Periodic Reporting for period 2 - ASCAPE (Artificial intelligence Supporting CAncer Patients across Europe)
Periodo di rendicontazione: 2021-07-01 al 2023-06-30
The latest cancer statistics show promising advances in decreasing mortality related to cancer. However, the number of patients living with cancer will grow significantly in the near future due to the fact that one in two people will be diagnosed with cancer in their lifetime, while at the same time the average life expectancy increases.
As far as breast cancer is concerned, according to the CONCORD-3 study (data from 2010 to 2014), the 5-year net survival age-adjusted probability in all adults, in the 28 countries of the EU, ranges from 79% in Croatia to 93% in Cyprus. In 2018, the 5-year prevalence for breast cancer was in the absolute number of 2,054,887, from a total of 12,132,287 total cancer prevalence.
Regarding prostate cancer, the approximate number of new cases in the EU in 2015 is about 365,000 and is the most frequently diagnosed type of cancer in men. The incidence rates (ASR: age-adjusted rate on the European standard population) in the EU range from ASR 175 in Sweden to ASR 34 in Greece. The 5-year prevalence of prostate cancer in the EU is about 1,300,000, while at the same time survival has raised in all the EU countries with the highest increase monitored in the Eastern countries. The introduction and wide use of Prostate Specific Antigen (PSA) testing and diagnostic procedures such as biopsy have changed the distribution of the disease.
According to the numbers reported before, breast and prostate cancer survivorship represent a huge health problem for European countries. Physical, social, and emotional scars could compromise return to everyday life. Different studies showed that almost a third of cancer survivors experienced changes in their work situation after treatment.
Motivated by the above, the aim of ASCAPE was to build an open Artificial Intelligence (AI) infrastructure for cancer patient support where valuable patient data-derived knowledge in the form of Deep Learning models from healthcare providers across can be collected and shared through the cloud while advanced technological means ensure patient data remain confidential. This data-derived knowledge is made available to doctors to aid them in their decisions and help provide a better Quality of Life trajectory to their patients. ASCAPE challenges the Iron Triangle of Health orthodoxy by offering opportunities for both Quality of Care and Access to Care to improve while the Cost of Care decreases.
Results from ASCAPE clinical pilots showed a high level of acceptance of the proposed AI- based interventions among both patients and physicians. Continuous data collection, combined with ASCAPE’s ability to provide the most relevant information through an intuitive UI to clinicians at their fingertips, proved that can lead to significant improvements in both the quality and speed of their clinical decisions. In a similar manner, the predictive power of the data and its analyzes proved that is more than feasible to reduce the occurrence of health disorders and comorbidities that in turn could affect the well-being of cancer patients.
As far as breast cancer is concerned, according to the CONCORD-3 study (data from 2010 to 2014), the 5-year net survival age-adjusted probability in all adults, in the 28 countries of the EU, ranges from 79% in Croatia to 93% in Cyprus. In 2018, the 5-year prevalence for breast cancer was in the absolute number of 2,054,887, from a total of 12,132,287 total cancer prevalence.
Regarding prostate cancer, the approximate number of new cases in the EU in 2015 is about 365,000 and is the most frequently diagnosed type of cancer in men. The incidence rates (ASR: age-adjusted rate on the European standard population) in the EU range from ASR 175 in Sweden to ASR 34 in Greece. The 5-year prevalence of prostate cancer in the EU is about 1,300,000, while at the same time survival has raised in all the EU countries with the highest increase monitored in the Eastern countries. The introduction and wide use of Prostate Specific Antigen (PSA) testing and diagnostic procedures such as biopsy have changed the distribution of the disease.
According to the numbers reported before, breast and prostate cancer survivorship represent a huge health problem for European countries. Physical, social, and emotional scars could compromise return to everyday life. Different studies showed that almost a third of cancer survivors experienced changes in their work situation after treatment.
