Figure 1: Test Positivity for Covid-19 in India over time (this chart as of November 27 2020)
(Source: ICMR and Ministry of Health and Family Welfare)
This graph shows the best possible estimate of test positivity from available official data.
Two months ago, following a piece here, I started posting the graph of cumulative and daily positivity. It was just beginning to turn up. Since then, a steadily rising daily positivity means that a higher proportion of people tested are testing positive for CoVID-19. Testing has gone up almost ten times since the end of April, increasing week after week, and even more recently. But, as shown in Figure 4, the growth in cases was consistently around 1% (± 0.1%) more than that of the growth in testing for the past six weeks, till the last week of July. Coupled with increased testing, the number of people testing positive are thus growing between 3% and 4% every day for the past month. The difference between 3% & 4% per day is this. At 3% one case becomes 42,000 at the end of a year, at 4%, it becomes 1.6 million.
This pattern has changed in the last week of July when this difference has started to decrease rapidly, with sharp increases in testing.
The daily positivity also had a weekly peak on Monday. Since 11 May, as shown in Figure 5, every Monday peak bar one, has shown a higher daily positivity than the previous Monday. The day of weekly minimum varies, but every minimum too has been higher than the previous minimum.
The weekly peak is caused by a fall in testing on Sunday, without a corresponding and proportionate fall in cases. This points to an interesting feature, which indicates that samples of people more likely to be positive are prioritised for testing, since persons more likely to be positive are tested even on days when a smaller number of samples are tested – which can only reflect the knowledge of frontline workers or a prioritisation of testing by category of sample, e.g., samples of high risk contacts are prioritised in testing, both of which are to be appreciated.
This pattern too has changed in the last week of July when this difference has started to decrease rapidly, with sharp increases in testing.
Comparability or lack thereof
This is actually raising an issue with the comparability of numbers in the daily updates, which have become less comparable with past data The sampling numbers reported by ICMR has been erratic recently with major ups and downs. One major factor behind such as pattern is that it now includes significant shares of antigen testing. While some states give a break-up of the kind and number of tests, ICMR reportedly only collates data for cumulative number of tests all inclusive and number of tests done in 24 hours after receiving data from state.
In Delhi, antigen testing is now 70% of all testing. In UP, it appears to be at similarly high levels. This increases sampling and decreases positivity, since sensitivity (ability to identify infected cases) of antigen tests is about half that of RT-PCR tests. This gives the false impression of falling positivity, and leaves a number of undetected persons in the population, who will infect other people. Given assumptions on relative sensitivity, and share of antigen testing, “true” positivity can be estimated.
ICMR can separately report antigen testing and positives from antigen testing by analysing data from its portal, even if states do not report separate data. Portal numbers may not tally with states, but that should not be a barrier as long as there is clear acknowledgement of the different sources of data.
States are expanding testing at a near identical pace
Before this confounding of data, the massive expansion of testing from around 40,000 at the end of April to around 400,000 by mid-July was almost proportional across states. In Figure 3b, which updates Figure 3 and Figure 3a, we see in that the difference in a state’s share in testing between June 10 and July 10 is minimal, with most states clustering on the 45 degree line and the slope coefficient close to 1. The improved explanatory power of the regression indicates that deviations from the overall trend are low. The stable structure that seemed in place last month seemed to be holding up.
This contribution from all states is also something to be noticed positively.
Note that though some states that are not testing enough (high positivity indicates the need for more testing), they are however testing a lot, like Maharashtra and Tamil Nadu. Karnataka, given its recent surge perhaps needs to test more. This is something that would benefit from coordination.
Test positivity continues to trend upward in almost all states
Figure 2b updates Figure 2 and 2a and shows that this rising test positivity is true for almost all states – few states are below the 45 degree line, i.e., the cumulative test positivity on July 10 was more than that on June 10, which indicates that the national upward drift is not concentrated in a few states, though high-positivity states like Maharashtra, Delhi, Gujarat and Tamil Nadu may be more affected. The fact that the slope coefficient is more than one implies that the increase is more, if the original value was more. The improved explanatory power indicates that deviations from the broad pattern of June 10 are further reduced over the period between May and June.
A rising tide of infection is thus affecting all states. This becomes more clear when we look at Figure 6, which shows the share of the top 1% of districts (just seven districts – Delhi and Mumbai are considered a single district for this purpose) in the number of affected cases. Immediately after the end of the first month of lockdown, the share of these seven districts increased substantially, reaching almost 60% of all infections at its peak. However, over June, their share dropped steadily but slowly, coming to about 54% by the end of June. After that, there has been a precipitous fall. In the last three weeks, the share has come down to 40%. This is also true if one takes the top 7 districts as of July 15 instead of May 1, indicating that the shift is not within the top – it is not the result of new high infection districts like Bengaluru replacing old high infection districts like Delhi and Mumbai (both of which remain comfortably in the top 7). Indeed, applying the Herfindahl Hisrchman Index (a measure used to measure concentration of market share by firms) indicates that the concentration of the infection is indeed dropping, a little slowly during June and rapidly in July.
The other side of the concentration coin is spread. The infection is now distributed much more across many other districts. In May just over 400 districts reported a case. Now it is reported from almost all districts. The share of infections in districts with less than 0.1% of cases has grown from 8% to 18%. On May 1, just 42 districts had more than 100 cases. On July 15, that number was 552.
