Scoring Systems in the Intensive Care Unit and Time Series Preprocessing. Orhan Konak Data Management for Digital Health Winter 2018

Size: px

Start display at page:

Download "Scoring Systems in the Intensive Care Unit and Time Series Preprocessing. Orhan Konak Data Management for Digital Health Winter 2018"

Pamela Wells
5 years ago
Views:

1 Scoring Systems in the Intensive Care Unit and Time Series Preprocessing Orhan Konak Digital Health Winter

2 Data Management & Foundations Real-world Use Cases Where are we? Oncology Nephrology Intensive Care Additional Topics Biology Recap Data Sources Data Formats Business Processes Processing and Analysis 2

Recap ICU Patients with life threatening illness Supporting failing organ systems Highly specialized environment Equipment Monitoring Organ support equipment Intravenous lines, feeding, suction,

3 Recap ICU Patients with life threatening illness Supporting failing organ systems Highly specialized environment Equipment Monitoring Organ support equipment Intravenous lines, feeding, suction, drains, catheter Constant monitoring of bodily functions Observing vital signs Key to improve patients survival Generates data like waveforms 3

Patient Data Management System (PDMS) Document vital parameters sampled by monitors Demands on PDMS have increased immensely PDMS are currently expected to assist clinicians at every

4 Patient Data Management System (PDMS) Document vital parameters sampled by monitors Demands on PDMS have increased immensely PDMS are currently expected to assist clinicians at every level of intensive care, e.g. Strategic level of physician orders and prescriptions Operational level Administrative level 4

5 Does Introduction of a PDMS Improve the ICU? PDMS implementation costs range from 15k to 20 keur per bed Costs and revenues increased continuously over the years No clear evidence for cost savings after the PDMS introduction PDMS has resulted in better patient outcomes 5

6 /Produktkatalog/metavisionperfusion/ /82/240_F_ _K7fTE7wPn pokyjssbqxa3zvaxtqrbcdm.jpg ainadvisor.com/traum a-pain/spect-vs-mrito-diagnosetbi/article/579126/ r.de/motivgalerie/karriereberuf/biologe.html PDMS IT Infrastructure HIS Patient Data Management System 6

Scoring Systems in the ICU Scoring system as clinical decision support Severity scales important to predict Patient outcome, Comparing quality-of-care, and Stratification for clinical trials.

7 Scoring Systems in the ICU Scoring system as clinical decision support Severity scales important to predict Patient outcome, Comparing quality-of-care, and Stratification for clinical trials. Essential part of improvement in clinical decisions and in identifying patients with unexpected outcomes Using logistic regression models 7

8 Scoring Systems in the ICU Scoring system usually comprises of two parts a score (a number assigned to disease severity) and a probability model (equation giving the probability of hospital death of the patients) A =10 Number P A Probability model Score 8

Types of Scoring Systems Commonly Used Adult ICU Scoring Systems First-day scoring systems Acute Physiology and Chronic Health Evaluation (APACHE) Simplified Acute Physiology Score (SAPS) Mortality

9 Types of Scoring Systems Commonly Used Adult ICU Scoring Systems First-day scoring systems Acute Physiology and Chronic Health Evaluation (APACHE) Simplified Acute Physiology Score (SAPS) Mortality Prediction Model (MPM) Repetitive scoring systems Organ System Failure (OSF) Sequential Organ Failure Assessment (SOFA) Organ Dysfunction and Infection System (ODIN) Multiple Organ Dysfunction Score (MODS) Logistic Organ Dysfunction (LOD) 9

10 Severity Scores in Medical & Surgical ICU Timeline APACHE SAPS APACHE II SAPS II MPM APACHE III MODS MPM II ODIN SOFA CIS SAPS III APACHE IV current 10

Glasgow Coma Score (GCS) Let s Take a Closer Look at One Score Neurological scale Give a reliable and objective way of recording the conscious Initially used to assess a person's level of

