lower limit

Upper limit

p Value

Pain sensation to pain affect Pain affect to suffering Anxiety to suffering Suffering to physical role limitations Pain sensation to physical functioning Pain sensation to physical role limitations Gender to pain sensation Age to pain sensation Age to physical functioning Age to physical role limitations Psychopathology to pain sensation Psychopathology to pain affect

0.25 0.54 0.31 )1.66 )1.93 )2.46 )5.14 )0.14 )0.70 )0.87 1.15 0.81

0.36 1.09 0.51 )0.30 )0.75 )0.95 )0.60 )0.02 )0.23 )0.05 5.63 2.56

0.001 0.001 0.001 0.005 0.001 0.001 0.022 0.014 0.001 0.034 0.006 0.001

the two variables were differentially related to other constructs in the model. Treating them as indicators of the same underlying construct would obscure these differences. The second strategy was to maintain the conceptual distinctions between the variables but to permit the residuals to be correlated. This is justified theoretically if one can specify variables outside of the theoretical system that might serve as common causes of the two constructs. Because specification of such variables is straightforward (e.g., health status in general, overall physical fitness), we adopted this strategy. 1 The revised model with correlated error was re-fit to the data and the model yielded good fit. The Bollen–Stine p value was 0.26 and all of the traditional indices of overall fit were satisfactory (v2 ¼ 5:31; df ¼ 3; p < 0:15; GFI ¼ 0.99; CFI ¼ 0.99, RMSEA 6 0.07; close fit test p value <0.29; standardized RMR ¼ 0.018). In addition, more focused fit tests (examination of modification indices, offending estimates, standardized residuals and evaluations of theoretical coherence) all suggested adequate model fit. We examined the path coefficients for this model and deleted all paths from the model that were not statistically significant. To control for chance effects across multiple tests of significance, we adopted a modified Bonferroni criterion for declaring statistical significance of a path coefficient based on the False Discovery Rate (FDR) method (Keselman et al., 1999). Using this method, a family of tests was defined as the path coefficients leading from the exogenous variables to a given endogenous variable. As convention dictates, withinfamily error rates were controlled using the FDR, but controls across families were not invoked. This yielded the model in Fig. 1. The trimmed model was re-fit and good model fit was still manifest. The Bollen–Stine p value was 0.26 and the traditional fit indices were 1

A third option was to introduce reciprocal causality between the two variables, which would be possible to pursue in the present case because of the presence of instrumental variables in the theoretical system. We chose not to pursue this formally because the logic of correlated errors seemed more compelling.

v2 ¼ 21:67; df ¼ 17; p < 0:20; GFI ¼ 0.97; CFI ¼ 0.99, RMSEA 6 0.04; close fit test p value <0.58; standardized RMR ¼ 0.046. The path coefficients in Fig. 1 are from the trimmed model. Both unstandardized and standardized path coefficients are presented, with unstandardized coefficients in parentheses. All residuals and correlations are in standardized metrics. Correlations between the exogenous variables are omitted for purposes of figure clarity. All path coefficients were statistically significant ðp < 0:05Þ. Table 2 presents the bias corrected confidence intervals from the bootstrap analyses for the unstandardized coefficients. None of the path coefficients in the trimmed model were impacted much by the trimming (i.e., their values in the more saturated model were comparable to those in the trimmed model). All were statistically significant before trimming and all remained statistically significant after trimming. This trimmed model did not differ appreciably in the overall fit indices from the non-trimmed model. We highlight results in terms of the conceptual questions outlined in Section 1. 3.2.1. Sequential model of pain processing The sequential model of pain processing predicts that the path coefficients from pain sensation to pain affect, from pain affect to suffering, and from suffering to self reported behavior should be statistically significant. In general, this was the case. However, there were several notable results that were contrary to the model when just the four central constructs of sensation, affect, suffering, and behavior are considered. First, suffering was a statistically significant predictor of only one of the self reported illness behaviors (physical role limitations). Second, pain sensation had statistically significant direct effects on self reported illness behavior independent of pain affect or pain suffering. Whereas previous research has suggested that the effects of pain sensation on behavior should be mediated by affective responses to pain, our results suggest that pain sensation can affect behavior independent of these mechanisms. To be sure,

