The Factory Floor Solution to Pandemic Preparedness

The COVID-19 pandemic exposed a brutal truth: the world had sophisticated systems for tracking shipping containers and optimizing supply chains, but no equivalent infrastructure for tracking and controlling a virus spreading across borders in real time. When epidemiologists needed to know how many people were actually infected, they were flying blind. When hospitals needed to predict ICU demand three weeks out, they had no reliable crystal ball. When policymakers weighed lockdowns against economic collapse, they were making trillion-dollar decisions with incomplete, lagging, and often contradictory data.

A new paper from researchers at the University of Pennsylvania's Department of Industrial and Systems Engineering argues that this chaos was not inevitable—and that the tools to prevent it already exist, borrowed from decades of work in manufacturing, logistics, and service industries. The paper, "Epidemic Informatics and Control: A Holistic Approach from System Informatics to Epidemic Response and Risk Management in Public Health" by Hui Yang and colleagues, proposes a unified framework for applying industrial systems engineering to epidemic management. The framework is called DMAIC—Define, Measure, Analyze, Improve, and Control—and the researchers argue it could transform how humanity responds to the next pandemic, whether it emerges next year or next decade.

"The sustained successes of system informatics in a variety of established industries such as manufacturing, logistics, services and beyond" have never been systematically applied to epidemic response, the authors note. Their paper attempts to bridge that gap, synthesizing insights from public health, statistics, operations research, and artificial intelligence into a coherent blueprint for future pandemic preparedness.

The Science

The paper emerged from an unlikely intersection: industrial engineering and epidemiology. Yang's home department at Penn is traditionally focused on optimizing factory floors, hospital scheduling, and supply chains. But during the COVID-19 pandemic, Yang and his colleagues watched as familiar problems—data collection errors, forecasting failures, resource allocation bottlenecks—played out in a context far more consequential than any manufacturing line.

"We realized that the same frameworks we use to improve automotive assembly quality or reduce hospital wait times could be applied to epidemic response," Yang explained in a subsequent discussion of the work. "The math is different, but the underlying challenge is the same: how do you make good decisions with incomplete data under time pressure?"

The research team included specialists in statistical sampling, health systems modeling, and artificial intelligence. Over roughly two years, they conducted an extensive literature review spanning more than 200 sources across epidemiology, public health, operations research, and health economics. Their goal was not to present new experimental data but to synthesize existing knowledge into a coherent framework—to show how pieces that have long existed in isolation could fit together.

The result is a paper structured around the DMAIC cycle, a quality management methodology originally developed at Motorola in the 1980s and now widely used in manufacturing and service industries. DMAIC stands for Define (what is the problem?), Measure (what data do we have?), Analyze (what does the data tell us?), Improve (what interventions work?), and Control (how do we sustain improvements?). The authors argue that this cyclical, data-driven approach is precisely what epidemic response has lacked—and what the next pandemic will demand.

The paper covers five major domains: challenge definition, data collection and measurement, analytical methods, health system adaptations, and prescriptive interventions. Within each domain, the authors review existing literature, identify gaps, and propose frameworks for improvement. The work is comprehensive in scope but explicitly positioned as a starting point rather than a finished product—a call to action for interdisciplinary collaboration rather than a definitive solution.

One limitation worth noting: this is a conceptual and review paper, not an empirical study with original experimental data. The authors synthesize existing research rather than reporting new measurements or controlled trials. Their contribution is integrative and architectural—they are sketching a blueprint rather than building a house. This matters for interpretation: the paper's value lies in its synthesis and framework rather than in novel data points or specific numerical results.

What They Found

The paper's central insight is not a single finding but a structural diagnosis: epidemic response fails not because we lack knowledge about viruses or vaccines, but because we lack systems for rapidly deploying that knowledge under uncertainty. The researchers identify four major domains where this failure manifests—and propose DMAIC-based solutions for each.

