Factors related to steroid treatment responsiveness in thyroid eye disease patients and application of SHAP for feature analysis with XGBoost

Predicting Visual Acuity after Retinal Vein Occlusion Anti-VEGF Treatment: Development and Validation of an Interpretable Machine Learning Model

Accurate prediction of post-treatment visual acuity in macular edema secondary to retinal vein occlusion (RVO-ME) is critical for optimizing anti-VEGF therapy and improving clinical outcomes. While machine learning (ML) has shown promise in ophthalmic prognostication, existing models often lack interpretability and clinical applicability for RVO management. This study developed and validated an interpretable ML model to predict visual acuity changes in RVO patients following anti-VEGF treatment. Using retrospective data from 259 RVO patients at the First Affiliated Hospital of Jinan University, we identified key predictive features through the Boruta algorithm and evaluated eight ML algorithms. The Extreme Gradient Boosting (XGBoost) model emerged as optimal, achieving an AUC of 0.91 (95% CI: 0.85-0.96) in the testing cohort with 0.83 accuracy, 0.88 sensitivity, 0.73 specificity, 0.87 F1 score, and 0.14 Brier score. Critical predictors included baseline visual acuity, systolic blood pressure (SBP), age, diabetic retinal inner layer dysfunction (DRIL), and disease subtype. Shapley Additive exPlanations (SHAP) analysis revealed baseline visual acuity as the most influential prognostic factor, followed by SBP and age. Our model seeks to bridge the critical gaps in current research: (1) systematically comparing the applicability and effects of different ML algorithms in RVO-ME visual acuity prediction, and (2) inherent interpretability through SHAP value visualization. The combination of high predictive performance (AUC > 0.9) with inherent clinical transparency may enable the practical implementation of this tool in guiding anti-VEGF treatment decisions. Future validation in multicenter cohorts could further strengthen its generalizability for personalized RVO management.

Construction of a machine learning-based interpretable prediction model for acute kidney injury in hospitalized patients

In this observational study, we used data from 59,936 hospitalized adults to construct a model. For the models constructed with all 53 variables, all five models achieved acceptable performance with the validation cohort, with the extreme gradient boosting (XGBoost) model showing the best predictive efficacy and stability (area under the curve (AUC), 0.9301). For the simpler models constructed with 39 significant variables screened by the random forest recursive feature elimination method, the XGBoost model also had the best performance (AUC, 0.9357). All the models showed significant net returns according to decision analysis curves, and the XGBoost model achieved the optimal results. In addition, the Shapley additive explanation (SHAP) importance matrices revealed that uric acid, colloidal solution, first creatinine value on admission, pulse and albumin represented the top five most important variables for both modeling strategies. With the external validation cohort based on 4022 hospitalized patients, the performance of all models declined, among which the Support vector machine (SVM) model showed the best predictive efficacy (AUC, 0.8230 and 0.8329), followed by the XGBoost model (0.8124 and 0.8316). Thus, our model can predict the occurrence and risk of acute kidney injury (AKI) up to 48 h in advance, enabling clinicians to assess the risk of AKI in hospitalized patients more accurately and intuitively and to develop necessary AKI management strategies.

Construction of machine learning-based models for screening the high-risk patients with gastric precancerous lesions

BACKGROUND

The individualized prediction and discrimination of precancerous lesions of gastric cancer (PLGC) is critical for the early prevention of gastric cancer (GC). However, accurate non-invasive methods for distinguishing between PLGC and GC are currently lacking. This study therefore aimed to develop a risk prediction model by machine learning and deep learning techniques to aid the early diagnosis of GC.

METHODS

In this study, a total of 2229 subjects were recruited from nine tertiary hospitals between October 2022 and November 2023. We designed a comprehensive questionnaire, identified statistically significant factors, and created a web-based column chart. Then, a risk prediction model was subsequently developed by machine learning techniques. In addition, a tongue image-based risk prediction model was established by deep learning algorithms.

RESULTS

Based on logistic regression analysis, a dynamic web-based nomogram was developed and it is freely accessible at: https://yz6677.shinyapps.io/GC67/ . Then, the prediction model was established using ten different machine learning algorithms and the Random Forest (RF) model achieved the highest accuracy at 85.65%. According with the predictive results, the top 10 key risk factors were age, traditional Chinese medicine (TCM) constitution type, tongue coating color, tongue color, irregular meals, pickled food, greasy fur, over-hot eating habit, anxiety and sleep onset latency. These factors are all significant risk indicators for the progression of PLGC patients to GC patients. Subsequently, the Swin Transformer architecture was used to develop a tongue image-based model for predicting the risk for progression of PLGC. The verification set showed an accuracy of 73.33% and an area under curve (AUC) greater than 0.8 across all models.

