The last author of the above study pointed out that there were five – and not four – studies from Singapore published in the May supplementary issue of Clinical Infectious Diseases. A clear oversight on my part.

The study – which was conducted at the National University Hospital (NUH) – evaluated the accuracy of an algorithm at identifying community-acquired pneumonia from the electronic medical records (EMR) of the hospital that has been made available via an anonymised patient portal (an excellent initiative that facilitates research while fulfilling the letter of the Personal Data Protection Act).

Study protocol – screen capture from the NUH paper published in CID.

Compared to the gold standard – which comprised the diagnostic acumen of a senior respiratory physician and an infectious diseases physician (it can be contentious why this is the gold standard), the investigators reported that the algorithm, which comprised of text mining using a natural language processor for:

• Admission diagnosis of community acquired pneumonia AND
• Discharge ICD-9-CM diagnostic codes for pneumonia AND
• Search term for pneumonia in the chest X-ray report

performed reasonably well, with a positive and negative predictive values of 75% and 87% respectively over the 200 cases evaluated.