risks and opportunities Peter Klimek Singapore, July 18 2017 Our - - PowerPoint PPT Presentation

Digital transformation in healthcare: risks and opportunities

Peter Klimek

Singapore, July 18 2017

Our health care systems are unsustainable.

Singapore Main challenge: noncommunicable diseases

Noncommunicable diseases

Leading cause of death (83% in Singapore, 68% world-wide, from those more than 40% premature) 1 out of 4 Singaporeans aged above 65y developed a chronic disease in 2015 (Singapore Life Panel) How to respond to this challenge

Will new cures solve this problem?

Remarkable advances for some specific diseases (certain types of cancers) ...

Will new cures solve this problem?

... but not so much for many chronic disorders where lifestyle & environment are important risk factors

Rate of productivity of developing new drugs is going down (USD 5billion per drug) Rate of failure in clinical trials goes up (attrition rate of 95%)

Will healthy ageing initiatives solve this problem?

Singapore WHO Europe USA

Healthy Ageing?

Singapore WHO Europe USA

There is no healthy ageing

in this sense. Instead, we need a new vision for medicine

Complexity Perspective

• Chronic diseases have many different risk

factors (genetic, social, environmental, ... networks)

• Example Type 2 Diabetes: ~30 genes known

that explain only about 10% of disease

• ccurrences
• Each risk factor might be involved in several

disorders  patients have multiple disorders

• Need to understand networks involved in

disease and how they influence each other to repair them and prevent disorders Metabolic pathways Gene-regulatory networks Protein-protein interactions Social networks

managing networks = preventing & curing diseases

Vision

What is medicine?

Phenotypic / comorbidity data

• If you have disease A, how likely are you to have another disease B?
• Disease-disease networks of 141,000 classified diseases
• 1,000 known drugs
• Disease-drugs networks

Health-care data

• Re-use of population-level health care (claims) data (safeguarded by HVB)
• Anonymized research dataset containing each medical treatment and its cost in Austria
• ver two years
• 100,000,000 doctor visits and 2,000,000 hospitalizations per year and health-care

provider networks with 8,000,000 patients and 12,000 health care provider Social, environmental, demographic data

• Social networks and computational social science
• Geo-localized data on exposure to risk factors (air-particulates, density of fast food

restaurants, ...)

• Demographic trends and population structure

Molecular / physiological data

• Human Genome Project (25,000 genes)
• Gene regulatory networks (transcription, replication, translation, ...)
• Protein-protein interaction networks (100,000 proteins)
• 1,000 metabolic pathways and their networks

Big Data – a gamechanger?

Phenotypic / comorbidity data

• If you have disease A, how likely are you to have another disease B?
• Disease-disease networks of 141,000 classified diseases
• 1,000 known drugs
• Disease-drugs networks

Health-care data

• Re-use of population-level health care (claims) data (safeguarded by HVB)
• Anonymized research dataset containing each medical treatment and its cost in Austria
• ver two years
• 100,000,000 doctor visits and 2,000,000 hospitalizations per year and health-care

provider networks with 8,000,000 patients and 12,000 health care provider Social, environmental, demographic data

• Social networks and computational social science
• Geo-localized data on exposure to risk factors (air-particulates, density of fast food

restaurants, ...)

• Demographic trends and population structure

Molecular / physiological data

• Human Genome Project (25,000 genes)
• Gene regulatory networks (transcription, replication, translation, ...)
• Protein-protein interaction networks (100,000 proteins)
• 1,000 metabolic pathways and their networks

Big Data – a gamechanger?

Know How

Big Data

Understanding disease networks

patients with diabetes patients with pancreatic cancer patients with diabetes and pancreatic cancer diabetes pancreatic cancer

• nodes = diseases
• links = diseases are often co-
• ccurring
• size of nodes = disease

prevalence

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Understanding disease networks

0-8 years 9-16 years 17-24 years 25-32 years 33-40 years 41-48 years 49-56 years 57-64 years 65-72 years

Disease risks are predictable! (85%-95% of incidences in population) Chronic diseases are in the center of the networks: risk factors for many other disorders Chronic diseases can not be “repaired”. But digital transformation allows to prevent and manage them

Chmiel A, Klimek P, Thurner S, New J Phys 16, 115013 (2014)

Comorbidity networks and prevention

Prevention of this disease hard Prevention easy

Probability of causation: 0.73

€1,000 €10

Identify comorbidities Check causation Treat causing diseases

From Molecule to Man

What is the optimal prevention strategy? Example HIV: In large cities social policies

• utweigh medical

improvements

Beyond the pill

Moment-by-moment quantification of the individual-level human phenotype in vivo using data from smartphones

JP Onnela, SL Rauch, Neurpsychopharmacology 2016, in press

Data-driven prediction of drug effects and interactions

• From adverse drug event reports

to novel drug interactions

• Automated pharmaco-

epidemiological studies given ADE reports (adjusting for selection biases, sampling variances, confounders, ...)

• First application to Stanford

University Hospital: 47 new validated drug interactions

NP Tatonetti, PP Ye, R Daneshhou, RB Altmann, Science Translational Medicine 2012, 4(125): 125ra31.

A network view on the healthcare system

Nodes in the network = healthcare providers (HCP), for example doctors, pharmacies, hospitals, ... Links in the network = Connect HCPs that share patients Which flows of patients between HCPs are reasonable from a healthcare perspective? Healthcare mangement = shape this network such that it is efficient, sustainable, and resilient.

Patient flow networks

Ideal example of a patient flow: Primary physician Specialist Pharmacy

Patient flow networks

More realistic example... Primary physician Specialist Pharmacy Hospital

Austrian patient flow network

Towards a data-driven era in health care management How robust is this system with respect to adverse events? How do changes to this system affect the efficiency of patient flows? Resilience of the health care system depends on its network structure.

Geographically embedded simulation tool

• f the health care network

What happens if you delete or add certain types of doctors? Resilience can then be quantified and assessed using network-based approaches

Patient flow networks

Economics and quality of health care

• How do billing and prescription practices

depend on such networks?

• Results: Phyisicians adapt their behavior

to match their colleagues  identify inappropriate billings?

• DA Kim, AJ O‘Malley, JP Onnela, submitted (2016)

Who should own your health data?

• Currently it‘s not you.
• With digital transformation the healthcare system becomes exposed

to all kind of cyber vulnerabilities, including breaches of privacy.

• Those we know how to deal with (though healthcare infrastructures

might need to build up competencies in this area)

• However, there is a different type of threat for which we currently

have no response measures:

• Using your health data to „game the system“

Step 1

Identify rare diseases in population…

Step 2

… treated with single drug with expired patent but no generic version.

Step 3

Acquire rights for drug and set price as you want (insurance will pay anyhow)

How to become a millionaire using health data (in 3 steps)

Who should own your health data?

Step 1

Identify rare diseases in population…

Step 2

… treated with single drug with expired patent but no generic version.

Step 3

Acquire rights for drug and set price as you want (insurance will pay anyhow)

How to become a millionaire using health data (in 3 steps)

Martin Shkreli and Turing Pharmaceuticals, February 2015

Conclusions

Big Data allows for the first time to describe health care systems as „complex systems“ in a holistic way, paving the road towards a new type of medicine New era in health care management? Formulation of personalized prevention strategies and data-driven management of health care processes. But how should we decide who might use such data for which purpose?