Running head: ENGENDERING PAIN MANAGEMENT PRACTICES Engendering pain management practices: The role of physician sex on chronic low-back pain assessment and treatment prescriptions

Publisher's copyright statement: This is the peer reviewed version of the following article: Bernardes, S. F., Costa, M. & Carvalho, H. (2013). Engendering pain management practices: the role of physician sex on chronic low-back pain assessment and treatment prescriptions. Journal of Pain. 14 (9), 931-940, which has been published in final form at https://dx.doi.org/10.1016/j.jpain.2013.03.004. This article may be used for noncommercial purposes in accordance with the Publisher's Terms and Conditions for self-archiving.


Introduction
There are sex-related differences in communication styles and primary-care preventive practices. As compared to males, female physicians provide more preventive services 5,20,21,33 , and are more patient-centred, i.e., more collaborative and emotionally responsive, gather more psychosocial information, engage in more psychosocial counseling and spend more time with patients. 10,33,45,46,47 . However, the effect of physician sex on dimensions of medical care such as treatment prescriptions and referrals has been less explored 7,11,12,49 , especially in pain management contexts. This paper contributes to bridging this gap.
Studies looking for sex-related differences in pain management practices (PMP) are scarce. Although some authors have not found significant differences in analgesic administration practices 43 , some have shown that female/male doctors prescribe more analgesics or opioids to female/male patients, respectively. This pattern emerges both in vignettes studies depicting chronic low-back pain patients (CLBP) 53,54 and in a prospective observational study of physicians' PMP in the ER. 48 . Therefore, like for primary-care preventive practices 10,12,21,49 , patient sex seems to have a different impact on male and female physicians' PMP. However, because patient sex is not the only cue that may influence physicians' clinical judgments 52 , it would be important to explore whether other relevant contextual cues (i.e., variables pertaining to the patient or the clinical situation) could have different impacts on male and female physicians' PMP.
A recent literature review 52 shows that (chronic) pain is often under-estimated and treated in the absence of diagnostic evidence of pathology (EP) and also when patients show distressed pain behaviors. Several authors have hypothesized that the impact of such cues on pain assessment/treatment may be accounted for by pain being psychologized and/or judged as less credible/legitimate. 22,26,51 . However, whether such This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 5 cues are equally weighed by male and female physicians in their PMP or whether such effects are accounted for by the same pain judgment processes (i.e., how pain is perceived as credible or attributed to psychological causes), to the best of our knowledge, has never been explored. This study aimed to explore whether physician sex moderates: (1) the effects of patients' pain behaviors and EP on prescriptions and referrals for CLBP, and (2) the mediating role of pain credibility judgments and psychological attributions on these effects.
Because a more patient-centred physician would place less emphasis on the visibility and objectivity of patients' signs and symptoms and tend to perceive patients as more unique individuals 36 , and also, because female physicians, probably as a consequence of gender socialization processes 37,39 , are more often patient-centred 10 41 . Two sensory (''sharp and cutting'') and two affective (''fearful and cruel'') descriptors were selected. The affective pain descriptors were aimed at conveying more emotional distress than the sensory pain descriptors.
As in our previous studies 8,9 several independent doctors and nurses checked for the credibility, realism and rigor of the scenarios. All the scenarios were perceived as simple and easy to read. It should also be noted that a CLBP scenario was chosen for two main reasons: (1) it is one of the most pervasive worldwide chronic pain conditions 16 ; and (2) it is equally prevalent among males and females, allowing us to build a more gender-neutral scenario. 31,32 .

Dependent variables
The dependent variables aimed to measure GPs' treatment and referral decisions.
Based on previous studies on pain assessment and treatment decisions 15,24,35 , and with the help of independent GPs, we adapted a set of items to operationalize the following dimensions: were rated on an evaluative scale from 1 (not at all) to 7 (extremely), where the anchors were adapted to the item's content (e.g., extremely genuine).
In order to check the factor structure of the measure in our sample, a principal axis factor analysis (oblique rotation) was undertaken (Table 1). Two factors were extracted with Eigenvalues above 1: 1) Psychological attributions to pain and 2) Pain credibility. Both factors showed good internal reliability (Table 1) and a low negative correlation (r = -.33, p <.001). Table 1 Principal factor analysis of pain judgment items (oblique rotation): factor loadings and consistency indices. This study was approved by the Institutional Review Board. The procedure was the same as the one we have used in our previous studies. 8,9 . Participants' were invited to collaborate on a study on memory and decision making processes in clinical contexts.

