BMC Digital Health (Aug 2025)

Evaluating the performance of an automated respiratory rate counter in detecting fast breathing pneumonia in children using a reference video expert panel

  • Ahad Mahmud Khan,
  • Md Shafiqul Islam,
  • Nabidul Haque Chowdhury,
  • Salahuddin Ahmed,
  • Rezwana Tabassum,
  • Sadia Afrin,
  • Zannatul Ferdush Amin,
  • Kazi Sazzadul Haque,
  • Afroza Yeasmin Rumi,
  • Jawata Rahman,
  • Rakib Bhuiyan,
  • Rizouan Ur Rashid,
  • Kamrun Nahar,
  • Robynne Simpson,
  • Ayaz Ahmed,
  • Md Mozibur Rahman,
  • Ting Shi,
  • Abdullah H. Baqui,
  • Steve Cunningham,
  • Eric D. McCollum,
  • Harry Campbell

DOI
https://doi.org/10.1186/s44247-025-00175-3
Journal volume & issue
Vol. 3, no. 1
pp. 1 – 10

Abstract

Read online

Abstract Background According to the World Health Organization’s Integrated Management of Childhood Illness (IMCI) guidelines, childhood pneumonia diagnosis relies on counting respiratory rate (RR). Counting RR by health workers is frequently inaccurate, leading to misdiagnosis and poor outcomes. Automated RR counters could potentially overcome these limitations. To address this gap, we introduced an automated RR counter and developed a reference video expert panel (VEP) to evaluate its performance. Methods We conducted a cross-sectional study involving children aged 0–59 months with suspected pneumonia in Bangladesh. The RR of children was counted using an automated counter (ChARM) and chest movements were simultaneously videotaped. These videos were interpreted by the VEP, trained to a standard procedure. We assessed ChARM’s accuracy in comparison to the RR generated by the VEP and summarised the time taken to count RR by ChARM. Results Among 339 enrolled children, ChARM successfully counted the RR of 294 children (86.7%). The VEP reached a consensus (i.e., RR count difference within two breaths per minute (bpm) between two VEP members) in 257 of the 294 children (87.4%). ChARM and the VEP agreed on RR counts within two bpm in 68.1% of children (n = 175/257), with a mean difference of 1.7 bpm and limits of agreement ranging from -6.7 to 10.2 bpm. ChARM classified age-adjusted fast and normal breathing with a sensitivity of 95.8% (95% CI: 88.1–99.1) and a specificity of 93.5% (95% CI: 89.0–96.6), demonstrating high agreement (kappa = 0.86). The median time to count the RR by ChARM was 66 s (interquartile range: 61–73 s). Conclusions ChARM counted RR accurately against a VEP reference, indicating a potential role in supporting health workers to diagnose pneumonia. However, it was unsuccessful for 1 in 8 cases, typically those more clinically challenging, suggesting a possible systematic bias. Further research is needed to address these issues and confirm ChARM’s reliability for broader use in real-world settings. Trial registration Current Controlled Trials ISRCTN14120515, registered retrospectively on 19 September 2024.

Keywords