Motivated by the above, the aim of ASCAPE was to build an open Artificial Intelligence (AI) infrastructure for cancer patient support where valuable patient data-derived knowledge in the form of Deep Learning models from healthcare providers across can be collected and shared through the cloud while advanced technological means ensure patient data remain confidential. This data-derived knowledge is made available to doctors to aid them in their decisions and help provide a better Quality of Life trajectory to their patients. ASCAPE challenges the Iron Triangle of Health orthodoxy by offering opportunities for both Quality of Care and Access to Care to improve while the Cost of Care decreases.
Results from ASCAPE clinical pilots showed a high level of acceptance of the proposed AI- based interventions among both patients and physicians. Continuous data collection, combined with ASCAPE’s ability to provide the most relevant information through an intuitive UI to clinicians at their fingertips, proved that can lead to significant improvements in both the quality and speed of their clinical decisions. In a similar manner, the predictive power of the data and its analyzes proved that is more than feasible to reduce the occurrence of health disorders and comorbidities that in turn could affect the well-being of cancer patients.
During the second half of the project several activities have been performed towards the realization of ASCAPE objectives as summarized below:
• Delivered ASCAPE ML/DL algorithms and models based on federated learning, differential privacy, and homomorphic encryption.
• Provided an initial proof of concept implementation of the actual ML/DL models and algorithms.
• Designed several initial simulation methods based on domain knowledge cancer-care modelling.
• Evaluated initial accuracy and privacy-accuracy trade-offs testing of DL/ML ASCAPE models and algorithms.
• The final ASCAPE platform was deployed and integrated on all the four pilots and the Open Call participants.
• Testing was conducted to ensure the reliability and functionality of cloud ML and edge components to verify the performance and compatibility of the implemented features.
• ML based cancer-care predictive analytics and support for cancer care decision making was realized.
• A mechanism has been developed to implement a privacy-preserving federated learning environment in ASCAPE.
• Interpretable global insights into the reasoning leading to QoL issue predictions were provide in form of visualizations of surrogate models.
• A version of the standalone Dashboard for Doctors web application was delivered along with a Visualizations Library, packaging the core functionality of the Dashboard, allowing for this functionality to be easily integrated into web-based Healthcare Information Systems.
• During the prospective data collection in the four pilot sites >25.000 data points from EHRs of the four different healthcare systems, >290.000 data points from wearables and >380.000 data points from weather related data were collected.
• Across all four pilots, 379 patients were enrolled including 121 (31.9%) patients with prostate cancer and 258 (68.1%) patients with breast cancer.
• The assessment of overall QoL over time showed a relative improvement of at least 20% from baseline until the end of the study was set as KPI.
• The overall satisfaction of patients regarding the ASCAPE follow-up strategy was 5.6 (Likert-scale questions from 1 for strongly disagree to 6 for strongly agree) with the respondents to be as moderately to highly satisfied with ASCAPE.
• The overall satisfaction of doctors regarding the ASCAPE usefulness, relevant of information, experience in integrating ASCAPE into practice, accessibility, interface quality, trustworthiness among others was over defined KPIs with the respondents to be as moderately to highly satisfied with ASCAPE.
• Dissemination and communication results: 29 scientific publications with over 18K views and 34 citations, 45 events attended, 7 workshops (co)-organized, synergies with 11 EU projects, joined 3 clusters, 11 videos, 30 blogs, and 7 promotional materials designed.
• Open Calls results: 3 projects concluded with more than 6155 patients involved, 10 clinical centers and 2 hospitals participated, 3PhD students involved.
• Exploitations results: 2 major innovations to be exploited, a full market research conducted, a business model for ASCAPE AI-based platform was developed.
• Delivered ASCAPE ML/DL algorithms and models based on federated learning, differential privacy, and homomorphic encryption.
• Provided an initial proof of concept implementation of the actual ML/DL models and algorithms.
• Designed several initial simulation methods based on domain knowledge cancer-care modelling.
