Group testing—the practice of pooling multiple individual samples into a single diagnostic test—is a remarkably powerful tool for resource conservation. Imagine screening a population of 100 individuals where only one person is infected. By dividing the population into 10 pools of 10, you can clear 90 people in just 10 tests. The remaining positive pool requires 10 individual tests to identify the infected individual, bringing the total to 20 tests instead of 100.
As demonstrated by Claudio Mayrink Verdun et al. in Group Testing for SARS-CoV-2 Allows for Up to 10-Fold Efficiency Increase Across Realistic Scenarios and Testing Strategies (Frontiers in Public Health 2021), this adaptive pooling strategy can scale up screening capacity by up to 10-fold under low-prevalence conditions. Similarly, Curtis Tatsuoka et al. show in BAYESIAN GROUP TESTING WITH DILUTION EFFECTS (2021) that a baseline infection rate of 2% allows adaptive systems to reduce the total test load more than eightfold.
However, this efficiency relies entirely on a critical assumption: that the target condition is rare. When the infection rate rises, group testing encounters a mathematical and physical wall known as the prevalence threshold.
Why Pooling Collapses: The Re-testing Explosion
To understand why group testing fails at higher infection rates, consider the probability that a pool of size $k$ returns a positive result. If the individual infection probability (prevalence) is $p$, the probability that a pool contains at least one positive sample is:
$$P(\text{Positive Pool}) = 1 - (1 - p)^k$$
As $p$ increases, this probability grows exponentially. If $p = 20\%$, a pool of 10 samples has an $89.3\%$ chance of being positive. In a laboratory setting, this means nearly every single pool will trigger secondary individual re-testing. Instead of saving resources, the lab must perform the initial batch of pool tests plus individual tests for almost the entire population. The strategy becomes more resource-intensive and logistically slower than testing everyone one-by-one from the start.
This phenomenon creates a strict efficiency ceiling. Paola Betancourt Ruiz et al. note in El papel de la ingeniería en la pandemia de COVID-19 (Revista 2020) that high infection rates neutralize the throughput gains of pooling due to this constant sub-testing loop. In A Recovery Algorithm and Pooling Designs for One-Stage Noisy Group Testing under the Probabilistic Framework (2021), Yining Liu et al. identify a clear performance ceiling around a 10% prevalence threshold, beyond which pooling complexity yields diminishing returns. When simulating large-scale urban screening, Douglas K. S. Ng et al. in Implementation of the compulsory universal testing scheme in Hong Kong: Mathematical simulations of a household-based pooling approach (Frontiers in Public Health 2022) place the absolute logistical ceiling near 17%, demonstrating that past this point, the overhead of managing multi-stage pools completely overwhelms laboratory infrastructure.
Physical and Information-Theoretic Boundaries
The threshold is not merely a logistical inconvenience; it is bounded by fundamental physical and information-theoretic limits.
The Dilution Effect
Physically, pooling multiple samples dilutes the target biomarker (such as viral RNA). Rahil D Morjaria et al. explore this in Density-Dependent Group Testing (aistats 2025), proving that severe dilution creates a mathematical boundary where the information gained from a pool drops to zero, making individual testing the only optimal choice. Conversely, the authors of Group testing via hypergraph factorization applied to COVID-19 (Nature Communications 2022) (David Hong et al.) argue that in severe resource crises, maximizing total population screening throughput can justify accepting this marginal loss in individual analytical sensitivity.
Information-Theoretic Limits
From a pure data perspective, pinpointing the exact crossover point where pooling loses its advantage remains a core challenge. Abhishek Agarwal et al. establish foundational impossibility results in Novel Impossibility Results for Group-Testing (arxiv 2018), mapping out the defect densities where individual testing becomes unbeatable. To resolve this transition zone, Larkin Flodin et al. in Probabilistic Group Testing with a Linear Number of Tests (arxiv 2021) analyze the "mildly sublinear" regime—where the number of positive cases scales relative to the population size divided by its logarithm—to define the precise mathematical boundary where collective sampling minimizes the total test count. Additionally, Dariusz Rafal Kowalski et al. show in Scalable and Efficient Non-adaptive Deterministic Group Testing (neurips 2022) that when dealing with capped feedback models, a distinct efficiency transition occurs when the feedback threshold reaches the square root of the target set size.
