Impact statement

This article demonstrates how even well-designed global mental health surveillance systems can systematically exclude entire regions, with consequences that ripple through research, policy and resource allocation. The absence of West Africa from the largest global student mental health survey, representing over 400 million people is not an oversight but a structural feature of how global health knowledge is produced. The framework we introduce helps researchers, policymakers and funders identify surveillance blind spots before they shape health policy. Our proposed principles and operational tools offer practical guidance for making invisible populations visible in digital health systems, with implications extending from epidemiology to artificial intelligence and population-level decision-making. This work matters for anyone concerned with ensuring that digital transformation in health serves all populations equitably, not just those already visible in global data systems.

Summary box

Cyber-epidemiology as a foundation for Population Health Intelligence

The increasing use of digital platforms, connected technologies, online interactions, electronic health records and computational systems has created new environments in which health-related information is generated, observed and acted upon. This development has contributed to the emergence of digital epidemiology, which uses digitally generated information to investigate population health and support surveillance and public health decision-making (Robotham et al., Reference Robotham, Satkunanathan, Doughty and Wykes2016; Naslund et al., Reference Naslund, Aschbrenner, Araya, Marsch, Unützer, Patel and Bartels2017). However, digital epidemiology has generally focused on the opportunities created by new data sources, while comparatively less attention has been given to how the architecture of digitally mediated environments may itself shape who becomes observable, what is measured and how epidemiological signals are interpreted.

In this context, we use cyber-epidemiology to refer to the interdisciplinary study of the distribution, determinants, measurement, transmission and population-health consequences of health-related phenomena within digitally mediated environments, including the examination of how digital infrastructures, platforms, interfaces, patterns of access, connectivity and governance shape the generation and interpretation of epidemiological evidence. Cyber-epidemiology therefore encompasses not only the use of digital data for epidemiological inference but also the critical examination of the conditions under which such data are produced and the ways in which those conditions may introduce systematic distortion into population-health intelligence (Robotham et al., Reference Robotham, Satkunanathan, Doughty and Wykes2016; Abimbola, Reference Abimbola2019).

This distinction is important because a digitally generated epidemiological signal is not necessarily a representative signal. Differences in connectivity, platform participation, language, digital literacy, socioeconomic position, institutional access and data architecture may determine which populations become visible to digital surveillance systems and which remain systematically underrepresented (Kumar et al., Reference Kumar, Gotz, Nutley and Smith2018; Veinot et al., Reference Veinot, Mitchell and Ancker2018). Consequently, the apparent precision of digitally derived estimates or predictive models may coexist with substantial uncertainty or inequity at the population level.

Our conceptual development therefore extends beyond cyber-epidemiology towards PHI: an integrated approach concerned not only with generating epidemiological information but also with determining whether that information is sufficiently representative, valid, analytically defensible, transparent and accountable to support equitable population-level decisions. In this sense, cyber-epidemiology provides an important foundation for understanding digitally mediated health evidence, whereas PHI provides a broader framework for evaluating and translating such evidence into trustworthy PHI. A comprehensive elaboration of the PHI framework, including its definition, core components and operational principles, will be developed in forthcoming work.

This Perspective argues that surveillance architectures reflect existing power structures, not epidemiological needs. We use the WMH-ICS Initiative as a sentinel case to demonstrate how even gold-standard global mental health surveillance can produce structural exclusion. We document West Africa’s complete absence from WMH-ICS, trace three cascading distortions that follow from this invisibility, situate this absence within the longer history of inequitable knowledge production and propose a research agenda and governance principles for making invisible populations visible. Figure 1 presents the proposed cyber-epidemiology framework illustrating how digital surveillance architectures shape population visibility and generate the three distortions described in the Surveillance Exclusion Triad, with downstream consequences for PHI.

Figure 1.

From surveillance exclusion to population health intelligence: a cyber-epidemiological framework for tracing representational inequity.

