ambulance_acceleration

This repository contains the Python code used for data processing, statistical analysis and visualization described in the following paper:

Belteki G, Morley CJ. Acceleration during neonatal transport and its impact on mechanical ventilation Arch Dis Child Fetal Neonatal Ed. 2022 June 15

Link to the paper: https://fn.bmj.com/content/early/2022/06/14/archdischild-2021-323498

Contact: [email protected]; [email protected]

---

The outputs (numbers, tables, graphs) of the cells of the Jupyter Notebooks (.ipynb files) have been suppressed in order to comply with copyrights. Some of the corresponding data and graphs can be found in the paper and its only supplementary material.

This code can be viewed in any web browser. If the code is displayed (rendered) in Github, copy and paste the URL (web adress) of the Notebook into nbviewer (https://nbviewer.jupyter.org) for a read-only display.

To run the code, you need to use Jupyter. The raw ventilator data are not shared but the user can run this code on his or her own data obtained in the same format.

---

Packages required to run this Notebook:

Python version: 3.7.4

IPython version: 7.10.2

pandas version: 0.25.3

matplotlib version: 3.1.1

Seaborn version: 0.9.0

NumPy version: 1.17.4

SciPy version: 1.3.1

I recommend downloading these packages using the freely availably Anaconda built: https://www.continuum.io/downloads

---

Jupyter Notebooks

1. Initial processing of raw ventilator slow (0.5 Hz) data

ventilator_data_1_150: This notebook imports all ventilator data of these recordings (including ventilator parameters, settings, alarms (0.5Hz sampling rate) and waveform data (150Hz sampling rate).

• Total: 150 cases
• 21 cases were removed as they were less than 15 minutes long (empty or partial recordings)
• 129 cases remaining

Dictionary containing the processed ventilation data exported as pickle archive: data_pars_1_150.pickle

ventilator_data_151_300: This notebook imports all ventilator data of these recordings (including ventilator parameters, settings, alarms (0.5Hz sampling rate) and waveform data (150Hz sampling rate).

• Total: 150 cases
• 1 case (AL000257) was removed it was fragmented with multiple in data
• 32 cases were removed as they were less than 15 minutes long (empty or partial recordings)
• 117 cases remaining

Dictionary containing the processed ventilation data exported as pickle archive: data_pars_151_300.pickle

ventilator_data_301_450: This notebook imports all ventilator data of these recordings (including ventilator parameters, settings, alarms (0.5Hz sampling rate) and waveform data (150Hz sampling rate).

• Total: 150 cases
• 19 cases were removed as they were less than 15 minutes long (empty or partial recordings)
• 131 cases remaining

Dictionary containing the processed ventilation data exported as pickle archive: data_pars_300_450.pickle

ventilator_data_451_600: This notebook imports all ventilator data of these recordings (including ventilator parameters, settings, alarms (0.5Hz sampling rate) and waveform data (150Hz sampling rate).

• Total: 150 cases
• 26 cases were removed as they were less than 15 minutes long (empty or partial recordings)
• 124 cases remaining

Dictionary containing the processed ventilation data exported as pickle archive: data_pars_451_600.pickle

ventilator_data_601_674: This notebook imports all ventilator data of these recordings (including ventilator parameters, settings, alarms (0.5Hz sampling rate) and waveform data (150Hz sampling rate).

• Total: 74 cases
• 7 cases were removed as they were less than 15 minutes long (empty or partial recordings)
• 67 cases remaining

Dictionary containing the processed ventilation data exported as pickle archive: data_pars_601_674.pickle

From AL000001-AL000674: Appropriate ventilator data are available for a total of: 129 + 117 + 131 + 124 + 67 = 568 cases

---

2. Initial processing of fast (125 Hz) ventilator data

1. fabian_fast_data_conversion: This file extracts fast (125 Hz) pressure, flow and volume data from the raw hexdecimal data and converts them into pandas DataFrames. Data are exported as csv files. Currently it does not work with every ventilator data file. List of files with errors is written in a log file.

