Search results
Found 30896 matches for
Sample selection bias in machine learning for healthcare
While machine learning algorithms hold promise for personalised medicine, their clinical adoption remains limited, partly due to biases that can compromise the reliability of predictions. In this paper, we focus on sample selection bias (SSB), a specific type of bias where the study population is less representative of the target population, leading to biased and potentially harmful decisions. Despite being well-known in the literature, SSB remains scarcely studied in machine learning for healthcare. Moreover, the existing machine learning techniques try to correct the bias mostly by balancing distributions between the study and the target populations, which may result in a loss of predictive performance. To address these problems, our study illustrates the potential risks associated with SSB by examining SSB’s impact on the performance of machine learning algorithms. Most importantly, we propose a new research direction for addressing SSB, based on the target population identification rather than the bias correction. Specifically, we propose two independent networks (T-Net) and a multitasking network (MT-Net) for addressing SSB, where one network/task identifies the target subpopulation which is representative of the study population and the second makes predictions for the identified subpopulation. Our empirical results with synthetic and semi-synthetic datasets highlight that SSB can lead to a large drop in the performance of an algorithm for the target population as compared with the study population, as well as a substantial difference in the performance for the target subpopulations that are representative of the selected and the non-selected patients from the study population. Furthermore, our proposed techniques demonstrate robustness across various settings, including different dataset sizes, event rates, and selection rates, outperforming the existing bias correction techniques.
A review of the role of EFEMP1 in ophthalmic disease.
EGF-containing fibulin extracellular matrix protein 1 (EFEMP1), or fibulin-3, is an extracellular matrix glycoprotein encoded by the EFEMP1 gene. The role of EFEMP1 in the human eye is incompletely understood, but there are well-reported associations between mutations in the gene and a variety of ophthalmic diseases, such as myopia, juvenile open-angle glaucoma (JOAG), primary open-angle glaucoma (POAG) and familial drusen formation in Malattia Leventinese (ML)/Doyne honeycomb retinal dystrophy (DHRD). Variants which interact with EFEMP1 have also been identified in genome-wide association studies (GWAS) for age-related macular degeneration (AMD). Many of these conditions form a large component of ophthalmology case-load and have incompletely characterized pathogenesis. In this review, we will describe the role of EFEMP1 in ophthalmic disease. We discuss the role of EFEMP1 in Mendelian eye disease, its polygenic contributions to common ophthalmic conditions, and the potential to target EFEMP1 for therapeutic purposes.
Tumor-specific CD8 T cell characterization in HR+ breast cancer reveals an impaired antitumoral response in patients with lymph node metastasis.
Most breast cancers express the estrogen receptor (ER), but the immune response of hormone receptor-positive (HR+) breast cancer remains poorly characterized. Here, dendritic cells loaded with tumor lysate are used to identify tumor-reactive CD8 T cells, which are detected in most HR+ breast cancer patients, especially those with early-stage tumors. When present, the circulating antitumor CD8 response contains cytotoxic T cells with diverse specificity and T cell receptor (TCR) repertoire. Additionally, patients with blood cancer-specific T cells have significantly more CD8 tumor-infiltrating lymphocytes (TILs). Moreover, tumor-reactive TCR sequences are detected in the tumor, but at a significantly lower proportion in patients with lymph node involvement. Our data suggest that HR+ breast cancer patients with lymph node metastasis lack tumor-specific CD8 T cells with capacity to infiltrate the tumor at significant levels. However, early-stage patients have a diverse antitumor CD8 response that could be harnessed to develop immunotherapeutic approaches for late-stage HR+ patients.
