ClimaSalud-IA

climate
clinical-ai
AI platform integrating meteorological, environmental, and clinical data to anticipate the impact of extreme heat on chronic disease patients. Hospital pilot at Mútua Terrassa University Hospital.
Keywords

climate change, extreme heat, chronic disease, AI, predictive model, heatwave, Mútua Terrassa, Meteocat, epidemiology, case-crossover, UPC

Full title ClimaSalud-IA: AI for extreme climate events and chronic disease risk
Status Active · hospital pilot phase
Clinical partner Mútua Terrassa University Hospital
Meteorological partner Meteocat — Servei Meteorològic de Catalunya
University Universitat Politècnica de Catalunya (UPC / IRIS)
B2SLab PI Alexandre Perera Lluna

Context: climate change and chronic disease

Climate change is not a future risk for healthcare systems — it is a present one. Extreme heat events are intensifying in frequency, duration, and severity across Europe, and the Mediterranean region is warming faster than the global average. For Catalonia’s ageing population, where more than half of residents live with at least one chronic condition, each heat episode represents a direct clinical threat.

The physiological challenge is straightforward: the human body maintains core temperature within a narrow range (37 ± 0.5 °C). Extreme heat impairs thermoregulatory capacity, triggering a cascade of effects — dehydration, cardiovascular strain, renal stress, endothelial dysfunction — that interact with the vulnerabilities of pre-existing disease. The result is measurable, and preventable, excess morbidity and mortality during heat episodes.

What was missing, until ClimaSalud-IA, was the quantitative epidemiological foundation and the AI infrastructure to translate that knowledge into prospective clinical action.

The research: heat exposure and health-care use

In collaboration with Mútua Terrassa University Hospital and Meteocat, B2SLab conducted a detailed epidemiological study of the short-term association between heat exposure and four health outcomes in the hospital’s patient population:

  • Mortality
  • Emergency department visits
  • Hospital admissions
  • Unplanned primary care visits

Study population

The analysis drew on linked administrative and clinical data from Mútua Terrassa and the Catalan public health system (eCap / CatSalut), covering the municipalities served by the hospital across the Vallès Occidental area.

Outcome Total events ≥65 years Cardiovascular comorbidity
Death 1,791 87.0% 65.3%
Emergency visit 134,566 26.8% 39.1%
Hospital admission 17,840 50.1% 58.8%
Unplanned primary care 34,329 26.6% 38.9%

Patients were classified into eight comorbidity groups based on their diagnoses:

Comorbidity group Included diagnoses
Cardiovascular Congestive heart failure; myocardial infarction; peripheral vascular disease; cerebrovascular disease; arrhythmias
Respiratory Chronic pulmonary disease; asthma
Endocrine-metabolic Diabetes mellitus; liver disease; thyroid disorders
Renal Chronic renal disease
Mental-psychiatric Depression; anxiety; psychotic disorders; bipolar disorder
Neurological Dementia; Parkinson’s disease
Cancer-immunologic Solid or haematological malignancy; metastatic cancer; HIV/AIDS
Rheumatic-inflammatory Rheumatic disease

Heat definition

There is no globally standardised definition of heat, and the choice of definition has material consequences for epidemiological analysis. The study used the operational thresholds of Meteocat, the Catalan Meteorological Service, which grounds heat alerts in the local climatological baseline:

Note

Meteocat heat alert thresholds

  • Low-level alert: daily maximum temperature > 98th percentile of the local June–August distribution
  • High-level alert: daily maximum temperature > 98th percentile + 2 °C
  • Heatwave: daily maximum temperature > 98th percentile for ≥ 2 consecutive days

Statistical method: case-crossover design

The study applied a time-stratified case-crossover design, which controls for confounding by stable individual characteristics by comparing the heat exposure preceding each health event to exposure on matched control days within the same calendar month. This approach isolates the transient effect of heat without requiring an unexposed control population.

