Psychological and psychosocial predictors of chronic postsurgical pain: a systematic review and meta-analysis

Abstract

Knowledge about psychological and psychosocial predictors of chronic postsurgical pain is important to identify patients at risk for poor outcomes. The objective of this systematic review with meta-analysis was to assess the effect of such predictors. A comprehensive search of the available literature on this topic was performed using the electronic databases PubMed, Scopus, Embase, and PsycInfo. Estimates of the effect of each predictor were extracted, and both a narrative synthesis and a quantitative synthesis of these estimates were performed. Multiple imputation was used to take into account the effect of nonsignificant estimates in case they were not reported by original studies. From a sample of 8322 records, 83 articles were included in the narrative synthesis and 41 studies were used to perform the meta-analyses. The narrative synthesis showed that evidence about the effect of psychological predictors is heterogeneous, with few expected predictors, such as optimism, state anxiety and psychological distress, consistently associated with chronic postsurgical pain. By contrast, the meta-analyses showed that state anxiety, trait anxiety, mental health, depression, catastrophizing and, to a lesser extent, kinesiophobia and self-efficacy have a weak but significant association with chronic postsurgical pain. In conclusion, this study showed that psychological predictors have a significant association with chronic postsurgical pain and that state anxiety is the most explicative one.

1. Introduction

The past 2 decades have shown a constant growth in studies on chronic postsurgical pain.²¹ The importance of identification of its modifiable predictors has led to an increasing interest in psychological and psychosocial factors, which are hypothesized to increase its incidence by interacting with the physiological underpinnings of pain, by shaping its perception, and by influencing pain behaviors.¹⁰ However, although the strength of the association between these factors and pain has been clearly established in patients with various chronic diseases,^6,25,32 studies focusing on postsurgical pain report conflicting results, with estimates ranging from moderate values³⁶ to weaker or absent ones.⁴⁵

A first review of 50 studies retrieved from an initial search of 800 articles from 1996 to 2006 tried to collate in a narrative manner the available evidence on this topic and found that depression, psychological vulnerability, stress, and late return to work were likely predictors of chronic postsurgical pain, whereas the role of anxiety was uncertain.¹³ However, it was impossible to provide a quantitative synthesis of the results due to variability among the studies in terms of assessment of predictors and outcomes and the inclusion of studies on both cancer and noncancer conditions. This led to uncertain results. A previous attempt to use meta-analytic methods was performed by Jackson et al.¹⁶ in a review of 47 studies (n = 6207, based on an initial search of 7604 records) on acute and chronic pain due to both cancer and noncancer conditions. The authors found a weak correlation between emotional distress and postsurgical pain (Pearson's r = 0.25, 95% CI = 0.20-0.29). However, this estimate was the result of pooling data from studies focusing on both acute and chronic pain, on both cancer and noncancer conditions, and on a heterogeneous set of predictors, merged under the category of “emotional distress.” This category includes predictors such as anxiety, depression, aversive mood states (eg, anger, fear, and guilt), cognitions (eg, catastrophizing), and coping strategies. Because the effects of each predictor computed by the authors through a moderation analysis included both acute and chronic pain and cancer and noncancer condition, the differential role of each predictor on chronic postsurgical pain needs to be studied in more detail. Theunissen et al.³⁷ studied the role of anxiety and catastrophizing on the incidence of chronic postsurgical pain by performing a systematic review of 29 cohort studies, case-control studies, and randomized controlled trials on both cancer and noncancer conditions (n = 6628, based on an initial search of 512 records). The authors estimated that these predictors have an overall effect of weak intensity (odds ratios between 1.55 and 2.10). An additional meta-analysis focusing on predictors of total knee arthroplasty (TKA)¹⁹ found weak effects of anxiety, depression, and catastrophizing on chronic postsurgical pain (n = 29,993, based on an initial search between 592 records). Both studies, however, based their analysis on a limited set of studies and pooled heterogeneous outcomes.

These meta-analyses extracted data from studies providing full information for effect size calculation. However, several studies in the field of predictors of postsurgical pain do not report the effect estimates of nonsignificant predictors or of predictors that are significant in bivariate analyses but fail to reach significance in multivariable analyses.^13,14,37 Attempts to retrieve unpublished data by contacting authors have poor success rates,⁴¹ and exclusion of these studies could be problematic because it may produce an artificial inflation of the overall effect.³¹ A possible way to overcome this issue is to use multiple imputation techniques.²⁹ Briefly, these techniques create multiple data sets substituting plausible values for missing nonsignificant data. The results of independent meta-analyses performed on each imputed data set are then averaged to produce an unbiased overall effect. Applying these techniques to studies in the field of chronic postsurgical pain could help to provide more precise estimates of the effects of each predictor because it enables the inclusion of studies with negative results.

Clarifying the role of psychological predictors of postsurgical pain might help clinicians to identify patients at risk for poor outcomes. In addition, more precise knowledge about the independent role of each predictor may be important for the selection of the most appropriate measurement instrument to be used in the presurgical stage and to plan possible educational or therapeutic interventions. Therefore, we planned to perform a systematic review on this topic and to provide a quantitative synthesis of the available studies.

2. Methods

2.1. Literature search

The reporting of this study follows the guidelines from the PRISMA statement²² This study followed a review protocol, which was not published. An initial search on PubMed, Embase, Scopus, and PsycInfo was performed on April 23, 2018, with no restriction on publication dates. An update of this search was performed on February 21, 2020, restricting the search to articles published between January 2017 and the date of the search. Keywords included “surgery,” the names of common surgical procedures, “pain,” “predictor,” “psychological,” and names of psychological and psychosocial potential predictors; synonyms for each keyword were also provided (details of the search are reported in the supplemental digital content, available at http://links.lww.com/PAIN/B113). In addition, the reference lists of reviews retrieved during the search were hand-searched to find potentially eligible records. We decided to use broad inclusion criteria to study the effects of any potential predictor of postsurgical pain in a comprehensive way. At the same time, we decided not to consider studies on acute postsurgical pain (ie, follow-up duration <3 months) because its characteristics are strictly related to the surgical procedure; and on cancer pain, because pain could be due to additional treatments (eg, chemotherapy or radiotherapy) or to disease progression. Therefore, studies were included if:

• (1) they used a prospective or retrospective longitudinal design;
• (2) the effect of at least one psychological or psychosocial predictor was estimated;
• (3) psychological or psychosocial predictors were assessed before surgery;
• (4) the presence of postsurgical pain or its intensity was considered as an outcome;
• (5) follow-up duration was at least 3 months.

Studies were excluded if:

• (1) study reporting was not in the form of a peer-reviewed full text;
• (2) they did not focus on adult participants;
• (3) they included cancer patients in their samples.

