Regular use of pharmaceutical opioids and subsequent risk of cancer: a prospective cohort study and Mendelian randomization analysis

Summary

Background

Opium consumption was classified as “carcinogenic to humans” by the International Agency for Research on Cancer (IARC). We investigated whether use of pharmaceutical opioids, derived from or synthesized to mimic opium, is associated with cancer risk using separate observational and genetic analyses.

Methods

Observational analysis included 472,955 participants in the UK Biobank prospective cohort (2006–2022). Genetic analysis included 2-sample Mendelian Randomization (MR) analyses using data from 14 independent genome-wide-association-studies (N = 9931–357,292). Adjusted hazard ratios (a-HR) or odds ratios (ORs) associated with regular opioid use were assessed for six established opium-related cancers (lung, pancreatic, bladder, esophageal, oropharyngeal, and laryngeal) and seven non-opium-related cancers (prostate, breast, colon, endometrial, kidney, ovarian, and brain).

Findings

In UK Biobank, regular opioid use was associated with increased risk of opium-related cancers among ever-smoking [a-HR = 1.33 (95% CI = 1.22–1.43)] and never-smoking participants [a-HR = 1.32 (1.10–1.59)], but not non-opium-related cancers [a-HR = 0.96 (0.91–1.02)]. Risk increased with opioid strength [a-HR = 1.30 (1.20–1.40) for weak opioids; a-HR = 1.86 (1.43–2.40) for strong opioids, p-trend < 0.0001] and duration of action [a-HR = 1.32 (1.22–1.42) for short-acting; a-HR = 1.65 (1.24–2.18) for long-acting opioids, p-trend < 0.0001]. Both observational and genetic analyses showed increased risks for most opium-related cancers, including lung [a-HR = 1.39 (1.27–1.53); MR-Odds Ratio (OR) = 1.17 (1.07–1.29)], pancreas [a-HR = 1.24 (1.01–1.52); MR-OR = 1.34 (1.11–1.62)], bladder [a-HR = 1.26 (1.02–1.56); MR-OR = 1.15 (1.03–1.29)], esophagus [a-HR = 1.18 (0.94–1.49); MR-OR = 1.24 (1.01–1.52)], and larynx [a-HR = 1.37 (0.85–2.20); MR-OR = 1.29 (1.04–1.61)]. Except for an inverse association with prostate cancer [a-HR = 0.83 (0.76–0.91); MR-OR = 0.99 (0.92–1.05)], associations were null for non-opium-related cancers.

Interpretation

Regular use of pharmaceutical opioids was associated with elevated risk for cancers caused by opium, but not other cancers.

Funding

US National Institutes of Health, French National Cancer Institute.

Research in context.

Evidence before this study

The use of opioids has increased rapidly over the past two decades, making it a global public health challenge. While the short-term consequences (e.g., overdose and toxicity) are well established, the long-term public health effects remain poorly understood. In 2020, the International Agency for Research on Cancer (IARC/WHO) classified opium consumption as “carcinogenic to humans”. However, this classification applied only to raw opium, which is commonly used in parts of Asia, and explicitly excluded pharmaceutical opioids. Given the chemical and biological similarities between opium and pharmaceutical opioids, concerns have emerged about whether pharmaceutical opioid use may also increase cancer risk. We searched seven electronic databases (PubMed, EMBASE, Web of Science, PsycINFO, International Pharmaceutical Abstracts, CINAHL, and Scopus) on 19 February 2025 for studies assessing the risk of developing cancer in relation to pharmaceutical opioid use in human populations, using the following Medical Subject Headings and relevant terms: (opium OR opiate∗ OR opioid∗) AND (neoplasm∗ OR carcinogen∗ OR malignant∗ OR tumor∗ OR tumour∗ OR cancer∗), with no language or date restrictions. We identified several ecological and registry-based studies, along with a few retrospective analyses, reporting increased cancer incidence or mortality associated with pharmaceutical opioid use. However, these studies were limited by major methodological constraints, including residual confounding (e.g., smoking, alcohol use, comorbidities), susceptibility to reverse causation, and lack of longitudinal exposure assessment. To our knowledge, no prospective, population-based studies have investigated the relationship between pharmaceutical opioid use in non-cancer populations and future cancer risk.

Added value of this study

This is the first prospective, population-based cohort study to examine the association between regular pharmaceutical opioid use and future cancer risk, and the first to apply triangulation using both longitudinal epidemiological data and independent genetic evidence (two-sample Mendelian Randomization). It is also among the first to investigate detailed opioid characteristics, including origin, receptor target, duration of action, and strength, and to explore dose–response relationships. Furthermore, this study used negative control exposures (paracetamol) and outcomes (non-opium-related cancers) to evaluate residual confounding and specificity of associations. In this study of 472,955 cancer-free individuals in the United Kingdom followed for a decade, regular use of pharmaceutical opioids was associated with an elevated risk of developing most cancer types previously identified to have a causal relationship with opium use, but not other cancer types. Associations showed dose-response patterns and remained robust across both observational and genetic analyses, even after extensive sensitivity testing. Associations with paracetamol (negative control exposure), and non-opium-related cancer sites (negative control outcomes) were mostly null, supporting the specificity of the findings.

