A classification model for predicting eye disease in newly diagnosed people with type 2 diabetes

Machine Learning-Based Diagnosis and Ranking of Risk Factors for Diabetic Retinopathy in Population-Based Studies from South India

This paper discusses the importance of investigating DR using machine learning and a computational method to rank DR risk factors by importance using different machine learning models. The dataset was collected from four large population-based studies conducted in India between 2001 and 2010 on the prevalence of DR and its risk factors. We deployed different machine learning models on the dataset to rank the importance of the variables (risk factors). The study uses a t-test and Shapely additive explanations (SHAP) to rank the risk factors. Then, it uses five machine learning models (K-Nearest Neighbor, Decision Tree, Support Vector Machines, Logistic Regression, and Naive Bayes) to identify the unimportant risk factors based on the area under the curve criterion to predict DR. To determine the overall significance of risk variables, a weighted average of each classifier's importance is used. The ranking of risk variables is provided to machine learning models. To construct a model for DR prediction, the combination of risk factors with the highest AUC is chosen. The results show that the risk factors glycosylated hemoglobin and systolic blood pressure were present in the top three risk factors for DR in all five machine learning models when the t-test was used for ranking. Furthermore, the risk factors, namely, systolic blood pressure and history of hypertension, were present in the top five risk factors for DR in all the machine learning models when SHAP was used for ranking. Finally, when an ensemble of the five machine learning models was employed, independently with both the t-test and SHAP, systolic blood pressure and diabetes mellitus duration were present in the top four risk factors for diabetic retinopathy. Decision Tree and K-Nearest Neighbor resulted in the highest AUCs of 0.79 (t-test) and 0.77 (SHAP). Moreover, K-Nearest Neighbor predicted DR with 82.6% (t-test) and 78.3% (SHAP) accuracy.

Dynamic nomogram prediction model for diabetic retinopathy in patients with type 2 diabetes mellitus

BACKGROUND

To develop a dynamic prediction model for diabetic retinopathy (DR) using systemic risk factors.

METHODS

This retrospective study included type 2 diabetes mellitus (T2DM) patients discharged from the Second Affiliated Hospital of Kunming Medical University between May 2020 and February 2022. The early patients (80%) were used for the training set and the late ones (20%) for the validation set.

RESULTS

Finally, 1257 patients (1049 [80%] in the training set and 208 [20%] in the validation set) were included; 360 (28.6%) of them had DR. The areas under the curves (AUCs) for the multivariate regression (MR), least absolute shrinkage and selection operator regression (LASSO), and backward elimination stepwise regression (BESR) models were 0.719, 0.727, and 0.728, respectively. The Delong test showed that the BESR model had a better predictive value than the MR (p = 0.04899) and LASSO (P = 0.04999) models. The DR nomogram risk model was established according to the BESR model, and it included disease duration, age at onset, treatment method, total cholesterol, urinary albumin to creatinine ratio (UACR), and urine sugar. The AUC, kappa coefficient, sensitivity, specificity, and compliance of the nomogram risk model in the validation set were 0.79, 0.48, 71.2%, 78.9%, and 76.4%, respectively.

CONCLUSIONS

A relatively reliable DR nomogram risk model was established based on the BESR model.

Classification of Gastroparesis from Glycemic Variability in Type 1 Diabetes: A Proof-of-Concept Study

BACKGROUND AND OBJECTIVE

Delayed gastric emptying is a substantial challenge for people with diabetes, affecting quality of life and blood glucose regulation. The complication is underdiagnosed, and current diagnostic tests are expensive or time consuming or have modest accuracy. The assessment of glycemic variations has potential use in gastroparesis screening. The aim of this study was to investigate the differences in glycemic variability between type 1 diabetes patients with gastroparesis and without a diagnosis of gastroparesis and the potential for using a classification model to differentiate between groups.

METHODS

Continuous glucose monitoring (CGM) from 425 patients with diabetes was included in the analytic cohort, including 16 patients with a diagnosis of gastroparesis and 409 without a known gastroparesis diagnosis. Sixteen features (9 daytime features and 7 nighttime features) describing glucose dynamics were extracted to assess differences between patients with and without a diagnosis of gastroparesis. A logistic regression model was trained using forward selection and cross-validation.

