UNITE: Uncertainty-based Health Risk Prediction Leveraging Multi-sourced Data

Developing deep learning-based strategies to predict the risk of hepatocellular carcinoma among patients with nonalcoholic fatty liver disease from electronic health records

Background: Non-Alcoholic Fatty Liver Disease (NAFLD) is a common co-morbidity with obesity. Recent studies in adults with obesity have shown an association between subclinical hypothyroidism (SH) and NAFLD. Few studies of children from Europe have also assessed the association of SH with NAFLD diagnosed by ultrasound (1). Little is known about this association in children from underrepresented minorities in the US, especially in children of Latina ancestry and those with biopsy-proven NAFLD. Thyroid hormone impacts major metabolic pathways and overt hypothyroidism has been associated with insulin resistance and dyslipidemia. Aim: To assess the prevalence of SH in children of primarily Latina ancestry with biopsy-proven NAFLD as compared to children of the same age range visiting the clinic during the same years. Methods: We performed a retrospective study of 4363 children aged 5-19 years seen in an urban ambulatory care setting primarily serving children of Latina ancestry (> 80%). Sixty-three of these children had NAFLD proven by liver biopsy. Results of thyroid, liver, and lipid studies as well as demographic data were extracted from the electronic health record. A descriptive evaluation was performed. TSH values were divided into quartiles: Q1: TSH ≤1.07 mIU/L, Q2: TSH 1.07-1.59 mIU/L, Q3: TSH 1.59-2.36 mIU/L, Q4: TSH ≥ 2.36 mIU/L. The association of TSH quartiles with NAFLD was assessed by univariate analysis and multivariate logistic regression adjusting for age, gender, and severity of obesity. All analyses were performed in R statistical software. Results: There was a higher proportion of boys (74% vs. 40%, p < 0.001) in those with NAFLD, and they were younger (12.9±3.1 vs. 14.7±3.7 years, p < 0.001). Children with biopsy-proven NAFLD had significantly higher BMI, TSH, and ALT when compared to those without NAFLD (TSH, cases: 2.9 ± 1.2 mIU/L, controls: 1.9 ± 1.2 mIU/L, p<0.0001). The proportion of children with abnormal ALT, AST, and GGT was significantly higher in those with NAFLD compared to the controls (p < 0.001). They also had a higher proportion of abnormal total, HDL, and non-HDL cholesterol (p = 0.01). The odds of having NAFLD was statistically higher for children in Q3 (OR 5.07, p=0.003) and Q4 (OR 8.48, p<0.0001) of TSH, compared to those in Q1. After adjusting for age, gender, and severity of obesity, biopsy-proven NAFLD approached significantly higher odds (2.7, p = 0.07) in Q3 and was statistically significant in Q4 (OR= 4.0, p=0.0113). Conclusion: Children of Latina ancestry with NAFLD diagnosed by histology are at higher odds of having SH. Based on the results of this study, we hypothesize that thyroid dysfunction plays a mediator role in the development of NAFLD and should be investigated in all children with obesity.

ObjectivesSarcopenia is a known risk factor for non-alcoholic fatty liver disease (NAFLD). Studies evaluating the association between the fat-to-muscle ratio (FMR) and NAFLD are limited. Therefore, the aim of our...

The increasing global prevalence of diabetes has created an urgent need for accurate and early risk stratification methods. Artificial intelligence (AI) has emerged as a powerful tool for predictive healthcare, particularly through the integration of multi-source healthcare data. This study presents a systematic review of 52 peer-reviewed articles published between 2015 and 2024, focusing on AI-based diabetes risk stratification models that utilize diverse data sources, including electronic health records (EHRs), wearable device data, demographic factors, and genomic information. The findings indicate that machine learning and deep learning models significantly enhance predictive performance. Across the reviewed studies, predictive accuracy ranged from 78% to 94%, while sensitivity and specificity averaged 85% and 83%, respectively. Models integrating multi-source data outperformed single-source approaches by 12–18% in accuracy and demonstrated improved generalizability. Additionally, studies incorporating real-time wearable data reported a 15–20% increase in early risk detection rates, enabling more proactive interventions. Ensemble learning techniques achieved the highest performance, with an average area under the curve (AUC) of 0.89–0.93, compared to 0.75–0.82 for traditional statistical models. However, several challenges persist. Only 27% of the studies included external clinical validation, and approximately 35% reported issues related to data heterogeneity and missing values. Concerns regarding model interpretability and patient data privacy were also highlighted in over 40% of the reviewed papers. Overall, the study demonstrates that AI-driven multi-source data integration substantially improves diabetes risk prediction and emphasizes the need for standardized, explainable, and clinically validated models to support real-world implementation.

