Machine Learning and Deep Learning Techniques Applied to Diabetes Research: A Bibliometric Analysis

BACKGROUND

The use of machine learning and deep learning techniques in the research on diabetes has garnered attention in recent times. Nonetheless, few studies offer a thorough picture of the knowledge generation landscape in this field. To address this, a bibliometric analysis of scientific articles published from 2000 to 2022 was conducted to discover global research trends and networks and to emphasize the most prominent countries, institutions, journals, articles, and key topics in this domain.

METHODS

The Scopus database was used to identify and retrieve high-quality scientific documents. The results were classified into categories of detection (covering diagnosis, screening, identification, segmentation, among others), prediction (prognosis, forecasting, estimation), and management (treatment, control, monitoring, education, telemedicine integration). Biblioshiny and RStudio were used to analyze the data.

RESULTS

A total of 1773 articles were collected and analyzed. The number of publications and citations increased substantially since 2012, with a notable increase in the last 3 years. Of the 3 categories considered, detection was the most dominant, followed by prediction and management. Around 53.2% of the total journals started disseminating articles on this subject in 2020. China, India, and the United States were the most productive countries. Although no evidence of outstanding leadership by specific authors was found, the University of California emerged as the most influential institution for the development of scientific production.

CONCLUSION

This is an evolving field that has experienced a rapid increase in productivity, especially over the last years with exponential growth. This trend is expected to continue in the coming years.

Design and Usability of an Avatar-Based Learning Program to Support Diabetes Education: Quality Improvement Study in Colombia

BACKGROUND

This quality improvement study, entitled Avatar-Based LEarning for Diabetes Optimal Control (ABLEDOC), explored the feasibility of delivering an educational program to people with diabetes in Colombia. The aim was to discover how this approach could be used to improve awareness and understanding of the condition, the effects of treatment, and strategies for effective management of blood-glucose control.

METHODS

Individuals with diabetes were recruited by Colombian endocrinologists to a human-centered study to codesign the educational program, using the Double Diamond model. Participants contributed to two phases. The first phase focused on gathering unmet educational needs and choice of curriculum. Three prototypes were developed as a result. During phase 2, a different group of participants engaged with the program for several weeks, before reporting back.

RESULTS

Thirty-six participants completed a Web survey during phase 1, and five were also interviewed by telephone. The majority (33 of 36; 91%) were receptive to the prospect of educational interventions and ranked the chosen topic of hypoglycemia highly. In phase 2, the three prototypes were tested by 17 participants, 10 of whom also gave feedback in focus groups. The response was overwhelmingly positive, with 16 of 17 (94%) stating they would use a program like this again. The 3D version was the most highly rated.

CONCLUSIONS

Immersive, avatar-based programs, delivered through smartphone, have the potential to deliver educational information that is trusted, engaging, and useful. Future work includes expansion of the curriculum, evaluation with a larger group, and exploration of the prospective role of artificial intelligence in personalizing this form of educational intervention.

Interval Safety Layer Coupled With an Impulsive MPC for Artificial Pancreas to Handle Intrapatient Variability

The aim of control strategies for artificial pancreas systems is to calculate the insulin doses required by a subject with type 1 diabetes to regulate blood glucose levels by reducing hyperglycemia and avoiding the induction of hypoglycemia. Several control formulations developed for this end involve a safety constraint given by the insulin on board (IOB) estimation. This constraint has the purpose of reducing hypoglycemic episodes caused by insulin stacking. However, intrapatient variability constantly changes the patient's response to insulin, and thus, an adaptive method is required to restrict the control action according to the current situation of the subject. In this work, the control action computed by an impulsive model predictive controller is modulated with a safety layer to satisfy an adaptive IOB constraint. This constraint is established with two main steps. First, upper and lower IOB bounds are generated with an interval model that accounts for parameter uncertainty, and thus, define the possible system responses. Second, the constraint is selected according to the current value of glycemia, an estimation of the plant-model mismatch, and their corresponding first and second time derivatives to anticipate the changes of both glucose levels and physiological variations. With this strategy satisfactory results were obtained in an adult cohort where random circadian variability and sensor noise were considered. A 92% time in normoglycemia was obtained, representing an increase of time in range compared to previous MPC strategies, and a reduction of time in hypoglycemia to 0% was achieved without dangerously increasing the time in hyperglycemia.

Clinical Factors Associated with High Glycemic Variability Defined by Coefficient of Variation in Patients with Type 2 Diabetes

Background

High glycemic Variability (HGV) has become a stronger predictor of hypoglycemia. However, clinical factors associate with HGV still are unknown.