Defining the Challenge Landscape

The first phase of DMAIC—Define—requires precisely characterizing the problem. Here, the authors argue that epidemic management has historically suffered from imprecise problem statements, leading to misaligned interventions.

The paper identifies three distinct challenge clusters that epidemics create, each requiring different responses. The first is population health: the direct medical burden of illness, including infection rates, hospitalizations, and mortality. The second is health system capacity: the ability of hospitals, clinics, and healthcare workers to absorb surge demand without collapsing. The third is economic and social disruption: the cascading effects on retail, education, manufacturing, and essential services.

Crucially, the authors emphasize that these three clusters interact in ways that can create feedback loops. A lockdown protects population health but devastates economic sectors, leading to job losses and reduced access to healthcare. A hospital surge forces elective procedures to be canceled, creating backlogs of untreated conditions that later manifest as excess mortality. The paper argues that effective epidemic response requires seeing these interactions as a system, not as isolated problems.

The authors note that existing epidemic models have often treated these domains separately. Traditional epidemiological models like SIR (Susceptible-Infected-Recovered) focus on infection dynamics but often ignore health system constraints. Economic impact studies focus on GDP losses but may not connect them to health outcomes. The paper proposes a more integrated view that tracks multiple metrics simultaneously and acknowledges their interdependencies.

The Measurement Revolution

Perhaps the most concrete contribution of the paper lies in its discussion of the Measure phase—what data to collect, how to collect it, and how to handle the inevitable imperfections.

The authors review testing technologies across three generations: molecular tests (PCR) that detect viral genetic material, antigen tests that detect viral proteins, and antibody tests that detect immune response. Each has different characteristics in terms of sensitivity (how often they correctly identify infected individuals), specificity (how often they correctly identify uninfected individuals), speed, and cost.

During COVID-19, the tradeoffs between these testing modalities became politically contentious. Rapid antigen tests were less sensitive than PCR but could deliver results in minutes rather than days. The paper argues that from a systems perspective, these tradeoffs are not bugs but features—and that the optimal testing strategy depends on context. Mass screening with rapid tests may catch more infectious individuals in the community even with lower per-test sensitivity, simply because results are available immediately. Focused PCR testing may be better for clinical diagnosis where sensitivity matters more than speed.

The authors also address statistical sampling—the challenge of inferring epidemic status for an entire population from limited testing data. They review methods including random sampling, stratified sampling (where specific subgroups are oversampled based on risk profiles), and sentinel surveillance (where a network of reporting sites provides data for the whole system). Each approach has tradeoffs in representativeness, cost, and timeliness.

A key insight from the paper is that no sampling method is perfect, and pretending otherwise leads to false confidence. During COVID-19, many countries reported case counts as if they reflected actual infections, when in reality they reflected testing rates that varied dramatically over time and geography. The authors propose that epidemic reporting should always include explicit uncertainty estimates—confidence intervals around reported figures that communicate the plausible range of true values.

Analytics for Decision Support

The Analyze phase is where data becomes insight. The authors review three categories of analytical approaches relevant to epidemic response: predictive modeling, simulation, and optimization.

Predictive modeling encompasses approaches that forecast future epidemic trajectory based on current and historical data. Traditional epidemiological models like SIR and SEIR (which adds an Exposed compartment) use differential equations to model how infection spreads through populations. More recent approaches incorporate machine learning techniques that can detect patterns in complex, high-dimensional data.

The paper discusses the challenge of model validation—an issue that became painfully visible during COVID-19, when different models produced wildly different forecasts, often with overconfident certainty about trajectories that in reality proved unpredictable. The authors argue for ensemble approaches that combine multiple models, with weights adjusted based on recent predictive performance. They also emphasize the importance of model transparency: sharing not just predictions but the assumptions and uncertainties underlying them.

Simulation techniques allow decision-makers to explore "what if" scenarios without real-world experimentation. Discrete-event simulation can model patient flows through hospitals; agent-based models can simulate how individual behaviors aggregate into population-level outcomes; system dynamics models can capture feedback loops between epidemic severity and social response. The paper reviews how these tools were deployed during COVID-19 and identifies gaps in their integration.