Psychological predictors were defined as mental states, beliefs, emotions, or attitudes that can potentially modify the individual's behavior; psychosocial predictors were defined as the interaction between psychological processes and social factors or factors related to the individual's relationships. No language restriction was imposed in the search. Records in languages other than English, Italian, Spanish, and French were considered for inclusion after their text was translated using Google Translate. All the records were exported in the Rayyan web software,²³ and their titles and abstracts were screened according to inclusion and exclusion criteria. The full text of the studies was then accessed, evaluated, and used to extract data. Both title, abstract, and full-text screening and data extraction were performed independently by a postdoctoral scholar and a doctoral student in Clinical Psychology. In each of these phases, discordance between the reviewers was resolved by consensus.

2.2. Data extraction

Data were extracted using a purposely defined form built using the Covidence web tool.¹⁵ Data extracted from each study included the country where it was conducted, its study design (ie, retrospective or prospective cohort study), the sex distribution of the sample, its mean age, diagnosis (eg, osteoarthritis and heart disease), surgical procedure (eg, TKA and lumbar fusion), number of subjects enrolled at baseline (considering patients who provided informed consent), number of subjects interviewed at the last follow-up available, name of any psychological predictor included, measurement instrument used to assess each predictor, and measurement instrument used to assess the outcome and results at each follow-up, considering both bivariate and multivariable analyses.

In the case of missing outcome data or if only data from multivariable analyses were available, an attempt was made to obtain the results of the bivariate analyses from the authors. All included studies were then evaluated according to their methodological quality; the resulting information was used during the qualitative and quantitative synthesis of the estimates.

2.3. Methodological quality assessment

The quality of the included studies was assessed using the Quality In Prognostic Studies (QUIPS) tool.¹¹ This tool was chosen for its specificity to assess risk of bias in studies focusing on health-related predictors. It includes 6 domains: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, and statistical analysis and reporting (see supplemental digital content, available at http://links.lww.com/PAIN/B113). Each domain includes multiple subitems that were used to provide an overall grade. For each domain, risk of bias was judged low, moderate, or high. A study overall grade was computed by attributing, for each domain, a grade of 2 in presence of low risk of bias, 1 in presence of moderate risk of bias, and 0 in case of high risk of bias.

Ranges for low, moderate, and high quality were 0 to 8, 9 to 10, and 11 to 12, respectively. These grades were used in the narrative synthesis to summarize the average quality of the evidence for significant and nonsignificant results and were included as fixed effects in the subsequent mixed-effect model meta-analysis (see below).

2.4. Narrative synthesis

All included studies were used in a preliminary narrative synthesis of the results. This procedure was performed to assess the consistency regarding the association of each predictor with pain, while considering also the methodological quality of the corresponding studies, and to provide a preliminary assessment of predictors that could not be pooled in the subsequent quantitative synthesis. First, studies were described using data extracted in the previous step. Then, the focus shifted to an individual predictor level. The consistency of estimates of the association between each predictor and postsurgical pain was examined, counting the number of significant and nonsignificant results. If both bivariate and multivariable analyses were reported, the results were considered separately.

2.5. Quantitative synthesis

Predictors being assessed by 4 or more studies were considered for quantitative synthesis. We decided not to pool data from 2 or 3 studies to avoid performing underpowered analyses.¹⁵ For each predictor, we decided to exclude individual estimates if:

• (1) the predictor was measured using nonvalidated instruments;
• (2) the predictor was operationalized in a markedly different way from the majority of other studies;
• (3) the outcome was operationalized other than from presence of pain or intensity of pain.

Outcomes such as change of pain scores over time (eg, “improvement” and “worsening”), pain trajectories over time in studies with multiple measures, or improvement above a cutoff (eg, improvement >70% of baseline score) were not included in the analysis. This was done because including these outcomes would have added an additional source of heterogeneity.

For each predictor, separate random-effect models were used to synthesize quantitative results. These models were chosen because the variability across studies in terms of surgery, measurement instruments, and time points did not enable, from either a theoretical or statistical point of view, the assumption of a common effect across studies.⁵ Pearson's r was used as effect size. Original data were presented in the form of correlations, unstandardized and/or standardized regression coefficients, raw mean values and SDs, t-tests, odds ratios, and risk ratios. All these parameters were first converted to Pearson's r using standard formulae,^5,26 then converted to Fisher's z and pooled. Finally, the result of each meta-analysis was backtransformed to Pearson's r for a more immediate interpretation (all formulae are reported in the supplemental digital content, available at http://links.lww.com/PAIN/B113). Cutoffs for interpretation of Pearson's r were 0.1, 0.3, and 0.5, indicating small, medium, and large effects, respectively.⁷

When both bivariate and multivariable analyses were present, estimates of the former were included in the analysis. Data from different time points, outcomes (eg, pain in adjacent areas), subscales (eg, the subscales of the Pain Catastrophizing Scale), outcome measurement instruments (eg, visual analogue scale and numeric rating scale), or subgroups (eg, primary TKA and revision TKA) within a single study, if comparable, were aggregated using standard formulae.⁵ Estimates from data points distanced more than 12 months were analyzed separately. A multiple imputation procedure was performed to account for nonsignificant missing data. Briefly, this procedure follows these stages^2,29:

• (1) Bounds of plausible nonsignificant results are computed using data from complete studies;
• (2) Parameters for multiple imputation are computed maximizing the likelihood that the imputed value would fall in these bounds;
• (3) Multiple data sets with plausible imputed values are created;
• (4) Independent random-effect meta-analyses are performed on each data set.
• (5) The results are averaged to obtain an overall effect.

To assess heterogeneity, I² was used. This index is usually used as a measure of inconsistency and can be interpreted as the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error. Values of 25%, 50%, and 75% were used as cutoffs for low, medium, and high inconsistency, respectively.¹²

For each predictor, a prespecified subgroup analysis was performed separating studies by outcome type (ie, presence of chronic postsurgical pain and pain intensity). Separate meta-analyses were performed on each group.

Mixed-model meta-analyses were then performed to determine sources of heterogeneity. Fixed factors included in the model were: percentage of females enrolled, mean age of the sample, surgical area (ie, leg, back, inguinal, thorax, and mixed), methodological quality, outcome instrument, follow-up duration, and type of analysis (ie, bivariate and multivariable). All fixed factors were included in a first mixed-effect model and then deleted through a backward procedure until only significant factors remained. The Knapp–Hartung method was used to perform significance tests for each predictor. This method was chosen because it has shown good performance in meta-analyses with a small number of studies.⁴⁰ Significant factors were then used to perform subgroup analyses using the same method described above.

Funnel plots were visually inspected to detect presence of publication bias or possible sources of heterogeneity.³³ In addition, a trim-and-fill method was used as a sensitivity analysis.⁹ This procedure provides an estimate of the number of missing studies due to publication bias and the effect that these studies might have on the overall effect.

The multiple imputation procedure is contained in the R (version 3.4.2) package metansue,²⁸ which was also used to compute the overall estimates for each predictor. The remaining analyses were performed using the R package metafor.³⁹

3. Results

The first database and reference lists search yielded 6329 records after duplicate removal. The update of the search yielded 1993 additional records. From 234 full-text articles assessed for eligibility, 83 (describing 73 studies, n = 40,784) were included in the narrative synthesis (Fig. 1).