11 Glasgow Coma Score (GCS) Let s Take a Closer Look at One Score Neurological scale Give a reliable and objective way of recording the conscious Initially used to assess a person's level of consciousness after a head injury Now used by first responders, EMS, nurses, and doctors Part of several ICU scoring systems, including APACHE II, SAPS II, and SOFA 11

http://www.isfsports.org/ gymnastics https://melibeeglobal.

g/2015/01/the-lack-of-nonverbal-communication-in-adigital-world/

com/ GCS Calculation Behavior Response Eye Opening Response 4

Oriented to time, person and place 4 Confused 3 Inappropriate words 2

12 gymnastics g/2015/01/the-lack-of-nonverbal-communication-in-adigital-world/ GCS Calculation Behavior Response Eye Opening Response 4 Spontaneously 3 To speech 2 To pain 1 No response Verbal Response 5 Oriented to time, person and place 4 Confused 3 Inappropriate words 2 Incomprehensible sounds 1 No response Total Score Mild Moderate 9 12 Severe 3 8 Motor Response 6 Obeys command 5 Moves to localized pain 4 Flex to withdraw from pain 3 Abnormal flexion 2 Abnormal extension 1 No response 12

13 GCS Calculation E V M 13

14 GCS Example 1 Infant, moves spontaneously towards objects and follows them, smiling and orienting towards interesting sounds. The infant opens the eyes spontaneously. E V M

15 GCS Example 2 Adult, moves the hand away when applying pressure on the nail bed. The patient can make words but not form sentences. The patient opens the eyes to pain, but not to speech. E V M

16 GCS Example 3 Adult, moves hand towards head when applying pressure above the eye socket. The patient is disoriented but able to form sentences. The patient opens the eyes in response to speech. E V M

Acute Physiology And Chronic Health Evaluation II (APACHE II) Calculation - Patient s Age and 12 Routine Physiological Measurements AaDO2 or PaO2 (depending on FiO2) Potassium (serum) Creatinine

17 Acute Physiology And Chronic Health Evaluation II (APACHE II) Calculation - Patient s Age and 12 Routine Physiological Measurements AaDO2 or PaO2 (depending on FiO2) Potassium (serum) Creatinine Temperature (rectal) Sodium (serum) Hematocrit Mean arterial pressure Respiratory rate White blood cell count ph arterial Heart rate Glasgow Coma Scale 17

18 Comparison of ICU Scoring ICU Scoring System Timing of data collected Physiological values Other required data Total data elements required Original reported mortality prediction performance SAPS III Prior to and within 1 hour of ICU admission 10 Age, six chronic health variables, ICU admission diagnosis, ICU admission source, LOS prior to ICU admission, emergency surgery, infection on admission, four variables for surgery type 26 AUC = 84.8% (n=16,784) APACHE IV First ICU day (16-32 h depending on time of admission) 17 Age, six chronic health variables, ICU admission diagnosis, ICU admission source, LOS prior to ICU admission, emergency surgery, thrombolytic therapy, Fio 2, mechanical ventilation 32 AUC = 88.0% (n=52,647) MPM III Prior to and within 1 hour of ICU admission 3 Age, three chronic health variables, five acute diagnosis variables, admission type (e.g., medical-surgical) and emergency surgery, CPR within 1 h of ICU admission, mechanical ventilation, code status 16 AUC = 82.3% (n=50,307) 18

19 Which Score to Use? APACHE, SAPS, MPM only of historic significance? APACHE II most widely used in USA SAPS II commonly used in Europe APACHE III not in public domain SAPS III, APACHE IV better design MODS, LODS uncommonly used 19

20 The Ideal Scoring System On the basis of easily/routinely recordable variables Well calibrated Applicable to all patient populations Can be used in different countries The ability to predict functional status or quality of life after ICU discharge No scoring system currently incorporates all these features 20

https://en.wikipedia.org/wiki/professor_frink https://en.wikipedia.org/wiki/dr.

