Sri Lanka’s Problematic Covid Data

Chandre Dharmawardana, whose preferred choice of title is“Is complacency marring  Sri Lanka’s Covid-19 effort?  Fatality figures and the number of “confirmed cases”

Professor Pranna Cooray, in a  Q&A  session (Island, 20-04-2020)  draws attention to  Sri Lanka’s Covid-19 Case Fatality ratio (CFR). This is the ratio of the number of deaths to the number of confirmed cases.  He pointed out, using the  April 13th data for  the CFR figures for neighbouring countries, viz.,  Sri Lanka’s  3.2 percent (total cases – 218), India 3.4 percent (10,541), Pakistan 1.7 percent (5,716), Bangladesh 4.9 percent (803), “that  it is unacceptably high” given Sri Lanka’s reputed public health system.

In the figure,  Sri lanka’s offical  data from up to 14-04-20 are analysed into two groups, viz., “confirmed public cases”, and total confirmed cases, where the latter includes cases inside quarantine centers, and hence NOT in the public pool. The data are modeled using a mathematical model similar to that for the competition between species in an ecosystem.  The change in trends after imposition of the curfew shows that such methods are effective. The predictions for the two sets of data  hold from the 14th up to about the 21st (i.e., for about a week). Then a new trend is seen.

Professor Cooray argues that “our inability to find the true infected load, or even to come near that, is reflected in our Case Fatality Rate (CFR, percentage of the number of deaths divided by the total number of positives)”. So he advocates  doing  more tests to determine the “true infected number”.  Other health experts have also called for more comprehensive testing.

Until results form such testing comes, we can turn the argument around, and ESTIMATE a likely number of  “confirmed cases” if we knew what the CFR for Covid-19 for a tropical country should be. In a report published in the Lancet only weeks before the interview with Prof. Cooray,  the famed Harvard epidemiologist  Marc Lipstich is seen debating  other expert  authors  on what the CFR should be.

Prof. Lipsitch writes “No expert thinks the 3·6% raw ratio of deaths to cases on March 1 is an accurate estimate of the CFR because it suffers from all these … biases. The authors make the situation worse … without correcting for ascertainment of mild cases inflating the estimates, bringing them further from what most experts believe are the true numbers, around the 1–2% range for symptomatic cases”.

I have also assumed that a proper value for the CFR in Sri Lanka might be about 1.8% in my models of Covid-19 (See Figure). The number of deaths so far reported in Sri Lanka is seven – too  small to be confidently used in statistical analysis. However, if we  use this fatality number, the “confirmed cases” should be 350-450 if full testing had been feasible. However, the reported confirmed cases stood at 218. So, the further discovery of 15 cases (all in quarantine centers) on 14th of April, then 5 new cases of which 4 were in quarantine centers on the 15th, and so on  to the discovery of  33 new cases on the 20th of April, 38 new cases on 23rd April etc., WITHOUT the death toll increasing  should not  be regarded as too worrisome.  But there are other troubling signs.

So, assuming as Prof. Cooray does, that all the infected cases have not been discovered,  one can expect  that the number of confirmed cases will indeed increase towards 350 to 450.  But the graphs suggests that  the upper bound of ~450 may not be valid! Is the fatality number 7 valid?


The early successes in the fight against Covid-19 has led to a complacency and a belief that putting in curfews and imposing physical distancing have solved the problem. However, the graph shows a definite new trend since 21st April due to hidden public cases. They should have been in quarantine.

The case of the  Piliyandala fish vendor in instructive.   An infected  arrival was presumably “quarantined” in the same house as the vendor? Given the housing, and the social habits of the populace, “agreeing to quarantining at home” is perhaps like the promises of politicians.  Instead, all suspected cases must go to monitored quarantine  and anything less is a waste of  the expense and effort implied in having curfews and lock-downs. Releasing after 14 days should be replaced by the principle of “releasing  only if tested negative”.