Figure 1.: PRISMA flowchart of the systematic literature review.

Among the studies, 56 were prospective and 17 were retrospective (ie, were performed a posteriori using data stored in medical records or collected for other research purposes) cohort studies. Studies were more likely to include women than men (mean female percentage: 59.5%) and middle-aged subjects (mean age: 60.2, SD: 10.2). The median sample size was 203.5. The most common surgical procedure was TKA (37 studies), followed by lumbar surgery (13 studies), total hip replacement (5 studies), hysterectomy or other gynecological surgeries (4 studies), anterior cervical decompression, with or without fusion (4 studies), coronary artery bypass graft with or without valve replacement (3 studies), surgery for groin hernia (3 studies), mixed surgeries (3 studies), total shoulder arthroplasty (2 studies); and ankle and hind foot reconstruction, liver donation surgery, pelvic laparoscopy, meniscectomy, arthroscopic rotator cuff repair, and total shoulder arthroplasty (1 study each). Depression (38 studies), anxiety (37 studies), and catastrophizing (28 studies) were the most frequently assessed predictors. Even if catastrophizing had both time-varying and stable characteristics,³⁴ all the studies included in this review that evaluated this predictor assessed and considered it as a trait. Anxiety was operationalized and assessed as a trait or pathological condition in 30 studies, whereas in 7 studies, it was operationalized also or only as a temporary state. The former will be from now on referred to as “trait anxiety,” and the latter as “state anxiety.” Eleven studies assessed the association between chronic postsurgical pain and the SF-12 or SF-36 mental component score. Even if some studies used it as a measure of depression, this score was considered as a separate predictor because its items refer to the impact of the general emotional state on the daily activities. From now on, this score will be referred to as “mental health.” Social support was the only psychosocial variable inserted as predictor by the included studies. Follow-up duration ranged from 1.5 months to 10 years, with a mode of 12 months and a median of 6 months. The full references of the included studies are reported in the supplemental digital content (available at http://links.lww.com/PAIN/B113). Description of their results are reported in Table 1.

Table 1 - Description and results of the studies included in the systematic review.