• Evaluated initial accuracy and privacy-accuracy trade-offs testing of DL/ML ASCAPE models and algorithms.
• The final ASCAPE platform was deployed and integrated on all the four pilots and the Open Call participants.
• Testing was conducted to ensure the reliability and functionality of cloud ML and edge components to verify the performance and compatibility of the implemented features.
• ML based cancer-care predictive analytics and support for cancer care decision making was realized.
• A mechanism has been developed to implement a privacy-preserving federated learning environment in ASCAPE.
• Interpretable global insights into the reasoning leading to QoL issue predictions were provide in form of visualizations of surrogate models.
• A version of the standalone Dashboard for Doctors web application was delivered along with a Visualizations Library, packaging the core functionality of the Dashboard, allowing for this functionality to be easily integrated into web-based Healthcare Information Systems.
• During the prospective data collection in the four pilot sites >25.000 data points from EHRs of the four different healthcare systems, >290.000 data points from wearables and >380.000 data points from weather related data were collected.
• Across all four pilots, 379 patients were enrolled including 121 (31.9%) patients with prostate cancer and 258 (68.1%) patients with breast cancer.
• The assessment of overall QoL over time showed a relative improvement of at least 20% from baseline until the end of the study was set as KPI.
• The overall satisfaction of patients regarding the ASCAPE follow-up strategy was 5.6 (Likert-scale questions from 1 for strongly disagree to 6 for strongly agree) with the respondents to be as moderately to highly satisfied with ASCAPE.
• The overall satisfaction of doctors regarding the ASCAPE usefulness, relevant of information, experience in integrating ASCAPE into practice, accessibility, interface quality, trustworthiness among others was over defined KPIs with the respondents to be as moderately to highly satisfied with ASCAPE.
• Dissemination and communication results: 29 scientific publications with over 18K views and 34 citations, 45 events attended, 7 workshops (co)-organized, synergies with 11 EU projects, joined 3 clusters, 11 videos, 30 blogs, and 7 promotional materials designed.
• Open Calls results: 3 projects concluded with more than 6155 patients involved, 10 clinical centers and 2 hospitals participated, 3PhD students involved.
• Exploitations results: 2 major innovations to be exploited, a full market research conducted, a business model for ASCAPE AI-based platform was developed.
During the second reporting period, the conducted research and development activities have progressed the state of the art in different areas. Representative cases follow:
• A uniform comprehensive and machine-processable FHIR HL7 data mode
• Open and flexible federated learning framework
• Data driven intervention suggestion and estimation of intervention effects
• Privacy-protection through Differential Privacy
• A privacy-preserving federated learning environment of together with a globally trained HE models on collected encrypted data that allows to study possible relationships between interventions and the evolution of QOL issues, but at the same time proposes interventions based on the data available in all participating site.
• Significantly improved patient counselling due to intuitive ASCAPE interface that enables doctors to effectively utilise their time with patients, track the evolution of patient characteristics, and generate new knowledge that is based on real world evidence.
• Improved QOL of patients participated and positive feedback received by both physicians and patients.
• Ability to ingest patient-reported outcomes (PROs) that can aid the development of relevant and data-driven policy strategies.
• A uniform comprehensive and machine-processable FHIR HL7 data mode
• Open and flexible federated learning framework
• Data driven intervention suggestion and estimation of intervention effects
• Privacy-protection through Differential Privacy
• A privacy-preserving federated learning environment of together with a globally trained HE models on collected encrypted data that allows to study possible relationships between interventions and the evolution of QOL issues, but at the same time proposes interventions based on the data available in all participating site.
• Significantly improved patient counselling due to intuitive ASCAPE interface that enables doctors to effectively utilise their time with patients, track the evolution of patient characteristics, and generate new knowledge that is based on real world evidence.
• Improved QOL of patients participated and positive feedback received by both physicians and patients.
• Ability to ingest patient-reported outcomes (PROs) that can aid the development of relevant and data-driven policy strategies.