BMJ Support Palliat Care 2:351-355 doi:10.1136/bmjspcare-2012-000212
  • Research

Lack of association between genetic variability and multiple pain-related outcomes in a large cohort of patients with advanced cancer: the European Pharmacogenetic Opioid Study (EPOS)

  1. Pål Klepstad1,5
  1. 1European Palliative Care Research Center (PRC), Norwegian University of Science and Technology (NTNU), Trondheim, Norway
  2. 2Department of Laboratory Medicine, Children's and Women's Health, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
  3. 3Section of Population Health, Institute of Applied Health Sciences, University of Aberdeen, Aberdeen, UK
  4. 4Department of Oncology, St. Olav's University Hospital, Trondheim, Norway
  5. 5Department of Anaesthesiology and Emergency Medicine, St. Olavs University Hospital, Trondheim, Norway
  1. Correspondence to Torill Fladvad, Pain and Palliation Research Group, Department of Laboratory Medicine, Children's and Women's Health, Faculty of Medicine, Norwegian University of Science and Technology (NTNU), Trondheim N-7489, Norway; torill.fladvad{at}


Objective This study examined whether the choice of pain-related outcome to represent opioid efficacy influenced findings in a genetic association study. Data from the European Pharmacogenetic Opioid Study, which used opioid dose as the outcome, were analysed in respect of six alternative outcomes: average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief and pain interference.

Design Cancer pain patients using an opioid for moderate or severe pain were included. The pain outcomes were obtained using the Brief Pain Inventory. Genetic variation was analysed for 112 single nucleotide polymorphisms (SNPs) in 25 candidate genes relevant for opioid efficacy. The patients were randomly divided into a development and a validation sample and linear regression was used to compare the equality of means in the six outcomes. The influence of non-genetic factors was controlled for, the regression analyses were stratified by country, and the results were corrected for multiple testing.

Results 2201 cancer pain patients were included. Their mean age was 62.4 years and mean average pain was 3.5. None of the examined SNPs exceeded p values corrected for multiple testing for any of the outcomes.

Conclusions None of the outcomes were associated with variation in the selected SNPs, as previously shown for opioid dose. Thus, we observed that findings related to associations between genetic variability and opioid efficacy were consistent for several alternative outcomes.


Opioids are the most used analgesics for cancer pain,1 but the dose required to achieve pain control varies between individuals.2 The variation in required opioid dose is related to several clinical factors such as pain intensity, neuropathic pain, breakthrough pain and psychological distress.3 ,4 The efficacy of opioids may also be related to genetic variability in genes coding for enzymes involved in opioid metabolism,5 opioid receptors,6–8 proteins transporting opioids through the blood–brain barrier,8 ,9 and proteins in other biological systems influencing the effects of opioids.9–13 However, the results from studies on genetics and opioids are conflicting, and the recent large European Pharmacogenetic Opioid Study (EPOS) of 2294 patients failed to observe any consistent relationships between putative candidate genes and opioid dose in cancer pain patients.2

The most studied genetic variation is rs1799971 (A118G) in the OPRM1 gene. Early preclinical data14 and clinical data6 suggested that patients with the variant allele of this single nucleotide polymorphism (SNP) needed more opioids to relieve pain. Later studies both supported15 and contradicted this finding.16 In a systematic meta-analysis, Walter and Lötsch did not find consistent evidence for the effect of rs1799971 on opioid efficacy.17 Also in the EPOS study, the largest study yet on genetics in cancer pain, no evidence for an association between rs1799971 and the need for opioids was observed. This lack of evidence for the most extensively studied genetic variant related to opioid pharmacology, suggests that it is difficult to reach firm conclusions concerning the impact of genetic variability. This lack of consistency also precludes the use of genotyping to guide opioid treatment.

