Underestimating attacks: comparing two sources of publicly-available data about attacks on health care in 2017

Attacks on health care represent an area of growing international concern. Publicly available data are important in documenting attacks, and are often the only easily accessible data source. Data collection processes about attacks on health and their implications have received little attention, despite the fact that datasets and their collection processes may result in differing numbers. Comparing two separate datasets compiled using publicly-available data revealed minimal overlap. This article aims to explain the reasons for the lack of overlap, to better understand the gaps and their implications.

We compared the data collection processes for datasets comprised of publicly-reported attacks on health care from the World Health Organization (WHO) and Insecurity Insight's Security in Numbers Database (SiND). We compared each individual event to compile a comparable dataset and identify unique and matched events in order to determine the overlap between them. We report descriptive statistics for this comparison.

We identified a common dataset of 287 events from 2017, of which only 33 appeared in both datasets, resulting in a mere 12.9% (n = 254) overlap. Events affecting personnel and facilities appeared most often in both, and 22 of 31 countries lacked any overlap between datasets.

We conclude that the minimal overlap suggests significant underreporting of attacks on health care, and furthermore, that dataset definitions and parameters affect data collection. Source variation appears to best explain the discrepancies and closer comparison of the collection processes reveal weaknesses of both automated and manual data collection that rely on hidden curation processes. To generate more accurate datasets compiled from public sources requires systematic work to translate definitions into effective online search mechanisms to better capture the full range of events, and to increase the diversity of languages and local sources to better capture events across geographies.