Predicting diabetic kidney disease for type 2 diabetes mellitus by machine learning in the real world: a multicenter retrospective study

Implementing novel complete blood count-derived inflammatory indices in the diabetic kidney diseases diagnostic models

Objectives

Hemogram inflammatory markers, including neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), red-cell distribution width (RDW), and mean platelet volume (MPV) have been associated with type 2 diabetes mellitus (T2DM) and its complications, namely diabetic kidney diseases (DKD). We aimed to develop and validate logistic regression (LR) and CatBoost diagnostic models and study the role of adding these markers to the models.

Methods

All individuals who were managed in our secondary care center from March 2020 to December 2023 were identified. After excluding the ineligible patients, train-test splitting, and data preprocessing, two baseline LR and CatBoost-based models were developed using demographic, clinical, and laboratory features. The AUC-ROC of the models with biomarkers (NLR, PLR, RDW, and MPV) was compared to the baseline models. We calculated net reclassification improvement (NRI) and integrated discrimination index (IDI).

Results

One thousand and eleven T2DM patients were eligible. The AUC-ROC of both LR (0.738) and CatBoost (0.715) models was comparable. Adding target inflammatory markers did not significantly change the AUC-ROC in both LR and CatBoost models. Adding RDW to the baseline LR model reclassified 41.7% of patients without DKD, in the cost of misclassification of 38.4% of DKD cases. This change was absent in CatBoost models, and other markers did not achieve improved NRI or IDI.

Conclusion

The basic models with demographical and clinical features had acceptable performance. Adding RDW to the basic LR model improved the reclassification of the non-DKD participants. However, adding other hematological indices did not significantly improve the LR and CatBoost models' performance.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40200-024-01523-2.

Evaluating Feature Selection Methods for Accurate Diagnosis of Diabetic Kidney Disease

Background/Objectives: The increase in patients with type 2 diabetes, coupled with the development of complications caused by the same disease is an alarming aspect for the health sector. One of the main complications of diabetes is nephropathy, which is also the main cause of kidney failure. Once diagnosed, in Mexican patients the kidney damage is already highly compromised, which is why acting preventively is extremely important. The aim of this research is to compare distinct methodologies of feature selection to identify discriminant risk factors that may be beneficial for early treatment, and prevention. Methods: This study focused on evaluating a Mexican dataset collected from 22 patients containing 32 attributes. To reduce the dimensionality and choose the most important variables, four feature selection algorithms: Univariate, Boruta, Galgo, and Elastic net were implemented. After selecting suitable features detected by the methodologies, they are included in the random forest classifier, obtaining four models. Results: Galgo with Random Forest achieved the best performance with only three predictors, "creatinine", "urea", and "lipids treatment". The model displayed a moderate classification performance with an area under the curve of 0.80 (±0.3535 SD), a sensitivity of 0.909, and specificity of 0.818. Conclusions: It is demonstrated that the proposed methodology has the potential to facilitate the prompt identification of nephropathy and non-nephropathy patients, and thereby could be used in the clinical area as a preliminary computer-aided diagnosis tool.

Risk of Microvascular Complications in Newly Diagnosed Type 2 Diabetes Patients Using Automated Machine Learning Prediction Models

Background/Objectives: In type 2 diabetes (T2D), collective damage to the eyes, kidneys, and peripheral nerves constitutes microvascular complications, which significantly affect patients' quality of life. This study aimed to prospectively evaluate the risk of microvascular complications in newly diagnosed T2D patients in Dubai, UAE. Methods: Supervised automated machine learning in the Auto-Classifier model of the IBM SPSS Modeler package was used to predict microvascular complications in a training data set of 348 long-term T2D patients with complications using 24 independent variables as predictors and complications as targets. Three automated model scenarios were tested: Full All-Variable Model; Univariate-Selected Model, and Backward Stepwise Logistic Regression Model. An independent cohort of 338 newly diagnosed T2D patients with no complications was used for the model validation. Results: Long-term T2D patients with complications (duration = ~14.5 years) were significantly older (mean age = 56.3 ± 10.9 years) than the newly diagnosed patients without complications (duration = ~2.5 years; mean age = 48.9 ± 9.6 years). The Bayesian Network was the most reliable algorithm for predicting microvascular complications in all three scenarios with an area under the curve (AUC) of 77-87%, accuracy of 68-75%, sensitivity of 86-95%, and specificity of 53-75%. Among newly diagnosed T2D patients, 22.5% were predicted positive and 49.1% negative across all models. Logistic regression applied to the 16 significant predictors between the two sub-groups showed that BMI, HDL, adjusted for age at diagnosis of T2D, age at visit, and urine albumin explained >90% of the variation in microvascular measures. Conclusions: the Bayesian Network model effectively predicts microvascular complications in newly diagnosed T2D patients, highlighting the significant roles of BMI, HDL, age at diagnosis, age at visit, and urine albumin.