Note: This framework illustrates how structural determinants shape surveillance architectures, which in turn determine population representation and measurement validity. When surveillance systems generate incomplete or biased data, this produces the Surveillance Exclusion Triad: epidemiological distortion (biased prevalence estimates), resource allocation distortion (funding following existing data rather than need) and policy translation distortion (interventions lacking contextual validity). These distortions feed into the Population Health Intelligence (PHI) pipeline, the systematic conversion of population data into intelligence for prediction, prioritisation and policy with downstream consequences for health equity. The feedback loop indicates that evidence from implementation and outcomes should inform redesign of surveillance architectures and inclusion strategies. PHI = Population Health Intelligence.

Figure 1. Long description

The flowchart is organized into seven horizontal rows with a vertical sidebar on the right and a feedback loop on the left.

1. Population and Context: Includes geographic location, demographics and vulnerability, socioeconomic and structural factors, historical and colonial legacies, digital access and infrastructure, and institutional capacity and research infrastructure.

2. Surveillance Architecture and Participation Mechanisms: Includes study governance and inclusion criteria, funding and resource allocation, data collection platforms and tools, recruitment and participation pathways, and data logistics and management.

3. Population Representation (Measurement Validity): Asks who is represented in the surveillance data and links to representational adequacy and measurement validity.

4. Surveillance Data Output: Focuses on surveillance data, noting it is biased if representation is incomplete or measurement is invalid.

5. The Surveillance Exclusion Triad: Three interconnected boxes showing 1. Epidemiological Distortion (biased prevalence), 2. Resource-Allocation Distortion (funding follows existing data), and 3. Policy-Translation Distortion (interventions lack contextual validity).

6. Population Health Intelligence (P H I) Pipeline: A linear process from Data to Analytics, Prediction, Prioritisation, and Decision support.

7. Downstream Consequences: Biased P H I leads to misallocation of resources and widening health inequities, resulting in inequitable population health outcomes.

A vertical sidebar on the right lists Cross-Cutting Principles and Safeguards: Governance and accountability, Equity and inclusion, Measurement validity, Data provenance and transparency, Community and public participation, and Uncertainty and continuous learning. A Feedback Loop arrow runs from the bottom stage back up to stage 2, indicating that evidence from outcomes should inform the redesign of surveillance architectures.

From the 10/90 divide to the digital representation divide

The representational inequities observed in contemporary digital mental health surveillance must be understood within the longer history of unequal global mental health research investment. More than two decades ago, Patel documented what became known as the “10/90 divide” – the observation that less than 10% of global mental health research funding was directed to LMICs, where more than 90% of the world’s population lived (Patel, Reference Patel2007). This funding disparity produced a corresponding knowledge disparity – regions with the greatest mental health needs generated the least evidence to inform policy and practice.

The 10/90 divide has not been resolved; it has transformed. Contemporary digital surveillance systems may reproduce, amplify or obscure these longstanding asymmetries. Research infrastructure, institutional capacity and funding networks that were established during the era of the 10/90 divide now shape who can participate in digital data generation. Countries and institutions that were historically under-funded in mental health research are also less likely to have the digital infrastructure, trained personnel, and institutional partnerships required for participation in global surveillance initiatives.

This historical continuity is essential for understanding why West Africa’s absence from WMH-ICS is not an isolated oversight but a pattern consistent with decades of unequal knowledge production. The mechanisms that produced the 10/90 divide, concentration of research funding in high-income countries, colonial and neocolonial patterns of academic collaboration, language barriers in knowledge production and differential institutional capacity now operate through digital architectures to produce what we term the digital representation divide. Contemporary surveillance systems do not merely reflect existing inequities; they embed them in platform architectures, data standards and analytical frameworks that determine whose mental health becomes visible and whose remains invisible.

Surveillance exclusion as structural and intersectional inequity

Representational inequity is rarely produced by geography alone. Geographic absence may intersect with historical power structures, colonial legacies in knowledge production, digital access, socioeconomic position, age, language and institutional research capacity. These factors can operate cumulatively, producing populations that are simultaneously less likely to be sampled, less likely to generate digital traces, less likely to be represented in research infrastructures and less likely to influence the construction of measurement systems.