Name	Country	Design	F%*	Age M (ds)	N	N (T1)†	Diagnosis	Surgery	QUIPS‡	Variable	Scale	Outcome	Follow-up	Biv.§	Multi‖
Aasvang, 2010	Denmark, Germany	Prospective	0	55.2 (13.3)	464	442	Groin hernia	Groin hernia repair	12	Depression Trait anxiety	HADS-D HADS-A	CPP	6		− −
Abbott, 2011	Sweden	Prospective	61.7	50.6 (10.4)	107	87	Degenerative disk disease	Lumbar fusion surgery	10	Mental health Kinesiophobia Self-efficacy Outcome expectancy Coping Catastrophizing	SF-36-MH TSK SES BBQ CSQ CSQ-C	VAS	24 to 36		− − − − − +
Amusat, 2014	Canada	Prospective	62	68	405	n/a	OA	TKA	9	Depression	CES-D	WOMAC	6	+	+
Archer, 2011	The United States	Prospective	58	59.1 (13.3)	141	120	Degenerative spine conditions	Spinal surgery	12	Depression Kinesiophobia	PHQ-9 TSK	BPI	3		− −
Seebach, 2012	The United States	Prospective	58	59.1	141	123	Lumbar or cervical degenerative conditions	Laminectomy and/or fusion surgery	12	Depression Positive affect Negative affect	PHQ-9 PANAS PANAS	BPI	3		− − −
Archer, 2014	The United States	Prospective	58	59.1 (13.3)	141	120	Degenerative spine conditions	Spinal surgery	11	Depression Kinesiophobia	PHQ-9 TSK	BPI	6		− −
Attal, 2014	France	Prospective	65	68.7 (8.9)	89	69	OA	TKA	12	Depression State anxiety Trait anxiety Passive coping Active coping TMT-A TMT-B ROCF-C ROCF-IR	BDI STAI STAI CSQ CSQ TMT-A TMT-B ROCF-C ROCF-IR	BPI	6, 12	+, − +, + −, − +, + −, − −, − +, + −, + −, +
Masselin-Dubois, 2013	France	Prospective	65	68.7 (8.9)	89	89	OA	TKA	11	Depression Catastrophizing State anxiety Trait anxiety	BDI PCS STAI STAI	BPI >3	3	− − + −
Bierke, 2017	Germany	Prospective	63	69 (7.9)	138	100	OA	TKA	8	Catastrophizing Trait anxiety	PCS STAI	NRS	6, 12	+, − +, +
Birch, 2019	Denmark	Prospective	52	67.3	859	615	OA	TKA, UKA	10	Catastrophizing	PCS	OKS pain	12		+
Bossmann, 2017	Germany	Prospective	66.1	68.8	56	47	OA	TKR	10	Catastrophizing	PCS	WOMAC	6	−	−
Brander, 2003	The United States	Prospective	55.2	66 (10.5)	116	112	OA	TKA	7	Depression Trait anxiety Stress	BDI STAI PSS	VAS	3, 6, 12	−, −, + −, −, + −, −, −
Brander, 2007	The United States	Prospective	55.2	66 (10.5)	116	83	OA	TKA	7	Depression Trait anxiety	BDI STAI	VAS	60	−, − −, −
Buvanendran, 2019	The United States	Prospective	65.3	65	311	245	OA	TKA	12	State anxiety Trait anxiety Catastrophizing Depression Mental health	STAI STAI PCS BDI SF-36	NRS >4	6	+ + + + +	+ − − − −
Cho, 2017	The United States	Prospective	58.7	65.7 (10.1)	60	46	OA	TSA	8	Depression Trait anxiety	HADS-D HADS-A	VAS	12	− −	− −
Choiniere, 2014	Canada	Prospective	21	61.9 (10.2)	1247	976	Heart disease	CABG and/or valve replacement	10	Depression Trait anxiety Catastrophizing Mental health Physical health	HADS-D HADS-A PCS SF12-MCS SF12-PCS	CPP	3-24		+ + + + +
Cunningham, 2019	The United States	Retrospective	50.4	65	1037		OA	TKA	8	Depression diagnosis	Charlson comorbidity index	VAS	12-24, 60		+
Dahl, 2014	Sweden	Retrospective	58.5	69.3 (8.7)	2883	2123	OA	TKA	9	Psychological distress	EQ-5D item 5	KOOS, VAS	12		+
Dailiana, 2016	Greece	Prospective	73.6	67	378	346	OA	TKA; THA	9	Social support	Marital status	WOMAC	12		−
D'Angelo, 2010	Italy	Prospective	40.7	46	142	108	Lumbar disk herniation	Microdiscectomy	8	Trait anxiety Depression	STAI ZUNG	VAS	12	+ −	+ −
De Groot, 1997	The Netherlands	Prospective	46	44	126	112	Disk herniation or canal stenosis	Lumbar surgery	9	State anxiety Coping	STAI TMSI	VAS	3	+ −
Den Boer, 2006	The Netherlands	Prospective	50	43	310	277	LRS	Surgery for LRS	11	Kinesiophobia Passive coping Outcome expectancy	TSK PCI 4 item scale	VAS	6		+ + +
Desmeules, 2013	Canada	Prospective	66	67 (9.3)	197	138	OA, rheumatoid arthritis	TKA	11	Social support Psychological distress	QHS PSI	WOMAC	6		− −
Divi, 2020	The United States	Retrospective	53.4	53	264	n/a	Cervical radiculopathy, myelopathy, radiculomyelopathy	ACD	4	Mental health	SF-12	VAS	12-46	+	−
Edgley, 2019	Australia	Prospective	30%	43 (median)	326	229	Orthopedic trauma	Trauma surgery	8	Psychological distress Catastrophizing	Kessler psychological distress scale PCS	CPP	3 3	− −	−
Escobar, 2017	Spain	Prospective	73.6	71.8 (6.7)	855	640	OA	TKA	9	Social support Mental health	Single item (assistance) SF-36-MH	WOMAC, SF36-BP	6		+, − +, −
Forsythe, 2008	Canada	Prospective	63.6	69	55	48	OA	TKA	8	Catastrophizing	PCS	MPI	24	+
Goh, 2019	Japan	Retrospective	40.4	56	104	n/a	Cervical myelopathy	ACDF	6	Mental health	SF-36	VAS neck, VAS arm	6, 24	+, +
Guimaraes pereira, 2016	Portugal	Prospective	29.9	68	310	288	Heart disease	CABG and/or heart valve replacement	10	Depression Catastrophizing Trait anxiety Self-esteem	DUKE-D PCS DUKE-A DUKE-SE	CPP	3	− + − +
Han, 2017	China	Prospective	100	n/a	966	870	n/a	Hysterectomy	12	Depression Trait anxiety	HADS-D HADS-A	BPI	3	+ +	+ +
Hegarty, 2012	Ireland	Prospective	45.5	39	53	53	Intervertebral disk herniation	Lumbar discectomy	9	Depression Trait anxiety Catastrophizing	HADS-D HADS-A PCS	70% pain reduction	3	− − −	− − −
Holtzman, 2014	Canada	Prospective	49.2	38.7	80	65	Healthy patients	Liver donation	8	Trait anxiety Pain-related anxiety Anxiety sensitivity Catastrophizing	STAI PASS-20 Anxiety sensitivity index PCS	NRS	6, 12	− + − −
Hoofwijk, 2015	The Netherlands	Prospective	56.4	53	1396	908	Mixed	Mixed	8	Psychological distress Catastrophizing Surgical fear Optimism Quality of life	EQ-5D item 5 PCS (6 items) SFQ LOT (4 items) EQ-5D	VAS >3	12	− − + + +	− − + + +
Hovick, 2016	Norway	Prospective	50.7	64.8	71	61	OA	TKA	9	Catastrophizing	PCS	BPI	2-12		−
Iversen, 1998	The United States	Prospective	58	69 (9)	257	228	Lumbar spine stenosis	Lumbar spine stenosis surgery	9	Depression Pain relief expectations Pain relief satisfaction	ZDS Nonvalidated Nonvalidated	6-Level scale	6		+ + +
Jarrell, 2014	Canada	Prospective	100	33 (7.5)	61	61	Nonacute pelvic pain	Pelvic laparoscopy	6	Depression Catastrophizing Pain disability	CES-D PCS PDI	VAS	6	+ + +	− + −
Jimenez-Ortiz, 2018	Spain	Prospective	74.2	70.86	265	260	OA	TKA	8	Depression Trait anxiety	HADS-D HADS-A	VAS	12	− +
Johansson, 2010	Sweden	Prospective	40	40 (8)	59	55	Lumbar disk herniation	Lumbar discectomy	6	Coping Catastrophizing Expected outcome Kinesiophobia	CSQ CSQ Item about return to work TSK	CPP	12		− − + −
Johansson, 2016	Sweden	Retrospective	39.3	n/a	59	56	Lumbar disk herniation	Lumbar discectomy	5	Behavioral variables Expectation of return to work	TSK+CSQ¶ Item about return to work	VAS	3, 24		−, − +, +
Judge, 2012	England	Prospective	62.1	71.3 (9.4)	3608	1991	OA, RA	TKA	8	Psychological distress	EQ-5D item 5	OKS	6	+	+
Kornilov, 2018	Russia	Prospective	94.9	63 (8)	100	79	OA	TKA	5	Depression Trait anxiety	HADS-D HADS-A	NRS walking >3	12	− −
