Machine Learning and Data Mining Methods in Diabetes Research

Digital diabetes: Perspectives for diabetes prevention, management and research

Digital medicine, digital research and artificial intelligence (AI) have the power to transform the field of diabetes with continuous and no-burden remote monitoring of patients' symptoms, physiological data, behaviours, and social and environmental contexts through the use of wearables, sensors and smartphone technologies. Moreover, data generated online and by digital technologies - which the authors suggest be grouped under the term 'digitosome' - constitute, through the quantity and variety of information they represent, a powerful potential for identifying new digital markers and patterns of risk that, ultimately, when combined with clinical data, can improve diabetes management and quality of life, and also prevent diabetes-related complications. Moving from a world in which patients are characterized by only a few recent measurements of fasting glucose levels and glycated haemoglobin to a world where patients, healthcare professionals and research scientists can consider various key parameters at thousands of time points simultaneously will profoundly change the way diabetes is prevented, managed and characterized in patients living with diabetes, as well as how it is scientifically researched. Indeed, the present review looks at how the digitization of diabetes can impact all fields of diabetes - its prevention, management, technology and research - and how it can complement, but not replace, what is usually done in traditional clinical settings. Such a profound shift is a genuine game changer that should be embraced by all, as it can provide solid research results transferable to patients, improve general health literacy, and provide tools to facilitate the everyday decision-making process by both healthcare professionals and patients living with diabetes.

Advancing Measurement of Diabetes at the Population Level

PURPOSE

The measurement and estimation of diabetes in populations guides resource allocation, health priorities, and can influence practice and future research. To provide a critical reflection on current diabetes surveillance, we provide in-depth discussion about how upstream determinants, prevalence, incidence, and downstream impacts of diabetes are measured in the USA, and the challenges in obtaining valid, accurate, and precise estimates.

FINDINGS

Current estimates of the burden of diabetes risk are obtained through national surveys, health systems data, registries, and administrative data. Several methodological nuances influence accurate estimates of the population-level burden of diabetes, including biases in selection and response rates, representation of population subgroups, accuracy of reporting of diabetes status, variation in biochemical testing, and definitions of diabetes used by investigators. Technological innovations and analytical approaches (e.g., data linkage to outcomes data like the National Death Index) may help address some, but not all, of these concerns, and additional methodological advances and validation are still needed. Current surveillance efforts are imperfect, but measures consistently collected and analyzed over several decades enable useful comparisons over time. In addition, we proposed that focused subsampling, use of technology, data linkages, and innovative sensitivity analyses can substantially advance population-level estimation.

Continuous Glucose Monitoring: Current Use in Diabetes Management and Possible Future Applications

The recent announcement of the production of new low-cost continuous glucose monitoring (CGM) sensors, the approval of marketed CGM sensors for making treatment decisions, and new reimbursement criteria have the potential to revolutionize CGM use. After briefly summarizing current CGM applications, we discuss how, in our opinion, these changes are expected to extend CGM utilization beyond diabetes patients, for example, to subjects with prediabetes or even healthy individuals. We also elaborate on how the integration of CGM data with other relevant information, for example, health records and other medical device/wearable sensor data, will contribute to creating a digital data ecosystem that will improve our understanding of the etiology and complications of diabetes and will facilitate the development of data analytics for personalized diabetes management and prevention.

Identifying people at risk of developing type 2 diabetes: A comparison of predictive analytics techniques and predictor variables

BACKGROUND

The present study aims to identify the patients at risk of type 2 diabetes (T2D). There is a body of literature that uses machine learning classification algorithms to predict development of T2D among patients. The current study compares the performance of these classification algorithms to identify patients who are at risk of developing T2D in short, medium and long terms. In addition, the list of predictor variables important for prediction for T2D progression is provided.

METHODS

This study uses 10,911 records generated in 36 clinics from the 15th of November 2008-15th of November 2016. Syntactic minority oversampling and random under sampling were used to create a balanced dataset. The performance of Neural Networks, Support Vector Machines, Decision Tress and Logistic Regression to identify patients developing T2D in short, medium and long terms was compared. The measures were Area Under Curve, Sensitivity, Specificity, Matthew correlation coefficient and Mean Calibration Error. Through importance analysis and information fusion techniques the predictors of developing T2D were identified for short, medium and long-term risk analysis.

