The Edmonton Classification System for Cancer Pain: a tool with potential for an evolving role in cancer pain assessment and management

ABSTRACT

Introduction: Undertreatment of cancer pain is associated with inadequate assessment and inconsistent or non-standardized classification, resulting in failure to both appreciate its multidimensional nature and appropriately target therapeutic interventions. This review examines the classification of cancer pain with a focus on the progressive development of the Edmonton Classification System for Cancer Pain (ECS-CP); the appropriateness of its constituent features, associated outcomes and its potential future development in cancer pain classification.

Areas covered: A Medline search from 1989 to November 2017, using combined terms ‘cancer’ or ‘oncology’, ‘Edmonton’, ‘pain’ or ‘analgesia’, and ‘staging’ or ‘classification’, identified 280 records. A total of 20 studies with empirical data relating to validation studies of the ECS-CP or evaluation of either its constituent or proposed domains were selected for inclusion in the core review.

Expert commentary: The ECS-CP is a tool in evolution and a valid template for further cancer pain classification development. The assessment of ECS-CP domains requires a standardized approach. The domain ratings can inform the therapeutic strategy, and are associated with pain management outcomes, particularly stable pain control. The ECS-CP enables standardized reporting, based on patients’ pain and related characteristics, and thus may improve the validity of comparisons across research study samples.

KEYWORDS:

• Cancer pain
• pain assessment
• pain classification
• pain management
• palliative care

1. Introduction

Pain is one of the most feared and distressing symptoms associated with cancer. Pooled prevalence estimates suggest that 55% (95% confidence interval, 46–64) of patients receiving active cancer treatment and 64% (58–75) of those with advanced disease experience cancer pain [1]. Cancer is associated with ageing [2], and given current demographic trends that signal a dramatic increase in the elderly proportion of the population [3], an informed approach to cancer pain assessment and management is of great importance. In a comprehensive pain assessment, classification of cancer pain is based on the multidimensional nature of cancer pain [4]. Identification of the multidimensional aspects of an individual’s cancer pain presentation is key to informing the most effective therapeutic strategy. This in turn is pivotal in optimizing the patient’s health-related quality of life, which is at risk of significant impairment due to poorly controlled cancer pain.

Health-related quality of life (QoL) reflects a complex multifactorial construct, comprisingmeasures of functioning in the physical, social, spiritual and psychological domains [5]. The relationship of health-related QoL with cancer pain is also complex, as represented in Figure 1. Depending on the stage of disease, cancer is also associated with other distressing symptoms, such as fatigue and dyspnea. Most cancer patients are elderly and many have multiple comorbidities, which further compromise QoL [6]. In the course of pain management, frailty, cognitive and renal impairment may increase the risk of opioid side effects. Furthermore, the burden and risk associated with disease modifying therapies, such as radiation and chemotherapy, may outweigh the benefit. Overall, the complications associated with advanced cancer and the presence of multiple comorbidities pose significant challenges in terms of achieving stable pain control and optimizing health-related QoL. These complex clinical scenarios and the often under appreciated palliative context of care warrants a holistic, multidimensional, and multidisciplinary assessment. Moreover, there is strong evidence to indicate that cancer pain is undertreated [7], and failure to perform a comprehensive assessment is recognized as a significant underlying factor [8,9]. Classification of cancer pain is regarded as a critical component of a comprehensive cancer pain assessment [10,11].

Figure 1. Cancer pain in the context of health-related quality of life.

Clinically, the main purpose of a cancer pain classification system is to inform pain management [10]. A classification system may identify pain or patient characteristics that portend difficulty in the task of achieving stable pain control, a major objective from the patient’s quality of life perspective. In addition to assisting with the choice of the most appropriate therapeutic intervention, a classification system may also inform the effective triaging of patients to sites or levels of care consistent with their complexity of need [12]. Meanwhile, from a research perspective, a cancer pain classification system enables standardized reporting. This may lead to better study sample characterization and possibly enable stratification of patients based on their pain and related characteristics; as a result, more valid comparisons may be made across study samples. A valid cancer pain classification system allows both clinicians and researchers to use a common language in or between their respective settings [13]. Despite the plausibility of these clinical and research frameworks, progress to date, in terms of both developing a valid and universally acceptable cancer pain classification system, and ensuring uptake in routine practice and in cancer pain research studies, has been mixed.

A systematic review of cancer pain classification identified three formal systematically developed cancer pain classification systems with varying degrees of validation [11]. The development of a chronic pain (both malignant and non-malignant) taxonomy was begun by the International Association for the Study of Pain (IASP) in the 1970s and recently updated 14 it represents a very extensive catalogue of pain conditions and respective codes but no prognostic functionality to indicate difficulty in achieving stability in pain relief. The Cancer Pain Prognostic Scale (CPPS) involves a comprehensive application of a potentially burdensome battery of assessment tools and aims to classify prognosis of pain relief within two weeks, based on a combined score from the assessment tool data; to date further studies regarding the CPPS have not been reported other than the original study [15]. Of the three cancer pain classification systems with some validation to date [11], the Edmonton Classification System for Cancer Pain (ECS-CP) is clearly the most widely studied and has involved almost three decades of development. This review will focus mainly on the ECS-CP as a template for cancer pain classification development and the objectives are: (1) to describe the challenges and progress in development of the ECS-CP; (2) to evaluate the appropriateness of its constituent domains and associated outcomes; and (3) to discuss its potential future development in cancer pain classification.

2. Methods

Medline was searched from 1989 to November 2017, using the combined terms ‘cancer’ or ‘oncology,’ ‘Edmonton,’ ‘pain’ or ‘analgesia,’ and ‘staging’ or ‘classification’ or ‘assessment.’ The search was limited to English language publications, humans, and adults. We initially retrieved 280 records; further screening identified 15 studies with empirical data relating to validation studies of the ECS-CP or evaluation of either its constituent or proposed domains. A search of the reference lists of these 15 references identified a further 5 studies with empirical data, resulting in a total of 20 empirical studies for inclusion in this review [16–35]. Similar searches were conducted in relation to each domain item of the ECS-CP in addition to pain intensity and age using relevant terms, see Appendix for details.

3. Results

3.1. Challenges and progress in the evolution of the ECS-CP

The twenty empirical investigation studies that reported any aspect of validity, reliability, clinical utility, or audit with the composite form of the ECS-CP, or validity aspects of current or potential ECS-CP domains will be discussed as they arise in relation to subsequent sections of the review. The earliest of these studies are summarized in Table 1. Bruera et al used the ‘TNM’ classification system for cancer staging as a model for the development of the first precursor of the ECS-CP, which was called the Edmonton Staging System for Cancer Pain (ESS)[16]. This tool had seven features that demonstrated good predictive value in estimating the prognosis of a patient reaching stable pain control. The features included mechanism of pain, incidental pain, daily opioid dose on admission, cognitive function, psychological distress, tolerance, and past history of alcohol or drug addiction. Using certain feature combinations, patients were classified as being in Stage I, II, or III and as having a good, intermediate, or poor prognosis, respectively, in terms of achieving good pain control. Subsequently, this version of the ESS was modified to reflect the results of a new study in which two if its features (cognitive function and daily opioid dose on admission) failed to demonstrate predictive potential for the achievement of good pain control, and consequently were removed from the tool [17].

Table 1. Validation and related studies (1989–2000) of the ESS, a precursor of the Edmonton Classification System for Cancer Pain (ECS-CP).

