Bhautesh Dinesh Jani

Background: Machine learning (ML) is increasingly applied to predicting chronic kidney disease (CKD) progression and kidney failure using routinely collected healthcare data, but the methodological quality and clinical applicability of these models remain uncertain. We hypothesised that ML models would show strong discrimination but be limited by inadequate external validation, inconsistent calibration, and high risk of bias (ROB), restricting clinical implementation. Methods: We conducted a systematic review according to PRISMA guidance. Ovid MEDLINE, Embase, and CINAHL were searched from inception for studies evaluating ML models predicting CKD progression, kidney failure, or kidney replacement therapy using routinely collected healthcare data. Studies were synthesised by modelling framework, and ROB was assessed using PROBAST. Results: Fifty-two studies were included. ML models demonstrated favourable discrimination, with 37/52 studies reporting AUROC or concordance estimates ≥ 0.80. Where directly comparable, gains over conventional statistical models and established risk scores, including the Kidney Failure Risk Equation (KFRE), were often modest. External validation and calibration assessment were each reported in 19/52 studies, while 41/52 studies were at high ROB. Kidney function, proteinuria-related variables, and age were the most influential predictors. Substantial heterogeneity was observed in cohort design, outcomes, prediction horizons, and validation methodology. Inconsistent handling of proteinuria, missing data, and model-development workflows further limited assessment of transportability and real-world applicability. Conclusions: ML models for CKD prognostic prediction demonstrate favourable discrimination across kidney outcomes. However, limited external validation, inconsistent calibration, and methodological heterogeneity restrict clinical implementation. Future progress requires rigorous validation, transparent methodology, and demonstration of real-world clinical utility.

Background and hypothesis: Large differences between glomerular filtration rate estimated with creatinine and cystatin C—termed estimated glomerular filtration rate (eGFR) discordance—have important implications for health. Several demographic factors and individual long-term conditions (LTCs) are associated with eGFR discordance. We hypothesize that the presence of multiple LTCs, i.e. multimorbidity is likely to be associated with eGFR discordance. Methods: Differences between creatinine-based and cystatin C-based eGFR [eGFR using serum creatinine (eGFRcr) and eGFR using serum cystatin C (eGFRcys), respectively] using European Kidney Function Consortium equations were calculated in over 400 000 UK Biobank participants. Large negative eGFR difference was defined as an eGFRcys >30% lower than eGFRcr, i.e. ([eGFRcys − eGFRcr]/eGFRcr) < −30%. Self-reported LTCs were counted and analysed overall and by type of multimorbidity (cardiometabolic, non-cardiometabolic, mixed mental and physical). Proportions of participants with large negative eGFR differences were cross-tabulated at different chronic kidney disease (CKD) stages and according to number of LTCs. The association between number of LTCs and eGFR discordance was assessed in logistic regression models adjusted for age, sex, body mass index, smoking status, and thyroid disease. Results: Multimorbidity (two or more LTCs of any type) was observed in 33% of the overall cohort and 63% of those with large negative eGFR differences. When CKD stage was assigned based on eGFRcr and then reassigned using eGFRcys, 15% of the cohort were reclassified to a more advanced CKD stage with frequency of reclassification increasing progressively with increasing counts of LTCs. Individuals with four or more LTCs of any type were more than four-fold more likely to have a large negative eGFR difference (adjusted odds ratio 4.64, 95% confidence interval 4.13–5.21) compared to individuals with no LTCs. Conclusion: Discordance between creatinine-based and cystatin C-based eGFR increased substantially with multimorbidity. Cystatin C testing should be considered (alongside creatinine) in individuals with multimorbidity to refine estimation of kidney function and risk stratification.

