Atul Anand

Artificial intelligence is increasingly being applied to the assessment, monitoring and care of older adults, with potential benefits across both population health and individual clinical practice. This narrative review examines current and emerging applications of AI in common conditions and care needs associated with ageing, including hypertension, stroke, myocardial infarction, Parkinson’s disease, osteoporosis, sarcopenia and supportive care robotics. Across these domains, AI-enabled analysis of electronic health records, electrocardiograms, imaging, video, wearable-sensor data and home-monitoring measurements may support earlier disease detection, more objective functional assessment, personalised risk prediction and treatment, and greater independence. Computer-vision approaches may provide detailed measures of movement following stroke and in Parkinson’s disease, while opportunistic analysis of existing imaging may improve recognition of low bone mineral density, vertebral fractures and reduced muscle health. Care robots may additionally assist with companionship, medication management and falls detection. However, much of the evidence remains based on selected datasets, proof-of-concept studies or controlled settings. Algorithmic bias, limited representation of frail and multimorbid older adults, privacy, governance, interpretability, interoperability, user burden and workforce consequences remain important barriers. AI should therefore complement rather than replace clinical judgement. Equitable implementation will require representative datasets, rigorous external and prospective validation, transparent models, user-centred design and evaluation of clinically meaningful outcomes.

Background: Co-existing mental and physical conditions (mental-physical MLTC) has been associated with poorer quality of life and worse health outcomes than physical conditions alone. Whether its demography and comorbidity patterns differ across mental conditions is unclear.Methods: Cross-sectional analysis of routine healthcare data for 702,342 adults in Lothian, Scotland, UK. Fifty-two physical and nine mental conditions were identified in linked primary care and hospital data. We examined the prevalence, demography and comorbidity patterns of mental-physical MLTC overall and across nine index mental conditions. We quantified the co-occurrence of mental and physical conditions using prevalence and age-sex standardised observed-to-expected ratios (sOER).Findings: Mental-physical MLTC affected 22·3% (n=156,479) of adults and accounted for 57·2% of all MLTC. Two-thirds (68·3%) of people with any mental condition had a physical comorbidity, rising from 38·7% at 18–39 years to 94·4% at ≥65 years. Prevalence was higher in the most deprived than the most affluent quintile for mental-physical MLTC (29·0% vs 18·1%) but not physical-only MLTC (13·0% vs 15·5%). Depression/anxiety featured in 76·8% of mental-physical MLTC and was the only mental condition in 50·5%; additional mental conditions were far more common in other subgroups (86·1–94·2%). Mental-physical MLTC involving personality disorders and PTSD had the highest proportion living in the most deprived areas (23·1% and 21·9% respectively, vs 11·8% overall). The strongest pairwise associations also involved these conditions: PTSD with chronic pain (sOER 3·65, 95% CI 2·96–4·50) and personality disorder with COPD (3·58, 2·74–4·67).Interpretation: Mental-physical MLTC is common, socially patterned and heterogeneous. Most cases involve depression/anxiety, but other mental conditions show distinct demographic profiles and patterns of physical comorbidity. Treating mental-physical MLTC as a single entity risks obscuring subgroups with markedly different needs, while the depression/anxiety subgroup offers a large and potentially tractable target for intervention.

Purpose of Review: Heart failure affects over 64 million people worldwide, yet diagnosis remains challenging, with nearly two-thirds of cases estimated to remain undiagnosed overall. This review synthesises current evidence on multimodal artificial intelligence (AI) for heart failure detection, covering fusion strategies, explainability, and implementation challenges. Recent Findings: Unimodal AI achieves good-to-excellent discrimination from routine clinical measurements, 12-lead electrocardiograms, and chest X-rays, while echocardiography and cardiac magnetic resonance enable measurement automation and phenotyping. However, subgroup performance and external validation remain limited. Multimodal AI combining ECG, echocardiography, radiography, and electronic health records improves discrimination and phenotyping compared to single-modality models, with the largest gains observed from foundation models pretrained on large imaging and waveform corpora. Summary: Multimodal AI could capture the multidimensional complexity of heart failure, but prospective validation, calibration, and generalisability remain critical gaps. Multicentre trials with standardised endpoints, co-designed interfaces, and regulatory frameworks are priorities for deployment-centric algorithms.