Implications of all the available evidence

These findings suggest that regular use of pharmaceutical opioids may increase the risk of developing cancers previously linked to opium consumption, raising important concerns for public health. Given the widespread and increasing global use of opioids, particularly for chronic non-cancer pain, even modest increases in cancer risk could have substantial population-level implications. While opioids remain indispensable for managing acute and cancer-related pain, caution is warranted regarding their long-term use in other settings. Importantly, this study evaluated pharmaceutical opioids as a drug class; future research is needed to assess individual opioid compounds, replicate these findings in more diverse populations, and use more precise assessments of opioid exposure to better guide clinical and regulatory decisions.

Introduction

Opioid use has increased rapidly over the past two decades. Over 17% of adults in the United States had at least one opioid prescription filled in 2017, with the opioid crisis officially designated as a public health emergency.¹ In the United Kingdom, the rate of opioid consumption was the highest globally in 2019, with opioid-related poisoning deaths increasing by 388% since 1993 in England and Wales.² High rates of opioid prescribing are also reported in Canada, Australia, and throughout Western Europe.¹ Short-term adverse public health impacts of opioid use are well documented, including overdose deaths and toxicities, but long-term health effects are poorly described.

In 2020, the International Agency for Research on Cancer (IARC/WHO) classified consumption of opium as “carcinogenic to humans”.³ This classification applied only to raw opium, which is the dried latex from the seed capsules of the opium poppy, and an addictive narcotic drug commonly used in central and western Asia. The classification applied to both smoking and ingestion of opium, and was based on sufficient evidence for causing cancers of the lung, larynx, and bladder, and limited evidence for cancers of the esophagus, pancreas, stomach, and pharynx.³ Pharmaceutical opioids are either derived from opium or synthesized to mimic its chemical structure and effects. Therefore, we hypothesized those pharmaceutical opioids might also increase cancer risk.⁴

In this study, we evaluated whether use of pharmaceutical opioids is associated with cancer risk by analyzing 472,955 participants in the UK Biobank cohort study. We assessed whether participants who self-reported opioid use had elevated risk of subsequent cancer diagnoses, and described associations with opioid strength, duration of action, origin, and receptor type. To provide another line of evidence, we also carried out risk analyses using genetic proxies of pharmaceutical opioid use in large cancer-specific genome-wide association studies (i.e., two-sample Mendelian Randomization [MR]), which minimize potential bias from confounding and reverse causation.⁵

Methods

Study design, participants, and measurements

UK Biobank is a population-based cohort study that recruited over 500,000 people aged 40-69 years between 2006 and 2010 from 22 centers throughout the United Kingdom.⁶ All participants provided written informed consent at recruitment and the study protocol was approved by the North West Multicenter Research Ethics Committee of the United Kingdom. This analysis was also approved by the Ethics Committee of the International Agency for Research on Cancer (IARC-WHO). In this analysis, we excluded individuals who had been diagnosed with any cancer (except non-melanoma skin cancer) before enrollment (n = 29,401) and those who withdrew their consent (n = 53).

At recruitment, participants were asked about their regular use of prescription medications, defined as use on a weekly, monthly, or quarterly basis. Self-reported regular medications do not include short-term medications, such as a 1-week course of antibiotics or analgesics. Dosage and duration of use were not collected. Medications were previously mapped to the Anatomical Therapeutic Chemical (ATC) Classification System codes.⁷ Table S1 lists the reported pharmaceutical opioids along with their ATC codes. For this analysis, we assessed pharmaceutical opioids as a class of medications that act on opioid receptors, including both analgesic opioids (primarily ATC category N02) and non-analgesic opioids (primarily ATC categories R05, N07, and A07). Opioids used for anesthesia (ATC category N01) were excluded, as they are not prescribed for regular use.

We also extracted information on regular use of paracetamol and non-steroidal anti-inflammatory drugs (NSAIDs) to account for specific biases (described further below) and for adjustment purposes, as these analgesics may reduce the risk of several cancers (File S1).

To assess the validity of self-reported regular opioid use, we compared it with available prescription records from 171,813 individuals in the 12 months prior to enrollment, and found substantial agreement between self-reported use and having repeated (3 or more) opioid prescriptions [total agreement = 96.0% (95% CI = 95.9%–96.1%), Kappa = 0.66 (95% CI = 0.65–0.67)]. However, for this analysis, we relied only on self-reported regular opioid use data, primarily because the sample size for self-report data was nearly three times larger than that of individuals with linked prescription records. This larger sample enabled analyses by individual cancer sites and population subgroups. Additionally, we found that self-reported data more effectively captured regular use of over-the-counter opioids and non-analgesic opioids (e.g., loperamide).

We classified pharmaceutical opioids based on their strength (weak or strong opioids based on Oral Morphine Equivalent [OME] conversion factor <1 or ≥1), duration of action (short-intermediate-acting or long-acting), origin (natural, semi-synthetic, or synthetic), and target receptors (multitarget or predominantly Mu-receptor agonists).⁸ A detailed description of these classifications for each pharmaceutical opioid is provided in File S2.

Chronic health conditions were considered present if the diagnosis date in linked hospital records preceded the participant's recruitment date, or if the patient self-reported the diagnosis at recruitment (Table S2). To assess socioeconomic position, we used self-reported highest educational achievement, and the area-based Townsend deprivation index. We classified smoking status as never, former mild-moderate, former heavy, active mild-moderate, or active heavy, with mild-moderate and heavy smoking divided at the median of 19 pack-years. We classified alcohol use as never, former, active light, or active heavy use, with heavy use exceeding three days per week.