First Author, Year, Country	Setting	Study Design	Sample Size (N)	Objectives	Assessment/Analysis	Findings
Bruera et al, [16] 1989, Canada	PCU	Prospective cohort	56	Evaluate ESS for cancer pain using prognostic factors for pain response	Stages 1–3 (good-poor prognosis) according to PM and pain characteristics, prior OD, CF, PD, tolerance and history of drug addiction or alcoholism.	18/22 (82%) patients in Stage 1 achieved good pain control vs. 2/22 (10%) in Stage 3; (P < 0.01). Sensitivity, specificity and negative predictive value of ESS were 0.75, 0.86 and 0.80, respectively.
Bruera et al, [17] 1995, Canada	PCU	Prospective cohort	276	To assess the accuracy of the ESS for CP	ESS completed at 1st consultation. ESS included PM, (NeuP versus non-NeuP), pain characteristics (continuous vs incidental), prior OD, CF, PD, tolerance, past history of alcohol or drug abuse.	High ESS Inter-rater agreement (kappa 0.72–0.76, p < 0.001). 93% Stage I patients achieved good pain control vs 55% Stage 2 and 3 patients (55%, P < 0.001). ESS sensitivity and specificity were 0.93 and 0.46, respectively. CF and OD not correlated with pain.
Bruera et al, [18] 2000, Canada	Regional Palliative Care Program	Retrospective cross-sectional	1,354	Compare frequency of pain prognostic indicators in TPCU vs Acute Care patients	ESS: prognosis for pain control based on absence of any factor (Stage 1) or presence of ≥ 1 of 5 factors (Stage 2): NeuP, incidental pain, opioid tolerance, history of alcoholism or drug abuse and somatization	The frequency of ESS Stage 2 was 110/127 (87%) in the TPCU and 397/608 (65%) in Acute Care (p < 0.0001)

Abbreviations (listed alphabetically): CP, Cancer Pain; CF, cognitive failure; ESS, Edmonton Staging System; NeuP, neuropathic pain; OD, opioid dose; PM, pain mechanism; PI, pain intensity; PCU, Palliative Care Unit; PPG, personalized pain goal; PD, psychological distress; TPCU, Tertiary Palliative Care Unit;

Over the course of the subsequent decade, difficulties were reported regarding the definition of some ESS features, particularly incidental pain and psychological distress. Furthermore, although the method of estimating opioid tolerance was based on calculating the opioid escalation index percentage (OEI%) as a standardized measure [36], this was not used consistently. A series of validation and other related studies were reported between 2005 and 2010 (see Table 2). In a subsequent revised version of the ESS, the rESS, the opioid tolerance feature was removed [19]. Also, given the pivotal importance of a patient’s cognitive status and their ability to give a pain history or not, the cognitive function feature was reinserted in the rESS [19]. The rESS was then validated in a regional multicentre study and inter-rater reliability for the items ranged from r = 0.67 for pain mechanism to r = 0.95 for presence of an addiction history [19]. In addition, there was another validation study using a modified Delphi survey process, which resulted in the retention of the five main rESS features: pain mechanism, incident pain, psychological distress, addictive behaviour and cognitive function. However, the operational definitions for incidental pain, psychological distress, addictive behavior, and cognitive function were all modified; the incidental term was replaced with the term incident [20]. The name of the tool, rESS was changed to the Edmonton Classification System for Cancer Pain (ECS-CP). (See Box 1). This name change was made so as to divest the tool of the ‘staging’ concept; study data had demonstrated the ultimate attainment of stable pain control in most patients [19], even though some pain and patient features were associated with longer time to achieve this. The change effectively shifted the focus from prognosis to complexity of management: it reflected the ECS-CP’s role in classifying patients on the basis of complex pain management needs rather than reflecting prognosis for the attainment of stable pain control [12]. Instructions and case examples to assist clinical practitioners and researchers in completing the ECS-CP are available in an online accessible manual [37].

Table 2. Validation and related studies (2005–2010) of the rESS, a precursor of the Edmonton Classification System for Cancer Pain (ECS-CP).

First Author, Year, Country	Setting	Study Design	Sample Size (N)	Objectives	Assessment/Analysis	Findings
Fainsinger et al, [19] 2005, Canada	Multisite: TPCU, Acute Care and Hospices	Prospective cohort	746	Assess the inter-rater reliability and predictive validity of the rESS in classifying CP.	rESS: assessed PM, presence or absence of incidental pain, presence or absence of PD and Addiction, and level of CF	Inter-rater reliability estimates range: 0.67 (PM) to 0.95 (Addiction). In multivariate analysis, only age (≤ 60), NeuP and incidental pain were associated with time to stable pain control (P < 0.05).
Nekolaichuk et al, [20] 2005, Multiple countries	PC, Pain, Oncology, Internal Medicine experts	Delphi survey of content experts	72	To gather construct validity evidence for a CP classification system.	Expert panels: regional (Panel A, n = 18) and national/international (Panel B, n = 52) palliative medicine and pain specialists in a Delphi survey	Definitions for incidental pain, PD, Addiction and CF were revised; guidelines developed for use. Renamed ECS-CP
Russell et al, [21] 2006, Scotland, United Kingdom	Hospice inpatients	Retrospective cross-sectional	118	Evaluation of an abbreviated ESS (aESS) vs the ESS and achievement of stable pain control	Compared aESS (only NeuP and incident pain features) and ESS groups regarding adjuvant use, opioid switches and achievement of stable pain (PI ≤ 2 on 3 days over 5 consecutive days)	ESS and aESS poor pain prognosis groups used more adjuvants, had more opioid switches, and fewer achieved stable pain control. aESS-derived poor pain prognosis: sensitivity was 0.91 and specificity was 0.39.
Lowe et al, [22] 2008, Canada	Multisite: TPCU, Acute Care and Hospices	Retrospective cohort	532	Evaluate opioid dose escalation in a CP classification system	Secondary analysis of rESS validation study [17] . OEI% as an index of opioid responsiveness. OEI% association with rESS domains	Approximately 44% of the patients had an OEI%>5%. No significant associations between OEI% and age, NeuP, incident pain, PD or A.
Amigo et al, [23] 2008, Canada	Acute Care, TPCU, and Community	Retrospective cross-sectional	1,719	Audit of clinical practice in a regional palliative care program	Compared the frequency of ECS-CP domains (NeuP, incident pain, PD, Addiction and CF) across all regional PC sites of care	The frequency of normal CF and the presence of all other ECS-CP domains was significantly higher in TPCU patient
Fainsinger et al, [24] 2009, Canada	Multisite: TPCU, Acute Care and Hospices	Retrospective cohort	591	Secondary analysis of PI in an rESS validation study [17]	PI at 1st assessment: mild (0–3), moderate (4–6), or severe (7–10). PI association with time to stable pain control, final OD, and number of adjuvants.	Predictors of time to stable pain control in multivariate model: moderate and severe PI, age, and NeuP. PI associated with higher final OD and greater use of adjuvant modalities.
Fainsinger et al, [25] 2010, Mutiple countries	Acute Care, TPCUs, Outpatient clinics, and Hospices	Prospective cohort study	1,100	Assess predictive validity of the ECS-CP and Pain Intensity (PI) in an international sample of cancer patients.	Patients recruited from 11 sites. PC specialists completed ECS-CP for patient. Daily pain ratings, number of BTP doses, types of pain adjuvants and ODs recorded until end point (stable pain control/discharge/death).	In multivariate analysis, age, NeuP, incident pain, PD and PI were associated with time to stable pain. Patients with NeuP, IP, PD or higher PI required more adjuvants and higher final ODs.

Abbreviations (listed alphabetically): BTP, breakthrough pain; CP, Cancer Pain; CF, cognitive failure; ESS, Edmonton Staging System; rESS: revised Edmonton Staging System; IP, incident pain; NeuP, neuropathic pain; OEI%, opioid escalation index %; OD, opioid dose; PM, pain mechanism; PL, pain localization; PI, pain intensity; PC, palliative care; PCU, Palliative Care Unit; PPG, personalized pain goal; PD, psychological distress; TPCU, Tertiary Palliative Care Unit;

Box 1. Components of the Edmonton Classification System for Cancer Pain.

