Why are Clinical Trials so Expensive, Why They Fail and How AI can Help.

This is literally a multi-billion dollar question per drug. Bringing a new pharmaceutical drug to market takes about 12 years and can reach upwards of $1 billion in R&D expenditures. But why ?

The Process

The process of drug development involves various stages :

• In Vitro/Cell-based studies

• In Vivo/Animal Model studies

• Phase 1 trials

• Phase 2 trials

• Phase 3 trials

• Approval

• Phase 4 trials

Ninety percent of new chemical entities (NCEs) entering clinical trials fail, more than 80 % fail at phase 1 and 2 level

Why is it so expensive and time consuming ?

There are many reasons why clinical trials take so long and cost so much. 1. Volunteer-based. For the most part, clinical trials depend on volunteers willing to participate in the studies.

2. Clinical evaluation. Not every volunteer is recruited. There is a careful process to recruit eligible patients. Clinicians today manually go through a lot of medical data regarding patients ( volume ), with lots of different types of fields in the data ( variety ). Also, this needs to be done quickly ( velocity). This data has to be matched against the eligibility criteria for each trial .

3. Number of trial participants. Each trial requires a large number of participants. Human body is complex, so is its reaction to the drugs. This high variability means that large number of participants are required to assess the responses accurately. Also, in many cases, the response is subjective and hard to measure, such as a reduction in pain with a pain killer. Phase 3 trials typically involve hundreds to thousands of trial candidates.

4. Multiple trials. Multiple trials are conducted in multiple locations ( multicenter trials). Also, multiple phases of trials are conducted - for safety, dosing, efficacy, adverse reactions and so on. This multiplies to the number of trial participants required.

5. High failure rate. All of this is compounded the fact that since failure rate is so high, many different candidates need to be tested extensively before one successful drug comes to market.

Failed Trials Are A Problem It used to be that failed drugs weren't a huge concern. This is because although billions were spent on failed drugs, the occasional blockbuster drug that successfully passed the trials would generate enough revenue to cover the R&D of these failed drugs by a huge margin.

This is not the case today. Today's drugs have a smaller subset of indications. This is because they focus on smaller target populations with the advent of precision medicine. This means the revenue from successful drugs isn't what it used to be. Therefore, failure rate of drug trials has become a big challenge to pharma.

Reasons for High Failure Rates

One of the top reasons is the lack of adoption of biomarkers in the clinical trial process. Studies show that disease programs that adopted selection biomarkers had higher success rates than normal. However, finding predictive biomarkers that actually work and also the companion diagnostic is challenging. This is a complex process involving a deep understanding of the underlying biology. It requires understanding the relations between genomics data and tissue data. It requires understanding the tumor cells (in the case of cancer) and the immune cells. It also requires development of assays which can cost millions of dollars.

However, big problems require big solutions. The industry needs to prioritize research on biomarkers to improve the clinical trial process and fix the failure rates.

How AI can Help

AI can have a big impact on reducing the cost of clinical trials. These are broadly the various ways in which advanced intelligence can help with this problem.

1. Clinical Trial Process.

- Web and app-based recruitment of trial participants. While this is not much of an AI problem, selecting patients automatically based on inclusion and exclusion criteria for trial protocols can be automated using AI. Also, the algorithms can improve the effectiveness of the

- Adherence. Startups are already working on the problem of medical adherence using computer vision. Here, AI is used to automatically track if a patient is adhering to the drug dosage schedule and not missing doses, using photos of the patient placing the drug in their mouth or other means of detecting anomalous responses ( potentially indicating lack of adherence ).

2. BioMarkers.

The need is to select the right patient for the right drug at the right time at the right dose. This requires a deep understanding of the immune system and its interaction with the tumor ( in the case of cancer).

- over 50% of drugs fail Phase III because of lack of efficacy. Here, AI can help in two ways. First, AI can use historic data, as well as genetic and clinical data to identify which drug candidates are more likely to pass the clinical trials. In other words, an AI-based intelligent design for drugs. Second, biomarkers can be used effectively to pull drugs early when they fail. There are imaging biomarkers as well as biomarker proteins, genes and other substances. Broadly speaking and without going into many of the details, one can argue that AI can help with discovery and measurement of these biomarkers.

3. Stratify patients.

Identifying and qualifying patients for clinical trials using biomarkers helps improve success rate. For example, in the case of the trial of Opdivo, it is suggested that targeting the drug for use on patients with higher PDL-1 levels could have made the trial successful.

4. Genomics research tools.

Better AI-based tools for genomics research leads to better understanding of how genes, along with lifestyle and environment affect the body's response to a drug.

6. NLP to quickly summarize and select relevant medical research. Summarizing the large corpus of PubMed data and based on relevance to the trial process can improve trial success.

References:

https://software.intel.com/en-us/articles/artificial-intelligence-powers-clinical-trials

https://www.clinicalleader.com/doc/the-high-price-of-failed-clinical-trials-time-to-rethink-the-model-0001