Search

We consider model-based clustering methods for continuous, correlated data that account for external information available in the presence of mixed-type fixed covariates by proposing the MoEClust suite of models. These models allow different subsets of covariates to influence the component weights and/or component densities by modelling the parameters of the mixture as functions of the covariates. A familiar range of constrained eigen-decomposition parameterisations of the component covariance matrices are also accommodated. This paper thus addresses the equivalent aims of including covariates in Gaussian parsimonious clustering models and incorporating parsimonious covariance structures into all special cases of the Gaussian mixture of experts framework. The MoEClust models demonstrate significant improvement from both perspectives in applications to both univariate and multivariate data sets. Novel extensions to include a uniform noise component for capturing outliers and to address initialisation of the EM algorithm, model selection, and the visualisation of results are also proposed., Published in Advances in Data Analysis and Classification

Factor-analytic Gaussian mixture models are often employed as a model-based approach to clustering high-dimensional data. Typically, the numbers of clusters and latent factors must be specified in advance of model fitting, and remain fixed. The pair which optimises some model selection criterion is then chosen. For computational reasons, models in which the number of latent factors differ across clusters are rarely considered. Here the infinite mixture of infinite factor analysers (IMIFA) model is introduced. IMIFA employs a Pitman-Yor process prior to facilitate automatic inference of the number of clusters using the stick-breaking construction and a slice sampler. Furthermore, IMIFA employs multiplicative gamma process shrinkage priors to allow cluster-specific numbers of factors, automatically inferred via an adaptive Gibbs sampler. IMIFA is presented as the flagship of a family of factor-analytic mixture models, providing flexible approaches to clustering high-dimensional data. Applications to a benchmark data set, metabolomic spectral data, and a manifold learning handwritten digit example illustrate the IMIFA model and its advantageous features. These include obviating the need for model selection criteria, reducing the computational burden associated with the search of the model space, improving clustering performance by allowing cluster-specific numbers of factors, and quantifying uncertainty in the numbers of clusters and cluster-specific factors., Comment: Published in Bayesian Analysis

Background Prostate cancer (PCa) represents a significant healthcare problem. The critical clinical question is the need for a biopsy. Accurate risk stratification of patients before a biopsy can allow for individualised risk stratification thus improving clinical decision making. This study aims to build a risk calculator to inform the need for a prostate biopsy. Methods Using the clinical information of 4801 patients an Irish Prostate Cancer Risk Calculator (IPRC) for diagnosis of PCa and high grade (Gleason ≥7) was created using a binary regression model including age, digital rectal examination, family history of PCa, negative prior biopsy and Prostate-specific antigen (PSA) level as risk factors. The discrimination ability of the risk calculator is internally validated using cross validation to reduce overfitting, and its performance compared with PSA and the American risk calculator (PCPT), Prostate Biopsy Collaborative Group (PBCG) and European risk calculator (ERSPC) using various performance outcome summaries. In a subgroup of 2970 patients, prostate volume was included. Separate risk calculators including the prostate volume (IPRCv) for the diagnosis of PCa (and high-grade PCa) was created. Results IPRC area under the curve (AUC) for the prediction of PCa and high-grade PCa was 0.6741 (95% CI, 0.6591 to 0.6890) and 0.7214 (95% CI, 0.7018 to 0.7409) respectively. This significantly outperforms the predictive ability of cancer detection for PSA (0.5948), PCPT (0.6304), PBCG (0.6528) and ERSPC (0.6502) risk calculators; and also, for detecting high-grade cancer for PSA (0.6623) and PCPT (0.6804) but there was no significant improvement for PBCG (0.7185) and ERSPC (0.7140). The inclusion of prostate volume into the risk calculator significantly improved the AUC for cancer detection (AUC = 0.7298; 95% CI, 0.7119 to 0.7478), but not for high-grade cancer (AUC = 0.7256; 95% CI, 0.7017 to 0.7495). The risk calculator also demonstrated an increased net benefit on decision curve analysis. Conclusion The risk calculator developed has advantages over prior risk stratification of prostate cancer patients before the biopsy. It will reduce the number of men requiring a biopsy and their exposure to its side effects. The interactive tools developed are beneficial to translate the risk calculator into practice and allows for clarity in the clinical recommendations.

