Psychosocial and Demographic Correlates of Employment versus Disability Status in a National Community Sample of Adults with Chronic Pain: Toward a Psychology of Pain Presenteeism

Abstract

Background

Although chronic pain is a source of work-related disability, relatively little research has addressed the psychological factors that differentiate individuals in chronic pain who leave the workforce from those who remain on the job despite their pain.

Objective

The present study examined a small set of attitudinal and coping-related factors as potential correlates of pain-related disability versus continued part- or full time employment over and above the role of well-known risk factors.

Methods

A large sample of adult men and women with chronic pain drawn from across the United States (N= 1293) by means of random digit dialing was subdivided into two groups: working (N = 859) and on disability (N = 434). Both groups were interviewed (by telephone) to complete a set of instruments (called the Profile of Chronic Pain: Extended Assessment [PCP: EA] Battery) measuring pain attitudes and coping methods.

Results

Logistic regression analysis revealed, as expected, that continued employment status was inversely related to pain severity and work status was positively related to higher education and being Hispanic. After controlling for severity and demographic factors, belief in a medical cure and catastrophizing tendencies were significant inverse predictors and task persistence was a positive predictor of continued employment.

Conclusions

Results revealed both demographic and attitudinal predictors of continued employment, and highlight the value of harnessing insights from the psychology of work engagement to better understand the processes underlying pain presenteeism. Interventions designed to keep persons with pain in the active work force should build upon and extend the present findings.

1. Introduction

Over the years, a consensus has emerged that persistent pain can create beliefs and attitudes that undermine personal autonomy as well as people’s sense of control and efficacy across a wide range of everyday activities including vocational pursuits [1,2]. For example, persons who are disabled by acute and chronic pain report more discomfort and distress relative to their employed peers [3], and extended absence due to musculoskeletal and other forms of pain constitutes a major physical and mental health risk as well as an economic drain [4–6].

Nonetheless, when considering the currently restricted employment opportunities in Europe and the United States, the durational and financial limits imposed by most disability insurance coverage, and the many socioeconomic disadvantages associated with chronic pain conditions, it is not surprising that persons with persistent pain often do not have the luxury of delaying or giving up their work lives [7,8]. Moreover, because contextual, psychological, and even neurobiological factors can serve to moderate the connection between chronic pain and vocational outcomes such as work engagement (job commitment), well-being, medical leave-taking, absenteeism, retirement, return-to-work patterns, and/or the quality of work, the task of predicting who will or will not be permanently or temporarily sidelined by pain has not been easy. We speculate, for example, that when they are at work, some employees with chronic pain engage in the active process of “job crafting” [9] that is, coordinating their varied job demands with their pain coping resources [10] as well as drawing strength from their continued ability to work [11] all in an effort to remain psychologically and physically present on the job.

We therefore sought to examine whether a set of psychological variables could differentiate working from non-working adults while controlling for known risks such as pain severity, a low education level, divorce, and aging. Participants, sampled from across the United States by means of random digit dialing procedures, were initially screened for the presence of chronic pain (via the Profile of Chronic Pain: Screen [12]) and subsequently interviewed using the sub-scales of the Profile of Chronic Pain: Extended Assessment (PCP: EA) battery [13]. Identifying attitudinal/belief, coping, or social response scales from the PCP: EA that reliably discriminate among a large segment of working and non-working adults, after controlling for pain severity as well as for demographic factors, promises useful insights into the nature of job-related pain management. Despite their correlational nature, findings from such an analysis enhances our nascent appreciation of the psychosocial factors linked to the critical but still incompletely understood process of pain presenteeism--the capacity to remain at work in a presumably productive manner despite the continued experience of pain.

2. Methods

2.1 Sample

The present study sample was a subsample of a larger national parent sample of individuals with chronic pain [13]. Table 1 provides descriptive information about the national parent sample and the subsample that was the focus of the present study. The subsample consisted of adults with chronic pain who were working full-time (n = 644), working part-time (n = 215), or who were on pain-related disability (n = 434). Because there were few differences between participants working part-time and full-time, we combined these two groups (n = 859).

Table 1.

