Fosbury Flop: This radically new style in the high jump introduced by the American athlete Dick Fosbury during the 1968 Olympics permanently changed the sporting event upside down (literally and figuratively). Until then, the landing surfaces for high jumpers used to be hard but then came sweeping changes during the 1960s in the form of deepfoam mattings that allowed athletes alternate options so that they did not necessarily need to land on their feet. Fosbury responded to this vital change by jumping backwards over the bar, landing on his shoulders, which gave him the ability to jump higher. And the rest was history.
Fosbury showed the world that one can become a winner if one quickly adapts to changes. The world of technology is constantly evolving with the latest buzzwords being analytics, clouds, mobility and social media. Like Fosbury, the pharma industry is responding to these technological advancements in a big way. This article will seek to understand how the pharma landscape is ready to jump higher with the advent of business analytics and machine learning. Areas like biostatistics have been using advanced hypothesis based traditional analytics for more than 30 years and hence pharma has been a front-runner in some sense. In this article we will discuss how the latest trends like machine learning and deep learning can potentially impact four key business functions within clinical research.
Risk-Based Monitoring
Clinical trials are conducted at hundreds of sites, each comprising a sizable chunk of patient population. This translates into an enormous volume of data. Risk-Based Monitoring (RBM) is a new paradigm that attempts to analyze data with the ultimate aim of ensuring quality and safety of clinical trials. RBM is built on the fundamental concepts of risk management, i.e., detection, assessment and mitigation of risks.
Business analytics can play a crucial role in determining the success of RBM using descriptive (visualization), predictive and prescriptive algorithms. Visualizations like histograms, scatter plots and box plots can detect the outlier sites that have a disproportionately higher number of patient safety issues, delays in data entry/data review, fraudulent data capture, delayed patient enrollment and other anomalies.
With the advent of machine learning algorithms, the potential of predictive analytics in this field is huge. There are many variables or features that affect the performance of a site, giving rise to the possibility of supervised learning techniques like a decision tree and unsupervised learning tools like clustering analysis (K-means). These machine learning systems can provide the ability to predict a site’s risks that can slip into potentially serious safety, deviation or fraudulent data issues. Moreover, prescriptive analytics can provide an optimal deployment algorithm wherein adequate resources get allocated to those sites carrying higher risks.
Safety Signal Detection
Safety signal is reported information on a possible causal relationship between an adverse event (a harmful side effect) and a drug. Early identification of the hazards associated with drugs is the main goal of signal detection.
Traditional detection algorithms use either the frequentist approaches like proportional rate ratio (PRR) or the Bayesian techniques like Multi Gamma Poisson Shrinkage. However, predictive analytics through data mining techniques can be effectively deployed to estimate the probability at which an adverse event is caused by a drug. There are studies on signal detection that use association rule learning and PRR from the user-contributed content available in social media.
Patient Recruitment
Studies reveal that more than 30 to 40 percent of research budgets go to patient enrollment and, at the same time, 80 percent of clinical trials get delayed due to recruitment targets not being met. Predictive algorithms that forecast patient enrollment rates can help set the right expectation at the beginning of a trial.
Geography, therapeutic area, competition, epidemiology, phase and duration of a trial are some key variables that influence enrollment. Multi-variate regression analysis can be a very useful tool in determining those variables that have the strongest influence on enrollment. In many cases, lack of patient awareness of the trial benefits can stifle recruitment plans; therefore, promotions targeting the potential population with the relevant information can be very effective. Cost-benefit analysis through simulations determines if there is any significant value in promotional spend.
The use of analytics to gain enrollment feasibility insights early in the life cycle can avert substantial delays.
Clinical Trial and Data Management
Advancement in technology has ushered in a new wave of technology like wearables, smartphones and their associated mHealth apps. Along with this, the proliferation of social media and cloud computing have resulted in a deluge of data confronting every business across the globe. This increased access to digitization has equipped the industry with a blizzard of metrics that can help identify risks, issues and outliers. These metrics certainly help in providing answers to what is going wrong in a clinical trial.
However, an immediate impact of such data explosion is the need to derive a meaningful understanding from the data. Too often, clinical trial and data management run the risk of being data rich and information poor.
Here again, the use of dimensionality reduction analytics techniques like principal component analysis, factor analysis and correspondence analysis can help in firmly establishing causal relationships, which in turn help answer why an event happened. Combined with powerful visual analytics that transmit messages faster than textual data, there is a tremendous potential for analytics in this sphere of clinical operations.
Conclusion
Clinical trials constitute a major share of the budget, and timelines that go into drug development as well as the demand to reduce costs, improve quality and decrease timelines will always be present. Just in these four key business functions mentioned above - risk-based monitoring, patient recruitment, safety signal detection and clinical trial & data management - machine learning and deep learning can be deployed to reduce risks and costs by mining and interpreting meaningful data to gain powerful insight. Like the Fosbury Flop did for the sport of high jumping, analytics holds the key to transform pharma research as the industry shifts from reactive approaches to ones more proactive in nature.
As Chiltern’s Vice President and General Manager of clinical analytics in India, Shankar Arun leads the company’s clinical data management, eClinical services and biostatistics operations in India. With more than 21 years of experience largely devoted to technology within the pharma industry, he brings rich expertise developing and implementing services around the eClinical, biostatistics, CDM regulatory and pharmacovigilance landscape. In his current role, Shankar has been instrumental in establishing FSPs and stand-alone services for large, multinational pharma customers in Chiltern’s India offices based in Bangalore and Pune. Before joining the life sciences industry, he worked as a consultant in the U.S. for corporations like Microsoft and Apple. A thought leader and technology enthusiast, Shankar has presented on the pharma IT landscape in various forums like SCDM, Analytics India Summit and Unicom Analytics Summit.