Data Acquisition: Patient Monitoring
                              
            (Introduction to Medical Informatics)
         (http://www.cpmc.columbia.edu/edu/textbook)

LAST REVIEWED:  1 October 1997

monitoring - definition
     repeated or continuous observations (many measurements)
          of physiologic function to guide therapy (or make diagnoses)

used primarily in ICU
     unstable physiologic systems (respiration on overdose)
     suspected life-threatening condition (heart attack)
     high risk for instability (e.g., just underwent chest surgery)
     but also seen in outpatient, such as holter monitor
     
     data usage in office
          70% history
          20% physical
          10% laboratory work
     
     data usage in ICU
          42% laboratory and blood gas
          22% drug and fluid balance
          21% clinical observations (mainly physical)
          13% physiologic monitoring

history
     1625 - body temperature and pulse (Santario & Galileo)
     1707 - scientific report on pulse
     1852 - scientific report on fever
     1896 - sphygmanometer
     1900 - arterial pressure monitoring
     1903 - ECG
     1920 - temp, pulse, BP, respiratory rate in charts:  "vital signs"
     1940s - transducer developed
     1950s - first ICUs
     1963 - CCU results in 60% drop in mortality for acute MI
     1969 - arrhythmia monitor (Cox; Washington Univ)
     1960s-70s - proliferation of ICUs
     1980s - 75000 ICUs in USA
     1990s - focus back to primary care

continuous vs. intermittent monitoring
     = an important decision:  storage volume, network requirements and
          signal quality depend on this
     deciding questions:
          - How rapidly can the parameter change?
          - How long after a dangerous change will irreversible
               damage ensue?
     answer:  parameters that can change rapidly and cause damage
          in a short time should be measured continuously
     example:
          heart rhythm:  measure continuously
          temperature:  measure intermittently

closed vs. open-loop monitoring systems
     closed:  measurement is used by computer to affect 
        treatment directly
          - example:  nitroprusside pump (regulates blood pressure)
     open:  computer gives advice based on measurement; human
        alters treatment directly
          - example:  ventilator management expert system

sensor (transducer)
     convert biological signal to voltage
          (signal itself may be a voltage, e.g., ECG)
     eg,
          pressure - blood pressure
          temperature
          light - pulse oximeter
          chemical - pH
          electric - heart rhythm

analog signal
     continuous waveform
     subject to noise

analog to digital conversion
     
     resolution - bits of information
          2 bits - only two levels
          8 bits - usually minimum useful
          12 bits - CD player
          ...
     
     "Nyquist" sampling frequency - must sample at twice
          rate of highest frequency wanted in the data
     
     can hear up to 20,000 cps => CDs sample at 40,000
     
     ECG shows waves at 1-3 per second, but QRS has
          frequency components up to 150 cps => 300-500
          "signal-averaged" ECG can detect components
     
     poor resolution or low sampling are forms of noise

digital version
     once converted, less subject to noise
     easier to manipulate by computer

computer processing
     pattern recognition (eg, find MI)
     feature extraction (eg, PR interval)
     monitor signal quality
     auto-calibration
     filter noise
     storage
     aggregate over time
     intelligent alarms
          ex:  verify tachycardia by comparing ECG and
                arterial pulse waveform
     software upgrade without new equipment

eg, computer-based electrocardiogram (ECG) interpretation
     much of medical informatics started here
     5 companies interpret 90% of ECGs
     $200,000 purchase for large hospital

ECG:  "leads" pick up small currents on skin
     signal is digitized
     detection modules
          find QRS (ventricles pumping)
          find P waves (atria pumping)
     typification modules
          cluster QRS
          cluster ST-T (ventricle relaxing)
     dominant beat selection
          select complexes for analysis and average them
     waveform templates:  matching waveforms to detect abnormalities
     segmentation
          define beginning, components, and end for
               QRS, T, and P
          use slope of voltage between two points to detect upswing
     classification
          estimate parameters (intervals)
          choose rhythm
          contour classification (hypertrophy, MI, ...)
     
     compare old ECGs
     print unverified report
     (physician verifies)
     
     research
     noninvasive methods (pulse oximeter)