Conflicting results from studies on genetics and opioids can be explained by several factors. Studies are performed in different populations, treatment strategies differ between centres, and the design of the genetic analyses varies. The identity and number of analysed SNPs vary and some studies combine SNPs in haplotypes.10 Finally, studies on genetic variability and opioid efficacy use different phenotypes as outcomes to assess opioid efficacy. In genetic clinical research the choice of phenotype is crucial in order to obtain results that truly reflect the biological phenomenon addressed in the study. In studies on cancer pain these outcomes may be opioid dose,2 pain relief,8 ,13 various assessments of pain intensity,17 pain interference of health related quality of life9 and the need for opioid rotation.11 As reflected by the number of outcomes used in previous studies, a consensus on which outcomes should be used in clinical studies on genetics and opioids has not yet been established.

The EPOS study included SNPs from 25 candidate genes for opioid efficacy.2 In the analyses of the development sample of 1475 patients, nine SNPs dispersed in HTR3E, ADRA2A, DRD2, HTR3B and COMT, showed an association with opioid dose, prior to correction for multiple testing. However, none of the associations exceeded the Benjamini–Hochberg false discovery rate (FDR) criteria or were replicated in the validation sample of 726 patients. This shows that observed associations are at risk of being false positives due to multiple testing or findings that are not replicated in confirmatory studies. In the EPOS study, opioid dose was chosen as the primary outcome to assess opioid efficacy. If all patients were ideally titrated to obtain adequate pain relief, opioid dose should reflect both nociceptive stimuli and opioid sensitivity. However, as in other cancer patient cohorts, pain intensity varies and a large proportion of the patients report moderate or severe pain. Therefore, poor opioid sensitivity could also result in more pain for those patients who are not adequately titrated with an opioid. This is an argument for also applying pain intensity as an outcome in studies of genetics and cancer pain.

In order to examine if the lack of positive associations between genetic variability and opioid efficacy could be influenced by the choice of opioid dose as outcome, we have re-examined the data from the EPOS study applying the outcomes previously used in other studies on genetic variability and opioid efficacy.


Patient samples

The analyses were performed on data from the EPOS study.2 The study included patients from 17 centres in 11 European countries. Adult patients (>18 years) with a malignant disease using an opioid for moderate to severe pain (step III on the WHO treatment ladder for cancer pain) were recruited. Patients unable to speak the language used at the study centre were not eligible for inclusion.

Study procedure

The patient's age, gender, weight, height, ethnicity, cancer diagnosis, and known localisation of metastases were recorded. The dose and route of scheduled opioids for the last 24 h, incident to breakthrough pain, and the duration of opioid treatment were recorded. Oral opioid equivalent morphine doses were calculated using standard tables.18 The mechanism of pain was registered as recommended in the revised Edmonton Staging System for Cancer Pain (visceral pain, bone or soft-tissue pain, neuropathic pain, mixed pain, unknown pain),19 cognitive function was assessed by the Mini Mental State Examination,20 and performance status was assessed by the Karnofsky Performance Status.21 For all instruments a validated version in the language of each study centre was applied. Serum creatinine and albumin concentrations were determined using standard analytical methods. The six studied outcomes of pain intensities and interference were assessed using the Brief Pain Inventory (BPI),22 where the patients give scores from 0 (no pain) to 10 (pain as bad as you can imagine) to describe their pain at its worst and least in the past week, their pain on average and pain right now. Pain relief is described in percent for the last week. Pain interference consists of seven sub-domains for description of interference with everyday life, with scores from 0 to 10, adding up to a maximum of 70.

Genotyping procedures

DNA was extracted from EDTA whole blood using the Gentra Puregene Blood Kit (QIAGEN Science, Germantown, Maryland, USA). The choice of the candidate SNPs was based on the expected clinical relevance of the variant alleles (allele frequency >0.10), previously described associations or putative functional effects related to pain and opioid pharmacology.2 The selected 122 SNPs were in the genes OPRM1, OPRD1, OPRK1, ARRB2, GNAZ, HINT1, STAT6, ABCB1, COMT, ADRA2A, MC1R, TACR1, GCH1, DRD2, DRD3, HTR3A, HTR3B, HTR2A, HTR3C, HTR3D, HTR3E, HTR1, HTR4, HRH1 and CNR1. The reference SNP ID number (‘rs no.’) for each selected SNP is given in online supplementary table S1. Genotyping was performed with the SNPlex System (Applied Biosystems, Foster City, California, USA) as described previously.2