1. Mechanism of Pain No No pain syndrome Nc Any nociceptive combination of visceral and/or bone or soft tissue pain Ne Neuropathic pain syndrome with or without any combination of nociceptive pain Nx Insufficient information to classify 2. Incident Pain Io No incident pain Ii Incident pain present Ix Insufficient information to classify 3. Psychological Distress Po No psychological distress Pp Psychological distress present Px Insufficient information to classify 4. Addictive Behavior Ao No addictive behavior Aa Addictive behavior present Ax Insufficient information to classify 5. Cognitive Function Co No impairment. Patient able to provide accurate present and past pain history unimpaired Ci Partial impairment. Sufficient impairment to affect patient’s ability to provide accurate present and/or past pain history Cu Total impairment. Patient unresponsive, delirious or demented to the stage of being unable to provide any present and past pain history Cx Insufficient information to classify.

aInstructions for completion of the ECS-CP are available via weblink [37]

Various studies to evaluate the ECS-CP and its constituent or related domains from 2010 to 2017 are summarized in Table 3. The utility of the ECS-CP has been examined in two international studies, each with just over 1000 patients from multiple countries and practice settings [25,28]. Although both studies demonstrated acceptable utility of the ECS-CP, the relatively low level of interrater reliability for some items such as psychological distress and incident pain remain a concern. The great challenge is to improve on the tool’s scientific metrics, yet to keep its functionality sufficiently simple enough to encourage routine practice use and audit.

Table 3. Validation and related studies (2011–2017) of the Edmonton Classification System for Cancer Pain (ECS-CP).

First Author, Year, Country	Clinical Setting	Study Design	Sample Size (N)	Objectives	Assessment/Analysis	Findings
Knudsen et al, [26] 2011, Multiple countries	Inpatients and Outpatients with cancer	Prospective cross-sectional	2,278	Test if variables as identified in the literature are associated with pain.	Assessment of PM, BTP and PD in relation to PI and/or response to pain treatment. Exploration of new variables	Variables most strongly associated with average pain, worse pain and pain relief were: BTP, PD, sleep, and OD.
Knudsen et al, [27] 2012, Multiple countries	Oncology and Palliative Care Centres	Combined cross-sectional (A) and prospective longitudinal B)	1,529 (A) and 352 (B)	To further develop an evidence-based platform for the content of an international CP classification system.	Applied regression models to test association of IP, PD, NeuP, PL, sleep disturbances, MEDD, and cancer diagnosis on pain outcomes (average, worst, relief)	IP, PL, MEDD, nonsteroidal anti-inflammatory drugs, and sleep were associated with ≥ 1 pain outcomes in A. Initial PI, initial pain relief, IP, PL, cancer diagnosis, and age were predictors in B analysis. Identified domains only explained 16% to 24% of the variability of the pain outcome.
Nekolaichuk et al, [28] 2013, Multiple countries	PCUs, Outpatient Clinics, Oncology Units Hospices, and Acute Care	Prospective Cross-sectional	1,051	Describe the prevalence of the five ECS-CP domains in a sample of patients with advanced cancer.	Patients recruited from 17 international sites. A clinician completed the ECS-CP. Additional information, including PI collected through patient self-reports, using touch-sensitive computers.	670 (64%) had pain: nociceptive (534; 80%); NeuP (113; 17%); IP (408; 61%); PD (212; 32%); A (30; 4.5%); normal CF (616; 92%). Prevalence of ECS-CP features and PI scores varied widely
Brunelli et al, [29] 2014, Multiple countries	PCUs, Outpatient Clinics, Oncology Units, Hospices, and Medical Wards in Acute Care	Cross-sectional	563	To compare physician clinical assessment with patient-rated evaluations for IP/BTP, NeuP, and PD.	Average pain in previous 24 hours as dependent variable in multivariate linear regression models, and IP/BTP, NeuP, and PD were tested as regressors; in assessing regressors, physicians used the ECS-CP, and patients used structured self-assessment questionnaires.	When tested in 2 separate models, patient ratings explained 20.3% of variability (95% confidence interval [CI] = 15.2–25.3%), whereas physician ratings explained 6.1% (95% CI = 2.2–9.8%). Patient ratings had higher discriminative capability.
Mercadante et al, [30] 2015, Italy	Two Acute Pain Relief and Supportive Care Units	Prospective cohort	166	Assess the prognostic value of initial PI and its duration to achieve pain control in cancer patients.	PI at admission (T0), at stable pain control (Ts) and 15 days pre- admission (T-15). Ts response s graded: (1) good pain control; (2) adequate pain control requiring opioid increase; (3) adequate pain control with adverse effects; (4) unable to achieve pain control within a week. OEIs	PI at T0 and at T-15, opioid doses, opioid therapy duration, and age associated with more complex analgesic therapies, which were effective in 133 (80%) ≤ 1 week. Concluded that high PI can reflect inadequate treatment and predicts more complex treatment. Opioid tolerance, younger age, may also predict PC outcomes
Pina et al, [31] 2015, Portugal	Pain Clinic in Cancer Centre	Prospective cross-sectional	371	To determine the characteristics and associations of initial pain intensity in a specialist cancer pain clinic	Standardized assessments with validated tools, including Brief Pain Inventory, Eastern Cooperative Oncology Group (ECOG) functional performance status, MEDD. Linear regression model was constructed to predict iPI.	285 (77%) had moderate (4–6) or severe (7–10) iPI. In the multivariable model, higher income, ECOG 3–4, cancer diagnosis (head & neck, genitourinary and gastrointestinal), adjuvant use and initial MEDD were associated with iPI (p < 0.05). The model’s R 2=18.6, explaining only 19% of iPI
Arthur et al, [32] 2015, U.S.A.	Outpatient Supportive Care Clinic in a Cancer Center	Retrospective cross-sectional	100	Study relationship of negative ECS–CP features with PI and OD. Also, to determine if sum of negative ECS–CP features associated with higher PI	The electronic charts of 100 were reviewed for variables like patient characteristics, initial ECS–CP assessment, MEDD, opioid rotation, Edmonton Symptom Assessment Scores, and use of adjuvant analgesics.	NeuP associated with higher median PI (7 vs 5, P = 0.007) and median MEDD (83 vs. 30, P < 0.013). PD associated with higher median PI (7 vs. 5, p < 0.042). A higher number of negative ECS–CP features was associated with higher PI (p < 0.01).
Arthur et al, [33] 2017, U.S.A.	Outpatient Supportive Care Clinic in a Cancer Center	Retrospective cohort	386	To explore the association between ECS-CP features and pain outcomes in patients managed by a specialist PC team.	Initial and follow-up clinical information of 386 eligible supportive care outpatients were retrospectively reviewed and analyzed. Data was analyzed regarding PPG	Neuropathic pain (p < 0.001) and ≥ 1 ECS-CP feature (p = 0.006) patients used more adjuvants. Fewer with neuropathic pain reached PPG (29 vs 72%, p = 0.015). No association between sum of ECS-CP features and pain outcomes.
Tanco et al, [34] 2017, U.S.A.	Outpatient Supportive Care Clinic in a Cancer Center	Prospective cohort	1,071	Determine rate of completion of the ECS-CP domains after revision and simplification of the response for each domain.	Electronic records of patient visits collected at 6 months before (pre-interventional), 6 and 24 months after (post-interventional) the simplification (responses changed to Yes/No) of the ECS-CP tool.	5/5 items were completed pre-intervention, 6 months post-intervention and 24 months post-intervention in 0/343 (0%), 131/341 (38%), and 222/328 (68%), respectively, (p < 0.001).
Fainsinger et al, [35] 2017, Canada	Acute Care hospitals and a hospital-based TPCU	Prospective cohort	300	Assess feasibility of obtaining PPG and compare a clinically based PPG definition with ESC-CP definition for stable pain.	The attending physician completed routine assessments, including a PPG and the ECS-CP, and followed patients daily until stable pain control or death, or discharge.	169/231 (73%) provided a PPG, with 113/169 (67%) reporting a PPG ≤3 (median = 3, range = 0–10). Using the PPG definition as the gold standard, sensitivity and specificity of the ECS-CP definition were 71.3% and 98.5%, respectively.

