The AI and Automation in Anaesthesia (TRIPLE‐A) Perioperative Database in Eastern Denmark 2017–: A Platform for Epidemiology, AI‐Driven Prediction, Quality Control and Automated Data Retrieval

ABSTRACT

Introduction

Electronic health records can be used to create high‐quality databases if data are structured and well‐registered, which is the case for most perioperative data in the Capital and Zealand Regions of Denmark. We present the purpose and development of the AI and Automation in Anaesthesia (TRIPLE‐A) database—a platform designed for epidemiology, prediction, quality control, and automated research data collection.

Methods

Data collection from the electronic medical record (EPIC Systems Corporation, WI, USA) was approved by the Capital Region, Denmark, and ethical approval was waived. The first TRIPLE‐A database version included surgical procedures performed in a public hospital in the Capital and Zealand Regions of Denmark from 2017 to 2024 without age restrictions. We received structured, timestamped raw data including event log, medicine administrations, blood samples, vital parameters, etc. In collaboration with surgical specialists, we classified surgical procedures based on procedure codes and names. Variables were coded and enriched using multiple data sources (e.g., acute kidney injury: increase in creatinine OR new dialysis or ICD‐10 codes) to improve quality and mirror the clinical reasoning of physicians. Variable coding, validation and improvement continuous in iterative processes with each project in the database. The database is scheduled for updates every 6 months and as needed, currently approved to December 2030 with plans for future extension. The database is used for multiple purposes in research with broad collaboration.

Results

The first version of the TRIPLE‐A database consists of 1,142,108 surgeries in 675,790 patients. Data were processed into 17 summary tables (anaesthesia, airway management, complications, etc.) including a total of 1617 variables. The database currently provides data for 34 research projects.

Conclusion

TRIPLE‐A is a continuously updated, granular, high‐quality perioperative database. It will be used for (1) high‐quality epidemiological research, (2) automated data retrieval into case report forms in cohort studies and randomized trials, (3) monitoring and evaluation of clinical implementation of interventions, and (4) development of perioperative prediction models for risk stratification and individualized treatment.

1. Introduction

Surgery accounts for a substantial portion of healthcare expenditures and is a leading contributor to serious adverse events, loss of work capacity, and personal adverse events such as chronic opioid use [1, 2]. Anaesthesiologists play a pivotal role in optimising perioperative care, and anaesthesia‐initiated new interventions have consistently demonstrated significant reductions in mortality and complications, while improving patient‐centred outcomes [3, 4, 5].

Large healthcare databases can be used to generate such novel hypotheses for improving care, but also to monitor clinical effects of known interventions and develop personalised treatment algorithms [6, 7]. As information technology has advanced and electronic medical record (EMR) systems have become widespread, it has become an economic, political, and ethical priority to make better use of healthcare data to improve research and clinical quality [8, 9]. With the use of structured data entry and checklists, surgical and perioperative data are often consistently and accurately recorded, making them well‐suited as input for high‐quality clinical databases [10]. To make data output clinically relevant, data and their processing must reflect clinical practise [11]. Hereby, clinically relevant “data warehouses” can be built to support multiple research objectives and eliminate cost‐ and time‐consuming processes related to data processing [12, 13].

The AI and Automation in Anaesthesia (TRIPLE‐A) database encompasses all surgical procedures from public hospitals in Eastern Denmark from 2017. The TRIPLE‐A database is structured as an organized database for (1) epidemiological studies that reflect everyday clinical practice, (2) automatic data retrieval and cleaning into automated case report forms (aCRFs) for prospective studies and randomized trials, (3) quality assurance studies for monitoring implementation of new interventions in clinical care, and (4) development and implementation of AI prediction models (Figure 1).

FIGURE 1.

TRIPLE‐A work fields.

The TRIPLE‐A design is intended to reduce or overcome frequent problems of (1) “garbage in, garbage out” in retrospective databases [12, 14], (2) costs and time consumption from manual entry in electronic case report forms for cohort studies and clinical trials [15], (3) lacking assessment of actual benefits or harms when interventions are broadly implemented in clinic care based on research findings from selected research populations [16], and (4) low clinical usability and generalisability of AI prediction models [17, 18, 19].

