Dynamic early warning scores for predicting clinical deterioration in patients with respiratory disease.

BACKGROUND: The National Early Warning Score-2 (NEWS-2) is used to detect patient deterioration in UK hospitals but fails to take account of the detailed granularity or temporal trends in clinical observations. We used data-driven methods to develop dynamic early warning scores (DEWS) to address these deficiencies, and tested their accuracy in patients with respiratory disease for predicting (1) death or intensive care unit admission, occurring within 24 h (D/ICU), and (2) clinically significant deterioration requiring urgent intervention, occurring within 4 h (CSD). METHODS: Clinical observations data were extracted from electronic records for 31,590 respiratory in-patient episodes from April 2015 to December 2020 at a large acute NHS Trust. The timing of D/ICU was extracted for all episodes. 1100 in-patient episodes were annotated manually to record the timing of CSD, defined as a specific event requiring a change in treatment. Time series features were entered into logistic regression models to derive DEWS for each of the clinical outcomes. Area under the receiver operating characteristic curve (AUROC) was the primary measure of model accuracy. RESULTS: AUROC (95% confidence interval) for predicting D/ICU was 0.857 (0.852-0.862) for NEWS-2 and 0.906 (0.899-0.914) for DEWS in the validation data. AUROC for predicting CSD was 0.829 (0.817-0.842) for NEWS-2 and 0.877 (0.862-0.892) for DEWS. NEWS-2 ≥ 5 had sensitivity of 88.2% and specificity of 54.2% for predicting CSD, while DEWS ≥ 0.021 had higher sensitivity of 93.6% and approximately the same specificity of 54.3% for the same outcome. Using these cut-offs, 315 out of 347 (90.8%) CSD events were detected by both NEWS-2 and DEWS, at the time of the event or within the previous 4 h; 12 (3.5%) were detected by DEWS but not by NEWS-2, while 4 (1.2%) were detected by NEWS-2 but not by DEWS; 16 (4.6%) were not detected by either scoring system. CONCLUSION: We have developed DEWS that display greater accuracy than NEWS-2 for predicting clinical deterioration events in patients with respiratory disease. Prospective validation studies are required to assess whether DEWS can be used to reduce missed deteriorations and false alarms in real-life clinical settings.

LMISA: A lightweight multi-modality image segmentation network via domain adaptation using gradient magnitude and shape constraint.

In medical image segmentation, supervised machine learning models trained using one image modality (e.g. computed tomography (CT)) are often prone to failure when applied to another image modality (e.g. magnetic resonance imaging (MRI)) even for the same organ. This is due to the significant intensity variations of different image modalities. In this paper, we propose a novel end-to-end deep neural network to achieve multi-modality image segmentation, where image labels of only one modality (source domain) are available for model training and the image labels for the other modality (target domain) are not available. In our method, a multi-resolution locally normalized gradient magnitude approach is firstly applied to images of both domains for minimizing the intensity discrepancy. Subsequently, a dual task encoder-decoder network including image segmentation and reconstruction is utilized to effectively adapt a segmentation network to the unlabeled target domain. Additionally, a shape constraint is imposed by leveraging adversarial learning. Finally, images from the target domain are segmented, as the network learns a consistent latent feature representation with shape awareness from both domains. We implement both 2D and 3D versions of our method, in which we evaluate CT and MRI images for kidney and cardiac tissue segmentation. For kidney, a public CT dataset (KiTS19, MICCAI 2019) and a local MRI dataset were utilized. The cardiac dataset was from the Multi-Modality Whole Heart Segmentation (MMWHS) challenge 2017. Experimental results reveal that our proposed method achieves significantly higher performance with a much lower model complexity in comparison with other state-of-the-art methods. More importantly, our method is also capable of producing superior segmentation results than other methods for images of an unseen target domain without model retraining. The code is available at GitHub (https://github.com/MinaJf/LMISA) to encourage method comparison and further research.

Machine learning can predict disease manifestations and outcomes in lymphangioleiomyomatosis.