Unveiling the utility of artificial intelligence for prediction, diagnosis, and progression of diabetic kidney disease: an evidence-based systematic review and meta-analysis

OBJECTIVE

The purpose of this study was to conduct a systematic investigation of the potential of artificial intelligence (AI) models in the prediction, detection of diagnostic biomarkers, and progression of diabetic kidney disease (DKD). In addition, we compared the performance of non-logistic regression (LR) machine learning (ML) models to conventional LR prediction models.

METHODS

Until January 30, 2024, a comprehensive literature review was conducted by investigating databases such as Medline (via PubMed) and Cochrane. Research that is inclusive of AI or ML models for the prediction, diagnosis, and progression of DKD was incorporated. The area under the Receiver Operating Characteristic Curve (AUROC) served as the principal outcome metric for assessing model performance. A meta-analysis was performed utilizing MedCalc statistical software to calculate pooled AUROC and assess the performance differences between LR and non-LR models.

RESULTS

A total of 57 studies were included in the meta-analysis. The pooled AUROC of AI or ML model was 0.84 (95% CI = 0.81-0.86, p < 0.0001) for analyzing prediction of DKD, 0.88 (95%CI = 0.84-0.92, p < 0.0001) for detecting diagnostic biomarkers, and 0.80 (95% CI = 0.77-0.82, p < 0.0001) for analyzing progression of DKD. The pooled AUROC of LR and non-LR ML models exhibited no significant differences across all categories (p > 0.05), except for the random forest (RF) model, which displayed a statistically significant increase in predictive accuracy compared to LR for DKD occurrence (p < 0.04).

CONCLUSION

ML models showed solid DKD prediction effectiveness, with pooled AUROC values over 0.8, suggesting good performance. These data demonstrated that non-LR and LR models perform similarly in overall CKD management, but the RF model outperforms the LR model, particularly in predicting the occurrence of DKD. These findings highlight the promise of AI technologies for better DKD management. To improve model reliability, future study should include extended follow-up periods as well as external validation.

Thyroid FT4-to-TSH ratio in the first trimester is associated with gestational diabetes mellitus in women carrying male fetus: a prospective bi-center cohort study

Background

Gestational diabetes mellitus (GDM) is one of the most common medical complications of pregnancy, which increases the risk of other pregnant complications and adverse perinatal outcomes. Thyroid dysfunction is closely with the risk of diabetes mellitus. However, the relationship between euthyroid function in early pregnancy and GDM is still controversial.

Aims

This study was to find the relationship between thyroid function within normal range during early pregnancy as well as glucose and lipids metabolisms as well as the risk of subsequent GDM.

Methods

A total of 1486 pregnant women were included in this prospective double-center cohort study. Free thyroxine (FT4), thyroid stimulating hormone (TSH) and antithyroid peroxidase antibodies (TPOAb) were tested during 6-12 weeks of gestation and oral glucose tolerance test (OGTT) was conducted during 24-28 weeks to screen GDM. Relative risks (RR) with 95% confidence intervals (CI) for subsequent risk of GDM by thyroid function quartiles were assessed adjusting for major risk factors.

Results

The incidence of GDM was 23.0% (342/1486). TSH, FT4 and the percentage of positive TPOAb were no significant difference between women with and without GDM, but FT4/TSH ratio was significantly higher in GDM group compared with NGT group [6.97(0.84,10.61) vs. 4.88(0.66,12.44), P=0.025)]. The linear trends of TC, TG, HDL-C, LDL-C, fasting glucose in the first trimester, insulin, C-peptide, HOMA-IR, fasting glucose during OGTT and incidence of GDM according to FT4/TSH ratio were all statistically significant. Further analysis based on fetal sex presented only the third quartile of FT4/TSH ratio in women carrying male fetus was associated with higher incidence of GDM statistically significant [RR (95% CI), 1.917 (1.143,3.216)], rather than in women carrying female fetus.