This manuscript describes the development, aims and perspectives of the TRIPLE‐A database.

2. Methods

The TRIPLE‐A database consists of all surgical procedures performed in children and adults at the public hospitals of Eastern Denmark—that is the Capital and Zealand Regions. It covers a period from 2017 when the EMR system “Sundhedsplatformen” (Epic Systems Corporation, WI, USA) was rolled out. The two regions hold a total population of ~2.76 million people (0–17 years: 19.2%, 18–39 years: 29.1%, 40–59 years: 25.9%, 60–79 years: 20.2%, 80–99 years: 4.8%, 100–108 years: 0.16%. 51% are female) [20]. Danish inhabitants have free access to healthcare and Denmark uses 11% of GDP on health expenditures ($6280 per capita compared to an OECD average of $4986) [21, 22]. All Danish citizens are assigned a unique 10‐digit personal civil registration number (CPR‐number) at birth, which is used in all societal contexts, including healthcare. Healthcare data can thereby be accurately linked across platforms and registers. This makes the Danish population optimal for epidemiological studies [12, 23].

2.1. Ethics and Approvals

TRIPLE‐A was deemed exempt for ethical approval by the Regional Ethics Committee (F‐25031347, 14 April 2025). Data extraction and the project was approved for both the Capital and Zealand regions by the Centre for Health, Capital Region, Denmark, Team for Medical Records Research (R‐25029803, 16 May 2025). Finally, the project was approved by the Capital Region Legal Department (p‐2025‐19193, 23 May 2025). In order to provide data for specific collaborating projects, individual approvals must be applied for. The above‐listed approvals cover the creation of the TRIPLE‐A database and consecutive updates currently up until 31 December 2030. We plan to apply for extensions as long as the database continues to support active research.

2.2. Collaboration

TRIPLE‐A is investigator‐initiated and collaborates with all hospitals in the Eastern Region of Denmark, primarily via the Collaboration for Evidence‐based Practice and Research in Anaesthesia (CEPRA) network [24]. Further, data scientists from the IT University of Copenhagen develop prediction models and user‐friendly interfaces for the implementation of these models in EMR systems (Figure 2).

FIGURE 2.

TRIPLE‐A collaboration.

We are continuously collaborating on projects to ensure that data meet each project's specific needs, to continuously validate and improve data quality, and to gain expert input for optimizing different subfields of the database and generate new variables. To optimize the use of the database, TRIPLE‐A invites other relevant parties—both anaesthesiologists, surgeons, and data scientists—to use our database rather than building parallel datasets from the same sources (a process that is both demanding and costly). External collaborators can contact us regarding project ideas at tripleateamdk@gmail.com or via the homepage www.triplea.dk. Ideas should be structured using the PICO format and guided by the downloadable TRIPLE‐A handbook, which is available from the website and contains a standard template for protocol and variable list [25].

2.3. Data Extraction and Database

Data were extracted by SP Sundhedsdata (“SP Health Data”), The Data Unit, Capital Region of Denmark from Sundhedsplatformen (Epic Systems Corporation, WI, USA) via SQL queries and exported as separate comma‐separated value (csv) files, each keyed by patient or surgery identifiers and, where relevant, data‐specific timestamps. To assemble these files—demographics, vital signs, administration of medication, etc.—we partnered with SP Sundhedsdata and began by mapping each required variable: navigating the Sundhedsplatformen interface to locate the correct backend table and column. We defined our study cohort once in a temporary SQL table, which then served as the foundation for all subsequent extracts. Extraction proceeded through iterative cycles of variable specification, query development, clinical expert review, validation against source records, and query refinement—repeated until accurately reflecting the intended measure.

