Tackling methodological challenges to sharpen infectious disease forecasts in Ghana

Models built for pandemics in wealthy countries routinely misread how infections spread in places like Ghana, where fragmented data, informal economies, and deep cultural diversity bend epidemic curves in ways standard frameworks never anticipated. That is the central warning from an international team of researchers led by Verena Struckmann of the Technical University of Berlin and the German West-African Centre for Global Health and Pandemic Prevention, publishing in the journal Global Health Research and Policy. The group, which includes virologist Christian Drosten of Charité Universitätsmedizin Berlin and global health researcher John Amuasi of Kwame Nkrumah University of Science and Technology in Kumasi, argues that epidemic modeling for low- and middle-income countries needs nothing short of a methodological rethink—and they lay out a concrete roadmap for how to achieve it.

The commentary’s timing is pointed. During the COVID-19 recovery period, Ghana simultaneously confronted outbreaks of Marburg virus and Mpox, all against a backdrop of rapid urbanization, increased human-wildlife interaction, shifting transmission dynamics, and environmental degradation. Each of these pressures exposed fresh cracks in modeling approaches that were largely conceived in high-resource settings and then, with often fatal assumptions intact, transplanted into radically different epidemiological soil. The authors contend that the resulting projections were not merely imprecise but potentially harmful, because policymakers relying on them may have misjudged transmission rates, disease severity, and the likely impact of interventions.

At the heart of the problem is data. As of 7 April 2024, Ghana had officially reported 172,075 confirmed COVID-19 cases and 1,462 confirmed deaths. But the researchers note that underreporting analyses suggest the true death toll may have been as much as 27.75 times higher. That staggering discrepancy is not a footnote; it is the difference between a model that calibrates correctly and one that drifts far from reality. The team illustrates the problem with mortality statistics drawn from parallel reporting systems: in 2020, the District Health Information Management System (DHIMS) recorded 38,429 deaths while the Births and Deaths Registry (BDR) logged 51,026; in 2021, DHIMS counted 43,569 against 55,349 in the BDR. Meanwhile, Ghana’s 2021 population census, covering June 2020 through June 2021, estimated 132,199 total deaths—meaning the two routine systems together missed more than half of all mortality events. A scaled-area visualization in the commentary shows BDR capturing roughly 41.9 percent of census-reference deaths and DHIMS2 roughly 32.9 percent, with Health and Demographic Surveillance System sites at Navrongo, Dodowa, and Kintampo adding partial coverage that overlaps with the others in ways revealed only through field triangulation.

The fragmentation runs deeper still. Ghana’s mortality and health data are dispersed across a patchwork of digital platforms—DHIMS, the Lightwave Health Information Management System (LHIMS), the Births and Deaths Registry, HDSS surveillance sites, individual mortuaries, and the Surveillance Outbreak Response Management and Analysis System (SORMAS), which was deployed during COVID-19 to produce daily situational updates. While these systems complement one another, their differing reporting mechanisms, collection methods, and coverage areas generate discrepancies and duplication in national statistics. The National Health Insurance Management System introduces another blind spot: it omits the substantial share of Ghanaians who pay out of pocket or carry private insurance. With only 54 percent of the population holding active National Health Insurance Scheme membership in 2021, models that depend on utilization and coverage data cannot be properly validated, and the effect of insurance coverage itself becomes analytically invisible.

Diagnostic constraints compound the data problem. COVID-19 detection in Ghana leaned heavily on polymerase chain reaction (PCR) testing, which requires samples collected during the acute phase of infection. Reconstructing past transmission dynamics therefore depends on archived samples from infection peaks—a resource made scarce by limited long-term storage capacity and the involvement of multiple laboratories with no centralized biobank. Serological surveys, the classic alternative for estimating cumulative infection, face their own technical hurdles: antibody evidence persists only within specific time windows, and assays can cross-react with the endemic human coronaviruses that circulate widely, requiring additional confirmatory testing that inflates both financial and human resource costs. The upshot, the authors write, is reduced confidence in any retrospective estimate of how the virus actually moved through the population.