Background: UK clinical guidance recommends that structured risk prediction tools and risk stratification should not be used in self-harm, to predict suicide or determine who is offered treatment. Underpinning this position is the premise that routinely collected health data contain no useful predictive signal, which has received little direct scrutiny. Objective: To test whether routinely collected electronic health record data can distinguish groups at higher and lower risk of severe outcomes following paracetamol overdose. Methods: We analysed 4,095 adults presenting to NHS Lothian emergency departments with paracetamol overdose (2017-2023). Elastic-net logistic regression was fitted to 37 routinely collected electronic health record features to predict a composite of death or mental health inpatient admission at 0-7, 8-30 and 31-365 days following attendance, evaluated on a held-out 20% test set with bootstrapping. Findings: Events occurred in 5.5% of patients at 0-7 days, 2.0% at 8-30 days and 7.9% at 31-365 days, dominated by mental health admission. Bootstrap AUROC 95% confidence intervals lay above 0.5 in every window (0.65-0.82, 0.63-0.90, 0.71-0.85): models ranked patients better than chance. Calibration slopes (1.04, 1.14, 1.07) were close to one. Ranking drew primarily on mental health-related features. Conclusions: Routinely collected health data carried predictive signal for severe outcomes after paracetamol overdose, although discrimination fell short of what is needed for individual-level clinical use. Clinical implications: These models are not proposed for clinical deployment; however, treating risk prediction as a settled question will redirect research efforts, potentially excluding this patient population from machine learning advances driving improvements in care in other medical specialties. Summary box: What is already known on this topic: NICE NG225 (2022), NHS England’s Staying Safe from Suicide framework (2025) and NCISH guidance (2024) recommend against structured risk prediction in self-harm, both for predicting suicide or repetition and for allocating treatment. An increasingly common reading of the underpinning literature is that routinely collected health data contain no useful predictive signal in this population. Evaluations of this claim in large-scale UK linked healthcare datasets remain scarce. What this study adds: In a whole-population cohort of adults attending emergency departments with paracetamol overdose, models using 37 structured electronic health record features separated groups at higher and lower risk of all-cause mortality or mental health admission across all three time horizons, with calibration slopes close to one. Discriminative signal was carried primarily by mental health-related features rather than demographics. Useful predictive signal is therefore recoverable from routinely collected UK healthcare data in this population. How this study might affect research, practice or policy: These findings are not evidence that suicide can be predicted at the individual level, and the models are not proposed for clinical use. They do, however, indicate that the empirical question underlying current guidance remains open, and should stay subject to scientific enquiry and to revision if sufficiently robust models emerge. Future research should expand the feature space beyond structured records, particularly into clinical free text, where the information on which psychosocial assessment relies already resides.

Objective: To identify and characterise primary care or community based interventions for patients with multimorbidity involving depression or anxiety, and to determine their effectiveness for improving patients’ mental health, physical health, and quality of life. Design: Systematic review with meta-analysis. Data sources: Medline, Embase, Cochrane Library, CINAHL, PsycInfo, and Web of Science databases, from inception to 11 November 2024. Eligibility criteria for selecting studies: Included studies were randomised controlled trials of primary care or community based interventions targeting adults with depression or anxiety disorders and one or more long term physical conditions. Risk of bias assessment used the Cochrane risk of bias tool. Interventions were categorised as organisational or patient oriented, and were subgrouped by intervention type. Intervention components were systematically categorised, and effects on mental health and quality of life outcomes were meta-analysed in groups defined by intervention type and assessment time point. Physical health outcomes were too heterogenous to meta-analyse and were synthesised without meta-analysis with Fisher's method for combining P values. Results: 29 randomised controlled trials comprising 9487 participants were included. High quality evidence was found for organisational interventions (n=10, including collaborative care, stepped care, and post-discharge interventions) which resulted in small improvements in symptoms of depression (standardised mean difference −0.25, 95% confidence interval (CI) −0.43 to −0.06) and quality of life (0.21, 0.01 to 0.41), but had no effect on symptoms of anxiety at the end of the intervention. No effect on depression or anxiety symptoms was observed, and no data for quality of life were found from organisational interventions at the late follow-up period (18-24 months). In the subgroup analysis, collaborative care resulted in sustained improvements in symptoms of depression at 18-24 months. Synthesis without meta-analysis showed evidence of benefit from organisational interventions (specifically collaborative care) on physiological (eg, haemoglobin A _1c levels), but not on functional (eg, disability) or global physical health outcomes. Low to moderate quality evidence was found for patient oriented interventions (n=19; interventions including exercise, psychotherapy, and psychoeducation) which led to small improvements in symptoms of depression (standardised mean difference −0.46, 95% CI −0.71 to −0.21) and quality of life (0.22, 0.14 to 0.29) at the end of the intervention. These effects were diminished at the late follow-up period (≥12 months). In the subgroup analysis, no reported data for the long term effects of exercise, psychotherapy, or psychoeducation (18-24 months after randomisation) were found. Synthesis without meta-analysis showed evidence of benefit from patient oriented interventions (primarily psychotherapy) on physiological, functional, and global physical health outcomes. Conclusions: The study showed that interventions improved mental health, physical health, and quality of life outcomes in people with multimorbidity involving depression or anxiety, but the effects were small and, for patient oriented interventions in particular, diminished over time. Systematic review registration: PROSPERO CRD420251004355.