BACKGROUND: Lymphangioleiomyomatosis (LAM) is a rare multisystem disease with variable clinical manifestations and differing rates of progression that make management decisions and giving prognostic advice difficult. We used machine learning to identify clusters of associated features which could be used to stratify patients and predict outcomes in individuals. PATIENTS AND METHODS: Using unsupervised machine learning we generated patient clusters using data from 173 women with LAM from the UK and 186 replication subjects from the US National Heart, Lung, and Blood Institute (NHLBI) LAM registry. Prospective outcomes were associated with cluster results. RESULTS: Two- and three-cluster models were developed. A three-cluster model separated a large group of subjects presenting with dyspnoea or pneumothorax from a second cluster with a high prevalence of angiomyolipoma symptoms (p=0.0001) and tuberous sclerosis complex (TSC) (p=0.041). Patients in the third cluster were older, never presented with dyspnoea or pneumothorax (p=0.0001) and had better lung function. Similar clusters were reproduced in the NHLBI cohort. Assigning patients to clusters predicted prospective outcomes: in a two-cluster model the future risk of pneumothorax was 3.3 (95% CI 1.7-5.6)-fold greater in cluster 1 than cluster 2 (p=0.0002). Using the three-cluster model, the need for intervention for angiomyolipoma was lower in clusters 2 and 3 than cluster 1 (p<0.00001). In the NHLBI cohort, the incidence of death or lung transplant was much lower in clusters 2 and 3 (p=0.0045). CONCLUSIONS: Machine learning has identified clinically relevant clusters associated with complications and outcome. Assigning individuals to clusters could improve decision making and prognostic information for patients.

The Barriers and Motivators to Using Human Tissues for Research: The Views of UK-Based Biomedical Researchers.

Introduction: The use of human-derived samples is vital to numerous areas of biological and medical research. Despite this, researchers often find or anticipate difficulty in sourcing samples. There are ongoing efforts to increase the visibility and accessibility of UK human tissue biobanking, but minimal (if any) research on the reasons behind researchers' choice of sample source has been undertaken. We have analyzed UK researchers' motivations on using their preferred sample sources and their perceived barriers to human sample use. Methods: The study was based on an online survey of academic and industry researchers, followed by focus groups, with participants across the United Kingdom. Both the survey and focus groups probed participants' views on the barriers to finding and using human samples in research. Results: One hundred ninety-eight academic and industry researchers completed the survey on their human sample use, and five focus groups consisting of 21 total participants took place. The top cited reasons for choosing sources included the availability of linked clinical data (40%), the geographical location of the resource (39%), and preexisting collaboration (33%). Focus group participants highlighted their strong preference for local or known sample sources, which were preferred because additional scientific and logistical input could be obtained for their work and they were more confident that the samples would be of good quality. Discussion: We found that there were significant perceptions of governance barriers to sample access. As a consequence, researchers preferred local and known suppliers because of the perception that these could assist with the governance, would be reliable, and able to provide the additional support they required. Equally, data availability was a major contributor to the selection of a new source of samples. These observations are of significant value to those seeking to improve the access to existing sample resources via online discovery tools.

Young adults' attitudes to sharing whole-genome sequencing information: a university-based survey.

BACKGROUND: Genomic services are increasingly accessible to young adults starting their independent lives with responsibility for their self-care, yet their attitudes to sharing genomic information remain under-researched. This study explored attitudes of university-based 18-25 year-olds towards sharing personal whole-genome sequencing (WGS) information with relatives. METHODS: We surveyed 112 young adults. Hypotheses were tested regarding the relationships between their preferences for sharing personal WGS information with relatives and factors including their gender, previous genetics-specific education, general educational attainment level and current study in a science, technology, engineering, maths or medicine (STEMM) field. RESULTS: Most participants were positive about both their intention to share their WGS results with their parents and siblings, and their desire to know their relatives' results. Being female and having a university-level genetics education were consistently positively correlated with intention to share one's results with parents and with siblings as well as the desire to know relatives' results. Additionally, females who had undertaken a genetics course at university had significantly greater intentions and desires than females who had not. Lower general educational attainment was related to a lower intention to share with siblings. Participants who were in a STEMM field had a greater desire to know their relatives' results. CONCLUSIONS: Participants' gender and prior genetics education were consistently related to their intentions to share WGS results with relatives and their desire to know relatives' results. Educational attainment was found to be positively correlated with intention to share with siblings. Being in a STEMM field was related to participants' desire to know their relatives' results. These findings indicate that gender and genetics education are particularly important influencers on young adults' stated sharing preferences. More research is required to examine the dependent variables studied to further understand their influence on attitudes to sharing WGS results. These findings are particularly interesting for information provision and support before genomic sequencing and post-results to improve the outcomes for individuals and their relatives.

