Empirical evidence on the efficiency of backward contact tracing in COVID-19

Abstract

Standard contact tracing practice for COVID-19 is to identify persons exposed to an infected person during the contagious period, assumed to start two days before symptom onset or diagnosis. In the first large cohort study on backward contact tracing for COVID-19, we extended the contact tracing window by 5 days, aiming to identify the source of the infection and persons infected by the same source. The risk of infection amongst these additional contacts was similar to contacts exposed during the standard tracing window and significantly higher than symptomatic individuals in a control group, leading to 42% more cases identified as direct contacts of an index case. Compared to standard practice, backward traced contacts required fewer tests and shorter quarantine. However, they were identified later in their infectious cycle if infected. Our results support implementing backward contact tracing when rigorous suppression of viral transmission is warranted.

Fig. 1. Schematic representation of the different testing and tracing strategies and which parts of the chain of transmission they can uncover.

On the left, a transmission chain is shown where COVID-19 spreads from a parent case to an index case and their sibling case at a shared source event. The index, sibling and child cases all spread their infection further. Black arrows show transmission events, while green diagonals show the infectious period of each case. The index case develops symptoms on day 0 and gets tested as soon as possible. Double full vertical lines highlight when each case is detected as a contact, considering a combined testing and tracing delay of 1 day and testing of identified contacts as soon as possible. The standard and extended contact tracing windows are shown above the timeline. The testing and contact tracing strategies and which additional cases they identify are shown on the right. As especially the parent case demonstrates, a possible drawback of backward contact tracing is that some infected contacts are detected at a later stage of their infection, decreasing the effectiveness of testing and quarantine measures. It must be noted that the directionality of transmission and thus the position of an infected individual in the transmission tree is usually difficult to ascertain in practice.

Fig. 2. Schematic representation of contact tracing strategies.

Thin black and thick green arrows indicate the directions of transmission and contact tracing respectively. I index case, C child case, P parent case, S sibling case. White circle: undetected case. Grey circle: case detected through symptomatic screening. Green circle: case detected through contact tracing (a) When an index case is diagnosed, the child case at event D-1 is identified through standard forward tracing. A source investigation would fail at this stage, because there is no indication of further infections at the source event. b Source investigation does succeed when a second index case I₂ is diagnosed independently of the initial index case I₁. As the source event becomes clear due to identification of multiple infections, all attendants are traced. c An extended tracing window quickly identifies parent, sibling and child cases as direct contacts of the first index case.

Fig. 3. Inclusion flowchart main study period.

The number of included and excluded cases and contacts is shown for the main alpha-dominant cohort.

Fig. 4. Outcomes, positivity rates and risk ratios for contacts of index cases with corresponding p values.

The dotted line indicates the positivity rate in the control group. The error bars indicate 95% two-sided confidence intervals (Clopper–Pearson). * indicates a statistically significant difference in comparison to the control group (p < 0.05) as assessed using a two-sided Chi-squared test, not adjusted for multiple comparisons. Section a tests the main hypothesis by comparing the extended tracing window to the symptomatic control group. Subgroups by the numbers of days from onset or test of the index case to the last interaction with the index case are shown in section b, c for the extended and standard tracing windows respectively. Section d shows subgroups according to presence at suspected source events, and subgroups by relationship type are shown in e.

Fig. 5. Infection stages of COVID-19-positive contacts.

Symptom onset in infected contacts relative to sampling (a) or symptom onset (b) of the index case. Asympt asymptomatic. Case–contact pairs that were excluded from the mean calculation because either or both were asymptomatic, are shown on the right. Panel a shows the delay between detection of an index case and symptom onset of their infected contact. Red dots show the estimated remaining fraction of transmission potential of the infected contact at the time of sampling of the index case. Forward traced symptomatic contacts were detected on average 1.8 days earlier in their infectious cycle than their backward traced counterparts, assuming equal delays between index case diagnosis and tracing of the contact. This resulted in a 28% lower mean remaining transmission potential for backward traced contacts at the time of index case testing. Panel b shows the delay between symptom onset of an index case and their infected contact. Horizontal lines indicate the 25th–75th percentile ranges of expected timings for parent, sibling and child cases, based on a published normal distribution of the serial interval. The observed timings are compatible with a high proportion of sibling cases and few parent or child cases in the backward traced group.

Fig. 6. Timing of RT-qPCR testing in contacts as performed in the study period and the diagnostic accuracy of such tests by day since last exposure.

Error bars indicate 95% confidence intervals. a shows the number of contacts who underwent a first and second tests at our test centre after their exposure. This demonstrates how testing immediately after exposure (“test to trace”) was most often complemented with testing after a latent period (“test to release”). While the former mainly supports iterative tracing and in some cases a shortened isolation period, the latter allows shortening of quarantine for non-infected contacts. As the delay between last exposure and symptom onset or testing of the index case increased, the percentage of contacts requiring two tests decreased. b shows the mean timing of first and seconds tests at our centre for contacts, relative to their last exposure. The difference in timing of the first and second tests is reduced as the contact tracing window is extended further back in time. c shows the test results of infected contacts by day after last exposure, demonstrating how the sensitivity of RT-qPCR testing increased rapidly in the first days after exposure.

Fig. 7. Outcomes, positivity rates and risk ratios for contacts of index cases with corresponding p values.

The contact took place in selected periods, differing with regards to the dominant variants of concern, immunity, level of viral circulation, social contact restrictions and government testing/quarantine strategy. The error bars indicate 95% two-sided confidence intervals (Clopper–Pearson). * indicates a statistically significant difference in comparison to the control group (p < 0.05) as assessed using a two-sided Chi-squared test, not adjusted for multiple comparisons. a repeats the main study outcomes from Fig. 4a, while the results from subsequent periods are shown in b, c.

Fig. 8. Cost-benefit analysis of backward versus forward contact tracing in our setting, using a simple branching process model of iterative contact tracing.

Asympt asymptomatic. a, b show the marginal benefits and costs respectively, per day that the tracing window is extended backward. Both are given as a fraction of the benefits and costs of a standard forward tracing window. c show the total cost/benefit ratio of a contact tracing window extended to 7 days before onset or test, relative to a standard forward tracing window. Combinations of three cost and three benefit measures are shown. “Averted infectivity” denotes the number of detected cases, multiplied with their remaining fraction of transmission potential according to Fig. 5a. This measure of benefit accounts for the observation that backward traced cases were detected later in their infectious cycle. In this figure, “averted infectivity” can be considered equivalent to the number of averted infections, with the important caveat that it only includes child cases of a detected case, not any subsequent averted branches of the transmission tree.

Fig. 9. Schematic representation of two possible strategies for backward contact tracing, based on our results.

a shows a hybrid strategy, which avoids testing contacts in the extended tracing window who were not present at the suspected source event. b shows an extended tracing window strategy with systematic testing of all contacts.