R. Lee Kennedy

Application of autonomous neural network systems to medical pattern classification tasks

This paper presents a study of the application of autonomously learning multiple neural network systems to medical pattern classification tasks. In our earlier work, a hybrid neural network architecture has been developed for on-line learning and probability estimation tasks. The network has been shown to be capable of asymptotically achieving the Bayes optimal classification rates, on-line, in a number of benchmark classification experiments. In the context of pattern classification, however, the concept of multiple classifier systems has been proposed to improve the performance of a single classifier. Thus, three decision combination algorithms have been implemented to produce a multiple neural network classifier system. Here the applicability of the system is assessed using patient records in two medical domains. The first task is the prognosis of patients admitted to coronary care units; whereas the second is the prediction of survival in trauma patients. The results are compared with those from logistic regression models, and implications of the system as a useful clinical diagnostic tool are discussed.

Free fatty acid receptors: emerging targets for treatment of diabetes and its complications

Fatty acids (FAs) are important as metabolic substrates and as structural components of biological membranes. However, they also function as signalling molecules. Recently, a series of G protein-coupled receptors (GPRs) for FAs has been described and characterized. These receptors have differing specificities for FAs of differing chain length and degree of saturation, for FA derivatives such as oleoylethanolamide, and for oxidized FAs. They are a critical component of the body’s nutrient sensing apparatus, and small molecule agonists and antagonists of these receptors show considerable promise in the management of diabetes and its complications. Agonists of the long-chain free fatty acid receptors FFAR1 and GPR119 act as insulin secretagogues, both directly and by increasing incretins. Although, drugs acting at short-chain FFA receptors (FFAR2 and FFAR3) have not yet been developed, they are attractive targets as they regulate nutrient balance through effects in the intestine and adipose tissue. These include regulation of the secretion of cholecystokinin, peptide YY and leptin. Finally, GPR132 is a receptor for oxidized FAs, which may be a sensor of lipid overload and oxidative stress, and which is involved in atherosclerosis. Regulation of its signalling pathways with drugs may decrease the macrovascular risk experienced by diabetic patients. In summary, FA receptors are emerging drug targets that are involved in the regulation of nutrient status and carbohydrate tolerance, and modulators of these receptors may well figure prominently in the next generation of antidiabetic drugs.

Thyroid cancer management

OBJECTIVE Thyroid cancer is the commonest endocrine malignancy, yet management remains controversial. Many endocrinologists advocate diagnosis by fine needle aspiration (FNA), treatment by thyroidectomy, ablative radioiodine (131I) and TSH suppression, together with follow‐up with 131I scans or thyroglobulin (Tg) measurements. 131I (therapy or diagnosis) Is given only when TSH Is >30 mIU/I. With this strategy in mind, the aim of the present study was to audit existing clinical practice In a large Edinburgh teaching hospital to establish whether a need existed for local guidelines for the management of thyroid cancer.

Dementia: opportunities for risk reduction and early detection in general practice

This project aimed to measure general practitioner (GP), practice nurse and patient health literacy about memory problems, dementia and its risk factors. Data were collected from general practices across Australia and a smaller sample in England. Questionnaires explored sources and adequacy of dementia knowledge and a randomised controlled trial tested the intervention of a dementia risk reduction pamphlet on patient knowledge of dementia risk reduction strategies. Data were analysed using SPSS software. The results of 621 questionnaires from patients aged over 30 years showed 37% had memory concerns, 6% recalled having a memory test, 52% would like a memory test and 15% had heard about dementia from their GP. Patients receiving the intervention were significantly more likely to be aware of dementia risk reduction strategies (P≤0.005). The results of 153 GP/nurse questionnaires indicated 64% thought a doctor should discuss dementia with patients despite only 21% assessing their dementia knowledge as adequate. There was no significant difference in responses between Australia and England (P≥0.05). The frequency of documentation of Mini-Mental State Examination and dementia diagnosis in computerised medical records of patients over 75 years was less than 0.01. These results demonstrate that many adult patients attending GPs have memory concerns, associate dementia with memory loss, and are receptive to information about dementia risk reduction. Most general practitioners and their nurses rate their dementia knowledge as inadequate with few testing for memory problems or discussing dementia with their patients.

A Prototype Neural Network Decision-Support Tool for the Early Diagnosis of Acute Myocardial Infarction

An application of the ARTMAP neural network model to the early diagnosis of acute myocardial infarction is described. Performance results are given for 10 individual ARTMAP networks, and for combinations of the networks using “pooled” decision making (the so-called voting strategy). Category nodes are pruned from the trained networks in different ways so as to improve accuracy, sensitivity and specificity respectively. The differently pruned networks are employed in a novel “cascaded” variation of the voting strategy. This allows a partitioning of the test data into predictions with a high and a lower certainty of being correct, providing the diagnosing clinician with an indication of the reliability of an individual prediction.