RESULTS

In total, 3 features were included in the model utilizing forward selection of features and cross-validation: mean absolute glucose (MAG), span, and standard deviation during the night. The Receiver operating characteristic (ROC) AUC for the classification model was 0.76.

CONCLUSIONS

Gastroparesis seems to have an impact on glucose variability, especially during the night. Moreover, CGM could possibly be used as a part of the screening process for delayed gastric emptying, but more studies are needed to determine a realistic accuracy.

Nomogram for Prediction of Diabetic Retinopathy Among Type 2 Diabetes Population in Xinjiang, China

PURPOSE

To establish an accurate risk prediction model of diabetic retinopathy (DR) using cost effective and easily available patients' characteristics and clinical biomarkers.

PATIENTS AND METHODS

Totally 18,904 cases diagnosed type 2 diabetes mellitus (T2DM) were collected, among which 13,980 cases were selected after quality screening. The least absolute shrinkage and selection operator (LASSO) regression models were used for univariate analysis and factors selection, and the multi-factor logistic regression analysis was used to establish the prediction model. Discrimination, calibration, and clinical usefulness of the prediction model were assessed using AUC/ Harrell's C statistic, calibration plot, and decision curve analysis. Both the development group and validation group were assessed.

RESULTS

Candidate variables were selected by Lasso regression and multivariate logistic regression analysis. Finally, the candidate predictive variables were included diabetic peripheral neuropathy (DPN), age, neutrophilic granulocyte (NE), high-density lipoprotein (HDL), hemoglobin A1c (HbA1C), duration of T2DM, and glycosylated serum protein (GSP) were used to establish a nomogram model for predicting the risk of DR. In the development group, the area under the receiver operating characteristic curve (AUC) was 0.882 (95% CI, 0.875-0.888). In the validation group, the AUC was 0.870 (95% CI, 0.856-0.881). Meanwhile, the optimism-corrected Harrell's C statistic were 0.878 and 0.867 in the development group and the validation group, respectively. Decision curve analysis demonstrated that the nomogram was clinically useful.

CONCLUSION

We constructed and verified nomograms that could accurately predict the risk of DR in T2DM patients, which could be used to predict the personalized risk of DR patients in Xinjiang, China.

Prognostic models of diabetic microvascular complications: a systematic review and meta-analysis

BACKGROUND

Many prognostic models of diabetic microvascular complications have been developed, but their performances still varies. Therefore, we conducted a systematic review and meta-analysis to summarise the performances of the existing models.

METHODS

Prognostic models of diabetic microvascular complications were retrieved from PubMed and Scopus up to 31 December 2020. Studies were selected, if they developed or internally/externally validated models of any microvascular complication in type 2 diabetes (T2D).

RESULTS

In total, 71 studies were eligible, of which 32, 30 and 18 studies initially developed prognostic model for diabetic retinopathy (DR), chronic kidney disease (CKD) and end stage renal disease (ESRD) with the number of derived equations of 84, 96 and 51, respectively. Most models were derived-phases, some were internal and external validations. Common predictors were age, sex, HbA1c, diabetic duration, SBP and BMI. Traditional statistical models (i.e. Cox and logit regression) were mostly applied, otherwise machine learning. In cohorts, the discriminative performance in derived-logit was pooled with C statistics of 0.82 (0.73‑0.92) for DR and 0.78 (0.74‑0.83) for CKD. Pooled Cox regression yielded 0.75 (0.74‑0.77), 0.78 (0.74‑0.82) and 0.87 (0.84‑0.89) for DR, CKD and ESRD, respectively. External validation performances were sufficiently pooled with 0.81 (0.78‑0.83), 0.75 (0.67‑0.84) and 0.87 (0.85‑0.88) for DR, CKD and ESRD, respectively.

CONCLUSIONS

Several prognostic models were developed, but less were externally validated. A few studies derived the models by using appropriate methods and were satisfactory reported. More external validations and impact analyses are required before applying these models in clinical practice.