Although new arrivals  now face mandatory  quarantine,  people sent off to “self-quarantine” since early-March probably  lived with others and are now a  reservoir  of infectious people who should  have been tested.  Adequate arguments for “herd immunity” from such a reservoir do not still exist.  EVERY CASE of respiratory illness referred to medical officers should be tested for Covid-19.

According to the Institute Pasteur, some 17% of infected people show no symptoms. So, using our estimate of some 400 infected people, there should have been some 70 more asymptotic carriers  in the community by 14th April. The challenge is to prevent  its  increase without becoming complacent.

If the incubation period is 14 days, curfews and physical distancing  into early May should be effective in low-incidence areas. However, the Western Province (WP) is the hub of economic activity.  It contributes 42% to the GDP, and 60%  to the service sector. Opening the country with the WP closed is like expecting a one-legged, one-handed, partially traumatized person to take his usual leadership role of the country. So it is in regions  like the WP,  that strict quarantine and more testing are needed.

However, even if  Sri Lanka had the means of a Germany, comprehensive testing is impossible within the required time. Germany carried out some  350,000 tests PER  WEEK  and managed to examine only about a million people, in a country of 67 million.  That is, only 5 tests per 1000 people. South Korea  did not do much better.

We have not been able to find out how many tests have been carried in Sri Lanka since 10th March, although one suspects that the number is about 6000-8000 for a population of 22 million.  Kerala with 36 million people  has tested some 20,000, compared to Canada’s 460,000 with a comparable population.   However,  it is not the number of tests per 1000  that matters, but the STRATEGIC deployment of the available testing.

Testing a person once is not sufficient due to  false negatives and false positives in tests.  Even with the resources of a Germany, to test ALL the people in Sri Lanka’s western province may take four months, while to adequately test (i.e., 1/e of them, e = 2.718, base of Napier’s Logarithm) will take one and a half months. But we need the information NOW to prepare  the hospitals for a possible surge.  This is why estimates from small amounts of strategically collected data using models (however limited they be), together with strategic inputs from hospitals  may become relevant. However, epidemiological  models (or weather prediction models) fail beyond even a week.  Simpler indicators, e.g,  the constancy of the fatality figure for a few weeks, would be a better harbinger of the end of the crisis. Even then, strategic testing must continue.

The method of tracking social news, i.e.,  gossip and information, spot checks, raids, and all the methods employed for the detection of criminals and terrorists become equally  effective in tracing absconding sick people and their asymptomatic contacts who may be infectious.  So, optimal selection of samples of people from areas under suspicion coupled with an investigative  approach is the best  strategy.  I have discussed this already in my article in the Island, 6th  of April (


Filed under accountability, charitable outreach, coronavirus, cultural transmission, governance, historical interpretation, landscape wondrous, life stories, politIcal discourse, security, self-reflexivity, sri lankan society, the imaginary and the real, transport and communications, unusual people, welfare & philanthophy, world events & processes

2 responses to “Sri Lanka’s Problematic Covid Data

  1. Pingback: Debating the Progress of Covid-19: Vibrant Viewpoints | Thuppahi's Blog

  2. Rohan Pethiyagoda

    As usual, a revealing analysis by CD: well done! I just want to point out that there is one easily obtained proxy for fatalities in general: the gross week-by-week mortality rate. All deaths in SL have to be reported to the local grama niladhari (GN) and the registrar of births and deaths. No funeral otherwise. So, comparing the deaths in each week this year against the same week last year / previous years gives an impression of changes in the death rate at a quite fine scale. These data are current and easily collated. Of course, adjustments need to be made for changes in mortality resulting from the lock down: fewer road accidents, probably more suicides etc. But cause of death is recorded on the death certificate, so even that is possible.Unofficial data from a few GN divisions show a distinct rise in mortality this year. C19 seems to have been killing in SL well before it was picked up officially, in mid-March. But there is nothing to prevent government from releasing the whole dataset, which will provide a crude proxy of C19 morbidity at a fine demographic scale, until mass screening becomes possible.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.