Intersectional representational inequity captures this complexity. A population can be statistically invisible not because of one exclusion mechanism but because several systems of exclusion reinforce one another. In the WMH-ICS case, West Africa’s absence may reflect not only geographic marginalisation but also:

Historical/colonial structures in knowledge production that have shaped research partnerships, funding flows and institutional relationships between African and non-African institutions. Socioeconomic disadvantage that limits institutional research capacity, digital infrastructure and the resources required for participation in international consortia. Digital access inequalities that affect both the feasibility of digital data collection and the populations that can be reached through digital platforms. Age-related factors: the WMH-ICS focuses on college students, a population that, while growing in West Africa, remains a small proportion of the region’s youth compared to higher-income regions. Language barriers in survey instruments and research communications that favour English and other European languages over the hundreds of languages spoken across the region. Institutional research capacity including trained researchers, ethics committees, data systems and administrative support required for participation in multinational surveys.

Surveillance exclusion should therefore be understood as an intersectional process rather than a simple binary distinction between participating and non-participating regions. This understanding has implications for how we interpret absence and design corrective strategies.

The sentinel case: West Africa’s structural absence from WMH-ICS

The WMH-ICS Initiative, led by the WHO World Mental Health Survey Consortium, represents one of the largest coordinated efforts to generate population-level mental health intelligence among young people globally. Since 2015, the initiative has collected data from more than 100,000 college students across multiple countries to estimate prevalence of mental disorders, treatment gaps and risk factors (Auerbach et al., Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016, Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018). Published findings have shaped the understanding of student mental health needs and informed policy recommendations worldwide, with participating institutions spanning North America, Europe, Asia, Latin America, and selected LMICs (Auerbach et al., Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016, Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018).

West Africa’s complete absence from WMH-ICS is particularly striking given the region’s demographic weight. Home to over 400 million people including the world’s youngest and fastest-growing population, the region faces substantial mental health needs amid rapid digital transformation and persistent infrastructure gaps. This exclusion raises urgent questions about whose mental health counts in global surveillance systems.

As of February 2026, no West African country is listed among participating institutions in published WMH-ICS reports (Auerbach et al., Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016, Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018; Cuijpers et al., Reference Cuijpers, Auerbach, Benjet, Bruffaerts, Ebert, Karyotaki and Kessler2019). Table 1 documents the geographic distribution of WMH-ICS participating countries. No institution in any West African country, Nigeria, Ghana, Senegal, Côte d’Ivoire, or the 12 other nations in the region has participated. This absence reflects a broader pattern documented in global mental health research whereby regions with the greatest need often generate the least data (WHO, 2021).

Table 1.

Geographic distribution of WMH-ICS participating countries

Table 1. Long description

The table consists of three columns: Region, Participating countries, and West Africa represented.

* North America: USA and Canada. West Africa represented: No.

* Europe: Belgium, France, Germany, Netherlands, Romania, Spain, Sweden, and U K. West Africa represented: No.

* Asia: Hong Kong and Saudi Arabia. West Africa represented: No.

* Latin America: Chile and Mexico. West Africa represented: No.

* Africa (non-West Africa): Kenya and South Africa. West Africa represented: No.

* Oceania: Australia and New Zealand. West Africa represented: No.

* West Africa: None. West Africa represented: No: 0 of 16 countries.

The final row for West Africa is emphasized in bold text to highlight the lack of representation from the 16 countries in that region.

Sources: Auerbach et al. (Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016, Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018), Cuijpers et al. (Reference Cuijpers, Auerbach, Benjet, Bruffaerts, Ebert, Karyotaki and Kessler2019).

Note: The table describes the geographic distribution of countries participating in the WMH-ICS studies identified in the cited sources. West Africa comprises 16 countries; none is represented in the listed WMH-ICS participating-country set.

However, we do not infer that this absence reflects intentional exclusion. The mechanisms are likely multiple and interconnected: concentration of research funding in regions with established infrastructure; institutional capacity constraints including limited trained researchers, overstretched ethics committees and inadequate data systems; established partnership networks that tend to reproduce themselves; and funders returning to familiar collaborators while new institutions struggle to enter consortia (Ezeh et al., Reference Ezeh, Kissling and Singer2020; Abimbola, Reference Abimbola2021; Abimbola et al., Reference Abimbola, Asthana, Montenegro, Guinto, Jumbam, Louskieter, Kabubei, Munshi, Muraya, Okumu, Saha, Saluja and Pai2021). These mechanisms are not separate from structural interpretations; they represent the operational channels through which epistemic hierarchies persist. We therefore treat West Africa as a sentinel case through which broader mechanisms of representational inequity can be examined, rather than as a uniquely or intentionally excluded region. Other regions, including parts of Asia, Eastern Europe and the Middle East, also show limited or no participation.