Background: Qualitative studies suggest discrepancies between older adults’ self-perceived and measured frailty. Quantification of this is limited. This study investigated the relationship between older adults’ self-perceived frailty and a measure of frailty derived from electronic heath record data (electronic Frailty Index [eFI] score). The eFI is derived routinely from available primary care electronic health record data and is based on the cumulative deficit model of frailty. Methods: One thousand people (≥ 70 years), randomly selected from a GP practice, were sent a survey, asking them to rate their frailty (ordinal and binary scale), and complete self-rated health (SRH) and PRISMA-7 questionnaires. We analysed (a) agreement between self-perceived frailty (ordinal scale) and eFI categorised frailty; (b) discrimination of self-report measures for eFI defined frailty (threshold ≥ 0.12); and (c) predictors of self-perceived frailty (logistic regressions). Results: 375 people were analysed (median age 76, 51% female). Agreement was ‘fair’ between self-perceived frailty and eFI (linear weighted Kappa 0.25, quadratic weighted Kappa 0.37). Agreement was higher with linear and quadratic weighted Gwet’s second order agreement co-efficient [AC2]), (0.65 and 0.81 respectively). As eFI increased, agreement with self-perceived frailty decreased. Disagreements commonly reflected self-perceived frailty reported as less severe than eFI. Self-perceived frailty poorly discriminated eFI defined frailty (AUC 0.59, 95%CI 0.55-0.63) as did SRH, while PRISMA-7 reached moderate discrimination (AUC 0.71, 95%CI 0.66-0.76). The optimal eFI cut-point for discriminating self-perceived frailty was 0.17. A multivariable regression model revealed increasing age (OR 1.10 per year, 95%CI 1.02-1.18) and depression (OR 1.51, 95%CI 1.31-1.74) were associated with self-perceived frailty, however, sex, anxiety, eFI score and deprivation were not. Conclusions: The mismatch between self-perceived and eFI categorised frailty has implications for the social acceptability of screening and for meaningful engagement with frailty interventions including advance care planning.

Cardiac troponin is central to risk stratification in patients with suspected acute coronary syndrome. High-sensitivity assays measuring cardiac troponin (hs-cTn) I or T are recommended by international guidelines. However, diagnostic thresholds for these assays are derived from different populations. We aimed to evaluate whether the change in assay influenced risk stratification. This is a secondary analysis of the TWITCH-ED study enrolling 25,849 patients with suspected acute coronary syndrome between October 2020 and October 2022. In October 2021, sites changed from a hs-cTnI to a hs-cTnT assay. Study outcomes included discharge from the Emergency Department following a single cardiac troponin measurement, a composite outcome of subsequent myocardial infarction, heart failure hospitalization, or cardiovascular death at one year, and all-cause death at one year. Patients were stratified as low-risk at presentation using established thresholds. Higher proportions of patients were identified as low-risk (57.3% versus 17.6%) and discharged following a single troponin measurement (47% versus 30%) using the hs-cTnI assay, compared to the hs-cTnT assay (P for both). Those identified as low-risk with the hs-cTnT assay were younger (median 52 [IQR 40-62] versus 41 [32-51] years) with fewer comorbidities, and had lower rates for the composite outcome (0.8% versus 0.2%) and all-cause death (2.0% versus 0.3%) (P for both) at one year. Risk stratification thresholds for hs-cTnI and hs-cTnT assays do not provide equivalent performance. The hs-cTnT assay and threshold identified fewer low-risk patients suitable for discharge, with fewer subsequent cardiovascular events at 1 year.