Medical Intelligence and Anomaly Detection: Towards a Global System for Morbidity Forecasting

Paolo Poletti, Lecturer in Information Security Management at Link Campus University of Rome; Senior Advisor, Artemisia Lab and UAP

Introduction

The COVID-19 pandemic dramatically revealed the vulnerability of healthcare systems that are essentially reactive: they wait for disease to erupt instead of anticipating it. The traditional sequence of hospitalisation – treatment – follow-up has shown its limits: high costs, variable outcomes, and territorial inequalities.

In this context, the need to transform healthcare into a system capable of predicting, identifying and intervening before morbidity manifests on a large scale has become increasingly evident.

Medical intelligence — the set of tools, processes, data and predictive models that make it possible to monitor population health and anticipate vulnerabilities or critical events — thus becomes a strategic lever. When integrated with artificial intelligence (AI) technologies and a transnational perspective, it enables the construction of early-warning surveillance systems that go beyond the national level.

This article explores:

• the global context of health forecasting and the challenges of morbidity;
• the concept of medical intelligence and the role of “anomaly indices”;
• the contribution of AI, big data and anomaly detection;
• the enabling conditions for their development at national and international level;
• operational proposals and scenarios for Italy and for global cooperation.

1. The global context of morbidity and forecasting

Worldwide, healthcare systems are grappling with ageing populations, increasing multimorbidity and chronic diseases, and resources that struggle to keep pace with growing demand. In advanced economies, an ever-larger share of healthcare expenditure is devoted to treatment and chronic care management.

The global dimension of health instead requires a predictive system capable of considering emerging signals from other countries as well. An unexpected increase in bilateral pneumonia in a specific geographic area, a surge in demand for respiratory assistance, or changes in regional mobility patterns are all signals that may precede broader health events.

The pandemic demonstrated how the lack of early surveillance amplifies crises. Medical intelligence can help bridge this gap.

Box 1 – Case study: HealthMap and web-based surveillance

HealthMap, a digital surveillance platform, was among the first systems to signal anomalies in Wuhan in 2019, demonstrating how OSINT analysis and predictive models can anticipate epidemics (https://www.healthmap.org/en/).

2. Medical intelligence and anomaly indices

By medical intelligence we mean the ability of a healthcare system to collect data (clinical, behavioural, environmental), analyse them predictively, identify deviating patterns, and activate targeted interventions before disease fully emerges.

Anomaly indices are signals indicating a significant deviation from a historical or geographic trend (such as the increase in bilateral pneumonia observed prior to COVID-19) and may foreshadow rising morbidity. Examples include:

• an unexplained sudden increase in pneumonia cases;
• a surge in outpatient demand for specific symptoms;
• peaks in physiological parameters detected by wearables in vulnerable populations;
• increased online searches or social media activity related to specific symptoms.

Anomaly indices signal deviations from historical trends and may anticipate increasing morbidity.

Box 2 – Operational example

A 30% increase in diagnostic codes for respiratory conditions in a region compared to the historical average represents an anomaly index that must be investigated immediately.

Focusing solely on national data is clearly insufficient and exposes systems to the risk of delay. An anomaly in another country may precede its impact on the domestic system, given the rapid circulation of people. Interoperability, data sharing and transnational comparison are therefore essential.

3. The contribution of AI and digital technologies

3.1 Machine learning, deep learning and anomaly detection

AI is the enabling element of medical intelligence. Machine learning (ML) and deep learning (DL) models can process large volumes of structured and unstructured data and identify hidden patterns (non-linear correlations) that elude traditional analysis. Anomaly detection techniques allow deviations from expected profiles to be identified.

3.2 Examples and recent literature

• A 2023 study applied ML algorithms in hospital settings to predict mortality and discharge, achieving accuracy levels above 85%.
• Another study showed that a DL algorithm applied to individual wearable time series could signal deterioration before clinical intervention became evident.

3.3 Enabling technologies and heterogeneous data sources

Sources include electronic health records, territorial healthcare systems, environmental sensors, mobility data, social media and web search data. Integrating these sources enables an open and multidimensional view of health.