Clinicians usually intend to get overall health status as well as specific health care information from patient's electronic health record (EHR). However, the amount of EHR data can be incredibly large, which makes it a very tedious and inefficient work for clinicians to extract valuable information. This paper first investigates the clinical relevance of EHR data content. Then it presents a Level of Detail (LOD) information navigation model, which classified patient's whole EHR into different detail levels according to their clinical relevance. A novel navigation and visualization method is brought forward based on this model. Finally, an application is designed and implemented based on this method. It can guide clinicians from summarized overview to detailed contents of the EHR smoothly. Both the evolution of specific health problems and other related EHR information can be visualized and located quickly.

The prevalence of nonalcoholic fatty liver disease (NAFLD) is increasing worldwide. Accumulating evidence suggests that serum ferritin and uric acid (UA) are strongly associated with the risk of NAFLD, but no consensus has been reached. We sought to demonstrate the association between serum ferritin, UA levels, and NAFLD risk in a large cohort study. We separated 2,049 patients into non-NAFLD and NAFLD groups. The NAFLD group had four subgroups based on serum ferritin and four subgroups based on UA quartile levels. We used binary logistic regression to evaluate the correlation between serum ferritin, UA, and NAFLD. Additionally, an area under the curve (AUC) of receiver operating characteristic analysis (ROC) was used to predict the diagnostic value of combined serum ferritin and UA for NAFLD. Serum ferritin and UA levels were higher in the NAFLD group compared with the non-NAFLD group. Serum lipid and liver transaminase concentrations were elevated with the increase of serum ferritin and UA. The logistic regression results showed an independent correlation between serum ferritin, UA, and NAFLD. In the NAFLD group, the AUC value of serum ferritin and UA was 0.771. Increased serum ferritin and UA levels are independent risk factors for NAFLD. Increased serum UA is a stronger risk factor for NAFLD than elevated serum ferritin. Serum ferritin and UA can be important predictors of NAFLD risk.

Higher levels of thyroid-stimulating hormone (TSH) in the euthyroid state can negatively affect the metabolic health, including nonalcoholic fatty liver disease (NAFLD). We studied the effect of TSH levels in the setting of normal levels of thyroid hormone on all-cause and cause-specific mortality stratified by NAFLD status. The National Health and Nutrition Examination Survey (NHANES) III from 1988 to 1994 and NHANES III-linked mortality data through 2015 were used. NAFLD was defined as ultrasonographically diagnosed hepatic steatosis without coexisting liver diseases. Subclinical hypothyroidism was defined as a TSH level over 4.5 mIU/L and "low-normal" thyroid function as higher TSH level (2.5-4.5 mIU/L) within the euthyroid reference range. The Cox proportional hazard model analyzed the all-cause mortality and cause-specific mortality. In a multivariate logistic regression analysis, individuals with low thyroid function demonstrated an association with NAFLD in a dose-dependent manner. During a median follow-up of 23 years, low thyroid function was associated with increased all-cause mortality only in the univariate model. Low thyroid function was associated with a higher risk for all-cause mortality in individuals with NAFLD and not in those without NAFLD. Furthermore, low thyroid function was associated with a higher risk for cardiovascular mortality in the entire population and among those with NAFLD but demonstrated no association with the non-NAFLD group. In this large nationally representative sample of American adults, low thyroid function was associated with NAFLD and a predictor of higher risk for all-cause and cardiovascular mortality in individuals with NAFLD.