The extracted data were sorted and then imported into the database into clinically relevant tables. The imported data were imported in long format, that is multiple measurements per patient—as extracted from the EMR. The database is automatically updated 30 days after 1 January and 1 July to include patients up to these dates with 30 days of follow‐up.

2.4. Summary Variables

First, we used the admission time log to reconstruct the entire surgical care pathway—from scheduling and preoperative preparations through anaesthesia and the operating room to recovery in the post‐anaesthesia care unit (PACU), the surgical ward, and subsequent post‐admission phases. This “sacred timeline” formed the basis of all subsequent variable coding. Based on the timeline, we created summary tables for demographic, anaesthesia, surgery, medications, fluids, invasive procedures, symptoms, vital parameters, ventilator parameters, blood samples, and complications (Table 1). These tables hold variables where each surgery has one value. For instance, all oxygen saturation measurements from surgery start to surgery end have been averaged into a ‘mean intraoperative saturation’.

TABLE 1.

Demographic characteristics.

Median (IQR) or frequency (%)	Overall n = 871,598
Age	55.0 (33.0; 71.0)
Female	494,234 (56.7%)
BMI	25.4 (22.1; 29.4)
ASA‐score
1	227,827 (26.1%)
2	396,985 (45.5%)
3	192,968 (22.1%)
4	18,218 (2.1%)
Smoking
Current smoker	132,729 (15.2%)
Former smoker	222,248 (25.5%)
Never smoked	274,230 (31.5%)
Missing information	242,391 (27.8%)
Charlson comorbidity index	2.2 (2.2; 2.2)
Type of anaesthesia
General anaesthesia	691,836 (79.4%)
Spinal	57,776 (6.6%)
Epidural	15,386 (1.8%)
Sedation	55,328 (6.3%)
Regional	14,188 (1.6%)
Missing data	32,943 (3.8%)
Combinations	4141 (0.5%)
Airways
Tube	399,748 (45.9%)
Laryngeal mask	268,395 (30.8%)
Spontaneous respiration	92,566 (10.6%)
Missing information	99,838 (11.5%)
Tracheostomy	693 (0.1%)
Missing	99,838 (11.5%)
Acute surgery	280,315 (32.2%)
Intraoperative IME (intravenous morphine equivalents)	10.0 (0.0; 20.0)
0–24 h postoperative IME	0.0 (0.0; 7.5)
Length of surgery (hours)	0.8 (0.4; 1.3)
Length of admission (days)	1.2 (0.3; 3.1)
90‐days all‐cause mortality	22,719 (2.6%)

Variables were designed to mirror the clinical reality and mimic a clinician's logic rather than storing raw data points (Figure 4). Instead of tracking whether, for example, an analgesic was given at any point during the hospital stay, the database summarizes whether typical standard multimodal analgesia was in place upon awakening from anaesthesia—as anaesthesiologists would assess it in practice. Likewise, intraoperative variables were consolidated to reflect clinically meaningful metrics, such as total fluid balance or the cumulative duration a patient spent below a given blood pressure threshold. The database integrates data across multiple sources to build robust, clinically relevant variables; for instance, acute kidney injury is identified through a combination of creatinine changes, dialysis procedure codes, ICD‐10 diagnoses, and even free‐text mining of notes, while thromboembolic events are captured from therapeutic‐dose low‐molecular‐weight heparin administrations, free‐text mining, and ICD‐10 codes. The specific definition and data sources for each variable are available at www.triplea.dk.

FIGURE 4.

Database input and output. AKI, Acute kidney injury; BIS, Bispectral index; CCI, Charlson Comorbidity Index; ICU, Intensive care unit referral; LIA, Local infiltration analgesia; MEQ morphine equivalent consumption; NMBA, Neuromuscular blocking agents; ORADEs, opioid‐related adverse events; TOF, Train‐of‐four measures.