214

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

there is evidence that some mediation does, in fact, occur (as reflected by supplemental tests based on the bootstrapped analyses that we performed demonstrating a statistically significant reduction in the path coefficient from pain sensation to physical role limitations when the mediators of pain affect and suffering were held constant (see Hoyle and Kenny, 1999 for a description of the general logic of such tests). But despite this, the data also suggest that pain sensation affects behavior over and above any effects that it has on these mediators. The model in Fig. 1 is consistent with the proposition that (1) pain affect mediates the effect of pain sensation on pain suffering and (2) pain suffering mediates the effects of pain affect on behavior (when pain affect is related to behavior). Supplemental analyses on the reduction of path coefficients when the mediators were held constant yielded results consistent with complete mediation. Both of these propositions are in accord with past research on the sequential model of pain processing. 3.2.2. Gender effects The statistically significant path coefficient in Fig. 1 from gender to pain suggests gender differences in pain sensation (holding age and psychopathology constant). The unstandardized coefficient reflects the adjusted mean difference between males and females and indicates that, on average, females report higher levels of pain sensation than males (by about three scale units). Although gender did not have direct effects on pain affect, suffering, or behavior, the model suggests that this is because the effects of gender on these variables are mediated by pain sensation. When pain sensation is held constant, gender effects on these more distal variables reduces by a statistically significant amount and, in fact, to non-significant levels of prediction. Thus, the key to understanding gender differences on the more distal outcome variables seems to be the differential effects that gender has on pain sensation (holding age and psychopathology constant). 3.2.3. Age Like gender, there was a statistically significant path coefficient between age and pain sensation, with pain sensation tending to decrease as age increases. Although pain sensation seems to mediate the effects of age on pain affect, pain suffering, and to some extent, behavior, age also had independent effects on behavior over and above these mediators. When the effects of pain sensation were held constant, older individuals tended to report greater problems related to physical health than younger individuals. 3.2.4. Trait anxiety The only statistically significant path coefficient for trait anxiety was from anxiety to pain suffering. In

general, higher levels of anxiety were associated with higher levels of suffering (holding constant pain affect). Anxiety also had an indirect effect on physical role limitations (with higher anxious patients exhibiting more limitations and problems), but these effects were mediated by pain suffering. When long-term pain suffering was statistically held constant, the effects of anxiety on physical role limitations were reduced to nonsignificance. 3.2.5. Psychopathology Psychopathology yielded two statistically significant path coefficients. The first was to pain sensation, with individuals characterized by higher levels of psychopathology reporting higher levels of pain sensation (holding gender and age constant). Psychopathology also had statistically significant effects on pain affect, some of which was due to the mediating role of pain sensation, i.e., because individuals with higher levels of psychopathology report higher levels of pain sensation and because those with higher levels of pain sensation report higher levels of pain affect, it follows that those with higher levels of psychopathology also report higher levels of pain affect. However, the model in Fig. 1 suggests that the effect of psychopathology on pain sensation cannot completely account for the effects of psychopathology on pain affect. Rather, psychopathology seems to have an independent effect on pain affect irrespective of pain sensation. These effects of psychopathology on pain sensation and pain affect work their way through the theoretical system to ultimately produce effects on both pain suffering and self reported behavior. However, the key mechanisms to understanding the impact of psychopathology on pain suffering and behavior seems to be those relating psychopathology to pain sensation and pain affect. 3.3. Supplemental analyses In addition to the above model tests, we conducted supplementary analyses to explore potential problems of model misspecification and parameter bias induced by the presence of measurement error. With respect to the former, we used traditional regression methods in conjunction with product terms to test for possible interaction effects between predictors of each endogenous variable in Fig. 1 (Jaccard and Wan, 2003). The regression equations were dictated by the limited information estimation approach to SEM described by Bollen (1996) and did not suggest the presence of any meaningful interaction effects. For the final trimmed model, we were able to explore possible reciprocal causation because of the presence of one or more instrumental variables (Bollen, 1989). We did not observe any significant improvement in model fit as a result of adding reciprocal influences between any of the four