Background: Preventing multimorbidity requires understanding how long-term conditions begin to accumulate. Clinical management of a first long-term condition may provide an opportunity for secondary prevention of multimorbidity. Cardiovascular-kidney-metabolic (CKM) conditions share interconnected risk factors and established monitoring and risk-reduction pathways, but whether progression differs after a first CKM versus non-CKM condition remains unclear.Methods: We conducted a retrospective cohort study of adults aged 18 years and older in Japan, the UK, and Denmark, with no or one of 22 harmonised conditions most relevant in primary care. We used a continuous-time multistate framework to examine progression from a first condition to two conditions (MM2), then three or more conditions (MM3+), and death after MM2. We estimated three-year absolute risks and HRs, using time-split Cox models (0–1, 1–3, and ≥3 years), stratified by sex and adjusted for age and calendar year.Findings: We included 3 258 240 participants in Japan, 1 570 985 in the UK, and 3 391 192 in Denmark. Three-year absolute multimorbidity risk was higher after CKM first conditions than non-CKM first conditions: Japan: 44·7% versus 38·8%, UK: 32·9% versus 22·2%, Denmark: 40·2% versus 32·8%. CKM-first trajectories had higher hazards of progression to multimorbidity than non-CKM-first trajectories across all countries and time bands. Corresponding HRs for non-CKM versus CKM first conditions ranged from 0·84 to 0·97 in Japan, 0·69 to 0·86 in the UK, and 0·70 to 0·90 in Denmark. After MM2, associations with MM3+ varied by pattern, age, country, and time; the non-CKM-to-non-CKM pattern had the highest mortality hazard across countries.Interpretation: A first CKM condition was associated with faster progression to MM2, but absolute risk after a first non-CKM condition remained substantial. After MM2, further accumulation and mortality varied by disease composition and recorded sequence. Secondary prevention for multimorbidity should address both CKM-first and non-CKM-first trajectories.

Obesity is a key risk factor for chronic disease and multimorbidity, yet variation in how obesity is defined may influence risk estimation in clinical practise. Waist‐to‐hip ratio (WHR) and waist‐to‐height ratio (WHtR) are better at capturing fat distribution than body mass index (BMI) and may have a role in obesity related multimorbidity prediction. To examine multimorbidity risk across obesity groups classified using BMI in combination with WHR/WHtR. A UK Biobank cohort free of comorbidity at baseline ( n = 179 876) was followed for incident first long‐term condition and multimorbidity (≥ 2 of 38 conditions) over a mean of 9.9 years. Six adiposity categories combining BMI with WHR/WHtR were analysed using Cox proportional hazards models. Subgroup analyses evaluated whether high central obesity (WHtR ≥ 0.6) provided additional risk stratification within BMI groups. Combined general with central obesity showed highest risk for the first condition (HR 1.11; 95% CI 1.087–1.138) and multimorbidity (HR 1.13; 95% CI 1.093–1.172). However, BMI‐stratified subgroup analyses, central obesity provided additional risk discrimination only among individuals in the overweight category (BMI 25–29.9 kg/m ² ), but not among those with BMI ≥ 30 kg/m ² . Waist‐based measures improved risk identification in adults with overweight BMI, supporting the potential role of combined BMI‐waist approaches for earlier multimorbidity prevention.

Background: Psoriatic arthritis (PsA) is a chronic inflammatory musculoskeletal disease associated with psoriasis, affecting an estimated 0.13% of adults worldwide. People living with PsA often experience multiple long-term conditions (MLTCs) or multimorbidity (the presence of two or more long-term health conditions), such as hypertension, diabetes, obesity, and metabolic syndrome. Multimorbidity and treatment burden which is the workload of healthcare experienced by individuals and its impact on wellbeing may exacerbate poor health outcomes and complicate disease management in PsA. However, the influence of MLTCs and/or treatment burden on adverse health-related outcomes in this population remains poorly understood. Objective: This systematic review protocol will outline the methods to evaluate the current evidence regarding the impact, if any, of MLTCs and/or treatment burden on mortality and other adverse health-related outcomes in individuals with PsA. Design: Systematic review of the literature. The following databases will be searched: MEDLINE, EMBASE, CINAHL, PsycINFO, and Scopus. Longitudinal quantitative studies will be eligible for inclusion. Study selection will follow predefined eligibility criteria, and methodological quality and risk of bias will be assessed using the Cochrane Quality in Prognostic Studies tool. A narrative synthesis will be undertaken, and meta-analysis will be considered where appropriate. This protocol follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA-P) 2015 guidelines. Conclusions: Understanding the impact of MLTCs and/or treatment burden on health-related outcomes in PsA is essential for improving future clinical management. This review will help identify existing knowledge gaps and advance precision medicine strategies to improve health-related outcomes for individuals living with PsA.