Follow-up in UK Biobank is conducted primarily through registry linkage.⁶ We restricted the outcomes of interest for this study to the 20 most common incident cancer types (ICD-10 groupings) in UK Biobank, and then further restricted to cancer types with available genome-wide association study (GWAS) results. The final group of 13 cancer types included 6 opium-related and 7 non-opium-related types. We defined “opium-related” cancer types as those identified in the IARC Monograph as having a causal (i.e., lung, bladder, and laryngeal cancers), or positive (i.e., pancreatic, esophageal, and oropharyngeal cancers) association with opium use.³ “Smoking-related” and “obesity-related” cancers were also defined based on IARC classifications (Table S2).^9,10 For pharyngeal cancer, we analyzed risk of oropharyngeal cancer (ICD-10 grouping), which represented the majority of pharyngeal cancer cases.

Statistical analysis

Longitudinal analysis of the UK Biobank data

We used Cox proportional hazards models to estimate the association between medication use and cancer risk. Entry was defined as the date of recruitment and exit as the earliest date of first cancer diagnosis, death, or last follow-up until 3 March 2022. A first set of models combined the opium-related and, separately, the non-opium-related cancer types, while a second set of models was fit for individual cancer types.

We first fit minimally (age and sex) adjusted models. Then, we further adjusted for education, socioeconomic position, smoking status, alcohol use status, body mass index (BMI) categories, self-reported regular NSAID use, and a propensity score for analgesic use based on five chronic health conditions.¹¹ We generated the propensity score by fitting a logistic regression model with regular analgesic use as the outcome and the following chronic conditions as covariates: gastroesophageal reflux disease, inflammatory bowel diseases, acute or chronic pancreatitis, musculoskeletal disorders, and pain in the month preceding recruitment that interfered with usual activities. Pain was included as it could indicate an undiagnosed chronic health condition with possible confounding effect. The propensity score was included as a continuous covariate in all adjusted models. To assess the validity of propensity score adjustment, we evaluated the overlap in propensity score distributions across exposure groups (Figure S1) and examined covariate balance before and after adjustment, using standardized mean differences and variance ratios (Table S3).

We tested the proportional hazards assumption for each model using Schoenfeld's global test. The assumption was met for all covariates in most multivariable models. In a few cancer-specific models, certain covariates exhibited time-varying effects. However, modeling these covariates as time-dependent did not materially alter the hazard ratio estimates for our main exposure of interest (regular opioid use). Therefore, the simpler Cox models were maintained for consistency and interpretability.

To assess statistical significance of dose-response relationships, we used two approaches: (i) treated opioid strength and duration of action as continuous variables, with consecutive integers assigned to consecutive categories, (ii) classified opioid strength and duration of action as three level ordinal factors and tested for linear trends.

To address potential reverse causation (i.e., that participants might be prescribed opioids because of pain from undiagnosed cancer), we repeated analyses after excluding the first 1–4 years of follow-up.¹² To assess potential residual confounding, we stratified analyses by sex, Townsend deprivation index, smoking status, BMI, analyzed risk of non-opium-related cancers related to smoking and obesity, and further adjusted the main models for cancer-specific risk factors. To address potential residual confounding by indication (i.e., that the underlying reason for opioid use could be associated with cancer risk), we assessed the relationship between paracetamol-derivatives as a negative control exposure and risk of opium-related and non-opium-related cancers.

We used multiple imputations to handle missing data (File S3) and compared the results with those of complete-case analysis.

All statistical tests were two-sided, and analyses were performed using Stata version 17 (UK Biobank longitudinal analysis), and R version 4.1.2 (UK Biobank longitudinal analysis - survival package [3.5–7], Mendelian Randomization analysis—TwoSampleMR package [0.6.8]).

Mendelian randomization analysis

In addition to the observational analysis, to provide another line of evidence, we applied Mendelian Randomization (MR) to assess the association between genetic proxies for medication use and cancer risk (details in File S4). Briefly, in our two-sample MR analysis,⁵ we used genetic instrumental variables for self-reported regular use of opioids, paracetamol, and NSAID which were previously identified from a large GWAS of medication use in UK Biobank.⁷ The exposure genetic instrumental variables, listed in File S5, were then extracted from summary statistics of 13 independent GWASs of 6 opium-related and 7 non-opium-related cancer types from European ancestry (File S4). We then estimated odds ratios (ORs) of cancer risk per log-odds increase in genetic predisposition towards medication use using random-effects inverse-variance weighted MR. We further performed various analyses to address potential biases in our MR approach; Instrument strength was assessed via F-statistics,¹³ weak instrument bias was assessed via MR-RAPS (Robust Adjusted Pleiotropy Score),¹⁴ and horizontal pleiotropy and reverse causality were investigated using weighted median regression,¹⁵ MR Egger regression (including intercept test),¹⁶ MR-PRESSO (Pleiotropy RESidual Sum and Outlier),¹⁷ Steiger filtering,¹⁸ and Cochran's Q test.¹⁹ We further conducted multivariable MR, adjusting for genetic IVs for alcohol consumption, BMI, smoking status, and chronic pain (genetic instruments listed in File S6, original GWASs cited in File S4). Finally, we used bidirectional MR to explore the direction of causality between regular opioid use and cancer development.²⁰

