One of the best ways the world has to get a clear overview of COVID-19 is being underused. It’s time to harness the power of random sampling.
Last September, the U.S. Centers for Disease Control and Prevention estimated that only one in four SARS-CoV-2 infections in the United States had been reported. Across Africa, the average is closer to one in seven. Why? Many people who are quite ill or worried about their symptoms cannot be tested. Those with mild or no symptoms often do not seek testing.
And the minority is getting worse. Reinfections and breakthrough infections are on the rise, but they are often mild, so people go untested. The onslaught of Omicron cases has far exceeded the testing capacity of many countries. Last December, a test site near me in Atlanta, Georgia, had a waiting time of three to four hours. In the US, lateral flow tests at home are finally becoming more available, so fewer people will seek PCR confirmation.
All this sub-count makes many important questions unanswerable. For example, if an increase in case decreases, is the transmission down, or is the test maximized? Waiting to find out means the hospitals can not prepare and the politicians are two to four weeks behind. Who can drive and see only in his rearview mirror?
Wastewater monitoring is an innovative part of the solution. It shows whether virus levels are rising or falling across a community, and it does not depend on people searching or reporting test results. In my home state of Massachusetts, wastewater was one of the earliest reliable indicators that infections were declining last month.
But wastewater cannot pinpoint who in a society gets infected and who gets sick. With Omicron, hospitalizations of children have reached record highs. Yet infections in this age group are often missed. Obviously there are more infections, but are these infections more serious? Knowing the importance of risk-benefit calculations around schooling, vaccinations and more.
Samples can answer that kind of question. As long as the participants are selected at random, they will, on average, mimic the characteristics of the wider population. Roughly speaking, testing fewer than 1,000 people can provide crucial information about 10 million or even more.
Illustrative examples of random sampling are the Coronavirus (COVID-19) Infection Survey, which is operated across the UK by the Office of National Statistics (ONS) and the Imperial College London REACT-1 study. The ONS initiative aims to achieve grafting test results at least every fortnight from around 180,000 people across the UK and blood tests every month from around 150,000 people. At the end of January, one in 20 people tested positive for current infection. But age really did matter: One in 10 of the youngest children tested positive, as did one in 15 of the older children. The results signaled a huge pool of infections and were quickly made available to guide policy and family decisions.
Predicting the course of the pandemic requires reliable estimates of current levels of infection. Without accurate knowledge of these levels, epidemiologists have to make many assumptions (with the likelihood that, for example, those infected will develop symptoms or be tested). That guesswork informs mathematical models and consequently public discussions about the trajectory of the pandemic. Models that overestimate the number of infections that have been missed overestimate the immunity of the population and may underestimate the risk of recurrence. These estimates are used for decisions on everything from opening schools to planning policies and targeted vaccination campaigns. Without random sampling, there is a vicious circle of guesswork.
The UK data is informative elsewhere, but generalizing too much from one country’s data is dangerous. In the United States, a few randomized trials have been conducted by health departments and academic partners, for example in Indiana, Georgia, and California. These have strengthened the local understanding of differences across races and ethnic groups. At the national level, researchers at Emory University in Atlanta (where I also work) conducted a representative household survey (PS Sullivan et al. Clin. Infect. Haze. https://doi.org/hfvm; 2021). A new round of antibody and nose inoculation tests is performed every four to nine months. But a rapidly evolving situation requires more frequent testing.
Why does random sampling for infection no longer occur? These studies require sustained resources and coordinated efforts. The patchwork of the US public health system makes cross-state cooperation challenging. The surveys also require a public that is willing and able to participate. Low participation in surveys is a major challenge. As an incentive to participate in the ONS survey, the UK government has offered more than £ 200 million (US $ 270 million) in purchase vouchers.
More than two years into the COVID-19 pandemic, it is clear that the SARS-CoV-2 virus will be circulating for a long time to come. Millions of people are infected daily, and the threat of new varieties threatens. Investing in samples can better prepare governments for the future. A single sampling frame can be used for multiple pathogens, such as influenza and other respiratory viruses. For infectious diseases, it will mean bad decisions if one cannot see the whole picture. Yes, random sampling will cost, but bad information is also expensive.
The author declares no competing interests.