Modeling and control of operator functional state in a unified framework of fuzzy inference petri nets.

BACKGROUND AND OBJECTIVE: In human-machine (HM) hybrid control systems, human operator and machine cooperate to achieve the control objectives. To enhance the overall HM system performance, the discrete manual control task-load by the operator must be dynamically allocated in accordance with continuous-time fluctuation of psychophysiological functional status of the operator, so-called operator functional state (OFS). The behavior of the HM system is hybrid in nature due to the co-existence of discrete task-load (control) variable and continuous operator performance (system output) variable. METHODS: Petri net is an effective tool for modeling discrete event systems, but for hybrid system involving discrete dynamics, generally Petri net model has to be extended. Instead of using different tools to represent continuous and discrete components of a hybrid system, this paper proposed a method of fuzzy inference Petri nets (FIPN) to represent the HM hybrid system comprising a Mamdani-type fuzzy model of OFS and a logical switching controller in a unified framework, in which the task-load level is dynamically reallocated between the operator and machine based on the model-predicted OFS. Furthermore, this paper used a multi-model approach to predict the operator performance based on three electroencephalographic (EEG) input variables (features) via the Wang-Mendel (WM) fuzzy modeling method. The membership function parameters of fuzzy OFS model for each experimental participant were optimized using artificial bee colony (ABC) evolutionary algorithm. Three performance indices, RMSE, MRE, and EPR, were computed to evaluate the overall modeling accuracy. RESULTS: Experiment data from six participants are analyzed. The results show that the proposed method (FIPN with adaptive task allocation) yields lower breakdown rate (from 14.8% to 3.27%) and higher human performance (from 90.30% to 91.99%). CONCLUSION: The simulation results of the FIPN-based adaptive HM (AHM) system on six experimental participants demonstrate that the FIPN framework provides an effective way to model and regulate/optimize the OFS in HM hybrid systems composed of continuous-time OFS model and discrete-event switching controller.

Can machine-learning improve cardiovascular risk prediction using routine clinical data?

BACKGROUND: Current approaches to predict cardiovascular risk fail to identify many people who would benefit from preventive treatment, while others receive unnecessary intervention. Machine-learning offers opportunity to improve accuracy by exploiting complex interactions between risk factors. We assessed whether machine-learning can improve cardiovascular risk prediction. METHODS: Prospective cohort study using routine clinical data of 378,256 patients from UK family practices, free from cardiovascular disease at outset. Four machine-learning algorithms (random forest, logistic regression, gradient boosting machines, neural networks) were compared to an established algorithm (American College of Cardiology guidelines) to predict first cardiovascular event over 10-years. Predictive accuracy was assessed by area under the 'receiver operating curve' (AUC); and sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) to predict 7.5% cardiovascular risk (threshold for initiating statins). FINDINGS: 24,970 incident cardiovascular events (6.6%) occurred. Compared to the established risk prediction algorithm (AUC 0.728, 95% CI 0.723-0.735), machine-learning algorithms improved prediction: random forest +1.7% (AUC 0.745, 95% CI 0.739-0.750), logistic regression +3.2% (AUC 0.760, 95% CI 0.755-0.766), gradient boosting +3.3% (AUC 0.761, 95% CI 0.755-0.766), neural networks +3.6% (AUC 0.764, 95% CI 0.759-0.769). The highest achieving (neural networks) algorithm predicted 4,998/7,404 cases (sensitivity 67.5%, PPV 18.4%) and 53,458/75,585 non-cases (specificity 70.7%, NPV 95.7%), correctly predicting 355 (+7.6%) more patients who developed cardiovascular disease compared to the established algorithm. CONCLUSIONS: Machine-learning significantly improves accuracy of cardiovascular risk prediction, increasing the number of patients identified who could benefit from preventive treatment, while avoiding unnecessary treatment of others.

