Meta-Analysis:
Prescription Opioid and Morphine Use and Risk of Stroke
Ibrahim Krenawi 1, Mimoun Azizi *2
*Correspondence to: Mimoun Azizi, Chief Physician, Senior Consultant Neurologist and Neurogeriatrician, Klinikverbund Südwest, Sindelfingen, Germany.
Copyright
© 2026 Mimoun Azizi, This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Received: 12 October 2026
Published: 01 November 2026
DOI: https://doi.org/10.5281/zenodo.23907221
Abstract
Prescription opioid analgesics are widely used for the management of acute, chronic, and cancer-related pain, yet their potential association with cerebrovascular events remains uncertain. This meta-analysis evaluated the association between prescription opioid/morphine exposure and incident stroke in adults. Eligible observational studies reporting adjusted effect estimates for incident stroke among prescription opioid or morphine users were included, while studies focused on heroin, illicit opium, passive exposure, or opioid use disorder without separable prescription-opioid exposure were excluded from the primary analysis. Adjusted effect estimates were synthesized using generic inverse-variance random-effects methods, with restricted maximum likelihood estimation and Hartung–Knapp inference used for the principal analysis.
Four independent studies contributed to the primary meta-analysis of overall incident stroke. Prescription opioid/morphine exposure was associated with a modest increase in stroke risk, with a pooled adjusted relative effect of 1.10 (95% CI 1.03–1.18). Statistical heterogeneity was low (I² = 0% in the conventional analysis). The association remained above the null when the analysis was restricted to studies reporting hazard ratios (1.11, 95% CI 1.03–1.19) and after exclusion of the largest, highly weighted cohort (1.16, 95% CI 1.05–1.29). Additional evidence suggested associations with specific ischemic and hemorrhagic stroke outcomes; however, differences in study design, opioid exposure, populations, and outcome definitions precluded reliable subtype-specific pooling.
Overall, prescription opioid/morphine exposure was associated with a small but consistent elevation in incident stroke risk. Nevertheless, the exclusively observational evidence remains susceptible to residual confounding, confounding by indication, and exposure misclassification and therefore does not establish causality. Further prospective studies incorporating opioid type, dose, duration, cumulative exposure, and time-varying use are required to define the cerebrovascular implications of prescription opioid therapy more precisely.
Keywords: Prescription opioids; Morphine; Stroke; Cerebrovascular disease; Ischemic stroke; Hemorrhagic stroke; Opioid analgesics; Morphine and Stroke, Opioid and Stroke.
Introduction
Stroke remains a major cause of death and long-term neurological disability, making identification of potentially modifiable risk factors an important clinical and public-health priority. Prescription opioid analgesics are widely used in the management of acute, chronic, and cancer-related pain; however, concerns regarding their adverse effects extend beyond dependence, overdose, and respiratory depression. Emerging observational evidence has raised the possibility that prescription opioid exposure may also be associated with cardiovascular and cerebrovascular events.
The relationship between prescription opioids and stroke remains uncertain. In the prospective REGARDS cohort, prescription opioid use was not significantly associated with incident stroke (adjusted HR 1.04, 95% CI 0.78–1.38), whereas subsequent large population-based studies have reported positive associations between opioid exposure and stroke outcomes. Morphine-specific evidence has also suggested possible associations in selected populations, although findings may be strongly influenced by treatment indication, underlying disease, and opioid exposure intensity [1]. More recent evidence from patients with chronic pain has further linked opioid analgesic use with particular cerebrovascular outcomes, including subarachnoid hemorrhage.
Interpretation of this literature is complicated by substantial differences in study populations, opioid agents, exposure definitions, stroke subtypes, and control of confounding. Importantly, evidence concerning prescribed opioid analgesics should be distinguished from studies of heroin, illicit opium, opioid use disorder, and passive opioid exposure, as these represent fundamentally different exposure patterns and clinical contexts. A focused quantitative synthesis of prescription opioid exposure is therefore necessary to clarify the available epidemiological evidence.
Accordingly, this meta-analysis evaluated the association between prescription opioid/morphine exposure and incident stroke in adults. The primary objective was to estimate the pooled association with overall incident stroke using adjusted study-level estimates. Secondary objectives were to examine the consistency of the association through sensitivity analyses and to characterize available evidence for ischemic and hemorrhagic stroke subtypes.