RESULTS

The findings show that the performance of analytics techniques depends on both period and purpose of prediction whether the prediction is to identify people who will not develop T2D or to determine at risk patients. Oversampling as opposed to under sampling improved performance. 16 predictors and their importance to determine patients at risk of T2D in short, medium and long terms were identified.

CONCLUSIONS

This study provides guidelines for an automated system to prompt patients for screening. Several predictors are reportable by patients, others can be examined by physicians or ordered for further lab examination, which offers a potential reduction of the burden placed upon the clinical settings.

A Random Forest classifier-based approach in the detection of abnormalities in the retina

Classification of abnormalities from medical images using computer-based approaches is of growing interest in medical imaging. Timely detection of abnormalities due to diabetic retinopathy and age-related macular degeneration is required in order to prevent the prognosis of the disease. Computer-aided systems using machine learning are becoming interesting to ophthalmologists and researchers. We present here one such technique, the Random Forest classifier, to aid medical practitioners in accurate diagnosis of the diseases. A computer-aided diagnosis system is proposed for detecting retina abnormalities, which combines K means-based segmentation of the retina image, after due preprocessing, followed by machine learning techniques, using several low level and statistical features. Abnormalities in the retina that are classified are caused by age-related macular degeneration and diabetic retinopathy. Performance measures used in the analysis are accuracy, sensitivity, specificity, F-measure, and Mathew correlation coefficient. A comparison with another machine learning technique, the Naïve Bayes classifier shows that the classification achieved by Random Forest classifier is 93.58% and it outperforms Naïve Bayes classifier which yields an accuracy of 83.63%. Graphical abstract Random Forest classifier for abnormality detection in retina images.

Futuristic biosensors for cardiac health care: an artificial intelligence approach

Biosensor-based devices are pioneering in the modern biomedical applications and will be the future of cardiac health care. The coupling of artificial intelligence (AI) for cardiac monitoring-based biosensors for the point of care (POC) diagnostics is prominently reviewed here. This review deciphers the most significant machine-learning algorithms for the futuristic biosensors along with the internet of things, computational techniques and microchip-based essential cardiac biomarkers for real-time health monitoring and improving patient compliance. The present review also discusses the recently developed cardiac biosensors along with technical strategies involved in their mechanism of working and their applications in healthcare. Additionally, it provides a key for the ontogeny of an effective and supportive hierarchical protocol for clinical decision-making about personalized medicine through combinatory information analysis, and integrated multidisciplinary AI approaches.

Analysis and Study of Diabetes Follow-Up Data Using a Data-Mining-Based Approach in New Urban Area of Urumqi, Xinjiang, China, 2016-2017

The focus of this study is the use of machine learning methods that combine feature selection and imbalanced process (SMOTE algorithm) to classify and predict diabetes follow-up control satisfaction data. After the feature selection and unbalanced process, diabetes follow-up data of the New Urban Area of Urumqi, Xinjiang, was used as input variables of support vector machine (SVM), decision tree, and integrated learning model (Adaboost and Bagging) for modeling and prediction. The experimental results show that Adaboost algorithm produces better classification results. For the test set, the G-mean was 94.65%, the area under the ROC curve (AUC) was 0.9817, and the important variables in the classification process, fasting blood glucose, age, and BMI were given. The performance of the decision tree model in the test set is relatively lower than that of the support vector machine and the ensemble learning model. The prediction results of these classification models are sufficient. Compared with a single classifier, ensemble learning algorithms show different degrees of increase in classification accuracy. The Adaboost algorithm can be used for the prediction of diabetes follow-up and control satisfaction data.

Incorporating machine learning approaches to assess putative environmental risk factors for multiple sclerosis

BACKGROUND

Multiple sclerosis (MS) incidence has increased recently, particularly in women, suggesting a possible role of one or more environmental exposures in MS risk. The study objective was to determine if animal, dietary, recreational, or occupational exposures are associated with MS risk.