Each summary variable was manually validated and corrected. Small extracts were systematically assessed directly in the system by clinicians to ensure objective and reliable summary‐variables. Summary variables were then classified as either ‘validated for epidemiology’ or ‘validated for automated case report form’. Variables validated for epidemiology may tolerate minor deviations (e.g., small differences in mean intraoperative saturation), whereas those validated for aCRFs must match the original recorded values exactly, as they mirror data used in formal clinical documentation. The validation process is ongoing and aims to continuously improve the database. For each new project, external collaborators receive a pilot dataset of 20–50 surgical cases, which they review in Sundhedsplatformen by checking every variable requiring further validation. All discrepancies between the dataset and the EMR, along with suggestions for variable optimisation, are documented and returned to the responsible TRIPLE‐A data manager. The data manager then troubleshoots the code until the dataset accurately reflects the EMR and incorporates the suggested optimisations. This process is repeated in several iterations until the dataset adequately mirrors the clinical EMR data. The status of the validation process of each individual variable can be found on www.triplea.dk.

2.5. Surgical Characteristics

Surgeons book procedures from a list of unique procedure names (6209 different names in our database) and after surgery they register procedure codes according to the Danish Medical Classification System (SKS) [26] (1–10 codes per procedure; 91,165 unique combinations in total). In some studies, an overall classification such as ‘major abdominal surgery’ may be prudent, while others may require specific details such as ‘cemented total hip arthroplasty’. By using the combination of booking names and procedure codes, we therefore created a branching tree structure with specialty in top of the hierarchy (e.g., orthopaedic surgery), followed by overall procedures (e.g., total hip arthroplasty) and specific procedures (e.g., cemented total hip arthroplasty). This process was carried out in collaboration with surgeons from each surgical specialty. Cases with more than five SKS‐codes occurring less than 10 times were considered atypical and were initially excluded (this may be reevaluated in later database versions). Further, cases that did not fit into one surgical procedure were excluded in our classification system (e.g., femur amputation and cholecystectomy). First, we manually classified the 4184 most frequent combinations of booking and SKS‐codes. Then we developed and used a K‐nearest Neighboor (KNN) machine‐learning model to predict the remaining classifications and used this as a support‐tool to assist manual classification of the next 4830 combinations This was primarily carried out as a collaborative exercise to establish a shared understanding of the dataset between physicians and data scientists involved in the project. Due to low accuracy of this KNN model, we instead coded a rule‐based script to support the manual classification of the remaining combinations. Each TRIPLE‐A project can freely use the TRIPLE‐A classification system to define its study subpopulation or use alternative approaches, such as selecting the population based on specific SKS codes, diagnoses, study participation or hospitals.

Beyond the KNN model, we intend to use large language models for text mining of medical notes. In this database manuscript, we used ChatGPT for minor language improvement [27].

2.5.1. Dissemination

A list of ongoing studies and published articles using TRIPLE‐A data will be available from the webpage. Further, video format short presentations of results and implications will be uploaded to our youtube canal www.youtube.com/@TRIPLEADENMARK.

3. Results

3.1. TRIPLE‐A Database Version 1.0 (1 January 2017 Until 31 December 2024)

3.1.1. Population

The database contains 1,142,108 surgical procedures. The procedures were performed in 675,790 patients. The surgical procedures were primarily carried out in adults (90.8%; n = 1,037,285) with a median age of 56 years (IQR: 40–72) (Table S1). The majority of surgical procedures were performed in the Capital Region of Denmark (75.2%) with over a quarter of all surgical procedures being performed at Rigshospitalet (26.6%) – the most highly specialised hospital in both regions and the only hospital performing thoracic, liver, and neurosurgery (Figure 3). The population entailed all surgical specialities, with the majority of procedures being orthopaedic (25.1%) or gastrointestinal (23.9%) (Table S1).

FIGURE 3.

Flowchart of included patients and procedures and included hospitals.

3.1.2. Surgical Procedures

The TRIPLE‐A surgical classification ended up consisting of 42 overall classifications, 368 overall procedures, and 1070 specific surgical procedures. In total, the classification system included 871,598 validated surgeries (Table S2).

3.1.3. Variables

The processed database consists of 17 summary tables with seven to 946 variables each and a total of 1617 variables (Table 2).