Descriptive Information for the National Parent Sample and Study Subsample

	National Parent Sample	Subsample
Race/Ethnicity
White, Non-Hispanic	80.9%	77.5%
Black, Non-Hispanic	8.9%	11.1%
American Indian or Alaskan Native, Non-Hispanic	2.0%	2.5%
Asian	0.3%	0.5%
Native Hawaiian or Pacific Islander, Non-Hispanic	0.1%	0.1%
Hispanic	3.5%	4.3%
Mixed/Other	4.2%	4.0%
Educational Attainment
High school diploma or less	44.3%	41.6%
Some college	27.1%	28.2%
Associate’s or bachelor’s degree	19.7%	22.1%
Advanced degree	8.9%	8.1%
Marital Status
Married or living as married	60.2%	59.4%
Single	16.6%	20.3%
Divorced	13.3%	15.9%
Widowed	9.9%	4.5%
Currently in Treatment for Pain Problem
Yes	72.8	69.1
No	27.2	30.9

2.2. Procedure

2.2.1 Screening and Recruitment of the National Parent Sample

The study involving the national parent sample [13] from which the present subsample was drawn was approved by an Institutional Review Board and consent was obtained from all potential participants. Random digit dialing (RDD) sampling (conducted by a survey research company) was used to select households from across the United States. Computer assisted telephone interviewing was employed to screen respondents for a chronic pain problem. The RDD procedure involved the generation of a list of phone numbers, randomly drawn from across the United States to be representative of population distribution patterns. Non-working, cellular, and non-residential numbers were excluded from this original list. Thus, households that did not have a landline or who relied exclusively on a cell phone were excluded. Calls were made throughout the day and evening on all days of the week. A working number was called a minimum of six times. Calling patterns were regulated in keeping with the time zone being called, so as to neither call too early in the morning nor too late in the evening.

Within each household, the person with the most recent birthday was targeted for screening. Participants were screened if they spoke English and if they indicated that they experienced pain within the past six months. A total of 9759 participants were screened via the Profile of Chronic Screen (PCP: S; see “Measures” below for more details about this tool). The PCP: S consists of three scales--severity, interference, and emotional burden [12]. To be eligible to participate in the study, respondents were required to score at least one standard deviation above the National age/gender-based normative mean on one or more of the three scales of the PCP: S [12]. Using the above criteria, 3050 respondents (or 31% of the total 9759 screened) were eligible for participation. Of this group, 2407 were willing to complete the PCP: EA via telephone interview.¹ Approximately 400 participants with chronic pain were interviewed in each of the six cells formed by crossing gender with three age groups (25–44, 45–64, and 65–80).

2.2.2 The Study Sample

As noted above, in the present study we included only those who were working full- or part-time (n=859) or who reported being disabled due to their pain (n=434). As shown in Table 1, the subsample was quite similar to the parent sample.

2.3. Measures

2.3.1. Profile of Chronic Pain: Screen (PCP: S)

The 15-item PCP: S [12] assesses the respondent’s pain experience during the previous six months with regard to pain severity, pain-related interference, and the emotional burden of chronic pain. These three dimensions were strongly recommended by the Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials ([14] a panel of pain experts drawn from government, academia, a self-help organization, and the pharmaceutical industry) as core dimensions in pain assessment. The three-factor structure has been confirmed and replicated using multiple samples, the PCP: S possesses minimal levels of social desirability response bias (coefficients ranging from −.03 to −.22), acceptable reliability (retest reliabilities ranging from .77 to .85 and alpha coefficients ranging from .89 to .91), and evidence of validity [12]. National norms are available for the PCP: S by gender and three age groups. and were used in the screening process to identify those with a pain problem (see “Screening and Recruitment of the National Parent Sample” above) [12].

2.3.2. Profile of Chronic Pain: Extended Assessment (PCP: EA)

The PCP: EA [13] consists of 110 items. Items related to pain history and quality, pain diagnoses, and painful areas of the body were not included in the present analyses. Demographics and the thirteen scales of the PCP: EA [13] that relate to pain coping, attitudes and beliefs, and social responses to pain were included in the present analyses. The scales provide information on (1) pain coping (15 items; the Guarding, Ignoring, Positive Self-Talk, and Task Persistence scales), (2) pain attitudes and beliefs (18 items; Perceived Disability, Pain Control, Belief in a Medical Cure, and Pain-Induced Fear), (3) catastrophizing (4 items in the Catastrophizing scale), and (4) the positive and negative social responses of the “most important person” in the subject’s social network (15 items; the Emotional and Tangible Support scales and the Insensitivity and Impatience scales).