A new accuracy measure based on bounded relative error for time series forecasting.

Many accuracy measures have been proposed in the past for time series forecasting comparisons. However, many of these measures suffer from one or more issues such as poor resistance to outliers and scale dependence. In this paper, while summarising commonly used accuracy measures, a special review is made on the symmetric mean absolute percentage error. Moreover, a new accuracy measure called the Unscaled Mean Bounded Relative Absolute Error (UMBRAE), which combines the best features of various alternative measures, is proposed to address the common issues of existing measures. A comparative evaluation on the proposed and related measures has been made with both synthetic and real-world data. The results indicate that the proposed measure, with user selectable benchmark, performs as well as or better than other measures on selected criteria. Though it has been commonly accepted that there is no single best accuracy measure, we suggest that UMBRAE could be a good choice to evaluate forecasting methods, especially for cases where measures based on geometric mean of relative errors, such as the geometric mean relative absolute error, are preferred.

Nottingham Prognostic Index Plus: Validation of a clinical decision making tool in breast cancer in an independent series.

The Nottingham Prognostic Index Plus (NPI+) is a clinical decision making tool in breast cancer (BC) that aims to provide improved patient outcome stratification superior to the traditional NPI. This study aimed to validate the NPI+ in an independent series of BC. Eight hundred and eighty five primary early stage BC cases from Edinburgh were semi-quantitatively assessed for 10 biomarkers [Estrogen Receptor (ER), Progesterone Receptor (PgR), cytokeratin (CK) 5/6, CK7/8, epidermal growth factor receptor (EGFR), HER2, HER3, HER4, p53, and Mucin 1] using immunohistochemistry and classified into biological classes by fuzzy logic-derived algorithms previously developed in the Nottingham series. Subsequently, NPI+ Prognostic Groups (PGs) were assigned for each class using bespoke NPI-like formulae, previously developed in each NPI+ biological class of the Nottingham series, utilising clinicopathological parameters: number of positive nodes, pathological tumour size, stage, tubule formation, nuclear pleomorphism and mitotic counts. Biological classes and PGs were compared between the Edinburgh and Nottingham series using Cramer's V and their role in patient outcome prediction using Kaplan-Meier curves and tested using Log Rank. The NPI+ biomarker panel classified the Edinburgh series into seven biological classes similar to the Nottingham series (p > 0.01). The biological classes were significantly associated with patient outcome (p < 0.001). PGs were comparable in predicting patient outcome between series in Luminal A, Basal p53 altered, HER2+/ER+ tumours (p > 0.01). The good PGs were similarly validated in Luminal B, Basal p53 normal, HER2+/ER- tumours and the poor PG in the Luminal N class (p > 0.01). Due to small patient numbers assigned to the remaining PGs, Luminal N, Luminal B, Basal p53 normal and HER2+/ER- classes could not be validated. This study demonstrates the reproducibility of NPI+ and confirmed its prognostic value in an independent cohort of primary BC. Further validation in large randomised controlled trial material is warranted.

KI67 and DLX2 predict increased risk of metastasis formation in prostate cancer-a targeted molecular approach.

BACKGROUND: There remains a need to identify and validate biomarkers for predicting prostate cancer (CaP) outcomes using robust and routinely available pathology techniques to identify men at most risk of premature death due to prostate cancer. Previous immunohistochemical studies suggest the proliferation marker Ki67 might be a predictor of survival, independently of PSA and Gleason score. We performed a validation study of Ki67 as a marker of survival and disease progression and compared its performance against another candidate biomarker, DLX2, selected using artificial neural network analysis. METHODS: A tissue microarray (TMA) was constructed from transurethral resected prostatectomy histology samples (n=192). Artificial neural network analysis was used to identify candidate markers conferring increased risk of death and metastasis in a public cDNA array. Immunohistochemical analysis of the TMA was carried out and univariate and multivariate tests performed to explore the association of tumour protein levels of Ki67 and DLX2 with time to death and metastasis. RESULTS: Univariate analysis demonstrated Ki67 as predictive of CaP-specific survival (DSS; P=0.022), and both Ki67 (P=0.025) and DLX2 (P=0.001) as predictive of future metastases. Multivariate analysis demonstrated Ki67 as independent of PSA, Gleason score and D'Amico risk category for DSS (HR=2.436, P=0.029) and both Ki67 (HR=3.296, P=0.023) and DLX2 (HR=3.051, P=0.003) as independent for future metastases. CONCLUSIONS: High Ki67 expression is only present in 6.8% of CaP patients and is predictive of reduced survival and increased risk of metastasis, independent of PSA, Gleason score and D'Amico risk category. DLX2 is a novel marker of increased metastasis risk found in 73% patients and 8.2% showed co-expression with a high Ki67 score. Two cancer cell proliferation markers, Ki67 and DLX2, may be able to inform clinical decision-making when identifying patients for active surveillance.