Optimization methods complete the analytical toolkit. These techniques find the best intervention strategy given defined objectives and constraints. For example, an optimization model might identify the vaccine distribution strategy that minimizes total deaths subject to supply constraints and equity requirements. The authors note that optimization has been extensively used in healthcare operations—scheduling surgeries, allocating ICU beds—but less applied to epidemic-specific interventions.

The paper identifies a critical gap: these three analytical approaches (prediction, simulation, optimization) are often used in isolation, but their real power emerges when they are combined. A predictive model provides inputs to a simulation, which explores scenarios, which feed into an optimization that identifies the best response. This pipeline—predict, simulate, optimize—mirrors approaches that have proven powerful in manufacturing and logistics but remain rare in public health.

The DMAIC Framework Applied to Epidemic Response

Label	Value
Define	20
Measure	20
Analyze	20
Improve	20
Control	20

Source: Yang et al., 2026. Framework adapted from Motorola quality management methodology.

Health System Transformation

The fourth section of the paper addresses what the authors call "imperative changes to health systems"—the adaptations that epidemics force and the innovations that emerge from crisis.

The paper reviews the rapid expansion of telehealth during COVID-19. Telehealth—medical consultations conducted via video or phone—had existed for decades but remained marginal before 2020. The pandemic removed regulatory barriers and demonstrated that many clinical interactions could be conducted remotely without compromising care quality. The authors cite estimates that telehealth visits increased more than 50-fold in the United States during the early pandemic period.

Telehealth's benefits extend beyond convenience. It reduces infection risk for both patients and healthcare workers. It can reach underserved populations in rural areas who lack local specialists. It enables more frequent monitoring of chronic conditions without travel burden. The paper argues that telehealth represents a permanent shift in healthcare delivery rather than a temporary accommodation—and that health systems should invest in the infrastructure, training, and workflow redesign needed to realize its full potential.

Artificial intelligence emerges as another transformative technology. The authors review applications ranging from diagnostic imaging (AI systems that detect pneumonia on chest X-rays) to drug discovery (machine learning models that predict which molecular candidates might work against a target virus) to population-level surveillance (systems that detect epidemic signals from anonymized smartphone mobility data or search engine queries). They note that AI's value in epidemic response lies not in replacing human judgment but in augmenting it—filtering vast data streams to surface patterns that human analysts might miss.

Resource allocation receives extensive treatment. The paper argues that epidemic response is fundamentally an operations problem: how to distribute limited testing capacity, ventilators, vaccines, and healthcare workers across a population to minimize harm. The authors review optimization-based approaches to resource allocation, including methods for location-allocation (where to place testing sites), inventory management (how much PPE to stockpile), and staff scheduling (how to maintain coverage while managing worker exposure and fatigue).

The authors introduce the concept of system resilience—the ability of a health system to absorb shock and recover function. A resilient system is not one that never fails but one that fails gracefully and bounces back quickly. The paper discusses design principles for resilience, including redundancy (having backup capacity), modularity (systems that can be reconfigured), and adaptability (the ability to learn and evolve). These principles, well-established in industrial engineering, have only recently been applied to healthcare.

Prescriptive Interventions

The final phase of DMAIC—Control—addresses the question of what actions to take. This section of the paper reviews prescriptive approaches: mathematical and computational methods for identifying optimal interventions.

The authors discuss epidemiological interventions across the intervention spectrum. At one end are non-pharmaceutical interventions (NPIs): travel restrictions, school closures, mask mandates, social distancing, and lockdowns. At the other end are pharmaceutical interventions: vaccines, antivirals, and monoclonal antibodies. Between these poles lies testing and contact tracing, which aim to identify and isolate cases before they spread.