Objective: To identify which chronic conditions and patient characteristics contribute to increases in the number of regularly visited facilities (RVF), a measure of polydoctoring, among adults with multimorbidity in Japan. Design: Retrospective cohort study using individual fixed-effects Poisson regression models to estimate within-person associations between the onset of each chronic condition and changes in RVF. This approach controlled for all time-invariant personal characteristics. Effect modification by age, geographic region and baseline multimorbidity was also assessed. Setting: A nationwide health claims database covering multiple insurance systems in Japan, including municipal National Health Insurance, Employees’ Health Insurance and the Long-Life Medical Care System. Participants: 4 696 790 adults with two or more chronic conditions and two or more consecutive years of follow-up. Results: The onset of examined chronic conditions was associated with changes in RVF, with substantial variation in effect size across conditions. The strongest association was observed for malignancy (RR=1.102; 95% CI 1.101 to 1.103), followed by osteoporosis (RR=1.065; 95% CI 1.064 to 1.065), arthritis (RR=1.061; 95% CI 1.060 to 1.061), obesity (RR=1.049; 95% CI 1.044 to 1.055), stroke (RR=1.044; 95% CI 1.041 to 1.048) and cardiovascular disease (RR=1.043; 95% CI 1.042 to 1.044). Associations between disease onset and RVF were consistently attenuated among individuals aged ≥75 years and those with higher baseline multimorbidity while differences between urban and rural residents were minimal. Conclusion: Substantial heterogeneity was observed in the magnitude of RVF increases across chronic conditions, with particularly strong associations for malignancy, osteoporosis and arthritis. Attenuated effects among older adults and individuals with higher multimorbidity suggest that patient capacity and treatment burden may influence patterns of healthcare utilisation. These findings underscore condition-specific drivers of care fragmentation and may inform more tailored care-coordination strategies for people living with multimorbidity.

Background: This study’s aim was to evaluate the performance of artificial intelligence (AI)-based programmes in transcribing and generating clinically accurate and complete summaries of general practice consultations in real-time. AI has seen many recent advancements, making it more accessible and implementable within the healthcare profession. New large language models (LLMs) such as ChatGPT and BioBART have presented the possibility of developing tools to handle common administrative tasks in general practice, with some options already appearing on the market without any peer-reviewed evaluations. A general practitioner (GP) in the UK, on average, spends 14% of their day on clinical administrative tasks of making entries in electronic health records (EHRs), and greater administrative burden is known to be associated with increased physician burnout. There is also a possibility of recall bias-related inaccuracies in addition to the administrative burden of making entries into EHRs as often these tasks are performed at the end of the clinic rather than at the end of the consultation. We wanted to explore the possibility of using LLMs for summarising GP consultations in real-time, by doing a proof-of-concept study. Methods: This study quantitatively evaluated four selected LLM (some of them were fine-tuned based on a bank of 17,000 patient-doctor dialogues). We used publicly available N=47 simulated GP consultations and applied the selected LLM in generating summaries for these consultations. The accuracy of these AI-generated summaries was evaluated independently by two reviewers independently by a modified form of the Q-note score, which has been previously validated to evaluate accuracy of EHR entries. Results: After preliminary evaluation, a model fine-tuned on GPT-4 was used for the full analysis. The model fine-tuned using GPT-4 produced 47 consultation summaries, with 80.9% of the summaries scoring 46 or above out of a total score of 52. The results were further analysed by dividing the data based on 12 different categories based on the symptom profile of presentations, and no significant performance difference was detected in the accuracy of AI-generated summaries (P value =0.48). The most frequent errors noted were related to failure in recording relevant negative findings, recording of over-the-counter medications, and recording of numerical findings from the physical examination. Conclusions: The cloud-based GPT-4 model was successful in consistently producing reasonably complete and accurate summaries of simulated primary care consultations. Further research is required to address some of the limitations in recording certain aspects of GP consultations and may require further training of LLM to address that. Future research also needs to evaluate the performance of these models in real-life situations which will help further iterations and refinements. Our study, however, is an important first step in safe implementation of AI/LLM in summarising doctor-patient consultations.

