Using Machine Learning to Optimise the Danish Drowning Formula (DROWN_DDF2)

Machine Learning-assisted Drowning Identification for the Danish Prehospital Drowning Data: Using Machine Learning to Optimise the Danish Drowning Formula

The Danish Drowning Formula (DDF) was designed to search the unstructured text fields in the Danish nationwide Prehospital Electronic Medical Record on unrestricted terms with comprehensive search criteria to identify all potential water-related incidents and achieve a high sensitivity. This was important as drowning is a rare occurrence, but it resulted in a low Positive Predictive Value for detecting drowning incidents specifically. This study aims to augment the positive predictive value of the DDF and reduce the temporal demands associated with manual validation.

Study Overview

• Status: Active, not recruiting
• Conditions: Drowning; Drowning, Near; Drowning; Asphyxia; Drowning and Nonfatal Submersion; Drowning and Submersion While in Bath-Tub; Drowning and Submersion While in Natural Water; Drowning and Submersion Due to Fall Off Ship; Drowning and Submersion While in Swimming-Pool
• Intervention / Treatment: Other: Drowning incident

Detailed Description

The DDF was published in 2023. It is a text-search algorithm designed to search the unstructured text fields in databases containing electronic medical records to identify all potential water-related incidents. The DDF consists of numerous trigger words related to submersion injury (e.g., "drukn"/ drown, "vand"/water, "hav"/ocean, and "båd"/ boat).

An ongoing study showed impressive performance metrics of the DDF as a drowning identification tool when applied to the Danish PEMR on unrestricted terms. However, the PPV was low for detecting drowning incidents specifically. This study aims to augment the DDF's positive predictive value and reduce the temporal demands associated with manual validation.

Data are extracted from the Danish nationwide Prehospital Electronic Medical Record using the DDF and manually validated before entered into the Danish Prehospital Drowning Data (DPDD).

Data from the DPDD from 2016-2021 will be split into 80% (training data) and 20% (test data) and used to train the machine learning.

Data from the DPDD from 2022-2023 will be used as validation data to calculate the performance metrics for the machine learning.

• Study Type: Observational
• Enrollment (Estimated): 1500

Contacts and Locations

Study Locations

• Denmark
  • Region Zealand
    • Næstved, Region Zealand, Denmark, 4700
      • Prehospital Center

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

All fatal and non-fatal drowning patients in Denmark treated by the Emergency Medical Services (EMS) between 2016 and 2023.

Description

Inclusion Criteria:

• The patient must have been experiencing respiratory impairment from submersion or immersion in liquid (including persistent coughing, respiratory arrest, and unconsciousness).
• The patient must have been in contact with the Danish prehospital Emergency Medical Services.

Exclusion Criteria:

• Duplets
• Invalid civil registration number

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Fatal drowning; Drowning incidents where the patient died within 30 days after the incident as a consequence of the submersion injury	Other: Drowning incident; Drowning was defined by the WHO in 2002 as "the process of experiencing respiratory impairment from submersion or immersion in liquid".
Non-fatal drowning; Drowning incidents where the patient survived to 30 days	Other: Drowning incident; Drowning was defined by the WHO in 2002 as "the process of experiencing respiratory impairment from submersion or immersion in liquid".

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity of the machine learning algorithm as a drowning identification tool	Sensitivity [TP / (TP+FN)] will be calculated to show the performance of the machine learning as a drowning identification tool.	The sensitivity of the trained machine learning will be calculated based on data from 2022 and 2023.
Specificity of the machine learning algorithm as a drowning identification tool	Specificity [TN / (FP+TN)] will be calculated to show the performance of the machine learning as a drowning identification tool.	The specificity of the trained machine learning will be calculated based on data from 2022 and 2023.
PPV of the machine learning algorithm	PPV [TP / (TP+FP)] will be calculated to show the machine learning test result.	The PPV of the trained machine learning will be calculated based on data from 2022 and 2023.
NPV of the machine learning algorithm	NPV [TN / (FN+TN)] will be calculated to show the machine learning test result.	The NPV of the trained machine learning will be calculated based on data from 2022 and 2023.

Collaborators and Investigators

• Sponsor: Prehospital Center, Region Zealand
• Investigators: Study Director: Helle Collatz Christensen, Ass. Prof., Prehospital Center, Region Zealand

Publications and helpful links

General Publications

• Breindahl N, Wolthers SA, Jensen TW, Holgersen MG, Blomberg SNF, Steinmetz J, Christensen HC; Danish Cardiac Arrest Group. Danish Drowning Formula for identification of out-of-hospital cardiac arrest from drowning. Am J Emerg Med. 2023 Nov;73:55-62. doi: 10.1016/j.ajem.2023.08.024. Epub 2023 Aug 15.
• Breindahl N, Wolthers SA, Moller TP, Blomberg SNF, Steinmetz J, Christensen HC; Danish Drowning Validation Group. Characteristics and critical care interventions in drowning patients treated by the Danish Air Ambulance from 2016 to 2021: a nationwide registry-based study with 30-day follow-up. Scand J Trauma Resusc Emerg Med. 2024 Mar 6;32(1):17. doi: 10.1186/s13049-024-01189-y.

Study record dates

Study Major Dates

• Study Start (Actual): January 1, 2024
• Primary Completion (Estimated): December 31, 2024
• Study Completion (Estimated): December 31, 2024

Study Registration Dates

• First Submitted: March 7, 2024
• First Submitted That Met QC Criteria: March 13, 2024
• First Posted (Actual): March 15, 2024

Study Record Updates

• Last Update Posted (Actual): March 25, 2024
• Last Update Submitted That Met QC Criteria: March 21, 2024
• Last Verified: March 1, 2024

More Information

Terms related to this study

• Keywords: Epidemiology; Database; Machine learning; Denmark; Prehospital; Utstein; Register; Drowning; Danish Drowning Formula; Text search
• Additional Relevant MeSH Terms: Pathologic Processes; Wounds and Injuries; Death; Asphyxia; Drowning; Near Drowning
• Other Study ID Numbers: DROWN_DDF2

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: The data are intended for use nationally and internationally by researchers to reduce the incidence, mortality, and morbidity of drowning. The data are available upon reasonable request from researchers after application to the corresponding author, provided the necessary approvals are obtained from the relevant authorities.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No