Non-alcoholic fatty liver disease (NAFLD) is strongly associated with cardiovascular disease in the general population. We aimed to assess the impact of NAFLD and liver fibrosis on intermediate-high cardiovascular risk in people living with HIV. We included people living with HIV from three cohorts. NAFLD and significant liver fibrosis were defined using transient elastography: controlled attenuation parameter ≥288 dB/m and liver stiffness measurement ≥7.1 kPa, respectively. Cardiovascular risk was assessed with the atherosclerotic cardiovascular disease (ASCVD) risk estimator in patients aged between 40 and 75years and categorised as low if <5%, borderline if 5%-7.4%, intermediate if 7.5%-19.9% and high if ≥20% or with the presence of a previous cardiovascular event. Patients with hepatitis B and/or hepatitis C virus co-infection, alcohol abuse and unreliable transient elastography measurements were excluded. Predictors of intermediate-high cardiovascular risk were investigated in multivariable analysis by logistic regression and also by stratifying according to body mass index (BMI; cut-offs of 25 and 30kg/m2 ) and age (cut-off of 60years). Of 941 patients with HIV alone included, 423 (45%), 128 (13.6%), 260 (27.6%) and 130 (13.8%) were categorised as at low, borderline, intermediate and high ASCVD risk, respectively. Predictors of intermediate-high ASCVD risk were NAFLD (adjusted odds ratio [aOR] 2.11; 95% confidence interval [CI] 1.40-3.18; p<0.001), liver fibrosis (aOR 1.64; 95% CI 1.03-2.59; p=0.034), duration of HIV (aOR 1.04; 95% CI 1.02-1.06; p<0.001), and previous exposure to thymidine analogues and/or didanosine (aOR 1.54; 95% CI 1.09-2.18; p=0.014). NAFLD was also associated with higher cardiovascular risk in normoweight patients (aOR 2.97; 95% CI 1.43-6.16; p=0.003), in those with BMI <30kg/m2 (aOR 2.30; 95% CI 1.46-3.61; p<0.001) and in those aged <60 years (aOR 2.19; 95% CI 1.36-3.54; p=0.001). Assessment of cardiovascular disease should be targeted in people living with HIV with NAFLD and/or significant liver fibrosis, even if they are normoweight and young.

Introduction: Non-alcoholic fatty liver disease (NAFLD), a primary hepatic manifestation of metabolic syndrome, is now recognized as a significant contributor to cardiovascular disease burden. It is unclear whether advanced NAFLD severity independently predicts major adverse cardiovascular events (MACE) because severity has not been consistently defined, and few considered potential confounding factors. Methods: Following the PRISMA 2020 guidelines, we looked through the Cochrane Library, Web of Science, Embase, and PubMed till January 2025 for longitudinal cohort studies assessing MACE incidence according to NAFLD severity. Eligible studies included adults with biopsy-verified (NASH or fibrosis ≥ F2) or imaging-verified (MRI-PDFF ≥10% or CAP ≥280 dB/m) NAFLD and stratified as mild vs moderate/severe, with an average follow-up of 1 year or greater. Two reviewers separately conducted the data extraction and assessed for bias using the Newcastle-Ottawa Scale. Ninety-five percent confidence intervals (CI) and pooled risk ratios (RR) were determined using a random-effects model. Heterogeneity (I²) and small-study effects (Egger's test) were assessed. Sensitivity analyses were performed to exclude studies at high risk of bias. Results: Three cohort studies met inclusion criteria for MACE outcomes (N = 136,789; mean age: 52–58; duration: 5.2–8.1 years). One study on advanced (F3–F4/NASH) and mild (F0–F1/steatosis) NAFLD calculated a fully adjusted RR of 1.50 (95% CI 0.97–2.32; P = 0.07), indicating a trend towards greater MACE risk. The pooled RR for NAFLD, compared to no NAFLD, was 1.19 (95% CI 0.80-1.79; I² = 94%) with sensitivity analysis RR (95% CI) 1.12 (0.85-1.48). There is a pooled RR for MACE among NAFLD cases across all severity strata of 1.26 (95% CI 0.92-1.73; I² = 88%). Egger's test did not report evidence of publication bias (P = 0.34). Only 1 study controlled for diabetes and dyslipidemia, and none for visceral adiposity. Conclusion: This meta-analysis suggests that histologically or imaging-defined advanced NAFLD may be associated with a clinically meaningful but statistically insignificant increased risk of MACE. Substantial heterogeneity and residual confounding affect definitive interpretations. Future studies should employ standardized severity cutoffs (such as MRI-PDFF ≥10% or fibrosis stage ≥ F2), adjust for cardiometabolic risk factors, and distinguish the atherosclerotic and non-atherosclerotic MACE components to strengthen the prediction of cardiovascular risk in NAFLD.