TABLE 2.

Summary variables of TRIPLE‐A version 1.0.

Summary table	Variables, n (total 1617)	Description
Admission	13	Patient and surgical IDs, time stamps for different events during admission.
Anaesthesia pre‐exam	21	Anaesthetist's assessment of patient comorbidities, prior anaesthesia incidences, airway evaluation and anaesthesia plan for the specific surgical procedure.
Anaesthesia	14	Type of anaesthesia and anaesthetics, use of special monitoring systems.
Airways	84	Specific details on airway management during anaesthesia, including airway device, type and size, number of airway attempts, Cormack‐Lehane Classification, tube depths, etc.
Charlson Comorbidity Index	19	Sub‐domains and total score.
Complications	7	0–90 days mortality, 0–30 days acute kidney injury, 0–30 reoperation, etc.
Demographics	38	Age, sex, BMI, hospital, ASA‐score, smoking status, etc.
Medication	98	Variables for pre‐, intra‐, and postoperative doses of anaesthesia, analgesics, antibiotics and all other medications. e.g., administration of paracetamol 0–6 h before end‐anaesthesia.
Medical notes	15	Admission note, full pre‐exam, surgery plan note, surgical description, discharge note. Further, alle medical notes −180 to 180 days from the date of surgery aggregated into timely variables (e.g., 0–7 days postoperative notes).
OPIAID	24	Intraoperative variables tailored for the period start‐surgery to end‐anaesthesia and ‘optimal intraoperative morphine‐equivalent dose’.
Radiology	9	All radiology descriptions −180 to 180 days from the date of surgery split into specific time periods.
Samples	946	Haemoglobin, CRP, Leukocytes, Creatinine, Sodium, Potassium, Troponin, CKMB, Lactate, HbA1C, etc., at different time periods and variable types, for example latest preoperative haemoglobin or highest CRP 0–7 days postoperatively.
Research studies	1	ID on every study participation, enabling identification of included patients in specific studies for automated data retrieval.
Symptoms	55	Pain, shivering, nausea, vomiting, sedation and other opioid‐related adverse events. These are available for different time periods before and after anaesthesia, e.g., worst pain rating 0–2 h postoperatively.
Times	25	Length of different phases during the admission.
Ventilator	223	Intraoperative ventilator details (tidal volume, respiratory frequency, compliance, end‐tidal CO2, etc.). Values are available as mean, median, max, min, mean of the first 5 min, etc.
Vitals	25	Intraoperative vital parameters; e.g., mean heart rate and lowest perfusion index.

3.1.4. Projects

The database currently supports 34 unique research projects (Table S3).

4. Discussion

The TRIPLE‐A database includes 1,142,108 surgeries classified into 603 different surgical procedures. For each unique procedure, more than 1617 clinically relevant variables are available. Data are used for perioperative epidemiological research, quality assurance, automated case report forms, and development of prediction models.

The database is intended to support these four distinct areas, for which we develop methodological setups that can be used as automated standard templates for most projects within each area.

4.1. Epidemiological Research

The database already hosts a multitude of epidemiological research (Table S3). Examples of ongoing projects are studies that assess the efficacy of different surgical approaches, create an updated overview of procedure‐specific postoperative pain outcomes, and investigate associations between peripheral perfusion index and mortality. Each specific project requires its own ethical and data approval. This is handled in a standardized form by TRIPLE‐A collaborators. When the project is approved, we use TRIPLE‐A data to bypass normal data retrieval, curation, and validation. In total, this bypass typically reduce the project timeline by half or a whole year and save 5000 to 10,000 € only in data retrieval costs depending on the project. In turn, for each project, the external collaborator helps improve the database by validating a subset of data by comparing it to that in the EMR and making suggestions for variable improvements. Our vision is to have a fully validated database by the end of 2028.

