How big data can detect outbreaks faster

There is an old adage in medicine that “if you don’t look for a diagnosis, you are unlikely to find it”. But every human being has cognitive blind spots. Even experienced doctors can be caught off-guard by uncommon illnesses presenting in atypical ways. In the recent hepatitis C virus outbreak at the Singapore General Hospital, experts in the Independent Review Committee (IRC) noted a gap in the current system, where an unusual hospital-acquired infection like hepatitis C might not be spotted quickly.

To effectively manage an outbreak requires early detection — the earlier, the better. An experienced team can keep a lookout for unusual events. But what about unknown unknowns, or those which are so rare that there is little institutional memory even for a team of medical veterans?

Technology is not a panacea, but it can be a useful support tool for risk horizon scanning, to help decision makers sense emerging problems earlier. As Singapore embarks on the Smart Nation initiative, we should also ride on using Big Data to help us beef up the detection and response to any disease outbreak.

Every healthcare system contains data that can be mined to be analysed in a big way. One key area is in symptoms and diagnoses. Patients feel symptoms when they first fall ill, while the diagnosis is made once the medical team understands the cause of the illness.

But even in First World hospitals, data can be spread across different departments, or a mixture of paper and electronic records. Even electronic records may be split between different software platforms. This fragmentation is a bottleneck for analytics.

For data analytics to be meaningful, it is critical to maintain timely electronic data on diagnoses made by doctors, and symptoms reported by patients.

In some hospitals, a diagnosis is “coded” digitally when the patient is discharged, or when the bill is being prepared — which may be many days after the patient has become ill. Ideally, this should be updated daily, or in real-time.

Data collection works best when it is not an uphill task for front-line staff and end users, so the implementation must avoid creating too much additional paperwork and red tape.

A second area we can find rich, useful data is through primary care providers — polyclinics and general practitioners (GPs) who are at the front line for picking up symptoms in the community, whether it is “flu”, a fever with a rash, or some other pattern of illness. Their data and experience can play an important role in detecting disease outbreaks at an early stage, before patients feel ill enough to visit a hospital.

But many GP clinics are not yet fully computerised. For those that have gone digital, their computer systems may not be able to share information readily and in-depth with public hospitals or national databases.

There is an open niche in the market for a software programme that is easy, intuitive and empowering enough to use such that GPs nationwide would readily adopt it, while still being compatible with public-sector clinical databases.

Whether in our public hospitals or in the private sector, it will be important to make future databases and software platforms inter-operable. This allows data to be more easily shared and minimises IT bottlenecks.

Such a database should also be linked to hospital laboratories that generate enormous amounts of raw data every day. Blood test results, for instance, can show up in a laboratory computer some time before a human eye can read or process them.

SEEING BETTER,

THROUGH TECHNOLOGY

A common concern associated with the use of big data in healthcare is privacy. This issue can be addressed by encryption and anonymisation of data.

As a hypothetical example, a computer need not know the exact identity of a person — just that this person recently travelled through an infection “hot zone”, and has now shown up as part of a patient cluster with identical symptoms. To balance public health and personal privacy, the computer would only provide the identity of individual cases upon approval of authorities under stringent guidelines.

Big Data can also help with known unknowns. Experts may not have had first-hand experience of rare disease outbreaks, but computer analytics can be told what to look for, based on case studies from other countries.

The biggest benefit may come in searching out the unknown unknowns. Here, Big Data could be coupled with “smart software”, using machine learning to help guide astute human observers.

For example, the software could raise a flag when an unusual pattern of data, which has a probability of one in a million or rarer as compared with historical data and trends, is recorded. Such a pattern might have happened by coincidence alone, but a human operator could still take a look and decide if the one-in-a-million occurrence is spurious, or a leading indicator of something to look into further.

Thoughtful, simple and elegant design will also be key — whether it is electronic medical records, disease notification processes, or software to enhance human decision-making.

To achieve these design goals, a combination of established experience and start-up (or up-start) thinking will be helpful. As a thought experiment, it would be useful to imagine how an Apple or Google might approach the problem.

In a world with ever-growing complexity and a healthcare system with ever more moving parts, technology paired with renewed vigilance can give us more edge to see over the horizon, intercept problems earlier and help patients in a more timely way. In caring for our patients, we must seek to empower our healthcare workers to do their best, and even better.

ABOUT THE AUTHOR:

Dr Tan Wu Meng is a medical doctor and Member of Parliament for Jurong GRC. This commentary is written in his personal capacity.