Objective

To determine clinical variables that were associated with a coefficient of variation (CV) above 36% evaluated by continuous glucose monitoring (CGM) in a group of patients with diabetes mellitus.

Methods

A cohort of patients with type 2 diabetes (T2D) was evaluated. Demographic variables, HbA1c, glomerular filtration rate (GFR) and treatment regimen were assessed. A bivariate analysis was performed, to evaluate the association between the outcome variable (CV> 36%) and each of the independent variables. A multivariate model was constructed to evaluate associations after controlling for confounding variables.

Results

CGM data from 274 patients were analyzed. CV> 36% was present in 56 patients (20.4%). In the bivariate analysis, demographic and clinical variables were included, such as time since diagnosis, hypoglycemia history, A1c, GFR and treatment established. In the multivariate analysis, GFR <45 mL/min (OR 2.81; CI 1.27,6.23; p:0.01), A1c > 9% (OR 2.81; CI 1.05,7.51; p:0.04) and hypoglycemia history (OR 2.09; CI 1.02,4.32; p:0.04) were associated with HGV. Treatment with iDPP4 (OR 0.39; CI 0.19,0.82; p:0.01) and AGLP1 (OR 0.08; CI 0.01,0.68; p:0.02) was inversely associated with GV.

Conclusion

Clinical variables such as GFR <45 mL/min, HbA1C>9% and a history of hypoglycemia are associated with a high GV. Our data suggest that the use of technology and treatments able to reduce glycemic variability could be useful in this population to reduce the risk of hypoglycemia and to improve glycemic control.

Factors associated with clinically significant hypoglycemia in patients with type 1 diabetes using sensor-augmented pump therapy with predictive low-glucose management: A multicentric study on iberoamerica

BACKGROUND AND AIMS

Despite using sensor-augmented pump therapy (SAPT) with predictive low-glucose management (PLGM), hypoglycemia is still an issue in patients with type 1 Diabetes (T1D). Our aim was to determine factors associated with clinically significant hypoglycemia (<54 mg/dl) in persons with T1D treated with PLGM-SAPT.

METHOD

ology: This is a multicentric prospective real-life study performed in Colombia, Chile and Spain. Patients with T1D treated with PLGM-SAPT, using sensor ≥70% of time, were included. Data regarding pump and sensor use patterns and carbohydrate intake from 28 consecutive days were collected. A bivariate and multivariate Poisson regression analysis was carried out, to evaluate the association between the number of events of <54 mg/dl with the clinical variables and patterns of sensor and pump use.

RESULTS

188 subjects were included (41 ± 13.8 years-old, 23 ± 12 years disease duration, A1c 7.2% ± 0.9). The median of events <54 mg/dl was four events/patient/month (IQR 1-10), 77% of these events occurred during day time. Multivariate analysis showed that the number of events of hypoglycemia were higher in patients with previous severe hypoglycemia (IRR1.38; 95% CI 1.19-1.61; p < 0.001), high glycemic variability defined as Coefficient of Variation (CV%) > 36% (IRR 2.09; 95%CI 1.79-2.45; p < 0.001) and hypoglycemia unawareness. A protector effect was identified for adequate sensor calibration (IRR 0.77; 95%CI 0.66-0.90; p:0.001), and the use of bolus wizard >60% (IRR 0.74; 95%CI 0.58-0.95; p:0.017).

CONCLUSION

In spite of using advanced SAPT, clinically significant hypoglycemia is still a non-negligible risk. Only the identification and intervention of modifiable factors could help to prevent and reduce hypoglycemia in clinical practice.

Defining High Glycemic Variability in Type 1 Diabetes: Comparison of Multiple Indexes to Identify Patients at Risk of Hypoglycemia

