Random testing versus clinical testing

The worldwide figures available on the COVID-19 pandemic, shared by official health systems, are apparently from clinical sources. The database https://www.worldometers.info/coronavirus/, as of 4/30/20, 5 p.m., showed 415 COVID-19 cases, and 29.3 deaths, per million persons (1Mpop) in the world. The Philippines, in particular, had 77 cases and 5 deaths, while Singapore had 2,764 cases/2 deaths, South Korea had 210 cases/5 deaths, Malaysia had 185 cases/3 deaths, Japan had 110 cases/3 deaths, Thailand had 42 cases/0.8 deaths, and Indonesia had 37 cases/3 deaths, all per 1Mpop.The cases and deaths known to the official health systems presumably include only those they tested. The same source shows the Philippines far short in testing, with only 903 tests, versus 24,600 in Singapore, 12,091 in South Korea, 4,953 in Malaysia, 2,551 in Thailand, 1,299 in Japan, and 318 in Indonesia, all per 1Mpop. Our lack of testing might explain why we officially have much fewer cases per 1Mpop than Singapore, South Korea, Malaysia, and Japan.

To evaluate the efficacy of a quarantine policy, tracking is more needed on the COVID-19 infection rate than on the clinically recorded cases and deaths. The direct scientific way to see the COVID-19 infection rate, and its movement over time, is to survey a random sample of the population, carefully observe the infection rate of the sample by testing, and then repeat the survey periodically.

The current tests of people with COVID-19 symptoms, people who have been in contact with COVID-19 patients, people feeling sick but with non-COVID-19 symptoms, new arrivals from abroad, VIPs, and the like are not tests of random Filipinos. Those are clinical tests for the purpose of taking some action, or non-action, regarding the specific individuals tested. The COVID-19 infection rate of the population cannot be statistically inferred from these tests.

Under random testing, the infection rate of the population—whether national, regional, or city—from which the sample came is statistically estimated by the infection rate of the sample, plus/minus the margin for sampling error. The sampling error depends on the absolute (not the relative) size of the sample: approximately +/- 3% for a random sample of 1,000, +/- 2% for a random sample of 2,500, +/- 1.4% for a random sample of 5,000, and +/- 1.0% for a random sample of 10,000. A small random sample is usable when carefully interpreted; a non-random sample, however large, is useful only for the specific persons tested.

A bigger sample is always more accurate, but more costly. A moderate sample may be best, since good policy guidance requires not only a single survey, but a series of surveys, say monthly, to track the infection rate until the pandemic is over. Every survey should take a fresh sample, and not repeat a previous one.

Survey research involves sending fieldworkers to a random sample of individuals or respondents in the field, i.e. in homes. Drawing a random national sample typically involves stepwise selection of areas, regions, provinces, cities/municipalities, barangays, dwellings, and one (or more, as needed) persons in the dwelling, using probability proportional to size at each step. (This bypasses residents in institutions, like military/worker barracks, prisons, asylums, monasteries, convents, dormitories, and seagoing vessels; also forest and sea folk, gypsies, rebel groups, and the homeless. Special populations need special studies.)

In surveying about COVID-19, the field staff should explain the purpose, assure respondents (one random family member? the entire family?) that the data are confidential, and ask permission to continue. Then they can take test swabs (for laboratory evaluation), and ask background questions relevant for the analysis—all done on the spot, as respondents cannot be summoned to a health center for testing.

Another quality factor would be the type of COVID-19 test used in the survey. There seems to be a “gold standard” the Department of Health likes, that gives the fewest false-positives or -negatives, which is somewhat expensive; maybe other types will do. A field team needs someone trained to administer the test properly, which adds to cost. The principle should be to use the best test one can afford, counting all costs; it will give better guidance than no test at all.

Testing everyone is impossible; but random testing works already (see Louis Kaplow, “If we can’t test everyone for coronavirus, this is the next best thing” and Katrin Bennhold, “With broad, random tests for antibodies, Germany seeks path out of lockdown,” New York Times, 4/20/20).

Contact [email protected].