Introduction

Pain is common in patients with advanced cancer. [1] Pain might be associated with a poorer prognosis, but the evidence is mixed. [2,3,4,5,6] Opioids, used to reduce pain, [7] might reduce immunity and survival, which might limit their use and increase suffering. [8,9,10,11,12,13] A systematic review showed that, although findings from individual studies were variable, opioids tended not to affect survival in the last days to weeks of life, but there was a possible association with shorter survival with longer term opioid-use. [14] These were mostly poor quality studies where the effect of opioids on survival was not the primary outcome. [14] The findings of this systematic review are supported by subsequent publications. [13, 15,16,17] Retrospective studies in patients in the last weeks of life have shown no statistically significant effect of opioids on survival. [15, 16] Prospective studies of patients with advanced cancer, in the last months of life, reported that patients taking opioids had a decreased survival compared to those not taking opioids. [13, 17] Thus, data to date report an association between opioids and survival in patients with advanced cancer only over months of administration.

Patients with aggressive or advanced disease are likely to have increased pain (triggering opioid prescribing) and shorter survival, confounding an independent association between opioids and survival. [6, 18,19,20,21,22] Symptomatic patients may also have worse Karnofsky Performance Status (KPS), (a measure of functional status), itself an independent predictor of survival. [4, 23,24,25,26] Long-term data exploring the association between pain, opioids and survival are limited, but it is vital to address this concern to more accurately judge the benefits and risks of opioids in advanced progressive disease. [8]

Aim

To investigate the relationship between pain, analgesic use and survival within the European Palliative Care Cancer Symptom (EPCCS) study. Our null hypothesis was that there is no relationship between these variables and survival.

Methods

Study design

Secondary analysis of participants in the prospective, longitudinal EPCCS cohort study. [27] Site recruitment, sites and patient eligibility criteria have been published. [27] The first clinical assessment (baseline) was performed upon study inclusion. Subsequent follow-up, either at hospital or by mail were monthly ± one week, for at least six months if possible.

Eligible participants were consenting adults (≥18 years) with advanced, incurable cancer enrolled in the centre’s palliative care programme and able to comply with study procedures. Patients with severe cognitive impairment (<4/8 on the four-item version of the Mini-Mental State Examination) and thought to be imminently dying by the clinical team were excluded.

Demographic data (age, sex) and clinical data (type and stage of cancer at diagnosis) was collected at baseline. The baseline demographics have been published previously. [27] Data collected at each visit can be seen in Supplementary Table 1.

A retrospective recording of date of death was performed in February 2014, around 6 months after the last patient was included. Details of these have been published. [27]

Statistical analyses

The characteristics of the patients are presented for the baseline assessment using mean (sd), minimum and maximum or n (%).

Survival was determined from the date of first assessment to the date of death or date of last follow-up. [, 28] To examine the association between pain, analgesics and time to death a multilevel Weibull survival analysis model was used. Multilevel models consider repeated observations to be ‘nested’ within participants and accommodate data in which the number and timing of observations vary among participants. Hazard ratios and 95% confidence intervals (CIs) are presented. Model fitting procedures were carried out using the mestreg command of STATA/SE-14. To account for individual heterogeneity, a random effect at the individual level was included. The statistical significance of each covariate was tested with a t-test (estimate/standard error). Statistical significance was tested on the univariable analysis for each covariate and multivariable models. Models 1–3 explored the relationship between pain scores and use of each type of analgesic medication (non-opioid analgesics, opioids, co-analgesics) and the association with survival, adjusted for age, sex, stage and KPS. The interaction of pain and each type of analgesic medication was also explored. The full model (Model 4) included all three types of analgesic medication (non-opioid analgesics, opioids, co-analgesics) and pain, adjusted for age, gender, stage of cancer and KPS. Model 5 included, in addition to the variables in Model 4, C-reactive protein (CRP), to explore the influence of systemic inflammation/disease activity on survival.

All analyses were undertaken on STATA SE (StataCorp.2015. Stata Statistical Software: Release-14. College Station, TX:StataCorp LP).