Characterising the genetic architecture of common elastic tissue disorders using UK Biobank
Background Elastic fibres are key extracellular matrix (ECM) components that provide elasticity to tissues throughout the body, allowing them to recover after deformation. Derangement of elastic fibres lead to cutaneous and systemic disorders that show elastic tissue pathology. To distinguish between rarer disorders of elastic tissue, and common disease, I have coined the term ‘elastopathies.’ Common elastopathies such as hiatus hernia, diverticular disease, haemorrhoids, inguinal hernia, varicose veins, female genital prolapse, umbilical hernia, aneurysmal disease, emphysema, pneumothorax, rectal prolapse, and femoral hernia, have a complex aetiology where genetic predisposition and environmental factors interplay to influence overall phenotypic expression. These elastopathies are highly prevalent and exert a significant healthcare and socioeconomic burden. However, their genetic basis remains poorly defined, with limited putative genes identified. Method To unravel the genetic architecture of the 12 elastopathies identified in the UK Biobank resource (hiatus hernia, diverticular disease, haemorrhoids, inguinal hernia, varicose veins, female genital prolapse, umbilical hernia, aneurysmal disease, emphysema, pneumothorax, rectal prolapse, and femoral hernia), genome-wide association study (GWAS) testing was performed across ~400,000 genotyped participants from the UK Biobank resource, with replication from ~410,000 participants from 23andMe, Inc. (California) for the varicose veins analysis. All 12 elastopathies were studied independently, and then combined in a final pan-elastopathy GWA study to identify the shared genetic architecture of common elastopathies in UK Biobank. A disparate analysis was performed combining all four types of hernia in a pan-hernia analysis to identify shared genetics. Genes and pathways were prioritised using a suite of bioinformatic approaches, and pharmacological targets identified using the Open Targets Platform. A genetic risk score was constructed to examine the genetic burden among participants with severe disease. For the pan-hernia analysis, multi-trait and multivariate meta-analysis approaches were deployed to uncover the shared genetic susceptibility to multiple hernia phenotypes. For the pan-elastopathy analysis, an individual patient data (IPD) meta-analysis was performed and the latent elastopathy phenotype was analysed using genomic structural equation modelling (SEM) to uncover shared genetic biology. Results Performing the largest two-stage GWAS of varicose veins in 810,625 participants, forty-nine signals at 46 susceptibility loci were discovered, including 29 previously unreported associations. Next, through (at the time) the first-ever GWA study of haemorrhoids in over 400,000 participants, 13 signals at 12 novel loci were discovered to associate with haemorrhoids. Association analysis of inguinal, femoral, umbilical, and hiatus hernia individually yielded 58 signals at 38 loci (34 new) associated with the four hernia phenotypes. When combined in a multi-trait meta-analysis, 12 biologically relevant putative loci were discovered to associate with multiple hernia phenotypes, demonstrating novel and robust evidence of shared susceptibility to hernia. Of significance, the genetic risk scoring correlated with disease severity across the varicose veins, haemorrhoids, and pan-hernia analyses, with patients undergoing surgery having a higher genetic burden than those managed non-surgically. Lastly, studying the 12 elastopathies in a pan-elastopathy IPD meta-analysis, 18 susceptibility loci were discovered to associate with the pan-elastopathy phenotype which were not discovered when the 12 elastopathies were studied individually. Moreover, employing common factor analysis to unveil the latent elastopathy phenotype, a further four loci were discovered to be integral to a shared genetic risk towards elastopathies. Collectively, over 250 independent susceptibility loci were discovered to associate with the 12 elastopathies, which were mapped to over 500 putative genes, many of which demonstrated profound evidence of a shared genetic biology, therapeutic tractability and clustered in pathways pertaining to core matrisomal components and ECM homeostasis. Conclusion Prioritised genes and pathways demonstrate significant biological plausibility, and represent promising candidates for further investigation of elastic tissue biology and potential pharmacological targeting. The genetic risk score correlated with disease across varicose veins, haemorrhoids and hernia disorders, representing an important proof-of-principle for the future use of genetic risk scoring in personalised medicine approaches to surgical disorders. Lastly, studying the 12 elastic tissue disorders together, a novel category of pathologically linked disorders defined by elastic tissue dysfunction were discovered— the elastopathies. To this end, this thesis advances the field of study around elastopathies and complex trait genetics.
Exploring cellular changes in ruptured human quadriceps tendons at single-cell resolution.
Tendon ruptures in humans have often been studied during the chronic phase of injury, particularly in the context of rotator cuff disease. However, the early response to acute tendon ruptures remains less investigated. Quadriceps tendons, which require prompt surgical treatment, offer a model to investigate this early response. Therefore, this study aimed to explore the early cellular changes in ruptured compared to healthy human quadriceps tendons. Quadriceps tendon samples were collected from patients undergoing fracture repair (healthy) or tendon repair surgery (collected 7-8 days post-injury). Nuclei were isolated for single-nucleus RNA sequencing, and comprehensive transcriptomic analysis was conducted. The transcriptomes of 12,808 nuclei (7268 from healthy and 5540 from ruptured quadriceps tendons) were profiled, revealing 12 major cell types and several cell subtypes and states. Rupture samples showed increased expression of genes related to extracellular matrix organisation and cell cycle signalling, and a decrease in expression of genes in lipid metabolism pathways. These changes were predominantly driven by gene expression changes in the fibroblast, vascular endothelial cell (VEC), mural cell, and macrophage populations: fibroblasts shift to an activated phenotype upon rupture and there is an increase in the proportion of capillary and dividing VECs. A diverse immune environment was observed, with a shift from homeostatic to activated macrophages following rupture. Cell-cell interactions increased in number and diversity in rupture, and primarily involved fibroblast and VEC populations. Collectively, this transcriptomic analysis suggests that fibroblasts and endothelial cells are key orchestrators of the early injury response within ruptured quadriceps tendon. KEY POINTS: Tendon ruptures in humans have regularly been studied during the chronic phase of injury, but less is known about the early injury response after acute tendon ruptures. This study explored the early cellular changes in ruptured compared to healthy human quadriceps tendons at single-cell resolution. Fibroblasts and endothelial cells seem to be the key orchestrators of the early injury response within ruptured quadriceps tendon. Therefore, these cell types are obvious targets for interventions to enhance tendon healing. Overall, this study highlights that the development of more effective therapeutic options for tendon injury requires better understanding of the cellular, extracellular, and mechanical landscape of tendon tissue.