Figure 2 visualises the global distribution of WMH-ICS participation, categorising countries as participating (documented), non-participating (documented as not included) or no participation information available. The figure highlights West Africa’s absence and the scale of populations excluded from this major surveillance initiative, while acknowledging the broader geographic patterns of representational inequity.

Figure 2.

Geographic distribution of World Mental Health International College Student (WMH-ICS) participation and the West African surveillance gap.

Note: This map visualises the global distribution of WMH-ICS participating countries based on published reports (Auerbach et al., Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016, Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018; Cuijpers et al., Reference Cuijpers, Auerbach, Benjet, Bruffaerts, Ebert, Karyotaki and Kessler2019; search updated February 2026). Participating countries are shown in blue (n = 20), with the two African participating countries (Kenya and South Africa) highlighted in green. West African countries (n = 16) are highlighted in red to emphasise the region’s complete absence from this major global mental health surveillance initiative. Non-participating countries or those with no documented participation are shown in grey. The table below the map provides regional population coverage statistics, demonstrating that while over 100,000 students have been surveyed globally, West Africa that is home to approximately 418 million people has zero population coverage in this surveillance system. The map reflects documented participation in published reports; absence from reports does not imply intentional exclusion. Sources: Auerbach et al., Reference Auerbach, Alonso, Axinn, Cuijpers, Ebert, Green, Hwang, Kessler, Liu, Mortier, Nock, Pinder-Amaker, Sampson, Aguilar-Gaxiola, Al-Hamzawi, Andrade, Benjet, Caldas-de-Almeida, Demyttenaere, Florescu, de Girolamo, Gureje, Haro, Karam, Kiejna, Kovess-Masfety, Lee, McGrath, O’Neill, Pennell, Scott, Ten Have, Torres, Zaslavsky, Zarkov and Bruffaerts2016; Auerbach et al., Reference Auerbach, Mortier, Bruffaerts, Alonso, Benjet, Cuijpers, Demyttenaere, Ebert, Green, Hasking, Murray, Nock, Pinder-Amaker, Sampson, Stein, Vilagut, Zaslavsky and Kessler2018; Cuijpers et al., Reference Cuijpers, Auerbach, Benjet, Bruffaerts, Ebert, Karyotaki and Kessler2019; population estimates from UN DESA 2023.

Figure 2. Long description

A world map at the top shows participating countries in dark blue, including parts of North America, Europe, Australia, and Saudi Arabia. Two African countries, Kenya and South Africa, are marked with blue dots. West Africa is highlighted in yellow. A legend in the top-left defines these categories.

In the bottom-left, an inset map zooms into West Africa, showing 16 countries in yellow: Senegal, The Gambia, Guinea-Bissau, Guinea, Sierra Leone, Liberia, Mali, Cote d’Ivoire, Burkina Faso, Ghana, Togo, Benin, Niger, and Nigeria. A dashed line connects this region to the main map.

A table titled W M H I C S Participation by Region is located in the bottom-center. It lists data for W H O regions:

* Europe: 9 participating countries, ~929 million population, ~5.1 percent represented.

* Americas: 4 participating countries, ~1,030 million population, ~2.7 percent represented.

* Western Pacific: 2 participating countries, ~2,430 million population, ~1.7 percent represented.

* Eastern Mediterranean: 1 participating country, ~720 million population, ~0.3 percent represented.

* Africa (outside West Africa): 2 participating countries, ~1,110 million population, ~1.3 percent represented.

* West Africa: 0 participating countries, ~418 million population, 0 percent represented.

* World Total: 18 participating countries, ~6,637 million population, ~1.6 percent represented.

On the bottom-right, a Key Message box emphasizes that West Africa has zero population coverage in the initiative. Data sources include Auerbach et al. 2016, 2018 and Cuijpers et al. 2019.

The Surveillance Exclusion Triad

The exclusion of West Africa from WMH-ICS is not merely a data gap; it produces three cascading distortions with material consequences for global mental health equity. This structural absence generates what we term the Surveillance Exclusion Triad: three interconnected distortions with material consequences for global mental health equity (Table 2).