Ethics

Ethics approval was obtained at each EPCCS recruiting site. The study was performed according to the Declaration of Helsinki and was registered in the ClinicalTrials.gov database (No.NCT01362816). No further ethical approval was necessary for this secondary analysis of anonymised data [http://www.hra.nhs.uk].

Results

Patient characteristics

The original study included 1739 patients with baseline data (mean age 65.8 years (SD: 12.4) range 21–97; men 50%). The baseline characteristics are shown in Table 1. During the study, 1090 patients died (date of death missing for 25) and 339 were alive at the last follow-up. Information about vital status was missing for 310 patients, hence there are 1404 with a date of death or last follow-up date for the remaining analysis. Table 1 shows this group was comparable to the total sample on the key characteristics.

Table 1 Descriptive characteristics at baseline
Full size table

Severity of pain and analgesic use

Baseline

Pain was reported by 76% (Table 1). Table 2 shows that 60% of patients were taking opioids, 51% were taking non-opioid analgesics and 24% co-analgesics. There was a statistically significant association between the presence of pain and each type of analgesic medication (Table 2). Patients were more likely to take analgesic medications with a more severe pain intensity (p < 0.001).

Table 2 Brief Pain Inventory (BPI): Average pain in the last 24 h by type of analgesic medication at baseline
Full size table

The relationship between pain, analgesic use and survival

The mean survival from baseline was 46.5 (1.5) weeks (95% CI:43.6–49.3). From the last assessment, the mean survival was 38.8 (1.5) weeks (95% CI:35.9–41.8).

Table 3 shows the univariable analysis. Increasing age, male gender and a higher KPS were associated with a decrease in the hazard of death (p < 0.001).

Table 3 Univariable analysis for covariates and survival
Full size table

There was a higher hazard of death, associated with moderate (1.24 (1.07, 1.45), p = 0.004) or severe (1.28 (1.05, 1.55), p = 0.013) pain, compared to no pain. For specific pain classifications, compared with no pain, those with nociceptive pain had an increased hazard of death (1.21 (1.06, 1.38), p = 0.005), but this relationship was not seen in those with neuropathic pain, either alone or in combination with nociceptive pain (p = 0.224). Opioid (1.88 (1.64, 2.17), p < 0.001) and co-analgesic use (1.22 (1.05, 1.41), p = 0.007) were associated with a higher hazard of death; non-opioid analgesic use (p = 0.142) was not. The survival curves for each type of analgesic medication, pain severity and pain mechanism illustrate these findings (Fig. 1).

Fig. 1
figure 1

Survival curves (days) for use of each type of analgesic medication (non-opioid analgesics, opioids, co-analgesics), average severity of pain in the last 24 h Brief Pain Inventory (BPI) and pain mechanism (p-values are presented in Table 3)

Full size image

To explore the relationship with pain, analgesics and survival, multivariable models were undertaken for each analgesic group and BPI pain and the interaction (models 1–3; Table 4), adjusted for age and gender, stage of cancer and KPS. As there was a strong association with BPI pain (p < 0.001), Edmonton Classification System for Cancer Pain (ECS-CP): Mechanism of pain was excluded in the models and BPI average pain in the last 24 h was used as the key variable for pain measurement.

Table 4 Multivariable analysis (hazard ratio (95% CI)) to investigate hazard of death †
Full size table

Model 1 (Table 4) shows that opioid-use was statistically significant (HR = 1.74 (95% CI:1.38–2.21), p < 0.001) and Model 2 shows that co-analgesics use was statistically significant (HR = 1.37 (95% CI:1.02–1.84), p = 0.034). Model 3 shows that non-opioid analgesics was not statistically significant (p = 0.250). In models 1–3, the interaction between pain and analgesic use was not statistically significant, so was not included in the full model (Model 4; Table 4). Model 4 included each type of analgesic medication and pain, adjusted for age, gender, stage of cancer and KPS. In this model, only opioid-use was associated with an increased hazard of death (HR = 1.59 (95% CI:1.38–1.84), p < 0.001).

