Forum: Have pilot studies before implementing AI in healthcare

I refer to the report “NUH doctors using AI to save lives of stroke patients” (May 18) and the letter by Dr Chua Jun Jin, “Time for wider adoption of AI in medical field” (May 22).

The application RapidAI helps to shorten only one component of part of the process in managing stroke. Its usefulness in improving clinical outcomes for stroke patients in Singapore remains to be proven.

The families of patients must first recognise the symptoms of stroke and send the patients with suspected stroke promptly to the hospital.

Emergency Department staff must recognise such stroke patients, and ask for an urgent CT scan or MRI of the brain. Once scanning is done, the images have to be interpreted promptly by the radiologist and the neurologist. RapidAI helps shorten this interpretation process, from about 20 minutes to 1 minute.

Once a diagnosis is made, the family has to be briefed on the procedure of endovascular treatment, its risks and benefits, before the procedure is arranged.

So RapidAI, as with many other healthcare AI applications, shortens only one component of a long and complex process. Sometimes, shortening the timing of one part does not change the overall outcome, as it may just create congestion at the next bottleneck.

In our local setting, the current bottleneck is at the Emergency Department, where waiting time to go to the ward, and the time taken to reach the interventional radiology suite, can be long.

The public should remember the acronym BEFAST. 

Family members should look out for sudden loss of Balance, Eye visual loss, an uneven smile on the Face, weakness in the Arm, abnormality in Speech/Communication, and be aware that Time is crucial and call for an ambulance once stroke is suspected.

As the National University Hospital is the only institution using RapidAI, the Ministry of Health should coordinate a clinical study to determine if this particular AI program helps save lives and improves rehabilitation outcome, by simply comparing outcomes among different hospitals. 

We should be cautious in implementing AI programs in healthcare by first testing them in pilot studies, and introducing them country-wide only if they are proven to be cost-effective.

Dr Desmond Wai

Join ST's Telegram channel and get the latest breaking news delivered to you.