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results

Descriptive statistics

Our study included 472,955 UK Biobank participants, among whom 27,846 (5.8%) reported regular pharmaceutical opioid use. The participants were followed for a median of 12.5 years (interquartile range = 11.4–13.3 years), with a total analysis time at risk of 5,602,581 person-years. Compared with those not reporting opioid use, those reporting opioid use were more likely to be older (median age 60 vs. 57 years), to be female (61% vs. 53%), have a BMI exceeding 30 (44% vs. 24%), have experienced pain in the last month (92% vs. 59%), have a smoking history (48% vs. 33%), and have lower socioeconomic position and educational level (Table 1). However, heavy alcohol use was less frequently reported by participants with regular opioid use (27% vs. 45%).

Table 1.

Baseline demographics in relation to reported regular use of opioids in the UK Biobank cohort study.

	Participants reporting regular opioid use	Participants reporting no opioid use
Total number
Number	27,846	445,109
Age
Median (interquartile range), years	60 (53–65)	57 (50–63)
Sex
Male	10,818 (39%)	207,922 (47%)
Female	17,028 (61%)	237,187 (53%)
Socioeconomic positiona
First quartile (lowest socioeconomic position)	10,876 (39%)	107,212 (24%)
Second quartile	6750 (24%)	111,324 (25%)
Third quartile	5554 (20%)	112,545 (25%)
Fourth quartile (highest socioeconomic position)	4627 (17%)	113,477 (26%)
Education
Below high school	9251 (34%)	69,759 (16%)
High school	7226 (27%)	117,460 (27%)
Tertiary non-university	3870 (14%)	72,444 (17%)
University	6776 (25%)	176,664 (40%)
Smoking statusb
Never smoking	11,987 (52%)	246,648 (67%)
Former light smoking (<19 packyears)	2846 (12%)	51,786 (14%)
Former heavy smoking (≥19 packyears)	4747 (20%)	41,820 (11%)
Current light smoking (<19 packyears)	847 (4%)	10,066 (3%)
Current heavy smoking (≥19 packyears)	2808 (12%)	20,131 (5%)
Alcohol consumption status
Never drinking	1943 (7%)	19,084 (4%)
Former drinking	2572 (9%)	14,219 (3%)
Current light drinking	15,632 (57%)	212,904 (48%)
Current heavy drinking	7562 (27%)	197,466 (45%)
Body mass index (BMI)
Normal weight (BMI < 25)	5519 (20%)	149,714 (34%)
Pre-obesity (25 ≥ BMI < 30)	10,120 (36%)	189,749 (42%)
Obesity (BMI ≥ 30)	12,207 (44%)	105,646 (24%)
Experiencing strong pain in the past month
Yes	25,539 (92%)	259,882 (59%)

The number of missing values for each variable was as follows: Townsend deprivation index (n = 590, 0.1% of the entire cohort), education (n = 9505, 2%), smoking status (n = 2756, 0.6%), smoking pack-years (n = 79,269, 16%), alcohol consumption (n = 1573, 0.3%), frequency of alcohol consumption (n = 1276, 0.3%), BMI (n = 2928, 0.6%),and pain in the last month (n = 2075, 0.4%).

^aSocioeconomic position was determined by the quartiles of the Townsend deprivation index, an area-based measure indicating levels of social deprivation.

^bLight and heavy smoking status were categorized based on the median packyears of lifetime smoking that was reported by participants who ever smoked.

Opioid use and risk of combined opium-related and non-opium-related cancers in UK Biobank

In UK Biobank, during a total of 5,602,581 person-years of follow-up combined risk of the six opium-related cancers (n = 7908) increased with any regular use of opioids (adjusted hazard ratio (a-HR) = 1.33, 95% CI = 1.23–1.43). Risk increases were similar for natural, semi-synthetic, and synthetic opioids (p-heterogeneity = 0.34), but stronger for opioids targeting predominantly the Mu-receptor (a-HR = 1.53) than for multitarget opioids (a-HR = 1.26) (p-heterogeneity = 0.0081). A dose-response relationship was observed by opioid duration of action (a-HR = 1.32 for short-intermediate-acting and a-HR = 1.65 for long-acting, p-trend < 0.0001) and for opioid strength (a-HR = 1.20 for weak and a-HR = 1.86 for strong opioids, p-trend < 0.0001) (Fig. 1a). In contrast, no association was observed between opioid use and combined risk of six non-opium-related cancer types (n = 17,875, Fig. 1b). We excluded prostate cancer from this analysis due to an inverse association that differed from other cancer types, as described below.

Fig. 1.

Characteristics of regular opioid use and risk of a) opium-related cancers combined (i.e., cancers of the lung, larynx, bladder, esophagus, stomach, pancreas, and pharynx); b) non-opium related cancers (other than prostate cancer) combined (i.e., cancers of the breast, colon, kidney, liver, brain, and oral cavity). Risk estimates were derived from Cox regression models that were adjusted for age, sex, education, quartiles of socioeconomic position, lifetime smoking status, lifetime alcohol drinking status, body mass index categories, propensity score for chronic health conditions associated with analgesic use, and self-reported regular use of non-steroidal anti-inflammatory drugs (NSAIDs).