Methods
Study Design and Eligibility Criteria
Studies were considered eligible according to prespecified criteria addressing the population, exposure, comparator, outcome, and study design. The meta-analysis was designed to evaluate the association between prescription opioid analgesic exposure, including morphine, and the risk of incident stroke in adults. The primary analysis was restricted to studies evaluating prescription opioid exposure to maintain clinical comparability and to avoid combining therapeutic opioid use with illicit opioid exposure or opioid use disorder.
Inclusion Criteria
Studies were eligible for inclusion if they: (1) included adult human participants; (2) evaluated exposure to prescribed opioid analgesics, including morphine or other prescription opioid agents; (3) included an appropriate comparator group without the opioid exposure of interest; (4) evaluated incident stroke occurring after assessment of opioid exposure; and (5) used an analytical observational design, including prospective or retrospective cohort, case-control, nested case-control, or case-crossover designs.
For quantitative synthesis, studies were additionally required to report an adjusted effect estimate, such as a hazard ratio (HR), odds ratio (OR), or risk ratio (RR), with a corresponding 95% confidence interval (CI), or to provide sufficient information for calculation of an effect estimate and its variance. When multiple adjusted models were reported, the most comprehensively adjusted estimate relevant to the prespecified exposure–outcome comparison was selected.
Exclusion Criteria
Studies were excluded if they were case reports, case series, reviews, editorials, commentaries, or non-human studies. Studies without an appropriate comparison group or without sufficient quantitative data for effect estimation were also excluded.
To preserve the clinical specificity of the exposure, studies evaluating heroin, illicit opium, passive opioid exposure, or opioid use disorder without a separately identifiable prescription-opioid exposure were excluded from the primary analysis. Studies examining opioid administration after stroke, opioid treatment for post-stroke pain, or associations between opioid use and prognosis, mortality, or functional outcomes among patients with established stroke were also excluded because they did not evaluate incident stroke risk.
Primary and Secondary Outcome Eligibility
The primary outcome was the incidence of overall stroke. Studies reporting an adjusted association between prescription opioid or morphine exposure and overall incident stroke were eligible for the primary quantitative synthesis.
Secondary outcomes included ischemic stroke and hemorrhagic stroke, with subarachnoid hemorrhage considered separately where reported. Studies reporting only stroke-subtype outcomes could contribute to secondary analyses but were not incorporated into the primary overall-stroke model unless an eligible overall-stroke estimate was also available.
When a study reported both overall stroke and individual stroke subtypes, only the overall-stroke estimate was entered into the primary meta-analysis to prevent duplicate weighting of the same study population.
Management of Overlapping Study Populations
Potential overlap between study populations was evaluated by comparing the underlying database, recruitment and study periods, participant eligibility criteria, exposure-assessment periods, and outcome follow-up periods. When substantial participant overlap was identified or could not reasonably be excluded, estimates from the overlapping populations were not treated as independent observations within the same pooled analysis.
Studies with materially different exposure contrasts, outcome definitions, or analytical designs could be retained as supporting or secondary evidence without being incorporated into the primary quantitative synthesis.
Literature Search and Study Selection
Information Sources and Search Strategy
A comprehensive literature search was designed to identify studies evaluating the association between prescription opioid analgesic exposure and incident stroke. Electronic database searches were supplemented by manual screening of the reference lists of eligible articles and relevant reviews to identify additional primary studies.
The search strategy combined controlled vocabulary, where applicable, with free-text terms relating to prescription opioids and cerebrovascular outcomes. Exposure-related terms included opioid, opioid analgesic, prescription opioid, morphine, oxycodone, fentanyl, tramadol, codeine, and buprenorphine. Outcome-related terms included stroke, cerebrovascular accident, cerebrovascular disease, ischemic stroke, cerebral infarction, hemorrhagic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage. Search terms were combined using appropriate Boolean operators and adapted to the indexing requirements of each database.
The search was structured to identify studies in which prescription opioid exposure preceded the occurrence of stroke. Studies focusing exclusively on opioid administration after stroke, treatment of post-stroke pain, or opioid exposure as a predictor of prognosis among patients with established stroke were not considered relevant to the exposure–outcome relationship of interest.