Laufenber-Feldmann, 2018	Germany	Prospective	51	58.8 (16.5)	158	143	Lumbar disk herniation	Lumbar discectomy	12	Trait anxiety	GAD-7	NRS (increase)	6	−	−
Lavernia, 2012	The United States	Retrospective	66.4	n/a	563	563	OA	THA and TKA	10	SF36 mental health SF36 physical functioning	SF36 SF36	WOMAC	12		+ −
Lingard, 2007	The United States, The United Kingdom, Australia, Canada	Prospective	60.3	n/a	952	682	OA	TKA	11	Mental health	SF36	WOMAC	3, 12, 24		+, +, +
Liu, 2020	China	Retrospective	10.2	n/a	256	236	Inguinal hernia	Inguinal hernia repair	12	Trait anxiety State anxiety	STAI STAI	NRS >3	3	− +	− +
Lopez-Olivo, 2011	The United States	Prospective	65	65 (9)	272	241	OA	TKA	11	Depression Social support Trait anxiety Stress Coping Locus of control Self-efficacy Optimism	DASS-21 MOS-SSS DASS-21 DASS-21 COPE MHLC ASES LOT-R	WOMAC	6	+ + + + + + + −	− − − − + + − −
Mayo, 2017a	The United States	Retrospective	42.3	49.2 (9.4)	52	52	Neck pain and radiculopathy	ACD	8	Mental health	SF12-MCS	VAS	3		+, −
Mayo, 2017b	The United States	Retrospective	68.2	41 (11)	110	110	Lumbar disk herniation	Lumbar discectomy	8	Mental health	SF12-MCS	VAS	3, 6		−
Merrill, 2018	The United States	Retrospective	51.4	60	111	111	Lumbar spine stenosis	Lumbar laminectomy without fusion	6	Depression	PROMIS-D	PROMIS-P	6	+
Mulligan, 2016	The United States	Retrospective	53.9	55	132	n/a	n/a	Ankle and hind foot reconstruction	8	Mood disorder	Medical chart	VAS	12	+	+
Noiseux, 2014	The United States	Prospective	58	61.7 (9.8)	215	193	OA	TKA	10	Depression Trait anxiety Catastrophizing	GDS STAI-T PCS	CPP#	6	+ + +	− + −
Papakostidou, 2012	Greece	Prospective	79.4	69.17 (79.4)	204	188	OA	TKA	11	Depression Social support	CES-D10 Marital status	VAS, WOMAC	3, 6, 12 12	+, +, +, + n/a	−, + −
Perruccio, 2019	Canada	Prospective	57	65	577	477	OA	TKA	12	Psychological distress	HADS	KOOS pain score	3		+
Pinto, 2012	Portugal	Prospective	100	49	203	186	Mixed benign conditions	Hysterectomy	11	Depression Trait anxiety Illness perception Surgical fear Catastrophizing Coping	HADS HADS IPQ-R SFQ CSQ-C CSQ	CPP	4		− + − − + −
Pinto, 2013	Portugal	Prospective	66.3	64 (7.86)	130	92	OA	TKA	10	Depression Trait anxiety Illness perception	HADS HADS IPQ	VAS >3	6	− + +	− − −
Potter, 2015	The United States	Prospective	n/a	n/a	89	70	Full-thickness rotator cuff tears	Arthroscopic rotator cuff repair	7	Depression Somatization	ZDS SPQ	VAS	12		− −
Powell, 2012	United Kingdom	Prospective	4.3	61.5 (12)	140	115	Inguinal hernia	Hernia repair surgery	9	Depression Trait anxiety Kinesiophobia Catastrophizing Coping Activity expectations Expected pain control Surgical fear Optimism	HADS-D HADS-A TSK CSQ CSQ Single item Single item Single item LOT	CPP, BPI-WPI	4	+ + − − − − + − +	− − − − − − − − +
Rice, 2018	New Zealand	Prospective	48	69	363	288	OA	TKA	10	Trait anxiety State anxiety Expected pain Catastrophizing Depression	STAI STAI Ad hoc question PCS BDI	WOMAC	6, 12	+, + −, − +, + +, − −, −	−, − −, − −, − −, − −, −
Riddle, 2009	The United States	Prospective	70.7	63.7	283	140	OA	Knee arthroplasty	8	Depression Trait anxiety Kinesiophobia Self-efficacy Catastrophizing	PHQ-8 GAD-7 TSK ASES PCS	WOMAC**	6		− − − − +
Rolfson, 2009	Sweden	Retrospective	56.9		6158	n/a	OA	THR	6	Psychological distress	EQ-5D item 5	VAS	12		+
Rosenberger, 2009	The United States	Prospective	44	48.2 (11.9)	191	180	Knee diseases	Meniscectomy or arthroplastic knee repair	10	Depression Surgery stress Optimism	CES-D SSS LOT-R	MPI	12	+ + +	− + +
Singh, 2010	The United States	Retrospective	62	68.8	702	n/a	OA	TKA	6	Pessimism	MMPI	Moderate or severe pain	24, 60		+, −
Singh, 2013	The United States	Retrospective	55	68 (10)	8672	5115	OA, RA, other	TKA or revision	2	Depression Trait anxiety	H-ICDA H-ICDA	CPP#	24, 60	+, + +, +	+, + +, +
Singh, 2016	The United States	Retrospective	55	67	684	684	Hip disease	THA or revision	4	Optimism	MMPI	CPP#	24		−
Sinikallio, 2007	Finland	Prospective	58	61.7	119	100	LSS	LSS surgery	10	Depression	BDI	VAS >14	3		+
Sinikallio, 2009	Finland	Prospective	58	61.7	119	95	LSS	LSS surgery	10	Depression	BDI	VAS >11	12		−
Sinikallio, 2011a	Finland	Prospective	58	61.7	119	93	LSS	LSS surgery	10	Depression	BDI	VAS >0	24		+
Sinikallio, 2011b	Finland	Prospective	60	61.7	119	90	LSS	LSS surgery	10	Life satisfaction Sense of coherence	Life Satisfaction scale SOC scale	VAS >0	24		− −
Tuomainen, 2018	Finland	Prospective	62.5	58	120	72	LSS	LSS surgery	8	Depression	BDI	VAS	120		−
Skeppholm, 2017	Sweden	Prospective	53%	47	151	136	Cervical radiculopathy	ACDF or artificial disk replacement	10	Depression	HADS-D	VAS	24	+
Sullivan, 2011	Canada	Prospective	60.8	67	n/a	120	OA	TKA	10	Depression Kinesiophobia Catastrophizing	PHQ-9 TSK PCS	WOMAC	12	+ + +	− − +
Yakobov, 2014	Canada	Prospective	61.2	67 (8.2)	116	n/a	OA	TKA	9	Perceived injustice Catastrophizing Kinesiophobia	IEQ PCS TSK	WOMAC	12	+ + −	+ − −
Theunissen, 2016	The Netherlands	Prospective	100	46.9 (7.1)	517	376	Benign gynecological conditions	Hysterectomy	4	Surgery-related worries Psychological robustness Social support	SFQ+PCS LOT-R + CES-D + WBQ12 MOS-SSS	BPI >4	3-12		+ − −
Utrillas-Compaired, 2014	Spain	Retrospective	69.3	73 (6.4)	215	202	OA	TKA	6	Depression Trait anxiety	HADS-D HADS-A	KSS	12		− −
VanDenKerkhof, 2012	Canada	Prospective	100	49 (11)	696	433	Gynecologic disease	Gynecologic surgery	9	Depression State anxiety Trait anxiety Catastrophizing Somatization	CES-D STAI-S STAI-T PCS (2 items) SSST	CPP	6	+ + − + +	− + − − −
Vogel, 2019	Germany	Prospective	n/a	66.3	79	n/a	OA	TKA	9	Mental health Kinesiophobia Trait anxiety Catastrophizing Depression Somatization	SF-36 TSK BSI PCS BSI BSI	WOMAC	12		− − − − − −
Vogel, 2020	Germany	Prospective	n\a	65.82	144	n/a	OA	TKA	10	Borderline personality	IPO	WOMAC	12	−	+
Wang, 2018	China	Prospective	67.6	49 (13.5)	266	209	Mixed	Mixed	9	Depression Trait anxiety Kinesiophobia Catastrophizing Surgical fear Self-efficacy	HADS-D HADS-A TSK PCS SFQ PSEQ	BPI	4	− − − − + −	− − − − + −
Wong, 2018	The United States	Prospective	49.2%	66.7	280	n/a	OA	TSA	9	Psychiatric diagnosis	Medical chart	VAS	12, 24		−
Wright, 2017	The United States	Prospective	61	63.9	123	119	OA	TKA or THA	10	Catastrophizing	PCS	VAS	3	+
Wylde, 2012	Ireland	Prospective	62	70 (9)	251	220	OA	TKA	12	Depression Trait anxiety Self-efficacy	HADS-D HADS-A PSEQ	WOMAC	12		+ + −
Wylde, 2017	Ireland	Prospective	64	70	266	233	OA	TKA	11	Depression Trait anxiety Catastrophizing Self-efficacy Social support	HADS-D HADS-A PCS PSEQ MOS-SSS	WOMAC	12, 60	−, − +, − −, − −, − −, −	−, − +, − −, − −, − −, −