METHODS

Least absolute shrinkage and selection operator (LASSO) regression was used to identify a subset of exposures with potential relevance to disease in a large population-based (Kaiser Permanente Northern California [KPNC]) case-control study. Variables with non-zero coefficients were analyzed in matched conditional logistic regression analyses, adjusted for established environmental risk factors and socioeconomic status (if relevant in univariate screening),± genetic risk factors, in the KPNC cohort and, for purposes of replication, separately in the Swedish Epidemiological Investigation of MS cohort. These variables were also assessed in models stratified by HLA-DRB1*15:01 status since interactions between risk factors and that haplotype have been described.

RESULTS

There was a suggestive association of pesticide exposure with having MS among men, but only in those who were positive for HLA-DRB1*15:01 (OR pooled = 3.11, 95% CI 0.87, 11.16, p = 0.08).

CONCLUSIONS

While this finding requires confirmation, it is interesting given the association between pesticide exposure and other neurological diseases. The study also demonstrates the application of LASSO to identify environmental exposures with reduced multiple statistical testing penalty. Machine learning approaches may be useful for future investigations of concomitant MS risk or prognostic factors.

Different strategies for detection of HbA1c emphasizing on biosensors and point-of-care analyzers

Measurement of glycosylated hemoglobin (HbA1c) is a gold standard procedure for assessing long term glycemic control in individuals with diabetes mellitus as it gives the stable and reliable value of blood glucose levels for a period of 90-120 days. HbA1c is formed by the non-enzymatic glycation of terminal valine of hemoglobin. The analysis of HbA1c tends to be complicated because there are more than 300 different assay methods for measuring HbA1c which leads to variations in reported values from same samples. Therefore, standardization of detection methods is recommended. The review outlines the current research activities on developing assays including biosensors for the detection of HbA1c. The pros and cons of different techniques for measuring HbA1c are outlined. The performance of current point-of-care HbA1c analyzers available on the market are also compared and discussed. The future perspectives for HbA1c detection and diabetes management are proposed.

Artificial Intelligence for Diabetes Management and Decision Support: Literature Review

Background

Artificial intelligence methods in combination with the latest technologies, including medical devices, mobile computing, and sensor technologies, have the potential to enable the creation and delivery of better management services to deal with chronic diseases. One of the most lethal and prevalent chronic diseases is diabetes mellitus, which is characterized by dysfunction of glucose homeostasis.

Objective

The objective of this paper is to review recent efforts to use artificial intelligence techniques to assist in the management of diabetes, along with the associated challenges.

Methods

A review of the literature was conducted using PubMed and related bibliographic resources. Analyses of the literature from 2010 to 2018 yielded 1849 pertinent articles, of which we selected 141 for detailed review.

Results

We propose a functional taxonomy for diabetes management and artificial intelligence. Additionally, a detailed analysis of each subject category was performed using related key outcomes. This approach revealed that the experiments and studies reviewed yielded encouraging results.

Conclusions

We obtained evidence of an acceleration of research activity aimed at developing artificial intelligence-powered tools for prediction and prevention of complications associated with diabetes. Our results indicate that artificial intelligence methods are being progressively established as suitable for use in clinical daily practice, as well as for the self-management of diabetes. Consequently, these methods provide powerful tools for improving patients’ quality of life.

Vitamin D levels and parathyroid hormone variations of children living in a subtropical climate: a data mining study

BACKGROUND

Vitamin D and intact parathyroid hormone (iPTH) play a crucial role in calcium homeostasis and bone health of children. Serum level of 25-hydroxyvitamin D (25-OHD) is considered to be the most accurate marker for vitamin D status. However, there have only been a few studies, with limited number of subjects, investigating the relationship between 25-OHD and parathyroid hormone (PTH) in children. The aim of this study was to evaluate the seasonal 25-OHD levels and its associations with intact parathyroid hormone (iPTH) in Turkish children at all pediatric ages; and then to define a critical decision threshold level for 25-OHD deficiency in Turkish children.

