Model-free estimation of COVID-19 transmission dynamics from a complete outbreak

Abstract

New Zealand had 1499 cases of COVID-19 before eliminating transmission of the virus. Extensive contract tracing during the outbreak has resulted in a dataset of epidemiologically linked cases. This data contains useful information about the transmission dynamics of the virus, its dependence on factors such as age, and its response to different control measures. We use Monte-Carlo network construction techniques to provide an estimate of the number of secondary cases for every individual infected during the outbreak. We then apply standard statistical techniques to quantify differences between groups of individuals. Children under 10 years old are significantly under-represented in the case data. Children infected fewer people on average and had a lower probability of transmitting the disease in comparison to adults and the elderly. Imported cases infected fewer people on average and also had a lower probability of transmitting than domestically acquired cases. Superspreading is a significant contributor to the epidemic dynamics, with 20% of cases among adults responsible for 65-85% of transmission. Subclinical cases infected fewer individuals than clinical cases. After controlling for outliers serial intervals were approximated with a normal distribution (μ = 4.4 days, σ = 4.7 days). Border controls and strong social distancing measures, particularly when targeted at superspreading, play a significant role in reducing the spread of COVID-19.

Fig 1. Daily new cases of COVID-19 in New Zealand from 26 February to 31 May 2020, split into cases with a recent history of international travel (blue) and those without (orange).

Fig 2. Reconstruction of New Zealand’s nine largest clusters.

Each panel shows one instance of the Monte Carlo method for reconstructing the epidemic transmission tree with cases plotted by date of symptom onset. Red = imported case; blue = domestic case; dark colours = clinical; light colours = subclinical; squares = age under 10 years; circles = age 10–65 years; triangles = age over 65 years. The title of each panel describes the setting and location (Akl = Auckland; Chch = Christchurch; Qtn = Queenstown) associated with each cluster; n is the total number of cases in the reconstructed tree including actual cases and simulated (missing) subclinical cases.

Fig 3. Across all categories of case children are under-represented.

Frequency of cases for the different types of case aged: children under 10 years; adults 10–65 years; elderly over 65 years. Dashed lines show New Zealand population. Error bars show the binomial 95% confidence interval.

Fig 4. Serial interval distribution with fitted distributions.

Red curve shows the normal distribution fitted to all data (~N(μ = 5.0, σ = 5.7)); yellow curve shows the normal distribution fitted to data excluding serial intervals longer than 20 days (~N(μ = 4.4, σ = 4.7)).

Fig 5. Increases in alert level and improved border measures resulted in decreases in secondary transmissions.

Expected value of R_eff each day for clinical domestic cases (A) and imported cases (B). Solid line show the median from 500 instances of the reconstructed transmission tree, dashed lines are the 95% range. Imported cases had a lower effective reproduction number than domestic cases. The date of secondary case exposure was assigned to be the date of symptom onset in the index case. Note that 10–13 April was the long Easter weekend and 27 April was a public holiday in New Zealand. The timing of the end of alert level four was announced on 22 April. Shading in (A) shows period of alert level 4 (26 March to 27 April). Shading in (B) shows progressively stricter border restrictions introduced 16 March, 20 March and 10 April.

Fig 6. Children under 10 tended to infect fewer people and were less likely to be superspreaders.

Distribution of individual case reproduction numbers for domestic and imported cases in each age group: under 10 years (blue), 10 to 65 years (red), over 65 years (yellow).

Fig 7. Children under 10 had a lower secondary attack rate across all case groups.

Secondary attack rate for each group by age.