Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
FP7-ICT-2013-10
ICT-WP-2013.5.1
Personalized health, active ageing, and independent living
Capturing Scientific Knowledge ...
SciKnow, K-CAP, Palisades, NY, Oct 2015
What are we doing & why?
 A model of evidence-based medical risk
 Clinical resea...
SciKnow, K-CAP, Palisades, NY, Oct 2015
Existing risk prediction tools
 Hardcoded statistical models
 Input age, height,...
SciKnow, K-CAP, Palisades, NY, Oct 2015
 Cardiorenal patients
 Elevated risks of comorbidities
 Risk calculation & deci...
SciKnow, K-CAP, Palisades, NY, Oct 2015
A sample risk factor association
 Diabetes causes ischemic heart disease
 Diabet...
SciKnow, K-CAP, Palisades, NY, Oct 2015
SciKnow, K-CAP, Palisades, NY, Oct 2015
Types of risk factor
 Environmental
 Demographic
 Genetic
 Behavioural
 Biome...
SciKnow, K-CAP, Palisades, NY, Oct 2015
Observables
 How do we know when a risk factor applies to a patient?
 Who fits t...
SciKnow, K-CAP, Palisades, NY, Oct 2015
Capturing risk factors
 Clinician-defined literature search methodology
 Identif...
SciKnow, K-CAP, Palisades, NY, Oct 2015
How did it go?
 93 risk factor associations
 Based on 45 different risk elements...
SciKnow, K-CAP, Palisades, NY, Oct 2015
How did it go?
 Relatively straightforward
 Identifying risk factors sometimes t...
SciKnow, K-CAP, Palisades, NY, Oct 2015
Observable expressions
 “Diagnosed AND/OR between 8% and 9%”
 In clinical litera...
SciKnow, K-CAP, Palisades, NY, Oct 2015
Lessons & future work
 “Hidden” clinical knowledge
 Standard clinical terminolog...
SciKnow, K-CAP, Palisades, NY, Oct 2015
acknowledgment
work funded under project CARRE: Personalized patient empowerment
a...
Upcoming SlideShare
Loading in …5
×

Capturing Scientific Knowledge of Medical Risk Factors

333 views

Published on

A. Third, E. Kaldoudi, G. Gotsis, S. Roumeliotis, K. Pafili, J. Domingue, Capturing Scientific Knowledge on Medical Risk Factors, K-CAP2015: 8th International Conference on Knowledge Capture, ACM, Palisades, NY, USA, Oct. 7-10, 2015 http://www.isi.edu/ikcap/sciknow2015/#papers

In this presentation (and paper), we describe a model for representing scientific knowledge of risk factors in medicine in an explicit format which enables its use for automated reasoning. The resulting model supports linking the conclusions of up-to-date clinical research with data relating to individual patients. This model, which we have implemented as an ontology-based system using Linked Data, enables the capture of risk factor knowledge and serves as a translational research tool to apply that knowledge to assist with patient treatment, lifestyle, and education. Knowledge captured using this model can be disseminated for other intelligent systems to use for a variety of purposes, for example, to explore the state of the available medical knowledge.