Conclusions

Thyroid function even in normal range is closely related to glucose and lipids metabolisms during the first trimester. Unappropriated FT4/TSH ratio in the first trimester is an independent risk factor of GDM in women carrying male fetus.

Machine learning for predicting in-hospital mortality in elderly patients with heart failure combined with hypertension: a multicenter retrospective study

BACKGROUND

Heart failure combined with hypertension is a major contributor for elderly patients (≥ 65 years) to in-hospital mortality. However, there are very few models to predict in-hospital mortality in such elderly patients. We aimed to develop and test an individualized machine learning model to assess risk factors and predict in-hospital mortality in in these patients.

METHODS

From January 2012 to December 2021, this study collected data on elderly patients with heart failure and hypertension from the Chongqing Medical University Medical Data Platform. Least absolute shrinkage and the selection operator was used for recognizing key clinical variables. The optimal predictive model was chosen among eight machine learning algorithms on the basis of area under curve. SHapley Additive exPlanations and Local Interpretable Model-agnostic Explanations was employed to interpret the outcome of the predictive model.

RESULTS

This study ultimately comprised 4647 elderly individuals with hypertension and heart failure. The Random Forest model was chosen with the highest area under curve for 0.850 (95% CI 0.789-0.897), high accuracy for 0.738, recall 0.837, specificity 0.734 and brier score 0.178. According to SHapley Additive exPlanations results, the most related factors for in-hospital mortality in elderly patients with heart failure and hypertension were urea, length of stay, neutrophils, albumin and high-density lipoprotein cholesterol.

CONCLUSIONS

This study developed eight machine learning models to predict in-hospital mortality in elderly patients with hypertension as well as heart failure. Compared to other algorithms, the Random Forest model performed significantly better. Our study successfully predicted in-hospital mortality and identified the factors most associated with in-hospital mortality.

Risk prediction models for diabetic nephropathy among type 2 diabetes patients in China: a systematic review and meta-analysis

Objective

This study systematically reviews and meta-analyzes existing risk prediction models for diabetic kidney disease (DKD) among patients with type 2 diabetes, aiming to provide references for scholars in China to develop higher-quality risk prediction models.

Methods

We searched databases including China National Knowledge Infrastructure (CNKI), Wanfang Data, VIP Chinese Science and Technology Journal Database, Chinese Biomedical Literature Database (CBM), PubMed, Web of Science, Embase, and the Cochrane Library for studies on the construction of DKD risk prediction models among type 2 diabetes patients, up until 28 December 2023. Two researchers independently screened the literature and extracted and evaluated information according to a data extraction form and bias risk assessment tool for prediction model studies. The area under the curve (AUC) values of the models were meta-analyzed using STATA 14.0 software.

Results

A total of 32 studies were included, with 31 performing internal validation and 22 reporting calibration. The incidence rate of DKD among patients with type 2 diabetes ranged from 6.0% to 62.3%. The AUC ranged from 0.713 to 0.949, indicating the prediction models have fair to excellent prediction accuracy. The overall applicability of the included studies was good; however, there was a high overall risk of bias, mainly due to the retrospective nature of most studies, unreasonable sample sizes, and studies conducted in a single center. Meta-analysis of the models yielded a combined AUC of 0.810 (95% CI: 0.780-0.840), indicating good predictive performance.

Conclusion

Research on DKD risk prediction models for patients with type 2 diabetes in China is still in its initial stages, with a high overall risk of bias and a lack of clinical application. Future efforts could focus on constructing high-performance, easy-to-use prediction models based on interpretable machine learning methods and applying them in clinical settings.

Registration

This systematic review and meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement, a recognized guideline for such research.

Systematic review registration

https://www.crd.york.ac.uk/prospero/, identifier CRD42024498015.