The final manuscript will report the complete database-specific search strategies, dates of the final searches, and search limits in supplementary material after completion and documentation of the reproducible electronic search.
Study Screening and Selection
Records identified through the literature search were assessed against the predefined eligibility criteria. Potentially relevant studies identified during title and abstract screening underwent full-text evaluation to determine eligibility for inclusion.
During full-text assessment, particular attention was given to the source and definition of opioid exposure, temporal relationship between opioid exposure and stroke occurrence, availability of an appropriate comparator group, definition of the cerebrovascular outcome, study design, and availability of an adjusted quantitative effect estimate.
Studies evaluating heroin, illicit opium, passive opioid exposure, or opioid use disorder without a separately identifiable prescription-opioid exposure were excluded from the primary analysis. Studies evaluating outcomes after an established stroke rather than incident stroke were similarly excluded.
Selection for Quantitative Synthesis
Studies meeting the general eligibility criteria were further evaluated for suitability for quantitative synthesis. The primary meta-analysis included independent studies reporting an adjusted association between prescription opioid or morphine exposure and overall incident stroke, together with a corresponding 95% confidence interval.
Studies reporting only ischemic stroke, hemorrhagic stroke, or another specific cerebrovascular subtype were considered for secondary analyses but were not automatically incorporated into the primary overall-stroke model. Similarly, studies employing substantially different exposure contrasts or analytical designs were evaluated separately to avoid introducing inappropriate clinical or methodological heterogeneity into the primary pooled estimate.
When multiple publications or analyses appeared to originate from potentially overlapping populations, the underlying database, enrollment period, exposure-assessment period, eligibility criteria, and follow-up period were compared. Potentially overlapping estimates were not treated as independent observations within the same pooled analysis.
Documentation of the Selection Process
The study-selection process will be documented using a flow diagram showing the numbers of records identified, duplicate records removed, records screened, full-text articles assessed, studies excluded with reasons, and studies ultimately included in the quantitative synthesis.
No numerical study-selection counts are reported at this stage because the final reproducible database search and deduplication process must be completed before these values can be reported accurately.
Data Extraction and Outcome Definitions
Data extracted from eligible studies included author and publication year, country, study design, data source, sample size, participant characteristics, opioid exposure definition, comparator, follow-up duration, stroke outcome, effect measure, 95% confidence interval (CI), and variables included in multivariable adjustment.
The primary outcome was incident overall stroke following prescription opioid or morphine exposure. Secondary outcomes included ischemic stroke and hemorrhagic stroke, including subarachnoid hemorrhage when separately reported. Stroke definitions reported by the original investigators were retained.
For observational studies reporting multiple statistical models, the most comprehensively adjusted estimate relevant to the prespecified exposure–outcome comparison was extracted.
Hazard ratios (HRs), odds ratios (ORs), and risk ratios (RRs) with corresponding 95% CIs were recorded. When multiple stroke outcomes were reported from the same population, the overall-stroke estimate was preferentially used in the primary meta-analysis to avoid duplicate inclusion of participants. Stroke-subtype estimates were reserved for secondary analyses.
Risk-of-Bias Assessment
Risk of bias was assessed at the study level across key methodological domains relevant to observational research, including participant selection, exposure ascertainment, outcome assessment, control of confounding, and completeness or stability of follow-up and exposure measurement. Particular attention was given to confounding by indication, residual confounding, and potential misclassification of opioid exposure.
Studies were categorized according to the overall extent of methodological concern, with judgments based on the design and limitations reported by the original investigators. Risk-of-bias assessments were considered when interpreting the pooled estimates and sensitivity analyses. Studies were not excluded solely based on risk-of-bias judgments; instead, important methodological limitations were incorporated into the interpretation of the certainty and robustness of the findings.
Statistical Analysis
Adjusted effect estimates and their 95% confidence intervals (CIs) were logarithmically transformed for quantitative synthesis. Standard errors were derived from the corresponding 95% CIs, and study-specific estimates were combined using the generic inverse-variance method. Because clinical and methodological variation across populations and opioid-exposure definitions was anticipated, a random-effects model was used. Between-study variance (τ²) was estimated using restricted maximum likelihood (REML), with Hartung–Knapp adjustment applied to account for uncertainty associated with the small number of studies.
