The Relationship between Cognitive and Emotional Factors and Healthcare and Medication Use in People Experiencing Pain: A Systematic Review

Abstract

Pain conditions are among the leading causes of global disability, impacting on global healthcare utilization (HCU). Health seeking behavior might be influenced by cognitive and emotional factors (CEF), which can be tackled by specific therapies. The purpose of this study was to systematically review the evidence concerning associations between CEF and HCU in people experiencing pain. Three databases were consulted: PubMed, Web of Science and EconLit. Risk of bias was assessed using the Downs and Black Checklist (modified). A total of 90 publications (total sample n = 59,719) was included after double-blind screening. In people experiencing pain, positive associations between general anxiety symptoms, depressive symptoms and catastrophizing and pain medication use were found. Additionally, there appears to be a relationship between general anxiety and depressive symptoms and opioid use. Symptom-related anxiety and psychological distress were found to be positively related with consulting behavior. Last, a positive association between use of complementary and alternative medicine and level of perceived symptom control was confirmed in people with pain. For other relationships no evidence or inconsistent findings were found, or they were insufficiently studied to draw firm conclusions, indicating that more research on this topic is needed.

1. Introduction

Pain is one of the most reported symptoms [1] and the second most common reason for consulting primary healthcare [2], implying a strong contribution to the global burden of disease [3,4]. Pain conditions are among the leading causes of global disability, in particular low back pain and headache disorders as these are the 2 leading causes of years lived with disability according to the Global Burden of Disease Project 1990–2017 [3,4,5,6]. This entails that pain is impacting on global healthcare utilization (HCU) and productivity loss [7], and especially for chronic pain, this is resulting in high socioeconomic burden due to excessive HCU [7,8].

The International Association for the Study of Pain (IASP) defines pain as “An unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage” [9]. Pain is not only a unique individual experience in terms of somatosensory characteristics (e.g., different intensity, spread and duration), but also the pain-related cognitive and emotional processes (e.g., pain catastrophizing, hypervigilance and fear) are unique and context-specific to each individual [10,11]. These differences in pain experience impact the, again unique, behavioral actions that people take in response to pain and the influence that pain has on daily life in general [10]. For example, not everyone suffering from pain will use healthcare services. Hence, HCU can be seen as a behavioral action [12], as proposed in the “Behavioral Model of Health Service Use” of Andersen [13].

Although it might seem logic that the propensity to seek care for pain is mainly determined by pain-specific characteristics, this is often not the case as other factors beyond the intensity or duration of the pain episode may be even stronger predictors [12]. According to Andersen’s model, health seeking behavior, which drives HCU, is mediated by predisposing (e.g., sex, age, cultural and social factors), enabling (e.g., access to care, financial factors) and need (e.g., patient and provider’s experience) factors [13]. People’s cognitions and beliefs toward their health status, including their pain symptoms, can be categorized under both predisposing and need factors and are therefore an important component of the model [14]. Additionally, other cognitive and emotional factors (CEF) possibly co-existing with pain symptoms, such as symptoms of depression and anxiety but also catastrophizing and kinesiophobia, may impact people’s need to seek healthcare. Several studies confirmed the presence of maladaptive CEF in at least subgroups of different populations experiencing pain, e.g., the presence of depressive symptoms in low back pain [15,16] and fibromyalgia [17]; the presence of kinesiophobia [18,19,20] and pain catastrophizing [18,20] in post-lumbar surgery patients and the presence of anxiety, depressive symptoms and pain catastrophizing in people suffering from osteoarthritis [21]. Moreover, the relationship between CEF and HCU in people experiencing pain has been suggested numerous times, for example, high levels of catastrophizing have been found to be associated with higher levels of HCU [7,22,23,24,25]. Moreover, Hirsch et al., (2014) [26] found that patients with low back pain and strong beliefs that activity causes pain (i.e., a kinesiophobic way of thinking) have higher HCU and costs, compared to patients with more positive cognitions and attitudes toward physical activity. Furthermore, associations between illness perceptions and HCU were found in a variety of populations, including people with pain [27,28,29,30].

Cognitive and emotional factors are often modifiable by targeted therapies, such as pain neuroscience education and/or cognitive-behavioral therapy [31,32,33]. Given a potential relationship between maladaptive CEF and excessive HCU in patients experiencing pain, interventions specifically targeting CEF could possibly lead to decreases in the need of seeking healthcare, which could imply a reduction of the socioeconomic burden related to pain. A first step toward accomplishing such a socioeconomic benefit would be to identify the cognitive and/or emotional factors that are most likely to affect HCU in this population. Subsequently, specific interventions which are targeting those particular factors can be developed and/or implemented. To the best of our knowledge, the current literature is lacking an extensive overview of those CEF that are associated with HCU in patients experiencing pain.

To address the above outlined knowledge gap, the objective of this systematic review is to answer the question whether a relationship can be confirmed between CEF and HCU, in terms of both amount of HCU and use of different types of healthcare, in people experiencing pain, by systematically reviewing and synthetizing the available literature. It was hypothesized that maladaptive CEF would be positively related with higher amounts of healthcare services and medication use and the use of different types of healthcare, whereas positive CEF were expected to be inversely related with HCU outcomes.

2. Methods

2.1. Protocol and Registration

This systematic review is reported in accordance with the PRISMA statement (Preferred Reporting Items for Systematic reviews and Meta-Analysis) [34]. The protocol was registered a priori in PROSPERO under the following registration number: CRD42018104980.

2.2. Search Strategy

All authors contributed to the development of the search strategy based on their own expertise. The final search (6 August 2019) was conducted by EH in 3 electronic databases: PubMed, Web of Science and EconLit. No limits were applied to the search.

The research question was composed according to the PICO (Population-Intervention-Comparison-Outcome) model [35,36]: “Is there a relationship between CEF (Outcome 1) and HCU (Outcome 2) in people experiencing pain (Population)?” The components “Intervention” and “Comparison” were not relevant for our research question and therefore not defined. The final search strategy was built by combining both free and MeSH terms. Within each separate part of the PICO model, i.e., “Population”, “Outcome 1” and “Outcome 2”, search terms were combined using the Boolean term OR. Between the complete search terms for “Population”, “Outcome 1” and “Outcome 2”, the Boolean term AND was used. The complete search strategy for PubMed can be found in Table A1 (Appendix A). After determining the search string for PubMed, it was adapted for each individual database.