Abbreviations (listed alphabetically): BTP, breakthrough pain; CP, Cancer Pain; CF, cognitive failure; ESS, Edmonton Staging System; IP, incident pain; iPI, initial pain intensity; MEDD, morphine equivalent daily dose; NeuP, neuropathic pain; OEI%, opioid escalation index %; OD, opioid dose; PM, pain mechanism; PL, pain localization; PI, pain intensity; PC, palliative care; PCU, Palliative Care Unit; PPG, personalized pain goal; PD, psychological distress; TPCU, Tertiary Palliative Care Unit;

3.2. Pain management outcomes and the ECS-CP

The outcome commonly linked to the ECS-CP in research studies is the achievement of stable pain, based on both the numerical rating of pain and the patient’s use of breakthrough opioid analgesia. Based on expert input in a Delphi process [20], ‘stable pain’ in the ECS-CP studies was defined as both a pain intensity score of ≤ 3 for 3 consecutive days and < 3 opioid breakthrough doses on each of these 3 days; the understanding here is that breakthrough doses are prescribed hourly as needed. This restrictive definition may present problems in clinical practice, as problems arise in practice in relation to patients with cognitive deficits, and more frequent administration of breakthrough opioid doses (some centers use a 30-minute lockout for continuous administration pumps that deliver patient-controlled analgesia). Defining stable pain in the context of cognitive impairment is challenging; this can be addressed by omitting the pain intensity stipulation for the ECS-CP stable pain definition and instead, using the second half of the definition, which refers to the number of breakthrough opioid doses used. Standardized interpretation may also present challenges in this context due to practice variation in the prescribed frequency of opioid breakthrough dose administration, as with continuous administration pumps or the use of newer rapid onset opioid formulations for breakthrough pain. Apart from the previously mentioned outcomes that are conventionally associated with the ECS-CP, efforts have been made to develop some degree of international standardization.

An international expert group of pain management and palliative care physicians held a conference in Milan in 2009 [38]. At this meeting, consensual recommendations were made regarding specific working proposals and international standards in relation to cancer pain assessment and classification. Participants identified pain intensity, pain relief, and the temporal pattern of pain as the most relevant outcomes from clinical practice and research perspectives. The numerical rating scale (0–10) was recommended for pain intensity assessment, and average pain intensity as a measure of intensity, using the previous 24 h or the course of the preceding week as reference. In the assessment of pain response, a reduction of ≥ 50% was considered a ‘substantial decrease’ and ≥ 30% ‘a meaningful decrease’ in pain intensity. Controlled or manageable pain was considered as average pain ≤ 3 (on a 0–10 scale); pain ≥ 4 but < 7 was considered moderately controlled and pain ≥ 7 was considered inadequately controlled pain. In summary, this meeting achieved important international consensus at the time over the most relevant pain outcomes in addition to agreeing on optimal cut-points for mild, moderate, severe, and controlled pain.

The importance of patient reported outcome measures (PROMS) is increasingly recognized. The onus on clinical practitioners is to respect patient wishes and achieve consensus in regard to the level of therapeutic interventions that aim to achieve stable pain. This involves a personalized approach that embraces individual pain treatment thresholds and desirability to receive analgesic interventions [39]. The Milan consensus conference on cancer pain assessment and management occurred before the publication of some important studies that examined the role of the personalized pain goal (PPG) as a pain management outcome [35,40]. A PPG has been defined as ‘the verbal or written goal stated by the patient describing the desired level/intensity of pain that will allow the patient to achieve comfort in physical, functional, and psychosocial domains.’ In light of increasing focus on person-centered medicine and PROMS, a PPG is an alternative outcome of cancer pain management that clearly merits consideration [41]. In a retrospective study, a PPG was available for 445/465 (95.5%) cancer patients; the median PPG was 3 (interquartile range, 2–3) at initial consultation and remained stable at follow up (median 14 days). Using a ≥ 30% or ≥ 2-unit reduction in pain intensity as a measure of a clinical pain response and the PPG as the gold standard for individual’s targeted pain response, the sensitivity of the PPG was highest (98%) for those with severe (7–10 intensity rating) pain and lowest (52%) for those with mild (1–4) pain [40]. In a prospective study of 231 patients with cancer pain in acute care settings, a PPG was obtainable for 169 (73%) and approximately two-thirds (67%) identified their PPG as ≤ 3 [35]. The ECS-CP defined stable pain outcome of ‘a pain intensity score of ≤ 3 for 3 consecutive days and < 3 opioid breakthrough doses on each of these 3 days’ was compared to the PPG as a gold standard. Approximately 3 out of 4 (71.3%) patients who reached their PPG also attained stable pain by the ECS-CP criteria. As an outcome measure, the specificity of the ECS-CP stable pain definition was 98.5%. This study demonstrated a role for the PPG and also provided validity for the ECS-CP stable pain outcome criteria. Taken together, these studies show that for cognitively intact patients, the PPG has feasibility as an outcome in relation to cancer pain classification and in response to therapeutic intervention; compared to many of the previously studied clinical response outcomes, it is inherently more closely aligned with patient goals.

3.3. Current ECS-CP domains

3.3.1. Pain mechanism

The pain mechanism domain of the ECS-CP classifies cancer pain into nociceptive, neuropathic, or mixed components of both. This is an important step with a view to informing the subsequent choice of therapeutic intervention to target the underlying mechanism [25,42]. Although neuropathic pain is opioid sensitive, the responsivity of neuropathic pain to opioids is weaker than that of nociceptive pain [43], and adjuvant analgesics such as specific anticonvulsants and tricyclic and other antidepressants may be indicated [44].

Although cancer treatment-related pain and pain directly related to the cancer is considered to be in the same broad ‘cancer pain’ category when using the ECS-CP, it should be recognized that the clinical character, mechanisms, and treatment strategies of these two types of pain are really quite distinct [45]. Neuropathic pain in cancer patients may either be directly disease related, as in tumor associated compression of a nerve plexus, or it may be indirectly related to cancer through its treatment, as with postsurgical or chemotherapy-related neuropathic pain. There is also the possibility of comorbid non-cancer related pain. Unlike the classical peripheral nociceptor activation in nociceptive pain, neuropathic pain is associated with abnormal pain processing. Neuropathic pain has been defined as ‘pain caused by a lesion or disease in the somatosensory system’ [46]. A nociceptive and neuropathic mix in many cancer pains has prompted the dimensional concept of pain as being ‘more or less’ neuropathic, rather than viewing it categorically as being explicitly present or absent [47].

Overall pooled conservative and liberal estimates of neuropathic pain prevalence were reported in a systematic review as 19% (95% CI, 9.4–28.4) and 39.1% (28.9–49.5), respectively [48]. The International Association for the Study of Pain have endorsed guideline criteria, based on history, physical examination, and imaging or other investigations [49]. Based on the degree to which these criteria are met, pain is graded as definite, probable, or possible neuropathic pain. In the ECS-CP, the Ne designation indicates ‘neuropathic pain syndrome with or without any combination of nociceptive pain,’ and is largely based on the physician’s clinical judgement, which may be inaccurate [50]. Studies of the ECS-CP to date have reported the Ne present (Ne+ve) designation in the 16.7–35% range [23,25,28,29,32–34]. Among these studies, Ne+ve was reportedly higher in a tertiary palliative care unit (37.1%) than two acute care hospitals (8.9–9.8%)[23]. Also, use of the pain detect questionnaire to assess neuropathic pain in one study resulted in the designation of ‘uncertain or likely’ neuropathic pain in 25% of the study sample compared to a non-standardized clinical estimation of 16.7% using the ECS-CP [29]. This suggests that screening with a validated tool may improve the sensitivity of neuropathic pain detection, as demonstrated in cancer pain studies with the Leeds Assessment of Neuropathic Pain Scale in Australia and the DN4 in Portugal [50,51]. Also, use of formal guidelines, such as the NeuPSIG guidelines or a standardized algorithm based on NeuPSIG criteria is likely to improve the accuracy of neuropathic pain diagnosis when compared to non-standardized clinical assessment [49,52].

Other ECS-CP studies have demonstrated mostly consistent findings. In a cross-sectional study, both greater pain intensity and opioid consumption was associated with a Ne+ve designation [32]. A higher frequency of adjuvant use was associated with Ne+ve in two studies [25,33]. Furthermore, the Ne+ve designation was associated with longer time to achieve stable pain control and higher final daily opioid dose on reaching stable pain control [25]. Overall, these data indicate that the presence of a neuropathic pain component augurs for longer duration and a higher level of complexity and challenge in ultimately achieving stable analgesia.

