Big data. Algorithms. Machine learning. The jargon rolls off Dr Hairil Rizal bin Abdullah’s lips as naturally as one breathes and eats. Dr Hairil, consultant at Singapore General Hospital’s Department of Anaesthesiology is on a mission to improve healthcare with the use of large-set data from about 200,000 surgical patients who have passed through the doors of SGH since 2013.

One of his projects is the CARES (Combined Assessment of Risk Encountered in Surgery) algorithm, which is a locally derived machine learning based surgical risk scoring system. CARES accurately predicts the probability of post-surgery complications and the need for intensive care. While such algorithms have been developed overseas earlier, these are based on primarily the Western population. In contrast, CARES generates numbers that are far more applicable to patients in this part of the world. 

The rise of data

The implementation of electronic medical records (EMR) has given clinicians live (and ever-growing) information about patients. With technological advancements in data analytics, healthcare professionals can now gain insights that are not apparent to the naked eye, Dr Hairil explained.

While administrative data that capture details like a patient’s length of stay and identification details had long been digitised, it wasn’t the case for clinical data. Before EMR, thick cases of clinical notes were handed to the medical records office and entered into the system by non-medically trained staff.

Today, EMR has changed the game. Over the years, comprehensive notes of hundreds of thousands of patients have accumulated – and so has the potential for mining new medical insights.

“Administrative data sources are limiting in that they may not accurately capture clinical conditions. Clinical data sources such as the EMR have been shown to be far superior because they are taken at bedside prospectively by clinicians and include accurate diagnosis of complications, among many other things,” said Dr Hairil.

Digitised information from a large set of patients translates to accurate historical data, and with clear view of precedence, doctors can accurately predict surgical complications, he explained.

But that’s not all. A good data set can lend itself to endless applications such as allowing healthcare professionals to predict with more certainty how long a surgery will take so operating theatres are optimally utilised, reducing patients’ waiting time at clinics and even minimising parking woes for visitors.

Data analysis also allows users to run simulations. “For instance, we can check if increasing the number of porters who transport patients to operating theatres, from five to 10, will help operations to start on time. We don’t have to hire them first to find out,” Dr Hairil said.

Start of the learning curve

While Dr Hairil now speaks authoritatively on the topic, it was an uphill task getting the hang of data analytics. “Most of us [clinicians] are not even trained in statistics, so this is a double jump,” he said, adding that it has taken him eight years to get to this point.

In 2011, Dr Hairil was assigned to drive the digitisation of the pre-operation assessment form.

“When doing this, I learnt the importance of having structured data. We designed the form with checklists to structure the capture of data so that it could be analysed,” he said.

His first study using this data, however, only come in 2014, when a peer asked how many surgical patients were anaemic. “I could pull numbers from the US but that may not be relevant to the local context,” Dr Hairil said. With the help of IT colleagues, he managed to extract the information from the data set of 100,000 patients at that time.

“We then asked, which groups were at higher risk of anaemia? And what happened to these anaemic patients? What were their outcomes?” said Dr Hairil, which then led to the start of CARES.

Getting your hands dirty

There’s no other way to get started on data analytics except to just get down to doing it, shared Dr Hairil. “There’re lots of free online tutorials. You can take 30 minutes, an hour each day to watch. There are also more learning opportunities and platforms for data analytics now within SingHealth.”

His advice is to start low and slow. “The first thing you want to learn is to clean a data set. This means making patient records structured, deciding what is important, find out the missing variables and how to map a disease code. Also, go with a small number of 5,000 or 10,000 cases with a focused clinical question,” he said.

Despite the initial hard work, all the extra hours that Dr Hairil put into data analysis have paid off. “I’m now more evidence-based and directed in what I do. I can assess risk accurately for patients, do risk counselling with actual numbers and conduct risk mitigation with confidence. Without good data, we will not be able to tailor our approach according to patients’ different needs,” he said.

Dr Hairil’s hope is that more clinicians will take the plunge into the field of data science. He said, “EMR is a gold mine of data. Now is the time to reap the rewards, so that we can improve patient outcome and experience and reduce healthcare costs.”

Brought to you by Strategy Retreat Group 7 – A Learning Healthcare System.

Data Analytics Training Courses in SingHealth

• SingHealth – Duke NUS Joint HSR and Data Science Training Programs
  Deep Learning for Image Processing
  Health Systems Modeling Using Systems Dynamics
  Advanced REDCap User Training
  Tableau Training
  De-identification Training
  Practical Data Science  Primer
• Citizen Data Science Training Programs
  Empowering SingHealth users to be equipped with advanced data literacy and data analytical skills

For more information, please contact Dr Sean Lam Shao Wei ([email protected]) from Health Services Research Centre (HSRC).