---

3. Processing of clinical data

clinical_data_1_665

• Total: 665 cases
• Clinical data available in 539 cases
• Only keep clinical data for cases where ventilation recordings (>15 minutes) are also available: 511 cases

DataFrame containing the processed clinical data exported as clin_df_1_665.pickle

This Notebook also generates and exports disease-specific Excel sheets: clinical_data_diseases_1_665.xlsx It also exports pickle archives: clin_df_RDS, clin_df_MAS, clin_df_PPHN, clin_df_cardiac (except PFO and PDA), clin_df_HIE, clin_df_CDH, clin_df_NEC, clin_df_surgical (except NEC and CDH)

---

4. Further processing of slow (0.5Hz) ventilator data

ventilator_data_processing_1_300

Imported: data_pars_1_150.pickle, data_pars_151_300.pickle, clin_df_1_300.pickle

• Total: 246 cases with ventilator data available (with >15 minutes recording time)
• Clinical data were not available for 4 cases
• Appropriate ventilator and clinical data are available for 242 cases

Exported: dictionaries containing ventilation data as data_pars_measurements_1_300.pickle, data_pars_settings_1_300.pickle, data_pars_alarms_1_300.pickle

ventilator_data_processing_301_600

Imported: data_pars_301_450.pickle, data_pars_451_600.pickle, clin_df_301_600.pickle

• Total: 255 cases with ventilator data available (with >15 minutes recording time)
• Clinical data were not available for 38 cases
• Appropriate ventilator and clinical data are available for 217 cases

Exported: dictionaries containing ventilation data as data_pars_measurements_301_600.pickle, data_pars_settings_301_600.pickle, data_pars_alarms_301_600.pickle

ventilator_data_processing_601_665

Imported: data_pars_601_674.pickle, clin_df_1_665.pickle

• Total: 67 cases with ventilator data available (with >15 minutes recording time)
• Clinical data were not available for 15 cases
• Appropriate ventilator and clinical data are available for 52 cases

Exported: dictionaries containing ventilation data as data_pars_measurements_601_665.pickle, data_pars_settings_601_665.pickle, data_pars_alarms_601_665.pickle

---

5. Further processing of recordings with mechanical ventilation

ventilator_data_processing_1_300_ventilated

This notebook ventilator data to keep only periods of mechanical ventilation. Recordings have also been inspected and trimmed manually to remove periods when the ventilator was working but the patient was not connected.

Imported: data_pars_measurements_1_300.pickle, data_pars_settings_1_300.pickle, data_pars_alarms_1_300.pickle, clin_df_1_300.pickle

• Total: 242 cases
• Containing >15 minutes of mechanical ventilation: 146 cases remaining
• After removing periods when the patient was not connected, containing >15 minutes of mechanical ventilation: 145 cases

Exported: data_pars_measurements_ventilated_1_300.pickle, data_pars_settings_ventilated_1_300.pickle, data_pars_alarms_ventilated_1_300.pickle, vent_modes_1_300.pickle, vent_modes_ventilated_1_300.pickle, ventilation_modes_1_300.xlxs, ventilation_modes_ventilated_1_300.xlsx

ventilator_data_processing_301_600_ventilated

This notebook ventilator data to keep only periods of mechanical ventilation. Recordings have also been inspected and trimmed manually to remove periods when the ventilator was working but the patient was not connected.