3.3.2. Incident pain

In the ECS-CP manual definition, incident pain is ‘when a patient has background pain of no more than moderate intensity with intermittent episodes of moderate to severe pain, usually having a rapid onset and often a known trigger.’[37] It further expands on this definition by identifying specific characteristics of incident pain: its intensity is significantly greater than background pain; its intensity is moderate to severe; the trigger is often known or predictable and is activity related, as with movement, defecation or swallowing; onset is rapid, with an intensity peak within 5 minutes; it is transient in nature and may return to baseline shortly after the trigger is stopped or removed; and it recurs intermittently, especially in the context of a recurrence of the trigger. Breakthrough pain (BTP) is a broader term for a spectrum of pain entities, including incident pain [53,54]. As originally defined, BTP is ‘a transitory increase in pain to greater than moderate intensity, which occurs on a baseline pain of moderate intensity or less’ [55] The ECS-CP definition however differs from this in that it refers to a ‘usually known trigger.’ The ECS-CP definition also encompasses BTP of moderate intensity in the context of mild background pain, and mild BTP in the context of no background pain. The term episodic pain has been proposed as an overarching term[56].

In the ECS-CP and related studies selected for this review, the prevalence of incident pain ranged from 28 to 62.3% [19,23–29,31–33,35]. An overall pooled prevalence estimate of 59.2% was reported in a systematic review, which also reported a wide range (39.9%-80.5%) across different sites of care [57]. Given that varying definitions exist in relation to BTP, there is a need for standardized assessment with a validated tool [54,58]. Tools with some degree of validation include the Alberta Breakthrough Pain Assessment Tool (ABPAT) [59,60]; its abbreviated version, the Questionnaire for Intense Episodic Pain (QUDEI)58 and the Breakthrough Assessment Tool (BAT)[61].

In cancer pain classification studies that used the ECS-CP or the rESS, the presence of incident pain or BTP with a trigger had a significant association with pain intensity in cross-sectional studies [26,27], and in longitudinal studies it has been associated with higher pain intensity [27], and longer time to achieve stable pain control [19,25]. Using the ABPAT to assess BTP in cancer patients, BTP was associated with higher pain intensity and poorer quality of life [62]. Also, higher pain intensity (≥ 7–10) over the last week was a strong predictor of BTP in multivariable analysis.

Although these data on incident and BTP in cancer patients collectively demonstrate issues in relation to definitions, assessment and consequently study comparability, it is clear that episodic pain (to use the broadly encompassing term) with or without a trigger has an association with poorer quality of life, higher pain intensity, and longer time to achieve stable pain control.

3.3.3. Psychological distress

The ECS-CP Manual emphasizes the importance of interpreting psychological distress in the context of the pain presentation and defines it as ‘a patient’s inner state of suffering resulting from physical, psychological, social, spiritual and/or practical factors that may compromise the patient’s coping ability and complicate the expression of pain and/or other symptoms.’ [37] A wide variety of sources of distress are within the scope of this definition; it also encompasses an individual’s ability to cope with these in the context of their pain expression; and consequently, their high level of psychological distress may be somatically manifested as increased pain expression. However, the relationship is complex and bidirectional, and the proportion of a patient’s pain expression that is attributable to psychological distress is not easy to determine in a standardized manner. It is perhaps therefore not surprising that inter-rater reliability estimation for psychological distress (0.68) was only moderate [19]. Furthermore, the very wide prevalence range reported for psychological distress (15–75%) in a large multisite study may reflect difficulty in rating this domain [28]. Higher pain intensity has been demonstrated in association with standardized assessments of clinical depression and anxiety [63]. However, psychological distress as an ECS-CP domain has a broader scope of inclusion beyond such clearly defined psychiatric diagnoses. The close relationship of psychological distress to the concepts of ‘suffering,’ ‘total pain’ and ‘coping’ are specifically emphasized in the ECS-CP Manual [37]. Although these concepts are not explicitly defined therein, there are multiple case examples that clinical practitioners might easily recognize and use to assist with the interpretation or operationalization of the ECS-CP in relation to this particular domain.

Although a comprehensive account of the complex relationship between psychological distress and pain expression is beyond the scope of this review, current evidence largely suggests that the extent of higher pain expression in a total pain or suffering context may relate to one or more of many factors. These include a person’s psychological distress in a broad sense [64,65], their suffering along multiple dimensions [66], their coping ability and self-efficacy [67,68]; the presence of past or intercurrent psychopathology such as depression or anxiety [63,69]; their spiritual distress 70 and the degree to which psychosocial support is provided to assist a person in adjustment to the realities of disease progression [65]. The assessment of ‘total pain’ and the designation of psychological distress as being present in completing the ECS-CP is clearly challenging in the context of the many dimensions or constructs of psychological distress that may co-exist and even overlap to varying degrees.

Tools to assess psychological distress vary in their psychometric properties and suitability for use in different settings of care. Although it is difficult to select one universally acceptable tool that will capture the multiple facets of psychological distress in its broadest sense, consistency and commonality in assessing psychological distress is arguably a necessary step to establish standardization in rating this domain in the ECS-CP. The Edmonton Symptom Assessment System (ESASr) 71–74], and the Distress Thermometer [71,75] have been recommended for brief screening, whereas the Hospital Anxiety Depression Scale [76] is recommended for more detailed assessment [71].

3.3.4. Addictive behavior

Many terms have been associated with addictive behavior, such as substance use, misuse and abuse; dependence; addiction; alcoholism and chemical coping. Despite the multiple terms used, a validation study demonstrated a high level of inter-rater agreement for this domain [19]. This might reflect the use of the Cut down, Annoyed, Guilty, Eye-opener (CAGE) [77] alcohol questionnaire as a standardized assessment for this domain. As a validated screening tool for alcohol use disorder, the CAGE is recommended as optional in the ECS-CP manual for assessment of addictive behavior [37]. The manual defines addiction as ‘a primary, chronic, neurobiologic disease, with genetic, psychosocial, and environmental factors influencing its development and manifestations. It is characterized by behaviors that include one or more of the following: impaired control over drug use, compulsive use, continued use despite harm, and craving.’ This definition represents a blending of earlier definitions derived from the American Society of Addiction Medicine (ASAM) and other bodies. The manual also highlights 5 key characteristics of addictive behavior: chronicity; multidimensionality in its development and expression, including genetic, psychosocial and environmental factors; compulsivity; persistent use despite harm; and craving. It further states ‘a remote history of alcohol or substance use may not be considered relevant as a complicating factor in ongoing pain assessment and management.’ The addictive behavior domain specifically does not include chronic tobacco use. The operationalization of the addictive behavior assessment is through patient and collateral history, observation over a series of practitioner visits, and optional use of the CAGE to assess for possible alcohol abuse.

The ECS-CP Manual’s descriptors for addictive behavior were published prior to the publication of the American Psychiatric Association’s latest Diagnostic Statistics Manual (DSM-5) [78]. In DSM-5, the substance use disorder (SUD) section lists 10 substances, including alcohol and tobacco; the term addiction is not included due to definitional concerns and potentially negative connotations. Also, SUD is defined on a continuum; its severity is based on the total number of the 10 symptom criteria that are met to define mild (2–3 criteria); moderate (4–5); and severe (≥ 6). The ECS-CP Manual’s descriptors were also published prior to the latest ASAM definition of addiction in 2011, which states ‘Addiction is a primary, chronic disease of brain reward, motivation, memory and related circuitry. Dysfunction in these circuits leads to characteristic biological, psychological, social and spiritual manifestations. This is reflected in an individual pathologically pursuing reward and/or relief by substance use and other behaviors.’ [79] Addictive behavior is included in the ECS-CP as a possible predictor of maladaptive ‘chemical coping’ if the patient has a past history of drug or alcohol abuse. Patients with an addictive history may thus use opioids and other psychoactive medications to relieve psychological distress in a manner that is inconsistent with the original medication prescription.