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

primary variables in the four sequence pain model. These results are consistent with the observations of Wade and Price (2000), but they do not definitively rule out the possibility of reciprocal influence and should be viewed with tentativeness. In terms of measurement error, we re-estimated the model in Fig. 1 but imposed an a priori determined amount of measurement error onto the observed measures using the strategy described by Joreskog and Sorbom (1997). The amount of unreliability imposed was based on the alpha coefficient for each scale (i.e., the proportion of random error due to measurement error was set to be 1 minus the alpha coefficient for the scale). None of the major conclusions drawn from the original significance tests were changed. It also is useful to provide perspectives on statistical power for the tests of the path coefficients so that one can better appreciate the possibility of a Type II error for statistically non-significant path coefficients. Power analyses for SEM models are complicated and often rest on assumptions that are impractical or not viable. We followed the practice recommended by Jaccard and Turrisi (1996) that provides a rough sense of statistical power by applying power analytic methods for OLS regression as applied to selected linear equations from the set of linear equations implied by the model in question. Given a sample size of 168 and a two tailed alpha level of 0.05, we evaluated the statistical power associated with a path coefficient that represents 5% explained variance over and above a set of five additional covariates. Based on the residuals in Fig. 1, we evaluated three scenarios where the initial set of covariates accounted for 10% of the variance, 20% of the variance, or 40% of the variance. The approximate statistical power in these three scenarios was 0.87, 0.89, and 0.97. For a path coefficient that represents 3% additional explained variance in the same scenarios, the approximate statistical power was 0.66, 0.72, and 0.84. Overall, the approximate power seems adequate for detecting paths that account for at least 5% of the variance of an outcome variable and in some cases, it also is adequate for coefficients that reflect only 3% unique explained variance.

4. Discussion 4.1. The sequential pain processing model This study evaluated models of the relationships between pain sensation, pain affect, long term suffering and behavioral functioning in relatively severely affected IBS patients using Price’s sequential pain processing model. SEM based methods were used to elaborate this model and elucidate psychological mechanisms of pain not observed by previous pain processing researchers with respect to IBS. Our results are consistent with the

215

foundations of the sequential model of pain processing. As predicted, we found statistically significant path coefficients from pain sensation to pain affect, from pain affect to suffering, and from long term suffering to self reported illness behavior. These data echo previous SEM analyses that found that the early stages beginning with pain sensation give rise to later stages that culminate in psychological and behavioral dysfunction. In some respects, our data provide stronger empirical support for a four-stage model relative to previous research. We found a statistically significant path coefficient linking suffering to pain-related behaviors that does not characterize the findings of Wade et al. (1996). This disparity may be due to differences between our assessment protocol and that used by Wade et al. Our data also differed from early pain processing research in important ways. To date, the pain processing research suggests that early stages beginning with pain sensation have a ‘‘domino effect’’ on subsequent stages such that pain sensation impacts on pain unpleasantness which, in turn, impacts long term suffering and subsequent self reported illness behaviors. Consistent with this, we found that: (1) the relationship between pain sensation and pain suffering was mediated by the immediate emotional unpleasantness of pain experience; (2) the effect of pain affect on illness behavior was mediated by suffering. That said, we also observed patterns of data suggesting that the order among stages is neither necessarily successive nor fixed. Instead, the data were consistent with a model whereby pain sensation exerts some direct influence on illness behaviors independent of immediate and long-term pain affect. This suggests that pain sensation can influence later pain stages either as part of a ‘‘domino effect’’ on subsequent stages or by ‘‘hop scotching’’ intermediate stages altogether and influencing illness behaviors directly. 4.2. Cognitive elaboration of pain stages Past researchers have argued that more distal pain stages are subjected to higher levels of cognitive processing based on findings that the magnitude of association between psychological processes and pain stages was greater for more distal components of the model. We found mixed support for this proposition. Our finding that a relationship between trait anxiety and self reported illness behaviors (physical role limitations) was mediated, in part, by the degree of long term suffering is consistent with Price’s view that later stages are characterized by elaborate cognitive processing. On the other hand, the influence of psychological factors did not seem disproportionably concentrated at more distal pain stages. The influence of psychosocial factors on pain stages extended from sensory and affective dimensions of pain to suffering to illness behaviors. The pattern of these data raise questions about whether the relative