Background: International consensus on the use of continuous glucose monitoring (CGM) recommends coefficient of variation (CV) as the metric of choice to express glycemic variability (GV) with a cutoff of 36% to define unstable diabetes. Even though, CV is associated with hypoglycemia in type 2 diabetes patients, the evidence on the use of one particular measure of GV in type 1 diabetes (T1DM) patients as a predictor of hypoglycemia is limited. Methods: A cohort of T1DM ambulatory patients was evaluated using CGM. Number and incidence rate of events <54 and <70 mg/dL were calculated. Bivariate and multivariate analysis of different glycemic indexes and clinical variables were performed to identify those associated with hypoglycemia. Receiver operating characteristic (ROC) curve analysis for each of the glycemic indexes was performed to define the best index and its optimal cutoff threshold to discriminate patients with events of hypoglycemia. Results: Seventy-three patients were included. A total of 128 events <54 mg/dL were recorded in 34 patients, and 350 events <70 mg/dL were registered in 51 patients. CV was the only variable significantly associated with hypoglycemia <54 mg/dL in the multivariate analysis (adjusted relative risk [aRR] 1.44, 95% confidence interval [CI]: 1.10-1.88, P = 0.008). CV, HbA1c (glycated hemoglobin), and mean glucose were associated with events <70 mg/dL. ROC curve analysis showed that, among GV metrics, CV had the best performance to discriminate patients with events <54 mg/dL (area under the curve [AUC] 0.87, 95% CI: 0.79-0.95) and events <70 mg/dL (AUC 0.79, 95% CI: 0.68-0.90) with optimal cutoff thresholds values of 34% and 31%, respectively. Among glycemic risk (GR) indexes, low blood glucose index (LBGI) showed the best performance. Conclusions: This analysis shows that CV is the best GV index, and LBGI the best GR index, to identify patients at risk of clinically significant hypoglycemia and hypoglycemia alert events in T1DM patients.

Different Indexes of Glycemic Variability as Identifiers of Patients with Risk of Hypoglycemia in Type 2 Diabetes Mellitus

INTRODUCTION

Recent publications frequently introduce new indexes to measure glycemic variability (GV), quality of glycemic control, or glycemic risk; however, there is a lack of evidence supporting the use of one particular parameter, especially in clinical practice.

METHODS

A cohort of type 2 diabetes mellitus (T2DM) patients in ambulatory care were followed using continuous glucose monitoring sensors (CGM). Mean glucose (MG), standard deviation, coefficient of variation (CV), interquartile range, CONGA1, 2, and 4, MAGE, M value, J index, high blood glucose index, and low blood glucose index (LBGI) were estimated. Hypoglycemia incidence (<54 mg/dl) was calculated. Area under the curve (AUC) was determined for different indexes as identifiers of patients with risk of hypoglycemia (IRH). Optimal cutoff thresholds were determined from analysis of the receiver operating characteristic curves.

RESULTS

CGM data for 657 days from 140 T2DM patients (4.69 average days per patient) were analyzed. Hypoglycemia was present in 50 patients with 144 hypoglycemic events in total (incidence rate of 0.22 events per patient/day). In the multivariate analysis, both CV (OR 1.20, 95% CI 1.12-1.28, P < .001) and LBGI (OR 4.83, 95% CI 2.41-9.71, P < .001) were shown to have a statistically significant association with hypoglycemia. The highest AUC were for CV (0.84; 95% CI 0.77-0.91) and LBGI (0.95; 95% CI 0.92-0.98). The optimal cutoff threshold for CV as IRH was 34%, and 3.4 for LBGI.

CONCLUSION

This analysis shows that CV can be recommended as the preferred parameter of GV to be used in clinical practice for T2DM patients. LBGI is the preferred IRH between glycemic risk indexes.

Impact of a Basal-Bolus Insulin Regimen on Metabolic Control and Risk of Hypoglycemia in Patients With Diabetes Undergoing Peritoneal Dialysis

INTRODUCTION

Clinical interventional studies in diabetes mellitus usually exclude patients undergoing peritoneal dialysis (PD). This study evaluates the impact of an educational program and a basal-bolus insulin regimen on the blood glucose level control and risk of hypoglycemia in this population.

METHODS

A before-and-after study was conducted in type 1 and type 2 DM patients undergoing PD at the Renal Therapy Services (RTS) clinic network, Bogota, Colombia. An intervention was instituted consisting of a three-month educational program and a basal-bolus detemir (Levemir, NovoNordisk) and aspart (Novorapid, NovoNordisk) insulin regimen. Prior to the intervention and at the end of treatment were conducted measures of HbA1c levels and continuous glucose monitoring (CGM).

RESULTS

Forty-seven patients were recruited. Mean HbA1c level decreased from 8.41% ± 0.83 to 7.68% ± 1.32 (mean difference -0.739, 95% CI -0.419, -1.059; P < .0001). Of subjects, 52% achieved HbA1c levels <7.5% at the end of study. Mean blood glucose level reduced from 194.0 ± 42.5 to 172.9 ± 31.8 mg/dl ( P = .0015) measured by CGM. Significant differences were not observed in incidence of overall ( P = .7739), diurnal ( P = .3701), or nocturnal ( P = .5724) hypoglycemia episodes nor in area under the curve (AUC) <54 mg/dl ( P = .9528), but a reduction in AUC >180 ( P < .01) and AUC >250 ( P = .01) was evidenced for total, diurnal, and nocturnal episodes.

