Sage Journals: Discover world-class research

Abstract

Background:

Cardiac autonomic neuropathy (CAN) is a diabetes-related complication with increasing prevalence and remains challenging to detect in clinical settings. Machine learning (ML) approaches have the potential to predict CAN using clinical data. In this study, we aimed to develop and evaluate the performance of an ML model to predict early CAN occurrence in patients with diabetes.

Methods:

We used the diabetes complications screening research initiative data set containing 200 CAN-related tests on more than 2000 participants with type 2 diabetes in Australia. Data were collected on peripheral nerve functions, Ewing’s tests, blood biochemistry, demographics, and medical history. The ML model was validated using 10-fold cross-validation, of which 90% were used in training the model and the remaining 10% was used in evaluating the performance of the model. Predictive accuracy was assessed by area under the receiver operating curve, and sensitivity, specificity, positive predictive value, and negative predictive value.

Results:

Of the 237 patients included, 105 were diagnosed with an early stage of CAN while the remaining 132 were healthy. The ML model showed outstanding performance for CAN prediction with receiver operating characteristic curve of 0.962 [95% confidence interval (CI) = 0.939–0.984], 87.34% accuracy, and 87.12% sensitivity. There was a significant and positive association between the ML model and CAN occurrence (p < 0.001).

Conclusion:

Our ML model has the potential to detect CAN at an early stage using Ewing’s tests. This model might be useful for healthcare providers for predicting the occurrence of CAN in patients with diabetes, monitoring the progression, and providing timely intervention.

Keywords

cardiac autonomic neuropathy (CAN)diabetes mellitus (DM)Ewing’s battery tests health informatics logistic regression predictive analysis

Introduction

Diabetes is one of the significant burden of diseases globally with growing prevalence and leads to increased morbidity and mortality.^1,2 The International Diabetes Federation estimated that by 2040, one in every 10 people in the world would have diabetes.³ Patients with diabetes often present with several co-morbidities, including cardiovascular diseases,^4,5 stroke, kidney diseases, depression,^6,7 which affect their quality of life⁸ and lead to increased morbidity. Cardiac autonomic neuropathy (CAN) is one of the most severe complications of diabetes resulting from a complex interaction of blood glucose control, duration of disease, age, and blood pressure (BP).^9–11 CAN affect nerve fibers in the blood vessels and the heart muscle in diabetic patients, thereby causing cardiac arrhythmias, exercise intolerance, myocardial injury, silent myocardial ischemia, stroke, and sudden cardiac deaths^12,13 and substantially impact the quality of life of patients.^10,14 Furthermore, CAN independently predict the progression of diabetic nephropathy and chronic kidney disease in diabetes.¹⁵ Longitudinal studies in patients with CAN have shown 5-year mortality rates of 16–50%, with a high proportion attributed to sudden cardiac death.¹⁶

Globally, there is an increase in the occurrence of CAN, particularly in patients with type 2 diabetes.^17,18 Previous studies estimated that 20–73% of type 2 diabetes patients and about 17% of type 1 diabetes patients develop CAN.^19,20 Subclinical CAN, manifested as changes in heart rate variability (HRV), may be detected within 1–2 years of diagnosis of diabetes.¹⁶ The clinical symptoms of CAN, however, often do not appear until long after diabetes onset, thus remaining undiagnosed until the disease progresses to an advanced stage.²¹ The early detection and appropriate management of CAN are essential to prevent future complications and has been widely recommended.^15,22

Despite the importance of early detection, currently, there is no specific method for predicting CAN and its progression.¹⁹ CAN diagnosis is commonly performed in the clinics using Ewing’s test that includes the assessment of HRV, orthostatic hypotension, and 24 h of BP profiles and other variables in clinical settings.²³ There is, however, still debate about the diagnostic criteria and staging of CAN using Ewing’s test.²³ A majority of the previous research conducted in this regard has focused on using Ewing’s battery test alongside a wide range of risk factors to diagnose CAN.^24–32 Although these methods have performed well, the progression of CAN remains unclear. This has a significant impact on the ability of healthcare providers to increase the awareness of patients and to intervene promptly. In this study, we aimed at addressing this gap by developing and validating a machine learning (ML) model to detect and predict the progression of CAN in diabetic patients.

