Contact tracing with quarantine has been crucial to reduce COVID-19 transmission but has not interrupted it. We need more effective tracing strategies for when transmission reduction is (again) necessary.
Most jurisdictions have focused on forward tracing. Forward tracing tries to interrupt transmission from persons infected by an index case. Close contacts are quarantined and/or tested if they last saw the index case during their assumed infectious window: from two days before symptom onset or diagnosis until the interview. Backward contact tracing extends the window further back to also identify the infector of the index case and others infected by them. This approach is particularly promising in COVID-19 as only a small fraction of cases infects most individuals. Longer delays are however inherent in backward tracing. This may reduce efficiency in terms of the number of tests and quarantine days required per avoided transmission.
In a large cohort study, we extended the contact tracing window from the standard 2 days to 7 days before symptom onset or testing. We applied this simple change in protocol to 659 Belgian university students diagnosed with COVID-19 between February and May 2021.
These students had 1608 ‘forward traced contacts’ (last seen in the standard tracing window), 788 ‘backward traced contacts’ (last seen in the extended tracing window) and 1,461 ‘symptomatic controls’ (students undergoing a first test for symptoms in the same period). The risk of infection was over two times higher in backward traced contacts than in symptomatic controls. Extending the tracing window identified 49% more contacts and 42% more infections.
A more targeted version of backward tracing, known as ‘source investigation’, has most often been implemented. This strategy focuses on contacts present at events where the index case was likely infected. We identified such an event in 80.6% of cases. If it could be identified, the risk of infection was around 4 times higher for those who attended (absolute risk 27.5%, CI 21.6-34.2%) than those who didn’t (absolute risk 6.9%, CI 4.6-9.8%). If it couldn’t, the risk of infection for extended window contacts was 17.1% (CI 11.9-23.6%). This validates the efficiency of the source investigation approach. However, it also highlights the need to shift to an extended contact tracing approach when there is no clear source event.
We then investigated whether backward tracing identifies infected contacts later in their infection. We calculated the delay between the diagnosis of an index and symptom onset in their infected contacts. Infected contacts in the extended tracing window were identified on average 1.8 days later than those in the standard window. We then combined this information with a model describing when transmission takes place, relative to symptom onset. We found that the average remaining transmission potential at identification was 28% lower for backward traced contacts.
We also looked at the number of tests and quarantine days required to do backward contact tracing, compared to forward tracing. We assumed that contacts underwent two tests. A first one immediately after identification, so they could be traced iteratively if infected (“test to trace”). Then, seven days after their last exposure, so their quarantine could end if not infected (“test to release”). We implemented both during the study. Backward traced contacts were detected 4.0 days longer after last exposure than forward traced contacts on average. So, a single test at identification was more likely to serve both purposes. Using our data on when contacts took place, we estimated that 17% less tests were needed per backward traced contact. Also, the average quarantine was 3.0 days (57%) shorter for backward traced contacts.
Follow-up rates were much lower in more recent periods. Due to system overload and looser testing criteria, the focus shifted from contact follow-up to simple notification. Self-tests were also increasingly available and less often reported. When the Delta variant was dominant, backward traced contacts had a similar positivity rate as forward traced contacts and symptomatic controls. When Omicron was dominant, the positivity rate of backward traced contacts remained high (13.3%, CI 8.5-19.5%). However, it was now lower than the positivity rates of forward traced contacts and controls.
Only contacts who were students underwent backward iterative tracing, as others were not traced by the university tracing team. To estimate how this strategy had performed if applied to all contacts, we used a simple branching process model. This allowed us to estimate several cost-benefit measures over multiple iterations of tracing. Compared to forward tracing, backward iterative tracing identified 55% more cases and 61% more asymptomatic cases. It averted 38% more infections. In exchange, it required 78% more contacts to be traced, 67% more tests and 40% more quarantine days. The further the tracing window was extended, the lower both the costs and benefits of tracing were.
Our results show that the immediate cost of backward contact tracing can be proportional to the benefits obtained. While we cannot infer the impact on community transmission, modelling studies suggest it can be major.
In conclusion: backward tracing merits consideration whenever contact tracing is required for COVID-19.