Imported: data_pars_measurements_301_600.pickle, data_pars_settings_301_600.pickle, data_pars_alarms_301_600.pickle, clin_df_301_600_pickle

• Total: 217 cases
• Containing >15 minutes of mechanical ventilation: 114 cases
• 8 cases removed as there was no flow sensor used (and hence no tidal volume measurement), 106 cases remaining
• After removing periods when the patient was not connected, containing >15 minutes of mechanical ventilation: 106 cases

Exported: data_pars_measurements_ventilated_301_600.pickle, data_pars_settings_ventilated_301_600.pickle, data_pars_alarms_ventilated_301_600.pickle, vent_modes_301_600.pickle, vent_modes_ventilated_301_600.pickle, ventilation_modes_301_600.xlxs, ventilation_modes_ventilated_301_600.xlsx

ventilator_data_processing_601_665_ventilated

This notebook ventilator data to keep only periods of mechanical ventilation. Recordings have also been inspected and trimmed manually to remove periods when the ventilator was working but the patient was not connected.

Imported: data_pars_measurements_601_665.pickle, data_pars_settings_601_665.pickle, data_pars_alarms_601_665.pickle, clin_df_1_665_pickle

• Total: 52 cases
• Containing >15 minutes of mechanical ventilation: 22 cases
• 1 case removed as there was no flow sensor used (and hence no tidal volume measurement), 21 cases remaining
• After removing periods when the patient was not connected, containing >15 minutes of mechanical ventilation: 21 cases

Exported: data_pars_measurements_ventilated_601_665.pickle, data_pars_settings_ventilated_601_665.pickle, data_pars_alarms_ventilated_601_665.pickle, vent_modes_601_665.pickle, vent_modes_ventilated_601_665.pickle, ventilation_modes_601_665.xlxs, ventilation_modes_ventilated_601_665.xlsx

---

6. Analysis of accelerometer data and associated ventilator data during invasive mechanical ventilation

accelerometer_ventilated_processing_1

Identification and processing of ventilated cases with matched accelerometer recordings.

• This notebook first reads the start time of all accelerometer recordings and calculates the stop time of accelerometer data. It corrects the timestamps of the recordings if the correct timestamps can be found out, otherwise it removes the recordings

• It imports ventilator data of ventilated recordings from pickles archives:

data_pars_measurements_ventilated_1_300.pickle, data_pars_settings_ventilated_1_300.pickle, data_pars_alarms_ventilated_1_300.pickle, data_pars_measurements_ventilated_301_600.pickle, data_pars_settings_ventilated_301_600.pickle, data_pars_alarms_ventilated_301_600.pickle, data_pars_measurements_ventilated_601_665.pickle, data_pars_settings_ventilated_601_665.pickle, data_pars_alarms_ventilated_601_665.pickle

• It identifies the start and stop times of ventilator recoridings
• It identifies overlapping accelerometers and ventilator recordings
• It imports relevant accelerometer recordings from the original files
• It trims ventilator and accelerometer data to contain only the overlapping regions: 153 recordings
• It exports trimmed accelerometer data as accelero_ventilated_1.pickle
• It exports trimmed ventilator data for ventilated_ recordings: data_pars_measurements_ventilated_accelero.pickle, data_pars_settings_ventilated_accelero.pickle, data_pars_settings_ventilated_accelero.pickle

accelerometer_ventilated_processing_2

Further processing of the matched ventilated and accelerometer recordings.

• It imports the following pickle archives data_pars_measurements_ventilated_accelero, data_pars_settings_ventilated_accelero, data_pars_alarms_ventilated_accelero, accelero_ventilated_1

• It starts from 153 recordings when both ventilator and accelerometer data were available

• Removed recordings when the net ambulance movement time was <10 minutes - 113 remaining

• The gravity (9.81 m/sec2) is subtracted from the vertical (Z) acceleration

• Absolute values of X (front-back), Y (side-to-side) and Z (up-down) accelerations are calculated

• The length (Euclidean (L2) norm) of the acceleration vector is calculated

• Accelerometer data are resampled with 0.5 Hz sampling rate using median and interquartile range (IQR) as aggregating function

• Ventilator data (0.5 Hz) are combined with accelerometer data (0.5 Hz, aggregating function: median and IQR)

• Ventilator data are aggregated over 1-minute periods using mean and standard deviation (SD)