Background and hypothesis: Guidelines recommend using risk prediction models for predicting kidney failure in chronic kidney disease (CKD). Many people with CKD have multiple long-term conditions (multimorbidity), which influences outcomes including kidney failure and mortality. This study validated the four-variable kidney failure risk equation (KFRE) in individuals with CKD, with and without multimorbidity, comparing performance of KFRE using creatinine and cystatin C to calculate estimated glomerular filtration rate (eGFR) and updated the model to account for competing mortality risks. Methods: Observational cohort study using research-based (UK Biobank) and population-based cohorts (Stockholm Creatinine Measurements project: SCREAM). Multimorbidity was defined as two or more long-term conditions in addition to CKD. Kidney failure was defined as long-term dialysis or kidney transplantation. KFRE performance assessment included discrimination, calibration, and overall fit at 2 and 5 years. An updated model (using the same variables as KFRE) accounting for competing mortality risks was developed and validated. Results: 14 998 of 24 489 individuals in UK Biobank (61.2%) and 30 147 of 42 902 individuals in SCREAM (70.3%) had multimorbidity. Discrimination of KFRE was good (area under curve $\ge $0.86 across eGFR equations in all cohorts, multimorbidity groups and time horizons). Kidney failure risk was under-estimated in people with multimorbidity in UK Biobank (observed/expected (O/E) ratio 1.75 at 5 years; eGFR creatinine). Conversely, calibration-in-the-large (O/E ratio) at 5 years in SCREAM was 1.05 in the multimorbidity group (eGFR creatinine). Using cystatin C compared to creatinine did not improve model performance. Cumulative incidence of death was higher with multimorbidity compared to no multimorbidity. An updated model considering competing mortality improved calibration performance over KFRE, O/E ratio 0.98 in multimorbidity group of the validation cohort (UK Biobank) at 5 years. Conclusion: Competing mortality risk is important when predicting kidney failure, particularly for people with multimorbidity. An updated model accounting for competing mortality risk, permits improved model performance.

Background: Exercise and self-management support may be clinically effective and cost-effective treatments for a range of individual long-term conditions (LTCs), as they activate multiple beneficial physiological and psychological mechanisms. We aimed to develop a complex intervention to deliver exercise and self-management support for people with multiple LTCs (MLTCs). Methods: Following the Person Based Approach to intervention development, we conducted ten co-development workshops with people with MLTCs, family and friends; healthcare providers; service commissioners and policymakers. The workshops iteratively identified the unmet needs of people with MLTCs and informed a programme theory outlining theoretical mechanisms of change and intervention strategies to change the targeted behaviours. They also identified ideas for efficient delivery and service providers’ training needs. Mixed methods feedback from the Personalised Exercise-Rehabilitation FOR people with Multiple long-term conditions (PERFORM) feasibility study (reported elsewhere) informed intervention refinement. Results: A diverse group of stakeholders (26 people with MLTCs/supporters, 13 service providers, 16 experts in chronic illness and 14 service commissioners) helped to develop the PERFORM intervention. This included 16 supervised exercise sessions and 16 ‘Health and Wellbeing’ self-management support sessions, delivered in hospital or community settings over eight weeks, plus check-in sessions at four and six months. The self-management sessions covered maintenance of exercise/physical activity, healthy eating and managing common symptoms (pain, fatigue, breathlessness, stress). Conclusion: The PERFORM intervention is a comprehensive, evidence-informed, theoretically driven self-management and exercise-based rehabilitation intervention, co-developed with people with MLTCs, service providers and service commissioners. PERFORM is now ready for evaluation regarding clinical effectiveness and cost-effectiveness.