The factor structure of the PCP: EA has been confirmed and replicated via individual confirmatory factor analyses (CFAs) conducted on (1) the four pain attitude scales, (2) the four pain coping scales, (3) the catastrophizing scale, and (4) the social response scales [13]. All of the final confirmatory factor analyses yielded comparative fit indices (CFIs) >.95, root mean square errors of approximation (RMSEA) ranging from .000 to .060, standardized root mean square residuals (SRMR) ranging from .005 to .048, and significant factor loadings, suggesting excellent fit for all models. Acceptable reliability for the 13 scales of the PCP: EA has also been found, with an average coefficient alpha of .77 and a mean retest reliability of .74 [13]. Negligible social desirability response bias has been observed for the 13 scales, with a mean correlation between scales and social desirable responding of .13 [13]. Finally, evidence of validity has been reported in research with community samples, internet samples, primary care outpatients, and college students involving such topics as pain resilience, evaluation of the efficacy of an online pain management program, pain and sexual functioning, depression, pain and daily functioning, and racial differences in the pain experience [13,15–22].

3. Results

3.1. Overview of Statistical Analyses

First, evidence is presented to confirm that participants who report that they are disabled are, in fact, more debilitated by their pain than their working peers. Second, the bivariate relations are assessed between employment status (0 = on disability due to pain, 1 = working part-time or full-time) and (a) demographic factors and (b) the pain-related scales derived from the PCP: EA. Third, via preliminary logistic regression analyses, a subset of the pain-related variables is selected to serve as predictors in the main analysis. In the main analysis, we seek to predict employment status from pain severity, a set of demographic variables, and a set of PCP: EA-derived pain measures.

3.2. Confirming Expected Differences for Participants Indicating Disability versus Current Employment

Table 2 presents the means and standard deviations for severity, interference, emotional burden, and perceived disability for participants on disability and those who are working. Although all participants met our criteria for chronic pain, we expected participants who were on disability, relative to those who were working, to have higher scores on all four scales. Because violation of the assumption of homogeneous variances did not alter the statistical significance of any of our independent-sample t-tests, we report results under the assumption of equal variances.

Table 2.

Percentage, Means and Standard Deviation for Validity, Demographic, and Pain-Related Variables by Employment Status

	On Disability			Working			Cohen’s d
Variable	%	M	SD	%	M	SD
Validity
Severity		25.51	3.84		22.23	4.28	-.79
Interference		23.60	8.49		13.59	8.37	−1.19
Emotional burden		18.20	6.05		12.82	6.29	−.86
Perceived disability		15.06	4.36		7.22	4.90	−1.66
Demographics
Sex
Male	51			56
Female	49			44
Age
25–44	28			54
45–64	60			36
65–80	12			10
Ethnicity
White, non-Hispanic	75			79
Black, non-Hispanic	16			9
Hispanic	3			5
“Other”	6			7
Marital Status
Married or living together as married	49			64
Single	20			21
Divorced or widowed	31			15
Education
HS degree or less	57			34
Some college or AA degree	34			39
BA degree or higher	8			27
Pain-Related
Guarding		8.66	4.31		6.46	4.11	−.50
Ignoring		13.64	4.62		14.89	4.33	.27
Positive self-talk		15.01	4.80		16.38	4.50	.28
Task persistence		11.54	5.36		15.35	4.06	.79
Pain control		16.88	5.81		19.73	4.77	.52
Beliefs in a medical cure		17.31	6.77		11.14	7.47	−.80
Pain induced fear		11.87	7.77		6.15	5.89	−.82
Catastrophizing		10.90	5.77		6.34	4.88	−.83
Emotional support		12.43	3.50		12.07	3.57	ns
Tangible support		14.85	5.87		13.01	6.21	−.28
Insensitivity		6.21	5.52		4.85	5.31	−.24
Impatience		6.44	5.91		4.93	5.22	−.26

Relative to working participants, those on disability were significantly (p < .001) higher on severity, t (1291) = −13.49, interference, t (1291) = −20.20, emotional burden, t (1291) = −14.80, and perceived disability, t (1282) = −28.00. Using Cohen’s standardized mean differences (d) as an index of effect size, the magnitude of the effects ranged from −.79 (severity) to −1.66 (perceived disability; see Table 2).