Results are reported in the final 2 columns. For each follow-up point, significant results are represented with the symbol “+,” nonsignificant results with the symbol “−,” References are reported in Appendix I (available at http://links.lww.com/PAIN/B113).

^*Percentage of female participants.

^†Number of participants participating at last follow-up.

^‡QUIPS total score.

^§Results of the bivariate analysis.

^‖Results of the multivariable analysis.

^¶Scale based on factor analysis of TSK and CSQ items of subscales self-statement, catastrophizing, control over pain, and ability to decrease pain.

^#Chronic postsurgical pain measured as No/mild vs moderate/severe pain.

^**Improvement from baseline.

ACD, anterior cervical discectomy; ACDF, anterior cervical discectomy and fusion; AIMS, arthritis impact measurement scales; ASES, arthritis self-efficacy scale; BBQ, back beliefs questionnaire; BDI, Beck depression inventory; BPI, brief pain inventory; BPI-WPI, brief pain inventory—worst pain intensity; CABG, coronary artery bypass graft; CES-D, center for epidemiologic studies—depression scale; CPP, presence of chronic postsurgical pain; CSQ, coping strategies questionnaire; CSQ-C, coping strategies questionnaire—catastrophizing subscale; DASS-21, depression anxiety stress scales; DUKE-A, duke anxiety-depression scale—anxiety subscale; DUKE-D, duke anxiety-depression scale—depression subscale; DUKE-SE, duke anxiety-depression scale—self-esteem subscale; FABQ, fear-avoidance beliefs questionnaire; GAD-7, generalized anxiety disorder-7; H-ICDA, hospital adaptation of the international code for diseases; GDS, geriatric depression scale; HADS-A, hospital anxiety and depression scale—anxiety subscale; HADS-D, hospital anxiety and depression scale—depression subscale; IEQ, injustice experiences questionnaire; IPO, inventory for the assessment of borderline personality organization; IPQ-R, illness perception questionnaire—revised; KOOS, knee injury and osteoarthritis outcome score; LOT, life orientation test; LOT-R, life orientation test—revised; LRS, lumbosacral radicular syndrome; MHLC, multidimensional health locus of control; MOS-SSS, medical outcome study—social support scale; MPI, McGill pain questionnaire; NRS, numeric rating scale; OA, osteoarthritis; OKS, oxford knee score; PCS, pain catastrophizing scale; PANAS, positive and negative affect scale; PCI, pain-coping inventory; PDI, pain disability index; PHQ-9, patient health questionnaire-9; PROMIS-D, PROMIS depression scale; PROMIS-P, PROMIS pain scale; PSI, psychological symptoms index; PSS, perceived stress scale; QHS, Quebec health survey; ROCF-C, Rey–Osterrieth complex figure—copy; ROCF-IR, Rey–Osterrieth complex figure—immediate recall; SES, self-efficacy scale; SF12-MCS, short form-12—mental component scale; SF12-PCS, short form-12—physical component scale; SF36-BP, short form-36 questionnaire—bodily pain subscale; SF36-MH, short form-36 questionnaire—mental health subscale; SFQ, surgical fear questionnaire; SPQ, somatic perception questionnaire; SSS, surgery stress sale; SSST, seven-symptom screening test; STAI, state-trait anxiety inventory; STAI-T, state-trait anxiety inventory—trait scale; TKA, total knee arthroplasty; TMT-A, trail making test—part A; TMT-B, trail making test—part B; TMSI, threatening medical situation inventory; TSA, total shoulder arthroplasty; TSK, Tampa scale for kinesiophobia; UKA, unicompartimental knee arthroplasty; VAS, visual analogue scale; WBQ12, well-being questionnaire-12; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index; ZDS, Zung depression scale.

3.1. Methodological quality

The methodological quality of the included studies was heterogeneous. Fifteen studies (20.3%) were judged of high quality, 29 (39.2%) of medium quality, and 29 (39.2%) of low quality (Fig. 2). Most of the studies showed low risk of bias in the study attrition (58.1%), predictor measurement (68.9%), and outcome measurement (86.4%) domains of the QUIPS. On the contrary, 41.8% of the studies showed low risk of bias and 27% high risk of bias in the statistical analysis and reporting domain.

Figure 2.: Methodological quality of the included studies according to the QUIPS tool.

3.2. Narrative synthesis

Estimates of the association between psychological predictors and postsurgical pain were conflicting and were therefore separated based on analysis type (Table 2). Most of the predictors were associated with postsurgical pain in bivariate analyses but not in multivariable ones. Depression was associated with the widest spread in significance rates between bivariate (16/27) and multivariable analyses (8/30), followed by trait anxiety (12/24 vs 7/29) and catastrophizing (10/19 vs 7/21). Neither the quality of studies nor other study variables, including type of surgery and follow-up duration, could explain such discrepancy. Only state anxiety, optimism and psychological distress were consistently associated with chronic postsurgical pain. High-quality studies seemed to indicate that kinesiophobia and self-efficacy are not associated with postsurgical pain. In addition, multivariable analyses did not corroborate the role of social support as a predictor of postsurgical pain. No high-quality studies supported the association of any other predictor.

Table 2 - Estimates of the association between psychological predictors and postsurgical pain.

	Bivariate				Multivariable
	S	Quality (S)	NS	Quality (NS)	S	Quality (S)	NS	Quality (NS)
Depression	16	M	11	M	8	M	22	M
Trait anxiety	13	M	11	M	7	M	22	M
Catastrophizing	10	M	9	M	7	M	14	L
Kinesiophobia	1	M	2	H	1	M	10	H
Mental health	3	M	0	n/a	4	M	5	M
State anxiety	5	H	1	M	3	H	1	M
Social support	1	M	1	M	1	M	6	M
Psychological distress	1	L	2	L	4	M	2	M
Optimism	3	M	1	M	4	M	2	M
Coping strategies	1	H	2	M	1	H	4	M
Self-efficacy	1	M	2	H	0	n/a	5	H
Surgical fear	2	L	1	M	2	L	2	M
Somatization	1	L	0	n/a	0	n/a	3	L
Expectations	2	L	1	M	3	L	4	M
Illness perception	1	M	0	n/a	0	n/a	2	M
Stress	2	M	1	M	1	M	1	M
Primitive defenses	0	n/a	1	M	1	M	0	n/a
Locus of control	1	H	0	n/a	1	H	0	n/a
Pain disability	1	L	1	L	0	n/a	0	n/a
Perceived injustice	1	L	0	n/a	1	L	0	n/a
Psychiatric diagnosis	1	L	0	n/a	1	L	1	M
Executive functions	1	H	0	n/a	0	n/a	0	n/a
Negative affect	0	n/a	0	n/a	0	n/a	1	H
Positive affect	0	n/a	0	n/a	0	n/a	1	H
Self-esteem	1	M	0	n/a	0	n/a	0	n/a
Anxiety sensitivity	0	n/a	1	M	0	n/a	0	n/a
Pain-related anxiety	1	L	0	n/a	0	n/a	0	n/a
Life satisfaction	0	n/a	0	n/a	0	n/a	1	M
Somatization	0	n/a	0	n/a	0	n/a	1	M
Physical health	0	n/a	0	n/a	1	M	0	n/a

H, high quality; L, low quality; M, medium quality; NS, count of nonsignificant estimates across studies; Quality, overall methodological quality across studies as measured by the QUIPS tool for significant (S) and nonsignificant (NS) estimates; S, count of significant estimates across studies.