Striving to recover - wrist splint or plaster cast : a qualitative study of patients' experience of recovery after a distal radius fracture.
AIMS: We sought to explore patients' experience of early recovery from a fracture of the wrist (distal radius). This study was nested in the DRAFT-CASP randomized controlled trial (RCT), which explores the effectiveness of two treatment pathways for patients with a fracture of the distal radius that does not require manipulation: a plaster cast which is removed in fracture clinic, versus a wrist splint that patients remove themselves without returning to hospital. METHODS: Qualitative interviews were undertaken with 21 adults (mean age 58.2 years (SD 13.96), six male), from eight NHS hospitals, four to ten weeks post injury. Interviews were informed by phenomenology and analyzed using reflexive thematic analysis. RESULTS: We identified the overarching theme 'striving to recover', which conveys patients' determination to get back to normal after a wrist fracture. To recover, patients needed to be comfortable, to adapt, and to be certain that their wrist was healing. Early in their recovery, they were unable to complete their daily activities, experienced pain, loss of strength, worry, and were cautious about using their wrist. Overall, both treatments were considered acceptable. The splint was advantageous for the freedom and control it provided. The cast was valued for the protection and safety it provided. Both groups required more information and reassurance, but had varied views on the need for follow-up appointments. CONCLUSION: The splint made life easier for patients and was an acceptable treatment. Patients wanted reassurance that their wrist was healing, but they felt this could be achieved in a variety of ways. Most patients coped without a follow-up appointment. Innovative ways to maximize recovery are required. These include support for patients to, manage their pain and provide comfort, be able to adapt, and feel certain of healing. Sharing patients' experiences may help future patients to make informed treatment and recovery decisions.
Facilitating trial recruitment: A qualitative study of patient and staff experiences of an orthopaedic trauma trial
<jats:title>Abstract</jats:title> <jats:p>Background Qualitative research has been used to explore patients’ and healthcare professionals’ experiences of surgical randomised controlled trials (RCTs). From this research, reasons why patients accept or decline participation and barriers to engaging clinicians in trials have been identified. In a trauma setting, recruitment to surgical trials can be particularly difficult as patients may require urgent treatment and their ability to consider their options, ask questions and reach a decision may be hindered by the impact of their injury. Little research however, has explored patients’ and healthcare professionals’ experiences of surgical RCTs in a trauma setting. This study aimed to understand participants’ and staffs’ experiences of an orthopaedic trauma trial. Methods We carried out semi-structured interviews with 11 participants and 24 staff (10 surgeons and 14 research associates) participating in a UK multi-centre feasibility trial comparing intramedullary nails versus distal locking plates for fractures of the distal femur (TrAFFix). Interviews explored patients’ experience of TrAFFix and their reason for participating and staffs’ experience of recruiting to TrAFFix and trauma trials more generally. Interviews were audio recorded and transcribed verbatim. Transcripts were analysed using thematic analysis. Results Three themes were identified. These were i) navigating research with patients after orthopaedic trauma, ii) knowing that it is the right decision and iii) making it work. These themes reflect: i) how research associates supported and guided patients through the consent process enabling them to participate, ii) the difficulty in engaging surgeons in a trial when individual equipoise and experience of the interventions is low despite the presence of community equipoise and iii) the way in which research teams worked together and encouraged the development of a research culture within the clinical teams in order to facilitate recruitment. Conclusions Our findings highlight the pivotal role of research associates (RAs) in facilitating trial recruitment. RAs supported patients to enable them to make a decision about participation and assisted in developing a research culture within the team by promoting studies and communicating research to clinical staff. Our findings also reinforce surgeons’ difficulty with equipoise and suggest that accepting community equipoise could facilitate recruitment.</jats:p>
A phase 1b, multicentre, dose escalation, safety and pharmacokinetics study of tilvestamab (BGB149) in relapsed, platinum-resistant, high-grade serous ovarian cancer (PROC) patients.