2.3. Eligibility Criteria

Full-text observational studies and arms of (quasi-)experimental studies investigating the relationship between CEF and HCU in patients experiencing pain were considered for inclusion in this systematic review.

Cognitive and emotional constructs eligible for inclusion included, but were not limited to, anger, general anxiety symptoms (i.e., general emotion characterized by apprehension and somatic symptoms of tension in which impending danger, catastrophe or misfortune is anticipated [37], not specifically due to the experience of somatic symptoms), symptom-related anxiety symptoms (i.e., anxiety symptoms due to or concerning somatic symptoms, e.g., pain), catastrophizing, depressive symptoms, fear-avoidance beliefs, illness beliefs, psychological distress, stress, self-compassion, symptom vigilance, pain acceptance, perceived symptom control and self-efficacy beliefs. To be suitable for inclusion, these constructs had to be measured using patient-reported instruments. Studies using instruments specifically designed for the diagnosis of psychiatric conditions (e.g., PRIME-MD, Anxiety Disorders Interview Schedule for DSM-IV) were excluded.

Healthcare utilization had to be measured in terms of amount of healthcare used (either for a particular type of HCU or for HCU in general) or presence/absence of a certain type of HCU. Studies only reporting healthcare costs were excluded.

Participants had to be adults (≥18 years old) experiencing some form of pain (acute, subacute or chronic). If (part of) the sample was not experiencing pain, and no subgroup analysis in people with pain was executed, the study was considered not suitable for inclusion. Moreover, papers studying children, patients with dementia, pain during labor or during surgical procedures were excluded.

Full eligibility criteria can be found in Table 1.

Table 1. Eligibility criteria.

Inclusion	Exclusion
Full text (arms of) (quasi-)experimental studies or observational studies	Case reports, systematic reviews and meta-analyses, narrative reviews, letters to the editor, expert opinions, conference abstracts, studies without available full-text version
English, French or Dutch written	Other languages
Evaluation of the relationship between CEF, including but not limited to, anger, anxiety symptoms, catastrophizing, depressive symptoms, fear-avoidance beliefs, illness beliefs, psychological distress, stress, self-compassion, symptom vigilance, pain acceptance, perceived symptom control and self-efficacy beliefs, and HCU	No investigation of the relationship between CEF and HCU
CEF assessed by means of patient-reported instruments	Instruments specifically designed for physicians to diagnose psychiatric conditions (e.g., PRIME-MD, Anxiety Disorders Interview Schedule for DSM-IV)
HCU reported in terms of amount of HCU (of a particular type of HCU or of HCU in general) or in terms of type of healthcare services used (absence/presence of certain types of HCU)	Studies only reporting healthcare costs, without mentioning utilization, those only investigating adherence to recommendations, medication misuse or substance abuse for non-medical purposes and studies concerning the use of assistive or ergonomic devices (e.g., prosthesis, orthosis and canes)
Participants had to be adults (≥18 years old) experiencing either acute, subacute or chronic pain.	Complete or part of the sample was not experiencing pain and no separate analysis for people with pain was executed. Studies on children, women experiencing labor pain, people suffering from dementia, intraoperative subjects and palliative patients.
Studies reporting a quantified association or relationship analyzed by using statistics.	Studies only reporting observations without quantitative analysis or studies only including qualitative analyses.

CEF: cognitive and emotional factors; HCU: healthcare use.

2.4. Study Selection

Following de-duplication, all retrieved articles were screened for title and abstract by 2 reviewers independently (EH and LL) using Rayyan online software [38,39]. Subsequently, the same 2 reviewers performed the full text screening independently from each other. Percentage agreement was calculated to assess inter-rater reliability. Discrepancies were discussed after each stage of the screening in a consensus meeting with both reviewers and a third independent reviewer (IC).

2.5. Data Extraction

The a priori determined data extraction form included the following items: author, year of publication, country, study design, type of population (including pain duration), sample size (including sex distribution and age), outcome measures for CEF and HCU (including the moment of assessment if relevant), objective relevant for the present systematic review, statistics used to investigate the relationship between both outcomes and main findings (including numerical data, e.g., effect sizes, if reported).

Data extraction was performed by the first reviewer (EH) and checked for correctness by the second (LL) and last author (IC). Any discrepancies were discussed in a consensus meeting with all 3 reviewers.

2.6. Risk of Bias Assessment

A modified version of the Downs and Black checklist [40] (see Supplementary Material Document S1) was used for risk of bias assessment. This instrument was specifically designed for the risk of bias appraisal of different study designs, including randomized and non-randomized trials and observational studies [40], making it an appropriate risk of bias checklist for the variety of study designs included in this systematic review.

To further increase its suitability for estimating risk of bias for the specific objective of this systematic review, the original checklist was modified. Such modifications have been done before in previous systematic reviews [41,42,43]. The answer option “Not applicable” was added to several questions, as for some study designs particular questions were not applicable, resulting in a different total score depending on the study design. Original question 8 (“Have all important adverse events that may be a consequence of the intervention been reported?”) was omitted from the instrument as our research question is not focusing on an intervention, and therefore, there is no specific interest in potential adverse events. One additional question focusing specifically on the assessment of HCU was added to the section “internal validity—bias”. The question reads as follows: “Was healthcare utilization primarily registered for scientific research?” with answer options (1) Yes, (2) No and (3) Unable to determine. This was deemed important to consider as subtracting HCU data from for example clinical patient files or secondary databases may imply a higher risk for (coding) errors. The last question of the original checklist concerning the power of the reported results was adapted because of the unclarity of the original question. The adjusted question now reads: “Was an a priori sample size calculation performed, and was the anticipated sample size reached, or was a post hoc power analysis performed, which suggested that the results were sufficiently powered?” with answer options (1) Yes; (2) No and (3) Unable to determine.

