ICT for Healthcare Challenges and Solutions Roberto Montemanni - - PowerPoint PPT Presentation

University of Lugano University of Applied Sciences of Southern Switzerland

IDSIA

Dalle Molle Institute for Artificial Intelligence

ICT for Healthcare

Roberto Montemanni

Challenges and Solutions

roberto@idsia.ch

Research Institute in Lugano, Switzerland since 1988

Dalle Molle Institute for Artificial Intelligence

Research fields:  Optimisation, simulation and decision support systems  Uncertain reasoning, data mining and big data  Machine learning and artificial neural networks  Cognitive and mobile robotics 57 staff members: 7 Professors 19 PHD Students 8 Senior Researchers 3 Master Students 17 Researchers 1 Secretary 2 Research Assistants Optimisation, simulation and decision support systems

IDSIA

Basic Research Applied Research

University of Lugano University of Applied Sciences of Southern Switzerland

IDSIA

Dalle Molle Institute for Artificial Intelligence

IDSIA

Contributors to this talk

Giorgio Corani

Senior researcher IDSIA-SUPSI

Gianni A. Di Caro

Senior researcher IDSIA-SUPSI

Jerome Guzzi

PhD student IDSIA-USI

Thi Viet Ly Nguyen

PhD student IDSIA-USI

Tomas E. Nordlander

Research manager SINTEF Norway Senior researcher IDSIA-SUPSI Associate professor Carnegie Mellon

Unsustainable Development

Unsustainable Development

Unsustainable Development

 More and more patients – population ageing along with the

skewed demographic development

 Increased quality - Public’s rising expectations of quality treatment  Less funding - Government desire to reduce the health expenditure

Unsustainable Development

Proportion of pupils that will need to work in Norwegian healthcare to manage the increase of patients

Source: Gunnar Bovim, CEO, Central Norway Regional Health Authority RHF, “Strategy 2020”

Possible Remedies?

 Organization changes  Clinical Pathways  Electronic patient journals  More efficient resources utilization  Use of ICT systems – both hardware and software Strategic Tactical Operational Software Hardware

Optimization in Health care?

 Health care, a business like any other

 Trying to utilize resources (equipment, staff, etc.)

efficiently at strategic, tactical, and operational level while involving multiple decision-makers with conflicting goals

 Problems are

 Too complex to manually find good solutions  Time demanding and repetitive  Strategic and financial heavy

 The health care sector lacks Optimization/Operations Research support tools

 Why?

Partitioning of Health Care Decision making processes

 Initially, small, clean cut

subproblems were considered and solved via Optimization tools

 Then larger and more complex

problem instances have been considered, but always respecting the given partition

 This approach never fully

convinced Health Care practitioners

Partitions weakness

 Simplified

 Optimization problems solved

today are isolated problems



Resource sharing across the

• rganisation is totally overlooked

 Outdated

 The problem partition rationale is

based on what algorithms and computers could handle decades ago

 However, time has passed and

the effectiveness of algorithms and computer power have drastically improved

Problem Interrelationship

Greater Computational and Algorithmic Power

 Greater Processing Power

 Moore’s law [Moore, 1975] predicts that processing power* will

double approximately every year and its prediction has held true for several decades.

 The increase in memory is also beneficial for the algorithms.

 Improved Methods

 Improved Algorithms  Recent developments in hybrid, parallel methods

*More precisely, the number of transistors on integrated circuits doubles approximately every two years, which roughly double the processing power. This prediction has roughly been true until 2013.

Example: Linear Programming solvers During the period 1987 to 2000, Bixby (2002) estimated a speedup increase of six orders of magnitude in solving power, where processing power and memory contributed by half and the remaining three orders of magnitude is due to improved algorithm: "A model that might have taken a year to solve ten years ago, can now solve in less than 30 seconds.”

Greater Computational and Algorithmic Power

Partitions changes needed

 When working with the old partitions, we disregard the larger picture and miss chances of really efficient solutions.

