Digging into data-driven healthcare

By Jill Arul

SMU Office of Research & Tech Transfer – From our nutrition choices to each visit to the clinic – every decision we make and the measurable consequences of such decisions can chart our health. With an abundance of data available, researchers can then work towards testing and improving healthcare processes to accomplish a variety of goals like early diagnosis, better treatment, improved patient experience and more.

At the Singapore Management University’s (SMU) Healthcare Analytics and Operations Workshop that was held on 19 December 2022, participants heard from researchers, practitioners and policymakers from Singapore and overseas as they presented their work on healthcare research and practice.

The workshop also included a panel that consisted of distinguished experts from organisations like KK Women’s and Children’s Hospital, Northeastern University, Purdue University, and John Hopkins University in the US as well as Singapore’s Ministry of Health.

The group shared their views on the Healthier SG movement and how investing in regular screening, counselling and vaccination can support preventative care and bolster the health and wellness of the nation. As a varied group of representatives, they discussed the opportunities for all stakeholders to get involved and work together to advance preventative care.

Mapping the ideal test threshold

SMU Assistant Professor Sarah Yini Gao’s research aims to reduce resource waste and improve patient outcomes with early diagnosis of colorectal cancer.

In the same vein, early testing must be encouraged and refined with robust research. To increase the chance of a positive outcome, diseases like cancer must be diagnosed early.

Assistant Professor of Operations Management at SMU’s Lee Kong Chian School of Business (LKCSB), Sarah Yini Gao, presented her work on optimising initial screening for colorectal cancer detection. Currently, colorectal cancer is the most common cancer in Singapore and has the highest rate of diagnosis. It is mostly detected at a later stage, resulting in a higher mortality rate.

Professor Gao explained that the current procedure for testing includes a preliminary Faecal Immunochemical Test (FIT) that identifies colorectal polyps and potential cancers by detecting traces of haemoglobin, a major component in blood, in a patient’s stool. 

However, the test is not completely accurate. Patients who test positive at the preliminary FIT are advised to get a colonoscopy – a far more invasive test with several possible side effects – to confirm the cancer. While this process is mostly effective, it falls short in the margins of false positives or negatives.

In Professor Gao’s research, she evaluates an ideal cut-off that would benefit patients without straining healthcare resources – reducing the threshold too much could lead to too many false positives that could either result in unnecessary colonoscopies or reduce trust in the tests.

Professor Gao’s work harnesses patient behaviour data to propose a single ideal cut-off that would maximise expected follow-ups, detect 969 more cancer incidences and prevent 37,820 unnecessary colonoscopies.

Decoding data for better decision making

SMU Associate Professor Daniel Zheng Zhichao and his team have developed a model that is capable of improving estimations when working with censored data.

While Professor Gao’s work surrounds patient decision-making, Associate Professor of Operations Management at LKCSB, Daniel Zheng Zhichao, works to identify how practitioners can make more informed decisions with the data available to them.

Specifically, Professor Zheng identifies different ways that the electronic health record (EHR) data can be distorted by various policies that physicians employ, especially the treatment effect-based policy. Treatment effect refers to a patient’s response after being given a specific treatment.

Professor Zheng explained that policies based on predicted treatment effects can lead to distorted observations. Often, only favorable outcomes under treatment can be observed because the treatment will not be administered if it is predicted to fail. As such, much of the data is censored – sometimes distorting the later estimation and probability calculations like patient health transition probabilities which are the backbone of many models and algorithms behind personalised medicine.

To address these blind spots, Professor Zheng and his team have developed a structural model and econometrics theories that are capable of recovering true patient health transition probabilities from censored observations.

The model was then applied to estimate patient health transitions after the removal of tubes that aid breathing in an intensive care unit. When removed at the right time, patients can reduce their length of stay in the ICU without compromising their long-term health.

With their proposed model, Professor Zheng and his team were able to arrive at more accurate estimates that could offer physicians more confidence when making prognosis-based decisions.

“By accounting and adjusting for the structural distortion in the EHR data, our method can provide a more accurate and consistent estimation of patient health transitions under different treatment options,” explained Professor Zheng. “Equipped with such information, the physicians can make better decisions on which course of action would have the largest benefit to the patient.”

Through robust research and collaboration between stakeholders, like the representatives present at SMU’s workshop, Singapore’s healthcare system can be continually improved for the benefit of patients and the country’s resources.

Back to Research@SMU February 2023 Issue