Opioid use and cancer-specific risk in UK Biobank and MR-analyses

Considering each cancer type individually, results from risk analyses using UK Biobank and MR (genetic) provided compatible results, both for Opium-related cancers and non-opium-related cancers (Table 2). Regular opioid use was associated with an increased risk for most opium-related cancer types, including lung cancer [a-HR = 1.39 (95% CI = 1.27–1.53); MR-OR = 1.17 (1.07–1.29)], pancreatic cancer [a-HR = 1.24 (1.01–1.52); MR-OR = 1.34 (1.11–1.62)], and bladder cancer [a-HR = 1.26 (1.02–1.56); MR-OR = 1.15 (1.03–1.29)]. Suggestive associations were observed for esophageal cancer [a-HR = 1.18 (0.94–1.49); MR-OR = 1.24 (1.01–1.52)] and laryngeal cancer [a-HR = 1.37 (0.85–2.20); MR-OR = 1.29 (1.04–1.61)] while no association was observed for oropharyngeal cancer [a-HR = 1.08 (0.76–1.66); MR-OR = 1.17 (0.88–1.54)].

Table 2.

Association between regular use of opioids and risk of subsequent incident cancers.

Cancer site	Epidemiologic analysis (longitudinal analysis in UK Biobank cohort)			Genetic analysis (2-sample Mendelian randomization)
	Case (n)	Non-case (n)	Adjusted HR (95% CI)a	Case (n)	Control (n)	OR (95% CI)b
Opium-related cancers
Lung cancer (n = 3977)
No regular opioid use	3419	441,690	1	29,266	56,450	1
Regular use of any opioid	558	27,288	1.39 (1.27–1.53)			1.17 (1.07–1.29)
Pancreatic cancer (n = 1257)
No regular opioid use	1144	443,965	1	9055	7203	1
Regular use of any opioid	113	27,733	1.24 (1.01–1.52)			1.34 (1.11–1.62)
Bladder cancer (n = 1143)
No regular opioid use	1040	444,069	1	13,790	343,502	1
Regular use of any opioid	103	27,743	1.26 (1.02–1.56)			1.15 (1.03–1.29)
Esophageal cancer (n = 1017)
No regular opioid use	928	444,181	1	4112	17,159	1
Regular use of any opioid	89	27,757	1.18 (0.94–1.49)			1.24 (1.01–1.52)
Oropharyngeal cancer (n = 328)
No regular opioid use	303	444,806	1	1980	18,166	1
Regular use of any opioid	25	27,821	1.08 (0.76–1.66)			1.17 (0.88–1.54)
Laryngeal cancer (n = 186)
No regular opioid use	164	444,945	1	2490	18,178	1
Regular use of any opioid	22	27,824	1.37 (0.85–2.20)			1.29 (1.04–1.61)
Non-opium-related cancers
Prostate cancer (n = 11,335)
No regular opioid use	10,847	434,262	1	79,148	61,106	1
Regular use of any opioid	488	27,358	0.83 (0.76–0.91)			0.99 (0.92–1.05)
Breast cancer (n = 9379)
No regular opioid use	8752	436,357	1	122,977	105,974	1
Regular use of any opioid	627	27,219	0.98 (0.90–1.07)			1.07 (0.99–1.16)
Colon cancer (n = 3861)
No regular opioid use	3620	441,489	1	5100	4831	1
Regular use of any opioid	241	27,605	0.93 (0.81–1.07)			1.02 (0.82–1.26)
Endometrial cancer (n = 1512)
No regular opioid use	1386	443,723	1	12,906	108,979	1
Regular use of any opioid	126	27,720	0.96 (0.79–1.16)			1.03 (0.93–1.15)
Kidney cancer (n = 1355)
No regular opioid use	1241	443,868	1	10,784	20,406	1
Regular use of any opioid	114	27,732	1.15 (0.94–1.41)			1.09 (0.94–1.26)
Ovary cancer (n = 977)
No regular opioid use	912	444,197	1	25,509	40,941	1
Regular use of any opioid	65	27,781	0.89 (0.69–1.16)			1.17 (1.03–1.33)
Brain cancer (n = 791)
No regular opioid use	753	444,356	1	12,496	18,190	1
Regular use of any opioid	38	27,808	0.78 (0.56–1.10)			1.08 (0.94–1.24)

n: number; HR: hazard ratio; 95% CI: 95% confidence interval; OR: odds ratio.

Bold values indicate statistical significance at p-value < 0.05.

^aRisk estimates were derived from Cox regression models that were adjusted for age, sex, education, quartiles of socioeconomic position, lifetime smoking status, lifetime alcohol drinking status, body mass index categories, propensity score for chronic health conditions associated with analgesic use, and self-reported regular use of non-steroidal anti-inflammatory drugs (NSAIDs).

^bRisk estimates for ytje genetic analysis were derived from inverse variance weighted (IVW) method.

Among non-opium-related cancer types, we found no associations with opioid use, except for an inverse association with prostate cancer in the epidemiologic analysis and a positive association with ovarian cancer in the genetic analysis (Table 2). Similarly, we found no associations between paracetamol use and cancer risk in either epidemiological or genetic analyses, except for an elevated risk for laryngeal cancer and an inverse association with breast cancer in the epidemiologic analysis (Table S4). While regular NSAID use was not associated with any risk increase, it was inversely associated with the risk of several cancer types in the epidemiological analysis (Table S5).