Items
They were told that, firstly, the study aimed at understanding to what extent the ability to recall clinical information was influenced by its presentation format, i.e., on a videotaped, audio taped or written format. Despite being told that they had been randomly assigned to the latter condition, all clinical scenarios were presented in a written format. All participants were told that, on a second part, the study aim was to analyze the influence of the recalled information on health-care professionals' attitudes towards a patient/clinical situation. After participants verbally consented to collaborate, they were randomly presented one of the eight written scenarios. They were given a maximum of two minutes to carefully read the information and form an impression of the pain patient, and told they could not refer back to the scenario after the two minutes were over. Afterwards, they were asked to recall several details of the clinical scenario (e.g., patient's symptoms, sex, age, pain duration, patient's emotional state, presence of evidence of pathology), most items were included to check the manipulations of the independent variables. Then, they were asked to judge the patient's pain, and to rate the likelihood of prescribing non-pharmacological treatment, to choose the most suitable pharmacological treatment and, finally, to rate the likelihood of referrals. Finally, sociodemographic information was collected, along with participants' personal and vicarious experience with chronic pain. Questionnaires were individually administered and took an average of 10 minutes to complete. Finally, all participants were debriefed.

Data analyses
This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 12 We started by conducting several analyses (Pearson correlations and t-tests) in order to check whether there were any significant effects of the socio-demographic and pain-related variables on GPs' pain judgments and treatment prescriptions and referrals.
No significant effects were found. Therefore, these variables were not included in further analyses.
Next, in order to check whether, controlling for patient sex, GP sex moderated the effects of EP and patient's pain behaviors on treatment prescriptions and referrals, univariate analyses of variance were conducted, over each one of the seven dependent variables, with the following between-subjects factors: 2 (GP sex) x 2 (EP) x 2 (Patient's pain behaviors) x 2 (Patient sex). Because patient sex did not show any significant interactions with any of the independent variables, and only showed a very small main effect on referrals to psychology/psychiatry (see Table 2), patient sex was excluded from the following analyses.
Finally, to explore the mediating role of pain judgments on the abovementioned effects, moderated mediation models were tested following the procedures proposed by  Tables 3, 4  It should be noted that due to the considerable amount of analyses, in order to prevent an inflated type I error, we started by considering a Bonferroni correction that would reduce our critical value to p ≤ .001. However, the Bonferroni correction is often criticized by being overly conservative 40 , controlling for type I error often at the expense of increasing type II error. Therefore, in order to find a balance between both types of error, we decided to reduce our critical value to p ≤ .01.

Manipulation checks
About 86% of the participants correctly recalled all the information presented in the scenarios. However, 12 men and 30 women, equally distributed across experimental conditions, failed to recall at least one piece of information (e.g., EP, patient's pain behaviors or age). These participants were not significantly different from the rest of the sample in terms of their socio-demographic characteristics and professional or vicarious experience with chronic pain. Therefore, they were excluded from the following analyses.

Mean differences in treatment prescriptions and referrals
Firstly, we aimed to explore whether physician sex moderates the effects of the patient's pain behaviors and EP on prescriptions and referrals for CLBP. In order to do this, a set of univariate analyses of variance were conducted (see Data Analyses section).
As can it be seen in Our second aim was to explore whether physician sex moderates the mediating role of pain judgments on the effects of the independent variables on prescriptions and referrals for CLBP. In order to do that, we conducted a set of moderated mediation analyses (see Data Analyses section).

Predictors of pharmacological treatment prescriptions (PTP)
Linear regression results for moderated mediation effects on PTP are shown in Table 3. As expected following the previously reported analyses of variance (Table 2)   Effects of evidence of pathology on pharmacological treatment prescriptions mediated by pain credibility judgments and moderated by the GP sex.
Note: Female GP = exterior b values; b values in parentheses correspond to the overall effect of EP on PTP and the b values that immediately follow correspond to the direct effect, controlling for pain credibility; * p <.05; ** p < .01; *** p<.001;