• Ventilator data (1/min, aggregating function: mean and SD) are combined with accelerometer data (1/min, aggregating function: median and IQR)

Exported: combined_ventilated_2sec.pickle (0.5 Hz aggregated accelerometer and unaggregated ventilator data), combined_ventilated_1min.pickle (1/min aggregated accelerometer and ventilator data) accelero_ventilated_2.pickle (unaggregated accelerometer data with absolute values and total vector length) ventilator_mode.pickle (ventilator modes used)

accelerometer_ventilated_analysis_1

Generating time series graphs for acceleration and selected ventilator data

• It imports the following pickle archives: accelero_ventilated_2.pickle,combined_ventilated_2sec.pickle, combined_ventilated_1min.pickle, ventilator_mode.pickle
• It generates and exports time series plots for the 113 matched recordings:
  1. 3-dimensional accelerations
  2. Length (Euclidean norm) of the acceleration vector)
  3. Unaggregated acceleration data and unaggregated ventilator data (VTemand, (RR), MV, PIP FiO2) (Graphs with RR can only be exported for some recording as during SIMV RR is not recorded by AcuLinkLog)
  4. Aggregated (2-sec median and IQR) acceleration data and unaggregated (0.5 Hz) ventilator data (VTemand, MV, PIP FiO2)
  5. Aggregated (1-min median and IQR) acceleration data and aggregated (1-min mean) ventilator data (VTemand, MV, PIP FiO2)
  6. Generic function to visualize the Euclidean norm of acceleration together with a chosen ventilator parameter

accelerometer_ventilated_analysis_2

Analysis of ventilator parameter data together with accelerometer data in ventilated cases with matched accelerometer recordings.

Imported: accelero_ventilated_2.pickle, combined_ventilated_2sec.pickle, combined_ventilated_1min.pickle, ventilator_mode.pickle, clin_df_1_665

• It starts with 113 recordings

• Removes cases that were > 46 weeks (term plus 6 weeks) of corrected gestations at the time of the transfer (4 recordings removed)

• 109 recordings remaining final dataset for analysis

• Calculates and exports descriptive statistics and plots on clinical data

• Calculates and exports duration of the final dataset

• Analyses ventilator modes in the final dataset

• Generates and exports histograms of accelerations using the original accelerometer data (~100 Hz sampling)

  1. X (front-back), Y (left-right) and Z (up-down) acceleration vectors individually or together (with linear or logarithmic Y-scale)
  2. Absolute values of X (front-back), Y (left-right) and Z (up-down) acceleration vectors individually or together (linear or logarithmic Y-scale)
  3. Length (Euclidean (L2) norm) of the 3-dimensional acceleration vectors (linear or logarithmic Y-scale)

• Calculates the median absolute values of X, Y, Z acceleration in each recording and studies the distribution of the medians and compares them.

• Analyse and visualise the correlation between 1-minute IQRs of acceleration in X (front-back) or Y (side-to-side) direction with IQR in Z (vertical) direction

• Analyse and visualise the correlation between the 1-minute median and IQR of the various components (X, Y, Z) of the acceleration vector

• Identify periods of vibration and sustained acceleration

• Compare periods of vibration with periods of no acceleration

• Compare periods of low and high sustained acceleration

Exported as pickle archives: combined_ventilated_1min, combined_ventilated_1min_all, low_1_vibr_frame, high_1_vibr_frame, low_1_accel_frame, high_1_accel_frame

accelerometer_ventilated_fast_data_analysis

Analysis of ventilator waveform data and loops during periods of high and low vibration

Imported from pickle archives: combined_ventilated_1min, combined_ventilated_1min_all, low_1_vibr_frame, high_1_vibr_frame, low_1_accel_frame, high_1_accel_frame

Also imported all ventilator fasta (waveform) data for the relevant recordings

• Generate composite P-V curves for the minute with the lowest and the highest vibration for each recording and export them as images
• Study the disorder of loops (expressed as the number of P-V data pairs during the given minute)