216

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

association between psychological factors and pain stages is a reliable index of stage-specific cognitive processing. We also found evidence consistent with the idea that the proximal stage of pain sensation influenced later stages (both measures of illness behaviors) without mediation by affective stages. This latter finding questions Price’s notion that cognitive appraisals of the long-term implications of pain necessarily ‘‘constitute the link between sensory features of pain and emotional feelings and their expressions’’ (Price, 1999, p. 58). 4.3. Trait anxiety and psychopathology We did not find support for predictions derived from Gatchel’s model (1996) about the impact of either anxiety or psychopathology on pain processing stages. Whereas predictions based on Gatchel’s model suggested that psychopathology would impact the later stage of suffering, we found significant relationships between psychopathology and earlier pain stages. Similarly, whereas Gatchel’s model suggests relationships between trait anxiety and early pain stages, we did not find significant path coefficients. The impact of anxiety, however, did seem to manifest itself at later stages of pain processing. Specifically, the data suggest that anxiety has an indirect effect on physical role limitations (i.e., more chronically anxious patients exhibited greater disruption in role activities) and that this is mediated by pain suffering. This result is reminiscent of the work of previous researchers (Affleck et al., 1992; Harkins et al., 1989; Wade et al., 1992) who found a link between the anxiety-related personality dimension of neuroticism and later stages of pain processing. A notable finding in our research concerns the relationship between psychopathology and the early stages of pain processing. Laboratory and clinical studies supporting the sequential model have found that the influence of psychological factors on early pain processing is limited to their effect on pain unpleasantness (Price, 2000, p. 1769). Consistent with this conclusion were our findings suggesting that psychopathology had both a direct influence on pain affect and an indirect influence through the mediating influence of pain sensation. That said, other aspects of our data challenge the assumption that psychological factors do not typically influence the stage of pain sensory processing. Specifically, our analysis yielded a significant path coefficient for psychopathology to pain sensation, with individuals with higher levels of psychopathology reporting higher levels of pain sensation (holding gender and age constant). These findings suggest that the impact of psychological factors on early pain processing stages may not be limited to pain affect but may extend to pain sensation. If so, negative emotional states may not strictly be a by-product of pain-related sensation but influence sensory mechanisms of nociceptive processing

(i.e., sensory discrimination), in much the same way as emotion modulates sensory processing of other perceptual states (Schupp et al., 2003). 4.4. Contextual factors The model in Fig. 1 suggests that neither gender nor age have a direct effect on pain affect, suffering or illness behavior but rather their effects are mediated by pain sensation. When pain sensation was held constant, gender and age effects on these more distal variables decreased to non-significant levels of prediction. Taken together these findings suggest that the effect of age and gender on pain responding (pain affect and behavior) is mediated by sensory pain. Simply put, the effect of gender and age on both pain affect and illness is largely due to their influence on the sensory-discrimination properties of pain. 4.5. Limitations and caveats There are limitations to the study that must be kept in mind when drawing conclusions. The sequential pain model is temporal in nature but it was tested using crosssectional data that asked people about current and past states and behavior. Although the model is best tested with longitudinal, prospective data, it does predict a certain covariance pattern among the cross-sectional measures in the present study and hence, is amenable to evaluation using structural equation modeling. Because only 19% of our sample was male, our conclusions about gender must be considered tentative and await more definitive confirmation with a larger sample. Our data reflected a subset of more severely affected IBS patients and therefore may not necessarily generalize to a mildly affected population. Another limitation relates to our use of the SF-36 to measure pain-related behaviors. As a general health measure, the SF-36 may lack sufficient sensitivity to differentiate behavioral dysfunction due to pain from other symptoms (intestinal or extraintestinal) associated with IBS. That said, the SF-36 appears preferable to existing self report measures geared toward measuring overt pain behaviors (e.g., frequent position changing) more characteristic of muscoloskeletal pain patients than IBS patients. The study was designed as a test of the pain-processing model using a select number of individual difference variables associated with IBS. Previous tests of the sequential pain-processing model have variously investigated the effects of variables we did not explore (e.g., extraversion, expectancy, ethnic identity). Further understanding of the sequential pain processing model as it relates to IBS and other visceral pain disorders could be enriched by testing the effects of these and other factors (e.g., attention, race, psychiatric diagnosis, predominant bowel habit, severity of GI symptoms, etc.) on pain