A hybrid model for automatic identification of risk factors for heart disease.

Coronary artery disease (CAD) is the leading cause of death in both the UK and worldwide. The detection of related risk factors and tracking their progress over time is of great importance for early prevention and treatment of CAD. This paper describes an information extraction system that was developed to automatically identify risk factors for heart disease in medical records while the authors participated in the 2014 i2b2/UTHealth NLP Challenge. Our approaches rely on several nature language processing (NLP) techniques such as machine learning, rule-based methods, and dictionary-based keyword spotting to cope with complicated clinical contexts inherent in a wide variety of risk factors. Our system achieved encouraging performance on the challenge test data with an overall micro-averaged F-measure of 0.915, which was competitive to the best system (F-measure of 0.927) of this challenge task.

Semi-Supervised Fuzzy Clustering with Feature Discrimination.

Semi-supervised clustering algorithms are increasingly employed for discovering hidden structure in data with partially labelled patterns. In order to make the clustering approach useful and acceptable to users, the information provided must be simple, natural and limited in number. To improve recognition capability, we apply an effective feature enhancement procedure to the entire data-set to obtain a single set of features or weights by weighting and discriminating the information provided by the user. By taking pairwise constraints into account, we propose a semi-supervised fuzzy clustering algorithm with feature discrimination (SFFD) incorporating a fully adaptive distance function. Experiments on several standard benchmark data sets demonstrate the effectiveness of the proposed method.

Automatic detection of protected health information from clinic narratives.

This paper presents a natural language processing (NLP) system that was designed to participate in the 2014 i2b2 de-identification challenge. The challenge task aims to identify and classify seven main Protected Health Information (PHI) categories and 25 associated sub-categories. A hybrid model was proposed which combines machine learning techniques with keyword-based and rule-based approaches to deal with the complexity inherent in PHI categories. Our proposed approaches exploit a rich set of linguistic features, both syntactic and word surface-oriented, which are further enriched by task-specific features and regular expression template patterns to characterize the semantics of various PHI categories. Our system achieved promising accuracy on the challenge test data with an overall micro-averaged F-measure of 93.6%, which was the winner of this de-identification challenge.

A supervised adverse drug reaction signalling framework imitating Bradford Hill's causality considerations.

Big longitudinal observational medical data potentially hold a wealth of information and have been recognised as potential sources for gaining new drug safety knowledge. Unfortunately there are many complexities and underlying issues when analysing longitudinal observational data. Due to these complexities, existing methods for large-scale detection of negative side effects using observational data all tend to have issues distinguishing between association and causality. New methods that can better discriminate causal and non-causal relationships need to be developed to fully utilise the data. In this paper we propose using a set of causality considerations developed by the epidemiologist Bradford Hill as a basis for engineering features that enable the application of supervised learning for the problem of detecting negative side effects. The Bradford Hill considerations look at various perspectives of a drug and outcome relationship to determine whether it shows causal traits. We taught a classifier to find patterns within these perspectives and it learned to discriminate between association and causality. The novelty of this research is the combination of supervised learning and Bradford Hill's causality considerations to automate the Bradford Hill's causality assessment. We evaluated the framework on a drug safety gold standard known as the observational medical outcomes partnership's non-specified association reference set. The methodology obtained excellent discrimination ability with area under the curves ranging between 0.792 and 0.940 (existing method optimal: 0.73) and a mean average precision of 0.640 (existing method optimal: 0.141). The proposed features can be calculated efficiently and be readily updated, making the framework suitable for big observational data.