For further data synthesis (see Section 2.7), all included studies were categorized as presenting low, moderate or high risk of bias by evaluating 3 criteria based on the results of the modified Downs and Black checklist for each study. These criteria were selected during a consensus meeting with EH, IC, LL and DB and were deemed to be the most relevant for estimating risk of bias for the present review. The 3 selected criteria were (1) use of reliable and valid outcome measures, (2) clear reporting of the study results and (3) the study results were generalizable. Each criterium was scored using a color code with, green (the study met the criterium), orange (partly meeting the criterium) and red (the criterium was not met). The first criterium was scored based on item numbers 2 (i.e., the main outcomes were clearly described), 19 (i.e., the outcome measures used were valid and reliable) and 20 (i.e., HCU was primarily registered for scientific research) of the modified Downs and Black checklist. Criterium number 2 was scored based on items 6 (i.e., clear description of the study findings) and 7 (i.e., reporting of estimates of random variability for the main outcomes) of the modified Downs and Black checklist. For the appraisal of the third criterium on generalizability, both items of the Downs and Black checklist (10 and 11) on patient representativeness were taken into account, together with item number 3 (i.e., clear description of the study sample). Studies scored green on a criterium if the study met all respective Downs and Black items for that criterium, orange if 1 item was not met and red if 2 or more items were not met. If a study scored green on all 3 criteria, risk of bias was deemed to be low; if there was uncertainty or absence of 1 out of 3 criteria, risk of bias was scored moderate; all other scenarios were scored high risk of bias.

The appraisal was performed by 2 reviewers (EH and LL) independently. Percentage agreement between both reviewers was calculated to assess inter-rater reliability. Discrepancies were discussed in a consensus meeting with both reviewers (EH and LL) and a third independent reviewer (IC).

2.7. Data Synthesis

All extracted data were categorized and synthetized in summarizing evidence tables with their accompanying explanatory results tables. For the reporting of study characteristics and details on the methodology of the included studies, a separate study characteristics table was created, containing the following items: author, year of publication, country, study design, population and duration of pain, sample size, sex and age of the sample, outcome measure(s) for CEF and HCU, moment of assessment, objective of the study relevant for the present systematic review and statistical analysis used.

Categorization was executed on the level of the individual outcome measures reported in a study; therefore, one study could be mentioned under different categories for HCU and/or CEF. The categories were determined based on the expertise of all co-authors and by consultation of additional experts in the field of psychology.

Two main categories of HCU outcomes were identified: (1) amount of HCU and (2) type of HCU. Healthcare use outcome measures were categorized under the first category if they measured the number or frequency of visits, treatments received, hospitalizations or medications used, the length of stay in healthcare facilities, or if the presence of HCU in general was reported without mentioning any particular type of HCU. The second category, “type of HCU”, contains HCU outcome measures reporting on the presence or absence of use of specific healthcare providers, services or medication types, without reporting anything about the amount of healthcare used. A separate summarizing evidence and results table was created for both categories (vide infra).

Within the main categories for HCU, further subcategories were created using a systematic approach. First, categorization was executed by the primary reviewer (EH). Next, all authors were granted the opportunity to provide their feedback. Last, a consensus meeting was organized between EH, LL and IC to discuss the remaining discrepancies and agree upon the final categorization. “Amount of HCU” contained the following subcategories (n = 7): amount of pain medication use (i.e., over-the-counter (OTC) and prescription pain medication use), consultations (i.e., all types of consultations with healthcare providers, excluding complementary and alternative medicine (CAM) visits and emergency visits), emergency HCU (i.e., visits to the emergency department and other unscheduled emergency consultations), hospitalizations (i.e., length of stay and number of individual hospitalizations), CAM use, invasive procedures (i.e., surgeries and other invasive interventions) and HCU in general (in case the study did not make any further specifications). For “Type of HCU”, outcome measures were categorized into the following subcategories (n = 12): pain medication (in case no further specification was made in the original article concerning the type of medication (i.e., OTC/prescription/opioids)), OTC pain medication, prescription pain medication (excluding opioids), opioids, consultations (in case the study did not make any further specifications regarding the level of care (i.e., primary/secondary/tertiary care)), primary care consultations, secondary care consultations, tertiary care consultations, emergency HCU, invasive procedures, hospital admissions and CAM use. A complete overview of the clustering of all HCU outcome measures reported in the included studies can be found in Supplementary Material Table S2.

Cognitive and emotional factors were clustered into 19 different construct groups using the same approach as described for the subcategorization of HCU outcome measures (Supplementary Material Table S3). Additionally, 2 experts in the field of psychology were contacted to provide their feedback about the clustering. The clustering process finally resulted in 15 maladaptive CEF clusters (anger, general anxiety symptoms, symptom-related anxiety symptoms, catastrophizing, depressive symptoms, fear-avoidance beliefs, frustration, health worry, helplessness, negative consequences of symptoms beliefs, negative illness beliefs, psychological distress, stress, symptom vigilance and thanatophobia), 8 positive CEF clusters (illness coherence, pain acceptance, perceived benefits, perceived symptom control, positive mood, psychological flexibility, self-compassion and self-efficacy beliefs) and 3 were CEF for which it was impossible to classify them as being either maladaptive or positive (health attribution, locus of control and perceived cause of symptoms). Maladaptive constructs were expected to be positively related with amount of HCU and presence of different types of HCU, whereas positive constructs were expected to be inversely related with amount of HCU and presence of different types of HCU.

After clustering, a detailed results table was created for each category of HCU outcomes (“amount of HCU” and “type of HCU”) presenting the main results for each analysis of a potential association between a particular HCU outcome and CEF. The table was structured based on the different combinations of a particular HCU subcategory and CEF cluster between which an association was investigated in the literature. For each analysis, the following items were included in the results table: author and year of publication of the original publication, sample size, considered population, CEF and HCU outcome measures, description of how the association was analyzed (including potential other factors considered for multivariate analyses) and main findings (i.e., effect size and significance level, if reported). Subsequently, summarizing tables were created for both categories of HCU outcome measures presenting all investigated associations between a particular subcategory of HCU and CEF cluster and their outcome in terms of a positive, negative or no association, in a more consumable way.

Additionally, a summarizing level of association score was assigned to each investigated association, based on the proportion of analyses reporting a positive, negative or no association and accounting for the risk of bias evaluation. The methodology for this summarizing appraisal was adopted from the method used by Sallis et al., (2000) [44], Van der Horst et al., (2007) [45], Hinkley et al., (2008) [46] and Lubans et al., (2010) [47]. Specifically, if 0–33% of analyses reported a significant association, the result was classified as “no association” (0); if 34–59% of analyses reported a significant association, or if fewer than 4 studies investigated that particular relationship, the result was classified as “inconsistent” or “uncertain”, respectively (?); and if ≥60% of the analyses reported a significant positive or negative association, the result was classified as “positive” (+) or “negative” (−), respectively, based on the direction of association. To account for risk of bias, the method of Lubans et al., (2010) [47] was followed, if after exclusion of high risk of bias studies the association (+/−) or absence of an association (0) was still supported by, respectively, ≥60% or 0–33% of the analyses reporting a positive or negative association, the summary score was up-/downgraded to ++/−−/00.

