Personalization is disrupting products and services in every sector – be it retail, e-commerce or the quick-service industry. Therefore, it makes good sense to consider personalization in the context of healthcare delivery and services as well. The WHO has promulgated that a 1:1000 doctor-population ratio is desirable. Yet 45% of the 194 member states of the WHO do not meet this criterion; they have less than 1 physician for a population of 1000. With this abysmal doctor-patient ratio, particularly in over-populated and developing third-world countries, healthcare personalization has the potential to bring about real impact and wide-spread health benefits to patients.
Even in countries that meet the golden ratio, each doctor has 100s of cases. They cannot possibly consider the diversity of factors, such as, the patient’s general behaviour, past medical history, available external support, environmental factors, etc. and their interactive effects for each of their patients, in entirety. The treatment and diagnosis given to a patient is a culmination of the doctor’s experience with other patients having similar symptoms, and the doctor’s expertise as it relates to the treatment options available.
The explosion of data in the form of genetics data, electronic health records (EHRs), sensor data, environmental and lifestyle-related data etc. have opened up new avenues of treatment. While it is difficult for physicians to utilize this data manually, machine learning algorithms can integrate this data coming in from various sources, and pattern-match personalized treatments for each patient, based on that patient’s individual characteristics. The key idea of personal medicine is to base medical decisions on individual patient characteristics (biomarkers) rather than on the expectations over entire populations. Such personalized treatments have better medication effectiveness and reduce adverse risks by using therapies that have a higher chance of showing positive results while minimizing negative side effects. This is also result in significantly lower healthcare costs, as a consequence of optimized therapies without much trial-and-error.
Deep learning architectures have already demonstrated significant technological advances across the spectrum of personalized medicine, due to their versatility in integrating multi-modal clinical data, training on large datasets, automatic feature identification and successful generalization on unseen datasets. Traditional machine learning algorithms can also deal with large datasets, but are often rendered ineffective as personalized data is highly multi-dimensional. In contrast, deep learning networks are very effective in highly complex and multi-dimensional spaces. Deep neural networks provide a higher-level semantic structure to understand correlations among input data, and are good at detecting patterns from unstructured data.
Deep learning networks are able to learn from large amounts of data, and were originally inspired by cognition in the human brain. They can generalize concepts and apply these to new data with high accuracy. Bejnordi et al. (2017) reported that 7 deep learning algorithms trained to detect metastases in hematoxylin-stained and eosin-stained tissue sections of lymph nodes of women with breast cancer outperformed a panel of 11 pathologists. Ding et al. (2019) proposed a deep learning network demonstrating an improved early prediction of the final diagnosis of Alzheimer’s disease (at 82% specificity and 100% sensitivity), on an average of 75.8 months prior to the eventual diagnosis. These are just a handful of successful research studies on deep learning networks in personalized medicine.
However, machine-driven personalized treatment is not without its own set of drawbacks and / or limitations. Orphan diseases, which affect only a very small fraction of the human population can result in highly imbalanced datasets. For example, there is only a single patient diagnosed with the ribose-5-phosphate isomerase deficiency. There are more than 5,000 such disorders which are categorized as rare. Another limitation of AI is the lack of universal EHRs. The average hospital in the United States uses 16 different EHR platforms, indicating that individual data is siloed across several platforms. For example, an individual patient may have data regarding the common cold stored on one platform, and data relating to his depression medications on another platform – both with differing formats and permissions. Thus, an AI system may thus not have access to all the information it needs to suggest a personalized treatment. There is also skepticism with respect to deep learning networks being a very important step in the rapidly evolving personalized medicine roadmap. This is largely because deep learning models are pretty much black-boxes – with limited interpretability and reduced focus on causality (versus making predictions). This inability to understand why an algorithm achieves generalization so well may be a critical factor inhibiting the clinical application of deep learning technologies.
While there have been significant achievements in using AI in many fields, it is essential to note that machine learning based diagnosis and treatments do not eliminate, or reduce the importance of doctors. Even though, some physicians have a tendency to view these algorithms as unwanted second opinions, they should not dismiss the valuable inputs provided by such systems. In fact, machines and doctors need to work hand-in-hand – the algorithm can learn from the doctor’s input to improve its skills over time. It is well-accepted that patients will always value the relationship they have with their caregivers. Machine learning algorithms and other technological developments in medicine provide the tools that can aid clinicians or augment their capabilities in order to dispense superior care to their patients.
Corbett, E. (2017). The Real-World Benefits of Machine Learning in Healthcare. Retrieved 23 February 2020, from https://www.healthcatalyst.com/clinical-applications-of-machine-learning-in-healthcare
Ding, Y., Sohn, J., Kawczynski, M., Trivedi, H., Harnish, R., & Jenkins, N. et al. (2019). A Deep Learning Model to Predict a Diagnosis of Alzheimer Disease by Using 18F-FDG PET of the Brain. Radiology, 290(2), 456-464. doi: 10.1148/radiol.2018180958
Ehteshami Bejnordi, B., Veta, M., Johannes van Diest, P., van Ginneken, B., Karssemeijer, N., & Litjens, G. et al. (2017). Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA, 318(22), 2199. doi: 10.1001/jama.2017.14585
Emmert-Streib, F., & Dehmer, M. (2018). A Machine Learning Perspective on Personalized Medicine: An Automized, Comprehensive Knowledge Base with Ontology for Pattern Recognition. Machine Learning And Knowledge Extraction, 1(1), 149-156. doi: 10.3390/make1010009
Fröhlich, H., Balling, R., Beerenwinkel, N., Kohlbacher, O., Kumar, S., & Lengauer, T. et al. (2018). From hype to reality: data science enabling personalized medicine. BMC Medicine, 16(1), 150. doi: 10.1186/s12916-018-1122-7
How Machine Learning Is Crafting Precision Medicine. (2019). Retrieved 23 February 2020, from https://www.forbes.com/sites/insights-intelai/2019/02/11/how-machine-learning-is-crafting-precision-medicine/#32436f865941
Papadakis, G., Karantanas, A., Tsikankis, M., Tsatsakis, A., Spandidos, D., & Marias, K. (2019). Deep learning opens new horizons in personalized medicine (Review). Biomedical Reports. doi: 10.3892/br.2019.1199
World Health Organization. (2018). Global Health Workforce Statistics. Retrieved from https://data.worldbank.org/indicator/SH.MED.PHYS.ZS
Zhang, S., Bamakan, S., Qu, Q., & Li, S. (2019). Learning for Personalized Medicine: A Comprehensive Review From a Deep Learning Perspective. IEEE Reviews In Biomedical Engineering, 12, 194-208. doi: 10.1109/rbme.2018.2864254