METHODS

A retrospective record review of 90,042 children, was performed on serum 25-OHD and for 3525 iPTH levels. They were measured by mass spectrometry method and by electrochemiluminescence immunoassay simultaneously.

RESULTS

25-OHD levels showed a sinusoidal fluctuation througout the year; being significantly higher in summer and autumn (p < 0,01). 25-OHD levels decreased with respect to age. The significant inverse relationship that was found between iPTH and 25-OHD suggests that the inflection point of serum 25-OHD level for maximal suppression of PTH is at 30 ng/ml.

CONCLUSION

As the rate of vitamin D deficiency decreases in the early years due to vitamin D supplementation, the recommendation should be set due to a clinical threshold level of 30 ng/ml for 25-OHD based on PTH levels in children of our population.

Prediction of dyslipidemia using gene mutations, family history of diseases and anthropometric indicators in children and adolescents: The CASPIAN-III study

Dyslipidemia, the disorder of lipoprotein metabolism resulting in high lipid profile, is an important modifiable risk factor for coronary heart diseases. It is associated with more than four million worldwide deaths per year. Half of the children with dyslipidemia have hyperlipidemia during adulthood, and its prediction and screening are thus critical. We designed a new dyslipidemia diagnosis system. The sample size of 725 subjects (age 14.66 ± 2.61 years; 48% male; dyslipidemia prevalence of 42%) was selected by multistage random cluster sampling in Iran. Single nucleotide polymorphisms (rs1801177, rs708272, rs320, rs328, rs2066718, rs2230808, rs5880, rs5128, rs2893157, rs662799, and Apolipoprotein-E2/E3/E4), and anthropometric, life-style attributes, and family history of diseases were analyzed. A framework for classifying mixed-type data in imbalanced datasets was proposed. It included internal feature mapping and selection, re-sampling, optimized group method of data handling using convex and stochastic optimizations, a new cost function for imbalanced data and an internal validation. Its performance was assessed using hold-out and 4-foldcross-validation. Four other classifiers namely as supported vector machines, decision tree, and multilayer perceptron neural network and multiple logistic regression were also used. The average sensitivity, specificity, precision and accuracy of the proposed system were 93%, 94%, 94% and 92%, respectively in cross validation. It significantly outperformed the other classifiers and also showed excellent agreement and high correlation with the gold standard. A non-invasive economical version of the algorithm was also implemented suitable for low- and middle-income countries. It is thus a promising new tool for the prediction of dyslipidemia.

Digital health technology and diabetes management

Diabetes care is largely dependent on patient self-management and empowerment, given that patients with diabetes must make numerous daily decisions as to what to eat, when to exercise, and determine their insulin dose and timing if required. In addition, patients and providers are generating vast amounts of data from many sources, including electronic medical records, insulin pumps, sensors, glucometers, and other wearables, as well as evolving genomic, proteomic, metabolomics, and microbiomic data. Multiple digital tools and apps have been developed to assist patients to choose wisely, and to enhance their compliance by using motivational tools and incorporating incentives from social media and gaming techniques. Healthcare teams (HCTs) and health administrators benefit from digital developments that sift through the enormous amounts of patient-generated data. Data are acquired, integrated, analyzed, and presented in a self-explanatory manner, highlighting important trends and items that require attention. The use of decision support systems may propose data-driven actions that, for the most, require final approval by the patient or physician before execution and, once implemented, may improve patient outcomes. The digital diabetes clinic aims to incorporate all digital patient data and provide individually tailored virtual or face-to-face visits to those persons who need them most. Digital diabetes care has demonstrated only modest HbA1c reduction in multiple studies and borderline cost-effectiveness, although patient satisfaction appears to be increased. Better understanding of the barriers to digital diabetes care and identification of unmet needs may yield improved utilization of this evolving technology in a safe, effective, and cost-saving manner.

