The growth rate of weekly average of the number of positive cases, Covid diagnosed/suspected deaths and hospital admissions need to inform unlock decisions
Nobody likes a lockdown. Not those who advocate them, impose them, implement them or suffer them. Yet, lockdowns have come to be seen as essential, if not inevitable, when transmission rates surge skywards and hospitals get overwhelmed with a deluge of desperate patients. Once implemented, what are the indicators that we need to track to give us the signal that it is safe to unlock?
The trackers that countries have used have varied over time. Last year, the UK and several European countries used a steady fall in daily death counts to decide on the time to open up. This year UK has been using hospital admission rates and infection counts to guide the decisions on the timing and extent of unlocking. The pressure on the hospital system is the main indicator now.
One indicator that has been widely discussed, in the Indian context, is the Test Positivity Rate (TPR). This indicates the proportion of viral detection tests which are reported as positive during a defined time period. The World Health Organization provided a guideline of TPR below 5% as the indicator of adequate testing in any country. Policymakers from India have varyingly set TPR thresholds of 10% or 15% for imposing a lockdown in ‘high intensity’ districts.
However, a scientific advisory from the Corona Resource Centre of the Johns Hopkins University unequivocally states: “It is important to note that test positivity is a measure of testing capacity and while it can provide important context about case totals and trends, it is NOT a measure of how prevalent the virus is in communities. Policy decisions, like openings and closings or interstate travel, should not be determined based on test positivity alone” (highlighted as in the original).
There are several methodological reasons as to why test positivity rates are to be viewed as vulnerable to inaccuracies in estimating the number of currently infected persons. The Hopkins paper points out that there are four different methods of estimating TPR, depending on how we define the numerator and the denominator. Each of them gives a different result. Do we count the number of positive tests or positive persons in the numerator? What do we do with persons who have been repeatedly tested during a short time period? These are only some of the questions that account for variability in the results.
The tests also tend to underestimate the infected cases, as false negative results are associated with each testing method. The RT-PCR test, which detects replication of the viral nucleic acid, is conventionally taken as the gold standard. The Emergency Use Authorisation (EUA) issued by the US Food and Drugs Administration Agency (FDA), to the Laboratory Corporation of America (Labcorp) for RT-PCR testing to detect the SARS-CoV-2 virus, states: “Negative results do not preclude SARS-CoV-2 infection and should not be used as the sole basis for patient management decisions. Negative results must be combined with clinical observations, patient history, and epidemiological information.”
This wisdom is not new. Labcorp stated this clearly over a year ago in 2020. Yet, we have till recently insisted on a positive test result as a prerequisite for admitting a symptomatic patient to hospital. Sections of the media speculate that false negative test results are a unique feature of the recent variants during the second wave. We know that RT-PCR is less likely to detect the virus early or late in the infection when the virus is not replicating in copious numbers. Other factors that affect the test result are the ability of technicians to collect a good sample on a nasal or throat swab, properly transport this to the laboratory and safely store this prior to testing. With all these constraints, a single RT-PCR test usually delivers with a sensitivity of 60-70%. It means it can miss 30-40% of infected cases. It is suggested that repeated testing, during the clinically infected period, can enhance the sensitivity by having more tries at capturing an actively replicating virus. Unfortunately, there can be some false positive tests too, due to nucleic acid fragments of ‘dead viruses’ which may be detected even if there is no active infection.
However, the RT-PCR tests are time-consuming and do not deliver results for several hours. When the number of samples rise and accumulate, the delay in processing and reporting can stretch to several days. This delays decision making. So, Rapid Antigen Tests (RAT) have been advocated for use instead. These detect protein antigens of the virus, rather than the nucleic acid. The processing time is short and results can be delivered in 30 minutes. This has led to a preference for RAT as the screening test when there is a surge in suspected cases. That is why Indian authorities have recently recommended that a higher proportion of RAT must now be employed across the country, in comparison to RT-PCR.
The RAT comes with the disadvantage of even lower sensitivity than RT-PCR. Compared to that ‘gold standard’, the RAT has been reported to have sensitivity values in the field of 50-80%. If we assume that RT-PCR has a sensitivity of 70% in a single-time testing and that RAT has a sensitivity of 70% compared to RT-PCR, we will miss half the infected cases by depending on a single RAT result. However, advocates of RAT argue that the test is easier to repeat twice or thrice over a week to reduce the number of missed cases. Can we really do that, during a surge? In a hospital setting perhaps, but less likely in a community setting.
The number of tests performed daily may also vary week to week. If initially fewer tests are done, and, later, the numbers rise substantially, it may mean that the criteria for deciding who to test may have been relaxed. Earlier, only persons considered to have a high probability of infection may have been tested, while the tests may have later been extended to include low-probability persons too. This is especially likely with ‘on demand’ testing. The test positivity rates will drop in such cases, even if the trajectory of the epidemic has not changed. If the switch from RT-PCR to RAT reduces the test positivity rate at least by a third and liberal use of testing in low-probability individuals reduces test positivity rate by a quarter or more, the combined impact on the test positivity rate can be substantial and misleading.
The TPR is a useful index, but cannot be the sole tracker. Note the caveats against “alone” and “sole” in the Johns Hopkins and FDA documents. The TPR has to be supplemented with other types of data and their time trends. These indicators must include moving weekly averages of new hospital admissions of test positive and clinically highly probable cases (overall admissions and those needing intensive care), new deaths of diagnosed or highly likely Covid cases. It will be difficult to track persons on home care. However, deaths out of hospital must be subjected to a quick symptom based verbal autopsy interview of the attendants.
These additional indicators are not easy to collect and each comes with its own uncertainties and inaccuracies. Last year, I held the view that the noise-to-signal ratio, in terms of variable errors of estimation, was less for Covid deaths than for case-counts which were vulnerable to frequently changing testing numbers, methods and mandated versus elective eligibility criteria. However, the large number of unclassified out of hospital deaths occurring this year increases the uncertainty of the error estimate in Covid deaths.
So, we will have to use a combination of several trends to make the decisions on when and how much to open up or unlock in stages. The growth rate in the weekly average of the number of new persons testing positive (rather than positive tests as a fraction of all tests performed, once or many times in new and old cases) must be one indicator. The moving weekly average of Covid diagnosed or highly suspected deaths (medical certification and verbal autopsy) must be another. Hospital admissions (especially those requiring oxygen or mechanical ventilation as per standard criteria) must be another indicator.
While each of these will not yield precise estimates of reality, they will indicate the trend of the epidemic if they all move in the same direction. That will provide greater confidence than relying on TPR alone. It is better to use the numbers of a combination lock to open than to rely on a single, rusty key.
The author, a cardiologist and epidemiologist, is president, Public Health Foundation of India (PHFI)
Views are personal