3. Results

3.1. Study Selection

The systematic search resulted in 3543 unique studies to be considered for screening, of which 90 were included in this systematic review (Figure 1). Percentage agreement between both reviewers for title and abstract screening and full text screening were 92% and 80%, respectively. Reasons for exclusion were ineligible outcome (n = 1661; e.g., studies using diagnostic tools instead of patient-reported CEF assessment tools and studies only reporting about healthcare costs), ineligible study design (n = 858; e.g., systematic/narrative reviews and letters to the editor), ineligible population (n = 855; e.g., pediatric populations and study samples where not all participants were experiencing pain), ineligible language (n = 49) and no full text available (n = 30).

Figure 1. PRISMA flowchart.

3.2. Study Characteristics

In terms of study designs, 2 randomized controlled trials and 47 cross-sectional, 38 cohort and 3 case-control studies were included, comprising a total sample of 59,719 subjects. A complete overview of the characteristics of the included studies can be found in Table A2 (Appendix B).

3.3. Risk of Bias Assessment

Sum scores on the Downs and Black checklist ranged between 8/16 and 15/16 (mean: 12.3/16) for cross-sectional studies (n = 47) [7,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93], 10/18 and 17/18 (mean: 12.5/18) for observational cohort studies (n = 26) [94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119], 7/21 and 16/21 (mean: 13.4/21) for single-group interventional cohort studies (n = 11) [25,120,121,122,123,124,125,126,127,128,129], 12/19 and 15/19 (mean: 13.3/19) for case-control studies (n = 3) [130,131,132] and 13/27 and 19/27 (mean: 16/27) for randomized controlled trials (n = 2) [133,134] and multiple-group cohort studies (n = 1) [135]. Percentage agreement between both reviewers was 93%. Most of the studies did not report on an a priori sample size calculation. Another main limitation was the unclarity about the representativeness of the study results for the target population. The complete risk of bias assessment can be found in Table A3 (Appendix C).

Based on the 3 most important criteria to estimate risk of bias for this systematic review, 18 studies showed low risk of bias, 46 moderate risk of bias and 26 high risk of bias. All but 30 studies appeared to have used valid and reliable outcome measures, and all but 7 studies reported their results in a clear and accurate way. The generalizability of the results was not entirely clear in 49 studies, while results did not appear to be generalizable in 16 studies. The remaining 25 studies were deemed to have generalizable results for their target population. The results of the summarizing risk of bias assessment for each study can be found in Table 2.

Table 2. Summarizing risk of bias assessment.

	Reliable and Valid Outcome Measures 1	Reporting of Results 1	Generalizability of Results 1	Risk of Bias (High/Moderate/Low) 2
Alschuler (2012) [48]				Moderate
Asmundson (2001) [49]				High
Biggs (2003) [50]				Moderate
Boyer (2009) [51]				Moderate
Buse (2012) [94]				Moderate
Carroll (2016) [96]				High
Carroll (2018) [95]				High
Ciechanowski (2003) [25]				Low
Citero (2007) [97]				Moderate
Cronan (2002) [135]				High
Cronin (2018) [93]				High
Cronin (2019) [52]				Moderate
Daltroy (1998) [133]				High
De Boer (2012) [53]				Moderate
Demmelmaier (2010) [98]				Low
Dobkin (2006) [99]				Moderate
Durá-Ferrandis (2017) [134]				Low
Elander (2003) [54]				Moderate
Elander (2014) [55]				Moderate
Engel (1996) [100]				High
Fink-Miller (2014) [56]				High
Gebauer (2019) [101]				High
Gil (2004) [102]				Moderate
Görge (2017) [120]				Moderate
Grant (2000) [57]				Moderate
Hadlandsmyth (2013) [103]				Moderate
Harden (1997) [130]				High
Harding (2019) [58]				Moderate
Hill (2007) [59]				Low
Howell (1999) [60]				Moderate
Huffman (2017) [121]				High
Jensen (1994) [128]				Moderate
Jensen (2006) [122]				High
Jordan (2006) [104]				High
Jöud (2017) [7]				Moderate
Kapoor (2012) [123]				High
Kapoor (2014) [61]				Moderate
Keeley (2008) [105]				Moderate
Kratz (2018) [62]				Moderate
Kuijper (2014) [106]				Moderate
Lee (2008) [63]				Moderate
Lentz (2018) [107]				Low
Levenson (2008) [108]				Moderate
Lozano-Calderon (2008) [131]				Low
Lozier (2018) [64]				Moderate
Macfarlane (1999) [65]				Moderate
Macfarlane (2003) [66]				Moderate
Mann (2017) [67]				Moderate
Mannion (2013) [68]				Moderate
McCracken (1997) [69]				Moderate
McCracken (2005; Pain) [109]				Low
McCracken (2005; Beh Res Ther) [124]				Low
McCracken (2007) [70]				Moderate
Mourad (2016) [72]				Moderate
Mourad (2018) [71]				Moderate
Musey (2018) [110]				High
Navabi (2018) [111]				High
Ndao-Brumblay (2010) [73]				High
Newman (2018) [74]				High
Nielsen (2015) [75]				Moderate
Osborne (2007) [129]				Moderate
Pagé (2019) [112]				Low
Philpot (2018) [125]				High
Pierce (2019) [76]				High
Primavera (1994) [127]				High
Rosenberg (2008) [77]				Moderate
Shmagel (2016) [78]				Moderate
Talley (1998) [79]				Low
Thorstensson (2009) [80]				Low
Torrance (2013) [81]				Low
Trask (2001) [82]				Moderate
Tremblay (2018) [113]				Moderate
Tsuji (2019) [83]				Low
Ullrich (2013) [114]				Moderate
Valdes (2015) [84]				Moderate
van Tilburg (2008) [115]				Low
Vervoort (2019) [116]				Moderate
Villani (2010) [85]				High
Vina (2019) [86]				Low
Von Korff (1991) [87]				High
Von Korff (2007) [132]				High
Walker (2016) [88]				Moderate
Wideman (2011) [126]				Moderate
Wijnhoven (2007) [89]				Moderate
Williams (2006) [90]				Low
Williams (2018) [117]				High
Wong (2019) [118]				High
Woodhouse (2016) [119]				Moderate
Zebenholzer (2016) [91]				Low
Zondervan (2001) [92]				Low

¹ Each criterium was scored using a color code: green (the study met the criterium), orange (uncertainty about the criterium) and red (the criterium was not met). ² Overall risk of bias score: (1) green on all 3 criteria: low risk of bias; (2) orange or red on 1 criterium: moderate risk of bias; (3) >1 orange or red criterium: high risk of bias.