Methods

Design

We conducted secondary analysis from a retrospective cohort study. We developed and evaluated the performance of an ML model for detecting and prediction of CAN progression using the following steps: develop an ML model for CAN detection and prediction of progress; investigate Ewing’s tests significance in the prediction of normal and early categories of CAN; and evaluate the proposed ML predictive model.

Participants, location, and data collection

We used the Diabetes Complications Screening Research Initiative (DiScRi) data set,^26,33,34 which was collected at Charles Sturt University in Australia.³⁵ The DiScRi data set contains more than 200 variables conducted on more than 2000 participants with type 2 diabetes in rural New South Wales, Australia, between 2011 and 2014. Patients were recruited through a public media campaign, including newspaper advertisements, radio and local television, and advertisements in general practice and community health centers. Potential participants were requested to contact the university if they wished to undergo a health check, and an appointment was made to attend the clinic. All participants older than 40 years were eligible to participate.³³ Participants with existing cardiovascular, respiratory, and renal disease as well as depression, schizophrenia, and Parkinson’s disease were excluded. The data collection procedure involved the following steps: All participants were required to stop smoking or to consume drinks like alcohol and coffee 24 h before being tested. They were required to fast, beginning from midnight prior to the testing day. The tests were conducted from 9:00 a.m. to 12:00 p.m.

Variables and measurements

The data set contains a record of participants’ details, including demographic data such as age and sex, and history of diabetes, heart attack, palpitations, and atrial fibrillation. The data set also contained measurements of BP, body mass index (BMI), blood glucose level (BGL), cholesterol profile, and electrocardiography (ECG), and tests on peripheral nerve function, Ewing’s battery tests, HRV, and attributes of different blood biochemistry.

In the current study, we used Ewing’s tests, which include the five standard tests for CAN proposed by Ewing and Clarke.³⁶ The five tests are (1) lying to standing heart rate (LSHR) change expressed by 30:15 ratio. Such test indicates the ratio of longest R-R interval (ranging from 20 to 40 beats) to the shortest R-R interval (ranging from 5 to 25 beats) produced by a change in position (from a horizontal position to vertical position); (2) deep breathing heart rate (DBHR) change, which refers to the evaluation of beat-to-beat heart rate variation (R-R variation) based on deep breathing; (3) valsalva maneuver heart rate (VAHR) change measuring the response of heart rate during and after increasing the intra-abdominal and intrathoracic pressure; (4) handgrip blood pressure (HGBP) change measuring the change in diastolic BP after using a handgrip dynamometer; and (5) lying to standing BP (LSBP) change measuring the difference in the baroreflex-mediated BP after a change in the position. The results of Ewing’s test are shown in Supplementary Table S2.

Based on the Ewing’s tests, CAN has been categorized into five main classes: (1) normal (all tests normal or one borderline); (2) early CAN (one of the three heart rate tests abnormal or two borderline); (3) definite CAN (two or more of the heart rate tests abnormal); (4) severe CAN (two or more of the heart rate tests abnormal plus one or both of the BP tests abnormal or both borderline); and (5) atypical CAN (any other combination of tests with abnormal results).²⁶ Because there were only a few patients with a severe, definitive, and atypical CAN in this study, we excluded them from the analysis and only included 237 patients with normal and early CAN.

Ethics

Written informed consent was obtained from all participants before data collection. The protocol for the DiScRi study was approved by the Ethics in Human Research Committee of the Charles Sturt University (ID # 03/164).

The framework of the ML predictive model

Data analysis: Figure 1 shows the high-level framework along the steps followed to construct the proposed model. We used logistic regression which is one of the most common ML algorithms to develop the model and compatibility with a wide range of tools and platforms like R-language using the following formula³⁷

P = \frac{1}{1 + e x p^{- z}}

(1)

where P denotes the probability of an outcome occurrence and z represents the linear combination function which is inclusive of independent variables. An expression z can also be given as follows

z = β_{0} + β_{1} v_{1} + β_{2} v_{2} + \dots β_{n} v_{n}

(2)

where $β_{0}$ is represents the expected mean value of P when all $ν = 0$ ; n denotes the number of the independent variables; $β_{n}$ is the regression coefficient of each independent variable which is the influence of each independent variable on the likelihood of value (P ), and ν_n represents the independent variables that are included. Through the application of the linear combination function given in equation (1) for the Ewing’s tests as independent variables to determine an outcome (CAN progress), the predictive model’s formula performs the computation of the possibility of the occurrence of CAN progress based on the predictability of Ewing’s tests as expressed below