21 Taking Off the Physician s Glasses But I'm seeing things very clearly! Everything s kind of blurry! 21

22 timierung-ultimo-go/ different-radiationtherapy-types/ om/cancer/vital-signsmonitor#1 Selected Data Source Electrocardiography (ECG) Hospital Intensive Care Unit Monitoring? ECG Signal wiki/electrocardiography 22

Function of ECG Process of recording the electrical

skin Electrodes detect the tiny electrical changes

23 Function of ECG Process of recording the electrical activity of the heart Electrodes placed over the skin Electrodes detect the tiny electrical changes on the skin Commonly performed to detect any cardiac problems

Simplified ECG Signal Chain Amplifier Filter Analog-to-digital converter https://www.boehm-elektromedizingmbh.de/shop/neue-produkte/philipsintellivue-mx550/ https://de.banggood.

24 Simplified ECG Signal Chain Amplifier Filter Analog-to-digital converter Monitoring-Hearbeat-Sensor-Module-for-Arduino-Monitor-Devices-p html?akmClientCountry=DE&gmcCountry=DE&currency=E UR&createTmp=1&utm_source=googleshopping&utm_medium=cpc_ bgcs&utm_content=zouzou&utm_campaign=pla-deg-label pc&ad_id= &cur_warehouse=cn 24

Amplifier Electronic device that increases the power of a

with a larger amplitude https://www.conrad.

25 Amplifier Electronic device that increases the power of a signal. It does this by taking energy from a power supply and controlling the output to match the input signal shape but with a larger amplitude stereo-verstaerker-bausatz-9-vdc-12-vdc-18-vdc- 20-w html 25

Electronic Filter Circuits which perform signal processing functions Remove unwanted frequency components from the signal Include the low-pass filter, the high-pass filter, the band-pass filter, and

26 Electronic Filter Circuits which perform signal processing functions Remove unwanted frequency components from the signal Include the low-pass filter, the high-pass filter, the band-pass filter, and the notch filter (or the band-reject or band-stop filter) Bandpass-Band-Resistant-All-Pass-Low-Pass-High-Pass-p html?akmClientCountry=DE&gmcCountry=DE&curre ncy=eur&createtmp=1&utm_source=googleshopping&utm_m edium=cpc_bgcs&utm_content=zouzou&utm_campaign=pladeg-ele-pc&cur_warehouse=cn 26

Analog-to-Digital Converter (ADC) Converts an analog voltage to a digital number Converts the output data into a series of digital values by approximating the signal with fixed precision

27 Analog-to-Digital Converter (ADC) Converts an analog voltage to a digital number Converts the output data into a series of digital values by approximating the signal with fixed precision Detecting binary signals: is the button pressed or not? These are digital signals Converts voltage as a binary 0 or

Sampling Rate Nyquist Shannon Sampling Theorem The minimum rate at which digital sampling can accurately record an analog signal is known as the Nyquist Frequency, which is double the highest

28 Sampling Rate Nyquist Shannon Sampling Theorem The minimum rate at which digital sampling can accurately record an analog signal is known as the Nyquist Frequency, which is double the highest expected signal frequency Nyquist frequency = 2 x highest expected frequency E.g. you are recording ECG in humans that has components which can reach up to 50 Hz as their highest expected frequency, so the minimum sampling rate should be 100 Hz 28

Data Acquisition Biological signals recorded via a data acquisition unit (DAQ) Converted to a digital signal by DAQ unit Resulting digital signal is sampled at regular intervals by

29 Data Acquisition Biological signals recorded via a data acquisition unit (DAQ) Converted to a digital signal by DAQ unit Resulting digital signal is sampled at regular intervals by analysis software Data stored and displayed on computer html 29

https://en.wikipedia.org/wiki/dr._nick What is a Time Series?

1750 15:51:01 1 25.2250 15:51:02 2 25.2500 15:51:03 3 25.2500 15:51:04 4 25.2750 15:51:05 5 25.3250 15:51:06 6 25.3500 15:51:07 7 25.