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

stages. The basic structure of the SEM analyses was dictated by the a priori causal structure implied by the sequential pain processing model. It is, of course, possible that other causal structures whose predictions are confounded with those of the present model can account for the data as well. We suggest the model in Fig. 1 be approached with the scientifically conservative perspective that although the observed data are consistent with it, the data do not rule out the possibility of alternative causal models that could account for the data equally well. Confidence in the model will be increased when its fundamental patterns are replicated in additional studies. Caution also should be exercised in generalizing these conclusions to patients with non-IBS pain. There is increased awareness that visceral pain is not simply a variant of somatic pain but has distinct sensory properties and underlying physiological mechanisms (Cervero and Laird, 1999). Our data, coupled with that of Price et al. with somatic pain patients (Wade et al., 1996), raise the possibly that the complexion of pain sequencing models (e.g., staging, relative impact of psychosocial factors) may differ across pain conditions. It cannot be ruled out that the observed link between psychopathology and pain sensation may, given the bidirectional neural pathways between emotional centers in the brain and enteric nervous system, be specific to IBS patients. Despite these limitations, the present research suggests useful agendas for future work. For the clinician, our work suggests that the pain processing model offers a theoretically grounded, empirically validated framework by which data regarding different aspects of pain experience of IBS patients can be organized and integrated to clarify patient’s problems, develop a suitable case formulation, and use this information to plan a suitable treatment. Whether the pain processing model has sufficient case formulation value to help clinicians select treatments targeting the underlying mechanisms of a presenting problem (Haynes and Williams, 2003) is an interesting question. For the researcher, one future direction involves placing the four-stage model of pain processing into broader nomological networks that include individual difference, situational, contextual, and temporal based variables. Such networks permit us to better understand the mechanisms by which more distal variables impact behavior (or other psychologically meaningful outcomes) and elucidate the mediational roles that the core dimensions of pain processing have with respect to such effects. This research also provides more stringent evaluations of the sequential pain model as one evaluates whether the relationships of the pain dimensions to each other as well as the distal variables behave in accord with what one would theoretically expect. In testing the sequential model, a methodologically rigorous SEM approach may have some unique advantages over more traditional laboratory studies

217

whose conditions introduce an experimental artifact (e.g., experimenter’s obligation to assure subjects that pain stimulus is benign, time limited, and tolerable; Price, 1999) that artificially suppresses unpleasantness ratings and makes impossible (due to the absence of sustained follow up) the assessment of suffering component. Our results highlight the potential utility of SEM based research paradigms by raising questions about some of the basic tenants of the pain processing model and by mapping out plausible and interesting causal dynamics between the core pain processing model components and gender, psychopathology, anxiety and age. Future research can explore in greater detail the intriguing leads suggested by the present analyses.

