Keynote Presentation by A/Prof Benjamin Ong, Director of Medical Services, Ministry of Health, at Symposium on Data-Driven Healthcare, 29 August 2017

Using Data Effectively to Deliver Patient-centric Care

Dr Lun Kwok-Chan, Chair, Symposium on Data Driven Healthcare

Distinguished speakers,

Ladies and gentlemen,

Good morning.

1.           I am delighted to be here at the Symposium on Data-driven Healthcare this morning.  A warm welcome to the many overseas guests amongst us today.

2.           I thought I would start with a simple example of data use.  Does the adage “An apple a day keeps the doctor away” actually hold true? If it is true, this would be a great preventive health intervention! One study published earlier this year pooled data from 95 studies worldwide and, sadly, debunked this old saying. It concluded that instead of one apple, we need to eat 8 to 10 servings of fruits and vegetables a day to significantly reduce the risk of cardiovascular disease, cancer and many other diseases.  I think all of you would agree with me that it would be hard to get everyone to adhere to this, but the results of this study should further encourage our Health Promotion Board’s programmes that encourage the increase of daily fruit and vegetable intake.

3.           This illustration was an interesting case of how data is being pooled across studies.  There are many ways data science is helping medical professionals be more evidence-based in the approach to healthcare delivery.  It can potentially help us be more accurate in our diagnosis and prescription of treatment, reduce costs of treatment, be more operationally efficient, and ultimately, help us be more effective in care intervention. There are all the aims of what we are trying to do.

Use of Big Data in Singapore Healthcare

4.           Over the last few years, we have implemented several data-driven projects in public healthcare.  These projects are part of our Smart Health efforts that support and enable the Ministry of Health’s three fundamental shifts: Beyond Hospital to Community, Beyond Healthcare to Health, and Beyond Quality to Value.  They serve to achieve two key objectives: 1) to make our population policies and programmes more effective, thereby improving population health; and 2) to make our healthcare system more productive, and address the common issue of optimising manpower in healthcare while augmenting the work of our healthcare professionals.  Let me share some of these projects with you today.

5.           Total knee replacement surgeries are more common today as our population ages.  These procedures can be associated with bleeding.  As such, there is a high prevalence of blood transfusion required in about a quarter of such surgeries.  You may be aware that blood transfusion itself also carries risks. To improve care outcomes, the National University Health System implemented the Value Driven Outcome (VDO) project across several use cases, including total knee replacements.  They collected relevant data, benchmarked and analysed various quality and cost indicators.  This enabled them to identify cost-effective clinical practices and reduce unnecessary variations.  We were happy to observe that it led to improvements in both cost and clinical outcomes. This is a great example of the shift Beyond Quality to Value.  In the case of total knee replacements, the percentage of patients requiring post-operative blood transfusion fell from 26% to 3%.  Moreover, there was a median savings of $955 per case achieved.  Most importantly, for patients, this means a shorter hospital stay and better care outcomes.  In fact, some NUH patients are now discharged from hospital just three days after surgery, down from the average in the past of about five days.  A/Prof James Yip, will be sharing more details on this initiative later.

6.           Even after patients are discharged from the hospital, some are readmitted.  This negatively impacts patients and families but also causes strain to our healthcare system.  Many factors lead to hospital readmission, including the lack of support outside the hospital, as well as lack of awareness of where to find appropriate support.  We started the Hospital to Home programme to support such post-discharge patients, to facilitate the shift into the Community.  I guess when you look at these slides, you realise that many hospital-based systems, and Singapore for many years was a hospital-based system, actually stop planning at the door of the hospital. Unless you start to link up community resources and national planning at scale, it becomes an issue for the patients’ care. The Integrated Health Information Systems (IHiS) developed a predictive model utilising over a thousand indicators to identify patients who are likely to have multiple hospital readmissions.  Using a combination of clinical theory and machine learning they combed through over 7 million records, and the model has an accuracy of seven in ten patients predicted.  This effort significantly reduced the manual screening that the nurses used to have to do, enabling them to have more time to directly support patients and provide appropriate intervention, such as arranging for interim home meals delivery for patients with mobility issues, changing of wound dressing, and many more.  To date, over 5,000 patients have benefitted from the programme implemented in our six public acute hospitals.

7.           One of the patients who was identified by the predictive model was 94 year-old Mdm Lee.  Prior to enrolment in the programme in May this year, she had been warded twice and was attended to at the emergency department once.  As part of the programme, this lady’s needs beyond the hospital were identified.  Caregiver training was then provided to her helper in the hospital and further reinforced and monitored at home to ensure the use of correct techniques in caring for her.  Referral to an occupational therapist was also made for home assessment and modification to increase safety in the home.  The therapist also taught her family members how they could utilise the home environment to encourage physical activity for strengthening of her limbs.  It has been three months since she started on the programme and she is doing well. I detailed what has been needed, I hope you also picked out that essentially the services are no longer hospital-based, but are community –based. Most of our resources are unfortunately still in the hospitals.

8.           In an effort to be more strategic in our War on Diabetes, we also embarked on developing a National Diabetes Database together with IHiS to consolidate and analyse the data on diabetic patients from multiple sources.  To be completed in phases, this project would help us to better monitor and evaluate the impact of our policies and programmes.  It would also enable us to derive new evidence-based insights for our clinicians to develop more effective and personalised intervention in both prevention and management of diabetes, this is an example of going Beyond Healthcare to Health.