3.3. Demographic Variables and Employment Status

Using χ² tests and Φ correlation coefficients as our index of effect size, we examined the bivariate relations between the demographic variables and employment status. Table 2 provides percentages separately for participants on disability and who were working. The association between gender and employment status was not significant (p =.08), χ² (1, 1293) = 2.30. In contrast, the associations between age, ethnicity, marital status, and education and employment status were all significant (highest p < .01). Compared to the working group, the group on disability was less likely to be young (28% versus 54%) and more likely to be middle-aged (60% versus 36%), χ² (2, 1293) = 79.37, Φ = −.25. Compared to the working group, the group on disability was more likely to be Black, Non-Hispanic (16% versus 9%), χ² (3, 1285) = 15.85, Φ = .11. Compared to the working group, the group on disability was less likely to be married or in a marriage-like relationship (49% versus 64%) and more likely to be divorced or widowed (31% versus 15%), χ² (2, 1291) = 47.12, Φ = .19. Finally, compared to the working group, the group on disability was less likely to have a BA or higher degree (27% versus 8%) and more likely to have at most a high school degree (57% versus 34%), χ² (2, 1289) = 82.21, Φ = .25.

3.4. Pain-Related Variables and Employment Status

Using independent sample t-tests and d as an index of effect size, we examined the bivariate relations between employment status and the pain-related variables. Table 2 presents the means and standard deviations separately for participants on disability and those who were working. Because violation of the assumption of homogeneous variances did not affect the statistical significance of any of our independent sample t-tests, we report results under the assumption of equal variances. With the exception of emotional support, all of the t-tests were significant (highest p< .05).

Relative to participants on disability, participants who were working had higher scores on (a) ignoring, t (1283) = 4.76, d =.27, (b) positive self-talk, t (1280) = 5.02, d = .28, (c) task-persistence, t (1278) = 14.14, d = .79, and pain control, t (1281) = 9.35, d = .52 . In contrast, relative to participants who were working, participants on disability had higher scores on (a) guarding, t (1281) = −8.90, d = −.50, (b) belief in medical cure, t (1282) = −14.39, d = −.80, (c) pain-induced fear, t (1278) = −14.68, d = −.82, (d) catastrophizing, t (1283) = −14.86, d = −.83, (e) tangible support, t (1232) = −5.00, d = −.28, (f) insensitivity, t (1237) = −4.19, d = −.24, and (g) impatience, t (1232) = −4.57, d = −.26.

3.5. Selection of Pain-Related Variables to Use as Predictors of Employment Status

Several of the pain-related variables were moderately correlated with each other. To reduce the potential bias in parameter estimation that can occur when multiple correlated predictors are included in a model, we carried out an initial analysis designed to reduce the set of pain-related predictors. Emotional support was excluded from this analysis because the bivariate analysis revealed a non-significant relation between this variable and employment status. The other 11 pain-related variables along with pain severity scores were entered in to a preliminary logistic regression model in which the dependent variable was employment status (0 = on disability, 1 = working part-time or full-time). This model revealed that, controlling for severity, three pain-related variables were significant (highest p< .008) predictors of employment status--belief in a medical cure, catastrophizing, and task persistence.

3.6. Main Analysis

In the main analysis (involving the selected subset of variables as described above), participants working part-time or full-time were coded 1 and participants on disability were coded 0. In carrying out the logistic regression analysis, a block-entry technique was used. In block 0, only the intercept was included in the model. This model is also called the null or baseline model. In block 1, pain severity scores were added to the model. In block 2, the demographic variables were entered into the model (sex, age, ethnicity, marital status, and education). In block 3, the pain-related variables (belief in medical cure, catastrophizing, and task persistence) were entered into the model.

All of the demographic characteristics were coded as categorical variables and a reference group was specified for each variable. The categories and reference groups are as follows: age: 25–44 (reference group), 45–64, 65–80; race/ethnicity: White Non-Hispanic (reference), Black Non-Hispanic, Hispanic, Other; marital status: married or living in a marriage-like relationship (reference), single, divorced or widowed; education: high school degree or less (reference), some college, bachelor’s or higher degree; sex: the reference group was male.