SYSTEMATIC REVIEW REGISTRATION

PROSPERO CRD42018105287.

Detecting diabetic retinopathy through machine learning on electronic health record data from an urban, safety net healthcare system

Objective

Clinical guidelines recommend annual eye examinations to detect diabetic retinopathy (DR) in patients with diabetes. However, timely DR detection remains a problem in medically underserved and under-resourced settings in the United States. Machine learning that identifies patients with latent/undiagnosed DR could help to address this problem.

Materials and Methods

Using electronic health record data from 40 631 unique diabetic patients seen at Los Angeles County Department of Health Services healthcare facilities between January 1, 2015 and December 31, 2017, we compared ten machine learning environments, including five classifier models, for assessing the presence or absence of DR. We also used data from a distinct set of 9300 diabetic patients seen between January 1, 2018 and December 31, 2018 as an external validation set.

Results

Following feature subset selection, the classifier with the best AUC on the external validation set was a deep neural network using majority class undersampling, with an AUC of 0.8, the sensitivity of 72.17%, and specificity of 74.2%.

Discussion

A deep neural network produced the best AUCs and sensitivity results on the test set and external validation set. Models are intended to be used to screen guideline noncompliant diabetic patients in an urban safety-net setting.

Conclusion

Machine learning on diabetic patients’ routinely collected clinical data could help clinicians in safety-net settings to identify and target unscreened diabetic patients who potentially have undiagnosed DR.

Cognitive impairment in elderly people with prediabetes or diabetes: A cross-sectional study of the NHANES population

AIM

To investigate the cognitive function in people without diabetes, with prediabetes and with diabetes.

METHODS/DESIGN

The study design used was a cross-sectional analysis of data in people above 60 years registered in NHANES from 2011 to 2014.Three assessments were used to characterize cognitive function: (a) CERAD Word Learning subtest assessing immediate and delayed learning ability, (b) The Animal Fluency test assesing categorical verbal fluency, and (c) The Digit Symbol Substitution test assessing processing speed, sustained attention, and working memory.

RESULTS

(A) Memory recall (-0.19, [-0.34; -0.039], p = 0.014) and Delayed memory recall decline was associated with diabetes (-0.285, [-0.503; -0.067], p = 0.01), but not in an adjusted analysis. (B) Animal Fluency score decline was associated with diabetes (-1.185, [-1.688; -0.682], p < 0.001). (C) Digit Symbol score decline was associated with diabetes (-6.897, [-8.491; -5.302], p < 0.001). Prediabetes was not associated with cognitive function.

CONCLUSIONS

This study demonstrates an association between cognitive dysfunction and diabetes. Results may also indicate that cognitive decline is not yet present in people with mild impairments of glucose homeostasis.

Personalized risk-based screening for diabetic retinopathy: A multivariate approach versus the use of stratification rules

AIMS

To evaluate our proposed multivariate approach to identify patients who will develop sight-threatening diabetic retinopathy (STDR) within a 1-year screen interval, and explore the impact of simple stratification rules on prediction.

MATERIALS AND METHODS

A 7-year dataset (2009-2016) from people with diabetes (PWD) was analysed using a novel multivariate longitudinal discriminant approach. Level of diabetic retinopathy, assessed from routine digital screening photographs of both eyes, was jointly modelled using clinical data collected over time. Simple stratification rules based on retinopathy level were also applied and compared with the multivariate discriminant approach.

RESULTS

Data from 13 103 PWD (49 520 screening episodes) were analysed. The multivariate approach accurately predicted whether patients developed STDR or not within 1 year from the time of prediction in 84.0% of patients (95% confidence interval [CI] 80.4-89.7), compared with 56.7% (95% CI 55.5-58.0) and 79.7% (95% CI 78.8-80.6) achieved by the two stratification rules. While the stratification rules detected up to 95.2% (95% CI 92.2-97.6) of the STDR cases (sensitivity) only 55.6% (95% CI 54.5-56.7) of patients who did not develop STDR were correctly identified (specificity), compared with 85.4% (95% CI 80.4-89.7%) and 84.0% (95% CI 80.7-87.6%), respectively, achieved by the multivariate risk model.

