VitalDB: fostering collaboration in anaesthesia research

In the April issue of British Journal Anaesthesia, Lee and colleagues1Lee S. H.-C. Chu Y.S. et al.Deep learning models for prediction intraoperative hypotension.Br J Anaesth. 2021; 126: 808-817Abstract Full Text PDF PubMed Scopus (11) Google Scholar reported development validation deep hypotension. The authors developed an algorithm that, unlike marketed algorithms,2Hatib F. Jian Z. Buddi al.Machine-learning to predict hypotension based on high-fidelity arterial pressure waveform analysis.Anesthesiology. 2018; 129: 663-674Crossref (176) Scholar, 3Davies S.J. Vistisen S.T. Hatib Scheeren T.W.L. Ability analysis–derived index future hypotensive events in surgical patients.Anesth Analg. 2020; 130: 352-359Crossref (62) 4Wijnberge M. Geerts B.F. Hol L. al.Effect a machine learning-derived early warning system vs standard care depth duration during elective noncardiac surgery HYPE randomized clinical trial.JAMA. 323: 1052-1060Crossref (111) 5Schneck E. Schulte D. Habig al.Hypotension Prediction Index protocolized haemodynamic management reduces incidence primary total hip arthroplasty: single centre feasibility randomised blinded prospective interventional trial.J Clin Monit Comput. 34: 1149-1158Crossref (22) make use multimodal biosignal waveforms, acquired using routine invasive noninvasive patient monitoring events. Using data from 3301 patients their database, they trained validated model. Although some aspects methodology may still be improved, such as (acausal) extraction events,6Vistisen Johnson A.E.W. Predicting vital sign deterioration with artificial intelligence or learning.J 2019; 33: 949-951Crossref (14) model demonstrates strong predictive performance up 15 min before its actual occurrence, particularly when inputs included combined rather than signals. What really sets study apart others kind, however, is that have released both code underpin findings.7Lee H-C. Jung C-W. Vital Recorder—a free research tool automatic recording high-resolution time-synchronised physiological multiple anaesthesia devices.Sci Rep. 8 (Nature Publishing Group): 1527Crossref (71) practices are changing, there too few motivations researchers share well-curated self-developed software. effort taken by create, document, release this unprecedented perioperative dataset, VitalDB along analysis should serve lesson community. creation easily accessible physiologic databases within intensive has created outstanding education opportunities over past two decades.8Johnson A.E. Pollard T.J. Shen al.MIMIC-III, freely critical database.Sci Data. 2016; 3: 160035Crossref (2497) 9Pollard Raffa J.D. Celi L.A. Mark R.G. Badawi O. eICU Collaborative Research Database, available multi-center database research.Sci 5: 180178Crossref (287) 10Hyland S.L. Faltys Hüser al.Early circulatory failure unit learning.Nat Med. 26: 364-373Crossref (76) These offered fundamental insights into platform interdisciplinary educational programmes projects. Here, we highlight accomplishments consider relatively new context other currently datasets field (Table 1).Table 1Overview large, openly care. DUA, digital user agreement; MIMIC, Medical Information Mart Intensive Care; PIC, Paediatric eICU, Database; HiRID, High time Resolution ICU Dataset; AmsterdamUMCdb, Amsterdam University Centre database; VitalDB, database.Namentime spanApproximate resolution signs (HR, MAP, etc.)Access requirementsGeneral remarksDUATraining/courseSubmission summary planned researchMIMIC-IV60 000+ patients2001–16Every 1 h, matched waveforms offer possibility higher resolutionRequiredRequiredNot requiredWidely used research, comprising tertiary academic medical Boston, MA, USA. latest version MIMIC-IV, includes broad range modalities, including chest radiograms, electrocardiogram structured observations, unstructured notes.PIC12 patients2010–8Varying because manual entries (during every 5 min)RequiredRequiredNot requiredComprises paediatric admission children's hospital China. measurements, medications, laboratory fluid balance recordings, diagnostic codes, demographic information, moreeICU 200 2014–5Every minRequiredRequiredNot more units across continental USA collected part telehealth programme. Includes demographics series observations signs, tests.HiRID33 patients2008–16Every 2 minRequiredRequiredRequiredHiRID published datasets, most importantly bedside variables recorded min. This creates unique reliably characterising status stay.AmsterdamUMCdb23 patients2003–16Up one value requiredThe dataset demographics, tests, medications admissions.VitalDB6000+ patients2016–7Every 1–7 s also availableRequiredNot requiredNot first (holding ECG, blood pressure, plethysmography, etc.). It holds surgery-related information procedure, comorbidities, outcomes (mortality), preoperative values, treatments (e.g. estimated loss) Open table tab own which was recently is, our knowledge, open large containing setting (www.vitaldb.net).7Lee 6000 were captured monitors (ECG, etc.), ventilator (airway capnography at rate 62.5 Hz), depth-of-anaesthesia monitor (EEG bispectral [BIS] monitor, sampling 180 Hz). Furthermore, infusion pumps (drug, volume) cardiopulmonary trending heart respiratory rate; s) captured, essentially include numbers displayed operating room given recorded. addition, 60 provided help interpret On top this, extensive descriptive treatments, loss). With plethora data, can think many potentially interesting observational studies conduct. addition released, cost, software creation, VitalRecorder. application allows institutions set similar but simple capture studies.11Kim J. H.C. Byun S.H. al.Frontal electroencephalogram activity emergence general children without delirium.Br 293-303Abstract (4) 12Lee J.H. Ji al.Evaluation intratidal compliance profile different PEEP levels healthy lungs: prospective, crossover study.Br 125: 818-825Abstract (3) 13Oh H. Choe Kim Y.J. Yoon H.-K. Park H.-P. Intraarterial catheter diameter dynamic response system: controlled (epub ahead print)Crossref (2) For instance, often want obtain specific patient. Data export solutions, whether device manufacturers third parties, been costly limited supporting only modality. Today, least solutions exist: VitalRecorder7Lee VSCapture.14Karippacheril Ho T. acquisition S/5 GE Datex anesthesia VSCapture: source.NET/Mono tool.J Anaesthesiol Pharmacol. 2013; 29: 423Crossref (12) Both support different, partly overlapping, devices. VitalRecorder provides real-time graphical feedback visually confirming whereas VSCapture seems underway visualisations real-time. source, allowing project-specific modifications able impressive array team maintains accessible, illustrated guide connecting these use, it not yet source. There valid reasons keep closed, impedes usability collaborative integration Many projects will require devices supported Developing slow cumbersome process, requires direct access device. Allowing community contribute could greatly increase applicability already remarkable tool. Making shared prompt international collaborations. recent years, wider electronic health record systems increasing recognition importance sharing.15Thoral P.J. Peppink J.M. Driessen R.H. al.Sharing responsibly under Society Critical Care Medicine/European Medicine Joint Science Collaboration: Centers Database (AmsterdamUMCdb) example.Crit https://doi.org/10.1097/CCM.0000000000004916Crossref (19) Several detailed patient-level become publicly decade, communities working together reuse analytical code. resource although exist area 1). One widely known (MIMIC), comprises de-identified associated >60 000 admitted USA.8Johnson Scholar,16Johnson Stone D.J. MIMIC Code Repository: enabling reproducibility research.J Am Med Inform Assoc. 25: 32-39Crossref (115) investigators engineers around world, helping drive informatics, epidemiology, learning. Following footsteps (PIC) paediatric-specific populated Children's Hospital Zhejiang School Zhejiang, PIC encompasses 13 499 distinct admissions 12 881 (aged 0–18 yr). Another notable multicentre >200 hospitals USA, Philips Institute.9Pollard As yet, none aforementioned so fills unmet need. European emerged. Swiss, single-centre, HiRID database10Hyland nearly 34 patients' resembles ways. Dutch AmsterdamUMCdb,15Thoral contains 23 admissions. EU, UK Biobank,17Sudlow C. Gallacher Allen N. al.UK Biobank: identifying causes wide complex diseases middle old age.PLoS 2015; 12e1001779Crossref (2819) EU. AmsterdamUMCdb credentialed access, aspect comply EU General Protection Regulation (GDPR). advocating sharing struggle national interpretation GDPR substantial difficulties figuring out how (or pseudonymised) possible. example days interpretations legislation, debated Denmark (and even if all) well-established registries legislation. legislation remains difficult fully grasp leaders very progressive aspect. Table 1, all mentioned described detail content, approximate resolution, access. Most after credentialing, completion training course signing off agreement mandating responsible handling adherence principle research. required typically completed 1–2 h. advances transfer speed computation led rapid routinely paper MIMIC-III, 2016, cited 1100 times according Scopus, 465 2020 alone, illustrating massive impact releasing dataset. readily undoubtedly facilitates interdisciplinary, computational too. exemplified statements creators database15Thoral Scholar: ‘Our main goal connect healthcare science’. By opening scientists, clinicians work together, turn learn hard, frustrating discipline getting acquainted complexities data. melting pot necessary ideas emerge generates focal point collaboration. Therefore, welcome being first, hold Organising writing draft editorial. Proofreading final approval manuscript: STV. contributing TJP. JNE. TWLS. TJP received funding US National Institute Biomedical Imaging Bioengineering (Bethesda, MD, USA) grant number R01EB030362 , Healthcare . TWLS grants honoraria Edwards Lifesciences (Irvine, CA, Masimo Inc. consulting lecturing Pulsion Systems SE (Feldkirchen, Germany) lecturing. All declare no conflicts interest. Deep hypotensionBritish AnaesthesiaVol. 126Issue 4PreviewIntraoperative risk postoperative organ dysfunction. study, aimed present algorithms predictions 5, 10, event. Full-Text Archive