Seroprevalence surveys, or sero-surveys in short, to assess the prevalence of antibodies against COVID-19 in various populations are catching up. After Delhi, Ahmedabad and Mumbai also recently released the early findings of their respective surveys, and surveys in other cities are underway. Officials are also planning to repeat surveys on a monthly basis to better understand changes in the populations’ antibody statuses.
Nearly 23% of individuals sampled in Delhi and 17% in Ahmedabad were found to have antibodies against COVID-19. Mumbai’s authorities reported that 57% of those from slums and 16% of those from three wards in other residential areas tested positive.
These positivity rates in three large Indian cities are far higher than those reported from other international surveys. What could explain this discrepancy? Some have been arguing that these cities might be moving towards a ‘herd immunity threshold faster than cities are doing so elsewhere. But before we can make any such assertion, we should carefully consider and contextualise the findings.
First, it is epidemiologically known that when the prevalence of antibodies is at the lower end, a community based sero-survey of 10,000 to 30,000 people is likely to report a higher prevalence. This will happen even if a very reliable test, with 95% sensitivity and 90% specificity, is used. In such a situation, when the actual prevalence is 10%, the sero-survey’s observed prevalence could be nearly 17% to 19%. So we need to exercise caution with the results.
Second, the sampling methods adopted in these surveys will also influence the final outcome. Therefore, the sampling approach should ensure the individuals to be surveyed are proportionately represented by age, social and economic status, educational level, areas of residence and other important parameters. For example, if disproportionately more working-age individuals and fewer children are included, or if disproportionately more samples are collected from containment zones, the survey will obviously report a higher prevalence.
So the people organising the survey should ensure the cohort to be tested is representative of the larger population to a sufficient level – such as at the ward level in Mumbai or at the district level in Delhi.
To ensure the results are reliable and will be interpreted properly, the organisers should also release the survey’s findings along with the statistical range – specifically, the 95% confidence interval – to provide a sense of the accuracy. For example, if 23% of the people sampled in Delhi’s sero-survey tested positive for antibodies to COVID-19, the true prevalence could range from 14% to 27-28%. So the sero-survey’s findings should be released with details of the sampled population and sampling methods adopted. The age distribution and representatives to the population are key if we need to correctly interpret the data.
There also needs to be more scientific rigour. The people involved in the Mumbai survey reportedly planned to collect 10,000 samples but ended up collecting only 7,000 due to administrative challenges. This may seem innocuous – but in scientific research or surveys, this could become a major limitation that skews the intended distribution of the subjects. With a smaller sample size, we can’t know if the population subgroups were proportionately sized as planned or if a subgroup was overrepresented.
If 15% of a city’s population resides in slums, 15% of the survey’s cohort should also be from slums. Mumbai’s sero-survey concluded that 57% of the people from slums and 16% of the people from the other residential areas had antibodies to COVID-19, the total prevalence wouldn’t be a simple average of 36% [(56+16)/2]. Instead, it would be a proportionate average of 22% [(56*0.15) + (16*0.85)].
Sero-surveys require people to volunteer their respective samples, and some people from the chosen cohort are likely to have refused. Those who agree to volunteer their samples are likely to be different from those who refuse because those who believe they have had COVID-19 in the past are more likely to be willing, as they might be keen to know their statuses. However, this can distort the positivity rate to be higher. So the refusal rate in a population survey should be revealed to help contextualise the findings.
Another aspect is the comparability. If different surveys adopt different sampling methods, their findings are unlikely to be comparable. For example, the first sero-survey in Delhi was led by the National Centre for Disease Control (NCDC), which developed the survey methodology. The methods for the next round are reportedly being developed by Maulana Azad Medical College, Delhi. The methods and sampling approaches of the old and new surveys should be harmonised to ensure their analytical methods and results can be compared with each other.
In addition, not every round of a sero-survey needs to have a similarly large sample size of 20,000 or so. Subsequent rounds could be smaller. In the long run, policymakers should simply be able to get a broad view of the country’s states and cities in a comparable manner. Institutes like the Indian Council of Medical Research, NCDC and/or others could consult with epidemiologists to draft a common protocol, sample sizes for multiple rounds, sampling methods, analytical techniques, etc. Having such a common set of guidelines will help keep the results comparable and allow multiple surveys to pool their findings and so develop a comprehensive national picture.
The sero-surveys to assess the prevalence of antibodies against COVID-19 are a good start. All of us are looking forward to achieving the herd immunity threshold in various populations. These surveys can provide useful insights on this front, and also guide policymakers to develop suitable interventions and optimally roll-out any vaccines as and when they become available. This is how India will surmount its COVID-19 epidemic – with a coordinated policy response led by science.
Dr Chandrakant Lahariya is an epidemiologist and public health specialist. He tweets at @DrLahariya. The views expressed here are the author’s own.