Published in: Science
  • Be the first to comment

  • Be the first to like this

Capturing Scientific Knowledge of Medical Risk Factors

  1. 1. FP7-ICT-2013-10 ICT-WP-2013.5.1 Personalized health, active ageing, and independent living Capturing Scientific Knowledge of Medical Risk Factors Allan Third, Eleni Kaldoudi, George Gkotsis, Stefanos Roumeliotis, Kalliopi Pafili, John Domingue 1st International Workshop on Capturing Scientific Knowledge Palisades, New York, 7th October 2015
  2. 2. SciKnow, K-CAP, Palisades, NY, Oct 2015 What are we doing & why?  A model of evidence-based medical risk  Clinical research identifies risk factors  “If you smoke, your risk of X is increased by…”  Used to advise patients on  Lifestyle changes  Risk mitigation  Chronic disease management
  3. 3. SciKnow, K-CAP, Palisades, NY, Oct 2015 Existing risk prediction tools  Hardcoded statistical models  Input age, height, weight, some lifestyle factors…  Output a single probability & some general recommendations  Less informative  Not extensible  New science: new model/tool
  4. 4. SciKnow, K-CAP, Palisades, NY, Oct 2015  Cardiorenal patients  Elevated risks of comorbidities  Risk calculation & decision support  Based on at-home monitoring  Scientific contribution:  Generalised & extensible risk model  Capable of fine-grained & hypothetical reasoning
  5. 5. SciKnow, K-CAP, Palisades, NY, Oct 2015 A sample risk factor association  Diabetes causes ischemic heart disease  Diabetic and male: 2.82 x more likely to develop IHD  Confidence interval 2.35 – 3.38  Source: Pubmed ID 24859435  Diabetic and female: 2.16 x more likely to develop IHD  Confidence interval 1.82 – 2.56  Source: Pubmed ID 24859435  Combine en masse for extensible risk calculation
  6. 6. SciKnow, K-CAP, Palisades, NY, Oct 2015
  7. 7. SciKnow, K-CAP, Palisades, NY, Oct 2015 Types of risk factor  Environmental  Demographic  Genetic  Behavioural  Biomedical
  8. 8. SciKnow, K-CAP, Palisades, NY, Oct 2015 Observables  How do we know when a risk factor applies to a patient?  Who fits the population criteria in the evidence?  Observations of a patient  Manual or automatic  Logical expression describing study population  E.g., “sex = “female” and diabetes = “diagnosed””
  9. 9. SciKnow, K-CAP, Palisades, NY, Oct 2015 Capturing risk factors  Clinician-defined literature search methodology  Identify quantified risk factors & relevant population  Custom (Drupal) forms reflecting model  Automatically convert into RDF and store  Review  Repeat..
  10. 10. SciKnow, K-CAP, Palisades, NY, Oct 2015 How did it go?  93 risk factor associations  Based on 45 different risk elements  [Pause for valiant attempt at a demo]
  11. 11. SciKnow, K-CAP, Palisades, NY, Oct 2015 How did it go?  Relatively straightforward  Identifying risk factors sometimes tricky  Familiar language& natural model for clinicians  Inconsistencies in clinical writing (“risk factor”)  Biggest issue:  Logical expressions for grounding in observables
  12. 12. SciKnow, K-CAP, Palisades, NY, Oct 2015 Observable expressions  “Diagnosed AND/OR between 8% and 9%”  In clinical literature but not easily made machine-readable  OR not in clinical literature  Tacit knowledge in clinical process  “?diabetes = “diagnosed” || (?HbA1c > 8 && ?HbA1c < 9)”
  13. 13. SciKnow, K-CAP, Palisades, NY, Oct 2015 Lessons & future work  “Hidden” clinical knowledge  Standard clinical terminology hides some generalities  “Positive” risk factors  Ongoing work on decision support/visualisation  Easy to generalise to other areas of medicine  Outside medicine?  Curation
  14. 14. SciKnow, K-CAP, Palisades, NY, Oct 2015 acknowledgment work funded under project CARRE: Personalized patient empowerment and shared decision support for cardiorenal disease and comorbidities co-funded by the European Commission under the Information and Communication Technologies (ICT) 7th Framework Programme Contract No. FP7-ICT-2013-611140 CARREhttp://www.carre-project.eu/ cite as A. Third, E. Kaldoudi, G. Gotsis, S. Roumeliotis, K. Pafili, J. Domingue, Capturing Scientific Knowledge on Medical Risk Factors, K-CAP2015: 8th International Conference on Knowledge Capture, ACM, Palisades, NY, USA, Oct. 7-10, 2015 http://www.isi.edu/ikcap/sciknow2015/#papers

×