Study design
This study was a nationwide, retrospective, registry-based cohort study with a 30-day follow-up.
Setting
Denmark has approximately 5.9 million inhabitants. According to the latest DCAR report from 2021, the annual incidence of OHCA was 4,807, corresponding to 81 per 100,000 citizens,(19) overall 30-day survival rate was 13% (10 per 100,000 citizens), with a bystander cardiopulmonary resuscitation (CPR) rate of 79%.(20)
This study applied an advanced text-search algorithm (Danish Drowning Formula) within the DCAR from January 1st, 2016, through December 31st, 2021, to identify potential drowning patients.
The Danish Cardiac Arrest Registry
In 2016, the Danish Emergency Medical Services (EMS) introduced a nationwide electronic medical reporting system recording all consecutive cases of OHCA in which a resuscitative attempt was initiated.(21) With the application of a manual validation process, this electronic system is the cornerstone of the DCAR and constitutes a solid base for identifying and verifying OHCA.(22) Furthermore, the DCAR is linked with the text fields from the electronic prehospital medical records in the Danish quality database for prehospital emergency medical services.(23) This system enables advanced text searching within all unstructured text fields in the DCAR for specific cases of OHCA(24) based on prespecified identifying trigger words.
Identification of drowning-related OHCA
During the manual validation of OHCA in the changeover period in 2016, 211 cases were labelled as potential drowning accidents. These cases were manually reviewed and validated as drowning if the person most likely had suffered cardiac arrest following submersion in liquid. Three reviewers (NB, TWJ, and MGH) manually reviewed the prehospital medical records of validated cases and identified trigger-words related to submersion injury. NB and SNFB combined these trigger-words by “OR” and build a text-search algorithm, which was used to search the unstructured text fields within the electronic prehospital medical records in the DCAR. The algorithm returned new medical records from the initial text search, which were manually reviewed and validated by three reviewers (NB, SAW, and TWJ) to identify new trigger-words to be added to the algorithm. This iterative process was repeated until no more drowning cases appeared, thereby improving the sensitivity of the algorithm. Subsequently, NB and SNFB added specific trigger-words to the existing algorithm by “NOT” to improve the specificity without decreasing sensitivity. The final text-search algorithm was named the Danish Drowning Formula.
Participants
All cases from the final search were manually reviewed by three reviewers (NB, SAW, and TWJ) for validation. Any discrepancies were solved by a senior investigator (HCC). Patients were eligible for inclusion if they had suffered OHCA while submerged or immersed in liquid in which a resuscitative attempt was initiated. Cases with clinical signs of irreversible death were excluded (decomposition, post-mortem lividity, and post-mortem rigidity) (Fig. 1). All Danish citizens have a unique identification number which is a personal identifier that enables accurate linkage between all Danish national registries. Follow-up was completed for all patients with a valid identification number to report 30-day survival.
Variables
The primary outcome was 30-day survival. The primary outcome was analyzed as univariable and as multivariable adjusted for age, sex, bystander-witnessed event, and initial rhythm. Bystander CPR was removed from the adjusted analysis due to multicollinearity with bystander-witnessed event. Secondary outcomes were annual incidence rates of drowning-related OHCA per 100,000 citizens, survival status at hospital admission, and geographical localization of drowning-related OHCA in Denmark.
Data sources
SNFB extracted data from the DCAR for all Danish OHCA from 2016–2021, including age, sex, date of occurrence, observation of occurrence, CPR performed by bystanders and EMS personnel, bystander use of an automated external defibrillator (AED), GPS coordinates, EMS response time, initial rhythm analyzed by EMS, return of spontaneous circulation (ROSC) at any time prehospital, ROSC at hospital admission, 30-day, and 1-year survival. Data on liquid type, location type, activity type, and cause of drowning (intentional/unintentional) were manually extracted from the electronic prehospital medical records on validated drowning cases. The municipal population density was used to estimate population density at the OHCA site stratified as low, intermediate, and high density according to the EUROSTAT degree of urbanization system (DEGURBA).(25) The location type was categorized as following: 1) Bathtub, 2) Swimming pool, 3) Harbor, 4) Coastline, 5) Open ocean, 6) Lake, 7) Streams, 8) Other. The location type “Swimming pool” was used for all in- or outdoor structures designed to hold a body of standing water. The location type “Other” included drowning accidents in wells, manure stores, etc. Activity type described what the person was doing at the time of the incident and was categorized as following: 1) Swimming/bathing, 2) Boat activity, 3) Diving, 4) Recreational fishing, 5) Traffic, 6) Harbor activity, 7) Water sports. The activity type “Swimming/bathing” included all forms of playing, bathing, and swimming. The activity type “Boat activity” included activities on a boat in open water, such as oceans and lakes, including fishing from a boat. However, if the boat was in port, the activity was defined as “Harbor activity”. The activity type “Diving” was used for freediving and scuba diving. The activity type “Traffic” was used for accidents involving driving into the water using a bike or motorized vehicle. The activity type "Water sports" was used for accidents involving swim practice, kayaking, water polo, and surfing.
Statistical methods
Baseline characteristics are presented separately for drowning-related OHCA and other causes of OHCA and fatal and non-fatal drowning accidents. Categorical variables are presented as frequencies (counts and percentages) and compared using the chi-squared test. Numerical variables are presented as medians with interquartile ranges (IQR) due to non-normality (inspected using histograms and Q-Q plots) and compared using the Mann Whitney U test.
The primary outcome was analyzed using a multiple logistic regression model and presented as odds ratios (ORs) with 95% confidence intervals (CIs) adjusted for age, sex, bystander-witnessed event, and initial rhythm.
Annual incidence rates per 100,000 citizens were calculated and presented with 95% CIs. P-values < 0.05 were considered statistically significant. There was no imputation of missing data. All drowning-related OHCA were presented on a map of Denmark. All analyses were performed using R statistical software (version 4.2.2).(26)