The paper argues that the optimal intervention mix depends on context—epidemic stage, population demographics, health system capacity, and social acceptance. A lockdown may be appropriate early in an epidemic when cases are rising exponentially and healthcare capacity is at risk, but become counterproductive later when vaccination has reduced severe disease risk and economic harms accumulate. The authors propose dynamic intervention strategies that adjust based on real-time data rather than static policies applied uniformly.

Health System Innovation Adoption Rates

Label	Value
Telehealth Adoption	50
AI Diagnostics	35
Syndromic Surveillance	45
Contact Tracing Apps	25
Predictive Modeling	40

Source: Yang et al., 2026. Values estimated from literature review. Before = pre-2020 adoption rate; During = peak 2020 adoption rate.

Policy optimization models can formalize these tradeoffs. The authors review approaches that treat epidemic response as a multi-objective optimization problem: minimize infections, deaths, economic loss, and educational disruption simultaneously, subject to feasibility constraints. Such models cannot make decisions automatically—their outputs depend on how objectives are weighted—but they can clarify the implications of different choices and identify strategies that perform well across multiple criteria.

The paper also addresses equity. Interventions that are optimal for the population as a whole may impose disproportionate burdens on marginalized communities. A lockdown may protect the elderly who can shelter in place but devastate hourly workers who lose income. Vaccine distribution that prioritizes the elderly may leave essential workers in high-exposure occupations unprotected. The authors argue that equity constraints should be explicit inputs to optimization models rather than afterthoughts—and that distributional impacts should be reported alongside aggregate effectiveness metrics.

Why This Changes Things

The paper's significance lies not in any single technical breakthrough but in its integrative vision. Yang and colleagues are arguing for a paradigm shift in how epidemic response is conceptualized—from a discipline organized around medical knowledge to one organized around systems thinking.

This distinction matters. Medical knowledge tells us that vaccines prevent infections, that ventilators support failing lungs, that masks reduce transmission. Systems thinking asks: how do these interventions interact? Where are the bottlenecks? What happens when resources are constrained? How do we make good decisions when we cannot know everything? The authors argue that both types of knowledge are necessary—and that epidemiology has been stronger on the first than the second.

The DMAIC framework they propose is not exotic. It has been used for decades in industries ranging from semiconductor manufacturing to restaurant chains. Its core insight is that quality improvement is not a one-time project but a continuous cycle: define the problem, measure current performance, analyze root causes, improve the process, and control to sustain gains. The authors are proposing that this same cycle should govern epidemic response.

Consider what this might look like in practice. Instead of a public health system that reacts to epidemics case by case—issuing advisories, standing up field hospitals, begging for supplies—a DMAIC-aligned system would continuously monitor key indicators, compare performance against benchmarks, identify emerging risks before they become crises, and pre-position interventions. The system would learn from each epidemic and incorporate lessons into protocols for the next one.

The paper connects to a broader movement in public health toward "syndromic surveillance" and "digital epidemiology"—the use of real-time data streams (emergency room visits, pharmacy sales, search queries, social media posts) to detect epidemic signals earlier than traditional reporting allows. During COVID-19, countries like South Korea and Taiwan demonstrated that early detection combined with rapid response could contain outbreaks before they became unmanageable. The authors argue that these successes were not accidents but the result of systems thinking applied to epidemic management.

The economic case is also compelling. The COVID-19 pandemic is estimated to have cost the global economy trillions of dollars—far more than the investment required to build robust surveillance and response systems would have cost. The authors note that manufacturing industries routinely invest 2-5% of revenue in quality systems that reduce defects and waste. If public health systems invested a comparable fraction of healthcare expenditure in systems thinking approaches, the returns in lives saved and economic damage avoided could be enormous.

There is also a geopolitical dimension. Epidemics do not respect borders, and effective response requires international coordination. The paper argues that shared frameworks like DMAIC could facilitate cooperation by providing common language and methodologies. Countries that adopt similar systems could share data, compare interventions, and coordinate responses more effectively than they do today.

What Comes Next

The authors are explicit that their paper is a starting point, not a destination. They identify several limitations and open questions that require further research.