P = \frac{1}{1 + \exp^{(- (β_{0} + β_{1} T_{1} + β_{2} T_{2} + β_{3} T_{3} + β_{4} T_{4} + β_{5} T_{5}))}}

(3)

where P denotes the possibility of the occurrence of an outcome (CAN progress) based on the selected independent variables. $β$ indicates the regression coefficients of Ewing’s test that has been included. $T_{1}, T_{2}, T_{3}, T_{4}, and T_{5}$ refer to Ewing’s tests LSHR, DBHR, VAHR, HGBP, and LSBP, respectively. The description of Ewing’s battery tests is presented below.

Figure 1.

A graphical description of the proposed model.

Subsequently, the Ewing’s tests coefficients were calculated using logistic regression. The results obtained from the application of logistic regression are presented in Table 1. For the logistic regression, the dependent variable is CAN, whereas the independent variables are Ewing’s tests. Table 1 shows a significant association between Ewing’s tests and CAN (p < 0.05), and as such, have been included in the predictive model.

Table 1.

The results of logistic regression analysis.

Test	Coefficient	p value	Odds ratio	Confidence interval (%)
Intercept	28.88	0.000	3.48
LSHR	−4.68	0.013	0.009	0.000−0.373
DBHR	−0.488	0.000	0.614	0.53−0.711
VAHR	−11.57	0.000	0.009	0.00−0.001
HGBP	−0.197	0.000	0.821	0.754−0.894
LSBP	0.11	0.003	1.116	1.038−1.199

DBHR, deep breathing heart rate; HGBP, handgrip blood pressure; LSBP, lying to standing blood pressure; LSHR, lying to standing heart rate; VAHR, valsalva maneuver heart rate.

As presented in Table 1, the $β_{0} is 28.88$ and coefficients of T₁ and T₂ tests were $β_{1} T_{1}$ = −4.68, p = 0.013, and $β_{2} T_{2}$ = −0.488, p = 0.000, respectively. The coefficients of T₃ and T₄ tests were $β_{3} T_{3}$ = −11.57, p = 0.000, and $β_{4} T_{4}$ = −0.197, p = 0.000, respectively. As for T₅, the coefficient was $β_{5} T_{5}$ = 0.11, p = 0.000.

Negative β coefficient values were obtained as the outcomes of T₁ (LSHR), T₂ (DBHR), T₃ (VAHR), and T₄ (HGBP) indicating that they have a negative influence on the occurrence of the disease. In other words, the influence of the tests on the occurrence of the disease is negative. In contrast, positive β coefficient values were produced by the T₅ (LSBP), which is indicative of the positive effect of the tests on the occurrence of the disease. This confirms the correctness of the predictive model depending on Ewing’s rules to diagnose the condition. Based on equation (3) and the outcomes in Table 1, the predictive model was built as follows

P = \frac{1}{1 + \exp^{(- (28.88 - (4.68 \times T_{1}) - (0.488 \times T_{2}) - (11.57 \times T_{3}) - (0.197 \times T_{4}) + (0.11 \times T_{5})))}}

(4)

Subsequent to the construction of the predictive model of CAN, if new tests are obtained from the clinic, for example, the predictive model can be used in calculating the probability of the occurrence of CAN. The following cases are classic illustrations of the application of the predictive model in determining the probability of CAN. Assume the Ewing’s tests for patient A are LSHR = 1.06, DBHR = 16, VAHR = 1.1, HGBP = 15, and LSBP = 10; patient B are LSHR = 1.4, DBHR = 10, VAHR = 1.23, HGBP = 15, and LSBP = 8; and patient C are LSHR = 1.1, DBHR = 14, VAHR = 1.1, HGBP = 14, and LSBP = 8. Through the application of equation (4) with the tests of the three patients, the probability that diabetic patient A will develop CAN can be computed as follows

P (A) = \frac{1}{1 + \exp^{(- (28.88 - (4.68 \times 1.06) - (0.488 \times 16) - (11.57 \times 1.1) - (0.197 \times 1.15) + (0.11 \times 10)))}} = 0.82