Quantitative approaches to energy and glucose homeostasis: machine learning and modelling for precision understanding and prediction

Obesity is a major global public health problem. Understanding how energy homeostasis is regulated, and can become dysregulated, is crucial for developing new treatments for obesity. Detailed recording of individual behaviour and new imaging modalities offer the prospect of medically relevant models of energy homeostasis that are both understandable and individually predictive. The profusion of data from these sources has led to an interest in applying machine learning techniques to gain insight from these large, relatively unstructured datasets. We review both physiological models and machine learning results across a diverse range of applications in energy homeostasis, and highlight how modelling and machine learning can work together to improve predictive ability. We collect quantitative details in a comprehensive mathematical supplement. We also discuss the prospects of forecasting homeostatic behaviour and stress the importance of characterizing stochasticity within and between individuals in order to provide practical, tailored forecasts and guidance to combat the spread of obesity.

Targeting weight loss interventions to reduce cardiovascular complications of type 2 diabetes: a machine learning-based post-hoc analysis of heterogeneous treatment effects in the Look AHEAD trial

BACKGROUND

The Action for Health in Diabetes (Look AHEAD) trial investigated whether long-term cardiovascular disease morbidity and mortality could be reduced through a weight loss intervention among people with type 2 diabetes. Despite finding no significant reduction in cardiovascular events on average, it is possible that some subpopulations might have derived benefit. In this post-hoc analysis, we test the hypothesis that the overall neutral average treatment effect in the trial masked important heterogeneous treatment effects (HTEs) from intensive weight loss interventions.

METHODS

We used causal forest modelling, which identifies HTEs, using a random half of the trial data (the training set). We applied Cox proportional hazards models to test the potential HTEs on the remaining half of the data (the testing set). The analysis was deemed exempt from review by the Columbia University Institutional Review Board, Protocol ID# AAAO3003.

FINDINGS

Between Aug 22, 2001, and April 30, 2004, 5145 patients with type 2 diabetes were enrolled in the Look AHEAD randomised controlled trial, of whom 4901 were included in the The National Institute of Diabetes and Digestive and Kidney Diseases Repository and included in our analyses: 2450 for model development and 2451 in the testing dataset. Baseline HbA1c and self-reported general health distinguished participants who differentially benefited from the intervention. Cox models for the primary composite cardiovascular outcome revealed a number needed to treat of 28·9 to prevent 1 event over 9·6 years among participants with HbA1c 6·8% or higher, or both HbA1c less than 6·8% and Short Form Health Survey (SF-36) general health score of 48 or more (2101 [86%] of 2451 participants in the testing dataset; 167 [16%] of 1046 primary outcome events for intervention vs 205 [19%] of 1055 for control, absolute risk reduction of 3·46%, 95% CI 0·21-6·73%, p=0·038) By contrast, participants with HbA1c less than 6·8% and baseline SF-36 general health score of less than 48 (350 [14%] of 2451 participants in the testing data; 27 [16%] of 171 primary outcome events for intervention vs 15 [8%] of 179 primary outcome events for control) had an absolute risk increase of the primary outcome of 7·41% (0·60 to 14·22, p=0·003).

INTERPRETATION

Look AHEAD participants with moderately or poorly controlled diabetes (HbA1c 6·8% or higher) and subjects with well controlled diabetes (HbA1c less than 6·8%) and good self-reported health (85% of the overall study population) averted cardiovascular events from a behavioural intervention aimed at weight loss. However, 15% of participants with well controlled diabetes and poor self-reported general health experienced negative effects that rendered the overall study outcome neutral. HbA1c and a short questionnaire on general health might identify people with type 2 diabetes likely to derive benefit from an intensive lifestyle intervention aimed at weight loss.

FUNDING

None.

Wearable Continuous Glucose Monitoring Sensors: A Revolution in Diabetes Treatment

Worldwide, the number of people affected by diabetes is rapidly increasing due to aging populations and sedentary lifestyles, with the prospect of exceeding 500 million cases in 2030, resulting in one of the most challenging socio-health emergencies of the third millennium. Daily management of diabetes by patients relies on the capability of correctly measuring glucose concentration levels in the blood by using suitable sensors. In recent years, glucose monitoring has been revolutionized by the development of Continuous Glucose Monitoring (CGM) sensors, wearable non/minimally-invasive devices that measure glucose concentration by exploiting different physical principles, e.g., glucose-oxidase, fluorescence, or skin dielectric properties, and provide real-time measurements every 1–5 min. CGM opened new challenges in different disciplines, e.g., medicine, physics, electronics, chemistry, ergonomics, data/signal processing, and software development to mention but a few. This paper first makes an overview of wearable CGM sensor technologies, covering both commercial devices and research prototypes. Then, the role of CGM in the actual evolution of decision support systems for diabetes therapy is discussed. Finally, the paper presents new possible horizons for wearable CGM sensor applications and perspectives in terms of big data analytics for personalized and proactive medicine.