A New RBFNDDA-KNN Network and Its Application to Medical Pattern Classification

In this paper, a new variant of the Radial Basis Function Network with the Dynamic Decay Adjustment algorithm (i.e., RBFNDDA) is introduced for undertaking pattern classification problems with noisy data. The RBFNDDA network is integrated with the k-nearest neighbours algorithm to form the proposed RBFNDDA-KNN model. Given a set of labelled data samples, the RBFNDDA network undergoes a constructive learning algorithm that exhibits a greedy insertion behaviour. As a result, many prototypes (hidden neurons) that represent small (with respect to a threshold) clusters of labelled data are introduced in the hidden layer. This results in a large network size. Such small prototypes can be caused by noisy data, or they can be valid representatives of small clusters of labelled data. The KNN algorithm is used to identify small prototypes that exist in the vicinity (with respect to a distance metric) of the majority of large prototypes from different classes. These small prototypes are treated as noise, and are, therefore, pruned from the network. To evaluate the effectiveness of RBFNDDA-KNN, a series of experiments using pattern classification problems in the medical domain is conducted. Benchmark and real medical data sets are experimented, and the results are compared, analysed, and discussed. The outcomes show that RBFNDDA-KNN is able to learn information with a compact network structure and to produce fast and accurate classification results.

Prevalence of hyponatremia in acute medical admissions in tropical Asia Pacific Australia.

OBJECTIVE To determine prevalence of hyponatremia in acute medical admissions in Northern Australasia. METHODS We studied 469 consecutive acute medical admissions to a hospital in Australias Far North Queensland during the colder months of June and July 2012. Prevalence of hyponatremia and its relationship with gender, age, diagnosis and prognosis in acute medical admissions were investigated. RESULTS On admission, hyponatremia (plasma sodium <136 mmol/L) was present in 39.4% of patients, with mild (130-135 mmol/L), moderate (126-129 mmol/L) and severe (<126 mmol/L) hyponatremia being present in 25.2%, 10.7% and 3.6% respectively. Overall, adding together admission hyponatremia with that developing during admission, 45.2% of patients were affected with 11.5% moderate hyponatremia cases and 4.1% severe ones. Hypokalemia and hyperkalemia were present in 17.0% and 18.1%, respectively. Overall, 275/469 patients (58.6%) presented with an electrolyte abnormality. There were significant correlations of hyponatremia with age but not with gender and in-hospital mortality. Prevalence of hyponatremia was high across all diagnostic categories. CONCLUSIONS The prevalence of hyponatremia appears to be high in the tropical North Australian population, being the highest prevalence reported amongst acute hospital admissions. The previously reported correlations with age and mortality do appear to hold good for this population with a high prevalence of electrolyte disorders. Further prospective analysis on a larger population in the area is needed to confirm our findings.

Therapeutic perspective: starting an angiotensin-converting enzyme inhibitor or angiotensin II receptor blocker in a diabetic patient

There are extensive data confirming involvement of the renin-angiotensin system in microvascular and macrovascular complications of diabetes. Blockade of the system with angiotensin-converting enzyme (ACE) inhibitors or angiotensin II receptor blockers (ARBs) is regarded as the first-line approach to managing hypertension and end-organ protection in patients with diabetes. ACE inhibitors are still the preferred agents for most patients. Dose should be lower with renal impairment unless an agent which is not excreted by the kidneys is chosen. Dose should be titrated up to the maximum tolerated to optimize end-organ protection, and intermediate-acting agents should be given in a twice daily divided dose when higher doses are used. Electrolytes should be checked before commencing, 1-2 weeks later, and after each dose increment. A modest decrease in estimated glomerular filtration rate (eGFR) and increase in creatinine often occurs with ACE inhibitors or ARBs. The agents may need to be discontinued if eGFR decreases by >15%, if creatinine increases by >20%, or if hyperkalaemia develops. Cough occurs in 5-10% of patients taking ACE inhibitor, but not with ARBs. Angioedema is probably equally common with ACE inhibitor or ARBs. It is not widely appreciated that ACE inhibitors may precipitate hypoglycaemia in patients taking glucose-lowering medication. The combination of ACE inhibitor and ARB is not routinely indicated for either hypertension or end-organ protection. While patients should not be denied the undoubted benefits of these important classes of drugs, we should also guard against their indiscriminate use in patients with diabetes. We must also ensure that patients receive appropriate counselling and monitoring.

Automatic covariate selection in logistic models for chest pain diagnosis: A new approach

A newly established method for optimizing logistic models via a minorization-majorization procedure is applied to the problem of diagnosing acute coronary syndromes (ACS). The method provides a principled approach to the selection of covariates which would otherwise require the use of a suboptimal method owing to the size of the covariate set. A strategy for building models is proposed and two models optimized for performance and for simplicity are derived via 10-fold cross-validation. These models confirm that a relatively small set of covariates including clinical and electrocardiographic features can be used successfully in this task. The performance of the models is comparable with previously published models using less principled selection methods. The models prove to be portable when tested on data gathered from three other sites. Whilst diagnostic accuracy and calibration diminishes slightly for these new settings, it remains satisfactory overall. The prospect of building predictive models that are as simple as possible for a required level of performance is valuable if data-driven decision aids are to gain wide acceptance in the clinical situation owing to the need to minimize the time taken to gather and enter data at the bedside.