For patient B, the probability of developing CAN’s disease can be calculated as follows

P (B) = \frac{1}{1 + \exp^{(- (28.88 - (4.68 \times 1.4) - (0.488 \times 10) - (11.57 \times 1.23) - (0.197 \times 15) + (0.11 \times 8)))}} = 0.76

The probability of patient C developing CAN’s disease can be calculated as follows

P (C) = \frac{1}{1 + \exp^{(- (28.88 - (4.68 \times 1.1) - (0.488 \times 14) - (11.57 \times 1.1) - (0.197 \times 14) + (0.11 \times 8)))}} = 0.86

Evaluation of the predictive model

The evaluation of the efficiency of the proposed predictive model was done using sensitivity, accuracy, specificity, type I error and type II error, and confusion matrix measurements. The model was validated using 10-fold cross-validation to obtain an unbiased evaluation of generalization error. The whole data set was randomly divided into 10 subsets, of which nine (90%) were used in training the model and the remaining 10% was used in testing the performance of the model. The training and testing procedures were carried out 10 times repeatedly. All experiments were performed using SPSS version 20.0 (SPSS, Inc., Chicago, IL, USA) and R programming language on a personal computer with Intel Core i5, CPU 3.4 GB, 16 GB RAM running Windows 10 operating system.

Performance metrics

The performance of the model was evaluated using the area under the curve (AUC), which is referred to as a receiver operating characteristic (ROC) curve using the following assessment measures:

Sensitivity indicates the number of patients with CAN that are correctly predicted

Senstivity = \frac{TP}{(FN + TP)}

(5)

Specificity indicates the number of controls that are correctly predicted

Specificity = \frac{TN}{(FP + TN)}

(6)

Accuracy exposes the total number of the patients and controls that are correctly predicted

Accuracy = \frac{(TP + TN)}{(Total number)}

(7)

Type I error (α) is the probability that patients will be diagnosed into the control group

Type I error = \frac{FN}{(FN + TP)} = 1 - Sensitivity

(8)

Type II error refers to the probability that control group will be diagnosed and categorized into patients’ group

Type II error = \frac{FP}{(FP + TN)} = 1 - Specificity

(9)

Results

A total of 237 patients with type 2 diabetes participated in this study (55% females, age range = 32–90 years), of which 105 were diagnosed with an early stage of CAN while the remaining 132 participants were found to be healthy and disease-free. Of the 132 healthy participants, the health statuses of 119 participants were correctly predicted as healthy by the predictive model, and of 105 patients, 88 were correctly predicted to be at the early stage of the disease. The confusion matrix (Table 2) shows the predictive model has achieved a predictive accuracy of 87.34%. This performance is achieved because all the included tests are significantly associated with the tested categories of CAN (p < 0.05).

Table 2.

The confusion matrix.

Class	Normal	Early
Normal	119	13
Early	17	88

The predictive model was able to achieve a sensitivity value of 87.12%, being the number of people that were correctly identified with the positive disease. Meanwhile, the model was able to achieve specificity of 87.5%. Also, the predictive model achieved values of 12.88% and 12.5% for type I error and type II error, respectively. This is indicative of the ability of the model to efficiently predict the probability of the occurrence of CAN. This model can be exploited with a wide range of health conditions.

The ROC curves of the predictive model are presented in Figure 2, while the AUC of the predictive model is enlisted in Table 3. The predictive model achieved an ROC of 0.962% with a 95% confidence interval of 0.939–0.984. In addition, the model achieved a significant difference from 0.5, given that p value (asymptotic significance) is less than 0.05, indicating that the proposed model achieved a significant prediction better than by chance.

Figure 2.

ROC curve of the predictive model.

Table 3.

The area under the curve of the predictive model.

Area under the curve
Test result variable(s): predicted probability
Area	Std. error	Asymptotic sig.^a	Asymptotic 95% confidence interval
Area	Std. error	Asymptotic sig.^a	Lower bound	Upper bound
0.962	0.012	0.000	0.939	0.984

P-value (Asymptotic Significance) is < 0.05.

