CASE STUDY: Predictive analytics for healthcare

Up until relatively recent years, clinical records were handwritten by the clinician responsible of assessing the patient. These records were stored in physical folders and constituted the backbone of the healthcare documentation. This methodology was a practical solution but had major throwbacks: each medical doctor had to write down the clinical history by hand, thus favoring discrepancies in the documentation methodology. The folder for each patient containing the clinical history required its own physical storing space -which resource wise is very costly-, and the access to the information needed by the medical services was slow and tedious, particularly when the patient had to be treated by different professionals from diverse medical services other than his or her regular general practitioner. These factors resulted in a significant loss of healthcare efficiency due simply to the documentation system.

Over the last decade, however, the situation has dramatically changed: authorities have taken responsibility on the matter and slowly but steadily have introduced the informatization of the clinical records. This has boosted the dynamism of the healthcare system and has offered numerous opportunities for new medical strategies. From a sheer practical standpoint, a biopsy or hematological analysis, for instance, can be now performed kilometers away, and the doctor in charge of the patient can have the results on his computer screen almost at real time.


On top of this, what the recording revolution has brought along, is the creation of centralized datasets. There is a constant flow of large volumes of data from numerous sources within the healthcare system that allows advanced analytical strategies, such as predictive analytics, Machine Learning (ML) and Big Data analysis. This breakthrough has brought light into fields such as Epidemiology or Oncology, and brings the opportunity to finetune resources, discriminate risk factors and highlight the variables that best predict the outcome of interest.

There are, however, challenges that need to be tackled so that the transition to this Promiseland can be fulfilled. A major one is how to have access to the data and circumvent privacy matters. Medical records are private and sensitive data, and the access of third parties to this information is a very delicate topic: standardized protocols that secure the rights of the patients and guarantee their-our- privacy are essential. In this regard, blockchain protocol technology such as the newly released Microsoft´s Coco framework might be the right tool to tackle the issue; blockchains are decentralized, distributed and transparent, and on top of it all, extremely safe due to cryptography. The technology is therefore available, but political commitment is needed for the regulation of its application.


However, there are disciplines within healthcare and medicine that can bypass the above-mentioned issue where ML and Artificial Intelligence (AI) are already proving to be of great success: image analysis is one of those. Microsoft has several research lines, being one of which InnerEye, a research project that, along with a group of experts from Cambridge´s Addenbrooke’s Hospital, uses state of the art AI to build innovative image analysis tools to help doctors treat diseases such as brain tumors in a more targeted and effective way. One of the features of this project is the use of decision forest models for delineation of the tumor core in images obtained by Computed Tomography (CT). Once the model has been trained and tested with a fair number of samples, it can accurately score the tumor and even discriminate the different anatomical regions within it, such as the necrotic core, the edema and the active rim; the software achieves the scoring in a matter of seconds.  This approach can assist the clinicians with the evaluation, saving valuable time, money and resources: in Oncology, regular meetings of the specialists are required where each case is discussed and images are analyzed to classify the score and severity of each patient´s illness based on the data available. This methodology is very expensive since it takes time from a number of experts, which dilatates the time window for the treatments to be applied. Time is crucial for this type of pathologies, and being able of accurately addressing the nature of the tumor in a matter of seconds by a single operator is an evolution of huge impact for the healthcare professionals, the patients and administrations.




Not only non-invasive procedures are getting inputs from advanced analytics. Biopsy analysis is another field that is expected to be boosted by ML and Unsupervised Deep Learning in the very near future. Biopsies are tissue samples from patients that are processed in several ways, being a very representative one the Histological analysis on glass slides. In histology, making a long story short, the sample is processed and sectioned in thin slides of approximately 5 micrometers thick. Depending on what the clinician seeks to observe, different stains that highlight specific features of the tissue are applied on the sections. The most frequent staining is haematoxylin eosin, which colors the nuclei of the cells with haematoxilin in blue/purple and the cytoplasm, with eosin, in pink. This methodology is a golden procedure for pathologists to observe the morphology of the tissue and address potential abnormalities. Despite being a very informative technique for trained eye, there is very useful information that can be mined by data analysis from these samples that is virtually impossible to extract by looking at it. Instead, digitalization of these images and the massive numbers of characterized samples existing, Machine Learning models can be trained and tested so that it can assist the pathologists with the quantification and classification.




To sum up, it is hard to believe that state of the art of analytics will have an impact only on a few medical areas like precision imaging or administration. New data are continuously being generated at each hospital and medical center, such as Biochemical analysis, patient monitorization, microbiological culture, allergy testing, and so on. These data need to be properly analyzed so that they provide a revenue to society, to professionals and to the administration. To wrap up this statement with an estimation, a 2011 McKinsey Global Institute report estimated that if the US healthcare system could successfully apply big data to only drive efficiency and quality, the annual potential realized value could be more than $300 billion, two-thirds of which would arise from an 8% reduction in expenditures. And this is only the financial big picture. The benefits that patients and healthcare professional can obtain are difficult to predict, but what can already be inferred is that it is not going to be one to be ignored.

Adopt a Cloud!

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Share This

Copy Link to Clipboard