Posted by

Artificial Intelligence (AI) is machine simulation of human intelligence processes including learning, reasoning, and self-correction. 1 The goal of AI is to build machines that can perceive the world and make decisions in the same way as humans do.

Evolution of AI in medicine: In early 1970s, MYCIN launched diagnostic decision support.2 Early AI medical systems relied on large scientific domain people, that were required to train the computers by encoding the clinical information. However, these AI systems suffered due to more labor, time taken to build methods and difficulty to update the process. In 90s and 2000s, advanced machine learning methods were developed which can train themselves to learn the methods. However, they were time-consuming due to lack of availability of data. Currently, this situation had changed due to two factors. The transformational changes of AI in deep learning and related machine learning methods, aided by improvements in technology and significant training datasets. The second factor is accessibility of medical information in digital forms. In recent years, we have seen AI success in medicine such as, by using electronic health records (EHR) data3 we can predict the clinical factors right from disease diagnosis to death, classification of skin cancers4 and diagnosis of diabetic retinopathy from imaging techniques5.

Clinical research deals with the safety and efficacy of drugs, devices and diagnostic products proposed for human use. Clinical trials can be defined as the “investigations that are intended to detect the outcome of one or more investigational drug.” The rule of a trial is that, aim to ensure the rights, safety and well-being of subjects are protected, and the results of clinical trials are credible.

Drug phases

Clinical drug development has remained unchanged for the last three decades. It takes about 10 to 15 years and 1.5 to 2 billion USD to bring a new drug to market; almost half of this time and investment are committed to clinical trials. Phase-III clinical trial being the most complicated, expensive and time consuming.6


Brief Overview of The Clinical Trials

The probability of success (POS) of compound drug to proceed from phases to phase differs in the trail, where only 10% of compounds entering trail advances to FDA approval. However, a failed clinical trial sinks not only investment but also drug discovery research and pre-clinical costs.

Dev cycle

Source: Trends in Pharmacological Sciences 6

The main reasons for high trial failure rates are involved with “recruitment of subject techniques and cohort selection” paired with the inability to monitor subject effectively during the clinical trials.

Current advancements in artificial intelligence (AI) can increase the success rate in clinical research. Machine learning (ML) and Deep Learning (DL) can automatically find the patterns of meaning in large dataset like text, verbal communication and images.


Relationship between AI, ML and DL


Artificial Intelligence (AI): AI is applied in healthcare through machine learning. The goal of AI is to build machines that can perceive the world and make decisions in the same way as humans do.

Association Rule Mining (ARM): ARM is an important data mining activity that has a wide range of its application in data-mining industry. Association Rule is a design that is developed to imply recurrence of events or traits in a database. Such information can be utilized in a decision making in scientific and commercial domains.

Brain–Machine Interface (BMI): BMI is a communication gateway between a wired or enhanced brain and an external device. It is also referred to as a brain–computer interface (BCI), a mind–machine interface (MMI), or a direct neural interface (DNI).7

Heuristics: Involving or serving as an aid to learning, discovery, or problem-solving by experimental and especially trial and error methods.8

Markov Decision Process (MDP): It is a framework for decision-making modeling where in some situations the outcome is partly random and partly based on the input of the decision maker.8

Deep Learning (DL): It is a class of ML methods based on artificial neural networks, inspired by information processing and distributed communication nodes in biological systems, that use multiple layers to progressively extract higher level features from raw input. ‘Deep’ relate to the quantity of layers from which the data is being transformed.

Deep Reinforcement Learning (DRL): Deep reinforcement learning (RL) is an area of ML that is concerned with building software agents that can take actions in an environment to maximize some notion of cumulative reward.9 DRL combines DL and RL principles to create efficient algorithms to achieve this task.

Human–Machine Interface (HMI): HMI is defined as a user interface or dashboard that connects a person to a machine, system, or device.33 This can be technically applied to any interface that allows an interaction between a user and a device.

Machine Learning (ML): The scientific study of algorithms that build a mathematical model of sample data to make predictions or decisions without being explicitly programmed to perform the task. ML is often considered to be a branch of AI.

Support Vector Machine (SVM): SVM is a supervised machine learning algorithm which can be used for both classification or regression challenges.10

Natural Language Processing (NLP): A sub field of AI concerned with the interactions between computers and human (natural) languages, how to program computers to process and analyze large amounts of natural language data. NLP derives from various disciplines like computer science and its linguistics.