3.4. Evidence for Associations between CEF and HCU in People Experiencing Pain

Results will be discussed for each relationship between a certain CEF and HCU outcome for which a conclusive result could be formulated (i.e., relationship was investigated 4 times or more in the literature). First, the results for relationships between CEF and amount of HCU in people experiencing pain will be reported, and next, the investigated associations with different types of healthcare used will be outlined. Within each of these paragraphs, relationships with maladaptive CEF were discussed first, followed by the positive CEF and the unclassified CEF.

3.4.1. Associations between CEF and Amount of HCU in People Experiencing Pain

The level of general anxiety symptoms was found to be consistently positively related with the amount of pain medication use in people experiencing pain, based on univariate analyses (4 analyses reporting a positive association [55,75,108,118] and 1 reporting no association [55]—80% agreement for a relationship), whereas it appeared to be unrelated with the number of consultations with healthcare providers (univariate: 1 analysis reporting a positive association [103], 2 a negative [125] and 5 no association [64,98,103,108]—13%; multivariate: 1 analysis reporting a positive association [103], 1 a negative [125] and 5 no association—14%) and the amount of emergency HCU (univariate: 1 analysis reporting a positive association [110] and 3 no association [85,125]—25%).

In people having pain symptoms, analyses for symptom-related anxiety symptoms showed a consistent positive relationship with the amount of consultations with healthcare providers based on univariate analyses (10 analyses reporting a positive association [60,72,92,103,113] and 1 no association [95]—91% agreement), while in multivariate analyses inconsistent results for this association were found (8 analyses reporting a positive association [60,71,72,95,103,113,120] and 8 no association [50,60,71,72,120]—50%).

Univariate analyses showed inconsistent findings for the association between catastrophizing and pain medication use (3 reporting a positive association [55,118,126] and 3 no association [54,55]—50%) and consultations with healthcare providers (4 reporting a positive association [74,123,126], 2 a negative [61,128] and 3 no association [54,98]—44%) in people with pain. For the latter, multivariate analyses showed absence of a direct relationship (7 analyses showing no association [25,61,74,123,126]—0%). Furthermore, no relationship was found between level of catastrophizing and the amount of emergency care used by people experiencing pain (both univariate and multivariate: 4 analyses reporting no association [97]—0%).

Level of depressive symptoms was found to be consistently positively related with the amount of pain medication use in univariate analyses (6 analyses reporting a positive association [55,75,100,108,118,126] and 1 no association [55]—86%) in people with pain; however, multivariate analyses were not able to confirm this relationship (2 analyses reporting a positive association [100,102] and 5 no association [62,102,108,126]—29%). Analyses investigating the relationship between depressive symptoms and the number of healthcare consultations resulted in inconsistent findings (univariate: 13 reporting a positive association [64,67,72,74,83,98,100,108,113,123,126,132], 1 a negative [61] and 11 no association [48,98,123,125]—52%; multivariate: 13 reporting a positive association [61,74,78,83,96,102,114,120,123] and 19 no association [25,48,50,64,67,71,72,100,102,108,113,117,125,126]—41%). Based on univariate analyses, it was unclear whether depressive symptoms were associated with emergency HCU (3 reporting a positive association [67,83,85], 1 a negative [125] and 3 no association [48,108]—43%) and HCU in general (2 reporting a positive association [48,135] and 3 no association [48,64,135]—40%) in people with pain; on the contrary, multivariate analyses showed absence of evidence for a relationship with both emergency HCU (1 reporting a positive association [83], 1 a negative [125] and 5 no association [67,102,117]—14%) and HCU in general (2 a positive [119,120] and 5 no association [48,57,58,135]—29%). Moreover, no evidence was found based on multivariate analyses for a relationship between depressive symptoms and amount and/or duration of hospitalizations in people having pain (1 analysis reported a positive association [83] and 8 no association [52,102,114,117,125]—11%).

There is absence of evidence for a multivariate relationship between negative illness beliefs and fear-avoidance beliefs and the amount of consultations with healthcare providers in people with pain (negative illness beliefs: 1 analyses reporting a positive association [50] and 3 no association [50,120,128]—20%, fear-avoidance beliefs: 1 analysis reporting a positive association [105] and 3 no association [105,120,126]—25%).

Level of psychological distress appeared to be unrelated with the amount of pain medication use based on univariate analyses in people experiencing pain (4 analyses reporting no association [82,91]—0%). With number of healthcare consultations, on the other hand, 100% agreement for a positive relationship was found based on 8 univariate analyses [63,87,88,111]; however, this relationship could not be confirmed by multivariate analyses (1 reporting a positive association [63], 2 a negative [50] and 10 no association [50,87,88,105,106]—8%).

Analyses investigating the relationship between stress (multivariate: 2 reporting a positive association [102,105] and 3 no association [102,105]—40%) and symptom vigilance (univariate: 2 reporting a positive association [69,72] and 2 reporting no association [98]—50%) and the amount of healthcare consultations showed inconsistent findings in people with pain.

Concerning the relationship between HCU and positive CEF in people experiencing pain inconsistent findings for a potential association between pain acceptance and the amount of pain medication use were found (univariate: 3 analyses reporting a negative association [55,109] and 5 no association [55,109,124]—38%; multivariate: 2 a positive [62,109] and 3 no association [62,109]—40%).