Nowadays, population is growing on large scale along with the problems faced by the people are also increasing. Thus, healthcare industry is making lot of technological advancements to provide effective, faster and cheaper treatment to people. Digitization of health records are also expanding in zettabytes. Electronic Health Record (EHR) containing all the patient’s medical history, demographics and other clinical data is also used in hospitals for improved care co-ordination. To avoid critical conditions of people from chronic diseases like hypertension, diabetes, hyperlipidemia etc. there is a need for building a health risk prediction model. But, when whole EHR data is provided to this risk prediction model causes overfitting of features. Overfitting is caused when model learns the details & noise from dataset, thus having negative impact on the performance. Hence, a feature selection approach is proposed for discarding redundant features from EHR. Improved sparse logistic regression method selects the best suitable parameters and forwards to risk prediction model. This regression method improvises the model with the use of logistic loss function that controls the sparsity factor. Neural network is used as a risk prediction model. This paper describes the risk prediction of hypertension disease. Thus, people could take preventive measures for the disease.

The impact of galectin-3 inhibitors on nonalcoholic fatty liver diseases (NAFLD)-related outcomes is currently under investigation in randomized clinical trials. Whether there is a causal association between plasma galectin-3 levels and NAFLD is unknown. To evaluate the causal effect of circulating galectin-3 levels on NAFLD as well as >800 other human diseases. Inverse variance-weighted (IVW) Mendelian randomization (MR) and phenome-wide MR. Summary statistics of genome-wide association studies. Participants of the UK Biobank, Electronic Medical Records and Genomics (eMERGE), FinnGen, Prevention of Renal and Vascular End-Stage Disease (PREVEND), and IMPROVE cohorts. Identification of independent single-nucleotide polymorphisms (SNPs) associated with galectin-3 levels (P < 5 × 10-8) in the PREVEND (14 SNPs) and IMPROVE (3 SNPs) cohorts. Presence of NAFLD in a meta-analysis of genome-wide association study of the eMERGE, UK Biobank, and FinnGen cohorts (3042 NAFLD cases and 504 853 controls), as well as >800 other human diseases in the UK Biobank and FinnGen. Using IVW-MR, we found no causal association between galectin-3 levels and NAFLD in the meta-analysis of the 3 cohorts or in each individual cohort. After correction for multiple testing, we found no causal association between galectin-3 levels and >800 human disease-related traits. This MR study revealed no causal associations between circulating galectin-3 levels and NAFLD or any other disease traits, suggesting that plasma galectin-3 levels may not be directly implicated in the pathogenesis of NAFLD or other human diseases.

PurposeProtein kinase B (PKB/AKT) has shown a high profile in the research of metabolic diseases. This research sought to determine whether the AKT1 gene’s single nucleotide polymorphisms (SNPs) and the risk of developing non-alcoholic fatty liver disease (NAFLD) were related.Patients and MethodsRecruited in this case–control study were 2693 subjects, including 815 with NAFLD and 1878 without NAFLD. Three SNPs of AKT1 (rs2494732, rs2494752 and rs1130233) were genotyped. To examine the correlation between SNPs and NAFLD susceptibility, logistic regression was performed.ResultsAfter adjusting for sex, age, triglyceride and glucose, AKT1 rs2494732-C (all P < 0.05 in co-dominant model, dominant model and additive model) and rs2494752-G (P < 0.05 in co-dominant model) were linked to a lower risk of NAFLD. The combined effect of both SNPs on NAFLD risk was statistically significant, showing a dose dependence (Ptrend = 0.010). Sex, body mass index, hypertension, hyperglycemia, hypertriglyceridemia, high-density lipoprotein-cholesterol, alanine aminotransferase, and beneficial alleles were all significant predictors of NAFLD risk (all P < 0.05). The prediction model achieved good discrimination, with an area under the receiver operating characteristic curve of 0.779. The Hosmer–Lemeshow test suggested an inadequate calibration of the model (χ2 = 21.073, P = 0.007).ConclusionAKT1 rs2494732 and rs2494752 may be related to Chinese NAFLD susceptibility. The prediction model combining both SNPs with clinical factors displays a strong ability to discriminate NAFLD patients. Both SNPs may be exploited to design new models for early screening of NAFLD high-risk population.