CONCLUSIONS

An intervention consisting of an educational program and a basal-bolus insulin regimen in type 1 and type 2 diabetes mellitus patients undergoing PD caused a decrease in HbA1c levels, and mean blood glucose levels as measured from CGM with no significant increases in hypoglycemia episodes.

Dissipation and effects of tricyclazole on soil microbial communities and rice growth as affected by amendment with alperujo compost

The presence of pesticides in surface and groundwater has grown considerably in the last decades as a consequence of the intensive farming activity. Several studies have shown the benefits of using organic amendments to prevent losses of pesticides from runoff or leaching. A particular soil from the Guadalquivir valley was placed in open air ponds and amended at 1 or 2% (w/w) with alperujo compost (AC), a byproduct from the olive oil industry. Tricyclazole dissipation, rice growth and microbial diversity were monitored along an entire rice growing season. An increase in the net photosynthetic rate of Oryza sativa plants grown in the ponds with AC was observed. These plants produced between 1100 and 1300kgha(-1) more rice than plants from the unamended ponds. No significant differences were observed in tricyclazole dissipation, monitored for a month in soil, surface and drainage water, between the amended and unamended ponds. The structure and diversity of bacteria and fungi communities were also studied by the use of the polymerase chain reaction denaturing gel electrophoresis (PCR-DGGE) from DNA extracted directly from soil samples. The banding pattern was similar for all treatments, although the density of bands varied throughout the time. Apparently, tricyclazole did not affect the structure and diversity of bacteria and fungi communities, and this was attributed to its low bioavailability. Rice cultivation under paddy field conditions may be more efficient under the effects of this compost, due to its positive effects on soil properties, rice yield, and soil microbial diversity.

Influence of green waste compost on azimsulfuron dissipation and soil functions under oxic and anoxic conditions

Concerns have been raised over the sustainability of intensive rice cultivation, where the use of chemical fertilizers and pesticides has been associated with numerous environmental problems. The objective of this study was to test the effect of the herbicide azimsulfuron on important soil functions as affected by amendment with a byproduct of the olive oil industry. Soil was collected from a Mediterranean rice field. Part of it was amended with alperujo compost (AC). Amended and unamended soils were incubated for 43days in presence or not of azimsulfuron, under anoxic-flooded (AF) and oxic-unflooded (OU) conditions. We monitored the dissipation of the herbicide azimsulfuron, C mineralization, soil microbial biomass (SMB) and dissolved organic carbon (DOC) content and its nature. Under AF conditions, the application of compost produced an increase in the dissipation of the herbicide (up to 12.4%). It was related with the higher DOC content, 4 times higher than under OU conditions. Though increases in carbon turnover (under AF and OU conditions) and reduction of SMBC after herbicide application (only under AF conditions) were observed, the differences were not statistically significant. The application of this organic amendment is presented as an efficient management strategy to increase C turnover in agricultural soils and reduce some of the negative effects derived from the application of azimsulfuron under flooded conditions.

Effect of soil organic amendments on the behavior of bentazone and tricyclazole

The effect of soil amendment with different organic residues from olive oil production on the sorption and leaching of two pesticides used in rice crops (bentazone and tricyclazole) was compared in order to understand their behavior and to improve soil properties by recycling an abundant agricultural residue in Andalucía (S. Spain). A residue from olive oil production (AJ), the organic compost derived from this organic waste (CA) and a biochar (BA) made from CA were used. A soil devoted to rice cultivation, IFAPA (I), was amended at 2% (w/w) of each amendment individually (I+AJ, I+CA and I+BA). In order to evaluate the effect of dissolved organic matter (DOM) from these amendments on bentazone and tricyclazole behavior, the DOM from the amendments was extracted, quantified and characterized by fluorescence spectroscopy and FT-IR. The affinity of DOM for soil surfaces was evaluated with (I) soil and two other soils of different physicochemical properties, ARCO (A) and GUAD (G). These studies revealed differences in DOM quantity, quality and affinity for the used soils among amendments which can explain the different sorption behavior observed for tricyclazole in the amended soils. Leaching assays under saturated/unsaturated conditions revealed a slight delay of bentazone in I+CA and I+BA soils when compared to I+AJ, that can be related to the higher DOM content and much lower specific surface area of AJ. In contrast, tricyclazole was not detected in any of the leachates during the leaching assay. Extraction of tricyclazole residues from soil columns showed that the fungicide did not move below 5cm in the higher sorptive systems (I+CA, I+BA). The sorption of DOM from amendments on soil during the transport process can decrease the mobility of the fungicide by changing the physicochemical properties of the soil surface whose behavior may be dominated by the adsorbed DOM.