To facilitate epidemiological research in the database, we are developing a web‐based data platform that hosts updated aggregate TRIPLE‐A data, where external researchers in co‐creation with a TRIPLE‐A collaborator can protocol studies based on TRIPLE‐A data (www.triplea.dk). The solution will include a step‐by‐step guided interface, where users conceptualize their studies in PICO format [25]. Because the platform will hold actual data distributions, visualization of patient characteristics can be made and the most appropriate statistical analyses can be protocolled prior to data analyses. When these steps are filled in by the researcher, an underlying program will use the information to automatically generate drafts for a standardized study protocol, application for ethical approval, variable list, and a database query. Our vision is to have this platform running by the end of 2026.

As a sub‐goal within the epidemiological branch of TRIPLE‐A, we use the platform's fast project generation capabilities to support research training during anaesthesiology specialist education in Eastern Denmark. In this setting, a group of anaesthesiologists in training—often with no prior research experience—is paired with experienced researchers to conduct an epidemiological study. This collaboration aims to provide hands‐on experience with research methodology and the TRIPLE‐A platform, to potentially facilitate future engagement in data‐driven clinical research during and after specialization. This process commences in 2026.

4.2. Automatic Data Retrieval and Cleaning

The aim of the automated CRF is to save time and research resources compared to conventional electronic CRFs, where large amounts of data are typically manually extracted from the EMR system and re‐recorded into another database. Initially, we focused on creating aCRFs for pain and airway management trials. We are currently validating variables to support the OPIAID study [28] and ROCVIDEO [29], VIDFACE [30], and VITION [31] trials with the TRIPLE‐A aCRF. We utilize the prospective database approval to continuously improve and build upon the aCRF setup. Hence, when the aCRF is fully developed and validated, the next steps are to create add‐ons for blinded and automated data analytical and statistical reports.

In the future, our data warehouse with the aCRF add‐on could also potentially be able to support adaptive platform trials in perioperative care [32].

4.3. Quality Assurance

Actual benefits or harms of broad clinical implementation of new medical/surgical treatments may differ from research findings in selected research cohorts, why monitoring implementation is key when evaluating clinical impact of interventions [16]. From a modern research perspective, healthcare interventions should be evaluated for patient‐related outcome measures (PROM), (such as pain or adverse events), organizational aspects such as length of PACU stay, economic aspects (such as total perioperative cost) and environmental outcomes (such as total carbon equivalent emission or waste incineration) [33, 34, 35]. By using TRIPLE‐A data and a standardized before‐and‐after study design, we work towards a setup where we can evaluate implementation of new interventions across these domains. To minimize the limitations of non‐randomization and temporal confounding, we will use interrupted time series regression to account for causal effects of local confounding factors and trend changes [36].

4.4. Prediction Models as Clinical Support Tools

Machine learning models are increasingly applied to EMR systems to support clinical decision‐making [37, 38]. TRIPLE‐A has planned a series of prediction projects to develop real‐time machine learning models for detecting patients at high risk of postoperative complications (pulmonary embolism, acute kidney injury, cardiac complications, major bleeding, and death) and persistent opioid use, using predictor variables from demography, anaesthesia, perioperative analgesia, surgery, vital signs, and recovery. These models are intended for EMR implementation as safeguards during the surgical admission and will eventually be able to alert personnel in high‐risk cases. To improve detection of postoperative complications, we intend to use large‐language models for data‐mining on 0 to 6 months postoperative EMR notes and readmission data. Alongside AI‐supported care, we strive towards a more datadriven healthcare system. Here is our vision: Surgeons and anaesthesiologists responsible for local guidelines at Region East hospitals will recieve procedure‐ and hospital‐specific annual reports of important variables, such as the frequency of moderate/severe pain and nausea 0‐24h after surgery. Thereby, clinicians can continuously monitor and improve care based on the most contemporary data. Further,recovery outcomes can be stratified on patient characteristics and this can be communicated to patients as an “individualised alignment of expectations for recovery”. The TRIPLE‐A platform at www.triplea.dk is intended for communicating these individual case alignments as text or via an AI‐avatar. This could be particularly relevant for paediatric surgery, where “what to expect” could be presented by e.g., Peppa the Pig.