Predictors of non-pharmacological treatment prescriptions (NPTP)
As for NPTP, Table 4 Figure 2 shows the decomposition of the estimated parameters for male and female GPs separately. For female GPs, the effect of EP on NPTP was partially mediated by psychological attributions (Sobel Z = 2.91, p <.01) to pain, i.e., without EP, pain was more attributed to psychological causes and, hence, NPTPs were more common. However, for male GPs, both EP and psychological attributions had no significant effects on NPTP. Table 4 Regression analyses for moderated mediation effects on non-pharmacological treatment prescriptions (NPTP): EP as a predictor. Effects of evidence of pathology on non-pharmacological treatment prescriptions mediated by psychological attributions to pain and moderated by the GP sex.
Note: Female GP = exterior b values; b values in parentheses correspond to the overall effect of EP on NPTP and the b values that immediately follow correspond to the direct effect, controlling for psychological attributions; * p <.05; ** p < .01; *** p<.001 Finally, Table 5 shows an effect of pain behaviors on psychological attributions  35,51 . That the effect of psychological attributions was qualified by the GP sex (b = -.65, SE = .15), suggested that the suppression effect was being moderated by this latter variable. Figure 3 shows the decomposition of the estimated parameters for male and female GPs separately. For male GPs, when the patient showed signs of distress, his/her pain was clearly more attributed to psychological causes but, as opposed to female GPs' reports, such attributions had no significant effects on NPTP.
Like male GPs, female GPs made more attributions to psychological causes when the patient showed signs of distress. However, equation 3 (Table 5) shows that pain behaviors had a positive indirect effect (1.3 x .64 = .83), through psychological attributions, but a negative direct effect on NPTP (-.88). Because the direct and indirect effects have a similar size but opposite signs, the total effect of pain behaviors on females' NPTP was suppressed. This means that the more a female GP interpreted the patients' distressed pain behaviors as signs of pain determined by psychological factors the more she was likely to prescribe NPT, but the less she shared such interpretation the less likely she was to prescribe NPT. In other words, the negative impact of distressed pain behaviors on NPTP only happened when such behaviors were not followed by psychological attributions to pain. This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 23 Table 5 Regression analyses for moderated mediation effects on non-pharmacological treatment prescriptions (NPTP): Pain behaviors as predictor.  Our findings have only very partially supported our first hypothesis, given that GP sex only moderated the effect of EP on referrals to psychology/psychiatry; as hypothesized, EP had a larger effect on male than female GPs' referrals. As expected, male GPs seemed to place a higher emphasis on the visibility/objectivity of patients' symptoms 36 making such referrals, i.e., they were more likely to refer the patient to a psychologist/psychiatrist in the absence of objective, visible EP than in its presence.
This may reflect a more pronounced Cartesian thinking, where the presence/absence of EP was interpreted as meaning pain of organic/psychological causes, respectively. Such reasoning is often inadequate, considering that EP is far from being a good criterion to infer a person's low-back pain severity, credibility or cause. 27,30. The fact that this effect was only found for referrals to psychology/psychiatry might be partially accounted for by the stronger gender connotations of the latter (as compared to other referrals and treatment decisions), i.e., stronger association with the female stereotype. 23,28 . In fact, other authors have suggested that the effects of GP sex on medical practices might be greater in clinical situations that are more strongly associated with gender stereotypes. 25 . This may eventually be explained by the fact that, by activating GPs' gender schemas (i.e., cognitive structures that encompass learned knowledge about the meanings of being a man or a woman) 7  Still, this argument does not explain why GP sex moderated the effect of EP but not of pain behaviors on psychology/psychiatry referrals. This may possibly be accounted for by the fact that, as compared to EP, distressed pain behaviors are a less ambiguous and more consensual cue when it comes to referrals to psychology/psychiatry, which would suppress the likelihood of a gender schematic processing. 13 GP sex as a moderator of the mediating role of pain judgments on the effects of contextual cues on PMP.
Our findings also partially supported our second hypothesis. It was predicted that the effects of EP and distressed pain behaviors on pain treatment and referral decisions would be mediated by pain credibility judgments and/or psychological attributions, but these effects would be stronger among male physicians. Such results were found for the effects of EP on PTP. While EP generally showed a large effect on PTP, it was mainly among male GPs that part of this effect was accounted for by pain credibility judgments. In fact, it was mostly among male GPs that EP showed a significant effect on pain credibility judgements: in the absence of EP pain was perceived as less credible.
Also, male GPs' pain credibility judgments had a larger impact on their PTP. Such results may suggest that male GPs are more likely to adopt the role of gate-keepers of PT, characterized by a more suspicious, less empathic and, as predicted and supported by former evidence, less patient-centred style of interaction. 10,33,45,46,47 . Another possible interpretation is that this result may also reflect the adoption of a more stringent biomedical approach to pain. This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004