Table S6 separately presents results of cancer-specific models with minimal adjustment, full adjustment (primary analysis presented above), and extended cancer-tailored adjustments. Further adjustment for cancer-specific risk factors had negligible impact.

Cancer risk in relation to opioid strength and duration of action

Analyses of opioid strength and duration of action supported a dose-response relationship between opioid use and cancer risk for multiple opium-related cancer types (strength in Fig. 2a and Table S7, duration of action in Figure S2a and Table S7). For lung cancer, a-HRs for weak and strong opioids (vs. no regular use) were 1.37 and 1.80 (p-trend < 0.0001), and a-HRs for short-intermediate and long-acting opioids (vs. no regular use) were 1.37 and 1.84 (p-trend < 0.0001). Similar dose-response associations were observed for bladder cancer (a-HR = 1.24 and 1.78, p-trend = 0.021 for strength; a-HR = 1.24 and 1.80, p-trend = 0.020 for duration of action), and pancreatic cancer (a-HR = 1.23 and 1.33, p-trend = 0.038 for strength). Suggestive dose-response associations were also seen for esophageal cancer (a-HR = 1.13 and 2.27, p-trend = 0.063 for strength, a-HR = 1.15 and 1.81, p-trend = 0.097 for duration of action), and laryngeal cancer (a-HR = 1.25 and 3.37, p-trend = 0.088 for strength, a-HR = 1.31 and 2.56, p-trend = 0.13 for duration of action). Modeling opioid strength and duration of action as ordinal variables showed significant dose-response associations for opium-related cancers combined, as well as for lung, esophageal, and laryngeal cancers, but not for non-opium-related cancers (Table S7).

Fig. 2.

Strength of pharmaceutical opioids used and risk of subsequent cancers in subgroups of a) opium-related cancers, b) non-opium related cancers. Risk estimates were derived from Cox regression models that were adjusted for age, sex, education, quartiles of socioeconomic position, lifetime smoking status, lifetime alcohol drinking status, body mass index categories, propensity score for chronic health conditions associated with analgesic use, and self-reported regular use of non-steroidal anti-inflammatory drugs (NSAIDs).

Among non-opium-related cancers, except for an inverse dose-response relationship with prostate cancer for opioid strength (a-HR = 0.84 and 0.71, p = 0.00016), there was no association between cancer risk and opioid strength or duration of action (Fig. 2b, Figure S2b).

Sensitivity analyses

There was no evidence for association between opioid use and risk of non-opium-related cancers that are causally linked to smoking and obesity,^9,10 providing reassurance against residual confounding (Table 3). Further, increased risk of opium-related cancers persisted across all subgroups of smoking status, sex, BMI, and socioeconomic position. Excluding the first 1–4 years of follow-up did not alter the association between opioid use and opium-related cancers, providing reassurance against reverse causation. For individual cancer types, results were comparable after excluding the first 4 years of follow-up and performing complete-case analysis instead of multiple imputation for missing data (Table S8). Restricting the analysis to never-smoking participants attenuated the risk estimate for lung cancer, but strengthened the estimates for pancreatic, esophageal, and bladder cancers (Table S8). Analyzing prostate cancer mortality showed results that were comparable to the incidence (a-HR = 0.79 (0.57–1.09)), providing evidence against screening effects.

Table 3.

Sensitivity and stratified analyses to address the robustness of the association between regular use of opioids at baseline and subsequent cancer risk in the UK Biobank cohort.

Analysis type/subgroupsa	a-HR (95% CI) related to self-reported regular opioid useb
	Opium-related cancers	Non-opium related cancers
	Combined (n = 7908)c	Prostate cancer (n = 11,335)	Non-prostate cancers combined (n = 17,875)d	Smoking-related cancers combined (n = 15,572)e	Obesity-related cancers combined (n = 17,084)f
Assessing reverse causation
All participants	1.33 (1.23–1.43)	0.83 (0.76–0.91)	0.96 (0.91–1.02)	0.97 (0.91–1.03)	0.97 (0.91–1.03)
Dropping the first 2 years of follow-up	1.32 (1.22–1.43)	0.81 (0.74–0.90)	0.96 (0.90–1.03)	0.97 (0.90–1.04)	0.97 (0.91–1.04)
Dropping the first 4 years of follow-up	1.37 (1.26–1.49)	0.83 (0.75–0.93)	0.97 (0.90–1.04)	0.97 (0.89–1.05)	0.97 (0.90–1.05)
Stratification by smoking status
Never smoking	1.32 (1.10–1.59)	0.91 (0.78–1.06)	1.00 (0.92–1.09)	1.00 (0.91–1.10)	0.99 (0.90–1.08)
Ever smoking	1.33 (1.22–1.43)	0.79 (0.70–0.89)	0.93 (0.85–1.01)	0.95 (0.86–1.04)	0.95 (0.87–1.04)
Stratification by sex
Male	1.30 (1.17–1.43)	0.83 (0.76–0.91)	0.99 (0.85–1.14)	1.01 (0.87–1.19)	1.01 (0.87–1.19)
Female	1.37 (1.23–1.53)	–	0.97 (0.90–1.03)	0.97 (0.90–1.04)	0.97 (0.91–1.04)
Stratification by BMI
BMI < 30	1.43 (1.31–1.57)	0.82 (0.72–0.92)	0.94 (0.87–1.03)	0.97 (0.89–1.06)	0.95 (0.88–1.04)
BMI ≥ 30	1.20 (1.06–1.35)	0.86 (0.74–1.01)	0.99 (0.91–1.08)	0.98 (0.89–1.08)	1.00 (0.91–1.09)
Stratification by deprivation index
Lower socioeconomic position	1.33 (1.22–1.45)	0.78 (0.69–0.89)	0.97 (0.89–1.05)	0.97 (0.89–1.05)	0.97 (0.90–1.05)
Higher socioeconomic position	1.31 (1.15–1.50)	0.90 (0.78–1.04)	0.95 (0.87–1.05)	0.97 (0.88–1.08)	0.96 (0.87–1.06)