Juxtaposition of system dynamics and agent-based simulation for a case study in immunosenescence.

Advances in healthcare and in the quality of life significantly increase human life expectancy. With the aging of populations, new un-faced challenges are brought to science. The human body is naturally selected to be well-functioning until the age of reproduction to keep the species alive. However, as the lifespan extends, unseen problems due to the body deterioration emerge. There are several age-related diseases with no appropriate treatment; therefore, the complex aging phenomena needs further understanding. It is known that immunosenescence is highly correlated to the negative effects of aging. In this work we advocate the use of simulation as a tool to assist the understanding of immune aging phenomena. In particular, we are comparing system dynamics modelling and simulation (SDMS) and agent-based modelling and simulation (ABMS) for the case of age-related depletion of naive T cells in the organism. We address the following research questions: Which simulation approach is more suitable for this problem? Can these approaches be employed interchangeably? Is there any benefit of using one approach compared to the other? Results show that both simulation outcomes closely fit the observed data and existing mathematical model; and the likely contribution of each of the naive T cell repertoire maintenance method can therefore be estimated. The differences observed in the outcomes of both approaches are due to the probabilistic character of ABMS contrasted to SDMS. However, they do not interfere in the overall expected dynamics of the populations. In this case, therefore, they can be employed interchangeably, with SDMS being simpler to implement and taking less computational resources.

Signalling paediatric side effects using an ensemble of simple study designs.

BACKGROUND: Children are frequently prescribed medication 'off-label', meaning there has not been sufficient testing of the medication to determine its safety or effectiveness. The main reason this safety knowledge is lacking is due to ethical restrictions that prevent children from being included in the majority of clinical trials. OBJECTIVE: The objective of this paper is to investigate whether an ensemble of simple study designs can be implemented to signal acutely occurring side effects effectively within the paediatric population by using historical longitudinal data. The majority of pharmacovigilance techniques are unsupervised, but this research presents a supervised framework. METHODS: Multiple measures of association are calculated for each drug and medical event pair and these are used as features that are fed into a classifier to determine the likelihood of the drug and medical event pair corresponding to an adverse drug reaction. The classifier is trained using known adverse drug reactions or known non-adverse drug reaction relationships. RESULTS: The novel ensemble framework obtained a false positive rate of 0.149, a sensitivity of 0.547 and a specificity of 0.851 when implemented on a reference set of drug and medical event pairs. The novel framework consistently outperformed each individual simple study design. CONCLUSION: This research shows that it is possible to exploit the mechanism of causality and presents a framework for signalling adverse drug reactions effectively.

A novel semisupervised algorithm for rare prescription side effect discovery.

Drugs are frequently prescribed to patients with the aim of improving each patient's medical state, but an unfortunate consequence of most prescription drugs is the occurrence of undesirable side effects. Side effects that occur in more than one in a thousand patients are likely to be signaled efficiently by current drug surveillance methods, however, these same methods may take decades before generating signals for rarer side effects, risking medical morbidity or mortality in patients prescribed the drug while the rare side effect is undiscovered. In this paper, we propose a novel computational metaanalysis framework for signaling rare side effects that integrates existing methods, knowledge from the web,metric learning, and semisupervised clustering. The novel framework was able to signal many known rare and serious side effects for the selection of drugs investigated, such as tendon rupture when prescribed Ciprofloxacin or Levofloxacin, renal failure with Naproxen and depression associated with Rimonabant. Furthermore, for the majority of the drugs investigated it generated signals for rare side effects at a more stringent signaling threshold than existing methods and shows the potential to become a fundamental part of post marketing surveillance to detect rare side effects.

A quantifier-based fuzzy classification system for breast cancer patients.