The term ‘chemical coping’ was coined by Bruera et al to reflect higher opioid dose usage in cancer patients who had a positive CAGE screen [80]. A positive CAGE screen (score ≥ 2/4) is reported in approximately 4–17% of cancer patients in palliative care settings [81,82]. Based on CAGE positivity, chemical coping is associated with higher symptom distress levels, higher opioid dosing, and more prolonged opioid use [81,83,84]. Prevalence rates of addictive behavior in the 4–23.2% range have been reported in studies of the ECS-CP, rESS, and related studies included as core studies in this review [19,23–25,28,31–33,35,82]. A remarkably wide prevalence range (0–50%) was reported in one multisite study [28]. In an ECS-CP audit conducted across Edmonton Regional Palliative Care Program sites, the prevalence of addictive behavior ranged from 7.2% in the Community Consult Team referrals to 13.7% in the Tertiary Palliative Care Unit [23]. However, multivariable analyses in three large studies demonstrated that addictive behavior had no independent association with the outcome of time to achieve stable pain [19,24,25]. In an Italian study of patients with advanced cancer, a CAGE positive prevalence of 5% was recorded, and positive CAGE status was associated with neither pain levels nor opioid use [85]. Higher opioid use has been reported in association with the presence of addictive behavior in one ECS-CP study [19]. A retrospective study of 300 cancer patients reported that former or current smokers were more likely to have a positive CAGE screen and an illicit drug use history when compared with those who never smoked; in addition, higher pain levels were reported by current smokers [86]. However, a Portuguese cancer pain study found no association between tobacco use and initial pain intensity [31]. Taken together, the data from studies of addictive behavior present a mixed picture and indicate that further studies are needed to clarify the role of ‘chemical coping’ in cancer pain management and how best to assess it in the process of cancer pain classification.

3.3.5. Cognitive function

In the earlier versions of the ECS-CP [16,17], the cognitive function domain was categorized as normal or impaired, whereas the current version has cognitive function categorized as no impairment, partial impairment, total impairment, and insufficient information to classify [20]. No impairment indicates that the patient is able to provide a past and current pain history [37]. The partial impairment category implies that the patient is impaired but responsive, as typically might occur in mild delirium or early dementia; regardless of the etiology of cognitive impairment, the patient is impaired to the degree that they are unable to provide an accurate past or present pain history [37]. The total impairment designation indicates that the patient is unresponsive, severely delirious or demented and is unable to provide any present or past pain history [37].

The assessment of pain in the presence of cognitive impairment has been the subject of increasing research interest [87]. This has occurred predominantly in the non-cancer population and generalizability to dementia patients with cancer pain may be limited. Nonetheless, data regarding the ability of patients with earlier stage dementia in the mild to moderate range to report pain intensity have been demonstrated in some studies [88,89]. In the absence of self-report of pain in patients with dementia, assessment will likely rely on observation tools or features, such as behavioral and facial changes in relation to pain occurrence and severity. The advent of technological detection of facial changes represents an exciting development in this regard [90], potentially offering more accurate assessment of pain in this population. Meanwhile, for operational purposes of the ECS-CP, rather than rely of pain intensity report from patients who are either consistently not at any time capable or even inconsistently capable of providing a valid or reliable pain history, the ECS-CP currently relies on use of breakthrough or rescue analgesia (BTA) doses as an outcome measure to define stable pain (< 3 BTA doses per day for 3 consecutive days) [37].

Although the cognitive function domain had prognostic implications for achieving good pain control as an outcome in the original ESS study [16], a subsequent validation study demonstrated that it had no independent association with this outcome [17]. The rationale for the inclusion of the cognitive function domain in the current version of the ECS-CP relates to the level of importance attributed to this measure in routine clinical practice, audit and research; it therefore provides for more complete reporting and when used it may facilitate valid comparisons between studies.

The reported prevalence of cognitive impairment (Ci category) in the ECS-CP studies and related studies among the core studies included in this review has ranged from 2.2% to 34% [19,23–25,28,31–33,35]. These estimates are likely to reflect the type of setting and the degree of cognitive assessment conducted by the clinician completing the ESS-CP classification. Although the ECS-CP manual [37] does not mandate a cognitive screening test, it states that the ECS-CP is intended to be used in conjunction with objective measures, such as the Mini-Mental State Examination (MMSE) [91]. Although there are many tools to choose from in assessing cognition and in delirium screening, it would seem prudent to use tools that have been validated in the cancer pain population, such as the Short Memory Concentration Test (SOMCT) [92] and the Confusion Assessment Method (CAM) [93], respectively.

3.4. Additional variables to consider for inclusion in the ECS-CP

3.4.1. Age

The relationship between age and both pain intensity and opioid requirements has been examined in many studies. A retrospective study reported that older persons with cancer had similar levels of pain intensity but lower opioid analgesic requirements than younger adults [94]. A similar lack of association between age and pain intensity was subsequently reported by others [1,31,95–98]. On the contrary, many studies have also reported an inverse relationship between age and cancer pain severity [99–103]. Moreover, many studies have reported an inverse relationship between age and opioid dose administration [94,102,104–106]. The complex relationship involving age, pain intensity, and opioid use therefore warrants further evaluation. A study that examined the age and cancer pain relationship highlighted the importance of controlling for both the physical and mental health quality of life components [104].

Studies examining the effectiveness of pain management have reported conflicting findings in comparing older and younger age groups: community prescribing of opioids were significantly lower for the older age group in two database studies [107,108], and a prospective observational study suggested that less effective management of cancer pain does not occur in older patients [95]. Some authors have highlighted the possibility that the elderly may be less informed regarding analgesic management, less effective in communicating their distress and more likely to be living alone, which together may increase their risk of less effective pain management [109]. The authors of a mixed methods study have suggested that the adaptation to cancer pain may differ in relation to age [87]. Use of accommodative coping was more likely to occur in older patients; this involved greater acceptance and modification of goals and activities.

Two studies examined the opioid escalation index in relation to age and both reported no difference in the opioid escalation index between older and younger patients [22,106]. However, two of the ECS-CP studies reported that age < 60 was associated with longer time to achieve stable pain control [19,25]. In summary, based on available data that are somewhat conflicting, it seems reasonable to recommend that researchers studying cancer pain classification should fully report the age demographics of their study sample as a minimum requirement.

3.4.2. Initial or baseline pain intensity

Two large ECS-CP studies, each with time to stable pain as a primary outcome, reported that initial or baseline pain intensity had an independent and positive association with time to achieve stable pain control [24,25]. A strong positive association between initial pain intensity and pain severity or relief measures at 2 weeks post-study inclusion was reported in a longitudinal study of patients with cancer [27]. Meanwhile, two other Italian studies, conducted in inpatient PCUs, concluded that baseline pain intensity per se does not predict a poorer opioid response, but may be an indication of inadequate opioid usage [30,110]. Although most studies suggest that initial pain intensity is a valid predictor of pain outcomes, the contribution of inadequate opioid dosing to the initial pain intensity warrants consideration [31,111]. These somewhat conflicting data likely reflect the complex multidimensional nature of initial pain intensity.

Given the reported association between initial pain intensity and neuropathic pain, incident pain, psychological distress and addictive history [26,32,112], confounding must be considered in examining their relationship with pain outcomes. Despite adjusting for covariates in some of these longitudinal studies [24,25], the possibility of other non ECS-CP variables contributing to initial pain intensity as a composite measure must be acknowledged. In addition, unreported or undetected psychological distress might also impact initial pain intensity. Initial pain intensity has been associated with non ECS-CP variables such as sleep, cancer diagnosis, location of metastases, opioid dose, localization of pain, and use of non-opioid analgesics [26,27]. Additionally, a Portuguese study identified an independent association between cancer diagnosis (genitourinary, head and neck, and gastrointestinal), poorer functional performance status, and higher income [31]. The inclusion of initial pain intensity and these related variables in the ECS-CP is worthy of debate in addition to further exploration through rigorously designed cohort studies. Of the aforementioned variables, pain intensity, sleep and localization of pain were endorsed in the Milan conference as core or candidate variables for inclusion in a cancer pain classification system [38]. Increasing the number of variables in the ECS-CP is also a concern because brevity, ease of use and general uptake may be compromised as a result. In summary, further studies are needed to better clarify the role of initial pain intensity and those associated variables that are not already included in the ECS-CP, in predicting cancer pain outcomes.

3.5. Integration of patient input

In a mixed methods study, Knudsen et al interviewed 33 patients with advanced cancer to determine their views on the relevance and priority of various domains for inclusion in a cancer pain classification system [113]. Patients acknowledged the relevance of a set of domains similar to those previously identified by Knudsen et al. [26,27] In order of priority, these were etiology, duration, intensity, coping, localization, physical functioning, psychological distress, breakthrough pain, cognitive function, pain mechanism, previous pain experience, and addiction. In addition, sleep was identified as a potential variable for inclusion in cancer pain classification.

The degree of variance in pain intensity that is explained by presence or absence of constituent domains in cancer pain classification models is disappointing [27,29]. This has prompted calls for both a more standardized approach to assessing these domains as well as incorporating patient input [29,114]. In a study of 563 patients with advanced cancer, Brunelli et al compared physician clinical assessment of incident/breakthrough pain, neuropathic pain and psychological distress with patient-rated evaluations using standard instruments to assess these domains, in the prediction of average pain intensity in the preceding 24 hours [29]. Based on the coefficient of determination, R² in regression models, the standardized patient input with validated tools had a much higher discriminative ability than physician’s clinical assessment in regression models, R² was 20.3% and 6.1%, respectively. These findings highlight the benefit of incorporating patient self-assessment tools for the assessment of these 3 domains. Furthermore, the results suggest that standardized measures for any of its constituent domains will likely enhance the validity of a cancer pain classification system.

3.6. Clinical utility of the ECS-CP and practice implementation

The clinical utility of the ECS-CP both in its current format and in its previous versions are clearly demonstrated in two Edmonton-based studies [18,23]. Both illustrate the use of the ECS-CP or its predecessor to guide appropriate patient triage; those with the most complex symptom needs, such as neuropathic pain, incident pain, psychological distress or addictive histories are triaged to the Tertiary Palliative Care Unit. As an audit of resource utilization, quality of care, and a service planning tool, frequent reports are generated regarding the ECS-CP profile of patients accessing services at the various sites of care. There is potential to use the ECS-CP on a larger scale as part of a minimum dataset on a regional or even national level and thus provide the opportunity for big data studies in examining pain management outcomes. Despite the published clinical utility data [18,23], and although it has been translated into Catalan and Spanish, the global uptake of the ECS-CP has been limited. Various attempts have been made by researchers to promote easier and more extensive use of the ECS-CP.

An abbreviated version of the ESS (aESS), which included only neuropathic and incident pain, was examined in a hospice study [21]. The aESS compared favorably to the ESS in terms of sensitivity and specificity. In ECS-CP development, the nomenclature, such as No, Nc, Ne, and Nx (see Box 1), was chosen to reflect the presence or absence of domains in a manner similar to the TNM classification. In view of reported user difficulty, particularly among new users, in decoding this nomenclature, a study of palliative medicine specialists was conducted to examine their completion rates for the ECS-CP before and after implementation of a simplified ECS-CP tool [34]. Changing the response options for the ECS-CP items to a simple ‘Yes’ or ‘No’ resulted in a statistically significant increase in completion of the 5 items from 0/343 (0%) at 6 months prior, 131/341 (38%) at 6 and 222/324 (68%) at 24 months post intervention. These findings clearly warrant further research with the aim of improving both ease of use and completion rates.

Arthur et al examined the potential advantages of using the sum of negative indicators from the ECS-CP domains in predicting pain management outcomes [33]. Apart from a statistically non-significant trend of increasing opioid dose with summed negative ECS-CP prognostic indicators, and greater adjuvant use in those with ≥ 1 ECS-CP negative prognostic indicators at referral, there was no association between incremental summation of ECS-CP negative prognostic indicators and pain treatment responses, estimated either as a 30% reduction in pain intensity or a patient reaching their personalized pain goal. These findings from a retrospective study are in contrast to those of Neckolaichuk et al, who developed a valid predictive model for time required to achieve stable pain, based on summation of negative ECS-CP prognostic indicators in a large international prospective study [115]. The potential to, therefore, develop an index score of complexity, based on summation of ECS-CP predictors would have clinical appeal and is clearly worthy of further research.

The Milan conference in 2009 had representation from many of the major cancer and pain scientific associations and did generate consensus on many working proposals. This included a recommendation that pain intensity, pain mechanism, breakthrough pain and psychological distress be included as core domains in a cancer pain classification system; it also identified ‘candidate domains,’ including pain-related cognitions, addiction, pain location, genetic variations, and other relevant symptoms such as sleep, depression, and anxiety [38].

The knowledge gaps in relation to cancer pain classification as discussed in this review and therefore using the ECS-CP as a template are summarized in Table 4. Pertinent studies to inform some of these knowledge gaps include a comparative effectiveness study of the ECS-CP. This would likely require a cluster randomized controlled trial to compare current standard pain outcomes with use of the ECS-CP domains versus standard care without ECS-CP classification. Although there is limited evidence to currently support the routine inclusion of sex [116], race and ethnic minority status as specific domains in a cancer pain classification system [97,117], these data warrant routine reporting in research studies.

Table 4. Summary of knowledge gaps in relation to the ECS-CP and cancer pain classification.

Main area	Specific knowledge gaps and related study designs where relevant
Standardized assessment of ECS-CP domains and of pain	• Evaluate role of screening with validated tools; • Evaluate use of standardized criteria such as the NeuPSIG guidelines or algorithm; [49,52] • Consensus (Delphi process and surveys) on the standardized operationalization of domain assessments, particularly psychological distress and incident pain; • Use of technology in pain assessment, as in the facial pattern change [90]
Role of additional ECS-CP domains	• Further evaluation of domains with some current validity data (see text) to potentially support their inclusion º Age; initial pain intensity; sleep; pain location; cancer diagnosis; metastatic sites; genetic variations; opioid dose; non-opioid analgesic use; inflammatory markers; and symptom clusters º Use of self-assessment tools for patient ratings of some domains such as neuropathic and incident pain and psychological distress [22]
Comparative effectiveness of the ECS-CP	• A cluster randomized controlled trial to evaluate the stable pain outcome (as per the current definition) of ECS-CP use versus standard care without ECS-CP use
Role of additional outcomes related to ECS-CP use	• Further evaluation of other reproducible outcomes in prospective multicentre cohort studies to assess the performance of the ECS-CP across multiple sites º Patient reported outcomes such as personalized pain goal º Outcomes (average pain intensity, pain relief and temporal pattern of pain) identified at the international consensus meeting in 2009 [38] • Further evaluation of other reproducible outcomes through routine data collection as part of a minimum dataset to facilitate audit, big data studies, and mainly retrospective cohort studies º Utility in triaging to the most appropriate setting of care º Length of inpatient stay; financial cost
Predictive score for pain management outcomes	• Development of predictive models to examine the association of combined ECS-CP domain ratings with a poor prognosis (for stable pain control) with the attainment of stable pain control and other pain management outcomes
Feasibility and uptake of ECS-CP use in clinical practice	• Mixed methods (quantitative and qualitative) studies that involve focus groups and surveys of patients and professional caregivers, and prospective patient cohorts to examine the feasibility of the current and future versions of the ECS-CP • Aiming to achieve an optimal balance between brevity and ease of administration of the ECS-CP with its psychometric properties • Determine burden of assessment from patient and clinical practitioner perspectives
Cancer pain classification: knowledge synthesis	• Update the most recent systematic literature review that focussed on cancer pain classification and was published in 2009 [11]

4. Conclusions

The ECS-CP in its current form and the many studies conducted over eighteen years to inform its sequential development have contributed a large body of knowledge to the entire process of cancer pain classification. Cancer pain classification very clearly hinges on multidimensional assessment. The ECS-CP encompasses those dimensions or domains that have been demonstrated to have an association with pain management outcomes, particularly the achievement of stable pain control. The ECS-CP has potential for greater use in cancer care research, particularly in capturing some of the characteristics of a study population, thus aiding the degree of comparability between studies in the knowledge synthesis process. It has the potential to facilitate communication by allowing clinicians and researchers to speak a common language.

The studies conducted to date have mostly highlighted the need for better standardization in assessing the specific domains of the ECS-CP. Further research is necessary regarding many of its domains and its related outcomes. This is essential in order to better evaluate the role of the current domains and other potential domains in classifying cancer pain; collaboration is key to future research and development. The resulting knowledge synthesis will guide therapeutic intervention strategies in cancer pain management and thereby optimize health-related quality of life.