Sequence analysis is an increasingly popular approach for analysing life courses represented by ordered collections of activities experienced by subjects over time. Here, we analyse a survey data set containing information on the career trajectories of a cohort of Northern Irish youths tracked between the ages of 16 and 22. We propose a novel, model-based clustering approach suited to the analysis of such data from a holistic perspective, with the aims of estimating the number of typical career trajectories, identifying the relevant features of these patterns, and assessing the extent to which such patterns are shaped by background characteristics. Several criteria exist for measuring pairwise dissimilarities among categorical sequences. Typically, dissimilarity matrices are employed as input to heuristic clustering algorithms. The family of methods we develop instead clusters sequences directly using mixtures of exponential-distance models. Basing the models on weighted variants of the Hamming distance metric permits closed-form expressions for parameter estimation. Simultaneously allowing the component membership probabilities to depend on fixed covariates and accommodating sampling weights in the clustering process yields new insights on the Northern Irish data. In particular, we find that school examination performance is the single most important predictor of cluster membership., Comment: Published in Journal of the Royal Statistical Society: Series A (Statistics in Society)

Classifying indolent prostate cancer represents a significant clinical challenge. We investigated whether integrating data from different omic platforms could identify a biomarker panel with improved performance compared to individual platforms alone. DNA methylation, transcripts, protein and glycosylation biomarkers were assessed in a single cohort of patients treated by radical prostatectomy. Novel multiblock statistical data integration approaches were used to deal with missing data and modelled via stepwise multinomial logistic regression, or LASSO. After applying leave-one-out cross-validation to each model, the probabilistic predictions of disease type for each individual panel were aggregated to improve prediction accuracy using all available information for a given patient. Through assessment of three performance parameters of area under the curve (AUC) values, calibration and decision curve analysis, the study identified an integrated biomarker panel which predicts disease type with a high level of accuracy, with Multi AUC value of 0.91 (0.89, 0.94) and Ordinal C-Index (ORC) value of 0.94 (0.91, 0.96), which was significantly improved compared to the values for the clinical panel alone of 0.67 (0.62, 0.72) Multi AUC and 0.72 (0.67, 0.78) ORC. Biomarker integration across different omic platforms significantly improves prediction accuracy. We provide a novel multiplatform approach for the analysis, determination and performance assessment of novel panels which can be applied to other diseases. With further refinement and validation, this panel could form a tool to help inform appropriate treatment strategies impacting on patient outcome in early stage prostate cancer.

The diagnosis and treatment of prostate cancer (PCa) is a major health-care concern worldwide. This cancer can manifest itself in many distinct forms and the transition from clinically indolent PCa to the more invasive aggressive form remains poorly understood. It is now universally accepted that glycan expression patterns change with the cellular modifications that accompany the onset of tumorigenesis. The aim of this study was to investigate if differential glycosylation patterns could distinguish between indolent, significant, and aggressive PCa. Whole serum N-glycan profiling was carried out on 117 prostate cancer patients&rsquo, serum using our automated, high-throughput analysis platform for glycan-profiling which utilizes ultra-performance liquid chromatography (UPLC) to obtain high resolution separation of N-linked glycans released from the serum glycoproteins. We observed increases in hybrid, oligomannose, and biantennary digalactosylated monosialylated glycans (M5A1G1S1, M8, and A2G2S1), bisecting glycans (A2B, A2(6)BG1) and monoantennary glycans (A1), and decreases in triantennary trigalactosylated trisialylated glycans with and without core fucose (A3G3S3 and FA3G3S3) with PCa progression from indolent through significant and aggressive disease. These changes give us an insight into the disease pathogenesis and identify potential biomarkers for monitoring the PCa progression, however these need further confirmation studies.

Objective To analyse the performance of the Prostate Cancer Prevention Trial Risk Calculator (PCPT-RC) and two iterations of the European Randomised Study of Screening for Prostate Cancer (ERSPC) Risk Calculator, one of which incorporates prostate volume (ERSPC-RC) and the other of which incorporates prostate volume and the prostate health index (PHI) in a referral population (ERSPC-PHI). Patients and Methods The risk of prostate cancer (PCa) and significant PCa (Gleason score ≥7) in 2001 patients from six tertiary referral centres was calculated according to the PCPT-RC and ERSPC-RC formulae. The calculators’ predictions were analysed using the area under the receiver-operating characteristic curve (AUC), calibration plots, Hosmer–Lemeshow test for goodness of fit and decision-curve analysis. In a subset of 222 patients for whom the PHI score was available, each patient's risk was calculated as per the ERSPC-RC and ERSPC-PHI risk calculators. Results The ERSPC-RC outperformed the PCPT-RC in the prediction of PCa, with an AUC of 0.71 compared with 0.64, and also outperformed the PCPT-RC in the prediction of significant PCa (P

