IEO-Monzino: Converting unstructured clinical data into structured data for scientific research with Google Cloud

When technicians and experts from the European Institute of Oncology (IEO) and the Monzino Cardiological Center began utilizing Google Cloud services in August 2022, their primary purpose was to build the Clinical Data Platform (CDP) to collect all data being produced in the caring processes into a large health data lake in order to support the clinical research and to enrich the clinical practices.

Their objective was to leverage this accurate data for sector-specific investigations, promoting accessibility to the medical community operating within the centers. First and foremost, the aim was to convert information derived from diagnostic tests, medical records, and reports into analyzable data consolidated within a centralized data center. Additionally, the focus was on standardizing, categorizing, and processing this data with optimal efficiency. A crucial aspect of this approach was to ensure data privacy and security by anonymizing the collected information through secured access to the platform.

A project, related to CDP, started from the great work of Elisabetta Munzone, IEO high specialized physician in Breast Cancer Medical Treatments, who studied 500 pathological anatomy reports in order to extract data, such as biomarkers and diagnosis, with the purpose not only to interoperate with medical records coming from other centers, but also to relate them with international dictionaries. The work lasted 6 months. "We took advantage of such an initiative to structure a Natural Language Processing (NLP) model," Annarosa Farina, IEO-Monzino CIO, says. "We built a proprietary model using the Vertex AI to train and bring it into production. The task took us about 45 days."

76,000 medical reports data standardized in just two months

The Natural Language Processing (NLP) model began to operate and improve quickly, allowing the collection of structured information on 76,000 medical reports. But time is what makes the real difference: "It required about 2 months for data upload and deployment." This meant the processing speed was 300 times faster compared to the previous manual activity.

Usually medical reports contain a large amount of relevant information, affected also by the writer's sensitivity, spread across the whole document, thus resulting in high complexity for the NLP algorithms. "To conduct research or clinical analysis, it is essential the transformation of all the necessary information into standardized and anonymized formats. Currently, we are developing four Artificial Intelligence (AI) models."

Enabling all the Institute departments to use Cloud and AI

The aim of the IEO-Monzino teams is to extend these tools to all the data produced within our hospitals, providing the opportunity for many other departments to expedite and enhance scientific research, internal management, and patient care, while ensuring the privacy and security of their data.

IEO and Monzino today have a total of 1.7 million patients and 3.7 million outpatient visits. "As pathological anatomy alone, we count 800,000 reports and 37 million laboratory analyses. On the CDP there is an engine to select the data from the patients who explicitly gave the institution the consent for research projects. Moreover, these data undergo anonymization procedures, ensuring the privacy and confidentiality of individual patient information. These functionalities allow us to have pseudo-anonymized data in compliance with the European privacy law (GDPR)."

The next steps are well-defined: "We will use this standardization process for breast radiology," Farina explains. "They need to link the data with the histological markers. We extend the NLP models when we understand that we can adapt them or we develop new ones with AI, such as the case of the therapies. We have already developed two additional AI models in the hemolymph pathological area: the former generates standardized diagnoses and the latter biomarkers from the pathological anatomy reports."

Looker Studio's dashboard filters to study patients

The future of hospitals heads toward simplifying data interoperating. "Our users are mainly doctors and medical staff who are professional users, but not IT specialists." This is where Looker Studio, Google Cloud's tool for interactive dashboards and instant reports, can help/comes into play. "Clinicians love tools to monitor the outcomes of their activities, managers need operational and governance reporting. We experienced that this is the best solution to meet both needs."

The medical staff of the Division of gastrointestinal and neuroendocrine tumors of IEO can now easily check and extract data about patients who are in charge of the NET multidisciplinary team through the use of dashboard filters. "Now they have all they need at hand in one easy and intuitive tool."

From dashboards to web apps, more power to analyze data

As a next step, IEO will be switching to the Firebase web app, already implemented in the genomics area for the IEO's research campus. "We transferred the files from the sequencer onto the Clinical Data Platform (CDP): in this way geneticists can validate the genomic mutations, annotations and consequently produce reports."

With the hospital's internal group of the Clinical Research Coordinators, one of the IEO and Monzino's goals is to quickly find patients eligible for trials through the adoption of the Google Cloud technologies. "There are many potentially recruitable patients, selecting the right ones is quite a complex task, because the choice is based on the matching between the requested clinical inclusion criteria and the patient's clinical profile. With web apps we will be able to strengthen filters to find the most suitable profiles, combining the current status of the pathology, consents, etc."

Automations accelerate information gathering for certifications

IEO, as a cancer center of excellence, has achieved many certifications. As a part of the project with the Division of gastrointestinal and neuroendocrine tumors of IEO, we used the Clinical Data Platform to respond to the European Neuroendocrine Tumor Society (ENETS), which evaluates a healthcare organization in order to establish its adherence to a set of excellence requirements (standards) designed to improve patient safety and the quality of healthcare through a strict audit.

"The certification of center of excellence issued by ENETS is annual. In the past, IEO had to manually collect all the information required by the assessment, which involves data from the other hospital divisions, and hundreds of clinical parameters as KPIs. Thanks to the Google Cloud and its machine learning tools, we have been able to achieve the adoption of a dashboard to calculate the KPIs through the automated extraction and processing of the source data. This translates into a significant improvement in the outcome reliability as well as time and workload savings for numerous researchers and doctors, enabling them to focus on other aspects of their jobs."