Discussion

In this article, we presented an ML model to predict the probability of CAN occurrence in patients with diabetes. The results suggest that an ML method can predict the risk of developing CAN in patients with diabetes in the primary care setting. The model provided high accuracy (87.34%), sensitivity (87.12%) and specificity (87.5%) as well as very high stability (0.962%). The cardiovascular autonomic function abnormalities in CAN may develop before diabetes is manifested and remain subclinical. Our results provide evidence of the suitability of the use of the ML model to detect CAN early with minimum resources. The predictive model also provided further evidence that Ewing’s tests can significantly predict the normal and early categories of this condition. Thus, our model can be used by healthcare providers to detect and follow the progression of CAN over time and take appropriate measures.

Our findings support and extend prior studies of CAN detection using Ewing’s tests and ML models. In a previous study by Jelinek et al.,²⁴ the neurological diagnostics of CAN was examined on the basis of five Ewing’s test attributes using decision tree classification like REPtree, J48, SimpleCart, and NBTree. In a study by Kelarev et al.,²⁵ Ewing’s test features were combined with extra features from the clinical data to detect CAN using feature selection based on random forest and multilevel ensemble classifiers. A research conducted by Abawajy et al.²⁶ focused on enhancing the accuracy of the classification of CAN using blood biochemistry features alongside the results of Ewing’s tests. They achieved the enhancement of the classification accuracy through the use of the automated iterative multitier ensemble (AIME), which makes use of a variety of ensemble classifiers in each layer. The AIME achieved a high level of accuracy (99.57%) and could be used where Ewing’s tests are not available. In the current study, however, we used logistic regression which is more robust to detect CAN and its progression rather than only detecting CAN as performed by previous researchers.

Previous studies have attempted to predict CAN using different approaches. A hybrid of wrapper filter feature selection was proposed by Huda et al.²⁷ using ECG features and Ewing’s tests. Based on Ripple down Rules, ensemble classifiers were proposed by Kelarev et al.²⁸ to predict CAN using Ewing’s tests. Kelarev et al.²⁹ proposed ensemble classifiers based on decision tree classifiers for the monitoring of diabetic patients through using ECG features and Ewing’s tests. A different study conducted by Abawajy et al.³⁰ proposed a multitier ensemble classifier for the prediction of CAN, based on the QRS features of ECG and Ewing’s tests. Jelinek et al.³¹ discussed an iterative multilayer attributes selection and classification model, which used Ewing’s tests and HRV for the prediction of CAN. A meta-ensemble model was proposed in another study³² to investigate the prediction of CAN using HRV. Although these models^24–32 demonstrated superior performances, they did not address the problem of predicting the probability of CAN occurrence in patients with diabetes. In addition, the models were more general, where they have been proposed to predict the categories of the disease (e.g. normal, early, severe, definitive, abnormal). In contrast, our ML model can measure the relationship between the CAN categories and Ewing’s tests by estimating probabilities using a logistic function, which is the cumulative logistic distribution and a simple novel method.

The main strength of the study is the applicability of the results to the primary care diabetes population in rural settings. This study has limitations that should be considered. First, the small sample size might not be representative of all diabetes patients. Therefore, the results should be interpreted with caution. Second, individuals in our study were diagnosed using only the Ewing’s test with normal and early categories. Third, the probabilities predicted by our model were not calibrated to true occurrence probabilities. Furthermore, we did not correlate our model with clinical findings and biochemical data. In future, we aim to compare the model with various clinical and biochemical tests in more extensive studies with a more varied population.

CAN is a serious medical complication of diabetes and an independent predictor of cardiovascular morbidity and mortality. The importance of early diagnosis of CAN is widely recognized. Diagnosis of CAN, however, is problematic due to the medical expertise required for evaluating the data from Ewing’s test even when resources for the test are available. More broadly, our proposed model points toward the application of innovative methods for the early detection of CAN. The unique feature of our model is that it is easy and simple to employ and does not require a complex code. Thus, our model can be applied as a mobile application or online tool,³⁸ for example, in combination with the National Heart Foundation of Australia heart age calculator which is commonly used by clinicians in general practice (https://www.heartfoundation.org.au/heart-age-calculator). These results might be useful for healthcare providers to predict the early occurrence of CAN in people with diabetes and support them in combination with digital technologies^39–42 such as wearable devices,⁴³ smartphone apps⁴⁴ and text messages^45,46 which have shown to be cost-effective.^47,48

Furthermore, our study suggests several directions for future work. The ML models can be potentially applied in primary healthcare settings to early detect CAN by general practitioners and specialist nurses where there is a lack of trained healthcare specialists. Clinicians may use these models to monitor the progress of CAN using cheap and straightforward digital health tools such as web-based and mobile phone applications in their clinics. This may lead to early detection and prevention of CAN-related complications and morbidity. Furthermore, the methods used in this study offer a principled way to help inform individualized care using routine data from clinics. Before implementation in clinical practice, however, further research using large local data sets and randomized trials evaluating the effectiveness and cost-effectiveness is recommended.