Table 2.

The Surveillance Exclusion Triad: distortions arising from structural absence in global mental health surveillance

Table 2. Long description

The table consists of four columns and three rows of data under a header row.

* Header Row: Distortion | Primary consequence | Mechanism | Illustrative evidence.

* Row 1: Epidemiological distortion. Primary consequence: Global prevalence estimates exclude West African populations. Mechanism: Sampling frames reflect research infrastructure rather than epidemiological need; infrastructure itself is produced through historical inequities. Illustrative evidence: W H O Mental Health Atlas 2020; Patel 2007.

* Row 2: Resource allocation distortion. Primary consequence: Funding flows towards already-visible regions. Mechanism: Documented burden determines allocation; undocumented burden remains unfunded. Illustrative evidence: Evans-Lacko and Thornicroft 2019.

* Row 3: Policy translation distortion. Primary consequence: Interventions may lack contextual validity. Mechanism: Instruments developed in one context are assumed to be transportable without validation. Illustrative evidence: Kohrt et al. 2014.

A note below the table explains that the triad conceptualizes three linked consequences of structural absence in mental health surveillance: distortion of the observed epidemiological evidence base, distortion of resource allocation, and distortion in the translation of evidence into contextually appropriate policy and interventions.

Sources: Patel (Reference Patel2007), Kohrt et al. (Reference Kohrt, Rasmussen, Kaiser, Haroz, Maharjan, Mutamba, de Jong and Hinton2014), Evans-Lacko and Thornicroft (Reference Evans-Lacko and Thornicroft2019), WHO Mental Health Atlas (2020).

Note: The triad conceptualises three linked consequences of structural absence in mental health surveillance: distortion of the observed epidemiological evidence base, distortion of resource allocation and distortion in the translation of evidence into contextually appropriate policy and interventions.

Epidemiological distortion: Structurally produced misrepresentation

Epidemiological distortion does not arise solely because research infrastructure is unevenly distributed. The distribution of infrastructure itself reflects historically accumulated inequalities in research funding, institutional power, knowledge production and geopolitical priority. Consequently, sampling frames reproduce prior decisions about whose health is considered sufficiently visible, measurable or policy-relevant. What subsequently appears as an empirical representation of disease burden therefore partly reflects the historical geography of research investment rather than the underlying geography of population need.

When West African populations are absent from global surveillance, prevalence estimates are not merely incomplete, they are systematically biased. The WMH-ICS produces knowledge about global student mental health that is actually knowledge about student mental health in participating regions. Policymakers and researchers who rely on these estimates unknowingly operate with a partial picture that systematically excludes the world’s youngest and fastest-growing population (WHO, 2021). This is not a technical problem of sampling; it is a structural problem of whose mental health counts in global knowledge production.

Resource allocation distortion: The self-reinforcing cycle

Research funding follows data. The World Health Organization’s Mental Health Atlas 2020 documents a stark inverse relationship: regions with the least data also receive the least funding, regardless of need (WHO, 2021). This creates a self-reinforcing cycle: absence from surveillance leads to underfunding, which perpetuates the infrastructure deficits that produced absence in the first place. Evans-Lacko and Thornicroft’s analysis of WMH-ICS implementation in LMICs identifies institutional capacity gaps as a primary barrier to participation, barriers that funding could address, but funding flows only where data already exist (Evans-Lacko and Thornicroft, Reference Evans-Lacko and Thornicroft2019).

The cycle operates as follows

• • Historical under-investment in research infrastructure

• • Limited capacity for participation in surveillance initiatives

• • Absence from global data systems

• • Invisible burden to funders and policymakers

• • Continued underfunding of research and services

• • Reinforced infrastructure deficits → return to step 1

This cycle cannot be broken by encouraging participation alone; it requires deliberate investment in research infrastructure in under-represented regions.

Policy translation distortion: The validity gap

Interventions designed for populations in WMH-ICS participating countries may not translate to West African contexts where symptom expression, help-seeking pathways and structural determinants differ substantially. Kohrt and colleagues’ seminal work on cultural concepts of distress documents that diagnostic instruments developed in one context often fail to achieve conceptual equivalence when deployed elsewhere (Kohrt et al., Reference Kohrt, Rasmussen, Kaiser, Haroz, Maharjan, Mutamba, de Jong and Hinton2014). The absence of West African data means these validation failures go undetected until interventions are already deployed.

