Developing automated methods for disease subtyping in UK Biobank: an exemplar study on stroke.
Ontology highlight
ABSTRACT: Background
Better phenotyping of routinely collected coded data would be useful for research and health improvement. For example, the precision of coded data for hemorrhagic stroke (intracerebral hemorrhage [ICH] and subarachnoid hemorrhage [SAH]) may be as poor as < 50%. This work aimed to investigate the feasibility and added value of automated methods applied to clinical radiology reports to improve stroke subtyping.Methods
From a sub-population of 17,249 Scottish UK Biobank participants, we ascertained those with an incident stroke code in hospital, death record or primary care administrative data by September 2015, and ≥ 1 clinical brain scan report. We used a combination of natural language processing and clinical knowledge inference on brain scan reports to assign a stroke subtype (ischemic vs ICH vs SAH) for each participant and assessed performance by precision and recall at entity and patient levels.Results
Of 225 participants with an incident stroke code, 207 had a relevant brain scan report and were included in this study. Entity level precision and recall ranged from 78 to 100%. Automated methods showed precision and recall at patient level that were very good for ICH (both 89%), good for SAH (both 82%), but, as expected, lower for ischemic stroke (73%, and 64%, respectively), suggesting coded data remains the preferred method for identifying the latter stroke subtype.Conclusions
Our automated method applied to radiology reports provides a feasible, scalable and accurate solution to improve disease subtyping when used in conjunction with administrative coded health data. Future research should validate these findings in a different population setting.
SUBMITTER: Rannikmae K
PROVIDER: S-EPMC8204419 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BackgroundIdentifying people at risk of cardiovascular diseases (CVD) is a cornerstone of preventative cardiology. Risk prediction models currently recommended by clinical guidelines are typically based on a limited number of predictors with sub-optimal performance across all patient groups. Data-driven techniques based on machine learning (ML) might improve the performance of risk predictions by agnostically discovering novel risk predictors and learning the complex interactions between them. We tested (1) whether ML techniques based on a state-of-the-art automated ML framework (AutoPrognosis) could improve CVD risk prediction compared to traditional approaches, and (2) whether considering non-traditional variables could increase the accuracy of CVD risk predictions.Methods and findingsUsing data on 423,604 participants without CVD at baseline in UK Biobank, we developed a ML-based model for predicting CVD risk based on 473 available variables. Our ML-based model was derived using AutoPrognosis, an algorithmic tool that automatically selects and tunes ensembles of ML modeling pipelines (comprising data imputation, feature processing, classification and calibration algorithms). We compared our model with a well-established risk prediction algorithm based on conventional CVD risk factors (Framingham score), a Cox proportional hazards (PH) model based on familiar risk factors (i.e, age, gender, smoking status, systolic blood pressure, history of diabetes, reception of treatments for hypertension and body mass index), and a Cox PH model based on all of the 473 available variables. Predictive performances were assessed using area under the receiver operating characteristic curve (AUC-ROC). Overall, our AutoPrognosis model improved risk prediction (AUC-ROC: 0.774, 95% CI: 0.768-0.780) compared to Framingham score (AUC-ROC: 0.724, 95% CI: 0.720-0.728, p < 0.001), Cox PH model with conventional risk factors (AUC-ROC: 0.734, 95% CI: 0.729-0.739, p < 0.001), and Cox PH model with all UK Biobank variables (AUC-ROC: 0.758, 95% CI: 0.753-0.763, p < 0.001). Out of 4,801 CVD cases recorded within 5 years of baseline, AutoPrognosis was able to correctly predict 368 more cases compared to the Framingham score. Our AutoPrognosis model included predictors that are not usually considered in existing risk prediction models, such as the individuals' usual walking pace and their self-reported overall health rating. Furthermore, our model improved risk prediction in potentially relevant sub-populations, such as in individuals with history of diabetes. We also highlight the relative benefits accrued from including more information into a predictive model (information gain) as compared to the benefits of using more complex models (modeling gain).ConclusionsOur AutoPrognosis model improves the accuracy of CVD risk prediction in the UK Biobank population. This approach performs well in traditionally poorly served patient subgroups. Additionally, AutoPrognosis uncovered novel predictors for CVD disease that may now be tested in prospective studies. We found that the "information gain" achieved by considering more risk factors in the predictive model was significantly higher than the "modeling gain" achieved by adopting complex predictive models.