Acknowledgements Preparation of this paper was supported in part by National Institutes of Health Grant DK-54211 and an Interdisciplinary Research and Creative Activities Fund grant through the University at Buffalo, SUNY. References Affleck G, Tennen H, Urrows S, Higgins P. Neuroticism and the painmood relation in rheumatoid arthritis: insights from a prospective daily study. J Consult Clin Psychol 1992;60:119–26. Arbuckle AJ, Wothke W. AMOS 4.0 user’s guide. Chicago: Small Waters Corp; 1999. Blanchard EB. Irritable bowel syndrome: psychosocial assessment and treatment. Washington: APA; 2001. Blanchard EB, Scharff L. Psychosocial aspects of assessment and treatment of irritable bowel syndrome in adults and recurrent abdominal pain in children. J Consult Clin Psychol 2002;70:725–38. Bollen KA. Structural equations with latent variables. New York: John Wiley & Sons; 1989. Bollen K. An alternative two-stage least squares (2SLS) estimator for latent variable equations. In: Marcoulides G, Schumacker R, editors. Advanced structural equation modeling: issues and techniques. Mahwah, New Jersey: Erlbaum; 1996. Bollen K, Stine RA. Bootstrapping goodness of fit measures in structural equation models. In: Bollen K, Long S, editors. Testing structural equation models. Newbury Park: Sage; 1993. p. 111–35. Bush FM, Harkins SW, Harrington WG, Price DD. Analysis of gender effects on pain perception and symptom presentation in temporomandibular pain. Pain 1993;53:73–80. Buytendyck FE. Pain. London: Hutchinson; 1961. Camilleri M, Choi M. Review article: irritable bowel syndrome. Aliment Pharm Therap 1997;11:3–15. Cassell E. The nature of suffering and the goals of medicine. New York: Oxford University Press; 1991. Cervero F, Laird JM. Visceral pain. Lancet 1999;353:2145. Chang L, Heitkemper MM. Gender differences in irritable bowel syndrome. Gastroenterology 2002;123:1686–701. Chapman CR, Gavrin J. Suffering: the contributions of persistent pain. Lancet 1999;353:2233–7. Chapman CR, Stillman M. Pathological pain. In: Krueger L, editor. Handbook of perception: pain and touch. New York: Academic Press; 1996. p. 315–40.

218

J.M. Lackner et al. / European Journal of Pain 9 (2005) 207–218

Cook IJ, van Eeden A, Collins SM. Patients with irritable bowel syndrome have greater pain tolerance than normal subjects. Gastroenterology 1987;93:727–33. Derogatis LR. Brief symptom inventory. Minneapolis, MN: National Computer Systems; 1993. Dersh J, Polatin P, Gatchel R. Chronic pain and psychopathology: research findings and theoretical considerations. Psychosom Med 2002;64:773–86. Drossman DA, Corazziari E, Talley NJ, Thompson GW, Whitehead WE. Rome II. The functional gastrointestinal disorders. Diagnosis, pathophysiology and treatment: a multinational consensus. McLean, VA: Degnon Associates; 2000. Drossman DA, Leserman J, Nachman G, Li Z, Gluck H, Toomey T, et al. Sexual and physical abuse in women with functional or organic gastrointestinal disorders. Ann Intern Med 1990;113:828– 33. Drossman DA, Toner BB, Whitehead WE, Diamant NE, Dalton CB, Duncan S, et al. Cognitive-behavioral therapy versus education and desipramine versus placebo for moderate to severe functional bowel disorders. Gastroenterology 2003;125:19–31. Drossman DA, Whitehead WE, Camilleri M. Irritable bowel syndrome: a technical review for practice guideline development. Gastroenterology 1997;112:2120–37. Dubreuil D, Kohn P. Reactivity and response to pain. Pers Indiv Differ 1980;7:907–9. Ellermeier W, Westphal W. Gender differences in pain ratings and pupil reactions. Pain 1995;61. Gatchel R. Psychological disorders and chronic pain: cause and effect relationships. In: Gatchel RJ, Turk DC, editors. Psychological approaches to pain management: a practitioner’s handbook. New York: Guilford Publications; 1996. p. 33–54. Gordon RS. Pain in the elderly. JAMA 1979;241:2191–2. Harkins SW, Price DD, Braith J. Effects of extraversion and neuroticism on experimental pain, clinical pain, and illness behavior. Pain 1989;36:209–18. Haynes S, Williams A. Case formulation and design of behavioral treatment programs: matching treatment mechanisms to causal variables for behavior problems. Europ J Psychol Assess 2003;19:164–74. Hoyle RH, Kenny DA. Sample size, reliability, and tests of statistical mediation. In: Hoyle R, editor. Statistical strategies for small sample research. Thousand Oaks: Sage; 1999. Jaccard J, Turrisi R. LISREL analyses of interaction effects in multiple regression. Newbury Park: Sage; 1996. Jaccard J, Wan C. LISREL analyses of interaction effects in in multiple regression. Newbury Park: Sage; 2003. Jensen MP, Karoly P, Harris P. Assessing the affective component of chronic pain: development of the pain discomfort scale. J Psychosom Res 1991:149–54. Joreskog K, Sorbom D. LISREL 8 user’s reference guide. Chicago: Scientific Software; 1997. Keselman HJ, Cribbie R, Holland B. The pairwise multiple comparison multiplicity problem: an alternative approach to familywise and comparisonwise Type 1 error control. Psychol Meth 1999;4:58–69. Lackner JM, Gurtman MB. Pain catastrophizing and interpersonal problems: a circumplex analysis of the communal coping model. Pain [in press]. Lee OY, Mayer EA, Schmulson M, Chang L, Naliboff B. Genderrelated differences in IBS symptoms. Am J Gastroenterol 2001;96:2184–93. Leon B. Pain perception and extraversion. Perceptual Motor Skills 1974;38:510.