Conclusion

The ML model developed in this study showed to predict early-stage CAN, which might be useful for both healthcare providers and patients for early intervention, thereby leading to the prevention of CAN-related complications.

Supplemental Material

sj-docx-1-tae-10.1177_20420188221086693 – Supplemental material for Prediction of cardiac autonomic neuropathy using a machine learning model in patients with diabetes

Supplemental material, sj-docx-1-tae-10.1177_20420188221086693 for Prediction of cardiac autonomic neuropathy using a machine learning model in patients with diabetes by Ahmad Shaker Abdalrada, Jemal Abawajy, Tahsien Al-Quraishi and Sheikh Mohammed Shariful Islam in Therapeutic Advances in Endocrinology and Metabolism

Footnotes

Acknowledgements

We are grateful to Herbert F. Jelinek from the School of Community Health, Charles Sturt University, Albury, Australia, for providing us access to the DiScRi data sets in an anonymous format.

Author contributions

Ahmad Shaker Abdalrada: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Project administration; Writing – original draft.

Jemal Abawajy: Conceptualization; Data curation; Formal analysis; Supervision; Validation; Writing – review & editing.

Tahsien Al-Quraishi: Conceptualization; Data curation; Formal analysis; Investigation; Project administration; Validation; Writing – original draft.

Sheikh Mohammed Shariful Islam: Conceptualization; Supervision; Validation; Writing – original draft; Writing – review & editing.

Conflict of interest statement

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: SMSI is supported by a Fellowship from the National Health and Medical Research Council (NHMRC) and National Heart Foundation of Australia.

ORCID iD

Sheikh Mohammed Shariful Islam

Supplemental material

Supplemental material for this article is available online.

References

Vos

Lim

Abbafati

, et al. Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 2020; 396: 1204–1222.

Islam

SMS

Purnat

Phuong

NTA

, et al. Non-communicable diseases (NCDs) in developing countries: a symposium report. Global Health 2014; 10: 81.

Ogurtsova

da Rocha Fernandes

Huang

, et al. IDF diabetes atlas: global estimates for the prevalence of diabetes for 2015 and 2040. Diabetes Res Clin Pract 2017; 128: 40–50.

Islam

SMS

Ahmed

Uddin

, et al. Cardiovascular diseases risk prediction in patients with diabetes: posthoc analysis from a matched case-control study in Bangladesh. J Diabetes Metab Disord 2021; 20: 417–425.

Islam

SMS

Mainuddin

Islam

, et al. Prevalence of risk factors for hypertension: a cross-sectional study in an urban area of Bangladesh. Glob Cardiol Sci Pract 2015; 2015: 43.

Islam

SMS

Ferrari

Seissler

, et al. Association between depression and diabetes amongst adults in Bangladesh: a hospital based case–control study. J Glob Health 2015; 5: 020406.

Islam

SMS

Rawal

Niessen

. Prevalence of depression and its associated factors in patients with type 2 diabetes: a cross-sectional study in Dhaka, Bangladesh. Asian J Psychiatr 2015; 17: 36–41.

Safita

Islam

SMS

Chow

, et al. The impact of type 2 diabetes on health related quality of life in Bangladesh: results from a matched study comparing treated cases with non-diabetic controls. Health Qual Life Outcomes 2016; 14: 129.

Vinik

Maser

Mitchell

, et al. Diabetic autonomic neuropathy. Diabetes Care 2003; 26: 1553–1579.

10.

Maser

Mitchell

Vinik

, et al. The association between cardiovascular autonomic neuropathy and mortality in individuals with diabetes: a meta-analysis. Diabetes Care 2003; 26: 1895–1901.

11.

Maser

Lenhard

DeCherney

SG.