30 What is a Time Series? A time series is a collection of observations made sequentially in time Time Time series ID Value Time Series? 15:51: :51: :51: :51: :51: :51: :51: :51: :51: :51: :52: :52: :52: :52: :52: :52:

31 Time Series are Ubiquitous! People measure things Everything which is on the monitoring screen (ECG, BP, RR, ) Angela Merkel s popularity rating The weather in Berlin German Stock Index DAX and things change over time 31

32 Time Series How do we work with very large databases? Why is working with time series so difficult? 1 hour of ECG data: 1 GB Typical weblog: 5 GB per week Space Shuttle database: 200 GB and growing Since most of the data lives on disk (or tape), we need a representation of the data we can efficiently manipulate 32

Time Series Database Time-series data applications are proliferating As a result time-series databases are in fashion Most adopt a NoSQL model Developers preferred NoSQL to relational databases for

33 Time Series Database Time-series data applications are proliferating As a result time-series databases are in fashion Most adopt a NoSQL model Developers preferred NoSQL to relational databases for time-series data by over 2:1 Reason for adopting NoSQL time-series databases comes down to scale 33

34 Time Series Miscellaneous data handling problems Why is working with time series so difficult? Differing data formats Differing sampling rates Noise, missing values, etc. 34

35 hythmias/a009.htm Time Series Let s Compare ECG Signals What are you doing there? I m comparing the curves and try to find similarities, respectively abnormalities. Let me show you how to do it. 35

36 Euclidean Distance Metric Comparing to Time Series Let s assume we want to compare two time series About 80% of published work in data mining uses Euclidean distance 36

measure the distance between two raw time series, we may get very unintuitive

very sensitive to some distortions in the data.

37 Preprocessing the Data Before Distance Calculations If we naively try to measure the distance between two raw time series, we may get very unintuitive results 4 most common distortions Offset Translation Euclidean distance is very sensitive to some distortions in the data. For most problems these distortions are not meaningful should remove them Amplitude Scaling Linear Trends Noise 37

38 Preprocessing the Data Offset Translation 38

machine learning algorithms 39 http://didawiki.cli.

39 Preprocessing the Data Amplitude Scaling Zero-mean Unit-variance Widely used for normalization in many machine learning algorithms 39

Preprocessing the Data Offset Translation Removing linear trend: Fit the best fitting straight line to the time series, then subtract that line from the

40 Preprocessing the Data Offset Translation Removing linear trend: Fit the best fitting straight line to the time series, then subtract that line from the time Remove linear trend Removed offset translation Removed amplitude scaling 40

points value with its neighbors http://didawiki.

41 Preprocessing the Data Noise The intuition behind removing noise is Average each data points value with its neighbors 41

42 Software Filter Low Pass Filter High Pass Filter 42

43 Discrete Fourier Transform Fourier showed that any periodic signal s(t) can be written as a sum of sine waves with various amplitudes, frequencies and phases For example, the Fourier expansion of a square wave can be written as 43

http://mriquestions.com/fourier-transform-ft.

44 Discrete Fourier Transform X k = N 1 n= 0 x n e i2 N k n Fourier series in

45 Discrete Fourier Transform Important signal processing tool Used to decompose a signal into its sine and cosine components Output of the transformation represents the signal in the Fourier or frequency domain Apply mathematical operations to eliminate certain frequency domains very easily Applying the inverse Fourier transform to recover the original time signal 45

Of course, sometimes the distortions are

46 Summary of Preprocessing The raw time series may have distortions which we should remove before clustering, classification etc. Of course, sometimes the distortions are the most interesting thing about the data, the above is only a general rule 46

What Do We Want to Do with the Time Series Data? http://didawiki.cli.di.unipi.

47 What Do We Want to Do with the Time Series Data?

48 What to take home? Patient Data Management Systems are currently expected to assist clinicians at every level of intensive care Currently, decision support via ICU scoring systems Data is generated by sensors Preprocessing of time series Chance to improve decision support with the help of machine learning 48

Biosignal Analysis Biosignal Processing Methods. Medical Informatics WS 2007/2008

Biosignal Analysis Biosignal Processing Methods Medical Informatics WS 2007/2008 JH van Bemmel, MA Musen: Handbook of medical informatics, Springer 1997 Biosignal Analysis 1 Introduction Fig. 8.1: The