No abstract

Cross-validation is crucial in preventing the over-fitting of the ANN models to the training data. Artificial neural network models are a potentially useful technique for classifying causes of death from verbal autopsies. Large training data sets are needed to improve the performance of data-derived algorithms, in particular ANN models.

A multivariate model partially predicts the outcome of carpal tunnel surgery, aids decision making, and helps to manage patient expectations. Muscle Nerve, 2018.

KEYWORDS: Clinical decision support systems, decision-making, diagnosis ContextInformation technology (IT) is now commonplace in almost every branch of healthcare. Electronic health records, eprescribing and digital medical imaging are well known to clinicians and have been implemented with varying degrees of success. 1 In addition, clinicians increasingly make use of online repositories such as PubMed and Google Scholar, 2 and specialised search engines such as FindZebra 3 to help answer clinical questions. One often overlooked set of IT tools are clinical decision support systems (CDSSs), which have been defi ned as systems that 'provide clinicians or patients with computer-generated clinical knowledge and patient-related information, intelligently fi ltered or presented at appropriate times, to enhance patient care'. 4 Such CDSSs have been the subject of academic computer science research for more than ABSTRACT 50 years 5 and offer the potential for better supported decision making by clinicians, improved compliance with medical standards and improved clinical effi ciency and safety. 6,7 Nonetheless, utilisation of CDSSs remains limited, and most healthcare IT systems do not include robust CDSS functions that can be widely employed across organisations, clinical presentations and domains. 8 Some of the challenges to implementation of CDSSs relate to the volume of high-quality data required for state-of-the-art systems, the translation of such data to machine-readable states and the mapping of CDSS processes to fi t with existing clinical workfl ows. As a result, successful implementation of CDSSs has tended to be site and domain specifi c, with major diffi culties replicating these successes more extensively throughout healthcare systems. 9 This is in contrast to commercial fi elds such as fi nance, where decision support technologies have been widely deployed. For example, risk-profi ling tools for fi nancial experts have been developed as easy-to-use programs that can assimilate information and guide users through complex fi nancial information and associated decisions tailored to individual customer needs. Healthcare decision making is signifi cantly more complex than fi nancial planning; however, some of the challenges in both domains are similar: large quantities of data need to be linked, integrated and translated to machine readable formats, and expert knowledge is required to contextualise and apply the data in a meaningful way. We discuss some reasons for the limited dissemination and adoption of CDSSs to date and refl ect on the major barriers that need to be overcome for these useful tools to be adopted more widely.

BackgroundTo investigate machine learning methods, ranging from simpler interpretable techniques to complex (non-linear) “black-box” approaches, for automated diagnosis of Age-related Macular Degeneration (AMD).MethodsData from healthy subjects and patients diagnosed with AMD or other retinal diseases were collected during routine visits via an Electronic Health Record (EHR) system. Patients’ attributes included demographics and, for each eye, presence/absence of major AMD-related clinical signs (soft drusen, retinal pigment epitelium, defects/pigment mottling, depigmentation area, subretinal haemorrhage, subretinal fluid, macula thickness, macular scar, subretinal fibrosis). Interpretable techniques known as white box methods including logistic regression and decision trees as well as less interpreitable techniques known as black box methods, such as support vector machines (SVM), random forests and AdaBoost, were used to develop models (trained and validated on unseen data) to diagnose AMD. The gold standard was confirmed diagnosis of AMD by physicians. Sensitivity, specificity and area under the receiver operating characteristic (AUC) were used to assess performance.ResultsStudy population included 487 patients (912 eyes). In terms of AUC, random forests, logistic regression and adaboost showed a mean performance of (0.92), followed by SVM and decision trees (0.90). All machine learning models identified soft drusen and age as the most discriminating variables in clinicians’ decision pathways to diagnose AMD.ConclusionsBoth black-box and white box methods performed well in identifying diagnoses of AMD and their decision pathways. Machine learning models developed through the proposed approach, relying on clinical signs identified by retinal specialists, could be embedded into EHR to provide physicians with real time (interpretable) support.Electronic supplementary materialThe online version of this article (doi:10.1186/1471-2415-15-10) contains supplementary material, which is available to authorized users.