3.4 Machine Learning Driven Prognostic Model
3.4.6 Online Clinical Prognostic Model
The next stage in the prognostic model development (as shown in Figure 3.4) is to get these novel prognostic models incorporated as part of the clinical workflows for primary and secondary care clinicians in the UK and US. This objective is reached through the implementation of cardiac chest pain and heart disease prog- nostic models as online clinical prototypes. These online clinical risk assessment prototypes are used for the clinical validation and evaluation purposes by consul- tant cardiologist, Professor Stephen Leslie from Raigmore Hospital and Professor Warner Slack from Harvard Medical School as well as primary care clinician (GP) from Edinburgh who utilised heart disease prognostic models for clinical trials us- ing real patient data. These online prognostic models could be used to collect new data for further research work and could to be used with an online training algorithm to improve performance of existing models and to optimise machine learning inputs. These online prognostic models have been developed using PHP scripts to acquire patient data and HTML front end was developed to provide the risk score.
3.5
Conclusion and Discussion
In this chapter, we proposed a novel ontology and machine learning driven hybrid clinical decision support framework for cardiovascular preventative care. The key components of the proposed framework are (1) Ontology driven clinical risk as- sessment and recommendation system (ODCRARS) and (2) The Machine Learn- ing Prognostic System (MLDPS). The key components are developed in close collaboration with cardiologists from UK and US hospitals. Clinical question- naires encoded in the ontology driven cardiovascular risk assessment and recom- mendation system were written by Professor Warner Slack from Harvard Medical School in the US. The machine learning driven prognostic models for the cardiac chest pain and heart disease are developed in collaboration with primary and sec- ondary care clinicians.These prognostic models could help clinicians reduce load on overly prescribed angiography treatments in a cost effective manner. Details
of development, design and validation of the key components will be provided in chapter 4 and 5.
The proposed framework will also pave the way for the development of cost effective and patient centric preventative care solutions for chronic diseases with high mortality rates, such as breast cancer and diabetes. These chronic dis- eases could be largely preventable through close partnership among healthcare providers, commercial partners and researchers working in the healthcare infor- matics domain towards developing innovative doctor-patient based interactive col- laborative care solutions. The proposed framework will facilitate development of the next generation commercial clinical decision support systems with a learning capability based on machine learning (for information exchange among key com- ponents for risk calculation for cardiac chest pain and heart disease conditions). This could be utilised by primary and secondary care clinicians in the UK and US as a cardiovascular preventative care solution. The proposed novel ontology and machine learning driven hybrid clinical decision support framework exploits both (ontology and machine learning driven) approaches. Our proposed frame- work combines both clinical expert’s knowledge (encoded in the knowledge-based ODCRARS) and evidence-based/data driven MLDPS in an intelligent manner to deliver an effective, holistic and cost effective cardiovascular preventative care solution.
Chapter 4
Ontology Driven Clinical Risk As-
sessment and Recommendation Sys-
tem (ODCRARS) for Cardiovas-
cular Preventative Care
Chapter 3 presented the proposed novel ontology and machine learning driven hybrid clinical decision support framework for cardiovascular preventative care. This chapter focuses on the design, development and clinical validation of the ODCRARS.
The ODCRARS is developed in order to provide a cardiovascular preventative care solution for primary and secondary care clinicians and patients. It provides clinicians with a snapshot of a patient’s medical history in the form of patient medical records, details of recommended lab tests and medication; provides rel- ative and absolute cardiac risk scores; cardiac chest pain and heart disease risk scores. This provides a holistic cardiovascular decision support as part of a triage mechanism for primary and secondary care clinicians. The ODCRARS is de- veloped under the close supervision of Consultant Cardiologist, Professor Calum MacRae from Brigham and Women’s Hospital, Harvard Medical School and of Clinical Informatics expert, Professor Warner Slack from Beth Israel Deaconess Medical Centre, Harvard Medical School.
The detailed design, development and validation details of various compo- nents of the ODCRARS which includes ontology driven intelligent context aware
information collection, patient medical records, patient semantic profile, ontol- ogy driven clinical decision support including NICE/Expert driven clinical rules engine components are provided in detail.
In the latter part, ontology driven clinical decision support which is imple- mented through the recommendation ontology and NICE/Expert driven clinical rules engine is presented. The development, design and validation of the recom- mended ontology which utilises the patient’s semantic profile for the recommen- dation of lab tests and prescription of medication is discussed in detail.
We also discuss development of the NICE/Expert driven clinical rules en- gine and its utilisation in the cardiovascular risk scores calculation for various cardiovascular diseases. It also helps to control the patient flow within the car- diovascular preventative care solution. The outcome general cardiac risk score calculation using the Framingham risk score calculator is also explained.
The integration of the machine learning driven prognostic models (cardiac chest pain and heart disease prognostic models) is discussed at the end. These prognostic models are developed using the machine learning driven prognostic system, further details of the machine learning drive prognostic system (MLDPS) will be provided in Chapter 5.
4.1
Implementation of the Ontology Driven Clin-
ical Risk Assessment and Recommendation
System (ODCRARS)
In chapter 3, we introduced a novel ontology and machine learning driven hybrid clinical decision support framework for cardiovascular preventative care, this sec- tion focuses on the design and development aspects of different components of the proposed ontology driven clinician risk assessment and recommendation sys- tem. The components of the proposed ODCRARS are chronologically numbered in Figure 3.1 for explanation and clarity purposes.
The proposed ODCRARS aims to provide a cardiovascular preventative care solution for primary and secondary care clinicians in the UK and US hospitals by
way of automating patient encounter with the physician where a standard panel of health information including basic physiological parameters such as weight or blood pressure and patient demographics information is collected to generate their medical records. The doctor-patient interaction/interviewing mechanism is mimicked using the ontology driven context aware information collection com- ponent. The proposed ODCRARS recommends a number of lab tests (e.g. for cardiovascular, diabetes, cholesterol and other common risk factors), potentially additional evaluations such as an ECG or stress test and with the results sees the consultant again, who, based on the results of the physical exam and the laboratory tests often prescribe one of several classes of medications, e.g. an aspirin, a statin, an ACE inhibitor or an Angiotensin receptor blocker. It also provides cardiac risk scores for various cardiovascular diseases along with cardiac chest pain and heart disease risk scores which are calculated through the evidence based MLDPS.
We utilise ontology based approach in the development of ODCRARS. On- tology driven approach offers several advantages over conventional software engi- neering techniques, Firstly, our proposed ODCRARS is more convenient to up- date as modifying the ontology layer can be done without the need for additional and costly software engineering work. The clean separation between core system functionalities and the knowledge base utilised by the system means that the lat- ter can me modified should requirements or clinical expert’s knowledge change. Secondly, the ontology layer enables the system to perform operations, such as clinical decision support, which are cumbersome to implement using database and distributed system technologies on their own.
The proposed clinical decision support framework as shown in Fig 3.1 could thus be used for automatically conducting patient pre-visit interviews. It will not replace a human doctor, but could be used as a triage system to prepare a patient’s summary/doctor’s notes and pre-order appropriate tests by facilitating clinicians to make better use of their consultation time in providing direct patient-centric care.
1. Ontology driven intelligent context aware information collection. 2. Patient Medical Records.
3. Patient Semantic Profile.
4. Ontology driven clinical decision support and NICE/Expert driven clinical rules engine.