But the commentary insists that data infrastructure is only one of three interlocking methodological challenges. The second concerns the evaluation of non-pharmaceutical interventions (NPIs)—mask mandates, social distancing, and similar measures. In higher-income countries, NPI implementation tends to be standardized and enforcement mechanisms are comparatively strong. Ghana, like many low- and middle-income countries, faces far greater variability in compliance, shaped by informal economies, communal living arrangements, and uneven trust in public health institutions. The effectiveness of any given intervention varies widely across regions depending on public trust, risk perception, socioeconomic conditions, and the presence of concurrent measures—which makes isolating the effect of any single intervention statistically treacherous. Existing studies of NPIs in Ghana have focused on selected areas or populations, leaving an incomplete nationwide picture, and the absence of standardized evaluation frameworks makes cross-community comparison nearly impossible.

The third challenge is what the researchers call social-behavioral parameterization: building the human element into the equations. Ghana’s population encompasses more than 70 ethnic and linguistic groups, with a pluralistic society in which regional practices, diverse religions, and community governance shape health behaviors, risk perceptions, and institutional trust. Roughly 30 percent of the population works in agriculture and subsistence farming, pastoralism concentrates in the north, and significant cross-border movement of people, goods, and livestock flows along all of Ghana’s borders, complicating biosecurity management. In rural areas especially, trust in informal networks—elders, religious leaders, traditional healers—strongly conditions public responses to official guidance. A compartmental model that assumes homogeneous mixing and uniform compliance simply cannot represent these dynamics; the parameters it needs, such as compliance levels and risk perception by demographic group, are precisely the ones hardest to measure and most often missing from LMIC datasets. Higher-income countries, by contrast, benefit from more homogeneous behavioral datasets and stronger institutional capacity for regular behavioral surveillance.

To close these gaps, the team proposes five targeted strategies. First, develop culturally adapted metrics for NPI effectiveness, grounded in regular community surveys that capture local attitudes and needs, so that model parameters reflect actual behavioral patterns rather than imported assumptions. Second, overhaul data infrastructure by integrating and centralizing health data from DHIMS, LHIMS, and HDSS into a unified digital repository—with phased implementation and capacity support—while expanding sentinel surveillance sites across diverse regions and implementing real-time reporting for mortality and compliance data. Third, establish national reference biobanks to centralize storage of biological samples, enabling nationwide retrospective studies; the authors are explicit that such investments demand sustained financing and technical capacity, and should be justified through cost-effectiveness evaluations that weigh implementation constraints like infrastructure and workforce availability. Fourth, create guidelines and partnerships for local model adaptation, including pilot testing and validation in selected regions before scaling, supported by regional partnerships, international collaboration, training curricula, and technology transfer programs. Fifth, strengthen a sustainable, needs-oriented financing strategy through increased domestic health funding, pooling of public and private resources, and Program-Based Budgeting that aligns spending with national priorities.

Underlying all five recommendations is a call for genuine interdisciplinarity. The authors argue that models will only track infections effectively while aligning with local realities when epidemiologists work alongside political scientists assessing risk perception and compliance, virologists supplying localized pathogen data, and health systems researchers mapping access disparities. Participatory research and policy co-design involving community leaders, they contend, improves not just model accuracy but also relevance, trust, and adherence—the social substrate on which any projection ultimately rests. At the same time, the team strikes a note of fiscal caution: investments in modeling methodology must be carefully weighed against potentially more critical healthcare needs, so that resource allocation remains responsive to Ghana’s broader public health priorities.

The stakes extend well beyond Ghana’s borders. The authors frame their argument as a test case for the broader problem of model transferability across resource settings—a difficulty documented in the methodological literature on combining data from multiple sources. If Ghana, with its high infectious disease burden, active regional role in surveillance, and strong political commitment to digital health strengthening, cannot make global modeling frameworks work, the implications are sobering for much of Sub-Saharan Africa and the wider Global South. Conversely, the researchers argue that success in Ghana could set a valuable precedent: a demonstration that context-sensitive, interdisciplinary, data-integrated modeling can deliver accurate, policy-relevant projections in settings where the classic assumptions of epidemic mathematics break down. By investing in disaggregated data, biobanks, community-engaged parameterization, and sustainable financing, the commentary concludes, Ghana can build a resilient public health architecture capable of protecting its population against future outbreaks—and, in doing so, help reshape the science of epidemic forecasting into something that works for everyone, not just for the countries whose data pipelines it was designed around.