Construction of diagnostic models for the progression of hepatocellular carcinoma using machine learning

Liver cancer is one of the most prevalent forms of cancer worldwide. A significant proportion of patients with hepatocellular carcinoma (HCC) are diagnosed at advanced stages, leading to unfavorable treatment outcomes. Generally, the development of HCC occurs in distinct stages. However, the diagnostic and intervention markers for each stage remain unclear. Therefore, there is an urgent need to explore precise grading methods for HCC. Machine learning has emerged as an effective technique for studying precise tumor diagnosis. In this research, we employed random forest and LightGBM machine learning algorithms for the first time to construct diagnostic models for HCC at various stages of progression. We categorized 118 samples from GSE114564 into three groups: normal liver, precancerous lesion (including chronic hepatitis, liver cirrhosis, dysplastic nodule), and HCC (including early stage HCC and advanced HCC). The LightGBM model exhibited outstanding performance (accuracy = 0.96, precision = 0.96, recall = 0.96, F1-score = 0.95). Similarly, the random forest model also demonstrated good performance (accuracy = 0.83, precision = 0.83, recall = 0.83, F1-score = 0.83). When the progression of HCC was categorized into the most refined six stages: normal liver, chronic hepatitis, liver cirrhosis, dysplastic nodule, early stage HCC, and advanced HCC, the diagnostic model still exhibited high efficacy. Among them, the LightGBM model exhibited good performance (accuracy = 0.71, precision = 0.71, recall = 0.71, F1-score = 0.72). Also, performance of the LightGBM model was superior to that of the random forest model. Overall, we have constructed a diagnostic model for the progression of HCC and identified potential diagnostic characteristic gene for the progression of HCC.

Causal relationships between blood metabolites and diabetic retinopathy: a two-sample Mendelian randomization study

Background

Diabetic retinopathy (DR) is a microvascular complication of diabetes, severely affecting patients' vision and even leading to blindness. The development of DR is influenced by metabolic disturbance and genetic factors, including gene polymorphisms. The research aimed to uncover the causal relationships between blood metabolites and DR.

Methods

The two-sample mendelian randomization (MR) analysis was employed to estimate the causality of blood metabolites on DR. The genetic variables for exposure were obtained from the genome-wide association study (GWAS) dataset of 486 blood metabolites, while the genetic predictors for outcomes including all-stage DR (All DR), non-proliferative DR (NPDR) and proliferative DR (PDR) were derived from the FinnGen database. The primary analysis employed inverse variance weighted (IVW) method, and supplementary analyses were performed using MR-Egger, weighted median (WM), simple mode and weighted mode methods. Additionally, MR-Egger intercept test, Cochran's Q test, and leave-one-out analysis were also conducted to guarantee the accuracy and robustness of the results. Subsequently, we replicated the MR analysis using three additional datasets from the FinnGen database and conducted a meta-analysis to determine blood metabolites associated with DR. Finally, reverse MR analysis and metabolic pathway analysis were performed.

Results

The study identified 13 blood metabolites associated with All DR, 9 blood metabolites associated with NPDR and 12 blood metabolites associated with PDR. In summary, a total of 21 blood metabolites were identified as having potential causal relationships with DR. Additionally, we identified 4 metabolic pathways that are related to DR.

Conclusion

The research revealed a number of blood metabolites and metabolic pathways that are causally associated with DR, which holds significant importance for screening and prevention of DR. However, it is noteworthy that these causal relationships should be validated in larger cohorts and experiments.

Easily applicable predictive score for MPR based on parameters before neoadjuvant chemoimmunotherapy in operable NSCLC: a single-center, ambispective, observational study

BACKGROUND

Neoadjuvant chemoimmunotherapy (NACI) is promising for resectable nonsmall cell lung cancer (NSCLC), but predictive biomarkers are still lacking. The authors aimed to develop a model based on pretreatment parameters to predict major pathological response (MPR) for such an approach.

METHODS

The authors enrolled operable NSCLC treated with NACI between March 2020 and May 2023 and then collected baseline clinical-pathology data and routine laboratory examinations before treatment. The efficacy and safety data of this cohort was reported and variables were screened by Logistic and Lasso regression and nomogram was developed. In addition, receiver operating characteristic curves, calibration curves, and decision curve analysis were used to assess its power. Finally, internal cross-validation and external validation was performed to assess the power of the model.