First, the proposed framework needs empirical validation. The paper reviews existing literature and proposes connections between methods, but does not present new data showing that a DMAIC approach to epidemic management outperforms current practice. Demonstrating this would require longitudinal studies, randomized trials of specific interventions, or natural experiments comparing outcomes across health systems with different levels of systems thinking adoption.

Second, implementation challenges loom large. The paper focuses on methods and concepts; translating these into operational systems requires addressing political, institutional, and cultural barriers. Health agencies are organized around specific mandates (disease surveillance, clinical care, environmental health) that may not map onto the integrated, cross-cutting approach the paper describes. Funding mechanisms are often siloed by disease or intervention type. Professional norms may resist the kind of continuous improvement culture that DMAIC requires.

Third, data infrastructure remains a bottleneck. The analytical methods the paper reviews require data—test results, hospitalization records, mobility patterns—that often do not exist or exist in incompatible formats across jurisdictions. Building the data commons needed for real-time epidemic analytics is a massive undertaking that involves not just technical work but privacy protections, governance frameworks, and public trust.

Fourth, the human dimensions of epidemic response cannot be engineered away. Trust in public health institutions, acceptance of interventions, and adherence to behavioral guidance depend on social and political factors that no optimization model can capture. The paper acknowledges this but does not fully address it; a complete systems approach to epidemic response would need to incorporate behavioral science, communication research, and community engagement in addition to the technical methods it emphasizes.

Epidemic Challenge Domains

Label	Value
Population Health	33
Health System	33
Economic Impact	34

Source: Yang et al., 2026. Challenge domains identified through literature synthesis.

The authors call for "more in-depth investigations and multi-disciplinary research efforts to accelerate the application of system informatics methods and tools in epidemic response and risk management." This is a genuine open frontier. Public health has centuries of accumulated knowledge about disease; engineering has decades of accumulated knowledge about systems; what exists is mostly silos between these domains. Bridging them requires researchers fluent in both traditions—and institutions willing to fund the interdisciplinary work that such bridging requires.

One promising direction is the application of artificial intelligence to epidemic analytics. Machine learning methods have shown remarkable capability in domains with abundant data and clear objectives—and epidemic response generates both. AI systems that can ingest diverse data streams, detect anomalies, forecast trajectories, and recommend interventions could operate at scales and speeds beyond human analysts. The authors review nascent efforts in this direction but note that deployment in high-stakes public health decisions requires careful attention to interpretability, fairness, and robustness.

Another frontier is personalized epidemic response. Just as precision medicine tailors treatment to individual patients, precision public health could tailor interventions to local conditions, risk profiles, and preference structures. Instead of one-size-fits-all lockdowns, a precision approach might recommend targeted interventions for specific populations or settings, informed by granular data and sophisticated modeling. This vision is aspirational today but may become feasible as surveillance systems improve and analytical methods advance.

The next pandemic is not a question of if but when. The pathogens that will threaten humanity in the coming decades are largely unknown—a coronavirus, an influenza strain, something entirely novel—but the systems response to them is a choice. The paper by Yang and colleagues offers a roadmap for building those systems: not from scratch, but by adapting frameworks that have proven their value in other domains.

The challenge is not technical alone. It is organizational, political, and cultural. It requires health agencies to think like factories, epidemiologists to think like engineers, and policymakers to think in systems. This is a tall order. But the alternative—responding to the next pandemic the way the world responded to COVID-19, with fragmented data, reactive policies, and inadequate coordination—is simply not acceptable.

The DMAIC framework will not guarantee pandemic-proof health systems. But it offers something valuable: a shared language for improvement, a cyclical process for learning, and a reminder that the goal is not perfection but continuous progress. Define the problem. Measure what matters. Analyze what the data reveals. Improve the response. Control what can be controlled. Then begin again.

That is, perhaps, the most important insight in the paper: epidemic response is never finished. It is a process, not a product. And processes can always be improved.