Statistical heterogeneity was evaluated using Cochran's Q and the I² statistic. Sensitivity analyses included restriction to studies reporting hazard ratios and leave-one-out analyses to assess the influence of individual studies on the pooled estimate.
The pooled result was expressed as an adjusted relative effect with a 95% CI because the primary studies reported different relative effect measures. Funnel-plot asymmetry and Egger's regression were not formally assessed because fewer than 10 studies contributed to the primary meta-analysis.
Results
Study Selection and Characteristics
Six studies met the eligibility framework and contributed relevant evidence on prescription opioid exposure and incident stroke. Four independent studies provided sufficiently comparable estimates of overall stroke and were included in the primary quantitative synthesis: Lee et al. (2013), Khodneva et al. (2016), Oh et al. (2025), and a 2025 UK chronic-pain cohort [2, 3, 4, 5]. Two additional studies, Dalgaard et al. (2021) and Moon et al. (2026), provided relevant ischemic-stroke data but were not incorporated into the primary overall-stroke model because of differences in outcome definition, exposure contrast, study design, or potential population overlap [6, 7].
The included evidence encompassed nested case-control and prospective or population-based cohort designs across Taiwan, the United States, South Korea, and the United Kingdom. Study populations ranged from cancer patients receiving morphine and individuals with chronic pain to general population-based cohorts. Exposure definitions similarly varied from morphine-specific treatment to broader prescription opioid analgesic use.
The four studies contributing to the primary analysis reported adjusted associations with overall incident stroke. Lee et al. evaluated morphine exposure among patients with cancer, whereas Khodneva et al. examined prescription opioid use in the prospective REGARDS cohort. Oh et al. contributed the largest population-based analysis of oral or transdermal prescription opioid exposure, while the UK cohort evaluated opioid analgesic use among individuals with chronic pain [2, 3, 4, 5].
Across the primary studies, adjusted relative effect estimates ranged from 1.04 to 1.24. Differences in population characteristics, exposure ascertainment, duration of follow-up, and confounder adjustment were considered when interpreting the pooled estimate and informed the prespecified sensitivity analyses.
Primary Meta-Analysis of Overall Stroke
Four independent observational studies contributed adjusted estimates to the primary meta-analysis of prescription opioid or morphine exposure and incident overall stroke. The individual effect estimates were 1.13 (95% CI 0.97–1.31) for Lee et al. (2013), 1.04 (95% CI 0.78–1.38) for Khodneva et al. (2016), 1.09 (95% CI 1.08–1.10) for Oh et al. (2025), and 1.24 (95% CI 1.06–1.45) for the 2025 UK chronic-pain cohort. Thus, three studies had point estimates above unity, although the confidence intervals of Lee et al. and Khodneva et al. included the null value [2, 3].
Using a random-effects model with restricted maximum likelihood estimation and Hartung–Knapp adjustment, prescription opioid/morphine exposure was associated with a pooled adjusted relative effect of 1.10 (95% CI 1.03–1.18) for incident stroke. The pooled estimate therefore indicated a modest positive association between prescription opioid exposure and stroke occurrence across the included observational studies. Because the contributing studies reported both hazard ratios and odds ratios, the summary measure was expressed as a pooled adjusted relative effect rather than being labelled specifically as a pooled hazard ratio or odds ratio.
Between-study statistical heterogeneity was low. In the conventional inverse-variance analysis, Cochran's Q was 2.92 with 3 degrees of freedom, corresponding to an I² of 0%. The conventional DerSimonian–Laird random-effects model produced a pooled estimate of 1.09 (95% CI 1.08–1.10) with τ² estimated as zero. However, because only four studies contributed to the analysis, the more conservative REML/Hartung–Knapp analysis was considered the principal inferential model. Under this approach, the confidence interval widened while the direction of the pooled association remained unchanged.
A substantial imbalance in statistical precision was evident across studies. Oh et al. (2025), a large nationwide population-based study, had a markedly narrower confidence interval than the remaining studies and consequently contributed approximately 99% of the conventional inverse-variance weight when τ² was estimated as zero [4]. In comparison, Lee et al., Khodneva et al., and the UK cohort each contributed relatively little weight to the conventional pooled estimate. This imbalance motivated additional sensitivity analyses to determine whether the overall finding depended disproportionately on the largest study [2, 3].