Table 3 presents summary statistics on the utility of the prediction model after variables have been added in each block. The Hosmer and Lemeshow (H & L) test was employed to provide information on the goodness of fit of the set of predictors in each block. With severity in the model, the H & L test was significant, χ² (8, 1252) = 15.77, p < .05, indicating that the fit of the model to the data was inadequate. When the demographic variables were added in block 2, the H & L test was no longer significant, χ² (8, 1252) = 6.02, p = .64. Similarly, the H & L test was not significant in the final model after the inclusion of the pain-related variables, χ² (8, 1252) = 9.25, p = .32. Thus, the fit of the model to the data was deemed adequate in blocks 2 and 3.

Table 3.

Summary Statistics for Logistic Regression Model Predicting Employment Status

Statistic	Block
	0	1	2	3
H & L χ2 test		15.77x	6.02	9.25
−2LL	1583.47	1406.41	1261.75	1049.96
χ2Model test		177.06 xxx	321.73xxx	533.51xxx
χ2Block change test		177.06 xxx	144.66xxx	211.79xxx
Nagelkerke
Pseudo R2		.18	.32	.48
Overall hit rate	.67	.73	.75	.80
Sensitivity rate	.00	.38	.49	.63
Specificity rate	1.00	.90	.88	.88
False + rate	NA	.35	.34	.28
False − rate	.33	.25	.22	.17

^xp<.05,

^xxxp<.001

At each block, the log is generated of the probability that the observed values of the dependent variables may be predicted from the observed values of the predictors. When this statistic is converted to the −2 log likelihood (−2LL) statistic, it can be used to compare the block 0 model with subsequent models. Also, it can also be used to compare each successive model with the preceding model. From block 0 to block 3, the −2LL statistic decreased from 1583.47 to 1049.96. As can be seen in the χ² model test row, compared to the null model, each block contributes significantly (p< .001) to the prediction of employment status. As can be seen in the χ² model test row and in the block change test row of Table 3, each successive block contributed significantly (p< .001) to the prediction of employment status, above and beyond the preceding block.

The overall hit rate refers to the probability that a participant’s predicted employment status corresponds to his or her actual employment status. At each successive block, the overall hit rate increased and it reached 80% for the final model. In the context of the present study, sensitivity refers to the probability that a participant on disability was correctly classified. From block 0 to block 3, the sensitivity rate increased from zero to 63 percent. Specificity, in contrast, refers to the probability that a participant who was working was correctly classified. From block 0 to block 3, the specificity rate declined from 100 percent to 88 percent. In the current study, the false positive rate refers to the probability that the prediction of being on disability was incorrect whereas the false negative rate refers to the probability that the prediction of working was incorrect. From block 1 to block 3, the false positive rate decreased from 35 percent to 28 percent. Furthermore, from block 0 to block 3, the false negative rate decreased from 33 percent to 17 percent.

Table 4 provides information regarding the contribution of the individual predictors.

Table 4.

Logistic Regression Coefficients, Odds Ratios, and 95 Percent Confident Intervals for Predictors of Employment Status

Predictor	Block
		1			2			3
	b	eb	CI	b	eb	CI	b	eb	CI
Severity	−.22xxx	.80	.78–.84	−.19xxx	.82	.79–.86	−.13xxx	.87	.84–.91
Female				.26	1.29	.98–1.72	.30	1.34	.98–1.84
Some college				.54xxx	1.71	1.27– 2.30	.37x	1.45	1.04–2.02
BA degree or higher				1.46xxx	4.30	2.79– 6.61	1.37xxx	3.94	2.44–6.38
Single				−.46x	.63	.44–.90	−.47x	.63	.42–.93
Divorced or widowed				−.82xxx	.44	.31–.62	−.77xxx	.46	.32–.68
45–64 years old				−1.07xxx	.34	.26–.46	−1.09xxx	.34	.24–.47
65–80 years old				−.56x	.57	.36–.92	−.77xx	.46	.27–.78
Black, non- Hispanic				−.59xx	.56	.36–.85	−.18	.83	.52–1.34
Hispanic				.52	1.68	.82–3.44	.84*	2.32	1.07–5.05
“Other”				−.03	.97	.56–1.69	.06	1.07	.57–1.98
Belief in a medical cure							−.07xxx	.93	.91–.95
Task Persist- ence							.14xxx	1.15	1.11–1.19
Catastro- phizing							−.08xxx	.92	.90–.95

Note. Employment status: On disability = 0; 1 = Working

^*p< .05,

^**p< .01,

^***p< .001

In the final model, 11 of the 14 variables were significant (p < .05) predictors of employment status. The log odds of working (rather than being on disability) was higher among participants who had some college (b = .37) and who had a BA degree or higher (b = 1.37) relative to participants who had a high school degree or less education. In addition, relative to white Non-Hispanic participants, Hispanic individuals (b= .84) were more likely to be working than on disability. Furthermore, as task persistence scores increased (b = .14), the log odds that participants were working increased.