3.3 Quantitative synthesis

3.3.1 Predictors included in the quantitative synthesis

Estimates of the predictive role of depression, trait anxiety, state anxiety, catastrophizing, kinesiophobia, mental health, optimism, and self-efficacy were meta-analyzed. We decided not to pool data from studies evaluating the variables coping, social support, surgical fear and expectations. Studies evaluating coping included different coping strategies (eg, “ability to decrease pain” and “active coping”^1,3). Social support was measured using different modalities, including questionnaires, nonvalidated single questions, and marital status.^11,20,25 Similarly, surgical fear and expectations were assessed using validated and nonvalidated questionnaires or single questions.^19,29,33,45 In all these cases, there were insufficient studies to analyze their operationalizations separately.

3.3.2. Meta-analysis—depression

Twenty-four studies were included in the analysis of the effect of depression on chronic postsurgical pain (n = 17,861). Five estimates were missing due to nonsignificance and were imputed. The overall effect was r = 0.16 (95% CI = 0.12-0.19 P < 0.001) (Fig. 3). Heterogeneity among the studies was low (I² = 0.44) and there were no differences between estimates based on outcome type (t = 0.346, P = 0.732). The mixed-effect model meta-analysis did not reveal any significant fixed effects. Visual inspection of the funnel plot revealed no asymmetry attributable to publication bias (see supplemental digital content, available at http://links.lww.com/PAIN/B113). The trim-and-fill method revealed that adding 2 missing estimates on the right side would lead to a slightly stronger overall effect (estimated r = 0.17, observed r = 0.16).

Figure 3.: Meta-analysis on the effect of depression on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

3.3.3. Meta-analysis—trait anxiety

Estimates of the role of trait anxiety on chronic postsurgical pain were drawn by 16 studies (n = 11,647). Only 2 nonsignificant estimates were missing and were imputed. The random-effect meta-analysis detected a weak effect of trait anxiety, with an estimate of r = 0.13 (95% CI = 0.09-0.17, P < 0.001) (Fig. 4). Heterogeneity between studies was medium (I² = 0.57). Test for the moderating effect of outcome type did not reach the significance threshold (t = −0.33, P = 0.75) and no fixed effect helped to explain between-study heterogeneity.

Figure 4.: Meta-analysis on the effect of trait anxiety on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

Inspection of the funnel plot suggested presence of publication bias. This was confirmed by the trim-and-fill method, which revealed that 2 missing studies were expected on the left side of the plot, whose inclusion would have lead to a slightly lower effect (estimated r = 0.11, observed r = 0.13) (see supplemental digital content, available at http://links.lww.com/PAIN/B113).

3.3.4. Meta-analysis—state anxiety

Five studies reported estimates of the effect of state anxiety on chronic postsurgical pain (n = 950). Only one nonsignificant result was imputed. The overall impact of this variable on chronic postsurgical pain was r = 0.24 (95% CI = 0.14-0.33, P < 0.001) (Fig. 5). Heterogeneity between studies was low (I² = 0.09). Outcome type did not affect the results (t = −2.61, P = 0.08). The mixed-model meta-analysis did not reveal any significant effect. The funnel plot was symmetrical and the trim-and-fill method did not suggest that missing estimates were likely (see supplemental digital content available at http://links.lww.com/PAIN/B113).

Figure 5.: Meta-analysis on the effect of state anxiety on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

3.3.5. Meta-analysis—catastrophizing

Twenty-one studies (n = 4119) were included in this analysis, including 6 studies with nonsignificant missing data. The overall effect was r = 0.19 (95% CI = 0.11-0.27, P < 0.001) (Fig. 6). No differences were noted between studies using dichotomous and continuous estimates of postsurgical pain (t = −1.46, P = 0.16). High heterogeneity was detected (I² = 0.77). No fixed effect could explain a significant portion of such heterogeneity. Visual inspection of the funnel plot revealed the presence of a slight asymmetry on the right side of the plot, indicating that missing estimates would increase the results (see supplemental digital content, available at http://links.lww.com/PAIN/B113). This was confirmed by the trim-and-fill method (estimated r = 0.24, observed r = 0.19; number of trimmed studies = 3).

Figure 6.: Meta-analysis on the effect of catastrophizing on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

3.3.6. Meta-analysis—kinesiophobia

This analysis included the estimates from 7 studies (n = 740). Two studies did not report nonsignificant data and their estimates were imputed. The overall effect was r = 0.13 (95% CI = 0.04-0.22, P < 0.001) (Fig. 7). Heterogeneity was low (I² = 0.02) and there were no differences between dichotomous and continuous estimates of the effect of kinesiophobia on presence of pain and on pain intensity (t = −0.03, P = 0.97). No significant moderators were found in the mixed-effect model meta-analysis. The funnel plot revealed absence of publication bias (see supplemental digital content, available at http://links.lww.com/PAIN/B113).

Figure 7.: Meta-analysis on the effect of kinesiophobia on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

3.3.7. Meta-analysis—mental health

The effects of mental health on chronic postsurgical pain were analyzed by 5 studies (n = 1103). No estimate was imputed. The effect of this predictor was r = −0.17; 95% CI = −0.23 to −0.11, P < 0.001) (Fig. 8). Heterogeneity was absent (I² = 0.00) and the effect of outcome type was not significant (t = −1.91, P = 0.15). The mixed model meta-analysis was not performed because there was no heterogeneity. The funnel plot revealed a slight left asymmetry, but adding the potentially missing estimates caused only a slightly stronger effect of mental health (estimated r = −0.18, observed r = −0.17).

Figure 8.: Meta-analysis on the effect of mental health on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

3.3.8. Meta-analysis—optimism

Four studies analyzed the predictive effect of optimism in postsurgical pain (n = 1444). No estimate was missing due to nonsignificance. The overall effect was nonsignificant (r = −0.12; 95% CI = −0.25 to 0.01, P = 0.07) (Fig. 9). High heterogeneity was detected (I² = 0.81).

Figure 9.: Meta-analysis on the effect of optimism on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

Neither outcome type (t = −0.04, P = 0.97) nor other fixed effects could explain significant portions of such heterogeneity. Visual analysis of the funnel plot revealed the presence of 2 distanced groups of estimates (see supplemental digital content, available at http://links.lww.com/PAIN/B113). This could be attributable to heterogeneity. The trim-and-fill method revealed that adding potentially missing estimates would lead to computing a slightly weaker effect (estimated r = −0.08, observed r = −0.12).