| S-EPMC6519796 | biostudies-literature
Project description:BackgroundPrevious studies have revealed the involvement of coffee and tea in the development of stroke and dementia. However, little is known about the association between the combination of coffee and tea and the risk of stroke, dementia, and poststroke dementia. Therefore, we aimed to investigate the associations of coffee and tea separately and in combination with the risk of developing stroke and dementia.Methods and findingsThis prospective cohort study included 365,682 participants (50 to 74 years old) from the UK Biobank. Participants joined the study from 2006 to 2010 and were followed up until 2020. We used Cox proportional hazards models to estimate the associations between coffee/tea consumption and incident stroke and dementia, adjusting for sex, age, ethnicity, qualification, income, body mass index (BMI), physical activity, alcohol status, smoking status, diet pattern, consumption of sugar-sweetened beverages, high-density lipoprotein (HDL), low-density lipoprotein (LDL), history of cancer, history of diabetes, history of cardiovascular arterial disease (CAD), and hypertension. Coffee and tea consumption was assessed at baseline. During a median follow-up of 11.4 years for new onset disease, 5,079 participants developed dementia, and 10,053 participants developed stroke. The associations of coffee and tea with stroke and dementia were nonlinear (P for nonlinear <0.01), and coffee intake of 2 to 3 cups/d or tea intake of 3 to 5 cups/d or their combination intake of 4 to 6 cups/d were linked with the lowest hazard ratio (HR) of incident stroke and dementia. Compared with those who did not drink tea and coffee, drinking 2 to 3 cups of coffee and 2 to 3 cups of tea per day was associated with a 32% (HR 0.68, 95% CI, 0.59 to 0.79; P < 0.001) lower risk of stroke and a 28% (HR, 0.72, 95% CI, 0.59 to 0.89; P = 0.002) lower risk of dementia. Moreover, the combination of coffee and tea consumption was associated with lower risk of ischemic stroke and vascular dementia. Additionally, the combination of tea and coffee was associated with a lower risk of poststroke dementia, with the lowest risk of incident poststroke dementia at a daily consumption level of 3 to 6 cups of coffee and tea (HR, 0.52, 95% CI, 0.32 to 0.83; P = 0.007). The main limitations were that coffee and tea intake was self-reported at baseline and may not reflect long-term consumption patterns, unmeasured confounders in observational studies may result in biased effect estimates, and UK Biobank participants are not representative of the whole United Kingdom population.ConclusionsWe found that drinking coffee and tea separately or in combination were associated with lower risk of stroke and dementia. Intake of coffee alone or in combination with tea was associated with lower risk of poststroke dementia.
| S-EPMC8594796 | biostudies-literature
Project description:ObjectiveFor the UK Biobank, standardized phenotype codes are associated with patients who have been hospitalized but are missing for many patients who have been treated exclusively in an outpatient setting. We describe a method for phenotype recognition that imputes phenotype codes for all UK Biobank participants.Materials and methodsPOPDx (Population-based Objective Phenotyping by Deep Extrapolation) is a bilinear machine learning framework for simultaneously estimating the probabilities of 1538 phenotype codes. We extracted phenotypic and health-related information of 392 246 individuals from the UK Biobank for POPDx development and evaluation. A total of 12 803 ICD-10 diagnosis codes of the patients were converted to 1538 phecodes as gold standard labels. The POPDx framework was evaluated and compared to other available methods on automated multiphenotype recognition.ResultsPOPDx can predict phenotypes that are rare or even unobserved in training. We demonstrate substantial improvement of automated multiphenotype recognition across 22 disease categories, and its application in identifying key epidemiological features associated with each phenotype.ConclusionsPOPDx helps provide well-defined cohorts for downstream studies. It is a general-purpose method that can be applied to other biobanks with diverse but incomplete data.
| S-EPMC9846671 | biostudies-literature
Project description:Metabolic syndrome (MetS) is characterized by a constellation of metabolic risk factors, including obesity, hypertriglyceridemia, low high-density lipoprotein (HDL) levels, hypertension, and hyperglycemia, and is associated with stroke and neurodegenerative diseases. This study capitalized on brain structural images and clinical data from the UK Biobank and explored the associations of brain morphology with MetS and brain aging due to MetS. Cortical surface area, thickness, and subcortical volumes were assessed using FreeSurfer. Linear regression was used to examine associations of brain morphology with five MetS components and the MetS severity in a metabolic aging group (N = 23,676, age 62.8 ± 7.5 years). Partial least squares (PLS) were employed to predict brain age using MetS-associated brain morphology. The five MetS components and MetS severity were associated with increased cortical surface area and decreased thickness, particularly in the frontal, temporal, and sensorimotor cortex, and reduced volumes in the basal ganglia. Obesity best explained the variation of brain morphology. Moreover, participants with the most severe MetS had brain age 1-year older than those without MetS. Brain age in patients with stroke (N = 1042), dementia (N = 83), Parkinson's (N = 107), and multiple sclerosis (N = 235) was greater than that in the metabolic aging group. The obesity-related brain morphology had the leading discriminative power. Therefore, the MetS-related brain morphological model can be used for risk assessment of stroke and neurodegenerative diseases. Our findings suggested that prioritizing adjusting obesity among the five metabolic components may be more helpful for improving brain health in aging populations.
