James Weatherall

Taming EHR data: Using semantic similarity to reduce dimensionality

Medical care data is a valuable resource that can be used for many purposes including managing and planning for future health needs as well as clinical research. However, the heterogeneity and complexity of medical data can be an obstacle in applying data mining techniques. Much of the potential value of this data therefore goes untapped. In this paper we have developed a methodology that reduces the dimensionality of primary care data, in order to make it more amenable to visualisation, mining and clustering. The methodology involves employing a combination of ontology-based semantic similarity and principal component analysis (PCA) to map the data into an appropriate and informative low dimensional space. Throughout the study, we had access to anonymised patient data from primary care in Salford, UK. The results of our application of this methodology show that diagnosis codes in primary care data can be used to map patients into an informative low dimensional space, which in turn provides the opportunity to support further data exploration and medical hypothesis formulation.

It’s a long shot, but it just might work! Perspectives on the future of medicine

What does the future of medicine hold? We asked six researchers to share their most ambitious and optimistic views of the future, grounded in the present but looking out a decade or more from now to consider what’s possible. They paint a picture of a connected and data-driven world in which patient value, patient feedback, and patient empowerment shape a continually learning system that ensures each patient’s experience contributes to the improved outcome of every patient like them, whether it be through clinical trials, data from consumer devices, hacking their medical devices, or defining value in thoughtful new ways.

The Tendril Plot—a novel visual summary of the incidence, significance and temporal aspects of adverse events in clinical trials

Background In contrast to efficacy, safety hypotheses of clinical trials are not always pre-specified, and therefore, the safety interpretation work of a trial tends to be more exploratory, often reactive, and the analysis more statistically and graphically challenging. Methods We introduce a new means of visualizing the adverse event data across an entire clinical trial. Results The approach overcomes some of the current limitations of adverse event analysis and streamlines the way safety data can be explored, interpreted and analyzed. Using a phase II study, we describe and exemplify how the tendril plot effectively summarizes the time-resolved safety profile of two treatment arms in a single plot and how that can provide scientists with a trial safety overview that can support medical decision making. Conclusion To our knowledge, the tendril plot is the only way to graphically show important treatment differences with preserved temporal information, across an entire clinical trial, in a single view.

Distinguishing prognostic and predictive biomarkers: An information theoretic approach

Abstract Motivation The identification of biomarkers to support decision-making is central to personalized medicine, in both clinical and research scenarios. The challenge can be seen in two halves: identifying predictive markers, which guide the development/use of tailored therapies; and identifying prognostic markers, which guide other aspects of care and clinical trial planning, i.e. prognostic markers can be considered as covariates for stratification. Mistakenly assuming a biomarker to be predictive, when it is in fact largely prognostic (and vice-versa) is highly undesirable, and can result in financial, ethical and personal consequences. We present a framework for data-driven ranking of biomarkers on their prognostic/predictive strength, using a novel information theoretic method. This approach provides a natural algebra to discuss and quantify the individual predictive and prognostic strength, in a self-consistent mathematical framework. Results Our contribution is a novel procedure, INFO+, which naturally distinguishes the prognostic versus predictive role of each biomarker and handles higher order interactions. In a comprehensive empirical evaluation INFO+ outperforms more complex methods, most notably when noise factors dominate, and biomarkers are likely to be falsely identified as predictive, when in fact they are just strongly prognostic. Furthermore, we show that our methods can be 1–3 orders of magnitude faster than competitors, making it useful for biomarker discovery in ‘big data’ scenarios. Finally, we apply our methods to identify predictive biomarkers on two real clinical trials, and introduce a new graphical representation that provides greater insight into the prognostic and predictive strength of each biomarker. Availability and implementation R implementations of the suggested methods are available at https://github.com/sechidis. Supplementary information Supplementary data are available at Bioinformatics online.

Disentangling Prognostic and Predictive Biomarkers Through Mutual Information

We study information theoretic methods for ranking biomarkers. In clinical trials, there are two, closely related, types of biomarkers: predictive and prognostic, and disentangling them is a key challenge. Our first step is to phrase biomarker ranking in terms of optimizing an information theoretic quantity. This formalization of the problem will enable us to derive rankings of predictive/prognostic biomarkers, by estimating different, high dimensional, conditional mutual information terms. To estimate these terms, we suggest efficient low dimensional approximations. Finally, we introduce a new visualisation tool that captures the prognostic and the predictive strength of a set of biomarkers. We believe this representation will prove to be a powerful tool in biomarker discovery.

Weak Classifiers Performance Measure in Handling Noisy Clinical Trial Data

Most research concluded that machine learning performance is better when dealing with cleaned dataset compared to dirty dataset. In this paper, we experimented three weak or base machine learning classifiers: Decision Table, Naive Bayes and k-Nearest Neighbor to see their performance on real-world, noisy and messy clinical trial dataset rather than employing beautifully designed dataset. We involved the clinical trial data scientist in leading us to a better data analysis exploration and enhancing the performance result evaluation. The classifiers performances were analyzed using Accuracy and Receiver Operating Characteristic (ROC), supported with sensitivity, specificity and precision values which resulted to contradiction of conclusion made by previous research. We employed pre-processing techniques such as interquartile range technique to remove the outliers and mean imputation to handle missing values and these techniques resulted to; all three classifiers work better in dirty dataset compared to imputed and clean dataset by showing highest accuracy and ROC measure. Decision Table turns out to be the best classifier when dealing with real-world noisy clinical trial.

Structured exploration of clinical trials data - finding the middle way

The traditional statistical analysis of data from clinical trials tends to follow a conservative approach centred around pre-specification, hypothesis testing and regulatory considerations. On the other hand, attempts to take a more exploratory approach are often criticised for being open-ended ‘data dredging’ exercises, lacking pre-specification, and failing to adequately control type I error. If it is possible to find a ‘middle way’ between these two extremes, we can uncover invaluable additional insights about medicines - whether in development or approved. At AstraZeneca, our Structured Exploration capability was created to do just that. Examining patient-level clinical trials data more thoroughly can: (1) refine the target population for a medicine; (2) enhance our understanding of benefit-risk; (3) reignite the development of a compound otherwise shelved as having insufficient benefit in the overall trial population. Our implementation of Structured Exploration centers around the selection of two specific data mining methods - Virtual Twins [1] and Inside-Out [2]. Virtual Twins has roots in the field of causal inference and counterfactuals, and provides a flexible framework within which to explore predictive subgroups. Inside-out turns the usual model formulation around, and instead of estimating the treatment effect on each adverse event separately, quantifies the ability of all the adverse events to classify patients to treatment.

Text Analytics for Surveillance (TAS) An Interactive Environment for Safety Literature Review

Pharmacovigilance regulations and guidelines state that literature databases should be searched at least monthly to detect safety signals from the published literature. In addition, periodic safety update reports (PSURs) should contain a summary and references from reports in the literature containing important safety findings. The volume of literature that needs to be reviewed is high, making manual review of the abstracts a resource-intensive process. Text Analytics for Surveillance (TAS) was developed as a software tool to improve the efficiency and consistency of the routine literature evaluation, tracking, and documentation process within a regulated pharmaceutical environment. Text Analytics for Surveillance uses natural language processing and includes a novel application of text analytics to assist with identifying the most relevant articles in the process of scheduled surveillance of published literature by enhancing categorized review, introducing consistency of approach, ensuring rigorous recording of activities, and aiding profile analysis. There are clear opportunities to reuse the TAS approach within other scientific and business areas where regular literature evaluation is important.