New technologies are changing the medical industry around the world, simplifying the complicated process of cancer treatment.
According to data published by the World Health Organization, cancer is the first or second leading cause of death among people under the age of 70 in 112 of 183 countries and ranks third or fourth in a further 23 countries.1 An important factor that generates these statistics is a complex and time-consuming process of treating cancer, which the doctor is sometimes unable to cope with on his own. This problem has also been described in the publication “Artificial intelligence in medicine”, the authors of which – Sebastian Dobrzek and Anna Woźniacka, note that the work of specialist doctors is largely predictable, as it is based on developed diagnostic and therapeutic schemes. According to them, patients whose symptoms or response to treatment deviate from the expected scenario become a major challenge. In such a situation, the oncologist conducting therapy often abandons all other tasks and begins the tedious selection of the most effective method of treatment, analyzing textbooks, consulting specialists in other narrow fields and reaching for resources in online medical databases. Searching for a specific issue in such a database, however, obtains a huge number of records, only a small part of which refers to the problem being explored, and finding it is often extremely time-consuming.2 Doctors then come across tools based on artificial intelligence, which can analyze a large amount of data and learn to draw the conclusions needed for therapy from them. One such tool is the PPM technique developed by a Polish woman – Dr Agata Błasiak. It helps oncologists – often using complex combination treatments – quickly find the answer to the key question: “What drugs and in what dose will bring the desired effect?”3
Optimizing personalized medicine with artificial intelligence
Combination therapy consists in combining drugs so that when used together they bring the best results for the patient. However, there are no known universal combinations of drugs that work in the same way in every case. Therefore, the physician must each time estimate the effects of a given combination, using the data collected from the patient. The next step is to adapt it to the rapidly changing condition of a particular person, which means that the previously used therapy may suddenly stop working. Drugs that interact with each other often cause quite different reactions in each patient, and the problem becomes more complex the more complex the combination treatment is – so the more drugs are combined. For all these reasons, despite the efforts of the medical team, the selected treatment often turns out to be suboptimal. This problem was solved by a team of scientists from the Department of Bioengineering of the National University of Singapore, headed by Dr Agata Błasiak. To put it simply, their PPM technique can be compared to the work of two infallible doctors who have access to key data on each patient, continuously monitor their condition and use all the knowledge about cancer treatment available in the world. This solution has been called phenotypic personalized medicine (PPM) and works based on two tools, compared above to two doctors – QPOP and CURATE.AI platforms. QPOP finds the most effective drug combination for a given patient, while CURATE.AI extracts data on the appropriate dosing plan for these drugs.
The first of the platforms allows the selection of drugs for combination therapy and the determination of their initial doses. It is of great importance due to the huge number of possible combinations of different drugs in combination therapy. Let’s imagine that as a doctor we know five drugs that can help our patient if we combine them properly and five possible doses in which we can give these drugs together. However, we do not know which combination of drugs and doses will be most effective. Five drugs in five possible doses provide more than 3,000 combinations, which makes it impossible for us to test all these combinations on one patient, and therefore the chances of finding the perfect therapy for him decrease. QPOP addresses this problem by combining data on phenotypic outcomes (measurable, visible effects in the patient’s body) with combinations of drugs at different dosage levels before any of them is given to the patient. The platform examines the relationships between these datasets and then ranks the combinations in order from least to most successful. Comparing them with each other, he rejects combinations that turn out to be less effective or less cooperative, which leads to leaving only those best suited to the patient.
The second platform used – CURATE.AI recommends the most effective dosing of drugs indicated by QPOP for a given patient. This is where the individual patient profile is created, which changes dynamically along with the changing condition of the monitored person. The platform records disease progression and regression, additional medications, dose changes and other information affecting the effectiveness of treatment. The platform uses this data to predict how a particular patient will be affected by the administration of given drugs in certain doses.