BackgroundTilvestamab is a highly selective humanised immunoglobulin G1 anti-AXL monoclonal antibody. This phase 1 study evaluated its optimal dose, safety, tolerability, immunogenicity and pharmacokinetics (PK) in relapsed platinum-resistant HGSOC patients.MethodsPatients received tilvestamab in three dose levels (1 mg/kg, 3 mg/kg and 5 mg/kg) via IV infusion every 2 weeks. Primary objectives included safety, tolerability and PK. Exploratory objectives included overall response, progression-free survival (PFS) and quality-of-life measures. Pharmacodynamic included AXL expression, gene and protein changes by transcriptomic and proteomic analysis.ResultsBetween 25 February 2021 and 4 February 2022, 16 patients were enroled across 8 sites in Singapore, Korea, United Kingdom, and Norway. Median treatment duration was 6.1 weeks. Grade 3 or higher treatment-emergent adverse events occurred in 62.5% patients, but none were tilvestamab-related. Common events included fatigue (38%), anorexia (38%) infections (31%), anaemia (25%) and dyspnoea (25%). No objective responses were observed, but 7 (44%) had stable disease at 6 weeks. PK showed dose-proportional exposure and steady-state by the second dose. Pharmacodynamic analyses revealed reduced fibrosis-related gene signatures and AXL protein expression. Epithelial-mesenchymal transition reversal was seen in 2 patients.ConclusionTilvestamab was well-tolerated and further studies to examine the efficacy of AXL inhibition in other indications are required.Clinical trial registrationThis trial is registered at https://clinicaltrials.gov .Registration numberNCT04893551. EudraCT Number: 2020-001382-36.
A systems-based approach to uterine fibroids identifies differential splicing associated with abnormal uterine bleeding.
BACKGROUND: Uterine fibroids (UFs), benign tumours prevalent in up to 80% of women of reproductive age, are associated with significant morbidity, including abnormal uterine bleeding, pain and infertility. Despite identification of key genomic alterations in MED12 and HMGA2, the pathogenic mechanisms underlying UFs and heavy menstrual bleeding (HMB) remain poorly understood. METHODS: To correlate systematically genetic, transcriptional and proteomic phenotypes, we conducted an integrative multi-omic approach utilising targeted DNA sequencing, RNA sequencing and proteomic methodologies, encompassing fibroid, myometrium, and endometrium tissues from 91 patients. RESULTS: In addition to confirming the presence of MED12 mutations, we identify variants in AHR and COL4A6. Multi-omic analysis of endometrium identifies latent factors that correlate with HMB and fibroid presence with driver mutations of MED12, AHR, and COL4A6, which are associated with pathways involved in angiogenesis, extracellular matrix organisation and RNA splicing. We propose a model, supported by in vivo evidence, where altered signalling of MED12-mutated fibroids influences RNA transcript isoform expression in endometrium, potentially leading to abnormal uterine bleeding. CONCLUSIONS: This study presents a comprehensive integrative approach, revealing that genetic alterations in UF may influence endometrial function via signalling impacts on the RNA splicing mechanism. Our findings advance the understanding of complex molecular pathways in UF pathogenesis and UF-associated endometrial dysfunction, offering insights for targeted therapeutic development.
Evaluation of the impact of artificial intelligence-assisted image interpretation on the diagnostic performance of clinicians in identifying endotracheal tube position on plain chest X-ray: a multi-case multi-reader study.
BACKGROUND: Incorrectly placed endotracheal tubes (ETTs) can lead to serious clinical harm. Studies have demonstrated the potential for artificial intelligence (AI)-led algorithms to detect ETT placement on chest X-Ray (CXR) images, however their effect on clinician accuracy remains unexplored. This study measured the impact of an AI-assisted ETT detection algorithm on the ability of clinical staff to correctly identify ETT misplacement on CXR images. METHODS: Four hundred CXRs of intubated adult patients were retrospectively sourced from the John Radcliffe Hospital (Oxford) and two other UK NHS hospitals. Images were de-identified and selected from a range of clinical settings, including the intensive care unit (ICU) and emergency department (ED). Each image was independently reported by a panel of thoracic radiologists, whose consensus classification of ETT placement (correct, too low [distal], or too high [proximal]) served as the reference standard for the study. Correct ETT position was defined as the tip located 3-7 cm above the carina, in line with established guidelines. Eighteen clinical readers of varying seniority from six clinical specialties were recruited across four NHS hospitals. Readers viewed the dataset using an online platform and recorded a blinded classification of ETT position for each image. After a four-week washout period, this was repeated with assistance from an AI-assisted image interpretation tool. Reader accuracy, reported confidence, and timings were measured during each study phase. RESULTS: 14,400 image interpretations were undertaken. Pooled accuracy for tube placement classification improved from 73.6 to 77.4% (p = 0.002). Accuracy for identification of critically misplaced tubes increased from 79.3 to 89.0% (p = 0.001). Reader confidence improved with AI assistance, with no change in mean interpretation time at 36 s per image. CONCLUSION: Use of assistive AI technology improved accuracy and confidence in interpreting ETT placement on CXR, especially for identification of critically misplaced tubes. AI assistance may potentially provide a useful adjunct to support clinicians in identifying misplaced ETTs on CXR.