Melzack R. The McGill pain questionnaire: major properties and scoring methods. Pain 1975;1:277–99. Melzack R. The short form McGill pain questionnaire. Pain 1987;30:191–7. Melzack R, Casey K. Sensory, motivational and central control determinants of pain: a new conceptual model. In: Kenshalo D, editor. The skin senses. Springfield: Charles C. Thomas; 1968. p. 423–39. Naliboff BD, Berman S, Chang L, Derbyshire SW, Suyenobu B, Vogt BA, et al. Sex-related differences in IBS patients: central processing of visceral stimuli. Gastroenterology 2003;124:1738–47. Price DD. Psychological and neural mechanisms of pain. New York: Raven Press; 1988. Price DD. Psychological mechanisms of pain and analgesia. IASP Press; 1999. Price DD. Psychological and neural mechanisms of the affective dimension of pain. Science 2000;288:1769–72. Price DD, Harkins SW, Baker C. Sensory-affective relationships among different types of clinical and experimental pain. Pain 1987;28:297–307. Riley J, Robinson M, Wade JB, Myers C, Price DD. Sex differences in negative emotional responses to chronic pain. J Pain 2001;2:354–9. Riley JL, Wade JB, Robinson ME, Price DD. The stages of pain processing across the adult lifespan. J Pain 2000;1:162–70. Schafer JL. Analysis of incomplete data multivariate data. London: Chapman & Hall; 1997. Schupp HA, Junghofer M, Wieike AI, Hamm AO. Emotional facilitation of sensory processing in the visual cortex. Psychol Sci 2003;14:7–13. Society for Women’s Health Research. Irritable bowel syndrome in women: the unmet needs. Washington, DC: Society for Women’s Health Research; 2002. Speilberger CD, Gorsuch RL, Lushene RE. Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologist Press; 1970. Talley NJ, Boyce PM, Jones M. Predictors of health care seeking for irritable bowel syndrome: a population based study. Gut 1997;41:394–8. Unruh AM. Gender variations in clinical pain experience. Pain 1996;65:123–67. Wade JB, Price DD. Nonpathological factors in chronic pain: implications for assessment and treatment. In: Gatchel RJ, Weisberg JN, editors. Personality characteristics of patients with pain. Washington, D.C.: American Psychological Association; 2000. Wade JB, Dougherty LM, Archer CR, Price DD. Assessing the stages of pain processing: a multivariate analytical approach. Pain 1996;68:157–67. Wade JB, Dougherty LM, Hart RP. A canonical correlation analysis of the influence of neuroticism and extroversion on chronic pain, suffering, and pain behavior. Pain 1992;51:67–74. Wade JB, Price DD, Hamer RM, Schwartz SM. An emotional component analysis of chronic pain. Pain 1990;40:303–10. Ware JE, Sherbourne CD. The MOS 36-item short form health survey (SF-36). Med Care 1992;30:473–83. Ware JE, Snow KK, Kosinski M. SF-36 health survey and interpretation guide. Lincoln, RI: QualityMetric Incoporated; 2000. Whitehead WE, Engel BT, Schuster MM. Perception of rectal distention is necessary to prevent fecal incontinence. Adv Physiol Sci 1990;17:203–9. Woodrow KM, Friedman GD, Siegelaub AB, Collen MF. Pain tolerance: differences according to age, sex and race. Psychosom Med 1972;34:548–56.