Pharmaceuticals

AI in the pharmaceutical industry

Healthcare is becoming increasingly complex as new treatments emerge, the roles of healthcare professionals change, and new laws, regulations and payment models are introduced. Artificial intelligence (AI) is also being used more often. The resulting applications can often perform one specific healthcare task effectively. In many cases, they already perform that task better than a doctor. AI applications can, for example, detect infections at an early stage or improve the accuracy of diagnoses. The coronavirus crisis has accelerated this development considerably. Technology companies, researchers and governments are focusing heavily on AI applications that could help develop a medicine or vaccine against the coronavirus. The need and interest in using AI applications are there, but unfamiliarity and a lack of knowledge often make these data science projects difficult to get off the ground. In this article, we use a few examples to discuss how AI can help drive innovation and generate value.

Drug discovery and development

Bringing a new pharmaceutical product to market is a lengthy process with many obstacles. Tests regularly fail to meet their objectives, which can cause delays and increase the cost of an already expensive process. In situations such as the current coronavirus crisis, where a virus spreads rapidly, it is important to obtain accurate results quickly to accelerate drug development.

AI is therefore used in various ways to develop medicines. This involves data from viruses that, for example, have similar structures. That data provides insight into how certain medicines affect the treatment of patients with a similar virus. Google’s sister company ‘DeepMind’, for instance, has used AI to identify patterns in the coronavirus and speed up vaccine development. DeepMind uses an AlphaFold system to predict the structures of various proteins associated with SARS-CoV-2 that have been studied to different degrees. These predictions have not been experimentally verified, but they may help scientists understand how the coronavirus works. This could be useful in developing an effective vaccine. Gaining insight into these still-unknown proteins normally takes months. By applying AI algorithms, however, predictions can be made much sooner based on the structures of similar proteins.

Selecting patients for clinical trials

Bringing a new medicine to market takes an average of 10 to 15 years. Around half of that time, and the associated costs, are spent on clinical trials. This often makes clinical trials the most expensive phase of medicine development. Researchers therefore want to be sure they select the right patients for a particular study. Despite substantial investment, clinical trials still have a low success rate. Failures are mainly due to inaccurate selection methods and ineffective patient monitoring. AI can help improve these processes and increase the success rates of clinical trials. For example, some AI applications can reduce heterogeneity within a population, leaving patients who respond better to treatment. Monitoring patients remotely helps researchers track their behaviour during clinical trials and identify possible side effects of medicines. This makes it possible to predict which patients may drop out. AI is also used to select a niche patient population to help reduce costs.

Source: ERT, Transforming Clinical Trials through the Power of AI

Improving diagnostic accuracy

When a new pandemic strikes, diagnosing people is difficult. Large-scale testing is complicated and often expensive. Many people with symptoms of COVID-19 worry that they have contracted the virus, even when those symptoms may point to a milder illness. AI can play an important role in diagnosing diseases. A hospital in Florida was one of the first to attract attention for using AI in COVID-19 diagnostics. When patients arrive at the hospital, they receive an automated facial scan that uses machine learning to detect whether they have a fever.

A recent study found that 12 million adult patients in the United States receive an incorrect diagnosis each year, and that diagnostic errors account for 10% of deaths. By harnessing big data and analytics, healthcare providers can improve diagnostic accuracy and reduce mortality. Many data analytics companies now offer solutions that use innovative data science techniques and machine learning algorithms to improve diagnostic accuracy. These predictive techniques analyse historical data, such as patient records, symptoms, habits, illnesses and genomic structure, to produce an accurate prediction.

Source: Deloitte Analysis

Improving medication prescriptions

Pharmacists need insight into how often incorrect prescriptions occur. AI can help reduce the risks associated with prescribed medicines. When a medicine is prescribed to a patient, AI applications can recognise patterns in historical data. This allows doctors to be alerted to deviations from standard treatment procedures. It helps healthcare providers improve health outcomes and prevent complications linked to incorrect prescriptions. Clustering and scoring models can also be used to determine which treatment or medication is recommended for patients, based on historical outcomes and the success rates of treatment programmes.

Predicting medicine stock levels in pharmacies

According to the Koninklijke Nederlandse Maatschappij ter bevordering der Pharmacie (KNMP), medicine shortages are becoming more common. In 2018, 769 medicines were in short supply. By 2019, that number had almost doubled to 1,492. Shortages can be caused by problems in production or distribution, or by other economic factors. Demand for specific medicines can also rise during certain seasons or periods. That demand cannot always be estimated accurately, resulting in supply shortages. How can we help pharmacies prepare better for particular uplift/downlift events? And how can we make sure their stocks are better able to meet demand at such times?

KNMP Farmanco contains information on medicine shortages nationwide and historical data from pharmacies. This data offers a starting point for identifying patterns and relationships that could help predict demand for medicines. The historical data contains a great deal of useful information. Take the hay fever season, which peaks in May and June. During this period, you are most likely to experience hay fever symptoms, so people are more likely to visit a pharmacy for hay fever medication. Machine learning algorithms can quickly identify a trend. This also applies to conditions that are not seasonal: it is possible to predict which medicine will be dispensed most frequently at a given time. Pharmacies can then be given access to these results. They can be integrated into the existing environment, allowing medicine purchasing to be matched more closely to demand.

Natural Language Processing for text data

Natural Language Processing (NLP) is a computer programme’s ability to understand human language. One of the best-known products that uses NLP is ‘Siri‘, Apple’s virtual assistant. Siri uses NLP techniques to turn speech into commands (speech recognition) for navigating the phone.

But where does NLP fit into healthcare? The sector collects a vast amount of data, for example through electronic patient records, reviews and other sources. Since healthcare began adopting advanced technologies, the amount of data collected has grown enormously. But how can we put this data to good use?

That is where NLP comes in. Pharmacies and health insurers hold a great deal of data. By applying topic and sentiment analysis to, for example, patient reviews of pharmacies or customer reviews of health insurers, we can identify what people experience positively and negatively. These insights can help pharmacies and health insurers improve their services, leading to greater customer satisfaction. We also live in the age of social media: a single negative post about a medicine not working properly can be enough for a pharmaceutical company to face lawsuits. NLP and other AI algorithms are now used to quickly identify patterns and relationships in sources such as social media, local news reports and measurement data. By ‘scraping’ early warning signs from the internet, the pharmaceutical industry can make better decisions about its safety information and help prevent potential backlash and associated risks.

Which solutions could be of interest to you?

Right now, the healthcare sector is focused on the fight against coronavirus. This crisis has accelerated the use of artificial intelligence, wearables and innovative healthcare apps. The examples in this article illustrate just some of the possibilities. If this article has sparked your interest, we would be glad to get in touch to exchange ideas, offer inspiration and explore which AI solutions would be relevant and make an impact for you! We’re shaping the future for you and with you!