Analyses and statistics

The candidate SNPs were rejected if there was evidence of violation of Hardy–Weinberg equilibrium (χ2 test, p<0.0005) or if the minor allele frequency was less than 5%. The outcomes assessed were average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief and pain interference. Linear regression was used to compare the equality of means across genotypes. Non-genetic factors that predicted opioid efficacy were determined using stepwise multiple regression techniques, and significant factors for each outcome (average pain: age, gender, Karnofsky Performance Status; pain right now: gender, Karnofsky Performance Status; worst pain intensity: age, Karnofsky Performance Status; pain at its least: gender; pain relief: gender, time on opioids; pain interference: age, Karnofsky Performance Status, bone metastases) were included as covariates in the analyses of genetic association. The regression analyses were stratified by country.

In order to mitigate the multiplicity issues, the following approaches were adopted. First, for each of the six outcomes, a two-step analysis where the patients were randomly divided into two samples in the ratio 2 : 1, was performed. The larger two-thirds was the ‘development sample’ and used for initial screening of the SNPs, and the remaining third was a ‘validation sample’ that was used for confirmatory tests on the SNPs initially detected. Secondly, an FDR of 10% was used for determining SNPs associated with the outcome measure using the Benjamini–Hochberg method.23 Finally, the genetic models in the analyses were pre-specified as using the co-dominant model, with other models (dominant, recessive, additive) being considered as secondary and exploratory analyses. All analyses were carried out using Stata V.11.0 (Stata, College Station, Texas, USA).


This study was conducted in accordance with the Helsinki declaration. The study protocol was approved at each centre's relevant board for research ethics. Written informed consent was obtained from each subject.



The EPOS study has been described in detail elsewhere.2 In brief, 2294 cancer pain patients were recruited (Denmark n=31, Finland n=30, Germany n=452, Great Britain n=295, Iceland n=150, Italy n=462, Lithuania n=54, Norway n=565, Sweden n=135 Switzerland n=115 and Greece n=5). Ninety three patients were not included in the genetic analyses due to non-Caucasian ethnicity (n=61), small sample (Greek patients, n=5), no available blood sample (n=22) or withdrawal from the study (n=11). The mean age in the study population was 62.4 years, mean Karnofsky Performance Status was 59.0, mean average pain measured with the BPI was 3.5 and the median opioid dose in oral morphine equivalents was 180 mg in the last 24 h (quartiles 80–400). Demographics, clinical characteristics and use of analgesics of the 2201 patients included in the genetic analyses are given in table 1.

Table 1

Patient demographics, use of opioids and symptoms

Of the patients, 827 received morphine as the primary opioid, 445 received oxycodone, 695 received fentanyl and 234 received other opioids. A total of 1292 patients were given their scheduled opioid orally, 901 parenterally, two by the intrathecal route and six by the epidural route. The number of patients using non-opioid analgesics were as follows: paracetamol n=663; NSAIDs n=651; gabapentin or pregabalin n=385; and corticosteroids n=1068. Detailed information on the characteristics of patients using each opioid has been given previously.2

SNPs and associations with pain-related outcomes

Ten SNPs had a minor allele frequency below 0.05 and were excluded from further analyses (see online supplementary table S1). Chromosomal positions and genotype frequencies for the remaining 112 SNPs in the development and validation samples, respectively, are presented in the online supplementary tables. Between two and 10 SNPs were associated (uncorrected p value <0.05) with the outcomes average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief and pain interference in the development samples (table 2), but none of the SNPs exceeded the Benjamini–Hochberg criterion.

Table 2

SNPs in the six assessed outcomes, with an uncorrected p value of less than 0.05 in the development sample

Similar results were obtained with analyses performed assuming dominant, additive or recessive genetic models (data not shown).


In this study, as previously shown for opioid dose,2 none of the outcomes average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief or pain interference were associated with variation in the selected SNPs. Thus, we observed that the lack of associations between genetic variability and opioid efficacy was consistent for several outcomes.

