A Significant Portion of COVID-19 Transmission Is Presymptomatic

Epistemic sta­tus: Not quite set­tled sci­ence, but preprints seem to agree.

Strong ev­i­dence points to presymp­tomatic sources as a ma­jor source of COVID-19 in­fec­tions, pos­si­bly the ma­jor­ity. The ex­act pro­por­tion is en­vi­ron­ment-de­pen­dent; aware­ness and pub­lic health mea­sures re­duce symp­tomatic trans­mis­sion more than they re­duce presymp­tomatic trans­mis­sion.

The main rea­sons for think­ing presymp­tomatic trans­mis­sion is sig­nifi­cant are di­rect mea­sure­ments of the se­rial in­ter­val and in­cu­ba­tion pe­riod, and the out­side view of what level of pub­lic health mea­sures have and haven’t suc­ceeded at con­tain­ment.

Be­fore delv­ing into pa­pers, a quick aside. If COVID-19 were only trans­mis­si­ble when peo­ple were cough­ing or fev­er­ish, con­tain­ing it would be pretty easy; just tell peo­ple to stay home if they have those symp­toms. Some peo­ple might try to go out any­ways, so you might also set up check­points where peo­ple have their tem­per­a­ture taken and have some­one listen to whether they’re cough­ing, but that would pretty much be suffi­cient. Em­piri­cally, how­ever, COVID-19 is suc­cess­fully spread­ing in coun­tries which have taken these mea­sures and other more ex­treme mea­sures, which is what we would ex­pect given presymp­tomatic trans­mis­sion, but not what we would ex­pect with­out it.

(Note: You might think this means that symp­tomatic peo­ple aren’t con­ta­gious, but ac­tu­ally it just means that peo­ple who show symp­toms are do­ing a good job of iso­lat­ing them­selves. Peo­ple with COVID-19 symp­toms are definitely con­ta­gious and need to iso­late them­selves and no­tify peo­ple they might have spread it to.)

(Note: Presymp­tomatic trans­mis­sion is a sep­a­rate is­sue from asymp­tomatic car­ri­ers. Presymp­tomatic trans­mis­sion is when some­one is con­ta­gious when they aren’t symp­tomatic yet. An asymp­tomatic car­rier is some­one who is con­ta­gious but who never de­vel­ops symp­toms. Asymp­tomatic car­ri­ers seem to be rare, though not com­pletely nonex­is­tent.)

Se­rial In­ter­val and In­cu­ba­tion Period

The se­rial in­ter­val is the av­er­age length of time be­tween trans­mis­sions in a trans­mis­sion chain; that is, given pairs of peo­ple A and B where A was in­fected on a known date and then in­fected B on a known date, the se­rial in­ter­val is the av­er­age amount of time be­tween those dates. The in­cu­ba­tion pe­riod is the amount of time be­tween when some­one is in­fected, and when they dis­play symp­toms.

If the se­rial in­ter­val is shorter than the in­cu­ba­tion pe­riod, this im­plies that a large frac­tion of trans­mis­sion must be presymp­tomatic. So, with that in mind, I went look­ing for stud­ies which mea­sure COVID-19′s in­cu­ba­tion pe­riod and se­rial in­ter­val. Th­ese are in two ta­bles be­low.

One of these stud­ies, Tapiwa Ganyani et al, es­ti­mated the pro­por­tion of trans­mis­sion which was pre-symp­tomatic: 48% (95% CI 32-67%) for Sin­ga­pore and 62% (95%CI 50-76%) for Ti­an­jin. No other stud­ies es­ti­mated this quan­ti­ta­tively, but most stated that their re­sults pro­vided qual­i­ta­tive ev­i­dence that presymp­tomatic trans­mis­sion is oc­cur­ring.

Es­ti­mates of the In­cu­ba­tion Period

StudyIn­cu­ba­tion Pe­riod (days)Sam­ple SizeData source
Stephen A. Lauer et al5.1 (Me­dian)181Trav­el­lers “in ar­eas with
no known com­mu­nity trans­mis­sion”
Wei-jie Guan et al4 (Me­dian) 1099China out­side Hubei
Qun Li et al5.2 (Mean)425Wuhan
Jantien A Backer et al6.4 (Mean)88Trav­el­lers from Wuhan
Sijia Tian6.7 (Me­dian)262Beijing
Lau­ren C. Tin­dale et al7.1, 9 (Mean)228Sin­ga­pore and Ti­an­jin
Kaike Ping et al8 (Mean)162Ghuizho, Chi­na