Additionally, inconsistent results were found in univariate analyses for a potential association between self-efficacy beliefs and the amount of healthcare consultations (3 reporting a negative association [55,126] and 5 no association [64,98]—38%). Based on multivariate analyses, no evidence for a relationship between the latter was found (1 reporting a negative association [67] and 4 no association [64,126,129]—20%).

There was absence of evidence for a relationship between locus of control and the amount of healthcare consultations in people having pain, based on multivariate analyses (1 reporting a positive association [106], 1 a negative [106] and 4 no association [106]—17%).

All remaining analyses investigating the association between CEF and the amount of healthcare used in people experiencing pain were investigated less than 4 times. All results for associations between CEF and amount of HCU were summarized in Table A4. More comprehensive details on the analyses reported in the literature that investigated this relationship and their results can be found in Supplementary Material Table S4.

3.4.2. Associations between CEF and Type of HCU in People Experiencing Pain

In people experiencing pain, a positive association was found between level of general anxiety symptoms (4 analyses showing a positive association [94,121] and 2 no association [122,130]—67% agreement for a relationship) and depressive symptoms (8 analyses showing a positive association [86,94,96,121,122] and 3 no association [61,74,130]—73%) and using opioids based on univariate analyses. However, for depressive symptoms, multivariate analyses indicated the absence of an association with opioid use (2 analyses showing a positive association [86,101] and 4 no association [62,86,101,121]—33%). Moreover, no evidence was found for a relationship between depressive symptoms and prescription pain medication use (univariate: 2 analyses showing a positive association [48,76] and 16 no association [48,49]—11%), OTC pain medication use (univariate: 1 analyses showing a positive association [48] and 4 no association [48,49]—20%), having primary care consultation (univariate: 2 analyses showing a positive association [112] and 5 no association [48,104]—29%), undergoing invasive procedures (univariate: 1 analysis showing a positive association [48] and 4 no association [48,131]—20%) and using CAM services (univariate: 6 analyses showing a positive association [48,73,115] and 23 no association [48,73,77,112]—21%; multivariate: 5 analyses showing no association [73,115]—0%) in people with pain.

Symptom-related anxiety symptoms were found to be consistently positively related with having primary care consultations in patients with pain (univariate: 3 analyses reporting a positive association [60] and 1 no association [65]—75%).

Multivariate analyses found a consistently positive association between the level of catastrophizing and using pain medication in people experiencing pain (4 analyses found a positive relationship [53,84,89] and 2 no association—67%).

Psychological distress was found to be unrelated with using opioids (multivariate: 4 analyses showing no association [107,111]—0%), having consultations (univariate: 12 analyses showing no association [66,68,79,87,90,91]—0%; multivariate: 1 analysis showing a negative association [87] and 7 no association [68,80,87]—13%), having emergency consultations (multivariate: 1 analysis showing a positive association [88] and 3 no association [88,107]—25%), undergoing invasive procedures (multivariate: 2 analyses showing a positive association [107,111] and 5 no association [107]—29%) and using CAM (univariate: 4 analyses showing no association [82]—0%) in people having pain. Inconsistent evidence was found for the potential relationship between psychological distress and having primary care consultations in people experiencing pain (multivariate: 2 analyses showing a positive association [59,65] and 2 no association [65,80]—50%).

Both univariate and multivariate analyses indicated a positive relationship between perceived symptom control and using CAM services in people with pain (univariate: 3 analyses showing a positive association [73] and 1 no association [73]—75%; multivariate: 3 analyses showing a positive association [73] and 1 no association [73]—75%).

Based on univariate analyses there appears to be no association between self-efficacy beliefs and having secondary care consultations (1 analysis showing a positive association [51] and 3 no association [51]—25%)

All remaining analyses investigating the association between CEF and type of healthcare services used by people experiencing pain were investigated less than 4 times. Summarized results for associations between CEF and type of HCU can be found in Table A5. More comprehensive details on the analyses investigating the relationship between CEF and type of HCU retrieved from the literature and their results can be found in Supplementary Material Table S5.

4. Discussion

4.1. Discussion of the Results

4.1.1. Summary of the Results

The present systematic review investigated whether a relationship between CEF and HCU, in terms of amount of HCU and type of healthcare services used, in people experiencing pain could be identified by synthesizing the existing literature. Based on univariate analyses on amount of HCU outcomes, a positive association between general anxiety symptoms and depressive symptoms and amount of pain medication use and between symptom-related anxiety symptoms and psychological distress and the number of consultations in people with pain could be confirmed. However, based on the results of multivariate analyses, it could be assumed that there is no direct relationship between depressive symptoms and pain medication use and between psychological distress and number of consultations. The level of general anxiety symptoms seems to have neither direct nor indirect relationship with the number of consultations with healthcare providers that people with pain are having. Additionally, no direct or indirect relationship could be confirmed between catastrophizing and the amount of emergency care use. In terms of type of healthcare services used, a univariate positive association was shown between general anxiety symptoms and depressive symptoms and the use of opioids in people with pain. However, for the latter, the relationship might be only indirect as multivariate analyses were not able to confirm the association. Overall, it can be stated that, apart from the univariate relationship with opioid use, depressive symptoms are not firmly related with the use of particular types of HCU. Moreover, psychological distress was found to be unrelated with the use of several healthcare services in people experiencing pain. For catastrophizing, on the other hand, the existing literature confirmed a multivariate positive relationship with using pain medication. Furthermore, a univariate positive association between symptom-related anxiety symptoms and having primary care consultations was found. Last, the literature indicated a positive association between perceived symptom control and the use of CAM services, based on both univariate and multivariate analyses.

Nevertheless, there is clearly an interest in this subject in the available literature, and many associations between particular combinations of CEF constructs and HCU categories were only scarcely studied, making it impossible to draw firm conclusions about the existence of these relationships. Furthermore, when associations were sufficiently studied, the findings were often inconsistent across the studies.

4.1.2. Discussion of Confirmed Associations

Although it can be stated that the literature on the relationship between many of the reported CEF and HCU outcomes is still inconclusive, it is possible to confirm the presence or absence of some associations.

