NC3Rs Maths in Medicine Case Study: Big data for Biologists
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
NC3Rs Maths in Medicine Case Study: Big data for Biologists Manasi Nandi PhD, FBPhS, FHEA Senior Lecturer Integrative Pharmacology King’s College London manasi.nandi@kcl.ac.uk Google images used throughout M Nandi is a co-inventor on IP presented in this presentation
The arterial pulse measured in 1738 …a horse and a glass tube Hales, S. Haemastatics, 3rd edition pg 1. 1738
The arterial pulse measured in 1854 Sphygmograph 1831 Julius Hèrisson 1854 Karl von Vierordt 1863 Étienne-Jules Marey Google images
Mohomed FA The physiology and clinical use of the sphygmograph Med times Gazette 1872; 1:62. M.F O’Rourke, Hypertension 1992; 19:212-217
Frederick Akbar Mahomed Guy’s Hospital 1869 “The pulse, ranks the first among our guides; no surgeon can despise its counsel, no physician shut his ears to its appeal. Since, then, the information the pulse affords is of so great importance and so often consulted, surely it must be to our advantage to appreciate fully all it tells us, and to draw from it all that it is capable of imparting….. …we should study the pulse in its marvellous changes of character and form, as recorded by the sphygmograph” Mohomed FA The physiology and clinical use of the sphygmograph Med times Gazette 1872; 1:62 Mohomed FA The physiology and clinical use of the sphygmograph Med times Gazette 1872; 1:62. M.F O’Rourke, Hypertension 1992; 19:212- 217
Source: Chung, M.K., and Rich, M.W. Introduction to the cardiovascular system. Alcohol Health and Research World 14(4):269–276, 1990.
Are we missing a trick? • As individual control systems in a plane start to fail, so the plane wobbles, turns, spirals and eventually crashes…. • Similarly, in the human body, there may be subtle changes in our own control systems that are changing but by the time we have ‘diagnosed‘ a patient – they have already ‘crashed’
Sepsis Sepsis: Early diagnosis, rapid treatment (fluids, antibiotics) Infection Lungs, GI tract, 20% mortality Genitourinary infection Burns or other open wounds Invasive surgery Septic shock: CVS stabilising agents Exaggerated immune response 30-50% mortality Blood pressure plummets Impaired organ perfusion Multiple organ failure Death 40,000 deaths p.a. in UK NHS cost £2500 per patient per bed day
24 hour model – Typical experience of the animal 24 hours of sepsis in a mouse Hypotension Tachycardia Bradycardia 0 hrs 18-24 hrs Metabolic acidosis Impaired renal function
Can we predict that a patient will crash, before they crash? Endotoxin Mean arterial blood pressure Crash Time
Macro versus Micro circulation 130/90mmHg Mouse 1 450 bpm Mouse 2 110/80mmHg 480 bpm Is there information in the waveform that predicts clinical deterioration above and beyond the ‘set point’? Mouse 3 120/90mmHg 420 bpm Sand et al., J Appl Physiol . 2014
Mini summary • The cardiovascular system is a complex with many homeostatic mechanisms that contribute. • Sepsis is a condition where these systems start to fail and then the patient suddenly crashes. • Data is collected at high fidelity so we collect entire waveform data…but don’t use it…. • Since 1854 scientists and doctors have considered that there is important information in the shape of the wave • How can we quantify the waveform shape?
Waveform shape 120mmHg 80mmHg
Can mathematics help?