Policy translation distortion is particularly concerning because it operates invisibly. When an instrument or intervention developed elsewhere is introduced in West Africa without local validation, failures may be attributed to implementation problems, patient non-compliance or local barriers rather than to fundamental mismatches between the instrument’s assumptions and local realities. Without data from the region, these validation failures cannot be identified, and policies built on invalid evidence may continue to guide resource allocation and service design.

From cyber-epidemiology to population health intelligence

The WMH-ICS case demonstrates why cyber-epidemiology must treat absence as rigorously as presence. Digital surveillance systems do not neutrally reflect epidemiological reality; they actively shape whose mental health becomes visible and whose remains hidden. Traditional epidemiology has sampling frames; cyber-epidemiology has platform-generated data. When platforms exclude populations, those populations disappear from view entirely (Abimbola, Reference Abimbola2019).

However, the implications extend beyond epidemiology. Contemporary health systems increasingly convert population-level data into PHI; actionable information used for prediction, prioritisation, resource allocation and policy decision-making. This intelligence is only as valid as the populations it represents.

Principles for equitable population health intelligence

From this analysis, five core principles for equitable PHI emerge:

From surveillance exclusion to algorithmic and generative AI inequity

The implications of surveillance exclusion extend beyond conventional surveillance systems. As artificial intelligence systems increasingly support prediction, classification, summarisation and population-level decision-making, representational inequities in the data ecosystem may propagate into algorithmic systems.

Populations that are poorly represented in epidemiological and digital health datasets may consequently have less influence on the data, concepts and contextual knowledge from which AI systems are developed and evaluated. This creates a representational deficit that operates at multiple levels:

• • Data representation: Under-representation in training datasets

• • Construct representation: The concepts measured and modelled may not reflect local realities

• • Contextual representation: Contextual knowledge about mental health in under-represented regions may be absent from training data

• • Evaluation representation: Systems may be evaluated primarily on populations similar to training data

The resulting problem is therefore not simply algorithmic bias, which suggests a technical problem fixable through algorithmic adjustments, but a deeper representational deficit within the PHI pipeline. Populations absent from surveillance data are also absent from the AI systems that increasingly inform health decisions.

For generative AI and large language models (LLMs), the concern is even broader. These systems are trained predominantly on data from US, European and Asian sources, with limited representation of African languages, concepts and contextual knowledge. As these systems are increasingly used for summarisation, translation, information retrieval and decision support in global health, their limitations for African populations may produce systematically poorer performance, misclassification or misrepresentation of mental health needs (van Kessel et al., Reference van Kessel, Hrzic, O’Nuallain, Weir, Wong, Anderson, Baron-Cohen and Mossialos2022).

This does not mean that all AI systems are biased against West African populations. It means that when a population is absent from the foundational data ecosystems, surveillance, clinical records and research studies, it is also likely to be absent from the AI systems built on those foundations. Addressing surveillance exclusion is therefore also an AI equity imperative. Figure 3 presents the PHI Continuous Cycle as an operational architecture for implementing equitable PHI, with six stages, cross-cutting principles, enabling foundations and feedback mechanisms for continuous learning and adaptation.

Figure 3.

The Population Health Intelligence (PHI) continuous cycle: an architecture for equitable, representative and actionable health intelligence.

Note: This figure presents an operational architecture for equitable Population Health Intelligence, the systematic conversion of population-level data into actionable intelligence for health systems. The cycle is organised into six stages: (1) population and context definition, (2) inclusive data generation, (3) data integration and curation, (4) analytics and modelling, (5) intelligence synthesis and validation and (6) feedback and continuous improvement. Cross-cutting principles; equity and inclusion, governance and accountability, measurement validity, data provenance and transparency, community participation and uncertainty recognition apply throughout the cycle. Enabling foundations include policy and regulatory environment, data and digital infrastructure, workforce and capacity, sustainable financing and partnerships and collaboration. The cycle is values-driven (equity, dignity, respect, transparency, solidarity, justice and do no harm) and places people and communities at the centre. Feedback loops from implementation and outcomes to surveillance design ensure continuous learning and adaptation, with the ultimate goal of equitable health outcomes: reduced disparities, more efficient and just resource allocation, contextually relevant interventions and stronger trust, participation and system resilience. PHI = Population Health Intelligence.