5. Expert commentary

Although a significant body of knowledge has emerged from many of the studies conducted in the development of the ECS-CP, there clearly are knowledge gaps, particularly in relation to the phenomenon of pain intensity. Many of the regression models constructed in studies examining the correlates of pain intensity, report that such models only explain a small amount (at most 26%) of the variance in pain intensity [27,31]. This means that either there are additional domains that ought to be considered for inclusion in cancer pain classification, or that the assessment accuracy of the current ECS-CP constituent domains is inadequate, or perhaps there exists a combination of both of these factors.

Given the concerns regarding the accuracy and standardization in the assessment of ECS-CP domains, examination of existing databases or study datasets for most domains is unlikely to provide more knowledge than what is already known. Further research will probably therefore require large prospective cohort studies with sufficient statistical power to generate valid estimates of the association of multiple variables with a specific pain outcome measure, such as the ECS-CP stable pain criteria. Large sample sizes will require multisite collaboration. Such studies will require standardized measures to assess the variables (domains) under investigation. Other domains previously referred to in this review as having potential for use in a cancer pain classification system include age, pain intensity, sleep, cancer diagnosis, metastatic sites, localization of pain, opioid dose, and use of non-opioid analgesics. Although preliminary exploration of genetic markers for pain outcomes has demonstrated limited success [118–120], this complex area is worthy of further exploration [38]. Similarly, preliminary data demonstrating a correlation between inflammatory markers and cancer pain intensity is also worthy of further investigation [121]. There is a need to more clearly define symptom clusters that include pain, and to determine their potential if any in cancer pain classification [122,123].

The issue of the most appropriate pain outcome measure for the ECS-CP is a moot point, particularly in view of controversy over cut-points on a 0–10 numerical rating scale and definition of response on this scale. Admittedly, progress was made at the Milan conference in reaching a consensus on what constitutes a pain response. However, this consensus predated many of the studies on PPG. The studies to date on PPG indicate that patients may often make trade-offs in their level of acceptance of therapeutic intervention (a choice of remaining on a lower dose of opioid rather than risk potential side-effects associated with an increase in opioid dose) with a consequent potential increase or decrease in pain intensity level. Meanwhile, although a pain intensity rating of 0/10 would clearly be desirable, most patients have identified a PPG of 3/10 as a reasonable target in clinical practice [35,40].

Can improvements be made with the format of the ECS-CP? There is some evidence to suggest that simplifying the format may ease the administrative burden and improve its completion rate in clinical settings [21,34]. However, the brevity of the ECS-CP in its current format or even a shorter format is both a strength and a weakness; brevity is a necessary attribute for uptake in routine clinical practice and universal usage; however, such brevity may not adequately capture the truly complex phenomena the underpin cancer pain classification both in research studies and also in clinical practice. While both adding more domains of proven relevance or validity and conducting more standardized assessment of constituent domains have merit, such additional burden may deter some users for completion of the ECS-CP ratings. Again, this is a potential trade-off situation; should accuracy of assessment be sacrificed to maintain a tool that is briefer to administer? The answer is likely to be an optimal balance between the extremes of brevity and accuracy.

The Milan conference advocated the use of similar assessment methods in classifying cancer pain both in clinical practice and research [38]. Furthermore, a knowledge translation workshop in Edmonton reached the same conclusion [124]. In addition to the potential to generate confusion with two tools, the difficulty in promoting the implementation of two tools as opposed to one is a valid argument to develop a single tool. However, the rigorous assessment requirements of a research study may not always be feasible in routine clinical practice; it may be possible that the same tool can be used for both practice and research but that more rigorous assessment might be required when the tool is used in the context of a research study.

Implementation of evidence-based research findings into everyday clinical practice is potentially a slow process and calls for a variety of implementation strategies including consensus among policy makers; generation and uptake of educational practice guidelines; social media platform promotion of developments; and identification of initial champion users. It is seven years since the first ECS-CP knowledge translation symposium and eight years since the Milan conference on cancer pain assessment and classification; in light of new data and some apparent waning of enthusiasm, further meetings are indicated to harness collaborative input.

6. Five-year view

In the 2009, Milan conference [38], all of the ECS-CP domains were recommended for inclusion as either core or candidate domains in a virtual or ideal future cancer pain classification system; furthermore, the ECS-CP was endorsed as a model template. However, the name Cancer Pain Assessment and Classification System (CPACS) was proposed for this ideal classification tool. In the interim, further studies have been published in relation to the ECS-CP and its constituent domains [26–35,110], and three diagnostic models exemplifying chronic cancer associated pain syndromes have been proposed: bone pain, pancreatic cancer pain and chemotherapy-induced peripheral neuropathy [125]. There would appear to be some fragmentation of effort in developing a universally acceptable cancer pain classification system. Hopefully, over the coming five years a consensual approach will be regenerated and both recent and new study data will help to inform and advance the process of cancer pain classification for clinical practitioners, researchers and administrators.

Key issues

• Classification of cancer pain is regarded as a critical component of a comprehensive cancer pain assessment and aids therapeutic decision-making in relation to the choice of interventions, for example, targeting a neuropathic pain mechanism or incident pain or psychological distress. Furthermore, the ECS-CP domains have been demonstrated to have an association with pain management outcomes, particularly the achievement of stable pain control.

• From a research perspective, a cancer pain classification system enables standardized reporting. This may lead to better study sample characterization and possibly enable stratification of patients based on their pain and related characteristics; overall, this may enable more valid comparisons to be made across study samples.

• In an international conference on cancer pain assessment and classification in 2009, the most relevant pain outcomes were deemed to be pain intensity, pain relief and the temporal pattern of pain. Other recommendations included the numerical rating scale (0–10) for average pain intensity assessment; In the assessment of pain response, a reduction of ≥ 50% was considered a ‘substantial decrease’ and ≥ 30% ‘a meaningful decrease’ in pain intensity. Controlled or manageable pain was considered as average pain ≤ 3 (on a 0–10 scale); pain ≥ 4 but ≤ 7 was considered moderately controlled and pain ≥ 7 was considered inadequately controlled pain.

• Recent data suggests that for cognitively intact patients the personalized pain goal has feasibility as an outcome in relation to cancer pain classification and in response to therapeutic intervention; compared to many of the previously studied clinical response outcomes, it is inherently more closely aligned with patient goals.

• The studies conducted regarding the ECS-CP to date have mostly highlighted the need for better standardization in assessing its specific domains. Further research is necessary regarding many of its domains and its related outcomes.

• The potential to develop an index score based on summation of negative ECS-CP predictors warrants further research.

Declaration of interest

P Lawlor has received support from the University of Lisbon and the Gulbenkian Foundation for funding a visiting academic appointment with the University of Lisbon. P Reis-Pina received honoraria from Laboratórios Vitória, S.A. Portugal and Grünenthal, S.A. Portugal and is on the Speaker’s Bureau for Grünenthal, S.A. Portugal. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Acknowledgments

The authors wish to thank R Fainsinger and C Nekolaichuk for their review of the manuscript prior to submission.

Additional information

Funding

This paper was not funded.

Appendix Summary of Specific Literature Search Strategies for Individual Current and Additional Potential Domains of Edmonton Classification System for Cancer Pain (ECS-CP)

Table

Medline was searched from 1989 to November 2017. Searches were conducted in relation to each domain item of the ECS-CP in addition to pain intensity and age using relevant terms (“neuropathic” or “neuropathy”; “incident” or “breakthrough” or “episodic”; “psychosocial” or “psychological” or “distress” or “emotional” or “coping” or “suffering” or “demoralization” or “spiritual” or “existential” or “psychiatric” or “depression” or “anxiety”; “addiction” or “substance abuse” or “substance misuse” or “substance disorder” or “alcohol” or “chemical coping” or “dependence” or “CAGE alcohol questionnaire”; “cognitive” or “delirium” or “confusion” or “capacity”; “age” or “elderly” or “older” or “younger”; “pain intensity”) for these in combination with “cancer” or “oncology”; and “classification” or “assessment” and “pain”.