n: number; a-HR: Adjusted hazard ratio; 95% CI: 95% confidence interval; BMI: body mass index.

Bold values indicate statistical significance at p-value < 0.05.

^aIn all strata the reference category was set as no regular use of opioids at baseline.

^bRisk estimates were derived from Cox regression models that were adjusted for age, sex, education, quartiles of socioeconomic position, lifetime smoking status, lifetime alcohol drinking status, body mass index categories, propensity score for chronic health conditions associated with analgesic use, and self-reported regular use of non-steroidal anti-inflammatory drugs (NSAIDs).

^cOpium-related cancers were pre-defined based on IARC Monographs and include cancers of lung, pancreas, bladder, esophagus, oropharynx, and larynx.

^dNon-opium-related non-prostate cancers include cancers of breast, colon, endometrium, kidney, ovary, and brain.

^eNon-opium-related smoking-related cancers were pre-defined based on IARC Monographs and include cancers of breast, colon, kidney, and ovary.

^fNon-opium-related obesity-related cancers were pre-defined based on IARC Handbooks for Cancer Prevention and include cancers of breast, colon, endometrium, kidney, and ovary.

For the MR analysis, analyses using MR-Egger, Weighted Median, MR-PRESSO, and MR-RAPS generally supported the conclusions from the IVW method, despite decreases in statistical power (Tables S9–S11). MR-PRESSO analyses did not indicate pleiotropy for opium-related cancers, but indicated potential pleiotropy affecting several non-opium-related cancers. The bidirectional MR analysis, with genetic liability for regular pharmaceutical opioid use as an outcome, showed no evidence that developing cancers per se contribute to elevated risk for regular opioid use (Table S12). Finally, adjusting for alcohol use, BMI, smoking status, and chronic pain in multivariable MR analyses did not substantially impact the risk estimates, suggesting these factors are not important confounders (Table S13).

Discussion

We analyzed the relationship between regular use of pharmaceutical opioids and cancer risk among 472,955 longitudinally followed participants in the UK, as well as using genetic data from large GWAS consortia through MR. We found that participants reporting regular opioid use had higher risk of most cancer types previously associated with opium use,³ including cancer of the lung, pancreas, bladder, esophagus, and larynx. In contrast, opioid use was not associated with increased risk for any cancer types not previously linked to opium. Results were similar in a complementary genetic analysis where genetic proxies for opioid use were used as exposure indicators.

Registry data linkage studies in Europe,²¹ Australia,²² and Asia²³ have consistently showed increased cancer incidence or mortality among individuals using pharmaceutical opioids. However, these studies are subject to important biases.²⁴ Confounding may occur due to lack of information on smoking, alcohol use, and comorbidities. It is typically not possible in registry linkage studies to determine the temporality of the associations, in other words, whether the exposure occurs before the outcome.²⁴ Further, these studies lack information about the receptor target, strength, or duration of action of opioids which may further affect the observed associations.

In this study, we analyzed a well-characterized prospective epidemiological cohort to reduce the influence of these biases. Several observations in our study suggest regular opioid use may play a role in the risk of developing opium-related cancers, warranting further investigation in studies with more detailed exposure assessment. These include (i) the genetic analysis gave results highly consistent with the epidemiological analyses, (ii) the risk of opium-related cancers showed a clear dose-response relationship with opioid strength and duration of action, and (iii) sensitivity analyses did not suggest concerns of reverse causation or residual confounding, and (iv) no association was observed between paracetamol (negative exposure control) and risk of opium-related cancer, nor between opioid use and risk of non-opium related cancers (negative outcome control). One unexpected observation was an apparent inverse association between opioid use and prostate cancer risk. The reason for this is unclear, although opioid-induced hypogonadism could be one of the potential explanations.²⁵ We emphasize that our study did not assess the relationship between pharmaceutical opioids and cancer progression (i.e., among patients with cancer) and should not be construed as evidence against prescribing opioids for cancer pain.