Finding patterns in data streams x z y Time series data
Lorenz attractor – a product of chaos theory
Can we plot blood pressure data in 3D? Floris Takens – Mathematician 1981 Philip Aston Mathematician We only have 1 data stream University of Surrey 2013
Jerome Di Pietro – E-learning technology Step 1 – 3D plot with time delays
Step 2 : Rotate
Step 3 : Add density
1 140 BP (mmHg) z 120 y 100 x 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Time (seconds) 140 2 If: 130 X = 108 mmHg z 120 y = 120 mmHg 110 Z = 138 mmHg 100 100 120 y 140 140 130 120 110 100 X 3 4 140 5 6 140 130 z 120 120 110 100 100 100 100 100 120 y 120 140 130 120 110 100 140 120 x 140
AR is fundamentally to normal CV measures…. • Data are viewed in reconstructed phase space enabling continual analysis of lengthy data streams (hours/days of recording). • Baseline wander is factored out in order to focus on subtle changes in the waveform shape/variability – the key inventive step. • Scalar measures from the attractor are used as quantitative physiological readouts of change in waveform shape and variability
Extracting features from the attractor What is the angle of How big is it? rotation? How wide are How dense are the sides? the hot spots?
Application to other periodic waveforms in any species BP ECG Pulse Oximetry Intra Cranial Pressure Central Venous Pressure Respiratory Gary Chaffey and Philip Aston – mathematicians, U of Surrey ; Physionet and other open access online sources
Telemetry continuous waveform data 1000Hz 10am-4pm naïve Chart beat detection 10am-4pm sepsis AR coding software Extraction of AR Extraction of SBP, DBP, measures, size, form, HR, HRV and PP density etc. every half hour every half hour Ying Huang – mathematical Anna Starr & Claire Sand coding Integrative pharmacologists University of Surrey King’s College London
‘Conventional’ ‘Attractor’ measures measures Head to head comparison ROC AUC Hitesh Mistry Statistics Manchester
Healthy ROC AUC =1 Septic Healthy Septic ROC AUC ~ 0.5 ROC AUC = 1 – good discrimination between healthy and septic ROC AUC = 0.5 – Poor discrimination/random chance
‘Conventional’ ‘Attractor’ measures measures Conventional Baseline Baseline vs. AR measure Baseline vs. Baseline vs. measure vs. saline sepsis saline sepsis Systolic BP 0.52 0.54 AR measure A2 0.51 0.78 Diastolic BP 0.52 0.86 AR measureA3 0.63 0.76 Pulse 0.54 0.82 AR measure A4 0.62 0.65 Pressure AR measure A5 0.57 0.82 MABP 0.71 AR measure A6 0.59 0.64 Heart Rate 0.58 0.86 AR measure A7 0.53 0.83 Heart rate 0.53 0.84 variability AR measure A9 (HRV RR) AR measure P4 0.52 0.99 HRV SDRR 0.65 0.62 AR measure P6 0.62 0.96 HRV RMSRR 0.61 0.54
Pilot studies using HESI data Pimombendan- PDE3 inhibitor – positive ionotrope Itraconozole- antifungal – negative ionotrope Hypothesis: AR can be used to extract information about changes in cardiac contractility from a peripheral BP waveform HESI meeting, 13-15th June 2017, Dublin
Summary
Jerome Di Pietro Pete Charlton Claire Sand Ying Huang KCL KCL KCL Surrey Sepsis models E-learning technologist Coding developments Carolyn Lam Coding developments Integrative pharmacologist Clinical data KCL Maths PhD student Biomedical engineer Pharmacologist; Data processing Philip Aston Jane Lyle Miquel Serna Pascual Surrey Surrey Coding developments KCL Mathematics Pharmacologist; Maths research PhD Hitesh Mistry Esther Bonet Luz Data processing student Manchester Mathematician Statistics, Machine learning applied mathematics and algorithms Phil Chowienczyk KCL Clinical Pharmacology Cardiovascular waveforms Gary Chaffey Surrey Ashley Noel Hirst Jordi Alastruey Anna Starr Richard Beale Coding developments Nuffield project student KCL KCL KCL/GSTT Maths research fellow In silico modelling Biomed engineering Sepsis models and AR analysis Critical care medicine Integrative pharmacologist Cardiovascular waveforms Cath Williamson Jenny Venton Guy’s & St Thomas’ KCL Women’s Health Mathematics Duncan Mcrae and Mary Anton Coding developments Royal Brompton Data processing Paediatric Intensive Care
http://ehealth.kcl.ac.uk/cardiomorph/
You can also read