<p>Machine learning models’ framework.</p>

Sensitive poliovirus surveillance is critical during the polio endgame and relies on systematic testing of stool samples from reported acute flaccid paralysis (AFP) cases. Collecting stool samples from healthy close contacts of AFP cases, preferably contacts aged under 5 years, is a potential method for enhancing surveillance. The introduction of type 2 novel oral polio vaccine (nOPV2) in Nigeria from 2021 under Emergency Use Listing was accompanied by an unprecedented intensification of contact sampling, but it is unclear how this affected surveillance sensitivity. Here, we analysed data from 71,002 AFP cases and 51,100 associated contacts sampled in Nigeria between January 2017 and November 2023 to quantify the added value of contact sampling and assess how it can be optimised. A total of 600 serotype-2 vaccine-derived poliovirus AFP cases were reported, of which 123 (20.5%) were identified through the presence of the virus in the stool of their contacts (‘false negative’ cases). Boosted Regression Trees ensemble models were used to identify the factors associated with the probability of a type-2 vaccine-derived poliovirus-positive contact and for the AFP case to be a false negative. AFP cases and contacts had a higher chance of concordant type-2 vaccine-derived poliovirus test results when stools were collected within 5 days of paralysis onset, whilst false negative AFP cases were most likely 5–15 days after paralysis onset. Our results suggest that contact sampling enhances the sensitivity of poliovirus surveillance, especially for AFP cases with inadequate stool samples or with stool samples collected 5–15 days after the onset of paralysis.