9.           Meanwhile, demand at our outpatient clinics is also increasing.  One of the challenges faced by these clinics and probably around the world is patient no-shows.  On average, about 25% of patients who made medical appointments are no-shows. This leads to rather inefficient use of hospital resources, and deprives others of timely appointments.  It may seem rather mundane, but IHiS also developed a predictive model that identifies patients based on their risk scores who are likely no-shows.  It is being piloted at KK Women’s and Children’s Hospital (KKH) and the National Dental Centre of Singapore.  Once the model identifies such high risk patients, administrators would contact these patients to remind them about their appointments or check if they need to reschedule their appointments.  About one year’s worth of data, or about 3 million records, was used to develop the no-show model.  Another year’s data was used to validate the model, which has an accuracy of seventy-seven per hundred patients predicted.  While evaluation of the model is currently in progress, early observations indicated a decrease in no-show rate. Healthcare professionals at KKH shared that it helped them to optimise clinic resource utilisation, and released appointment slots would be given to other patients who needed them.  Moving forward, the model would be progressively implemented at SingHealth Polyclinic and I understand Sengkang General Hospital will also be adopting it.

10.       Several of the analytics initiatives that I just shared are made possible by the Business Research Analytics Insights Network (BRAIN).  The unique and secure platform was developed to enable analytical queries to be answered “on the fly”, pulling data from the disparate sources in real-time, anonymising, and harmonising them to derive meaningful insights.  The delivery of faster insights can certainly help us to be more responsive and timely in our design of appropriate programmes to help Singaporeans achieve better health.

Overseas use of data

11.       While we have implemented many different data-driven projects in public healthcare, there are still many other exciting possibilities that we can introduce to deliver better healthcare and help us be more effective and productive.  In Kaiser Permanente researchers developed a risk score that enabled clinicians to predict diabetic patients likely develop dementia in future, and resources were shifted to those in the high risk group for more targeted intervention. In case any of you are interested, I am a neurologist so this is quite close to my heart. There are interventions that you can personally take when it comes to reducing the risk of dementia, regardless of what your genetic pooling might be. One of the risk factors is diabetes, and better managing this is important. In another interesting example, computer scientists at Stanford University developed an algorithm to sift through hours of heart rhythm data to find life-threatening irregular heartbeats.  The algorithm could sort the data remotely and was observed to be very precise.  Another algorithm, developed also by the same computer scientists, used a database of nearly 130,000 skin disease images to reliably and accurately diagnose skin cancer.

12.       We are also doing something similar in Singapore.  However, instead of skin cancer, we are developing a digital retinopathy screening software that utilises artificial intelligence and deep learning technology to predict diabetes-related complications. This is currently undergoing testing.

Challenges to data science

13.       I thought I would talk a little bit about the challenges from my personal perspective. The use of big data has been on the rise all over the globe and its potential in healthcare is vast.  In fact, healthcare is quite a bit of a laggard with regard to this. If we compare ourselves to retailers, to banks, even insurance, they have used both secondary and primary source data to ensure the right messages are delivered. Their aim of course, is to maximise targeted sales and they have been doing this for a very long time.  So what are the obstacles to its adoption?  For one, not all data is recorded and structured in the same way.  I would posit that this is a necessary challenge to overcome as a patient’s health and health seeking behaviour is usually influenced by many different factors apart from clinical ones.  The BRAIN architecture was developed help overcome this problem.  Yes, we can have a more efficient analytics ecosystem, if we could harmonise the process of data collection right from the outset, this would be ideal.  Whenever possible, we should do this, but we need not wait.  Other industries already use what they have.

14.       Second, advanced analytics in healthcare requires multi-disciplinary expertise, bringing together people with technical, operational, and clinical knowledge as well as experience to interpret the large variety of data.  For our Hospital to Home programme, we adopted an end–to-end approach, and brought together teams from respiratory and critical care, family medicine and continuing care, bio-informatics, health economics, health delivery, health research, and data science and analytics, to ensure that the predictive model would be effective.

15.       Finally, and most importantly, we must have the drive and ability to transform, I was going to use the word disrupt, the way we organise and deliver care so that the insights and data we derive really facilitates the best outcomes for the patient and system.  What do I mean? In my journey with healthcare informatics all these years, the thing that has changed the least is the way we organise and deliver care. We have moved our paper records to EMRs, we have used the insights to drive change, but we still organise ourselves roughly the same way. Most other industries have changed, are we waiting for someone or something to disrupt us? I think that will come if we are not careful. This requires a clear strategy and resolve on how to utilise data and its deployment at scale.

16.       I think there are still many things yet to be discovered in our treasure trove of data.  We certainly need big data to help us change the way healthcare is delivered.  The greatest breakthroughs, insights, and progress, are usually made from collaboration and learning from each other. I think collaboration and who we work with will change, it will increase, it will not just be what we traditionally think of as healthcare professionals, healthcare administrators, data scientists and researchers. It will go way beyond that. What has been missing is the patient. Essentially we are deriving insights so that we can optimise outcomes to patients. In this whole journey, unless we can also utilise information that come from secondary sources that patients generate, I think it will be a lot more difficult for us to get to where we want to in terms of behavioural change.

17.       My team and I look forward to working with many more of you to deliver better healthcare, and I wish everyone a very fruitful conference.  Thank you.