The individual study estimates nevertheless showed a broadly consistent direction. The two least precise estimates, Lee et al. and Khodneva et al., were individually compatible with no association, whereas the large Korean study and UK chronic-pain cohort reported estimates above unity with confidence intervals excluding 1.00 [2, 3]. The forest plot demonstrates the study-specific adjusted estimates and corresponding 95% CIs together with the pooled estimate (Figure 1).
Study-specific adjusted relative effect estimates and 95% confidence intervals are shown for the four studies included in the primary quantitative synthesis. The vertical reference line at 1.00 represents no association. The pooled estimate was calculated using a random-effects model. Because the included studies reported different relative effect measures, the summary estimate is presented as an adjusted relative effect rather than as a single effect-measure-specific parameter.
This gives us the central finding properly, but there is one distinction we should preserve throughout the paper:
Primary inference: REML/Hartung–Knapp = 1.10 (95% CI 1.03–1.18).
Conventional model: DL = 1.09 (95% CI 1.08–1.10).
Sensitivity and Influence Analyses
Sensitivity analyses were performed to determine whether the primary association was dependent on the choice of effect measure or disproportionately influenced by an individual study. Overall, the direction of the pooled association remained stable across the prespecified analyses.
Because Lee et al. (2013) reported an odds ratio whereas the remaining studies contributing to the primary analysis reported hazard ratios, an analysis restricted to studies reporting HRs was conducted. Exclusion of Lee et al. produced a pooled adjusted relative effect of 1.11 (95% CI 1.03–1.19). The persistence of the association in this analysis indicated that the primary finding was not dependent on combining the OR from the morphine-specific study with HRs from the cohort studies [2].
Leave-one-out analyses were subsequently performed by sequentially excluding each study from the primary model. Exclusion of Khodneva et al. (2016) resulted in a pooled relative effect of 1.11 (95% CI 1.05–1.18), while exclusion of the 2025 UK chronic-pain cohort yielded an estimate of 1.09 (95% CI 1.08–1.10) [3].
Particular attention was given to Oh et al. (2025), which had substantially greater statistical precision than the other studies and therefore dominated the conventional inverse-variance weighting. When this study was excluded, the pooled association remained above the null, with an adjusted relative effect of 1.16 (95% CI 1.05–1.29). Thus, although removal of the largest study increased the magnitude and substantially widened the confidence interval, it did not reverse the direction of the association [4].
Across the leave-one-out analyses, pooled estimates ranged from approximately 1.09 to 1.16, and none of the corresponding 95% CIs included 1.00. These findings indicate that the direction of the primary association was not dependent on any single included study. Nevertheless, the considerable difference in statistical precision between studies and the small number of studies available for pooling warrant cautious interpretation of the apparent robustness.
Secondary Stroke Outcomes
Evidence for individual stroke subtypes was more limited and methodologically heterogeneous than that available for overall stroke. Consequently, ischemic and hemorrhagic outcomes were evaluated separately from the primary quantitative synthesis, and a pooled subtype estimate was not generated where clinical or methodological comparability was insufficient [8].
Ischemic Stroke
Lee et al. (2013) reported an adjusted OR of 1.08 (95% CI 0.92–1.27) for ischemic stroke among morphine-exposed cancer patients compared with those without morphine exposure. The confidence interval included unity, providing no statistically significant association within this population [2].
More recent evidence was provided by a study that evaluated chronic prescription opioid use among individuals with noncancer pain using the Korean National Health Insurance Service database. Chronic opioid exposure, defined as opioid prescription for ≥90 days, was associated with ischemic stroke compared with non-opioid use (HR 1.16, 95% CI 1.12–1.21). The propensity-score-matched analysis included 36,300 chronic opioid users and 334,590 controls.
This study was retained as secondary ischemic-stroke evidence rather than incorporated into the primary overall-stroke model because its endpoint was specifically ischemic stroke and its underlying database required consideration of potential overlap with other Korean population-based evidence [7].