| S-EPMC10310693 | biostudies-literature
Project description:Background Therapeutic advances have reduced cardiovascular death rates in people with cardiovascular diseases (CVD). We aimed to define the rates of cardiovascular and noncardiovascular death in people with specified CVDs or accruing cardiovascular multimorbidity. Methods and Results We studied 493 280 UK residents enrolled in the UK Biobank cohort study. The proportion of deaths attributed to cardiovascular, cancer, infection, or other causes were calculated in groups defined by 9 distinct self-reported CVDs at baseline, or by the number of these CVDs at baseline. Poisson regression analyses were then used to define adjusted incidence rate ratios for these causes of death, accounting for sociodemographic factors and comorbidity. Of 27 729 deaths, 20.4% were primarily attributed to CVD, 53.6% to cancer, 5.0% to infection, and 21.0% to other causes. As cardiovascular multimorbidity increased, the proportion of cardiovascular and infection-related deaths was greater, contrasting with cancer and other deaths. Compared with people without CVD, those with 3 or more CVDs experienced adjusted incidence rate ratios of 7.0 (6.2-7.8) for cardiovascular death, 4.4 (3.4-5.6) for infection death, 1.5 (1.4-1.7) for cancer death, and 2.0 (1.7-2.4) for other causes of death. There was substantial heterogeneity in causes of death, both in terms of crude proportions and adjusted incidence rate ratios, among the 9 studied baseline CVDs. Conclusions Noncardiovascular death is common in people with CVD, although its contribution varies widely between people with different CVDs. Holistic and personalized care are likely to be important tools for continuing to improve outcomes in people with CVD.
| S-EPMC8751922 | biostudies-literature
Project description:BackgroundThe cause of the most common form of dementia, sporadic Alzheimer's disease (AD), remains unknown. This may reflect insufficiently powered studies to date for this multi-factorial disorder. The UK Biobank dataset presents a unique opportunity to rank known risk factors and determine novel variables.MethodsA custom machine learning approach for high dimensionality data was applied to explore prospectively associations between AD in a sub-cohort of 156,209 UK Biobank participants aged 60-70 including more than 2,090 who were subsequently diagnosed with AD.ResultsAfter the possession of the APOE4 allele, the next highest ranked risk factors were other genetic variants within the TOMM40-APOE-APOC1 locus. When stratified by their apolipoprotein epsilon 4 (APOE4) carrier status, the most prominent risk factors in carriers were AST:ALT ratio, the "number of treatments/ medications" taken as well as "time spent in hospital" while protection was conferred by "Sleeplessness/Insomnia". In non-APOE carriers, lower socioeconomic status and fewer years of education were highly ranked but effect sizes were small relative to APOE4 carriers.ConclusionsPossession of the APOE4 allele was confirmed as the most important risk factor in AD. Other TOMM40-APOE-APOC1 locus variants further moderate the risk of AD in APOE4 carriers. Liver pathology is a novel risk factor in APOE4 carriers while "Sleeplessness/Insomnia" is protective in AD irrespective of APOE4 status. Other factors such as "Number of treatments/ medications" suggest that multimorbidity is an important risk factor for AD. Future treatments aimed at co-morbidities, including liver disease, may concomitantly lower the risk of sporadic AD.
| S-EPMC10293768 | biostudies-literature
Project description:Parkinson disease (PD) has become one of the most rapidly growing causes of disability among the older population and social isolation is a major concern in the PD community. However, the relationship between social isolation and future risk of PD remains unclear. This study included 192,340 participants aged 60 or older who were free of dementia and PD at baseline from the UK Biobank study. Social isolation was measured using a composite score derived from three questions on number in household, frequency of friend/family visits, and leisure/social activities. Incident PD cases were identified through electronic health records. Multivariable-adjusted Cox regression models were used to compute the hazard ratio (HR) and 95% confidence interval (CI). Among the 192,340 participants (mean [standard deviation] age, 64.2 [2.9] years; 103,253 [53.7%] women), 89,075 (46.3%) participants were in the least isolated group and 26,161 (13.6%) were in the most isolated group. Over a median follow-up of 12.5 years, 2048 incident PD cases were documented. Compared to the least isolated group, the multivariable-adjusted HRs (95% CIs) for PD were 1.00 (0.91-1.10) for the moderately isolated group and 1.19 (1.05-1.36) for the most isolated group (P-trend = 0.04). The observed association was independent of the genetic susceptibility to PD and consistent in subgroup analyses. Social isolation was associated with a higher risk of PD regardless of genetic risk. Our findings highlighted the importance of developing screening and intervention strategies for social isolation among older adults to reduce the risk of PD.