BackgroundDeep learning models have attracted significant interest from health care researchers during the last few decades. There have been many studies that apply deep learning to medical applications and achieve promising results. However, there are three limitations to the existing models: (1) most clinicians are unable to interpret the results from the existing models, (2) existing models cannot incorporate complicated medical domain knowledge (eg, a disease causes another disease), and (3) most existing models lack visual exploration and interaction. Both the electronic health record (EHR) data set and the deep model results are complex and abstract, which impedes clinicians from exploring and communicating with the model directly.ObjectiveThe objective of this study is to develop an interpretable and accurate risk prediction model as well as an interactive clinical prediction system to support EHR data exploration, knowledge graph demonstration, and model interpretation.MethodsA domain-knowledge–guided recurrent neural network (DG-RNN) model is proposed to predict clinical risks. The model takes medical event sequences as input and incorporates medical domain knowledge by attending to a subgraph of the whole medical knowledge graph. A global pooling operation and a fully connected layer are used to output the clinical outcomes. The middle results and the parameters of the fully connected layer are helpful in identifying which medical events cause clinical risks. DG-Viz is also designed to support EHR data exploration, knowledge graph demonstration, and model interpretation.ResultsWe conducted both risk prediction experiments and a case study on a real-world data set. A total of 554 patients with heart failure and 1662 control patients without heart failure were selected from the data set. The experimental results show that the proposed DG-RNN outperforms the state-of-the-art approaches by approximately 1.5%. The case study demonstrates how our medical physician collaborator can effectively explore the data and interpret the prediction results using DG-Viz.ConclusionsIn this study, we present DG-Viz, an interactive clinical prediction system, which brings together the power of deep learning (ie, a DG-RNN–based model) and visual analytics to predict clinical risks and visually interpret the EHR prediction results. Experimental results and a case study on heart failure risk prediction tasks demonstrate the effectiveness and usefulness of the DG-Viz system. This study will pave the way for interactive, interpretable, and accurate clinical risk predictions.

A population-based recalibration method for updating survival neural networks models for cardiovascular risk prediction in United Kingdom and China.

BackgroundWith no approved pharmacological treatments for non-alcoholic fatty liver disease (NAFLD) in Taiwan, identifying protective and risk factors is crucial for preventing disease progression. Given the clinical heterogeneity of NAFLD, this study aimed to identify clinically meaningful NAFLD phenotypes using electronic medical records (EMRs) and unsupervised clustering, stratify risk across different clusters, identify factors associated with disease progression, and derive a parsimonious set of predictors for high-risk phenotypes.MethodsThis study was a retrospective cohort study conducted in three steps with iterative model training. In step 1, patients diagnosed with NAFLD were identified, and all relevant patient data were extracted, followed by clustering analysis using the k-prototype algorithm. In step 2, survival analysis and Cox regression were applied to perform risk stratification across clusters. In step 3, Lasso regression, logistic regression, and receiver operating characteristic (ROC) curve analysis were used to identify potential protective and risk factors associated with NAFLD and to derive a parsimonious set of predictors for high-risk phenotypes across different risk strata.ResultsStep 1: The analysis of 6,023 patients identified four distinct phenotypic clusters. The first cluster had the most severe disease, the second the least. Step 2: Among 4,998 patients, the first cluster faced the highest risk for all outcomes, with a median survival of 3.06 years, significantly different from the others. There was no significant risk difference between the second and third clusters. Step 3: A comparison of the highest-risk and lowest-risk clusters finally identified 17 potential variables.ConclusionsUsing multiple analytical models, this study identified 17 potential risk factors associated with NAFLD progression. Their combined assessment may inform future risk stratification and hypothesis generation. Further validation is required before clinical application.