Figure 3. Long description

At the center of the diagram is a core circle labeled People and Communities at the Center, featuring a world map and a diverse group of people icons. Below this, the text Values Driving the Cycle lists equity, dignity, respect, transparency, solidarity, justice, and do no harm.

Surrounding the core is a circular flow of seven numbered stages connected by clockwise arrows:

1. Population and Context Definition: Define target populations, map needs, identify determinants, and engage communities.

2. Inclusive Data Generation (Surveillance and Beyond): Use multiple data sources, ensure inclusive participation, and document gaps.

3. Data Integration and Curation: Link and harmonize data, assure quality, and apply interoperability standards.

4. Analytics and Modeling (Prediction and Inference): Descriptive and inferential analysis, machine learning / A I models, and fairness audits.

5. Intelligence Synthesis and Validation: Validate across settings, assess transportability, and co-interpret with stakeholders.

6. Decision Support and Translation: Translate into actionable intelligence and assess equity implications.

7. Implementation, Monitoring and Learning: Implement interventions, monitor outcomes, and document lessons.

A horizontal box at the bottom, Feedback and Continuous Improvement, connects the cycle back to surveillance design.

Flanking the cycle are two vertical columns:

- Left (Cross-Cutting Principles): Equity and Inclusion, Governance and Accountability, Measurement Validity, Data Provenance and Transparency, Community Participation, and Uncertainty Recognition.

- Right (Enabling Foundations): Policy and Regulatory Environment, Data and Digital Infrastructure, Workforce and Capacity, Sustainable Financing, and Partnerships and Collaboration.

The diagram concludes at the bottom with a gold bar labeled Equitable Health Outcomes, listing goals such as reduced disparities, just resource allocation, and stronger system resilience.

An operational framework for equitable population health intelligence

The principles outlined above require operational translation into practical tools that researchers, policymakers and health system managers can use. We propose a six-domain operational framework for assessing and improving representational equity in PHI systems (Table 3).

Table 3.

Operational framework for equitable population health intelligence

Table 3. Long description

The table consists of four columns: Domain, Core question, Example indicator(s), and Failure consequence.

* Representation: Asks who is represented and what proportion of the target population is covered. Indicators include population coverage proportion and subgroup representation rates. Failure leads to invisible populations and biased intelligence.

* Measurement validity: Asks if constructs and digital traces are valid and comparable. Indicators include measurement invariance tests and cultural equivalence assessments. Failure leads to misclassification and invalid comparisons.

* Data architecture: Asks how data is generated and what platforms shape participation. Indicators include platform access analysis and digital literacy demands. Failure leads to digital selection bias and structural exclusion.

* Analytical validity: Asks if statistical inference is defensible. Indicators include sensitivity analyses for missing data and subgroup performance assessment. Failure leads to spurious precision and incorrect predictions.

* Translation: Asks how intelligence is translated into policy and if limitations are preserved. Indicators include resource allocation alignment. Failure leads to equity distortion and misallocated resources.

* Accountability: Asks who can challenge representational gaps. Indicators include representation audit processes and community advisory structures. Failure leads to uncorrected exclusion and perpetuated invisibility.

Note: The framework links population representation, measurement validity, data-generation architecture, analytical validity, translation and accountability. Each domain identifies a potential point at which representational inequity can enter or propagate through a population health intelligence system.

Assessment approach: For each domain, systems can be assessed as:

• • Adequate: Meets standards for equitable intelligence.

• • Incomplete: Has gaps that require acknowledgment and mitigation.

• • Insufficient: Cannot support valid intelligence for the target population.

• • Action implications: Domains assessed as incomplete or insufficient should trigger specific remedial actions, from improved sampling and data collection to measurement validation studies to community engagement and capacity building.

A research agenda for equitable population health intelligence

The WMH-ICS sentinel case identifies several research priorities for advancing equitable Population Health Intelligence.

Implications for policy and practice