Optical Character Recognition (OCR): It is defined as a field in AI that researches at pattern recognition and computational vision aimed at the electronic conversion of images of typed, handwritten, or printed text into machine-encoded text, whether from a scanned document, a photo of a document, a scene-photo, or from subtitle text superimposed on an image.32


AI and ML tools have a broad range of applications in clinical R&D. It includes identifying molecules that are potential, finding patient populations that fit inclusion and exclusion criteria, and reviewing the scans, claims reports and other healthcare data to interpret trends to improve and accelerate decision making. Nevertheless, these tools can be leveraged only if organizations have access to them, expertise and extensive data sets to design and effectively build algorithms and interpret the results.

Healthcare data is often unstructured and tough to access, requiring extreme protection to ensure privacy. Although artificial intelligence had not had a significant impact on clinical research, AI based models are helping trial design, used for patient recruitment, and monitoring systems aim to boost study adherence and decrease dropout rates.


Two major areas “Clinical trial operations and Production evidence” have a prominent role in shaping the future of clinical trials.

The clinical trial operation is to improve the design of trial and recruitment and, subject selection. Production evidence drives across the whole spectrum from research: molecule design and development; genomics; clinical trial data and the patient health data.

The genomic data along with clinical data can be placed together to understand the patterns in the patient cohort and response rates. With the availability of patient data in the clinical trial and longitudinally before they enter the clinical trial, a comparative study can be initiated. To predict favorable outcome, or populations which are going to be successful, algorithms and artificial intelligence play a vital role in the clinical trial. This can be one way to improve precision medicine. There are significant applications that artificial intelligence and machine learning can be applied in clinical trial paradigm.11 


The prime candidates for improvement in clinical trial are better protocol design, patient enrolment, monitoring and retention.


Patient selection plays key role in the clinical trial. The participating patients in the clinical trial should have features like suitability, eligibility, motivation and empowerment to enroll. Selection of patients under the desired timeline is a major challenge and deems as reason for delay in clinical trials. About 86 percent of clinical trials do not meet recruitment timelines.12 AI and ML tools expand the patient cohort composition and helps with patient recruitment.

Cohort selection: Clinical trials are not intended to show the high rate of response of treatment drug in a smaller sample of the general population, but aim to select in a subset population, if one exists, can be swiftly demonstrated. This strategy can be referred as ‘clinical trial enrichment’.13 Enrolling high number of suitable subjects does not assure success of a clinical trial, but recruiting unsuitable patients increases probability of failure of the trial.14

Artificial intelligence can improve cohort selection by the following as per the Food and Drug administration (FDA):

  • Reduced population heterogeneity
  • Prognostic enrichment and predictive enrichment.

Electronic phenotyping focuses on reducing population heterogeneity, the process of identifying patients with specific characteristics of interest. Electronic phenotyping applications across study types are “Cross-sectional, Association (case-control/cohort) and Experimental.15 Prognostic enrichment can be termed as “selection of patients who are expected to have measurable clinical end point.” Predictive enrichment is “detecting the population group which can respond to treatment drug.” 13


Patient enrollment is a crucial step, which is a deciding factor in the trial success or failure. In trial recruitment, eligibility criteria complexity concerning to medical language and numbers is challenging for a patient to understand their own eligibility. Obtaining significant information manually from this unstructured data source is a major task which impact healthcare professionals and patients as well.

Artificial intelligence methods can help finding the needles in electronic medical records (EMRs) haystack: Clinical Natural Language Processing (CNLP), is a paradigm for pre-screening patients for clinical trials. NLP16 is used to understand the written and verbal language from both structured (demographic, clinical laboratory and medication information combined with International Classification of Diseases (ICD) and Current Procedural Terminology (CPT) reimbursement codes) and unstructured narrative data types (clinical notes, discharge summaries, radiology and pathology reports). Machine learning (ML) and deep reinforcement learning (DRL) allows systems to study the output quality by underlying algorithms.17 Algorithms based study optimizations are aimed at data points from previous clinical trials and real-world data (RWD) sources to identify risk factors and guide on the clinical trial optimization.18

Artificial Intelligence and Machine Learning techniques like NLP and optical character recognition (OCR) are used to mine freely available digital content such as trial declarations, digital trial databases and social media to classify trials of relevance and definite subjects.


In clinical drug development, monitoring the enrolled subjects is a crucial factor. Sponsors and stakeholders should collaboratively work to ensure safety monitoring. In trail, enrolling the eligible subjects is a challenging and immense investment in terms of time and money. Returns on this can only be recognized through successful accomplishment of the clinical trial. Therefore, it is crucial that enrolled patients stick to the procedures and guidelines laid in the clinical trial, where the data points for monitoring the drug effect would be collected efficiently and reliably.

The dropout rate in clinical trials is about 30 percent and only 15 percent of trials does not have this problem of dropout rate.12 This is due to lack of adherence to clinical trial procedures which might require additional enrolling of the patients. This can delay the trial period and lead to extensive costs.