OBJECTIVES: Recent studies of breast cancer data have identified seven distinct clinical phenotypes (groups) using immunohistochemical analysis and a range of different clustering techniques. Consensus between unsupervised classification algorithms has been successfully used to categorise patients into these specific groups, but often at the expenses of not classifying the whole set. It is known that fuzzy methodologies can provide linguistic based classification rules. The objective of this study was to investigate the use of fuzzy methodologies to create an easy to interpret set of classification rules, capable of placing the large majority of patients into one of the specified groups. MATERIALS AND METHODS: In this paper, we extend a data-driven fuzzy rule-based system for classification purposes (called 'fuzzy quantification subsethood-based algorithm') and combine it with a novel class assignment procedure. The whole approach is then applied to a well characterised breast cancer dataset consisting of ten protein markers for over 1000 patients to refine previously identified groups and to present clinicians with a linguistic ruleset. A range of statistical approaches was used to compare the obtained classes to previously obtained groupings and to assess the proportion of unclassified patients. RESULTS: A rule set was obtained from the algorithm which features one classification rule per class, using labels of High, Low or Omit for each biomarker, to determine the most appropriate class for each patient. When applied to the whole set of patients, the distribution of the obtained classes had an agreement of 0.9 when assessed using Kendall's Tau with the original reference class distribution. In doing so, only 38 patients out of 1073 remain unclassified, representing a more clinically usable class assignment algorithm. CONCLUSION: The fuzzy algorithm provides a simple to interpret, linguistic rule set which classifies over 95% of breast cancer patients into one of seven clinical groups.

Characteristics of basal cytokeratin expression in breast cancer.

Breast cancer is recognised to be a heterogeneous disease and the second most common cause of morbidity and mortality worldwide in women. Basal-like breast cancer (BLBC) is associated with aggressive characteristics including development of recurrent disease and reduced survival. BLBC has been defined in some studies as tumours lacking both oestrogen receptor and progesterone receptor protein expression. Gene expression studies have shown that these tumours are also associated with expression of basal-type cytokeratins, the phenotypic patterns of basal cytokeratin expression in BLBC have not been widely studied. A well-characterised series of 995 invasive breast cancers with a long-term follow up were investigated using immunohistochemical staining for four basal cytokeratins (CK5, CK5/6, CK14 and CK17). The data were analysed using univariate and clustering analysis. As a result BLBC, as defined by negativity for ER and HER2 showed variable positivity for basal cytokeratin expression: 61.7 % CK5, 50.5 % CK5/6, 24.2 % CK14 and 23 % CK17. These characteristics were associated with poor outcome characteristics including high histological grade, mitosis, pleomorphism and tumour size >1.5 cm. CK5 positivity was more associated with ER(-), PgR(-), TN and double ER(-)PgR(-), than the other cytokeratins. Four different clusters of basal cytokeratin expression patterns were identified: (1) negativity for all basal cytokeratins, (2) CK5(+)/CK17(-), (3) CK5(-)/CK17(+) and (4) CK5(+)/CK17(+). These patterns of basal cytokeratin expression associated with differences in patient outcome, clusters 1 and 3 showed better outcomes than cluster 4 and 2, with cluster 2 having the poorest prognosis. In conclusion, four basal cytokeratin expression patterns were identified in human breast cancer using unsupervised clustering analysis and these patterns are associated with differences in patient outcome.

Using rule-based machine learning for candidate disease gene prioritization and sample classification of cancer gene expression data.

Microarray data analysis has been shown to provide an effective tool for studying cancer and genetic diseases. Although classical machine learning techniques have successfully been applied to find informative genes and to predict class labels for new samples, common restrictions of microarray analysis such as small sample sizes, a large attribute space and high noise levels still limit its scientific and clinical applications. Increasing the interpretability of prediction models while retaining a high accuracy would help to exploit the information content in microarray data more effectively. For this purpose, we evaluate our rule-based evolutionary machine learning systems, BioHEL and GAssist, on three public microarray cancer datasets, obtaining simple rule-based models for sample classification. A comparison with other benchmark microarray sample classifiers based on three diverse feature selection algorithms suggests that these evolutionary learning techniques can compete with state-of-the-art methods like support vector machines. The obtained models reach accuracies above 90% in two-level external cross-validation, with the added value of facilitating interpretation by using only combinations of simple if-then-else rules. As a further benefit, a literature mining analysis reveals that prioritizations of informative genes extracted from BioHEL's classification rule sets can outperform gene rankings obtained from a conventional ensemble feature selection in terms of the pointwise mutual information between relevant disease terms and the standardized names of top-ranked genes.