Dalgaard et al. (2021) provided additional evidence using a case-crossover design examining short-term ischemic-stroke risk in relation to individual prescription medications. Associations varied across opioid agents, including morphine (OR 1.75, 95% CI 1.56–1.98), oxycodone (OR 1.34, 95% CI 1.22–1.47), fentanyl (OR 1.61, 95% CI 1.37–1.91), buprenorphine (OR 1.21, 95% CI 1.07–1.36), and tramadol (OR 1.10, 95% CI 1.03–1.17). Because this study assessed short-term medication exposure using a case-crossover design and provided multiple drug-specific estimates from the same study population, these estimates were not treated as independent studies or combined with the longitudinal cohort estimates [6].
Overall, the ischemic-stroke findings were directionally suggestive of increased risk in several analyses, particularly for chronic opioid exposure and selected individual opioid agents. However, substantial differences in population, exposure duration, analytical design, and opioid agent precluded a clinically defensible pooled ischemic-stroke estimate.
Hemorrhagic Stroke and Subarachnoid Hemorrhage
Evidence for hemorrhagic stroke was more limited. Lee et al. reported an association between morphine exposure and hemorrhagic stroke (OR 1.36, 95% CI 1.02–1.82), in contrast to the less conclusive estimate observed for ischemic stroke [2].
In the 2025 UK chronic-pain cohort, opioid analgesic use was associated with subarachnoid hemorrhage (HR 1.92, 95% CI 1.25–2.95). Among 203,972 participants with chronic pain, the investigators documented 5,331 overall strokes, including 4,304 ischemic strokes, 872 intracerebral hemorrhages, and 485 subarachnoid hemorrhages during follow-up. Opioid exposure was associated with overall stroke and specifically with subarachnoid hemorrhage.
A pooled hemorrhagic-stroke estimate was not calculated because the available evidence did not represent sufficiently equivalent cerebrovascular endpoints: hemorrhagic stroke as reported by Lee et al. and subarachnoid hemorrhage in the UK cohort should not be assumed to represent interchangeable outcomes. The limited number of independent estimates further precluded a reliable subtype-specific quantitative synthesis.
Accordingly, stroke-subtype findings were retained as secondary evidence and interpreted independently from the primary meta-analysis of overall incident stroke.
Risk-of-Bias Findings
Risk-of-bias assessment identified methodological concerns across the included observational studies, although their nature and extent differed. No study was considered completely free from potential bias, primarily because opioid exposure was not randomly assigned and residual confounding could not be eliminated.
Khodneva et al. (2016) had several methodological strengths, including its prospective design, detailed assessment of baseline covariates, propensity-score analysis, and expert-adjudicated cardiovascular outcomes. Nevertheless, the study lacked information on opioid dose, duration of use, and longitudinal changes in exposure, and residual confounding by indication remained possible. Overall, the study was considered to have a moderate risk of bias [3].
Oh et al. (2025) benefited from a very large nationwide cohort and extensive propensity-score matching. However, opioid dose was unavailable, several potentially important lifestyle factors—including smoking, alcohol consumption, and body mass index—were not incorporated into the analysis, and opioid exposure could change during follow-up. These limitations raised concerns regarding residual confounding and exposure misclassification; the study was therefore considered to have a moderate risk of bias, despite its high statistical precision [4].
The UK chronic-pain cohort was also judged to have a moderate risk of bias. Medication use and several participant characteristics were assessed at baseline, and changes in medication exposure during long-term follow-up were not captured. The investigators additionally acknowledged the healthy-volunteer characteristics of the UK Biobank population, which may limit generalizability [5].
Lee et al. (2013) was considered to have a moderate-to-high risk of bias. Although the analysis adjusted for major vascular comorbidities, morphine treatment was evaluated among patients with malignancy, creating particular susceptibility to confounding by indication and differences in underlying cancer-related characteristics between exposed and unexposed patients [2].
Overall, the evidence base was therefore characterized predominantly by moderate risk of bias, with the principal concerns being residual confounding, confounding by indication, incomplete characterization of opioid dose and duration, and potential exposure misclassification. These limitations were considered when interpreting the magnitude and certainty of the pooled association.
Discussion
This meta-analysis identified a modest but statistically robust association between prescription opioid/morphine exposure and incident stroke. In the principal random-effects analysis using REML with Hartung–Knapp adjustment, the pooled adjusted relative effect was 1.10 (95% CI 1.03–1.18). Importantly, the association remained above the null across sensitivity analyses, including restriction to studies reporting hazard ratios and exclusion of the highly influential Korean nationwide cohort. When Oh et al. (2025) was removed, the pooled estimate remained elevated at 1.16 (95% CI 1.05–1.29). Collectively, these findings suggest that the observed association was not entirely dependent on a single study or on combining different relative effect measures.