Algorithmic changes needed

 In recent years, chip manufacturers needed to change the architecture and started to produce multi-core processors to allow them to continue doubling the processor power.  In addition, driven by the computer game race for ever more impressive graphics, more powerful programmable Graphics Processing Units (GPUs) were produced.  Also, we now have Accelerated Processing Units (APUs)  Most classic algorithms still use a sequential optimisation paradigm, that was not an issue while we had an exponential increase of processor clock frequency, now things have changed, and there are great margins of improvements

Robust Optimization for Home Health Care (HHC)

 Phases in Home Health Care Planning:

 rostering  assignment  routing  scheduling

 Current Situation: Phases addressed

 Individually and separately  With simplified HHC models

 Aims of the project:

 Providing

an integrated approach to tackle the different phases together

 Considering realistic features (e.g.,

uncertainty, workload balancing, loyalty, etc.) within the model

MILP model to represent the problem

• 1. Minimize the number
• f unscheduled tasks
• 2. Remain a consistent

patient-nurse loyalty

• 3. Minimize the total
• f overtime cost
• 4. Minimize the total cost of time

window violation

• 5. Minimize the total waiting time
• 6. Maximize the value of the minimum

number of tasks for each nurse

Methodology: the model

19

Constraints:

• Loyalty
• Assignment
• Routing
• Balance workload
• Labor regulations
• Hard time window
• Visiting
• Variables domains

Methodology: the model

Methodology: dealing with uncertainty

Robust optimization*: Optimization field aiming at retrieving solutions resistant against uncertainty

*Soyster (1973); Ben-Tal and Nemirovski (1997)

Conservativeness degree**:

Desired level of protection against uncertainty of the final solution

**Bertsimas and Sim (2003)

Example: in robust home health care service the Conservativeness degree 1 is the level of pessimistic thinking of the manager about the budget for a fraction of the number of missing nurses Uncertainty: Unpredictability of problem data

Solution cost in the best case Robustness minimum cost sol most robust sol compromise sols

In our approach robustness is embedded within the MILP

Methodology: dealing with large-scale real instances

 Robust MILP exact models intractable for real instances  Need for computationally tractable methods  Heuristic algorithms?

 Trading optimality for speed  Might generate very suboptimal solutions  Robustness concepts difficult to integrate

 MatHeuristic hybridizations!

 Inner robust MILP model for subproblems  Outer Genetic Algorithm (population based heuristic algorithm)

Matheuristics

Heuristic: Genetic algorithm Mathematical Programming: MILP model

Summary

OmniProfiler: Monitoring elderly people at home

 Goals:

 to monitor elderly people living at home  to track their behavior, detecting

anomalous patterns which might show the early stages of a decline of cognitive faculties  Technology involved:

 Intelligent home automation systems  Sensors to collect both environmental and biometric data

 Monitored variables:

 Amount and time of sleep  Physical activity (number of steps during the day)  Body weight  ...

 Profiling technique:

 Statistical analysis of control charts

Statistical analysis via control charts

 Green points: observed values  Red/blue points: anomalous data  Yellow points: estimated trend over time

Body weight  By control charts, it is detected whether the past week has shown anomalous patterns  Moreover it is estimated how the monitored variables evolve over time

Heart rate

Statistical analysis via control charts

Sleeping hours Physical activity

Alma: Ageing without Loosing Mobility and Autonomy

 Target: Support the autonomous mobility, navigation, and orientation of the mobility- impaired person (elderly, temporarily

• r

permanently disabled)  Instruments: Realization and combination of a set of advanced hardware and software technologies into an integrated and modular cost-effective system  Validation: Two pilot applications with different scenarios and therapeutic issues for primary (elderly, patients) and secondary (care givers) end-users  Long-term

• bjective:

Bring technological results to the market!

Alma: Ageing without Loosing Mobility and Autonomy

University of Lugano University of Applied Sciences of Southern Switzerland

IDSIA

Dalle Molle Institute for Artificial Intelligence

Healthcare is in urgent need of ICT