27
On the other hand, psychological attributions to pain were only significantly related to female GPs' NPTPs. Although both male and female GPs attributed pain to psychological causes more in the absence of EP or in the presence of distress cues, such attributions only showed a significant effect on female GPs' NPTPs: the more pain was attributed to psychological causes the more likely they were to prescribe NPT. Also, it was mainly among female GPs that psychological attributions to pain accounted for the effects of EP and pain behaviors on NPTP. First, psychological attributions partially mediated the effect of EP on NPTP: the absence of EP lead to more psychological attributions to pain, which in turn lead to a higher likelihood of NPTP. Second, psychological attributions entirely suppressed the effect of distressed pain behaviors on NPTP; in the presence of distressed pain behaviors, the more pain that was attributed to psychological causes the more likely was the prescription of NPT. However, if patients' distress was not attributed to psychological causes, female GPs were less likely to prescribe NPT. Such results corroborate former evidence suggesting that female doctors are more likely to take into account psychosocial factors when diagnosing or prescribing treatment. 10,33,45,46,47 . In this particular case, it was mainly among female doctors that psychological attributions played a significant role in predicting NPTP, which may be logical considering that walking, massage or hydrotherapy may be effective in diffusing distress, which often heightens pain experiences. 2,3 . It should be noted, however, that this effect was only found for NPTP, not for PTP. In fact, again similar to former studies 52 , our data showed that both male and female GPs are less willing to prescribe pain medications when pain is attributed to psychological causes.
Taken together, these results suggest that female doctors' NPTP may be compensating for their unwillingness to prescribe stronger PT when pain is psychologized.
Notwithstanding, it should be stressed that it was mainly among female GPs that such This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 28 psychological factors showed a significant predictive role of NPTPs. This could mean that female GPs may be more willing to refer patients to multidisciplinary treatment programs, where biological and psychosocial dimensions of pain experiences are taken into account.
Limitations, future directions for research and implications Some methodological limitations may be pointed out to this study, primarily regarding its ecological validity. First, because written vignettes are limited representations of real clinical scenarios, the generalization of our findings should be tentative. This is particularly true for results pertaining to the effects of pain behaviors.
In fact, EP and patient sex are easy to accurately operationalize to represent their theoretical constructs, but this is not so for pain behaviors, given their richness and complexity. Moreover, because the operationalization of distressed pain behaviors represented extreme situations (total absence of distress cues vs. presence of several distress cues), the generalization of our results to less extreme (and more common) scenarios should be made with reservations.
Second, because the NPTP variable did not specify the many and distinct treatments that it could encompass (e.g., walking, massage, acupuncture), it is difficult to say if the same pattern of results would be found for the separate prescription of these different NPT. The same could be said regarding the PTP variable, namely, the prescription of the distinct drugs (e.g., non-opioid analgesics, non-steroidal antiinflammatory, weak and strong opioids).
Third, our results have partially supported our hypotheses in a CLBP scenario but it is not possible to predict whether similar results would be found in other pain-This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 29 related clinical scenarios (e.g., in a more gendered pain situation like migraines or fibromyalgia).
Fourth, it is difficult to understand if our small effect sizes reflect the real effect sizes in clinical situations or are a consequence of the use of vignettes. The fact that we requested that participants systematically process the vignettes' information may have reduced the activation of participants' gender schemas 13 , accounting for the small effect sizes.
Fifth, although several cross-cultural studies suggest the contents of gender representations are shared by many western societies 55 , it would be interesting to check whether our results would be replicated in other cultures.
Finally, because our participants were recruited in scientific meetings, a selection bias should be considered given that they could be more motivated and involved in professional self-actualization than other GPs who did not have the opportunity to participate in our study. We suspect that a stronger pattern of biases would be found among the latter.
In sum, due to ecological and construct validity constraints some of our findings should be interpreted with parsimony. Future studies using more ecologically valid methodologies (e.g., video clips) and more representative samples and pain-related clinical scenarios could be useful to overcome these shortcomings.
Despite the shortcomings, this study has both theoretical and practical implications. Theoretically, our results suggest that GP's PMP may in certain situations be influenced by their gender assumptions and representations, although this contention should be directly tested. If this is true, it contradicts the ideology of the socially, culturally and, hence, gender neutral physician, which still prevails within the medical establishment. 4,29,44 . In fact, despite all the efforts of medical institutions to produce This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 30 "neutral" doctors, such socialization processes may seem unable to entirely overcome earlier gender socialization processes, in general, and in pain, in particular 37,39 ; where, as compared to men, women seem to be taught to be more accepting of their own and other's pain and its associated distress. In practical terms, this suggests that instead of ignoring such influences, they should be addressed, e.g., by the integration of gender awareness training programs in medical curricula. 18,44 . Helping doctors to become aware of how gender influences their own PMP may well be an important step toward ensuring gender equity in pain management services. This is a postprint version of a manuscript published in THE JOURNAL OF PAIN available at http://dx.doi.org/10.1016/j.jpain.2013.03.004 31