CONCLUSIONS

Our study developed machine learning and deep learning-based models for predicting the risk for progression of PLGC to GC, which will offer the assistance to determine the high-risk patients from PLGC and improve the early diagnosis of GC.

A Comprehensive Review of Thyroid Eye Disease Pathogenesis: From Immune Dysregulations to Novel Diagnostic and Therapeutic Approaches

Thyroid eye disease is a complex inflammatory disorder of the orbit that has gained tremendous interest over the past years, and numerous scientific efforts have been deployed to elucidate its pathophysiology for novel drug development. Our manuscript will delve into the molecular dysregulations involved in the pathogenesis of thyroid eye disease that led to its clinical manifestations. Abnormalities within the apoptotic pathway, inflammatory cascade, and autoimmune regulatory systems will be covered. We will further discuss the challenges involved in its diagnosis and management and provide a summary of the current diagnostic tools (i.e., molecular biomarkers, diagnostic scores) from the perspective of clinicians. Finally, our comprehensive literature review will provide a thorough summary of most recent preclinical and clinical studies around the topic of thyroid eye disease, with an emphasis on the manuscripts published within the last five years. We believe our manuscript will bring novelty within the field by bridging the fundamental sciences with the clinical aspect of this disease. This review will be a great tool for clinicians in better understanding the pathogenesis of thyroid eye disease while providing an outlook on future perspectives (i.e., liquid biopsies, artificial intelligence).

Prediction of Post-Treatment Visual Acuity in Age-Related Macular Degeneration Patients With an Interpretable Machine Learning Method

Purpose

We evaluated the features predicting visual acuity (VA) after one year in neovascular age-related macular degeneration (nAMD) patients.

Methods

A total of 527 eyes of 506 patients were included. Machine learning (ML) models were trained to predict VA deterioration beyond a logarithm of the minimum angle of resolution of 1.0 after 1 year based on the sequential addition of multimodal data. BaseM models used clinical data (age, sex, treatment regimen, and VA), SegM models included fluid volumes from optical coherence tomography (OCT) images, and RawM models used probabilities of visual deterioration (hereafter probability) from deep learning classifiers trained on baseline OCT (OCT0) and OCT after three loading doses (OCT3), fluorescein angiography, and indocyanine green angiography. We applied SHapley Additive exPlanations (SHAP) for machine learning model interpretation.

Results

The RawM model based on the probability of OCT0 outperformed the SegM model (area under the receiver operating characteristic curve of 0.95 vs. 0.91). Adding probabilities from OCT3, fluorescein angiography, and indocyanine green angiography to RawM showed minimal performance improvement, highlighting the practicality of using raw OCT0 data for predicting visual outcomes. Applied SHapley Additive exPlanations analysis identified VA after 3 months and OCT3 probability values as the most influential features over quantified fluid segments.

Conclusions

Integrating multimodal data to create a visual predictive model yielded accurate, interpretable predictions. This approach allowed the identification of crucial factors for predicting VA in patients with nAMD.

Translational Relevance

Interpreting a predictive model for 1-year VA in patients with nAMD from multimodal data allowed us to identify crucial factors for predicting VA.

Prediction models of intravenous glucocorticoids therapy response in thyroid eye disease

Background

Thyroid eye disease (TED) is an autoimmune orbital disease, with intravenous glucocorticoid (IVGC) therapy as the first-line treatment. Due to uncertain response rates and possible side effects, various prediction models have been developed to predict IVGC therapy outcomes.

Methods

A thorough search was conducted in PubMed, Embase, and Web of Science databases. Data extraction included publication details, prediction model content, and performance. Statistical analysis was performed using R software, including heterogeneity evaluation, publication bias, subgroup analysis, and sensitivity analysis. Forest plots were utilized for result visualization.

Results

Of the 12 eligible studies, 47 prediction models were extracted. All included studies exhibited a low-to-moderate risk of bias. The pooled area under the receiver operating characteristic curve (AUC) and the combined sensitivity and specificity for the models were 0.81, 0.75, and 0.79, respectively. In view of heterogeneity, multiple meta-regression and subgroup analysis were conducted, which showed that marker and modeling types may be the possible causes of heterogeneity (P < 0.001). Notably, imaging metrics alone (AUC = 0.81) or clinical characteristics combined with other markers (AUC = 0.87), incorporating with multivariate regression (AUC = 0.84) or radiomics analysis (AUC = 0.91), yielded robust and reliable prediction outcomes.