RESULTS

In total, 206 eligible patients were recruited in this study and 53.4% (110/206) patients achieved MPR. Using multivariate analysis, the predictive model was constructed by seven variables, prothrombin time (PT), neutrophil percentage (NEUT%), large platelet ratio (P-LCR), eosinophil percentage (EOS%), smoking, pathological type, and programmed death ligand-1 (PD-L1) expression finally. The model had good discrimination, with area under the receiver operating characteristic curve (AUC) of 0.775, 0.746, and 0.835 for all datasets, cross-validation, and external validation, respectively. The calibration curves showed good consistency, and decision curve analysis indicated its potential value in clinical practice.

CONCLUSION

This real world study revealed favorable efficacy in operable NSCLC treated with NACI. The proposed model based on multiple clinically accessible parameters could effectively predict MPR probability and could be a powerful tool in personalized medication.

External validation of a minimal-resource model to predict reduced estimated glomerular filtration rate in people with type 2 diabetes without diagnosis of chronic kidney disease in Mexico: a comparison between country-level and regional performance

Background

Patients with type 2 diabetes are at an increased risk of chronic kidney disease (CKD) hence it is recommended that they receive annual CKD screening. The huge burden of diabetes in Mexico and limited screening resource mean that CKD screening is underperformed. Consequently, patients often have a late diagnosis of CKD. A regional minimal-resource model to support risk-tailored CKD screening in patients with type 2 diabetes has been developed and globally validated. However, population heath and care services between countries within a region are expected to differ. The aim of this study was to evaluate the performance of the model within Mexico and compare this with the performance demonstrated within the Americas in the global validation.

Methods

We performed a retrospective observational study with data from primary care (Clinic Specialized in Diabetes Management in Mexico City), tertiary care (Instituto Nacional de Ciencias Médicas y Nutrición Salvador Zubirán) and the Mexican national survey of health and nutrition (ENSANUT-MC 2016). We applied the minimal-resource model across the datasets and evaluated model performance metrics, with the primary interest in the sensitivity and increase in the positive predictive value (PPV) compared to a screen-everyone approach.

Results

The model was evaluated on 2510 patients from Mexico (primary care: 1358, tertiary care: 735, ENSANUT-MC: 417). Across the Mexico data, the sensitivity was 0.730 (95% CI: 0.689 - 0.779) and the relative increase in PPV was 61.0% (95% CI: 52.1% - 70.8%). These were not statistically different to the regional performance metrics for the Americas (sensitivity: p=0.964; relative improvement: p=0.132), however considerable variability was observed across the data sources.

Conclusion

The minimal-resource model performs consistently in a representative Mexican population sample compared with the Americas regional performance. In primary care settings where screening is underperformed and access to laboratory testing is limited, the model can act as a risk-tailored CKD screening solution, directing screening resources to patients who are at highest risk.

Integrated machine learning and deep learning for predicting diabetic nephropathy model construction, validation, and interpretability

OBJECTIVE

To construct a risk prediction model for assisted diagnosis of Diabetic Nephropathy (DN) using machine learning algorithms, and to validate it internally and externally.

METHODS

Firstly, the data was cleaned and enhanced, and was divided into training and test sets according to the 7:3 ratio. Then, the metrics related to DN were filtered by difference analysis, Least Absolute Shrinkage and Selection Operator (LASSO), Recursive Feature Elimination (RFE), and Max-relevance and Min-redundancy (MRMR) algorithms. Ten machine learning models were constructed based on the key variables. The best model was filtered by Receiver Operating Characteristic (ROC), Precision-Recall (PR), Accuracy, Matthews Correlation Coefficient (MCC), and Kappa, and was internally and externally validated. Based on the best model, an online platform had been constructed.

RESULTS

15 key variables were selected, and among the 10 machine learning models, the Random Forest model achieved the best predictive performance. In the test set, the area under the ROC curve was 0.912, and in two external validation cohorts, the area under the ROC curve was 0.828 and 0.863, indicating excellent predictive and generalization abilities.

CONCLUSION

The model has a good predictive value and is expected to help in the early diagnosis and screening of clinical DN.