C-reactive protein (CRP) was excluded from the main analysis above as data were only available at baseline in a small subgroup (n = 219; 16%) who had a date of death or last follow-up. An exploratory sensitivity analysis with a multivariable model including CRP (Table 4, Model 5) revealed CRP was statistically significantly related, with a higher hazard of death for a unit increase in CRP (Hazard ratio = 1.004 (95% CI:1.002–1.006), p = 0.001, Model 5). Only opioid-use remained statistically significant in this model, although the strength of relationship between decreased survival and opioid-use weakened (HR = 1.38 (95% CI:1.03–1.85), p = 0.029).

Discussion

Overall analysis of these secondary data from unselected adults with advanced, incurable cancer, treated in palliative care units, showed that the only independent association with poorer survival was opioids. Conclusions cannot be drawn regarding causality from these observational data and thus should be interpreted with caution. A subgroup analysis of those who had also had CRP measures showed that the strength of relationship with opioids reduced. This raises the hypothesis that the observed relationship between opioids and survival may be driven by both tumour-related factors (inflammatory response from disease activity) and treatment-related factors (opioids).

Effect of demographics on survival

There was an increased hazard of death (1.96 (1.60, 2.40) for males compared to females; breast cancer had better survival than all other cancers. This is consistent with published literature. [29] This might be partly explained by some cancers in women being more amenable to treatment. [30] Older age was also associated with a shorter survival but the hazard ratio was marginal. Being older has been shown to correlate to decreased survival, but tumour site, type and stage of disease are also important. [4, 29,31,31] The relationship in our study might be weaker as they were a less well-mixed cancer population and mostly in the last year of life. Metastatic/disseminated disease was not statistically associated with a shorter survival but a higher hazard ratio was observed for metastatic/disseminated, again consistent with published literature. [4, 29, 31]

Effect of opioids and pain on survival

In the adjusted model, opioid-use was associated with an increased hazard of death; non-opioid, co-analgesics and severity of pain were not. Although this was adjusted for known variables (age, gender, stage of cancer and KPS), there are unknown confounders such as the pre-opioid pain score for patients on opioids at baseline, the clinical decision making, monitoring and compliance that surrounds opioid-use, opioid dose, type, or route of administration. The finding that opioid-use was associated with decreased survival in patients in the last months of life is supported by previous studies. [13, 17]

Pain severity and opioid-use were not related (Table 2); opioids reduce pain, so opioid-use might be a bystander effect. [6, 32] In the univariable model, moderate and severe pain were associated with a decreased survival (Table 3); however, in the multivariable model the effect of opioids on survival dominates (Table 4). Some previous studies have shown an association between higher pain scores and a poorer prognosis, but the evidence is mixed. [2,3,4,5,6] The lack of association between pain and survival in this analysis might be because although opioids are triggered by pain, their subsequent use reduces pain and so in themselves confound the relationship between pain and survival; opioids have taken over as proxy for disease progression. Clinically, increased pain triggers opioid-use, opioids in turn reduce pain, as opioid dose increases this potentially displaces any subsequent independent effect of pain and survival. It is still the underlying pain that dictates opioid-use (even though intensity is reduced), and it’s the underlying disease that dictates pain. There are many causes of cancer-related pain, the physical aspect being modulated by psycho-social-spiritual factors. [1, 7, 33] Physical pain related to the cancer may be due to both extent and the characteristics of the tumour; an aggressive inflammatory tumour may be more painful. [18,19,20,21,22, 34]

There was an increased hazard of death for nociceptive pain (combined with visceral and/or bone or soft tissue pain) compared to no pain, but no increased hazard for neuropathic pain syndrome with or without any combination of nociceptive pain compared to no pain. However, numbers of patients with neuropathic pain might have been too small (19% vs 50% nociceptive pain) to detect an effect. Pain characterization was limited; one question from the ECS-CP on mechanism of pain on clinical assessment. [35] There are different causes for neuropathic pain in patients with cancer, some of these might not be associated with systemic inflammation and thus less impact on survival. [1]

Effect of CRP on survival

In the exploratory sensitivity analysis increasing CRP was associated with worse survival (p < 0.001) and the effect size for opioid was reduced. This suggests systemic inflammation might be an important confounder in the opioid-survival relationship. The effect size for CRP was small with wide confidence intervals in keeping with an exploratory analysis with a small sample (n = 219). However, the findings are consistent with the literature observing a relationship between higher CRP and worse prognosis. [26, 36, 37] Advanced disease does not fully explain survival, but tumour activity might be more important. [18, 19] Increased tumour activity might be a driver for systemic inflammation which could be associated with a shorter survival and be painful. [18, 19] This exploratory analysis suggested that that the opioid-survival relationship is likely modified by disease activity. This indicates that information on disease activity in needed in future studies.