| S-EPMC11001945 | biostudies-literature
Project description:ImportanceAn increasing body of evidence indicates an association between consuming sugar or its alternatives and cardiometabolic diseases. However, the effects of the consumption of sugar-sweetened beverages, artificially sweetened beverages, and natural juices on kidney health remain unclear.ObjectiveTo investigate the association of the intake of sugar-sweetened beverages, artificially sweetened beverages, and natural juices with the risk of chronic kidney disease (CKD), and the effect of substituting these beverage types for one another on this association.Design, setting, and participantsThis prospective, population-based cohort study analyzed data from the UK Biobank. Participants without a history of CKD who completed at least 1 dietary questionnaire were included. The follow-up period was from the date of the last dietary questionnaire until October 31, 2022, in England; July 31, 2021, in Scotland; and February 28, 2018, in Wales. Data were analyzed from May 1 to August 1, 2023.ExposuresConsumption of sugar-sweetened beverages, artificially sweetened beverages, and natural juices.Main outcomes and measuresThe primary outcome was incident CKD. Multivariable Cox proportional hazards models were used to estimate the associations between the 3 beverage types and incident CKD. A substitution analysis was used to evaluate the effect on the associations of substituting one beverage type for another.ResultsA total of 127 830 participants (mean [SD] age, 55.2 [8.0] years; 66 180 female [51.8%]) were included in the primary analysis. During a median (IQR) follow-up of 10.5 (10.4-11.2) years, 4459 (3.5%) cases of incident CKD occurred. The consumption of more than 1 serving per day of sugar-sweetened beverages was associated with higher risk of incident CKD (adjusted hazard ratio [AHR], 1.19 [95% CI, 1.05-1.34]) compared with not consuming sugar-sweetened beverages. The AHR for participants consuming more than 0 to 1 serving per day of artificially sweetened beverages was 1.10 (95% CI, 1.01-1.20) and for consuming more than 1 serving per day was 1.26 (95% CI, 1.12-1.43) compared with consuming no artificially sweetened beverages. By contrast, there was no significant association between natural juice intake and incident CKD (eg, for >1 serving per day: HR, 0.99 [95% CI, 0.87-1.11]; P = .10). Substituting sugar-sweetened beverages with artificially sweetened beverages did not show any significant difference in the risk of CKD (HR, 1.03 [95% CI, 0.96-1.10]). Conversely, replacing 1 serving per day of sugar-sweetened beverage with natural juice (HR, 0.93 [95% CI, 0.87-0.97]) or water (HR, 0.93 [95% CI, 0.88-0.99]) or replacing 1 serving per day of artificially sweetened beverage with natural juice (HR, 0.90 [95% CI, 0.84-0.96]) or water (HR, 0.91 [95% CI, 0.86-0.96]) was associated with a reduced risk of incident CKD.Conclusions and relevanceFindings from this cohort study suggest that lower consumption of sugar-sweetened beverages or artificially sweetened beverages may reduce the risk of developing CKD.
| S-EPMC10902724 | biostudies-literature
Project description:MotivationEpidemiological cohorts typically contain a diverse set of phenotypes such that automation of phenome scans is non-trivial, because they require highly heterogeneous models. For this reason, phenome scans have to date tended to use a smaller homogeneous set of phenotypes that can be analysed in a consistent fashion. We present PHESANT (PHEnome Scan ANalysis Tool), a software package for performing comprehensive phenome scans in UK Biobank.General featuresPHESANT tests the association of a specified trait with all continuous, integer and categorical variables in UK Biobank, or a specified subset. PHESANT uses a novel rule-based algorithm to determine how to appropriately test each trait, then performs the analyses and produces plots and summary tables.ImplementationThe PHESANT phenome scan is implemented in R. PHESANT includes a novel Javascript D3.js visualization and accompanying Java code that converts the phenome scan results to the required JavaScript Object Notation (JSON) format.AvailabilityPHESANT is available on GitHub at [https://github.com/MRCIEU/PHESANT]. Git tag v0.5 corresponds to the version presented here.
| S-EPMC5837456 | biostudies-literature
Project description:Genome-wide association studies (GWASs) of medication use may contribute to understanding of disease etiology, could generate new leads relevant for drug discovery and can be used to quantify future risk of medication taking. Here, we conduct GWASs of self-reported medication use from 23 medication categories in approximately 320,000 individuals from the UK Biobank. A total of 505 independent genetic loci that meet stringent criteria (P < 10-8/23) for statistical significance are identified. We investigate the implications of these GWAS findings in relation to biological mechanism, potential drug target identification and genetic risk stratification of disease. Amongst the medication-associated genes are 16 known therapeutic-effect target genes for medications from 9 categories. Two of the medication classes studied are for disorders that have not previously been subject to large GWAS (hypothyroidism and gastro-oesophageal reflux disease).
| S-EPMC6478889 | biostudies-literature