3.4 Challenges: bias, data quality, interpretability, privacy

Obstacles remain: data representativeness, algorithm explainability, the risk of reinforcing inequalities, and the protection of citizens’ rights. Governance, shared standards and continuous auditing are indispensable.

3.5 Non-health data sources and social predictive intelligence

(OSINT, SOCMINT, Deep/Dark Web, CLOSINT)

Reliable morbidity forecasting cannot rely on clinical data alone. Social, economic and behavioural dynamics often precede the statistical emergence of disease. Medical intelligence must therefore integrate heterogeneous sources:

• OSINT: online media, public reports, forums, institutional portals;
• SOCMINT: conversational and semantic trends on social networks, useful for capturing weak signals (reported symptoms, health fears, OTC purchases);
• Deep Web / Dark Web: lawful and targeted monitoring of exchanges concerning drugs, symptoms or unofficial practices that may anticipate uncensored phenomena;
• CLOSINT (closed, licensed/paid sources): proprietary and premium databases (e.g. insurance claims, aggregated prescription and pharmaceutical consumption data, mobility or retail panel data, wearable and telemonitoring datasets, market research), often more timely and structured.

AI techniques (NLP, anomaly detection, multimodal data fusion) identify linguistic, behavioural and statistical patterns linked to potential morbidity not yet captured by clinical flows. Combining OSINT, SOCMINT, Deep/Dark Web and CLOSINT allows triangulation and cross-validation of signals, reducing false positives.

All of this must operate under rigorous governance: clear legal bases and licences, data minimisation and anonymisation, model auditing, source traceability and a human-in-the-loop clinical validation before alerts are activated.

4. Enabling conditions: national and international

Integrated datasets with interoperable standards (e.g. HL7 FHIR, OMOP) are a prerequisite, alongside predictive analytics infrastructures, supranational governance and cooperation among countries. The model must include territorial medicine and accredited private providers, with shared rules and protection of rights. The involvement of accredited private healthcare providers, healthcare SMEs and technology companies is essential to scale the model.

In this regard, the approval of the European Health Data Space (EHDS) represents a decisive step for medical intelligence.

Thanks to a common European framework for interoperability, security and data reuse, the EHDS will enable real-time sharing of clinical information and epidemiological indicators among Member States, facilitating the identification of transnational anomalies and the development of continent-scale predictive models.

However, the geographical horizon must be broader still: a supranational framework for data sharing, rapid response protocols and cross-country comparison is needed.

Predictive systems must ultimately translate into effective territorial services: proactive medicine, monitoring and early patient engagement.

At the same time, as already noted, disease forecasting raises ethical (profiling, stigmatisation) and regulatory (data, algorithms, liability) issues that must be addressed.

5. Operational proposals and scenarios

• Establish a National Medical Intelligence Centre for predictive prevention: an institutional body that collects data and models, acts as a predictive hub and coordinates local and international responses.
• Launch international pilot projects for anomaly detection in respiratory and chronic diseases, such as:
  • early monitoring of respiratory morbidity in Southern Europe;
  • wearable-based monitoring in elderly populations;
  • analysis of variations in outpatient diagnostic codes across European or extra-European regions.
• Define global early-warning indicators (changes in hospital admissions, diagnostic codes, emerging symptoms).
• Integrate predictive AI into public and private healthcare networks.
• Establish an innovation fund to support predictive digital healthcare for SMEs.

Conclusions

The healthcare of the future must be proactive: predicting in order to prevent.

Medical intelligence, supported by AI and anomaly analysis, makes it possible to anticipate disease, improve sustainability and make the universal right to health truly effective.

It is time to build a global network that transforms knowledge into prevention, uniting public and private actors, technology and territory.

For Italy, Europe and global cooperation, the challenge is clear: to create an early-warning, forecasting and intervention system that transcends national borders, leverages innovation and integrates all healthcare stakeholders.

The goal is a vigilant, proactive and sustainable healthcare system — no longer merely responding, but anticipating. Because prevention is not only more efficient: it is more just.