An example of an AI‐prediction model that is currently being developed based on TRIPLE‐A data is the Perioperative Opioid Algorithms for Individualised Dosing (OPIAID) prediction tool for perioperative opioid dosage [28]. While major advances in one size‐fits‐all multimodal analgesic interventions have been achieved by results from large RCTs [39, 40], still, postoperative opioids are required for most procedures and a significant proportion of patients experience pain, large opioid requirements and opioid‐related adverse events [41, 42, 43]. The OPIAID project uses TRIPLE‐A to construct a dataset of candidate predictor variables and to define the outcome variable ‘optimal opioid dose’ based on actual administered intraoperative opioid doses and postoperative symptoms. A machine‐learning model is then developed to provide real‐time, patient‐specific dose recommendations [28].

4.5. Limitations

The database has several important limitations. The underlying data were originally collected as part of clinical care rather than for research purposes, meaning that certain clinically relevant details were not systematically recorded and may be missing or inconsistently captured. To reduce this limitation, prospectively, we will seek to improve data registration in the EMR across the included hospitals using our collaborative network. Some variables may have been oversimplified with the risk of overlooking important nuances—for example from categories of antibiotic into a binary variable of antibiotic treatment yes/no. In sub‐studies that need specific details, new and more detailed variables can be coded on request. Missing data from incomplete documentation or variation in adherence to local procedures may have introduced bias even if assumed ‘missing at random’. Treatment choices in the extracted data also holds a risk of ‘confounding by indication’, since treatment choices and perioperative decisions may be linked to unrecorded patient or procedural characteristics. Finally, the lack of linkage to external data sources, such as pharmacy or long‐term follow‐up registers, constrains the broader interpretability of our findings. Therefore, we are currently integrating external sources from Statistics Denmark such as redeemed prescription data, socioeconomic status and ethnicity.

5. Conclusion

The TRIPLE‐A database is a granular, high‐quality perioperative database that is updated biannually and whenever necessary for assisting data retrieval in collaborative studies. The current Version 1 holds 1.1 million surgical procedures and 1617 clinically meaningful variables to support epidemiology research, clinical evaluation of medical interventions, automated data capture in clinical studies, and development of machine‐learning‐based prediction tools. TRIPLE‐A invites other research groups to collaborate and use this high‐quality database to develop resource‐ and cost‐optimized research across the entire perioperative field.

Author Contributions

The database was designed by A.P.H.K. and M.H.O. Data extraction was carried out by A.P.H.K. and M.H.O. Data processing was carried out by M.H.O., M.M., Anton Kokholm. The first version of the manuscript was written by A.P.H.K. and M.H.O.; all authors revised the manuscript; and all authors and collaborators (see below) approved the final version.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

C.S.M.: Founder of a start‐up company, WARD24/7 ApS, with the aim of pursuing the regulatory and commercial activities of the WARD‐project (Wireless Assessment of Respiratory and circulatory Distress, a project developing a clinical support system for continuous wireless monitoring of vital signs). WARD24/7 ApS has obtained license agreement for any WARD‐project software and patents. One patent has been filed: “Wireless Assessment of Respiratory and circulatory Distress (WARD), EP 21184712.4 and EP 21205557.8”. None of these related to the present manuscript. The other authors declare no conflicts of interest.

Supporting information

Table S1: Surgical specialties and age distribution.

Table S2:. Surgical procedures with n ≥ 1000 included in database per July 2025.

Table S3:. Planned TRIPLE‐A studies and data deliveries per July 2025.

Karlsen A. P. H., Møller M. Ø., Pedersen N. K., et al., “The AI and Automation in Anaesthesia (TRIPLE‐A) Perioperative Database in Eastern Denmark 2017–: A Platform for Epidemiology, AI‐Driven Prediction, Quality Control and Automated Data Retrieval,” Acta Anaesthesiologica Scandinavica 69, no. 9 (2025): e70116, 10.1111/aas.70116.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.