Prediction of postprandial blood glucose under uncertainty and intra-patient variability in type 1 diabetes: a comparative study of three interval models

The behavior of three insulin action and glucose kinetics models was assessed for an insulin therapy regime in the presence of patient variability. For this purpose, postprandial glucose in patients with type 1 diabetes was predicted by considering intra- and inter-patient variability using modal interval analysis. Equations to achieve optimal prediction are presented for models 1, 2 and 3, which are of increasing complexity. The model parameters were adjusted to reflect the "same" patient in the presence of variability. The glucose response envelope for model 1, the simplest insulin-glucose model assessed, included the responses of the other two models when a good fit of the model parameters was achieved. Thus, under variability, simple glucose-insulin models may be sufficient to describe patient dynamics in most situations.

Insulin dosage optimization based on prediction of postprandial glucose excursions under uncertain parameters and food intake

Considering the difficulty in selecting correct insulin doses and the problem of hyper- and hypoglycemia episodes in type 1 diabetes, dosage-aid systems are very useful for these patients. A model-based approach to this problem must unavoidably consider uncertainty sources such as large intra-patient variability and food intake. In the present study, postprandial glucose is predicted considering this uncertain information using modal interval analysis. This approach calculates a safer prediction of possible hyper- and hypoglycemia episodes induced by insulin therapy for an individual patient's parameters and integrates this information into a dosage-aid system. Predictions of a patient's postprandial glucose at 5-h intervals are used to predict the risk for a given therapy. Then the insulin dose and injection-to-meal time with the lowest risk are calculated. The method has been validated for three different scenarios corresponding to preprandial glucose values of 100, 180 and 250mg/dl.

Comparison of interval and Monte Carlo simulation for the prediction of postprandial glucose under uncertainty in type 1 diabetes mellitus

In this paper, the problem of tackling uncertainty in the prediction of postprandial blood glucose is analyzed. Two simulation approaches, Monte Carlo and interval models, are studied and compared. Interval simulation is carried out using modal interval analysis. Simulation of a glucoregulatory model with uncertainty in insulin sensitivities, glucose absorption and food intake is carried out using both methods. Interval simulation is superior in predicting all severe and mild hyper- and hypoglycemia episodes. Furthermore, much less computational time is required for interval simulation than for Monte Carlo simulation.

Computing the risk of postprandial hypo- and hyperglycemia in type 1 diabetes mellitus considering intrapatient variability and other sources of uncertainty

OBJECTIVE

The objective of this article was to develop a methodology to quantify the risk of suffering different grades of hypo- and hyperglycemia episodes in the postprandial state.

METHODS

Interval predictions of patient postprandial glucose were performed during a 5-hour period after a meal for a set of 3315 scenarios. Uncertainty in the patient's insulin sensitivities and carbohydrate (CHO) contents of the planned meal was considered. A normalized area under the curve of the worst-case predicted glucose excursion for severe and mild hypo- and hyperglycemia glucose ranges was obtained and weighted accordingly to their importance. As a result, a comprehensive risk measure was obtained. A reference model of preprandial glucose values representing the behavior in different ranges was chosen by a xi(2) test. The relationship between the computed risk index and the probability of occurrence of events was analyzed for these reference models through 19,500 Monte Carlo simulations.

RESULTS

The obtained reference models for each preprandial glucose range were 100, 160, and 220 mg/dl. A relationship between the risk index ranges <10, 10-60, 60-120, and >120 and the probability of occurrence of mild and severe postprandial hyper- and hypoglycemia can be derived.

CONCLUSIONS

When intrapatient variability and uncertainty in the CHO content of the meal are considered, a safer prediction of possible hyper- and hypoglycemia episodes induced by the tested insulin therapy can be calculated.

Prediction of glucose excursions under uncertain parameters and food intake in intensive insulin therapy for type 1 diabetes mellitus

Considering the difficulty in the insulin dosage selection and the problem of hyper-and hypoglycaemia episodes in type 1 diabetes, dosage-aid systems appear as tremendously helpful for these patients. A model-based approach to this problem must unavoidably consider uncertainty sources such as the large intra-patient variability and food intake. This work addresses the prediction of glycaemia for a given insulin therapy face to parametric and input uncertainty, by means of modal interval analysis. As result, a band containing all possible glucose excursions suffered by the patient for the given uncertainty is obtained. From it, a safer prediction of possible hyper-and hypoglycaemia episodes can be calculated.