From discovery and research to final production, the full process for bringing a drug to market usually lasts 10-15 years and costs approximately $2.6B. While only 14 percent of all new drug candidates reach testing and gain FDA approval, medication developers and manufacturers are already investing heavily in AI as a way to discover new drug compounds quicker—with less miscalculations—leading to higher approval ratings. Growth in this area will continue to boom.
With such a lengthy and expensive process to bring a new drug to market, discovery must be streamlined to make the correct investments, as at present billions of data points are examined when judging a potential drug candidate. Drug development has historically been an iterative process using high-throughput screening (HTS) labs to physically test thousands of compounds a day, with an expected hit rate of one percent or less. AI/ML offers the potential to add efficiency and scale. Machine learning technology is utilized to correlate vast amounts of data, uncover hidden relations, and generate new solutions. These systems are currently being used to search for new candidate compounds, speed complex computer simulations, and propose different routes of synthesis for new drugs.
ICD-11 is coming. The World Health Organization (WHO) announced in June 2018 the latest list of International Classification of Diseases (ICD-11) and presented the list to Member States in May 2019. The list, which contains four times the number of codes in ICD-10, will come into effect on January 1, 2022, including 10,000 proposals for revisions from ICD-10. This number is impossible for a human to correctly interpret and code.
When ICD-10 was implemented in October 2015, the number of codes rose from 13,000 (ICD-9-CM) to 68,000 (ICD-10-CM), according to the Centers for Medicare and Medicaid Services (CMS). While some healthcare facilities have become early adopters and have begun using some form of natural language processing (NLP) or ML, the majority of facilities still rely solely on human coders. This often leads to inaccuracies in interpreting provider notes, especially with unique codes and modifiers. For example, “pecked by a turkey” is ICD-10 code W61.43, while “pecked by a large chicken” is code W61.43, which may be easily confused by a human coder. Once the provider enters their note, AI/ML is better able to determine the correct code and recommend it for reimbursement.