The real lost opportunity is that we did not invest the necessary additional resources in controlling the infection at that time by supporting these districts with resources to test, trace and test, and supporting those who needed to be treated and isolated. Could that have arrested this relentless increase?
One can only hypothesise, and that will have to be another post.
`ICMR is currently testing over 90 thousand samples a day, a huge increase from one thousand a day in mid-March. This does not mean persons – because some of these samples are from people already tested positive who need to be tested again to see if they have recovered or whether they are still positive.
Ideally, the test positivity would be the number of persons testing positive divided by the number of persons tested, but this data is not released by official sources (it used to be). Table 1 below shows the kind of information related by ICMR at different times. Between 9 to 19 April, ICMR released the most information, number of total samples tested, the number of total people tested, the number of persons found positive, but since 25th April, they have only released the total number of samples tested. The MOH&FW used to also release the data on number of positive persons, which did not always match with ICMR. Since 25th April, only MOH&FW releases this data. The MOH&FW website continues to say their “figures are being reconciled with ICMR”.
The ICMR data can have shortcomings, e.g. there may be mis-identification of the location, duplicate records due to the multiple samples per patient, non-use of the unique ICMR ID for subsequent tests (see the explanation given by Brihanmumbai Municipal Corporation), but it is telling that reconciliation is proving extremely difficult and even after more than three weeks, it has not been possible to reconcile the data.
Test positivity is trending upward in almost all states
The updated Figure 2a shows that test positivity has moved up in almost all states – which would be the case, if the national test positivity is steadily trending upward. Within these, some states, notably Madhya Pradesh, West Bengal, Chhattisgarh and especially Punjab, the test positivity is trending down. The improved explanatory power indicates that deviations from the broad pattern of May 10 are lesser over the past month than the churn between April and May.
States are expanding testing at a similar pace
Compared to the earlier graph, Figure 3, we see in Figure 3a that the increase in testing between May 10 and June 10 has been much more proportional across states, with most states clustering on the 45 degree line and both the coefficient getting closer to 1 and the explanatory power of the regression increasing, indicating that deviations from the trend are not large. This is a welcome trend, subject to qualifications below.
Discourses that excoriate states for not testing enough miss this trend. Yes, some states needed to catch up, from their initial deficit in testing capacity, but by 10 May, there was a stable structure that seems to have been put in place.
Maharashtra, Delhi, Bihar, Gujarat and Tamil Nadu perhaps need to be testing more, given their rise in test positivity but we see that their share of testing has either remained same or has gone down. That would need attention and importantly, support – so that unused testing capacity in other states can be put to use. West Bengal, on the other hand, has both increased its share of testing and shown a reduced positivity, indicating that they appear to have managed the infection spread from returning residents thus far.
What does it mean if the test positivity remains constant, as the number of daily tests are increased almost 100X?
One would expect that as more people are tested the share of those infected would fall, but that has proved not to be the case at the national scale, though in many states, as seen in Figure 2, there has been a decline in test positivity over time. All the states below the 45 degree line show a lower test positivity on May 10, than on April 10, reflecting a decline as testing is increased. Gujarat and Maharashtra show a contrary trend and since they have a large number of cases, their contribution to the national ratio is high.
The constancy in the national test positivity masks significant churn within the states. This data is not released officially by the MOH&FW, but is compiled by crowdsourcing platforms from various official announcements. This data shows significant variation across districts and over time. As seen in Figure 2, there is considerable variation across states in test positivity. Over time, as testing as increased, the test positivity of most states has declined with the notable exception of Maharashtra and Gujarat. However, the crowd sourced data does not match with the data released by MoH&FW. Indeed, the test positivity is a little lower in the crowd sourced data than in the official release. There is also a lot of change in the testing being carried out across states as seen in Figure 3. It is important to put in reliable process to release data at state and district level if a sensible understanding of the crisis is to be attempted.
But, it is better news than a rising test positivity. Many of those who are being tested are likely to be infected because they were high risk contacts of people who tested positive for the virus. So, one would expect that the population prevalence to be much lower than that of the tested population. After the core group of contacts of people who have tested positive – both symptomatic and asymptomatic – there are other groups like SARI patients and symptomatic people with ILI in hotspots. But, a large number of patients are asymptomatic. Dr Gangakhedkar, ICMR’s head of epidemiology and communicable diseases said that “if we look at the number of tests done, so far 31% belong to the symptomatic category and the rest 69% would fall under asymptomatic.” Similarly, in Maharashtra, which releases these figures daily, over 70% of the cases are asymptomatic. Are all these people asymptomatic contacts? If not, how did they come to be tested? Till we know who is being tested, in comparison to the population, it is difficult to answer the question as to what this test positivity tells us about infection in the general population more precisely. The experience of pilgrims who returned from Nanded to Punjab, where about 30% tested positive, but were asymptomatic, indicates that there is much to know still about its transmission and virulence. But, beyond broad generalities, there is limited information on who is being tested.
ThePrint India carried this analysis by Partha Mukhopadhyay. Access the piece here.
The data on positive persons is released by the Ministry of Health and Family Welfare (MOH&FW) at 8AM every day.
The number of samples tested is released by the ICMR at 9AM
The figure will be updated on a daily basis. The associated data is available here.