Evolution and emergence of novel human infections

Arinaminpathy, N.; McLean, A. R.

In practice, it is highly unlikely that the underlying reproductive numbers for an evolving pathogen could be known with any accuracy. Here we focus instead on monitoring ongoing outbreaks for early signs of emergence, the primary concern being rapid notification. We suggest a simple non-parametric method for detection of emergence that does not rely on knowledge of the reproductive fitness parameters. An outbreak in progress is given a 'rarity score', with respect to past outcomes, defined as 2log 10 (proportion of past outcomes that were more extreme). We compare two approaches: for the 'single rarity' approach, a record is kept of the total outbreak size from every introduction. For an outbreak in progress, a rarity score can then be calculated for the cumulative number of cases. When this exceeds a given threshold, the alarm is triggered. The 'double rarity' approach likewise monitors outbreak sizes, and additionally monitors the daily incidence. A record is kept of the greatest incidence reached following every introduction, and thus a rarity score is calculated, in real time, for an outbreak in progress. An alert is then triggered where at least one rarity score exceeds a given threshold. Specificity is measured by the number of introductions causing a false alarm (introductions causing an alarm and subsequently going extinct). Sensitivity is measured, in the event of a true emergence, by the number of cases before the first alarm. Figure 6 shows measures of (b) Graphs of algorithm performance, calculated over 250 simulated emergences, with the alarm silenced for the first 400 introductions. Black bars, single; grey bars, double. (i, ii) Specificity is measured by number of false alarms before an emergence, (iii, iv) while sensitivity is measured by the number of cases before an alarm occurs, in the event of a genuine emergence. Left-and right-hand panels refer to punctuated and gradual scenarios, respectively.