Es­ti­mates of Se­rial Interval

S­tudySe­rial in­ter­val (d)Sam­ple SizeData source
Shi Zhao et al4.4 Mean
3.0 SD
21 chains, 12 pairsHong Kong pub­lic data
Nishiura H et al.4.0 or 4.6 Me­dian28 or 18 pairsPub­lished case re­port­s
Chong You et al4.41 Mean
3.17 SD
71 chain­sChina, out­side Hubei
Qun Li et al7.5 Mean5 cluster­sHubei case cluster­s
Zhan­wei Du et al3.96 Mean
4.75 SD
468 pairsChina, out­side Hubei
Tapiwa Ganyani et al5.21 Mean,
3.95 SD
226 cas­esSin­ga­pore and Ti­an­jin cluster­s
Lau­ren C. Tin­dale et al4.56 Mean,
4.22 SD
228 cas­esSin­ga­pore and Ti­an­jin cluster­s
Shi Zhao et al5.2 Mean48 pairsHong Kong and Shen­zhen
Kaike Ping et al6.37 Mean57 cas­esGuizhou, China

Tapiwa Ganyani et al and Lau­ren C Tin­dale et al ap­pear to have used over­lap­ping pub­lic data sources. The sam­ple size column for se­rial in­ter­val stud­ies is un­usu­ally painful, as sam­ple-size columns go, be­cause many of the stud­ies needed to ac­count for un­cer­tainty in who in­fected who; as such, sam­ple sizes are re­ported vary­ingly in units of (in or­der from most to least re­li­able per sam­ple) pairs, chains, clusters, and cases.

The study with the longest es­ti­mated se­rial in­ter­val, Qun Li et al, looks at a small num­ber of clusters and guesses which cases in­fected which other cases. While it es­ti­mates a mean se­rial in­ter­val of 7.5, Its data is also com­pat­i­ble with an in­ter­pre­ta­tion in which the mean se­rial in­ter­val is shorter and some of the trans­mis­sions are in­di­rect. This change in in­ter­pre­ta­tion would bring it in line with other stud­ies in this set, which es­ti­mate shorter in­ter­vals.

One of these stud­ies, Zhanwe Du et al, es­ti­mated the se­rial in­ter­val us­ing when peo­ple be­came symp­tomatic (rather than when they were ex­posed), and found that in 13% of cases, the in­fectee showed symp­toms be­fore the in­fec­tor did. This would im­ply that ei­ther in those cases the in­fec­tor trans­mit­ted presymp­tomat­i­cally, the in­fec­tor had a rel­a­tively long in­cu­ba­tion pe­riod, and the in­fectee had a rel­a­tively short in­cu­ba­tion pe­riod; or that this data set had ma­jor is­sues iden­ti­fy­ing who af­fected who. The dis­tri­bu­tion of SIs fits a nice Gaus­sian, which is some ev­i­dence that it’s the former.

Anec­do­tal Re­ports and Case Studies

To un­der­stand what presymp­tomatic trans­mis­sion of COVID-19 would look like, I went look­ing for anec­dotes and case stud­ies of known COVID-19 trans­mis­sion events. You can’t use these to in­fer much about rates, but they’re helpful for in­ter­nal­iz­ing what presymp­tomatic trans­mis­sion would look like.

“I be­lieve I caught it when at­tend­ing a small house party at which no one was cough­ing, sneez­ing or oth­er­wise dis­play­ing any symp­toms of ill­ness. It ap­pears that 40% of the at­ten­dees of this party ended up sick.”


(via Google Trans­late) “On Jan­uary 24, Li and his grand­father, grandma, and father went to aunt’s house for din­ner, a to­tal of 9 peo­ple. On Jan­uary 28, Li de­vel­oped fever. … all 9 peo­ple par­ti­ci­pat­ing in the din­ner were con­firmed as con­firmed cases.”


Prac­ti­cal Implications

The main prac­ti­cal im­pli­ca­tion is that con­tact trac­ing is re­ally im­por­tant.

Con­tact trac­ing is where, when you find some­one with COVID-19, you iden­tify ev­ery­one they might have spread it to and warn them that they’ve been ex­posed. Peo­ple who’ve been ex­posed are ex­pected to quaran­tine them­selves for 14 days, which is long enough that if they are in fact in­fected, there’s only a 1% chance they are in­fected but not yet symp­tomatic . Back in Jan­uary, this served two pur­poses: it en­sured that if they had a cough, they wouldn’t brush it off as some­thing minor and keep go­ing to work, and it en­sured that if they didn’t have a cough, they wouldn’t trans­mit it while presymp­tomatic. The first is­sue is now less of a con­cern; ev­ery­one knows that if some­one has a cough, they aren’t sup­posed to go to work, even if it’s definitely rhinovirus. The sec­ond is­sue is ex­actly as much a con­cern as it was be­fore.