Remote monitoring systems and refining training methods in the trial may reduce the dropout and non-adherence rates.19 Artificial intelligence techniques with wearable devices and sensors in trial offer new methods to develop portable, power efficient, timely and personalized digital monitoring systems.


In order to meet the adherence criteria, drug regimen records of the patient along with the data points associated with bodily functions, medication response and study protocols needs to be maintained. However, the patients can be non-adherent due to such time-consuming repetitive tasks. This can be mitigated by using the wearable sensors and remote monitoring. Wearable, video monitoring and companion applications can be used to collect patient data. This can reduce the patient task. Additionally, with these sensor applications, drug treatment can be delivered at the appropriate time, even for unforeseeable episodic indications. Machine learning and particularly deep learning methods can be used to evaluate data in real-time for identifying and recording the events. Artificial Intelligence and Machine learning approaches are also used to predict jeopardy of dropout rate for definite patient, which can detect the behavior of patient that suggests that he/she might be having adhering issues to the study protocol.

Applying deep learning for object detection in photos and investigating the real-time data from wearable device sensors, pilot testing and exploring artificial intelligence-based patient monitoring has been kicked off20 or successfully accomplished.22, 23 The innovation of wearable sensory devices with the medically health sensing technology24, furthermore running advanced deep learning methods25, 26 on such mobile platforms, can permit additional sensor combos to be investigated for a huge range of diseases.

While Health Insurance Portability and Accountability Act (HIPAA) compliant environments represent the information security baseline, advanced generations of AI-based tools scrutinize for making certain traceable and reliable multiparty communication and exchange of data.27-29


In PV, significant quantities of unstructured and structured data should be integrated and reviewed to make sure the quality and oversight. Artificial intelligence and machine learning tools are addressing several challenges that PV unit face in harnessing the ability of this data via new levels of insight and proactive analytics to reinforce the quality and oversight. AI methods such as OCR, NLP and deep neural networks are used to evaluate and format data for faster and efficient safety reviews.


There are three challenges that stand harnessing the power of artificial intelligence in clinical trials. They are “digitalization and availability of EMR data, accessibility of skilled resource and the opinion of pharmaceutical stakeholders and the regulators.”

Accessibility of EMR data is the critical one. In order to use artificial intelligence and machine learning in trials more efficiently, large quantities of data to run the algorithms is required. However, some algorithms can be used to reduce the amount of the data, but to get control of AI, we need to have large data. Hence, the pharmaceutical firms and medical data providers must mutually work to collect data more effectively. Subsequently, to run these artificial intelligence and machine learning programs and develop the codes, skilled workforce is required. A lot of pharmaceutical data scientists who are experienced in these fields are moving to finance industries or digital media sectors. Therefore, attracting these folks or hiring new individuals is a challenging task. The final challenge is that pharmaceutical stakeholders are typically conservative in approaches during new drug launch into the market. In fact, they would wait for the regulator to give guidance on artificial intelligence and machine learning. Also, the powerfully regulated legal background strictly limits third-party access to patient information and even makes it troublesome for patients themselves to access their own information. This is a major hurdle to make healthcare sector more efficient, and significant investments by governments and medical organizations can overcome this problem. The legal frameworks, the US Health Insurance Portability and Accountability Act (HIPAA) and the EU General Data Protection Regulation (GDPR) still evolve as protecting and governing the confidential health information. Alike with EMR data mining, for clinical trial meeting privacy and security with a degree of explainability of artificial intelligence methods need to be adopted to ensure that AI-based systems are functional and acquire the regulatory approval. This FDA approach can be a signal measuring the willing to embrace this new paradigm.


Over the past five years, recent AI techniques have advanced to the tier of maturity that permits them to use in real-life conditions to help human decision-makers in computer vision and in some cases of medical and health care environment 30. Simultaneously, pharma and healthcare sectors are still among the foremost extremely regulated and risk-averse industries. Such AI technology must be tested along with the existing technology that it aims to enrich or replace, and it needs to be demonstrated in explainable, accountable, ethical, repeatable, and unbiased way not solely to users but additionally to regulative bodies. Based on this approach, AI methods can be gradually implemented into trial, thus making it faster and minimizing research and development expenditure and trial failure percentages.6 Regulators continue to expand frameworks for assessing AI-based technologies in healthcare.31Similarly, transforming the clinical trials paradigm alone would not flip the efficiency of pharmaceutical research and development cycle from decay to growth, AI isn’t a magic bullet that may build the success rates of clinical trials skyrocket instantaneously. Transforming clinical trial design and utilizing the techniques are notable in the drug development cycle.