Cardiovascular autonomic neuropathy: the clinical significance of its determination. Endocrinologist 2000; 10: 27.

12.

Retnakaran

Zinman

Type 1 diabetes, hyperglycaemia, and the heart. Lancet 2008; 371: 1790–1799.

13.

Banthia

Bergner

Chicos

, et al. Detection of cardiovascular autonomic neuropathy using exercise testing in patients with type 2 diabetes mellitus. J Diabetes Complications 2013; 27: 64–69.

14.

Maser

Lenhard

MJ.

Cardiovascular autonomic neuropathy due to diabetes mellitus: clinical manifestations, consequences, and treatment. J Clin Endocrinol Metab 2005; 90: 5896–5903.

15.

Pop-Busui

Boulton

AJM

Feldman

, et al. Diabetic neuropathy: a position statement by the American Diabetes Association. Diabetes Care 2017; 40: 136–154.

16.

Pop-Busui

Cardiac autonomic neuropathy in diabetes. A clinical perspective. Diabetes Care 2010; 33: 434–441.

17.

Hazari

MAH

Khan

Reddy

, et al. Cardiovascular autonomic dysfunction in type 2 diabetes mellitus and essential hypertension in a South Indian population. Neurosciences 2012; 17: 173–175.

18.

Ziegler

Cardiovascular autonomic neuropathy: clinical manifestations and measurement. Diabetes Rev 1999; 7: 300–315.

19.

Dimitropoulos

Tahrani

Stevens

MJ.

Cardiac autonomic neuropathy in patients with diabetes mellitus. World J Diabetes 2014; 5: 17.

20.

Voulgari

Papadogiannis

Tentolouris

Diabetic cardiomyopathy: from the pathophysiology of the cardiac myocytes to current diagnosis and management strategies. Vasc Health Risk Manag 2010; 6: 883–903.

21.

Low

Benrud-Larson

Sletten

, et al. Autonomic symptoms and diabetic neuropathy: a population-based study. Diabetes Care 2004; 27: 2942–2947.

22.

Stranieri

Abawajy

Kelarev

, et al. An approach for Ewing test selection to support the clinical assessment of cardiac autonomic neuropathy. Artif Intell Med 2013; 58: 185–193.

23.

Tesfaye

Boulton

Dyck

, et al. Diabetic neuropathies: update on definitions, diagnostic criteria, estimation of severity, and treatments. Diabetes Care 2010; 33: 2285–2293.

24.

Jelinek

Abawajy

Kelarev

, et al. Decision trees and multi-level ensemble classifiers for neurological diagnostics. Aust J Med Sci 2014; 1: 1–12.

25.

Kelarev

Stranieri

Yearwood

, et al. Improving classifications for cardiac autonomic neuropathy using multi-level ensemble classifiers and feature selection based on random forest. In: Proceedings of the tenth Australasian data mining conference, vol. 134, Sydney, NSW, Australia, 5–7 December 2012, pp. 93–101. Sydney, NSW, Australia: Australian Computer Society, Inc.

26.

Abawajy

Kelarev

Chowdhury

, et al. Enhancing predictive accuracy of cardiac autonomic neuropathy using blood biochemistry features and iterative multitier ensembles. IEEE J Biomed Health Inform 2016; 20: 408–415.

27.

Huda

Jelinek

Ray

, et al. Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a hybrid of wrapper-filter based feature selection. In: 2010 sixth international conference on intelligent sensors, sensor networks and information processing (ISSNIP), Brisbane, QLD, Australia, 7–10 December 2010, pp. 297–302. New York: IEEE.

28.

Kelarev

Dazeley

Stranieri

, et al. Detection of CAN by ensemble classifiers based on ripple down rules. In: Richards

Kang

(eds) Knowledge management and acquisition for intelligent systems. PKAW 2012. Berlin: Springer, 2012, pp. 147–159.

29.

Kelarev

Stranieri

Yearwood

, et al. Empirical study of decision trees and ensemble classifiers for monitoring of diabetes patients in pervasive healthcare. In: 2012 15th international conference on network-based information systems (NBiS), Melbourne, VIC, Australia, 26–28 September 2012, pp. 441–446. New York: IEEE.

30.

