Informatics
mainstream, our ability to find and interpret new information patterns from this complex data and make more reliable and accurate predictions of clinical outcomes has improved drastically5-7. Machine learning (ML) is one of the methods to
make accurate predictions of future outcomes via pattern recognition to improve patient selection, eg different rates of disease progression. It also helps provide predictive long-term outcomes around safety and efficacy and reduces the time and cost of clinical trials which could vastly improve the drug development process5,8. For example, a recent study reported how ML helps detect drug effects that would be missed entirely by conventional sta- tistical tests9. Many ML techniques have also been adapted to predict survival or time to progression, but they have not been scalable in performance or interpretation. Now, using Deep Learning (DL) techniques, cog-
nitive abstraction can be leveraged to identify actionable biomarkers and predict clinical out- comes. The application of such techniques finally enables us to achieve the full potential of siloed data and helps us answer questions that were pre- viously unimaginable. In addition to helping us make more informed
decisions around clinical trial cohort selection, ML can also assist with data preparation such as automating large parts of data curation which is critical to achieving harmonisation across different datasets in trial datasets. Missing historical data can seriously compromise inferences from ran- domised clinical trials, but ML applications help with imputation of such missing data. ML can also be used on real world evidence data
combined from various sources such as electronic medical records, insurance claims data, prescription data, etc to compare patient populations enrolled in clinical trials for a specific disease. It can also iden- tify potential adverse events that may be correlated to certain unknown subgroup populations10.
Best practices for leveraging ML Although the opportunities of ML applications seem limitless, there are also challenges analysing clinical trial data with ML. While ML success in processing large amounts
of structured or image-based data, eg high content screening, is evident, there are many moving parts to clinical trial data, especially segregated or embargoed patient data. These acute challenges increase the complexity for life science companies seeking to leverage ML in Pharma6,8. This calls for best practices to be employed.
Some include: Drug Discovery World Fall 2019
Creating cleaner data Good quality data remains a foundation of strong drug research and development. Since ML applica- tions train on data to improve their predictive mod- elling capabilities, it is imperative that the data we intend to use is high quality, something which is not always readily available in historical datasets. Messy data (eg inaccurate, incomplete, improperly formatted or duplicated) can significantly distort predictive models. Therefore, organisations need to consider deploying automated tools that not only address data cleansing but also handle data normal- isation to make sure that clean and informative data is being fed into ML algorithms.
“A particular biomarker, for example, can be used to identify appro-
priate candidates for a clinical trial, such as those patients likely to respond to treatment. This can make it easier and faster to recruit patients and may result in a shorter time for drug approval.” Janet Woodcock, Director, US FDA Center for Drug Evaluation and Research
Providing readable/processable data ML applications are reliant on data that is machine readable/processable. Steps to make data machine readable could include, for example, preparing text data by removing words (tokenisation) which is not trivial when it comes to extracting the right data from medical notes or case report forms (CRFs). Appropriate data parsing, or using curation tools that can automate this process, should also be a pri- mary step in any AI-readiness strategy11.
Building an appropriate data infrastructure One of the biggest considerations for running ML applications is the need to have the right infrastruc- ture in place. This involves building an infrastructure that can dynamically scale storage capabilities as data volume grows. It also needs to provide ample computing resources including CPUs and GPUs to support performance and ensure proper data access mechanisms for successful, secure and efficient deliv- ery of data to appropriate users in the organisation. Furthermore, it is important to have high network bandwidth and low-latency work to enable ML applications to operate at an enterprise level11.
Committing to change management Enabling meaningful change and ensuring stake- holders adapt to and embrace new technologies can be challenging. Roadblocks range from solutions being rendered too complex or unusable by the
11
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68