Authors’ Profile

Rajiv Dinakar Babu, Senior Executive- Pharmacovigilance at FMD K&L

Rajiv Dinakar picture

Rajiv Dinakar is a Pharmacovigilance expert with an experience of 6+ years and holds Masters degree in Pharmaceutics. He has dealt with multiple clinical trials and post-marketing surveillance projects and worked on a wide range of therapeutic areas. As an SME and trainer he has an expertise in handling Clinical Trials (CT), spontaneous and literature cases.

Srilekha Dinakar, Senior ExecutivePharmacovigilance at FMD K&L

Srilekha picture

Srilekha Kola is a Pharmacovigilance professional with 5+ years of experience in handling patient safety information for post-marketing and clinical trial ICSRs and held the responsibility of trainer. Currently at FMD K&L she serves as a Senior Executive- Pharmacovigilance and holds an expertise in case processing and quality review of SAEs and NSAEs, MedDRA coding, labelling assessment and safety narrative writing.



  1. www.technologyreview.com/s/612404/is-this-ai-we-drew-you-a-flowchart-to-work-it-out/
  2. Clancey, W.J. and Shortliffe, E.H. (1984) Readings in Medical Artificial Intelligence: The First Decade, Addison-Wesley Longman
  3. Topol, E. (2019) High-performance medicine: the convergence of human and artificial intelligence. Nat. Med. 25, 44–56.
  4. Esteva, A. et al. (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542, 115–118.
  5. Gulshan, V. et al. (2016) Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA 316, 2402–2410.
  6. Stefan Harrer et al. Artificial Intelligence for Clinical Trial Design. Trends in Pharmacological Sciences. Vol. 40, No. 8.
  7. https://www.absolutecentral.com/2019/08/how-brain-machine-interface-bmi-technology-could-create-an-internet-of-thoughts/
  8. https://www2.deloitte.com/nl/nl/pages/data-analytics/articles/part-2-artificial-intelligence-techniques-explained.html
  9. https://en.wikipedia.org/wiki/Reinforcement_learning
  10. https://www.analyticsvidhya.com/blog/2017/09/understaing-support-vector-machine-example-code/
  11. https://www.clinicalinformaticsnews.com/2018/12/14/are-we-ready-for-ai-in-clinical-trials.aspx
  12. http://www.clinpal.com/clinpal-blog/recruitment-infographic/
  13. http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM332181.pdf
  1. http://www.wsj.com/articles/bristol-myers-opdivo-failed-to-meet-endpoint-in-key-lung-cancer-study-1470400926
  2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6583807/
  3. Young, T. et al. (2018) Recent trends in deep learning based natural language processing. IEEE Comput. Intell. Mag. 13, 55–75
  4. LeCun, Y. et al. (2015) Deep learning. Nature 521, 436–444
  5. http://www.techemergence.com/ai-machine-learning-clinical-trials-examining-x-current-applications/
  6. http://www.appliedclinicaltrialsonline.com/mhealth-perspective-patient-centricity
  7. http://www.ibm.com/blogs/research/2017/04/monitoring-parkinsons-disease/
  8. https://builtin.com/artificial-intelligence
  9. https://britehealth.co/
  10. Labovitz, D.L. et al. (2017) Using artificial intelligence to reduce the risk of nonadherence in patients on anticoagulation therapy. Stroke 48, 1416–1419
  11. http://www.forbes.com/sites/davidphelan/2018/09/27/apple-watch-series-4-when-will-the-ecg-feature-arrive-in-the-states-and-worldwide/#6c4061193a09
  12. https://code.fb.com/ml-applications/delivering-real-time-ai-in-the-palm-of-your-hand/
  13. https://developer.apple.com/documentation/coreml
  14. https://medium.com/@kabagak/blockchain-in-pharma-clinical-trial-data-management-556ee88b46b9
  15. http://www.iconplc.com/insights/blog/2018/06/21/harnessing-blockchain-technology/
  16. https://pharmaphorum.com/views-and-analysis/how-blockchain-will-revolutionise-clinical-trials-clinical-trials/
  17. Walczak, S. (2018) The role of artificial intelligence in clinical decision support systems and a classification framework. Int. J. Comput. Clin. Pract. 3, 31–47
  18. http://www.fda.gov/newsevents/newsroom/pressannouncements/ucm604357.htm
  19. https://en.wikipedia.org/wiki/Optical_character_recognition
  20. https://www.inductiveautomation.com/resources/article/what-is-hmi

DISCLAIMER: The information provided in this White Paper is strictly the perspectives and opinions of individual authors and does not represent the opinions and statements of the iMEDGlobal (an FMD K&L Company).