The magnitude of the association was modest and should be interpreted in the context of the underlying observational evidence. Prescription opioid use frequently occurs among patients with chronic pain, malignancy, multimorbidity, reduced physical activity, and other characteristics that may independently influence vascular risk [11, 12]. Residual confounding and confounding by indication therefore remain important alternative explanations. This concern is particularly relevant to the morphine study in patients with cancer and cannot be eliminated by statistical adjustment alone. Similarly, several included studies lacked detailed information on opioid dose, cumulative exposure, or changes in treatment over time. Khodneva et al., for example, acknowledged limitations relating to opioid dose, duration, and residual confounding despite extensive adjustment and expert-adjudicated stroke outcomes.
The subtype findings provide additional, although less definitive, evidence. Chronic prescription opioid exposure was associated with ischemic stroke in the Korean NHIS analysis (HR 1.16, 95% CI 1.12–1.21), while individual opioid agents showed varying associations with acute ischemic stroke in the case-crossover study [9]. Hemorrhagic evidence was more limited: morphine exposure was associated with hemorrhagic stroke in Lee et al., while opioid analgesic use was associated with subarachnoid hemorrhage in the UK chronic-pain cohort. These outcomes were not pooled because ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage represent biologically and clinically distinct cerebrovascular events [2].
Several mechanisms could plausibly contribute to an opioid–stroke association, including opioid-related respiratory depression and hypoxemia, hypotension, cardiovascular effects, reduced mobility, and interactions with underlying comorbidity. However, the present data do not establish that these mechanisms mediate the observed epidemiological association. The possibility that opioid exposure functions partly as a marker of underlying disease burden rather than as an independent causal determinant must therefore remain central to interpretation [10].
The findings have potential clinical relevance because prescription opioids are widely used for pain management, and even a modest relative association could be important in populations with substantial baseline cardiovascular risk. Nevertheless, these results do not demonstrate that clinically indicated opioid therapy causes stroke and should not, in isolation, alter prescribing decisions. Instead, they support careful assessment of vascular risk, treatment indication, dose and duration, particularly among patients requiring prolonged opioid therapy.
The principal strengths of this meta-analysis are its focused definition of prescription opioid exposure, exclusion of heroin and illicit-opium evidence from the primary synthesis, use of adjusted study-level estimates, and multiple sensitivity analyses. Important limitations include the small number of studies available for primary pooling, heterogeneity in populations and exposure definitions, reliance on observational evidence, residual confounding, potential exposure misclassification, and the disproportionate statistical precision of one large cohort. Furthermore, the limited number of eligible studies prevented meaningful assessment of small-study or publication bias and restricted quantitative analysis of individual stroke subtypes.
Overall, the available evidence supports a small positive association between prescription opioid/morphine exposure and incident stroke, which remained evident under more conservative statistical and sensitivity analyses. The finding should nevertheless be interpreted as an association rather than evidence of causality. Future studies with repeated exposure assessment, detailed opioid dose and duration data, rigorous control of pain severity and indication, and standardized adjudication of ischemic and hemorrhagic stroke are required to determine whether the observed association reflects an independent vascular effect of prescription opioids.
Conclusion
This meta-analysis demonstrates a modest positive association between prescription opioid/morphine exposure and incident stroke, with the direction of the association remaining consistent across the principal and sensitivity analyses. The findings suggest that cerebrovascular risk deserves consideration when evaluating prolonged prescription opioid therapy, particularly in patients with existing vascular risk factors or substantial comorbidity. However, the available evidence does not establish a causal relationship, as all contributing studies were observational and remained vulnerable to residual confounding, confounding by indication, and changes in opioid exposure over time.
The evidence for individual stroke subtypes remains less definitive because of differences in study populations, opioid agents, exposure duration, and outcome definitions. Future prospective research should incorporate detailed information on opioid type, dose, duration, cumulative exposure, and time-varying use, alongside rigorous adjustment for pain severity and cardiovascular risk factors. Such studies are necessary to determine whether the observed association represents an independent effect of prescription opioids and to identify patient groups in whom cerebrovascular risk may be greatest.
References