The mechanisms by which pharmaceutical opioids may cause cancer remain unknown. Some potential mechanisms for the carcinogenicity of opium use would not apply to pharmaceutical opioids, such as exposure to polycyclic aromatic hydrocarbons (PAHs).⁴ However, the recent identification of a distinct mutational signature in esophageal tumor tissues associated with opium exposure suggests mechanisms beyond the effects of smoking, PAHs, and other known carcinogens.²⁶ Other suggested mechanisms relate to genotoxic effects for certain opium alkaloids,²⁷ which may be shared among natural and semi-synthetic opioids that have similar alkaloid composition.²⁸ Finally, several potential mechanisms are linked to the activation of opioid receptors, which are overexpressed in tumor tissues of some opium-related cancers.²⁹ These effects may be shared among natural, semi-synthetic, and synthetic opioids. Activation of opioid receptors (particularly Mu-receptor) triggers complex intracellular signaling mechanisms, some of which play crucial roles in cancer biology, such as activation of cell cycle progression, angiogenesis, and impairment of immune function.^28,30 Opioids can also activate other intracellular signaling pathways, potentially contributing to tumorigenesis through interactions with non-opioid receptors.^28,30 Paradoxically, tumor-inhibitory pathways, such as the promotion of apoptosis, have been also described by experimental studies.^28,30 Mechanistic studies are needed to improve our understanding of whether, and how, opioids may influence cancer development.

Our study has several limitations. We had no information on duration and frequency of self-reported opioid use, which limited analysis of dose-response relationships. The single-timepoint assessment of opioid use is also a limitation, as opioid use is often dynamic, and continuous assessment during follow-up may more accurately assess the associations of interest. For the genetic analysis, we used the only available GWAS on regular opioid medication use in Europeans. This definition can be considered as genetic predisposition to opioid use, rather than specifically capturing their use and clinical effects, and may encompass traits leading to the need for opioid medication (i.e., a disposition to certain health conditions), and factors leading to accurate self-report. Accordingly, despite extensive sensitivity analysis, we cannot rule out a biasing of our MR results towards the observational. Therefore, future genetic analyses would benefit from drug target MR studies or a more refined GWAS on opioid medication use. Most importantly, our analysis of cancer risk was conducted within a single cohort and evaluated pharmaceutical opioids as a broad class of medications acting on opioid receptors. Future studies should examine individual opioid medications to better inform clinical practice, and replication in diverse populations is essential. To support this, our team is leading the Opioid Cohort Consortium (OPICO), which will include 10 prospective epidemiological cohorts from 4 continents. OPICO will use linked medication dispensing records to achieve detailed characterization of opioid use over time.

Since a randomized trial to evaluate the potential carcinogenicity of pharmaceutical opioids would be neither feasible nor ethical, the highest level of evidence on this topic must come from high-quality epidemiological studies together with triangulation by other methods, including convincing mechanistic evidence. In line with the recently established carcinogenic effects of opium consumption, our study suggests that regular use of pharmaceutical opioids may increase risk for several cancer types. Considering the widespread use of these medications, additional research is urgently needed to further characterize the importance of regular opioid use in cancer etiology.

Contributors

MS and HAR contributed to conceptualizing and designing the study; MS, SAP, and MOP contributed to exposure characterization; MS, RL, VV, JZ, PB, MJ, GDS, and HAR contributed to developing the analysis plan; MS, AD, KA, RL, DM, SV, EE, TD, TEG, LAK, NR, SK, and JZ contributed to data extraction; MS, AD, KA, RL, DM, XF, and JZ contributed to analysis; MS and HAR drafted the manuscript with significant contributions from AD, KA, RL, MJ, and GDS; all authors reviewed the manuscript and contributed to the interpretation of the results. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. MS, AD, KA, RL, DM, XF, and JZ had access to data and verified all analyses and results. MS and HAR are the guarantors. All authors read and approved the final version of the manuscript.

Data sharing statement

This study accessed relevant UK Biobank data under application number 80725. UK Biobank data are available through a procedure described at http://www.ukbiobank.ac.uk/using-the-resource, where timeframe information can also be found.

For the MR analysis, full summary statistics are publicly available via: the Open GWAS database [https://gwas.mrcieu.ac.uk] for prostate cancer (accession number: ieu-b-85), ovarian cancer (accession number: ieu-a-1120), and endometrial cancer (accession number: ebi-a-GCST006464); the European Bioinformatics Institute GWAS Catalog [https://www.ebi.ac.uk/gwas] for lung cancer (accession number: GCST004748) and esophageal cancer (accession number: GCST003739); and the Breast Cancer Association Consortium [bcac.ccge.medschl.cam.ac.uk] for breast cancer. PanScan and PanC4 GWAS data are available through dbGAP (accession numbers phs000206.v5.p3 and phs000648.v1.p1, respectively), Bladder GWAS data are available through dbGaP (accession number phs003342.v1.p1). Application to the relevant GWAS consortium is required for full summary statistics for the remaining cancer sites.

Declaration of interests

Peter Sarich received conference registration support from Sydney Cancer Partners to attend the 2022 New South Wales Cancer Conference and the 2023 New South Wales Cancer Conference. Marianne F. Weber is a member of the Data Development Working Group and the Program Guidelines Working Group of the National Lung Cancer Screening Program Advisory Committee for the Australian Department of Health and Aged Care. Marianne F. Weber received conference registration support from Cancer Council for the Oceania Tobacco Control Conference in 2024 as an invited speaker, and her institution received contract funding from the Australian Department of Health and Aged Care and Cancer Australia, and also received grant funding from the Australian Government Medical Research Futures Fund. George Davey Smith received grant funding from the Medical Research Council and consultation fees from the Bristol Myers Squibb, Relation Therapeutics, and Insitro. Meredith C.B. Adams received funding from the National Institutes of Health (NIH). Other authors declared no conflicts of interest.