Conclusion

This meta-analysis comprehensively reviews the predictive models for IVGC therapy response in TED. It underscores that integrating clinical characteristics with laboratory or imaging indicators and employing advanced techniques like multivariate regression or radiomics analysis significantly enhance the efficacy of prediction. Our research findings offer valuable insights that can guide future studies on prediction models for IVGC therapy in TED.

Advances in artificial intelligence in thyroid-associated ophthalmopathy

Thyroid-associated ophthalmopathy (TAO), also referred to as Graves' ophthalmopathy, is a medical condition wherein ocular complications arise due to autoimmune thyroid illness. The diagnosis of TAO, reliant on imaging, typical ocular symptoms, and abnormalities in thyroid function or thyroid-associated antibodies, is generally graded and staged. In recent years, Artificial intelligence(AI), particularly deep learning(DL) technology, has gained widespread use in the diagnosis and treatment of ophthalmic diseases. This paper presents a discussion on specific studies involving AI, specifically DL, in the context of TAO, highlighting their applications in TAO diagnosis, staging, grading, and treatment decisions. Additionally, it addresses certain limitations in AI research on TAO and potential future directions for the field.

Thyroid stimulating immunoglobulin concentration is associated with disease activity and predicts response to treatment with intravenous methylprednisolone in patients with Graves' orbitopathy

Background

Thyroid stimulating immunoglobulins (TSI) play a central role in the pathogenesis of Graves' orbitopathy (GO), while soluble interleukin-2 receptor (sIL-2R) is a marker for T-cell activity. We investigated TSI and sIL-2R levels in relation to thyroid function, disease activity and severity and response to treatment with intravenous methylprednisolone (IVMP) in patients with GO.

Methods

TSI (bridge-based TSI binding assay), sIL-2R, TSH and fT4 levels were measured in biobank serum samples from 111 GO patients (37 male, 74 female; mean age 49.2 years old) and 25 healthy controls (5 male, 20 female; mean age 39.8 years old). Clinical characteristics and response to treatment were retrospectively retrieved from patient files.

Results

Higher sIL-2R levels were observed in GO patients compared to controls (p < 0.001). sIL-2R correlated with fT4 (r = 0.26), TSH (r = -0.40) and TSI (r = 0.21). TSI and sIL-2R concentrations were higher in patients with active compared to inactive GO (p < 0.001 and p < 0.05, respectively). Both TSI and sIL-2R correlated with total clinical activity score (CAS; r = 0.33 and r = 0.28, respectively) and with several individual CAS items. Cut-off levels for predicting active GO were 2.62 IU/L for TSI (AUC = 0.71, sensitivity 69%, specificity 69%) and 428 IU/mL for sIL-2R (AUC = 0.64, sensitivity 62%, specificity 62%). In multivariate testing higher TSI (p < 0.01), higher age (p < 0.001) and longer disease duration (p < 0.01) were associated with disease activity. TSI levels were higher in patients with a poor IVMP response (p = 0.048), while sIL-2R levels did not differ between responders and non-responders. TSI cut-off for predicting IVMP response was 19.4 IU/L (AUC = 0.69, sensitivity 50%, specificity 91%). In multivariate analysis TSI was the only independent predictor of response to IVMP (p < 0.05).

Conclusions

High TSI levels are associated with active disease (cut-off 2.62 IU/L) and predict poor response to IVMP treatment (cut-off 19.4 IU/L) in GO. While sIL-2R correlates with disease activity, it is also related to thyroid function, making it less useful as an additional biomarker in GO.

Machine learning application in autoimmune diseases: State of art and future prospectives

Autoimmune diseases are a group of disorders resulting from an alteration of immune tolerance, characterized by the formation of autoantibodies and the consequent development of heterogeneous clinical manifestations. Diagnosing autoimmune diseases is often complicated, and the available prognostic tools are limited. Machine learning allows us to analyze large amounts of data and carry out complex calculations quickly and with minimal effort. In this work, we examine the literature focusing on the use of machine learning in the field of the main systemic (systemic lupus erythematosus and rheumatoid arthritis) and organ-specific autoimmune diseases (type 1 diabetes mellitus, autoimmune thyroid, gastrointestinal, and skin diseases). From our analysis, interesting applications of machine learning emerged for developing algorithms useful in the early diagnosis of disease or prognostic models (risk of complications, therapeutic response). Subsequent studies and the creation of increasingly rich databases to be supplied to the algorithms will eventually guide the clinician in the diagnosis, allowing intervention when the pathology is still in an early stage and immediately directing towards a correct therapeutic approach.

Exploring factors affecting patient satisfaction in online healthcare: A machine learning approach grounded in empathy theory

Objective

Empathy between doctors and patients is crucial in enhancing patient satisfaction with medical consultations. This study, grounded in empathy theory, employs natural language processing and machine learning algorithms to explore the factors influencing patient satisfaction in online healthcare services, particularly the impact of doctor-patient empathy.