Limitations

This study was not designed to assess the effect of opioids from initiation, with survival as the primary endpoint. [14] Multiple interacting factors influence survival such as tumour, type of analgesic medication and other treatment, or patient and clinician/service characteristics. Tumour related factors include the type, stage, location and aggressiveness of the cancer. There are insufficient numbers in this dataset to analyse by type and only a subgroup (n = 219) had CRP measurements. Patient characteristics include co-morbidities, the meaning of the pain, depression, coping, family, as well as compliance with medication. We have accounted for the variables for which we have sufficient data, but there are other confounders, which are either unknown, or for which the dataset is too small to support analysis. A larger prospective cohort is needed to address this limitation.

Although consecutive patients were recruited, those with cognitive impairment were excluded. There was limited information collected on analgesic medications. It is only known if patients were on opioids, non-opioid analgesics or co-analgesics. The opioid dose and type, doses or type of other analgesics, their titration, the clinical situation, monitoring, and how side effects were managed were not recorded. Thus, dose-response relationships could not be determined. Cancer pain is often mixed nociceptive, neuropathic and inflammatory, and clinical, radiological and neuroanatomical description to accurately determine the mechanism of pain was beyond the scope of this study. [35] There might be a predominant element but in the collected data nociceptive pain included visceral and/or bone or soft tissue pain and these might have neuropathic elements. Information about CRP was only on a small subgroup of patients, this limits the inferences that can be made about tumour activity and systemic inflammation.

Implications for clinical practice

This study raises a preliminary hypothesis that the observed association between opioids and survival may be partly driven by tumour-related inflammation; causality cannot be attributed. In clinical practice, patients cannot be left in pain, it is thus important that patients who have cancer-related pain are given effective analgesia and monitored appropriately to minimize potential harm. Furthermore, epidemiological data from UK and Canada suggest that only 42–48% of patients that die from cancer receive an opioid, and median treatment duration is around 9 weeks. [38,40,40] These studies suggest that under-treatment with opioids rather than over-treatment might be the more important clinical challenge. Opioid-use is likely to reflect more aggressive painful disease and is thus a “biomarker” of a poor outcome, patients still need to be monitored carefully as they are likely to deteriorate due to their underlying cancer. Figure 2 shows the associations between cancer, inflammation, pain, opioids and survival.

Fig. 2
figure 2

Associations between cancer, inflammation, pain, opioids and survival

Pain is common in patients with advanced cancer, some of this is via inflammation caused by the immune response to cancer. [1] Opioids are prescribed for pain. [7] Survival is dependent on multiple factors, including the underlying cancer, the systemic inflammatory response and opioids. [8,9,10,11,12, 14, 18,19,20,21,22] Pain might also affect survival, [2,3,4,5] but this is offset by opioid-use in this analysis.

Full size image

Implications for research

Large prospective cohort studies with the effect of opioids on survival as a primary outcome, where pain is known before and after opioids have started, is needed to understand the relationship between opioids and survival further. Careful depiction of the clinical situation, opioid dose, titration, and monitoring for individual patients is needed to further understand the relationship of opioids with clinical outcomes. As this exploratory analysis indicates systemic inflammation related to tumour activity might be a driver for increased pain, it is recommended that inflammatory markers (including CRP) should be measured.

Conclusions

This secondary data analysis of the EPCCS data set of adults with advanced, incurable cancer, showed an association of opioid-use with decreased survival, although causality cannot be determined. With a sensitivity analysis of a subset with CRP measures, the relationship with opioid-use reduced. Tumour-related systemic inflammation could be the independent variable explaining the observed relationship with opioids. Further prospective study is needed to test this hypothesis and accurately judge the effects of opioids, pain and tumour-related inflammation in patients with advanced progressive disease.