Four alternative exposure definitions were compared to assess the sensitivity of results to methodological choices about how heat is measured:

  • A. Single-day exposure — temperature at a single lag day
  • B. Window-averaged exposure — mean temperature over a 3-day window (lags 5–7)
  • C. Finite memory exposure — mean temperature over the 6 preceding days (lags 0–5)
  • D. Decaying exposure — exponentially weighted temperature over 15 days, with a fitted half-life

Key finding: distinct temporal associations

Tip

Heat affects health outcomes at fundamentally different time scales

The central finding of this work — captured in the study title — is that health outcomes of differing severity are associated with heat exposure over distinct time scales, reflecting differences in the underlying vulnerability and care-seeking behaviour that drive each outcome.

Using a decaying-exposure model to estimate the effective time scale over which heat acts on each outcome:

Outcome Mean effective exposure time scale
Death 1.84 days — mortality responds to very recent heat
Emergency visit 3.13 days — acute presentations lag by a few days
Hospital admission 3.78 days — hospitalisation follows acute deterioration
Unplanned primary care 10.24 days — primary care contacts reflect a much longer-duration cumulative exposure

The implication for clinical prediction is direct: a heat wave management system that monitors only the preceding 1–2 days will capture mortality risk but will systematically miss the excess primary care burden, which accumulates over nearly two weeks of sustained heat. An effective early warning system must account for this heterogeneity.

Across all outcomes and exposure definitions, a 1 °C increase in heat exposure was associated with significant increases in all four healthcare outcomes, with the strongest effects in mortality (OR ~1.06–1.08 per 1 °C increase, short lags) and hospital admissions. Among comorbidity groups, neurological disease showed the sharpest hospital admission risk at lags 0–4 days, while cardiovascular and endocrine-metabolic groups showed consistent effects across all outcomes.

The ClimaSalud-IA platform

Building on this epidemiological evidence, B2SLab designed and implemented ClimaSalud-IA — an AI platform that integrates heterogeneous data streams to anticipate the clinical impact of heat episodes prospectively.

Note

Data integration

ClimaSalud-IA draws on two data systems:

  • HCIS (Mútua Terrassa) — the hospital clinical information system: patient diagnoses, comorbidity groupings, medication, care history
  • eCap / CatSalut — the Catalan primary care system, linked by individual patient identifier (CIP), providing population-level coverage of chronic disease burden in the catchment area

Meteorological inputs — daily temperature forecasts and historical climatology — are provided by Meteocat, ensuring that heat alerts are anchored to the localised thresholds that govern the Catalan public health warning system.

The predictive model uses the temporal association structure identified in the epidemiological study — applying outcome-specific exposure windows so that mortality risk, hospitalisation risk, and primary care demand are each modelled at their own appropriate time scale.

Dashboard

ClimaSalud-IA dashboard showing patient risk stratification and meteorological data integration

ClimaSalud-IA data dashboard: integrating meteorological forecasts and patient risk stratification for clinical decision support during heat episodes.

The platform includes a clinical decision support interface designed for hospital and primary care management teams. During heat episodes, the dashboard surfaces:

  • Patients at elevated risk by comorbidity group and age band
  • Expected demand increase by care level (primary care, emergency, hospital)
  • Temporal trajectory of risk over the coming days

Hospital pilot: Mútua Terrassa

The hospital pilot at Mútua Terrassa University Hospital is the current phase of the project. Its objectives are to:

  1. Connect the platform to live hospital and CatSalut data in a compliant, secure configuration
  2. Complete final model training with post-2022 patient data, including cost-of-care dimensions
  3. Train clinical and operational staff
  4. Collect prospective data on clinical utility and operational impact during the 2026 heat season
  5. Document the results as the evidence base for territorial expansion

All activities are conducted in compliance with the General Data Protection Regulation (GDPR) and Catalan digital health regulations.

Tip

Why Mútua Terrassa

Mútua Terrassa is a mutual insurance and university hospital serving the Vallès Occidental area of Catalonia. Its combination of integrated primary and secondary care data, a substantial chronic disease population, and an institutional commitment to digital health innovation makes it an ideal partner for the first real-world deployment of the ClimaSalud-IA platform.

The data infrastructure spanning the HCIS hospital system and eCap primary care provides end-to-end patient visibility — from a GP consultation triggered by 10 days of accumulated heat stress to an emergency presentation on an acute heat day — which is precisely what the platform’s multi-scale temporal model requires.

Funding and partners

ClimaSalud-IA is an active collaboration between: