A blueprint for a testing strategy| Analysis
There is a way to minimise transmission risks and maximise detection of hidden casesUpdated: Apr 22, 2020 18:49 IST
India has now been locked down for a month and, finally, there is a large supply of testing kits, though their quality is, apparently, a real concern. Lockdowns are the best time to test, since movement and new contacts are minimal. While it may be necessary to tailor testing strategies to the states’ local context, are there common principles that should govern when, where, how and whom to test?
Since mid-March, testing has increased 30 times, to around 30,000 tests daily. But, even if a million additional tests were conducted, it would amount to only one test per 1,000, keeping India a low-test country. So, even without quality issues, efficient test strategies are needed to learn quickly and to achieve the two key goals of testing — first, finding infected persons, even if they are very few and asymptotic, and to treat them and prevent transmission; and second, generating data to implement a smart containment strategy, reducing the need for wholesale lockdowns.
Should states begin by testing in hotspots? If people test positive and are asymptomatic, they will be isolated and hospitalised, if symptoms develop. Ideally, testing should change action, but, in a hotspot, they are quarantined anyway.
But, hotspot testing can be informative. If positivity is less than expected, or clustered, it provides information about transmission; if higher, it signals a surge in demand for hospitalisation. This information is best obtained by antibody tests, if accurate, but learning is possible only with proper randomisation, i.e., testing as per a pre-decided statistical plan. Within the hotspot, more testing further from the core, where less infection is expected, can indicate the extent of spread. Importantly, it can also help decide when an area is no longer a hotspot.
But, focusing on current hotspots alone cannot find infected persons in areas where the virus has yet to be reported. Can testing prevent future hotspots by locating asymptomatic infections?
The Indian Council of Medical Research (ICMR)’s current recommended testing strategy only allows testing of asymptomatic persons if they are direct and high-risk contacts of a confirmed case. But, ICMR also says that about 69% of confirmed coronavirus disease (Covid-19) patients are asymptomatic. Considering that the surveillance of Severe Acute Respiratory Infections (SARI) patients indicated a positivity of over 2%, we may miss many infected persons. If even 0.1% is Covid-positive and asymptomatic, 20,000 infected persons could be infecting others in Delhi alone, as the lockdown is eased.
Just going beyond ICMR’s recommendations, and testing a sample of the population, is likely to find few infected persons. One of us has computed that if 0.1% of 10 million are infected and 1,000 tests are randomly administered, the chance of not finding even one positive case is more than 40%! An efficient testing design should thus maximise the chances of finding an infection, especially those more vulnerable to the disease.
This can be done in three ways.
First, test asymptomatic individuals who can become “super-spreaders”, i.e those susceptible to infection, but who interact frequently with others, even during lockdowns, such as health workers and the police, but also civic workers in essential services and street vendors. Many in these groups were found to be infected after they became symptomatic. These groups can be tested at work, using Reverse Transcription Polymerase Chain Reaction (RT-PCR) methods, with samples pooled for those who work together. If needed, their contacts can be traced to where they live. Some states already plan to do this. Such potential super-spreaders can also form a sentinel network that is checked weekly for symptoms.
Second, do risk stratification of areas, ie demarcate areas with initially high expected risk of transmission and/or high vulnerability such as dense cramped settlements, with larger numbers of elderly people and, this is very important, choose persons, within areas, in an explicit statistically-structured randomised manner and test them using RT-PCR methods. The chances of finding infected persons can improve by first using local information — people reporting symptoms of influenza-like illness (ILI), and then, drawing randomly from the area’s voter list.
Along with data on occupation, age, gender, and any morbidities and/or recent illnesses, basic information on the intensity of contact with others should be collected from all those tested, including in hotspots. Despite increased testing, the share testing positive has stayed steady around 4.5% — every 100 people tested led to five new Covid-19 patients. Is this because high-risk contacts are selected or because prevalence is worryingly high? If it’s, hopefully, the former, it means that contact tracing processes are not standardised in practice, since test positivity varies considerably among states.
Third, states should anonymise and release this data, along with associated test results. The randomisation would enable real-time, high-quality analysis of possible determinants of transmission, even in places not directly tested. Information — this is important — from the initial stages must be used to iteratively improve sampling designs after each round of tests. This can be analysed internally, but crowd-sourcing will exponentially increase the speed and quality of analysis, thus allowing a better understanding of the disease and improving policy response.
This suggested ICMR-plus strategy tests more asymptomatic persons and uses structured randomisation, to enable active learning by governments. It minimises the risk of transmission from super-spreaders, maximises the chance of detecting hidden infections, and shapes strategy in real-time for future testing, to optimise scarce medical and testing resources. Importantly, it can also calibrate containment strategies. States should consider using these principles as they embark on their testing journeys, which may, sadly, last longer than expected.