Challenges in AI implementation and development
Artificial Intelligence is one of the new technologies that have not yet found wide application in all areas of our lives. This is the result of problems that seem to be related to the still low trust in modern solutions and the insufficiently rapid development of tools based on artificial intelligence. According to the NIK report, in Poland, compared to many other European Union countries, there are huge delays in the introduction of modern technologies in the treatment of diseases, including, above all, cancer (waiting time for their implementation is counted in years)4. In addition, medical databases lack data that would be sufficiently diverse for tools based on these technologies to be able to plan treatment for each type of disease and regardless of the different characteristics of each patient. Occasionally, the dataset contains only patients with textbook-defined symptoms and clear directions for dealing with those symptoms. When more complex and unusual treatments are needed, the operation of artificial intelligence may not be accurate enough and it is these unique cases that are the biggest challenge for doctors in the real world. In addition, collecting, describing and integrating medical data needed for the development of personalized medicine with the use of artificial intelligence requires large financial resources, which are still lacking. It means that the treatment of cancer patients in an egalitarian way – that is, adapted to their characteristics and disease history based on knowledge derived from AI – despite its huge potential, still cannot be used on a large scale in the medical sector.
The effectiveness of Artificial Intelligence systems in the oncological treatment
The use of artificial intelligence in oncology research enables faster and more accurate diagnosis of the disease and selection of treatment tailored to a particular patient. It also does not require testing combinations of different drugs directly on patients. This is because AI can combine a lot of data at the same time, and then plan the most effective treatment and estimate the patient’s reaction before administering drugs. CURATE.AI developed by Agata Błasiak’s team has been clinically verified in a study for mCRPC – a malignant prostate cancer with metastases that has stopped responding to therapy with the drugs used so far. The study proved the ability of this platform to guide combination therapy in patients in even such difficult cases. Thanks to the initially confirmed effectiveness of the method, the team that developed PPM, led by a Polish woman, expects that the discovery will allow for the widespread use of AI in the medical sector, and thus relieve people working in healthcare. Moreover, according to another study, the accuracy of the assessment of a given neoplastic lesion by artificial intelligence may be greater than the accuracy of the assessment of this lesion made by experienced doctors (W. Bulten et al., 2020)5. It proves that AI can be successfully used not only at the stage of planning drug dosing but also when diagnosing a given type of cancer. The report “The size of the artificial intelligence market in the healthcare sector in Europe” (2021) provides great hopes for improving the statistics mentioned at the beginning, which predicts that by 2027 the AI market used in medicine will increase by almost 50%.6 If these estimates come true, new technologies have the potential to change the medical industry forever, improving the length and quality of life of a huge number of people with cancer.
- Global health estimates: Leading causes of death, World Health Organization, www.who.int/data/gho/data/themes/mortality-and-global-health-estimates/ghe-leading -causes-of-death (accessed 06/11/2022)
- Artificial intelligence in medicine, Sebastian Dobrzek, Anna Woźniacka, Wydawnictwo UMED, www.wydawnictwo.umed.pl/wp-content/uploads/2022/08/MONOGRAFIA_06_2022__PATRZYK_S.pdf (accessed 06.11.2022)
- CURATE.AI: Optimizing Personalized Medicine with Artificial Intelligence, Agata Błasiak, Jeffrey Khong, Theodore Kee, www.pubmed.ncbi.nlm.nih.gov/31771394/ (accessed 06.11 .2022)
- Information on the results of the audit: Availability and effects of cancer treatment, Supreme Audit Office, Department of Health, www.nik.gov.pl/plik/id,16371,vp,18897.pdf (accessed 06.11.2022)
- Automated deep-learning system for Gleason grading of prostate cancer using biopsies: a diagnostic study by Wouter Bulten, Hans Pinckaers, Hester van Boven, Robert Vink, Thomas de Bel, Bram van Ginneken, Jeroen van der Laak, Christina Hulsbergen-van de Kaa, Geert Litjens, www.pubmed.ncbi.nlm.nih.gov/31926805/ (accessed 06.11.2022)
- The size of the artificial intelligence market in the healthcare sector in Europe, www.graphicalresearch.com/industry-insights/1777/europe-healthcare-artificial-intelligence-market (accessed 06.11.2022)