Abawajy

Kelarev

Stranieri

, et al. Empirical investigation of multi-tier ensembles for the detection of cardiac autonomic neuropathy using subsets of the Ewing features. In: Proceedings of the workshop on new trends of computational intelligence in healthcare applications, CI-health, Sydney, NSW, Australia, 4 December 2012, pp. 1–11, https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.416.7758&rep=rep1&type=pdf

31.

Jelinek

Abawajy

Cornforth

, et al. Multi-layer attribute selection and classification algorithm for the diagnosis of cardiac autonomic neuropathy based on HRV attributes. AIMS Med Sci 2015; 2: 396–409.

32.

Abdalrada

Abawajy

Chowdhury

, et al. Meta learning ensemble technique for diagnosis of cardiac autonomic neuropathy based on heart rate variability features. In: 30th International conference on computer applications in industry and engineering, CAINE 2017, San Diego, CA, 2–4 October 2017, pp. 169–175. Winona, MN: International Society for Computers and Their Applications.

33.

White

Wang

Jelinek

HF.

Management of hypertension in patients with diabetes mellitus. Exp Clin Cardiol 2010; 15: 5–8.

34.

Abdalrada

Abawajy

Al-Quraishi

, et al. Machine learning models for prediction of co-occurrence of diabetes and cardiovascular diseases: a retrospective cohort study. J Diabetes Metab Disord. Epub ahead of print 12 January 2022. DOI: 10.1007/s40200-021-00968-z.

35.

Jelinek

Wilding

Tinely

An innovative multi-disciplinary diabetes complications screening program in a rural community: a description and preliminary results of the screening. Aust J Prim Health 2006; 12: 14–20.

36.

Ewing

Martyn

Young

, et al. The value of cardiovascular autonomic function tests: 10 years experience in diabetes. Diabetes Care 1985; 8: 491–498.

37.

Nampak

Pradhan

Manap

MA.

Application of GIS based data driven evidential belief function model to predict groundwater potential zonation. J Hydrol 2014; 513: 283–300.

38.

Dening

Islam

SMS

George

, et al. Web-based interventions for dietary behavior in adults with type 2 diabetes: systematic review of randomized controlled trials. J Med Internet Res 2020; 22: e16437.

39.

Chow

Ariyarathna

Islam

SMS

, et al. mHealth in cardiovascular health care. Heart Lung Circ 2016; 25: 802–807.

40.

Islam

SMS

Maddison

. Digital health approaches for cardiovascular diseases prevention and management: lessons from preliminary studies. Mhealth 2021; 7: 41.

41.

Moses

Adibi

Shariful Islam

, et al. Application of smartphone technologies in disease monitoring: a systematic review. Healthcare 2021; 9: 889.

42.

Frost

Mahmud

Kaiser

, et al. Innovative approaches to strengthening health systems in low-and middle-income countries: current models, developments, and challenges. Health Policy Technol 2021; 10: 100567.

43.

Islam

Cartledge

Karmakar

, et al. Validation and acceptability of a cuffless wrist-worn wearable blood pressure monitoring device among users and healthcare professionals: a mixed-method study. JMIR mHealth uHealth 2019; 7: e14706.

44.

Daryabeygi-Khotbehsara

Islam

SMS

Dunstan

, et al. Smartphone-based interventions to reduce sedentary behavior and promote physical activity using integrated dynamic models: systematic review. J Med Internet Res 2021; 23: e26315.

45.

Shariful Islam

Niessen

Ferrari

, et al. Effects of mobile phone SMS to improve glycemic control among patients with type 2 diabetes in Bangladesh: a prospective, parallel-group, randomized controlled trial. Diabetes Care 2015; 38: e112–e113.

46.

Shariful Islam

Farmer

Bobrow

, et al. Mobile phone text-messaging interventions aimed to prevent cardiovascular diseases (Text2PreventCVD): systematic review and individual patient data meta-analysis. Open Heart 2019; 6: e001017.

47.

Islam

SMS

Peiffer

Chow

, et al. Cost-effectiveness of a mobile-phone text messaging intervention on type 2 diabetes – a randomized-controlled trial. Health Policy Technol 2020; 9: 79–85.

48.

Shariful Islam

Lechner

Ferrari

, et al. Mobile phone use and willingness to pay for SMS for diabetes in Bangladesh. J Public Health 2016; 38: 163–169.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB