The global rise in prescription opioid use has contributed to an opioid epidemic, associated harms, and unintentional deaths in several western countries. Opioids however continue to be regularly prescribed for acute pain and in the chronic pain context due to limited treatment options. Currently there are no accurate tools that help predict which patients prescribed opioids may be at risk of death, which depends on the cultural context and varies across countries. Existing models do not account for statistical considerations such as censoring and competing risks. Using nationally representative data from the United Kingdom from 1,026,139 patients newly prescribed an opioid, we developed three competing risk time-to-event models: a regression model, a random forest, and a deep neural network to predict opioid-related deaths using UK primary care records. The models were externally validated in an external cohort of 337,015 patients. The models exhibited good discrimination and positive predictive value during internal validation (C-statistic for the regression model, random forest, and neural network: 84.3%, 84.4% and 82.1% respectively), and external validation (C-statistic for the regression model, random forest, and neural network: 81.8%, 81.5% and 81.5% respectively). Prior substance abuse, lung and liver comorbidities, morphine, fentanyl, or oxycodone at initiation and co-prescription of gabapentinoids were some of candidate predictors associated with a higher risk of opioid-related mortality within the models. These results demonstrate how routinely collected data from a nationally representative dataset may be used to develop and validate opioids risk algorithms to better help clinicians and patients predict risk to this serious adverse outcome.

OBJECTIVES:Artificial intelligence (AI) and machine learning applications are rapidly expanding across healthcare. Successful implementation of AI technologies in rheumatology will depend not only on technical performance but also on the perceptions and preparedness of end-users. This study evaluated the current opinions, expectations, and concerns of AI among healthcare professionals and researchers in rheumatology across the UK. METHODS:A 19-item survey was designed and distributed through national and regional networks aimed at the rheumatology workforce between June 2025 - January 2026, targeted at consultant rheumatologists, doctors-in-training, allied health professionals, specialist nurses, and non-clinical researchers in rheumatology. The questions included respondent background data, current applications of AI in clinical care and research, opinions about AI in terms of perceived impact, concerns, educational needs and expected performance. RESULTS:Of the 218 respondents, 39% to 40% reported daily or weekly use of AI in research and clinical practice respectively. The most common clinical uses were using LLMs to look up medical facts (45%), to improve grammar/spelling of clinical documentation (28%), generate differential diagnoses (22%) and use of ambient scribes (17%). 86% anticipated that AI would substantially impact clinical practice in five years or less. Administrative tasks (85%) and musculoskeletal imaging (63%) were perceived as the areas likely to experience the greatest impact from AI. Highlighted concerns included data security/privacy (70%), medical liability (70%), followed by lack of explainability (47%). One in four reported excellent confidence in using digital technology, with only 6% self-rating their AI knowledge as excellent. A strong interest in education about AI was expressed regarding several areas including the ethical and safe use of AI (66%), safe and efficient use of LLMs in clinical practice (64%), and ambient AI scribes (59%). CONCLUSIONS:AI is already being used frequently in UK rheumatology practice and research, with most anticipating a considerable impact on clinical care within the next five years or less. However, despite enthusiasm for adoption, important concerns regarding data security, liability, and explainability remain, alongside low self-reported AI knowledge, highlighting the need for targeted education, robust governance, and safe clinical implementation strategies.

Abstract Background/Aims To investigate the heterogeneity of treatment response in patients with psoriatic arthritis (PsA) who are commencing a biologic (b) or targeted synthetic (ts) DMARD and identify associations between different responses and clinical features at drug initiation. Methods Patients from the UK prospective observational cohort study, Outcomes of Treatment in Psoriatic Arthritis Study Syndicate (OUTPASS) had demographic and clinical data collected at 0, 3, 6, and 12 month follow-up time points after initiating b/tsDMARD therapy. Clusters of patients were identified based on their DAS28 composite and PsARC subcomponent scores over time via Group Based Trajectory Modelling. Only patients with at least 1 data point recorded in each component over the 12-month follow-up were included. Associations between identified clusters, demographic and clinical factors at treatment initiation were characterised through univariable and multivariable logistic regressions. Univariable models were corrected for multiple testing with 5% false discovery rate threshold correction. Missing risk factor data were imputed for regression analyses. Results Of the 569 patients, two clusters were uncovered following b/tsDMARD therapy: responders (55%) and non-responders (45%). The non-responder group was characterised by higher DAS28 and PsARC component scores across all time points and worsening of the patient global assessment component of the PsARC score after 3 months. Univariable models identified asthma, older age, higher baseline DAPSA, and initiating ustekinumab as significantly associated with non-response after multiple testing correction (Table 1). While depression, angina, myocardial infarction, and female gender were significantly associated with non-response using univariable analysis, the association was not maintained following correction for multiple testing. Multivariable logistic regression identified asthma, higher baseline DAPSA , lower baseline CRP, and ustekinumab therapy as significantly associated with non-response. Conclusion Our data identified two patterns of response in PsA patients commencing an advanced therapeutic. Associations found in previous PsA cohorts such as lower CRP and higher DAPSA correlating with non-response were replicated in this study, but presence of depression, older age, and female gender were not. Asthma has not previously been reported to be associated with non-response in PsA cohorts and may be confounded by presence of concurrent psoriasis or IL-17 dysregulation. Disclosure D. Chong: None. M. Khasru: None. S. Shoop-Worrall: None. A. Barton: None. H. Chinoy: None. M. Jani: None. J. Bluett: Grants/research support; JB has received a research grant award from Pfizer and travel/conference fees in the last 3 years from Fresenius Kabi and Novartis.

OBJECTIVES:IL-17A inhibitors are therapeutic options in PsA, but response is not universal. Evidence from other inflammatory arthritides suggests differential gene expression may predict the outcomes. This study aimed to identify transcriptomic predictive biomarkers of response in PsA patients commencing secukinumab. METHODS:Participants were recruited to OUtcomes of Treatment in Psoriatic Arthritis Study Syndicate (OUTPASS), a prospective observational cohort study of patients with PsA initiating advanced therapeutics. Samples for this analysis were chosen based on extreme phenotype response. Whole-blood RNA sequencing was performed longitudinally in 13 secukinumab-treated patients at baseline (pre-treatment) and at 3 months post-treatment, with response evaluated at 3 months using the DAS28 criteria and the Psoriatic Arthritis Response Criteria. Differential gene expression analysis, Ingenuity Pathway Analysis (IPA), and weighted gene co-expression network analysis (WGCNA) were performed to identify significantly differentially expressed genes (DEGs) (adjusted P < 0.05, |log2 fold change| ≥ 1), enriched pathways (P < 0.05), and co-expressed gene modules. Immune cell subset proportions were estimated by deconvolution, and hub genes were identified by integrating DEGs and WGCNA, with overlapping genes defined as potential driver genes. RESULTS:IGHV3-64D and IGHV1-46 were differentially expressed at baseline, 3 months, and sustained over time in the responder group (adjusted P < 0.05). Five overlapping genes (GMPR, CDC34, DMTN, UBXN6 and SLC25A39) were identified as potential drivers. Functional analysis indicated a potential contribution of metabolic pathways to the modulation of therapeutic response. CONCLUSION:We identified two genes as pre-treatment predictive biomarkers of secukinumab response that persisted over time. Integration with WGCNA revealed five additional candidate genes. These genes are implicated in metabolic pathways, which may modulate the secukinumab response. These findings warrant further validation.

Opioids are associated with serious adverse outcomes, including premature death. Respiratory depression is among the most severe opioid-related events, yet data on its incidence in non-cancer pain remain limited. Pharmacological differences suggest varying respiratory risks across opioid drugs. This study evaluated the comparative risk of respiratory depression by opioid drug and dose, and the impact of concomitant gabapentinoids and benzodiazepines in hospitalised patients with non-cancer pain. A retrospective cohort study was conducted using electronic health records from a large tertiary hospital in Northwest England. Adult inpatients (≥ 18 years) receiving opioids for non-cancer pain were included. Opioid exposure was defined from drug administration records. Respiratory depression was identified using National Early Warning Scores or naloxone administration. Incidence rates were estimated by time-varying opioid exposure, opioid drug, and daily morphine milligram equivalent (MME). Associations with incident respiratory depression were examined using a Cox regression adjusted for confounders. Effect modifiers including co-administration of gabapentinoids or benzodiazepines were assessed for their impact on respiratory depression risk, in addition to opioids. Among 32,909 inpatients, fentanyl (HR: 3.36, 95

Pain is a frequent symptom in people with inflammatory arthritis (IA), which has substantial impact on their quality of life. Analyses of electronic health record data indicate that UK pain care in people with IA often involves prescribing long-term opioids and gabapentinoids, despite absent trial evidence for efficacy. Patient survey data suggest that non-pharmacological pain care with supportive trial evidence is underused. A UK-specific guideline on pain management for people with IA is required to address this. This comprehensive life-course guideline is the first British Society for Rheumatology Guideline to specifically address pain in people with IA. It provides evidence-based recommendations on how pain can be best managed in people with IA. It was developed using the methods outlined in the British Society for Rheumatology's 'Creating Clinical Guidelines' protocol by a multidisciplinary Guideline Working Group, comprising healthcare professionals with expertise in paediatric and adult rheumatology and people with lived experience. By undertaking and considering the evidence from several systematic literature and umbrella reviews, 23 recommendations were developed. These address how pain should be assessed in people with IA alongside the role of the following treatments in IA pain management: DMARDs, glucocorticoids, analgesics, neuromodulators, exercise and physical activity, psychological interventions, ergonomic and orthotic interventions (excluding orthoses for foot pain), education, weight management and diet, addressing sleep problems, fatigue management, digital technologies and medical devices, complementary therapies, and support from others. An audit tool is provided to support the Guideline's implementation, and key recommendations made for future research.

Opioid use for chronic non-cancer pain remains common in the UK, despite limited evidence of long-term effectiveness. Delirium, a serious acute confusional state associated with increased mortality, is a known adverse effect of opioid use. Pharmacological differences between opioids may influence delirium risk, but comparative evidence is scarce. This study evaluated the association of opioid type and dosage with the risk of in-hospital delirium in non-cancer patients. We conducted a retrospective cohort study using electronic health records (EHRs) from a tertiary care hospital in northwest England (September 26, 2014–December 31, 2020). Adult (≥ 18 years) without cancer who were administered with opioids during admission were included. Delirium was identified using the 4 ‘A’s Test or through a combination of ICD-10 codes and new-onset confusion scores (= 3) on the National Early Warning Score. Daily opioid doses were converted to daily morphine milligram equivalents (MME/day) to assess the effect of dose across different opioid types. Incidence rates were calculated by opioid type and opioid dosage. Cox regression models, adjusted for confounders, were used to evaluate delirium risk. Among 50,586 opioid-exposed patients (mean [SD] age, 55 [20] years; 53

IntroductionTransformer-based models have shown strong potential for clinical prediction using electronic health record data, yet their performance can vary depending on modelling decisions and data characteristics.MethodsIn this study, we trained a BEHRT model on hospital-based UK Biobank data and evaluated its performance across four clinical prediction tasks, including next-visit diagnosis and longer-term diagnosis prediction up to five years. We exhaustively assessed the impact of model size, medical terminology (CALIBER vs ICD-10), and data split strategies.ResultsThe large model consistently outperformed the smaller one in long-term prediction tasks (AUROC = 0.874 vs 0.858 at 5 years), while differences were marginal in 6-months prediction tasks. Performance was also sensitive to the vocabulary size, with CALIBER model yielding higher average precision scores (Average Precision Score = 0.773 vs 0.678 using ICD-10).DiscussionOur results show that transformer models can achieve high predictive performance across diverse clinical scenarios, but outcomes vary considerably depending on modelling choices, particularly in long-term prediction tasks.

OBJECTIVE:Up to one in five patients with axial spondyloarthritis (AxSpA) or psoriatic arthritis (PsA) newly initiated on opioids transition to long-term use within the first year. This study aimed to investigate individual factors associated with long-term opioid use among opioid new users with AxSpA/PsA. METHODS:Adult patients with AxSpA/PsA and without prior cancer who initiated opioids between 2006 and 2021 were included from Clinical Practice Research Datalink Gold, a national UK primary care database. Long-term opioid use was defined as having ≥3 opioid prescriptions issued within 90 days, or ≥90 days of opioid supply, in the first year of follow-up. Individual factors assessed included sociodemographic, lifestyle factors, medication use and comorbidities. A mixed-effects logistic regression model with patient-level random intercept was used to examine the association of individual characteristics with the odds of long-term opioid use. RESULTS:In total, 10 300 opioid initiations were identified from 8212 patients (3037 AxSpA; 5175 PsA). The following factors were associated with long-term opioid use: being a current smoker (OR: 1.62; 95%CI: 1.38,1.90), substance use disorder (OR: 2.34, 95%CI: 1.05,5.21), history of suicide/self-harm (OR: 1.84; 95%CI: 1.13,2.99), co-existing fibromyalgia (OR: 1.62; 95%CI: 1.11,2.37), higher Charlson Comorbidity Index (OR: 3.61; 95%CI: 1.69,7.71 for high scores), high MME/day at initiation (OR: 1.03; 95%CI: 1.02,1.03) and gabapentinoid (OR: 2.35; 95%CI: 1.75,3.16) and antidepressant use (OR: 1.69; 95%CI: 1.45,1.98). CONCLUSIONS:In AxSpA/PsA patients requiring pain relief, awareness of lifestyle, sociodemographic and prescribing characteristics associated with higher risk of long-term opioid use can prompt timely interventions such as structured medication reviews and smoking cessation to promote safer prescribing and better patient outcomes.

Whilst pharmacological treatments have transformed the management of rheumatic and musculoskeletal diseases (RMDs) they are often associated with adverse events (AEs), ranging from mild to severe and may lead to hospitalisation or even death. Clinical prediction models (CPMs) are statistical tools that could assist in predicting the risk of AEs at an individual level and enable more informed, risk-stratified decisions shared with patients. This systematic review aims to identify, summarise, and evaluate the methodological quality of existing CPMs predicting AEs associated with RMD medications. We conducted a search in PubMed, Embase, and Medline databases from inception to March, 2024. Studies were included if they developed at least one multivariable CPM for predicting AEs in adult patients using RMD medications. Exclusion criteria included studies focused on illicit/recreational drugs, randomised controlled trials, systematic reviews, and meta-analyses. Three reviewers independently screened titles and abstracts, followed by full-text review. Conflicts were resolved by a fourth reviewer. Data extraction and quality assessment were conducted using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) and Prediction model Risk Of Bias Assessment Tool (PROBAST) checklists, to ensure, consistent reporting and assess the risk of bias (ROB). The protocol was registered in PROSPERO. Out of 2406 studies, 1734 titles and abstracts were screened, and 38 studies were selected for full-text review. Finally, 11 studies were included in the final review contributing to 16 individual CPMs. The majority of those CPMs (75.0%) were focused on rheumatoid arthritis and disease modifying anti-rheumatic drugs (DMARDs), such as methotrexate (68.7%) and biologic drugs (12.5%). Most models employed Cox proportional hazards or logistic regression models. Twelve models (75.0%) had high overall ROB due to inappropriate variable selection methods and sample size. This is the first systematic review that summarises the progress to date in developing and reporting CPMs for AEs related to RMD medications. It underscores the need for careful attention to variable selection and transparency around sample size. Additionally, future models could include a broader range of RMDs and medications commonly prescribed in such conditions. Emerging statistical techniques such as machine learning with the ability to model complex interactions, and multi-outcome CPMs to predict several AEs to one class of drug may demonstrate future potential in this context. C. Diomatari: None. D. Jenkins: None. G. Martin: None. M. Jani: None.

Objective: Large language models (LLMs) are attracting increasing interest in healthcare. This commentary evaluates the potential of LLMs to improve clinical prediction models (CPMs) for diagnostic and prognostic tasks, with a focus on their ability to process longitudinal electronic health record (EHR) data. Findings: LLMs show promise in handling multimodal and longitudinal EHR data and can support multi-outcome predictions for diverse health conditions. However, methodological, validation, infrastructural, and regulatory chal- lenges remain. These include inadequate methods for time-to-event modelling, poor calibration of predictions, limited external validation, and bias affecting underrepresented groups. High infrastructure costs and the absence of clear regulatory frameworks further prevent adoption. Implications: Further work and interdisciplinary collaboration are needed to support equitable and effective integra- tion into the clinical prediction. Developing temporally aware, fair, and explainable models should be a priority focus for transforming clinical prediction workflow.

Background Corticosteroids are widely prescribed in primary care for a broad range of inflammatory and autoimmune conditions. Despite their therapeutic benefits, corticosteroids are associated with adverse effects, particularly with long-term use. We aimed to describe corticosteroid prescribing patterns to better understand the current landscape of corticosteroid use in England. Methods We conducted a population-based cohort study using electronic health-care records from the Clinical Practice Research Datalink Aurum. The study population included adults (aged 18 years or older) who were alive, registered with a general practice in England, and had at least one corticosteroid prescription between Jan 1 and Dec 31, 2023 (the period of follow-up). The study population (age, sex, and ethnicity) and corticosteroid characteristics (drug substance and duration of use) were described by route of administration. Outcomes included pre-defined clinical indications across disease systems and were defined in the 2-year window prior to the patients’ first corticosteroid prescription up until their last corticosteroid prescription and described by route of administration. Finally, combinations of disease systems were described. This study was registered with Clinical Practice Research Datalink (erap number 25_004991). Findings Between Jan 1 and Dec 31, 2023, 12 029 952 unique corticosteroids were prescribed to 2 564 729 patients. Median age of patients was 56·5 years (IQR 40·5–70·5), 1 466 264 (57·2%) of the population were female, 1 999 601 (78·0%) were White, 207 228 (8·1%) were South Asian, 77 205 (3·0%) were Black, 33 239 (1·3%) were Mixed, 53 202 (2·1%) were of Other ethnicity, and 194 254 (7.6%) had missing ethnicity. The median number of corticosteroids prescribed per person over the year of follow-up was two (IQR 1–6). Inhaled corticosteroids accounted for the highest number of prescriptions (44·2%), followed by cutaneous (19·6%), nasal (14·3%), and oral (14·0%) routes. 1 267 685 (49·4%) patients had at least one pre-defined clinical indication recorded up to 2 years prior. Asthma (801 924 [63·3%]), eczema (229 382 [18·1%]), and chronic obstructive pulmonary disease (220 759 [17·4%]) were the most frequently recorded indications for all corticosteroids. Among patients with multiple indications, the most common combination was respiratory and dermatology-related disease with 53·8% of patients receiving corticosteroid treatment for both types of diseases. Interpretation Corticosteroid prescribing remains widespread in primary care in England. These findings highlight the need to evaluate safer therapeutic alternatives and update current guidelines for potentially reducing corticosteroid exposures. Funding None.

Rheumatology research increasingly relies on diverse real-world data sources to complement insights from randomised controlled trials. Real-world evidence or observational data derived from disease or treatment registries, administrative claims datasets, electronic health records, and distributed data networks can enable large-scale analyses of treatment effectiveness, safety, and healthcare utilisation that can improve patient care and outcomes. This review provides a structured overview of the key real-world data sources currently used in rheumatology, highlighting their strengths, limitations, and opportunities. While no single dataset is without limitations, aligning the right source to the right clinical research question requires careful attention to data provenance, data quality, generalisability, and reproducibility. We outline 10 key considerations to guide both healthcare professionals and researchers who work with observational data to critically appraise real-world studies and design robust, fit-for-purpose research. With the increasing use of artificial intelligence and machine learning being applied to health data, the review provides timely guidance on data considerations to reduce potential training and algorithmic biases. Recognising the trade-offs of different data sources and applying rigorous, transparent methods are essential to generate evidence that not only withstands scientific scrutiny but also meaningfully advances patient care and rheumatology research.

Constipation is a frequent adverse event associated with opioid medications that can have a considerable impact on patients’ quality of life. In patients who require opioids for pain relief, less is known about the risk conferred by specific opioids given their diverse pharmacology and the effect of daily dose and potency. The aim of the study was to evaluate the comparative risk of severe constipation by opioid type and dose in patients with non-cancer pain admitted to hospital. We conducted a retrospective cohort study using hospital electronic health records in Northwest England between December 1, 2009, and December 31, 2020. Patients who were ≥ 18 years and without a history of cancer were included. Opioid exposure was measured using administered drug information in hospital. The outcome was a severe constipation event defined as administration of an enema or suppository. Incidence rates by opioid use status, type of opioid class and morphine milligram equivalent (MME) per day were calculated, and a Cox regression model was used to determine associations with incident constipation after adjusting for confounders. The study included 80,475 eligible patients who were administered an opioid in hospital. Compared to codeine, morphine (HR 1.59, 95

Over the past two decades, England has experienced a notable rise in prescription opioid use, leading to some of Europe’s highest opioid-related hospital admissions, which nearly doubled between 2008 and 2018. Patients with rheumatic and musculoskeletal diseases (RMDs) are especially at risk due to multimorbidity and long-term opioid use. Machine learning-based clinical prediction models (ML-CPMs) could improve current risk assessment by addressing nonlinear relationships. We aimed to develop and evaluate CPMs based on regression and ML, using nationally representative UK data, to estimate the risk of opioid-related hospitalisations in RMD patients with non-cancer pain. We conducted a retrospective cohort study using the Clinical Research Practice Datalink, a large-scale UK primary care electronic health record database from 2006-2021. Patients were included if they were ≥18 years old and new opioid users diagnosed with one or more RMDs: osteoarthritis, fibromyalgia, rheumatoid arthritis, ankylosing spondylitis, systemic lupus erythematosus, or psoriatic arthritis. Patients with a history of cancer were excluded. Hospital admissions related to opioid-related harms (including opioid poisoning and abuse, fractures and falls, delirium, sleep disorders and constipation) were identified using Hospital Episode Statistics Admitted Patient Care and Accident and Emergency Attendances. Candidate predictors were selected based on clinical relevance from prior literature. Patients were considered ‘at risk’ from the time they received their first new opioid prescription and were censored at death, loss of follow-up or two years without an opioid prescription. Predictive models using Cox proportional hazards (CPH), Random Survival Forest (RSF) and DeepHit Neural Network were trained using time-to-event patient data. Predictive performance of the models was evaluated, including calculating the area under the receiver characteristic operator curve (AUC) using 5-fold cross validation. Over the 15-year study period, 1,127,357 unique patients (60.7% female, median age [IQR]: 62 [25]) who were newly initiated on opioids were analysed. Opioid-related hospitalisations occurred in 76,033 patients (6.7%). In the CPH model, the factors most associated with an increased risk of opioid-related hospitalisation included history of suicide attempts and self-harm (hazards ratio [HR]: 1.96, 95% CI: 1.82, 2.14), high comorbidity score (HR: 1.68, 95% CI:1.59,1.73) and very high initial MME/day (HR: 1.65, 95% CI:1.56,1.72). Alcohol dependence (HR: 1.59, 95% CI: 1.49-1.65) and concomitant benzodiazepine use (HR: 1.12, 95% CI: 1.07-1.18) were also associated to a higher risk. The CPH, RSF and DeepHit models achieved mean AUC values of 0.63, 0.64 and 0.75, respectively. In this study, the DeepHit clinical prediction model demonstrated superior performance compared to the others. Higher risks of opioid-related hospitalisation were associated with concomitant use of benzodiazepines, prior alcohol dependence, and history of self-harm. Greater vigilance where these factors exist could guide targeted interventions to reduce opioid-related hospitalisations. C. Ramirez Medina: None. J. Benitez-Aurioles: None. D. Jenkins: None. M. Lunt: None. W. Dixon: None. N. Peek: None. M. Jani: None.

Background: Self-care for people with musculoskeletal conditions often includes taking pain-relieving medication at times of need and organising help with daily activities (social care). Information on this self-care is not systematically documented by healthcare providers but could be generated by patients themselves. Aims: To assess the feasibility, acceptability and added value of electronic patient-generated data on opioid, over-the-counter medication and supplement use, and social care use in people with musculoskeletal conditions. Methods: Adults with a musculoskeletal condition from an outpatient department in England completed a web-based questionnaire on their medication and/or social care use every two weeks for six months. Using a sequential mixed-methods approach, we analysed questionnaire response rates and completeness, and transcripts from interviews with patients. We also compared patient-generated self-care data to data recorded in electronic health records. Results: Of 102 people invited, thirty-six consented to take part; reasons to decline included no access to technology or unwillingness to use it for collecting self-care data, and physical limitations due to people’s condition. All those consenting completed at least one questionnaire, with twenty-two of them completing ≥80% of questionnaires. Participants identified potential benefits of data collection (e.g., to support self-monitoring and patient-led consultations) and considered it feasible to continue this longer term. We found that patient-generated self-care data contributed new information on (changes over time in) medication (e.g. medication frequency, side effects, and supplement use) and social care use (e.g. level of formal/informal support received) compared to data in the electronic health record. Conclusions: People with musculoskeletal conditions found it feasible and acceptable to collect electronic patient-generated data on self-care, which complemented information recorded by clinicians in electronic health records, including changes in self-care use over time. Healthcare providers should therefore consider collecting patient-generated self-care data to enhance service delivery and patient outcomes, and to enrich musculoskeletal research. Clinical trial number: Not applicable

Machine learning has increasingly been applied to predict opioid-related harms due to its ability to handle complex interactions and generating actionable predictions. This review evaluated the types and quality of ML methods in opioid safety research, identifying 44 studies using supervised ML through searches of Ovid MEDLINE, PubMed and SCOPUS databases. Commonly predicted outcomes included postoperative opioid use (n = 15, 34%) opioid overdose (n = 8, 18%), opioid use disorder (n = 8, 18%) and persistent opioid use (n = 5, 11%) with varying definitions. Most studies (96%) originated from North America, with only 7% reporting external validation. Model performance was moderate to strong, but calibration was often missing (41%). Transparent reporting of model development was often incomplete, with key aspects such as calibration, imbalance correction, and handling of missing data absent. Infrequent external validation limited the generalizability of current models. Addressing these aspects is critical for transparency, interpretability, and future implementation of the results.

Objective To quantify the changes in opioid prescribing over time to a population with high rates of opioid use to understand the impact of longer elective wait times during the covid-19 pandemic.Design With the approval of NHS England, a retrospective cohort study using linked electronic health record data in OpenSAFELY-TPP.Setting Primary and secondary care electronic health records of people registered at general practices in England that use TPP SystmOne software, covering about 43% of the total registered population in England, linked to data from the Waiting List Minimum Dataset (WLMDS) within the OpenSAFELY-TPP platform, which is part of the NHS England OpenSAFELY covid-19 service.Participants 63 850 eligible patients on the waiting list for elective trauma procedures or orthopaedic procedures whose wait ended in admission between May 2021 and April 2022.Main outcome measures Opioid prescribing to eligible patients before referral to the waiting list, while waiting for treatment, and after discharge from treatment. Opioids were classified based on their strength (weak, moderate, or strong opioids) and duration of action (immediate release v modified release opioids).Results Of 63 850 people on elective trauma or orthopaedic waiting lists whose wait ended during the study period (median age 61 years, 54.6% female), 20.5% waited for more than 52 weeks to be admitted. In the three months before their waiting list referral date, 9890 (15.5%) participants had three or more opioid prescriptions, and 3790 (5.9%) were prescribed a strong opioid. Weekly opioid prescribing rates per 100 people on the waiting list were stable over time, with prescription rates peaking immediately after treatment and plateauing about three months after treatment. Comparing the three month period before the waiting list referral date to the period four to six months after the waiting list end date, changes in the proportion of people with three or more prescriptions for an opioid during that period were −1.6% (95% confidence interval −2.2% to −1.0%) for people on the waiting list for 18 weeks or less, −1.1% (−1.7% to −0.5%) for people waiting for 19-52 weeks, and −0.5% (−1.4% to 0.4%) for people waiting for more than 52 weeks.Conclusions In this study, one in five people who received treatment for an elective orthopaedic procedure between May 2021 and April 2022 waited for more than one year. Nearly one in seven people were prescribed opioids long term before their referral date to the waiting list, and only small reductions in long term opioid prescribing were observed after a patient’s procedure, regardless of length of time spent on the waiting list.