Strategies for Pushing the Ceiling
To maintain the benefits of group testing as prevalence rises, researchers have developed three primary mechanisms to bypass or shift the efficiency threshold.
1. Prevalence-Adaptive Batch Sizing
Instead of using static pool sizes, systems can dynamically shrink or grow pools based on real-time epidemiological data. Balram Rai et al. demonstrate in Pool Testing for COVID-19: Suitable Splitting Procedure and Pool Size for India (Disaster 2020) that calculating the expected number of tests as a direct function of individual infection probability allows labs to optimize the "sweet spot"—keeping pools large enough to save kits but small enough to avoid constant re-testing. This power-law relationship between prevalence and pool size is leveraged by Li Zhang et al. in Exploring Chlamydia trachomatis screening in Shenzhen: a cost-effective sample pooling strategy for pre-marital and pre-pregnancy populations (Frontiers in Public Health 2025) to minimize the total test load across massive urban cohorts.
2. Informative Triage (Smart Pooling)
If you can predict which individuals are highly likely to be positive, you can alter the math entirely.
[Population] ───> [ML Triage Model] ───> High Risk ───> Individual Testing
───> Low Risk ───> Highly Efficient Pools
In Smart pooling: AI-powered COVID-19 informative group testing (Scientific 2022), María Escobar et al. utilize machine learning on clinical and sociodemographic data to triage high-risk individuals into immediate individual testing. By removing these high-probability samples, the effective prevalence of the remaining population drops significantly. This ensures that the resulting pools are highly likely to be entirely negative, preserving pooling efficiency even during rampant outbreaks.
3. Exploiting Community and Network Structures
Standard group testing assumes that infections are independent random variables. In reality, infections cluster within families, classrooms, or contact networks.
Arlei Lopes da Silva et al. address this in Group Testing on a Network (aaai 2021) by formalizing pool formation as a graph partitioning task based on transmission networks, maintaining testing utility across high disease prevalence. Similarly, Pavlos Nikolopoulos et al. demonstrate in Group testing for connected communities (aistats 2021) and Community-Aware Group Testing (IEEE Trans. Inf. Theory 2023) that grouping highly correlated individuals allows the system to isolate positive clusters collectively. This community-aware approach reduces required tests by 55% to 75% compared to traditional methods, preventing the efficiency collapse that occurs when high-prevalence populations are pooled blindly.
Optimal Design and Decoding
To achieve these theoretical limits, the underlying pooling matrices and decoding algorithms must be precisely calibrated. Dino Sejdinovic et al. establish in Note on Noisy Group Testing: Asymptotic Bounds and Belief Propagation Reconstruction (arxiv 2010) that the optimal probability for including any single item in a test pool is inversely proportional to the expected number of defective items in the population ($1/k$). When this density is maintained, message-passing algorithms like Belief Propagation can rapidly decode results even in noisy environments.
Finally, ensuring that these designs remain robust across shifting defect rates requires robust mathematical frameworks. Huseyin A. Inan et al. demonstrate in On the Optimality of the Kautz-Singleton Construction in Probabilistic Group Testing (arxiv 2018) that classic deterministic structures like the Kautz-Singleton construction achieve order-optimal performance when evaluated under a probabilistic covering framework, enabling sub-linear time decoding across varying growth patterns of infection.
Key Foundational Work
Compressed Sensing with Probabilistic Measurements: A Group Testing Solution (2009) — contributed 1 concept elements to this direction.
Statistical physics of group testing (2008) — contributed 1 concept elements to this direction.