“In recent years, hospitals have no longer been able to obtain the analyses and answers they need from simple biostatistics: Data are becoming more heterogeneous and complex and now include textual data, signal data, genomic data, big data, and real world data, as well as images, all of which come from different information sources. Researchers and clinicians know that there is information in that data. However, we need something completely innovative to get it out: New models instead of those that have been taught for the past 20 years. Such methods are being developed through the sharing of transdisciplinary competences, and a place like Politecnico offers a chance to be on the cutting edge of exposure to all possible stimuli in this context.” Francesca Ieva is a professor of Statistics, Applied Statistics and Biostatistics in the Department of Mathematics of Politecnico di Milano. She has been doing analysis and statistical modeling of biomedical data for over ten years and has developed various methods of estimating, forecasting, and sizing problems derived from healthcare system requirements. “Over the years, we have worked primarily on data related to cardiovascular diseases. However, these methods have been designed to be adaptable to other diseases, and, in particular, we are working on two projects to apply them to the COVID situation. Firstly, in Lombardy, and, once the model is ready, it can be scaled up to areas of any size.”
The Coronavirus pandemic had a massive impact on our healthcare system’s ability to provide services, diagnoses, treatments, care, etc. The REVEAL project allows for an interpretative analysis of how access to all healthcare services has changed for the Italian population and forecasts how many people have been “left behind” due to healthcare system saturation and lockdown measures. “We must accurately estimate our ‘waiting list’ in order to use resources efficiently. Moreover, in some cases, patients will never catch up, as their health worsened due to lack of intervention and, therefore, they will impact even more on the system than they would have in normal conditions.” Starting from these data, the model builds an index for each patient, via artificial intelligence algorithms, that indicates how much this person is at risk of developing serious adverse events. These indices will be helpful both to institutions, for a correct estimate of resources to invest in the healthcare system, and to doctors, for creating personalized therapeutic treatments.
“REVEAL takes into consideration the patient’s current situation and clinical history. For each patient, the method allows us to automatically summarize their history with their present condition in a single output that places them in a specific risk category. The purpose of this indicator is not to replace doctors but to provide them with a support tool that assists in doing assessments, setting up monitoring processes or creating personalized therapies.”
Therefore, the objective of REVEAL is to build a decision-making tool for doctors and healthcare institutions that highlights possible scenarios of personalized intervention with a more rational and efficient use of resources. “If I estimate that a certain number of patients will now be classified as “at high risk”, this means I am associating a series of procedures with specific economic and human resource costs with this risk. Forecasting is fundamental in moments of extreme uncertainty, such as now, as it allows a healthcare system to allocate the correct amount of resources for its needs and not find itself overwhelmed in an emergency.” You can find an article that shows the REVEAL model applied to cardiovascular diseases at this link. Data related to COVID studies have not been published yet.
COVIDEMIA, the second project that Ieva’s group is working on, is also moving towards personalized medicine, while generating a healthcare system report to help staff forecast possible onsets of high-risk situations. While the REVEAL project is taking a close look at how well the healthcare system functions in extreme conditions, COVIDEMIA is studying the behavioral and psychological consequences affecting patients’ lives, i.e. their level of compliance with prescribed therapies. Here too, the method developed by Ieva for cardiocirculatory diseases is ready to be adapted to the COVID scenario.
“When patients with multiple or chronic conditions are released from hospital, they are prescribed a pharmacological therapy or routine. However, they are not always able to maintain compliance due to forgetfulness, indolence, contraindications, or for organizational or economic reasons. COVIDEMIA describes development scenarios based on a series of questions. Was the patient prescribed everything needed for their specific condition? Is the patient following their therapy correctly? If not, why? Etcetera. Each of these options has an impact on the healthcare system, as a patient with good compliance will be re-hospitalized less often and require less unscheduled appointments, while a non-compliant patient will likely develop further problems. We are measuring the impact of this behavior by integrating statistical models with artificial intelligence, machine learning, and text mining tools. We elaborate data using techniques that allow for appropriate pre-processing, which is then followed by modeling and clinical-epidemiological analysis, through parametric and non-parametric provisional models e.g. mixed-effect hierarchical models, linear and non-linear regression models, and multistate and survival models, to which longitudinal information is integrated via Functional Data Analysis techniques.”
Using regional administrative databases, researchers analyze biomedical and healthcare data with the objective of understanding how different behaviors translate into different results and in what measure. “We can extract and integrate very heterogeneous administrative or clinical data. This allows us to do scenario analyses that describe possible evolutions in the state of health of many patients. We use machine learning techniques to analyze non-structured and non-digital data, such as texts, images, or genomic information. All this information is then elaborated and appropriately summarized so it can be input into a provisional model and provide an accurate and easy-to-read output. The last step is to create information dashboards that provide easily interpretable results, even by clinicians and administrative staff.”
Tools such as REVEAL and COVIDEMIA are of great use both to doctors, who obtain objective criteria on which to base therapeutic decisions for individual patients, and to the system as a whole, as they allow for understanding what healthcare needs could be required in various scenarios within a specific time period. “When we talk about personalized medicine, one of the most important aspects is privacy protection. All these data are provided and handled anonymously. Each person is a statistical unit being studied for its behavior and their identity cannot be traced directly because data owners encrypt identities in accordance with laws regarding privacy and sensitive data handling. Analysis results are then projected in terms of patient risk. Nobody has or will ever have access to the personal data of individual citizens.” The timeline for both projects is 18 to 24 months.
This article was first published on 3 July by Politecnico di Milano.