The remaining significant predictors exhibited inverse associations with the log odds of working rather than being on disability. As (a) severity (b = −.13), (b) belief in a medical cure (b = −.07), and (c) catastrophizing (b = −.08) increased, the log odds that participants were working decreased. The log odds of working was lower among participants who were single (b = −.47) and who were divorced or widowed (b = −.77) relative to participants who were married or living in a marriage-like relationship. In addition, relative to adults who were 25–44 years old, participants who were 45–64 years old (b= −1.09) or who were 65–80 years old (b= −.77) were less likely to be working than on disability.

To probe a potenial suppressor effect in which the inclusion of the block 3 predictors amplified the magnitude of the relation between employment status and Hispanic versus White, Non-Hispanic, six additional analyses were undertaken. In separate models, belief in a medical cure, catastrophizing, and task persistence were entered singly into the model in block 3. Furthermore in separate models, each pair of these three variables was entered into the model in block 3. The comparison of Hispanic versus White, Non-Hispanic participants was significant in only one of these six models. When belief in a medical cure and task persistence were entered in block 3, Hispanics were 2.51 times more likely than White, Non-Hispanics to be working as opposed to being on disability (p =.02).

Next, the two ethnic groups were compared on belief in a medical cure and task persistence. Although not statistically significant, we found that (a) Hispanics (M = 14.04, SD = 8.24) had higher belief in a medical cure scores than White, Non-Hispanics (M = 12.94, SD = 7.71), t (1024) = 1.02, p = .31; and (b) Hispanics (M = 13.56, SD = 5.68) had lower task persistence scores than White, Non-Hispanics (M = 14.39, SD = 4.60), t (1024) = −1.27, p = .20. Thus, when the modest disadvantages that Hispanics have relative to White, Non-Hispanics with respect to their higher belief in a medical cure and their lower task persistence scores are eliminated by statistically equating the two ethnic groups on these variables, the magnitude of the association between being Hispanic and working is enhanced.

4. Discussion

Viewing chronic pain in work settings in a manner consistent with contemporary person-centered motivational models of both pain and work [10,23–27], the current study sought to discover whether a small set of self-reported psychological variables could distinguish among two broad classifications: remaining in the work force despite experiencing chronic pain (so-called pain presenteeism) versus being on disability, over and above the effects of the usual risk factor suspects---pain severity and demographics.

Pursuing these questions in stages, we discovered several noteworthy patterns. First, and as expected on the basis of existing research, participants reporting that they were currently receiving disability indicated higher levels of pain severity, interference, and emotional burden relative to the working group. Next, employment status, i.e., disability versus working, was significantly associated with age (the disabled group tended to be middle-aged rather than young), ethnicity (the disabled group was more likely to be African-American rather than Non-Hispanic White or Hispanic), marital status (the disabled group was less likely to be married and more likely to be divorced relative to the working group) and education (the disabled group was more likely to have a high school education relative to the working group which was more likely to possess a college degree).

An unanticipated finding was that Hispanic ethnicity was a significant predictor of working status. This pattern, observed only in the multivariate analysis, has rarely been investigated. Over the years, comparisons between Caucasian (Non-Hispanic Whites) and African-American persons with pain have been the most common, with the data suggesting lower pain threshold levels among persons in the latter group [28]. Consistent with this finding, in our bivariate analysis being African-American was associated with an increased likelihood of being disabled due to pain. Others have suggested that racial and ethnic differences in pain and pain adjustment tend to be small when the groups are matched on relevant confounding factors such as SES [29]. Admittedly, the present set of comparisons tends to conflate race and ethnicity. That is, our Hispanic ethnic sample may well contain individuals who identify as White or Black in addition to their cultural identification as Hispanic. Nonetheless, our findings suggest that race and ethnic differences in working versus being disabled among adults with chronic pain merit further study.