Concerning pain medication use, the literature confirms an at least indirect relationship between the level of general anxiety and/or depressive symptoms and the amount of pain medication used [55,75,100,102,108,118,126,133] and the odds of using opioids [86,94,96,121,122]. Furthermore, the level of catastrophizing was directly related with the odds for using pain medication in general [53,84,89]. The latter are important findings in the light of the current opioid epidemic [136] in Northern America, Canada and Australia [137] and to a lesser extent in European countries, such as Germany and the UK [138]. Although all patients with pain should receive an appropriate treatment, and opioids can be useful in the treatment of short-term acute pain episodes [136], the opioid epidemic is an example of how HCU for pain can become excessive with major negative impact on the individual and society. This is characterized by, for instance, abusive prescription practices of medical practitioners, illicit opioid overdose-related deaths and high socio-economic burden with an estimated economic liability of $78 billion a year in the USA [139], not even including the costs of decreased quality of life, psychological distress and social dysfunction [136,140]. The relationship between the odds for opioid use, and by extension pain medication use in general, and the level of symptoms of depression, general anxiety and catastrophizing suggests that it might be possible to decrease excessive opioid use by managing depressive symptoms better [48]. Moreover, it has been suggested that there might be a positive relationship between symptoms of depression [55,141,142,143,144] and/or anxiety [55,141,142,144] and non-adherence to medication recommendations (e.g., pain medication dependence, overuse or taking someone else’s prescription). Such non-adherence to medication recommendations is one of the causes of the opioid epidemic, indicating that it might be useful to target these CEF in clinical practice in an attempt to decrease medication misuse, withholding great potential for impacting upon the opioid epidemic. Although more research is needed to confirm these assumptions.

Maladaptive CEF were expected to be positively related with consultation behavior. Based on the available literature, we cannot confirm this hypothesis for many CEF, but there is potential for a positive association between symptom-related anxiety symptoms and both the number of healthcare consultations [60,71,72,92,95,103,113,120] and the odds for having primary care consultations [60]. However, it should be stated that this might only be an indirect relationship as only univariate analyses showed a consistent positive association. This can possibly be explained by the notion that patients who are experiencing pain-related anxiety and/or fear often deem their condition to be threatening, which drives them into safety-seeking behavior, such as avoiding symptom-provoking activities [145] and frequent consultations with healthcare providers [71,72,146]. Additionally, there is potential for a positive association between the propensity to have healthcare consultations and catastrophizing [7,89] and psychological distress [59,65,82], although these relationships were only scarcely studied, making it hard to draw firm conclusions about them. The presence of symptoms of anxiety and/or fear and catastrophizing might go hand in hand in some patients experiencing pain, high anxiety and/or fear levels and could lead to more catastrophizing about a painful episode, and catastrophizing could in turn result in more anxiety and/or fear, suggesting a bidirectional relationship [145]. Moreover, it is known that anxiety [147], catastrophizing [147,148,149,150] and psychological distress [151,152,153] can enhance pain intensity and related disability. This augmented pain experience combined with the fact that catastrophizers often view their condition as threatening might lead to a faster decision to consult a healthcare provider.

Remarkably, most of the maladaptive CEF for which positive associations with HCU were found are part of the fear-avoidance model (i.e., depressive symptoms, general anxiety symptoms, symptom-related anxiety/fear symptoms and catastrophizing) [145,154]. This leads to the consideration that the influence of these CEF on pain intensity and disability might also play a role in their relationship with HCU, which has been suggested in the literature before [154]. Although for some factors consistent independent relationships were shown, based on multivariate analyses. Additionally, Alschuler et al., (2012) [48] could not confirm that presence of depressive symptoms had a moderator effect on the relationship between pain intensity and HCU. It should also be considered that the fear-avoidance model was designed based on chronic pain conditions, whereas different types of pain conditions were included in this systematic review. However, when looking at the results for chronic and acute pain conditions separately, we did not find any outstanding differences. Furthermore, it should be taken into account that due to their involvement in the fear-avoidance model, these CEF are possibly more popular in research, making them more extensively studied compared to some of the other CEF included in this systematic review.

Some additional aspects that might have led to the presence or absence of a confirmation of certain associations in particular studies should be considered. In the introduction, it was already mentioned that the modifiable CEF considered in this systematic review can be categorized under the need and/or predisposing factors of Andersen’s model of health services use, next to many other demographic and clinical patient-related characteristics [13,14,155]. Although it would go beyond the scope of this systematic review to go into too much in detail, it is worth considering that the results of the included studies might have been influenced by factors from the third component in van Andersen’s model: enabling factors [12,13,14] [WU1]. These enabling factors can be individual-specific (e.g., income), but many of these factors apply to an entire community, healthcare system or patient population. Therefore, the amount of and whether people seek care can be dependent on the healthcare system in which they are residing, based on for instance, differences in the accessibility [13,14,50,51,68,155,156] and cost of care [155]. This can explain why certain associations between CEF and HCU outcomes might only be present in specific countries with their particular healthcare and/or health security systems.

Another factor that should be considered when interpreting HCU among different conditions and pathologies is the fact that for some health conditions, the healthcare trajectories are more predetermined than for others. Therefore HCU of some people experiencing pain will be more selective and therefore to a greater extent determined by the free will of the patient, while for others care seeking behavior will be highly influenced by the fact that a standard care trajectory is available for their condition [157]. It might be assumed that for the latter CEF will play a less important role in the decision to have care.

4.2. Directions for Future Research

Several potential associations between particular CEF and HCU outcomes were investigated less than 4 times; wherefore, it was deemed impossible to draw firm conclusions about the effective existence of these relationships. However, the limited results for some of them point towards a confirmation of our hypotheses (i.e., ≥60% of the analyses confirm hypothesis). More specifically, concerning maladaptive CEF, there is potential for an, at least indirect, positive relationship between (1) general anxiety symptoms and the propensity to seek emergency care [110] and to use CAM services [115], (2) catastrophizing and the odds of using prescription pain medication [84], using opioids [84,122], having consultations with healthcare providers in general [7,89] and having tertiary care consultations in particular [56], (3) depressive symptoms and having hospitalizations [52], (4) fear-avoidance beliefs and the amount of pain medication use [126] and the chance of having a healthcare consultation [68], (5) frustration and using pain medication [59], (6) health worry and number of consultations with healthcare providers [132], (7) helplessness and the amount of healthcare consultations [128] and the odds for having secondary care consultations [116], (8) the level of beliefs of negative consequences of health condition and the propensity to use pain medication [59] and to have primary [59] and secondary care consultation [116], (9) negative illness beliefs and the chance of using pain medication [59], having healthcare consultations in general [68] and primary care consultations in particular [59] and the amount of HCU in general [120], (10) psychological distress and the number of emergency room visits [111] and hospital admissions [111] and the propensity of using pain medication in general [59] and prescription pain medication in particular [81,111], and of having primary care consultations [59,65,82] and (11) symptom vigilance and the amount of healthcare consultations [69].