3.3.9. Meta-analysis—self-efficacy

Four studies tested self-efficacy as a predictor of postsurgical pain (n = 1030); one of them did not report nonsignificant estimates. Because one study included estimates from data points distanced more than 12 months, we included both of its effect sizes.⁴⁵ The meta-analysis showed that self-efficacy had a very low association with postsurgical pain, with an estimated effect of r = −0.08 (95% CI = −0.14 to −0.01, P = 0.02) (Fig. 10). Heterogeneity was negligible (I² < 0.001). Outcome type was constant across the studies, and no subgroup analysis was performed. The mixed-effect model meta-analysis was not performed because the number of studies was too limited. The funnel plot did not reveal the presence of asymmetry.

Figure 10.: Meta-analysis on the effect of self-efficacy on postsurgical pain. “Estimates are Pearson's r, lower and upper limits of 95% confidence intervals are also reported. Imputed estimates are represented in gray in the forest plot. Abbreviations: FU = Follow-up duration; LL = Lower Limit of the confidence interval; UL = Upper limit of the confidence interval”

4. Discussion

The aim of this systematic review and meta-analysis was to assess the role of psychological and psychosocial factors in predicting presence and intensity of chronic postsurgical pain. The results showed that depression, state anxiety, trait anxiety, catastrophizing, mental health, kinesiophobia, and self-efficacy have a weak, but significant, association with these outcomes.

The quantitative synthesis showed that state anxiety is the most explicative predictor among psychological factors. This was surprising because this factor is generally overlooked in favor of more persistent characteristics such as catastrophizing and trait anxiety.^16,37 This could be due to the specificity of postsurgical pain. Unlike other forms of chronic pain, this condition evidently is linked to a specific event and could be therefore more affected by the emotional state of the patient. Hypothetically, presurgical state anxiety might trigger a worse experience in the perioperative period and enhance exaggerated responses to perceived threats. More stable variables that were found to be associated with chronic postsurgical pain such as catastrophizing, depression, and mental health could exert their influence when activated by the outcome of surgery (eg, presence of a complication or unexpected pain). According to the fear-avoidance model, these factors contribute to a vicious circle linking disuse of the painful body component, more disability and, in turn, psychological distress.⁴² The weak effects found in the meta-analysis could reflect the fact that activation of this vicious circle does not depend only on the presurgical characteristics of the patient. This might also explain why other stable characteristics such as trait anxiety, kinesiophobia, self-efficacy, and optimism had a weaker or absent role on the development of chronic pain.

It is worth noting that demographic and clinical predictors have similar, or even weaker, associations with chronic postsurgical pain compared to those of the predictors included in this review.^19,43 This is due to the complexity of this condition, which should be understood as the result of the simultaneous interaction between different and heterogeneous predictors.

The overall effects of psychological predictors are weaker than those computed by Jackson et al.¹⁶ A potential explanation for this difference might be that the authors combined various psychological predictors in the category of “emotional distress,” included noncancer and cancer conditions, and considered both acute and chronic postsurgical pain as outcomes. In addition, this study used a more extensive search and multiple imputation techniques to account for nonsignificant missing data. These choices could have led to more precise, but weaker, estimates.

Both the quantitative synthesis and the narrative synthesis of the results might help to shed light on the presence of conflicting evidence about predictors of postsurgical pain. Most of the predictors had only weak associations with chronic postsurgical pain. This might explain why factors such as depression, trait anxiety, and pain catastrophizing are significantly associated with chronic postsurgical pain in >50% of the bivariate analysis and in <50% of the multivariable ones. Lower rates of significant results in multivariable analyses were expected because estimates were adjusted for potential confounding factors, including presurgical pain and demographic characteristics. However, other aspects should be considered to explain these results. Most of the studies included in this review had the aim to test a high number of pathophysiological, clinical, and psychological factors to find the most explicative ones. As a result, in each study, a different set of predictors competed for the explanation of outcome variance and the significance of each predictor was dependent on: (1) other predictors included in the analysis, (2) strength of their effects (3) interactions between these effects, and (4) sample size.

Psychosocial predictors were rarely assessed by the included studies. Surprisingly, the present review identified only the perception of social support as a potential predictor. This is not attributable to the comprehensiveness of the search, and it is probably due to lack of studies in this field. Variables such as quality of close relationships, type of support from caregivers, and work-related variables have shown significant associations with other types of chronic pain.^3,17,33,37 Longitudinal investigations are therefore needed to address the role of these factors on chronic postsurgical pain.

These findings have multiple research and clinical implications. Knowledge of modifiable predictors of postsurgical pain has a great importance for patient education and for the early identification of patients at risk for poor outcomes. During the preoperative period, the assessment of the emotional state of the patient could be useful for this purpose. Because the association with postsurgical pain is of weak intensity, the presence of high levels of state anxiety or catastrophizing should not be used for patient selection. Rather, they could represent the primary target of preoperative psychological interventions. Several specific and brief presurgical treatments are available, ranging from group classes that combine education with skill acquisition to cope with distressing sensations and thoughts, to proper mindfulness-based stress reduction treatments.^8,46

The main limitation of this study is the fact that overall estimates from meta-analyses were based on heterogeneous sets of data. Studies were different in terms of samples, sample characteristics, measurement instruments used to assess both predictors and outcomes, follow-up duration, and statistical analyses. As a result, estimates must be interpreted with caution even in the cases where heterogeneity was not detected by statistical tests. In addition, it was not possible to explain the heterogeneity among the estimates of the predictive effects of catastrophizing, trait anxiety and optimism, which might be due to moderators not addressed by this study or by the interaction of multiple moderators.

At the same time, several study strengths should be acknowledged. The comprehensiveness of the search helped to include a high number of studies and to perform the analyses on more than 40,000 subjects while focusing on noncancer pain (compared to fewer than 10,000 in other reviews on predictors of postsurgical pain^16,37). This helped us also to reach conclusions about predictors that were rarely addressed by existing reviews. The simultaneous use of a narrative synthesis and meta-analytic methods helped to overcome the limitations of each procedure and to address both the consistency of significant and nonsignificant estimates in bivariate and multivariable analyses and their pooled effect. In addition, various methods were used to overcome biases that are typical in this field. Among them, the use of imputation methods helped to counter the presence of nonsignificant missing estimates, which was regarded as a threat for the interpretation of pooled results by other authors.¹⁶

In conclusion, this study showed that psychological predictors have a significant association with postsurgical pain and that state anxiety is the most explicative one. Therefore, a preliminary assessment of this factor could help to identify patients at risk for unfavorable surgical outcomes and could guide specific early treatment. Catastrophizing, trait anxiety, depression, and kinesiophobia are other significant predictors of postsurgical pain and could be addressed in synergy with state anxiety.

Conflict of interest statement

The authors have no conflicts of interest to declare.

Appendix A. Supplemental digital content

Supplemental digital content associated with this article can be found online at http://links.lww.com/PAIN/B113.

Supplemental video content

A video abstract associated with this article can be found at http://links.lww.com/PAIN/B155.