More importantly, as depicted in Table 2, a number of psychosocial dimensions as assessed on the Chronic Pain Screen: Extended Assessment (EA) battery [13] also emerged as differentiating the two employment statuses. Individuals who remain at work relative to persons on disability reported an adaptive pattern (e.g., ignoring pain, more positive self-talk, higher levels of task persistence, less guarding, less pain-induced fear, less catastrophizing, and the like) suggestive of greater self-directedness and emotional control. Even after accounting for the interrelatedness among our multiple predictors and the role of pain severity (by means of a logistic regression), three EA measures, namely, belief in a medical cure, catastrophizing, and task persistence, remained as significant predictors of work status along with such demographic factors as having some college or a bachelor’s degree, being Hispanic, being married, and being in our youngest age group (25 to 44 years of age). Although neither age nor ethnicity are viable change targets, cognitive-behavioral interventions are available to facilitate educational attainment skills, marital satisfaction, and, most notably, pain-related attitudes and values [30] in an effort to promote vocational persistence. Thus, the findings may help to establish an interventive agenda for persons with chronic pain who wish to continue working. It is also worth pointing out that a recent study [31], de Vries and colleagues compared workers with musculoskeletal pain to those on sick leave across a number of psychological characteristics and found those on sick leave to be higher in pain catastrophizing and fear avoidance and lower in self-efficacy and pain acceptance. Our findings, along with those of de Vries and associates[31], serve to strengthen the possibility of eventually constructing a psychological profile for predicting who will and will not remain at work with chronic pain.

4.1. Toward a Psychology of Pain Presenteeism versus Pain Absenteeism

As humans are said to have evolved complex reactions to stress [32], and as pain can be viewed as a specialized, neurally mediated, and experientially shaped stress modulation system [33,34], we believe that investigators are justified in searching out distinct pain management phenotypes that reflect context-sensitive or socially situated modes of response calibration, resource allocation, and emotion regulation. We submit that pain absenteeism and pain presenteeism may represent two broad phenotypic patterns that, with continued research, may be further differentiated into, say, reluctant versus enthusiastic subtypes [35].

Recently, job crafting theorists have suggested that workers can learn to identify the discretionary aspects of their typically non-discretionary work life, and thereby enhance their ability to allocate their limited resources to fluctuating contextual demands [9]. However, with chronic pain serving as a resource drain, the task of job crafting would appear to require additional strategic elements. Although the current research provides no information directly relevant to the real time strategic allocation of vocational resources among workers with pain, we can nonetheless speculate that factors such as task persistence, reduced levels of catastrophizing, and lower levels of belief in a medical cure might contribute to the ability of some working adults to remain “present” on the job despite their pain. Of course, the present research is not without potential weaknesses. Most obviously the cross-sectional design does not permit causal inferences. Additionally, although our use of a national sample obtained via random digit dialing affords a noteworthy methodological advantage in terms of generalizability, it nonetheless means that we could not obtain medical records to confirm pain diagnoses or ascertain other relevant medical conditions. Neither could we verify current and past employment status. Because we can conceive of no incentives for participant misrepresentation, we view these omissions as relatively minor. Finally, the exclusion of cell phone numbers in our sampling frame poses a potential for bias due to undercoverage. This is a real and expanding threat to survey research conducted only on landline telephones. Although the current estimate is that 20% of households do not have a landline, at the time of data collection, an estimated 7% of households relied exclusively on a cellphone [36]. Thus, while generalization is limited to those individuals with chronic pain who live in a home with a landline, the coverage extended to approximately 93% of households in our original sampling frame.

Nonetheless, despite its limitations, the present study has yielded novel findings pertaining to the demographic and psychosocial correlates of pain presenteeism versus work disability. To more fully address the means of overcoming the varied personal and economic losses brought about by chronic pain, clinical researchers are enjoined to incorporate the conceptual and empirical contributions emerging from the domain of applied occupational health [9,10,23,25,27,35]. Although remaining at work when physically sick (the usual meaning of “presenteeism”) is a public health threat, numerous advantages accrue to persons with pain who can manage to avoid premature job turnover by rescripting their attitudes, reallocating their resources, and acquiring the requisite self-regulatory skills for vocational persistence.