Methods

Utilizing the three dimensions of the Jefferson Scale of Physician Empathy, seven variables were extracted from patient-doctor dialogs as independent variables, with patient satisfaction as the dependent variable. Employing machine learning algorithms, a classification model was constructed to identify the best-fitting model for exploring the pivotal factors influencing patient satisfaction in online medical services. The optimal model was then chosen to investigate the essential factors impacting patients' satisfaction with online healthcare.

Results

A total of 7586 data points were collected, with 5447 consultation dialogs (71.8%) receiving a satisfactory rating from patients. LightGBM emerged as the best-performing model, achieving an F1 score of 0.78 and an area under the curve value of 0.81. Factors within the Standing in Patient's Shoes and Perspective Taking dimensions were identified as key determinants of patient satisfaction in online healthcare services.

Conclusion

This study broadens the conventional scope of applying empathy theory, signifying its crucial role in cultivating doctor-patient empathy within the realm of online healthcare and elevating the overall quality of medical services. The findings indicate that two pivotal factors influencing patients' satisfaction with online healthcare are doctors' perceived competence and ability to empathize, understanding patients' perspectives, and offering assistance.

Predictive markers for anti-inflammatory treatment response in thyroid eye disease

Anti-inflammatory treatment is the primary and vital therapeutic approach for active, moderate-to-severe thyroid eye disease (TED). Accurate pretreatment prediction of treatment response is of paramount importance for the prognosis of patients. However, relying solely on the clinical activity score asa determinant of activity has led to unsatisfactory treatment outcomes. In recent years, significant advancements have been made in identifying predictive markers for anti-inflammatory treatment response in TED, clinical markers, body fluid biomarkers and imaging biomarkers. Several clinical studies have developed prediction models based on these markers. However, there is still a lack of comprehensive elucidation or comparison between the different markers. Therefore, this review aims to provide a detailed analysis of the definition, characteristics, and application of predictive markers for anti-inflammatory treatment response in TED. Through detailed literature search, 26 articles applying anti-inflammatory treatment effect prediction with a total of 1948 TED patients were used for analysis and discussion. By gaining a better understanding of the current research on predictive markers, we can accelerate and guide the exploration of treatment prediction strategies, leading us towards an era of precise therapy for TED.

Statins in Graves Orbitopathy: A New Therapeutic Tool

PURPOSE

Graves orbitopathy (GO) is the most common extrathyroidal manifestation of Graves disease. Although its pathogenesis is not fully elucidated, GO is commonly considered an autoimmune disease due to loss of self-tolerance against autoantigens shared by thyroid epithelial cells and orbital fibroblasts. High-dose intravenous glucocorticoids (ivGCs) are the most used treatment for moderate-to-severe, active GO, but the addition of other immunomodulating treatments can improve the efficacy of ivGCs. Among the various risk factors that can affect the occurrence of GO, cholesterol may be worthy of interest. Since 2015 the role of cholesterol and cholesterol-lowering medications has been investigated. The purpose of this review is to discuss this topic, thereby offering new therapeutic opportunities for patients with GO.

METHODS

We searched PubMed for studies published between January 1, 1980 and June 1, 2023, using the search terms "Graves orbitopathy," "thyroid eye disease," "Graves ophthalmopathy," "thyroid ophthalmopathy," "thyroid-associated ophthalmopathy," "endocrine ophthalmopathy," "cholesterol," "lipids," "statins," "low-density lipoprotein," "atorvastatin," and "cholesterol-lowering drugs." Only English-language articles were included.

RESULTS

A correlation between low-density lipoprotein cholesterol and the risk of GO development has been reported. Furthermore, low-density lipoprotein cholesterol has been proposed as a risk factor that can affect the course of GO and the response to ivGCs. The protective role of cholesterol-lowering medications in preventing GO has been also investigated. Statin treatment was found to have potential benefits in reducing the risk of GO in patients with Graves disease. Given these findings, measurement of low-density lipoprotein cholesterol and treatment of hypercholesterolemia in patients with moderate-to-severe, active GO may be considered before starting ivGCs administration. Recently, a randomized clinical trial aimed at investigating the effects of statins in GO suggested that the addition of oral atorvastatin to ivGCs improves the overall outcome of moderate-to-severe, active GO in hypercholesterolemic patients given ivGCs.

CONCLUSIONS

Overall, statins seem to have a preventive and therapeutic role in moderate-to-severe active GO. Their efficacy can be related to cholesterol-lowering activity, pleiotropic actions, and interaction with methylprednisolone.