Objectives To analyse the clinical utility of a prediction model incorporating both clinical information and a novel biomarker, p2PSA, in order to inform the decision for prostate biopsy in an Irish cohort of men referred for prostate cancer assessment. Patients and Methods Serum isolated from 250 men from three tertiary referral centres with pre-biopsy blood draws was analysed for total prostate-specific antigen (PSA), free PSA (fPSA) and p2PSA. From this, the Prostate Health Index (PHI) score was calculated (PHI = (p2PSA/fPSA)*√tPSA). The men's clinical information was used to derive their risk according to the Prostate Cancer Prevention Trial (PCPT) risk model. Two clinical prediction models were created via multivariable regression consisting of age, family history, abnormality on digital rectal examination, previous negative biopsy and either PSA or PHI score, respectively. Calibration plots, receiver-operating characteristic (ROC) curves and decision curves were generated to assess the performance of the three models. Results The PSA model and PHI model were both well calibrated in this cohort, with the PHI model showing the best correlation between predicted probabilities and actual outcome. The areas under the ROC curve for the PHI model, PSA model and PCPT model were 0.77, 0.71 and 0.69, respectively, for the prediction of prostate cancer (PCa) and 0.79, 0.72 and 0.72, respectively, for the prediction of high grade PCa. Decision-curve analysis showed a superior net benefit of the PHI model over both the PSA model and the PCPT risk model in the diagnosis of PCa and high grade PCa over the entire range of risk probabilities. Conclusion A logical and standardized approach to the use of clinical risk factors can allow more accurate risk stratification of men under investigation for PCa. The measurement of p2PSA and the integration of this biomarker into a clinical prediction model can further increase the accuracy of risk stratification, helping to better inform the decision for prostate biopsy in a referral population.

The decision to proceed to biopsy for the diagnosis of prostate cancer in clinical practice is a difficult one. Prostate cancer risk calculators allow for a systematic approach to the use of patient information to predict a patient’s likelihood of prostate cancer. In this paper, we validate the two leading prostate cancer risk calculators, the prostate cancer prevention trial (PCPT) and the European Randomized Study of Screening for Prostate Cancer (ERSPC) in an Irish population. Data were collected for 337 men referred to one tertiary referral center in Ireland. Calibration analysis, ROC analysis and decision curve analysis were undertaken to ascertain the performance of the PCPT and the ERSPC risk calculators in this cohort. Of 337 consecutive biopsies, cancer was subsequently diagnosed in 146 men (43 %), 98 (67 %) of which were high grade. The AUC for the PCPT and ERSPC risk calculators were 0.68 and 0.66, respectively for the prediction of prostate cancer. Each calculator was sufficiently calibrated in this cohort. Decision curve analysis demonstrated a net benefit via the use of the PCPT and ERSPC risk calculators in the diagnosis of prostate cancer. The PCPT and ERSPC risk calculators achieve a statistically significant prediction of prostate cancer in this Irish population. This study provides external validation for these calculators, and therefore these tools can be used to aid in clinical decision making.

This paper describes the process of incremental pipeline filling and the phased commissioning of a real-time leak detection system for the 1768 km long BTC crude oil pipeline. Due to stringent environmental requirements, it is essential for the leak detection system to work from the moment that crude oil is introduced into the pipeline. Without any prior operational data and with the pipeline partially filled, it is challenging for the leak detection system to monitor the integrity of the pipeline throughout the whole filling process. The application of the pig tracking software to track the oil front as the crude displaces nitrogen is also discussed.Copyright © 2006 by ASME

This paper addresses the specific issues of transient leak detection in crude oil pipelines. When a leak occurs immediately after pumps are switched on or off, the pressure wave generated by the transients dominates the pressure wave that results from the leak. Traditional methods have failed to detect such leaks. Over the years, NWO has developed and implemented various leak detection systems both in-house and by commercial vendors. These systems work effectively under steady-state conditions but they are not able to detect leaks during transients. As it is likely for a leak to develop during transients, NWO has decided to have the ATMOS Pipe statistical leak detection system installed on their pipelines. This paper describes the application of this statistical system to two crude oil pipeline systems. After addressing the main difficulties of transient leaks, the field results will be presented for both steady-state and transient conditions.Copyright © 2004 by ASME