Clinical studies on genetic variability and the effects of opioids use several different domains for defining the phenotype. These include pain intensity,17 pain relief8 ,13 and need for opioid rotation.11 Furthermore, in preclinical studies genetic variability has been examined with outcomes such as experimental pain,24 or proxy measures for opioid effects such as pupil constriction24 ,25 and cerebral pain processing.26 The variation in use of outcomes limits the possibility to compare results between studies or summarise results in meta-analyses. The lack of agreement on which outcome to use in clinical genetic studies on pain and the use of opioids is related both to which domain to assess and to which assessment tool to use. This reflects the general lack of consensus on outcomes in cancer pain studies in general.27–31

For studies where several outcomes may be relevant as phenotypes, there is a risk that the selection of a particular outcome determines the conclusions of the study. Therefore, it is important to predefine a specific outcome in the primary analysis of a study. In EPOS this was, as stated above, opioid dose. In the present analyses we aimed to explore if our previous conclusion was influenced by our choice of opioid dose as outcome. The results of this exploratory analysis also revealed that with the alternative relevant outcomes average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief and pain interference, no significant associations with genetic variability were found. This finding shows that the lack of significant associations applied to several pain-related outcomes. However, as there were no statistically significant positive relationships, we do not know whether a positive relationship with one particular outcome would have been reflected by similar positive relationships for other pain outcomes.

In a cancer pain population it is not possible to distinguish between genetic influence on assessed pain and opioid response. More pain and demands for a higher opioid dose can reflect cancer progression, but can also be due to lower opioid efficacy. Patients with similar cancer characteristics and pain may also report different levels of pain, and the request for opioids can vary because of subjective factors such as different expectations of how much pain should be tolerated, fear of addiction, or not wanting to bother the nurses. Additionally, physician education and practice can influence opioid therapy. A longitudinal study design as applied by Campa et al8 can relate efficacy to changes in symptoms and opioid doses, but again subjective factors will also change over time and response shift may occur. As the cancer progresses, patients may expect and tolerate more pain and adverse effects may be more pronounced at higher doses, which lead to compromises between pain relief and adverse effects. Patients’ responses to questions about pain relief (ie, changes in symptoms) may also be biased by difficulties recalling experiences.32 These concerns illustrate that a genetic associations study involving self-reported outcomes must be interpreted in light of the clinical setting of the study.

This study shows that different outcomes may give similar results, but there is still a need for consensus. Consensus is required if information on genetic variation is to be used in clinical decision making, for example, as has been done estimating warfarin dose, which is based on both clinical and genetic data.33 Also, for genetic association studies as for the other studies in cancer pain, standardisation of the use of assessment tools and cancer pain domains is needed in order that studies can be compared and results combined in systematic reviews.34

We recognise that several methodological considerations other than the choice of outcomes should be addressed. SNP analyses could be expanded for exploring the effect of joint genes7 and it can be appropriate to combine coupled SNPs in haplotypes. Also, different genetic association studies use various statistical approaches to allow for multiple significance testing. This may influence the conclusion of a study; for example, omitting the validation section of the EPOS study would have resulted in several statistical associations being reported.

In conclusion, none of the outcomes average pain intensity, pain right now, worst pain intensity, pain at its least, pain relief and pain interference were associated with variation in the selected SNPs. These findings showed that lack of association between genetic variability and opioid efficacy was consistent for several outcomes in the EPOS study.


The Norwegian Research Council, the European Union's 6th Framework Programme (Contract No. 037777) and The Liasion Committee between the Central Norway Regional Health Authority (RHA) and the Norwegian University of Science and Technology financially supported this study.


  • Contributors TF performed the genotyping; TF and PF analysed the data; TF, PF, FS, SK and PK contributed to interpretation of results and the final version of the paper.

  • Competing interests None.

  • Ethics approval The study protocol was approved by each contributing centre's relevant board for research ethics.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Received 8 February 2012.
  • Accepted 7 August 2012.
  • Published Online First 27 September 2012


Free sample issue
This recent issue is  free  to all users to allow everyone the opportunity to see the full scope and typical content of BMJ Supportive & Palliative Care.

View free sample issue >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.