Moreover, for some relationships between positive CEF clusters and HCU outcomes that were investigated less than 4 times, the limited results met our expectations of showing a negative association. This was the case for the association between: (1) illness coherence and the odds for pain medication use [59], (2) pain acceptance and the propensity of using opioids [62], (3) perceived symptom control and the chance of using pain medication [59] and having consultations [66], (4) positive mood and the amount of emergency HCU [102] and hospitalizations [102], (5) psychological flexibility and the amount of pain medication use [70] and consultations with healthcare providers [70] and (6) self-efficacy beliefs and the amount of pain medication use [55,75,126], emergency HCU [67,93] and HCU in general [135] and the odds for using prescription pain medication [81].

It would be erroneous to assume that these results provide us with conclusive evidence for a relationship between these variables, but also considering the impact that revealing these associations could have on socioeconomic burden, it is clear that further research on this topic is needed.

Next to some associations that can be assumed to be non-existent, there is evidence for the presence of several relationships between CEF and HCU. Although for many it is not clear yet what the mechanism behind these associations is and whether they are independent relationships. This should be further investigated before conclusions can be drawn about potential causal interactions between CEF and HCU. In case the suggested causal interactions can be confirmed, further research could focus on the development and/or implementation of interventions that address CEF in an attempt to keep HCU to an optimal level and avoid excessive use.

4.3. Implications for Clinical Practice

The findings of this systematic review indicate that there is evidence for a relationship between several maladaptive CEF (in particular for general and symptom-related anxiety symptoms, catastrophizing and depressive symptoms) and HCU measures in patients experiencing pain, whereas for others more research is needed to confirm a potential relationship. Although it is impossible to conclude anything about causal interactions, it can carefully be suggested that interventions specifically targeting the former CEF could lead to decreases in HCU towards an optimal level, which potentially implies a socioeconomic benefit. An example of such a therapy option is pain neuroscience education, which has been found to effectively address maladaptive CEF and enhance positive CEF in several patient populations experiencing pain [158,159], including patients with chronic spinal pain [32,160,161,162], knee osteoarthritis [163,164,165], fibromyalgia [166] and chronic fatigue syndrome [167]. Moreover, Louw and colleagues found that one preoperative session of pain neuroscience education in patients undergoing surgery for lumbar radiculopathy resulted in large long-term decreases of postoperative healthcare costs [168,169], which are inseparably linked to HCU. Based on the findings of this systematic review, it could be suggested that these results on healthcare costs might be mediated by the direct effect of pain neuroscience education on CEF.

4.4. Strengths and Limitations

To our knowledge, this is the first systematic review specifically focusing on the relationship between CEF and HCU in patients experiencing pain. Although many different constructs of CEF and different ways of approaching HCU were reported in the included studies, making it a complex matter, the authors aimed at giving a digestible overview of the evidence so far. This was achieved by clustering outcomes and creating summarizing tables as has been done before in previous systematic reviews investigating associations between certain variables in non-pain populations [44,45,46,47]. The results of this systematic review indicate that some modifiable CEF are associated with HCU in patients experiencing pain, which might imply that targeted interventions could eventually lead to decreased HCU.

Furthermore, this review has several methodological strengths that ensure minimization of potential bias: the double-blind screening of the literature and risk of bias appraisal, including consensus meetings when necessary; the fact that several databases where consulted aiming at a complete representation of the literature and the final inclusion of a large number of studies (n = 90) comprising a large sample of participants with pain (n = 59,719).

When interpreting the results of this systematic review, a number of limitations related to the included studies should be taken into account. First, most of the studies showed moderate risk of bias (51%), with some studies even showing high risk of bias (29%), and the minority were rated as having low risk of bias (20%). In light of this, readers should keep in mind that the generalizability of some of the results might be questionable. Second, the research question of the present systematic review was often only a secondary objective of the included studies, meaning that results were not always described in detail and the included studies might have been underpowered for this particular research question, which increases the risk for type II errors, or false negatives.

Next, some limitations related to the methodology of this systematic review should be considered. In an attempt to give a comprehensive overview of the literature about the relationship of CEF and HCU in patients experiencing pain, it was necessary to apply broad in- and exclusion criteria. Due to the amount of included studies, their analyses and the variety in outcome measures, it was necessary to cluster outcomes, making the results more consumable. Inevitably, this led to a loss of information; however, thanks to this clustering, it became possible to draw firm conclusions about particular relationships. Moreover, to our knowledge there are no standardized or validated criteria available yet to rate the level of associations in systematic reviews especially focusing on relationships between variables. Therefore, we aimed at using a methodology that was suitable for our research question and had been used before in comparable systematic reviews. This was the case for the chosen methodology, which had been used 4 times before [44,45,46,47] and was deemed to be reliable to make an estimation of the level of association for the pooled results. Furthermore, this review focused solely on associations between CEF and HCU, wherefore, no conclusions about causal relationships could be made based on the results.

5. Conclusions

Based on the available evidence, an at least indirect positive association between general anxiety symptoms, depressive symptoms and catastrophizing and the amount of pain medication use can be confirmed in people with pain. Additionally, general anxiety and depressive symptoms appear to be univariately related with the propensity to use opioids. In terms of consultation behavior in people experiencing pain, an at least indirect relationship with, symptom-related anxiety and psychological distress was found. Use of CAM services appeared to be positively related with the level of perceived symptom control. For other relationships, no evidence was found, inconsistent findings were reported, or they were insufficiently studied to draw firm conclusions. However, in the latter case, the limited results for some relationships pointed towards a confirmation of our hypothesis that maladaptive CEF were related to more HCU, and the other way around for positive CEF, indicating that more research on this topic is needed. Although it is impossible to draw conclusions about causal interactions, the results of this systematic review carefully suggest that it could be important to address maladaptive CEF, such as anxiety symptoms, catastrophizing and psychological distress, in clinical practice to potentially decrease excessive healthcare seeking behavior in people experiencing pain and the socio-economic burden related to pain.