KK Women's and Children's Hospital (KKH) has adopted artificial intelligence to improve service levels and operational efficiency in portering services—with promising results to show.
KKH's Patient Transport Services (PTS) department runs a brisk operation. The team receives over 350 requests on any given day for portering services, from transporting patients from one spot to another to delivering medical consumables, equipment and lab specimens.
Like masterful traffic controllers or puzzle fixers, PTS controllers assign job requests to hospital porters with the aim of ensuring efficiency and equitable task distribution. All this, on top of the occasional need to recall and re-assign jobs when changes are made by the requestor.
"Controllers in the PTS department are constantly making decisions to put up the next available and suitable porter for a task," shares Mr Tey Yew Wei, Assistant Director, PTS, KKH. On exceptionally busy days, the manual effort to ensure patients receive optimal service and porters are efficiently deployed can become overwhelming.
Since July 2023 however, artificial intelligence (AI) has been a productivity game-changer at the PTS department.
Swift and smart decision-making
While AI isn't a new feature in clinical work and has been used in reading diagnostic images and picking up diseases, for example, incorporating it into hospital's support operations is still a novelty, says Ms Sally Oh, Director, Patient Support Services Division, KKH.
Even before the pandemic, Ms Oh, Mr Tey and their colleagues have been exploring ways to improve service and productivity. AI or machine learning became a point of interest—it would soon shape up to become the AI-powered Portering Task Assignment System.
The plan was to develop an AI adoption strategy to categorise and prioritise tasks, in order to optimise porter resources and improve patient service delivery. The system would consider factors such as task specifics, porter specifics, porter locality and workload, to sort and assign work tasks to the most suitable porter in an effective and efficient manner. Leveraging this predictive AI model took portering task assigning to a whole new level.
From the outset, its interface allows controllers to quickly see which porters are on duty, the number of tasks they have completed and outstanding tasks for each person. Aside, the most valued feature of the system is its ability to shortlist the most suitable porters for each job request.
It is not just physical proximity to the job location that makes a porter suitable for a particular request. "The system also considers a porter's pending tasks, break times, and attributes required for the job request such as whether he or she is able to push a trolley bed," explains Mr Tey.
Currently, the system churns out a maximum of five recommended porters for each request. The controller then makes the assignment. In time, the plan is to allow the system to assign tasks automatically.
Promising results to date
The team looked into several metrics to assess the effectiveness of the portering system, one of which was: would the system be able to assign tasks with the same judgement as a human controller?
So far, results have been positive, with the system matching up to a human controller's decision making at about 75%, measured through cross-validation of the AI system's recommendation against actual assignments by controllers. With more production data generated every day, the model learns and improves its accuracy over time.
"Portering assignments require human judgement so there will never be full objective accuracy," explains Ms Oh. Exceptionally busy days and changes to job requests also naturally require controllers to make manual adjustments.
On the ground, the PTS team has benefitted from the new system. Team Leader Mdm Maimoon Binte Ramli shares, "With the system doing the assignments, I can focus on other things such as managing requests via phone calls and putting together simple reports, which I wasn't able to do before."
Meanwhile, Patient Transport Assistant Mdm Keng Lee Hwa enjoys the smoother workflow. "Task assignment is more accurate, and I don't have to move long distances between jobs. Patient service has improved because jobs are completed more quickly."
Previously, an average of 45 seconds was needed for a Controller to assign a task manually, but with AI, the time taken has been significantly reduced to 10 seconds. This translates to a 5 to 10% improvement in a patient's wait time for a porter.
The invisible work behind the scenes
Putting the AI-powered system together was a massive undertaking. Mr Tey and his team have had to gather three years' worth of data to kick start the machine learning process.
That wasn't all. The team had to ensure that the data "makes sense" to the machine.
Preparing datasets for training and testing was essential, and so was making sure that the data contained features that were relevant to the problem they were solving.
"We went through a gruelling phase of scrubbing the data to remove wrongly assigned jobs, and anomalies that came up during COVID-19. We needed the machine to learn only data reflecting a 'business as usual' situation," explains Ms Oh. This amounted to hundreds of thousands of data points, which the team took months to pore over.
Even picking a vendor for the project was a drawn-out process as the team had to prioritise data security and ensure the vendor had a stellar track record.
It would have been easy to give up on the project, professes Ms Oh. "It has taken a lot of time to get here, on top of managing our daily work, but our team persevered and the management continued to support us."
What's next
Since the roll out of the AI system in July 2023, the team has continued to check on the accuracy of AI-generated assignments and fine-tune the algorithm to improve the AI model as more production data is being generated every day.
At the same time, a new project team has been convened under the Patient Support Services (PSS) Division to explore the use of predictive AI to transform other support functions, such as clinical coding in the Health Information Management Services Department.
The AI-powered Portering Task Assignment System will only get more accurate with time as it continues to learn new data every day—and it can potentially improve workflows and productivity beyond KKH's PTS office.
