Leader: Antonio Gasbarrini (UNICATT); Other collaborator(s): Maria Benedetta Donati (NEUROMED), Elisa Gremese (UNICATT); Licia Iacoviello (NEUROMED); Carlo Pappone (UNISR), Antonio Cherubini (INRCA), Francesca Cecchi (UNIFI), Giuseppe Rengo (UNINA)
Task 6.4 activities will focus on the impact of climate change parameters and environmental pollutants exposure on cluster of age-related diseases. Task 6.4 activities will be based on data gathered in task 6.3 and will include the implementation of complex modeling methods
Brief description of the activities and of the intermediate results: The research activities focused on the definition of the study protocol for assessing the impact of climate change and air pollution on clusters of chronic diseases and multimorbidity in selected cohorts of older adults. Specifically, heat waves (defined as the presence of maximum apparent temperature above the monthly threshold for two or more consecutive days) and the concentrations of the main air pollutants recommended by the World Health Organisation (PM2.5, PM10, ozone, nitrogen dioxide, sulphur dioxide, carbon monoxide) will be used as exposure variables. Climatic/atmospheric information will be gathered through the interrogation of databases of meteorological and environmental data and satellite data. Data on patients admitted to the Fondazione Policlinico Gemelli Hospital in Rome will be used to test this approach. The effects of climatic/environmental variables will be assessed for their association with conditions whose incidence increase in the event of heat waves and/or environmental pollution (e.g. acute cardiovascular, neurological, infectious events). Following anonymisation, health information will be collected using codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) retrieved from emergency department (ED) and hospital discharge records.
The extraction of data from emergency department databases and hospital discharge records for the last 10 years has begun and the variables have been coded. Data on heat waves or cold spells (i.e. prolonged periods of extremely high or extremely low temperatures for a particular region) were extracted from the Copernicus database. Clinical and climate databases are being cleaned and analysis activities are being planned.
In the third reporting period of 2024, research activities focused primarily on the definition of the study protocol for assessing the impact of climate change on the incidence of mobility disability and other negative health outcomes in the cohort of older adults with physical frailty and sarcopenia enrolled in the SPRINTT study. Data on heat waves and cold spells have been extracted for all regions where study participants resided during the trial (i.e., 16 cities across 11 European countries). Exploratory phase has been started, which encompassed the exploration and cleaning of clinical and environmental datasets, as well as the identification of key variables for the analysis.
During the last reporting period, the Task 6.4 working group focused its activities on exploring the possible associations between environmental variables (temperature, humidity, pollution) and the incidence of adverse events (functional, cardiovascular, respiratory, muscular, hormonal and neurological) in over 1500 older people aged over 70 years with physical frailty and sarcopenia enrolled in the SPRINTT project (Sarcopenia and Physical fRailty IN older people: multi-componenT Treatment strategies). During the SPRINTT study period (about five years), environmental data for different European locations (i.e. 16 cities in 11 European countries) were collected by Copernicus (the European Union's Earth observation programme). We are now carrying out analyses to assess how extreme events, such as heat or cold waves, or pollutants (including carbon monoxide, nitrogen oxides, black carbon and organic carbon) affect the health of SPRINTT participants. We plan to explore hybrid methods that combine classical statistical techniques with innovative AI solutions to address gaps in clinical data and ensure consistency and reliability of analyses. The next phase will focus on identifying correlations between clinical and environmental data using both classical algorithms and machine learning, with the aim of maximising insights and deepening understanding of the interactions between environmental variables and health outcomes.
During January, we obtained the first preliminary results about the potential correlations between environmental conditions and the physical health of elderly participants, particularly focusing on the 400-meter walking test (DISWALK). The team selected the most vulnerable participants (SPPB<8) and established a binary outcome (presence or absence of disability). Data preprocessing involved integrating high-resolution environmental datasets (e.g., temperature, humidity, PM2.5, NOx) with clinical data. The first round of supervised learning using a Random Forest model showed limited predictive capacity (accuracy ~57%), although certain variables like NOx, CO, and humidity appeared influential. These early findings indicated that environmental factors alone were not sufficient for robust predictions. In February, the team deepened the DISWALK analysis with pairwise combinations of the most informative environmental variables through logistic regression. While some pairs, such as TEMP-15 (average temperature over 15 days) with DEW-7 (dew point over 7 days), offered relatively better F1-scores and recall, the overall predictive power remained weak. A parallel unsupervised analysis using K-Means clustering sought to identify whether natural groupings in environmental data could align with the outcome of motor disability. Although cluster structures showed moderate internal consistency (Silhouette scores >0.52), they did not significantly differentiate the outcome, reinforcing the conclusion that environmental variables alone have limited explanatory power regarding physical decline in the sampled population. March focused on analyzing subjective motivations reported by participants who did not complete the 400-meter test. Using a Large Language Model (LLM), free-text reasons were classified into six semantic categories (e.g., musculoskeletal pain, general malaise, psychological barriers). These were then cross-analyzed with environmental data. Both supervised (Random Forest) and unsupervised (K-Means) methods failed to uncover significant correlations—p-values and correlation coefficients remained outside acceptable thresholds. These results suggested that subjective test avoidance is likely influenced by personal, medical, or logistical reasons rather than environmental factors. The team concluded that while the methodologies were sound, a broader scope of variables is needed to fully understand health-related outcomes in the elderly.
Coming soon