Computer-aided classification of sickle cell retinopathy using quantitative features in optical coherence tomography angiography

As a new optical coherence tomography (OCT) imaging modality, there is no standardized quantitative interpretation of OCT angiography (OCTA) characteristics of sickle cell retinopathy (SCR). This study is to demonstrate computer-aided SCR classification using quantitative OCTA features, i.e., blood vessel tortuosity (BVT), blood vessel diameter (BVD), vessel perimeter index (VPI), foveal avascular zone (FAZ) area, FAZ contour irregularity, parafoveal avascular density (PAD). It was observed that combined features show improved classification performance, compared to single feature. Three classifiers, including support vector machine (SVM), k-nearest neighbor (KNN) algorithm, and discriminant analysis, were evaluated. Sensitivity, specificity, and accuracy were quantified to assess the performance of each classifier. For SCR vs. control classification, all three classifiers performed well with an average accuracy of 95% using the six quantitative OCTA features. For mild vs. severe stage retinopathy classification, SVM shows better (97% accuracy) performance, compared to KNN algorithm (95% accuracy) and discriminant analysis (88% accuracy).

Estimating Disease Onset Time by Modeling Lab Result Trajectories via Bayes Networks

The true onset time of a disease, particularly slow-onset diseases like Type 2 diabetes mellitus (T2DM), is rarely observable in electronic health records (EHRs). However, it is critical for analysis of time to events and for studying sequences of diseases. The aim of this study is to demonstrate a method for estimating the onset time of such diseases from intermittently observable laboratory results in the specific context of T2DM. A retrospective observational study design is used. A cohort of 5,874 non-diabetic patients from a large healthcare system in the Upper Midwest United States was constructed with a three-year follow-up period. The HbA1c level of each patient was collected from earliest and the latest follow-up. We modeled the patients' HbA1c trajectories through Bayesian networks to estimate the onset time of diabetes. Due to non-random censoring and interventions unobservable from EHR data (such as lifestyle changes), naïve modeling of HbA1c through linear regression or modeling time-to-event through proportional hazard model leads to a clinically infeasible model with no or limited ability to predict the onset time of diabetes. Our model is consistent with clinical knowledge and estimated the onset of diabetes with less than a six-month error for almost half the patients for whom the onset time could be clinically ascertained. To our knowledge, this is the first study of modeling long-term HbA1c progression in non-diabetic patients and estimating the onset time of diabetes.

An adaptive threshold determination method of feature screening for genomic selection

Background

Although the dimension of the entire genome can be extremely large, only a parsimonious set of influential SNPs are correlated with a particular complex trait and are important to the prediction of the trait. Efficiently and accurately selecting these influential SNPs from millions of candidates is in high demand, but poses challenges. We propose a backward elimination iterative distance correlation (BE-IDC) procedure to select the smallest subset of SNPs that guarantees sufficient prediction accuracy, while also solving the unclear threshold issue for traditional feature screening approaches.

Results

Verified through six simulations, the adaptive threshold estimated by the BE-IDC performed uniformly better than fixed threshold methods that have been used in the current literature. We also applied BE-IDC to an Arabidopsis thaliana genome-wide data. Out of 216,130 SNPs, BE-IDC selected four influential SNPs, and confirmed the same FRIGIDA gene that was reported by two other traditional methods.

Conclusions

BE-IDC accommodates both the prediction accuracy and the computational speed that are highly demanded in the genomic selection.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-017-1617-9) contains supplementary material, which is available to authorized users.