CONCLUSIONS

Accurate prediction of progression to STDR in PWD can be achieved using a multivariate risk model whilst also maintaining desirable specificity. While simple stratification rules can achieve good levels of sensitivity, the present study indicates that their lower specificity (high false-positive rate) would therefore necessitate a greater frequency of eye examinations.

High -density lipoprotein cholesterol as a predictor for diabetes mellitus

Background:

Diabetes is a prevalent chronic disease around the world. To evaluate the risk of diabetes comprehensively, we developed a score model for risk prediction with HDL-C as a protective factor.

Methods:

We extracted physical examination data of 2728 individuals. The data contain 18 demographic and clinical variables. To identify the statistical significant feature variables, the backward stepwise logistic regression was used based on the data of the “exploratory population”. To ascertain the cutoff value of the selected variables, we used the Youden index. Then we assigned each variable level a score according to the estimated regression model coefficients and then calculated the individual’s total score. We gained the cutoff value for the total score through the Youden Index and stratified the total score into four levels. We employed the data of “validation population” to test the performance of the score model based on the area under the ROC curve.

Results:

Age, LDL-C, HDL-C, BMI, family history of diabetes, diastolic blood pressure and TCHO were selected as statistically significant variables. The diabetes risk score range varied from 0 to 17. The risk level categorized by the total score was low, middle, high and extremely high, with a score range of 0-2, 3-7, 8-12 and 13-17, respectively.

Conclusions:

The score model based on physical examination data is an efficient and valuable tool to evaluate and monitor the potential diabetes risk for both healthy and unhealthy people at an individual level.

Clinical worthlessness of genetic prediction of common forms of diabetes mellitus and related chronic complications: A position statement of the Italian Society of Diabetology

AIM

We are currently facing several attempts aimed at marketing genetic data for predicting multifactorial diseases, among which diabetes mellitus is one of the more prevalent. The present document primarily aims at providing to practicing physicians a summary of available data regarding the role of genetic information in predicting diabetes and its chronic complications.

DATA SYNTHESIS

Firstly, general information about characteristics and performance of risk prediction tools will be presented in order to help clinicians to get acquainted with basic methodological information related to the subject at issue. Then, as far as type 1 diabetes is concerned, available data indicate that genetic information and counseling may be useful only in families with many affected individuals. However, since no disease prevention is possible, the utility of predicting this form of diabetes is at question. In the case of type 2 diabetes, available data really question the utility of adding genetic information on top of well performing, easy available and inexpensive non-genetic markers. Finally, the possibility of using the few available genetic data on diabetic complications for improving our ability to predict them will also be presented and discussed. For cardiovascular complication, the addition of genetic information to models based on clinical features does not translate in a substantial improvement in risk discrimination. For all other diabetic complications genetic information are currently very poor and cannot, therefore, be used for improving risk stratification.

CONCLUSIONS

In all, nowadays the use of genetic testing for predicting diabetes and its chronic complications is definitively of little value in clinical practice.

Toward Big Data Analytics: Review of Predictive Models in Management of Diabetes and Its Complications

Diabetes is one of the top priorities in medical science and health care management, and an abundance of data and information is available on these patients. Whether data stem from statistical models or complex pattern recognition models, they may be fused into predictive models that combine patient information and prognostic outcome results. Such knowledge could be used in clinical decision support, disease surveillance, and public health management to improve patient care. Our aim was to review the literature and give an introduction to predictive models in screening for and the management of prevalent short- and long-term complications in diabetes. Predictive models have been developed for management of diabetes and its complications, and the number of publications on such models has been growing over the past decade. Often multiple logistic or a similar linear regression is used for prediction model development, possibly owing to its transparent functionality. Ultimately, for prediction models to prove useful, they must demonstrate impact, namely, their use must generate better patient outcomes. Although extensive effort has been put in to building these predictive models, there is a remarkable scarcity of impact studies.