An observational study in ASA-PS Classification, sponsored by Erasmus Medical Center. Completed at 1 site in Netherlands. Open to participants aged 18 Years and older. Per ClinicalTrials.gov, last updated 2025-05-18.
Sponsored by Erasmus Medical Center · Observational
The development of a machine learning algorithm that predicts American Society of Anesthesiologist-Physical Status (ASA-PS) based on preoperative variables would not only improve clinical decision-making in patient risk stratification but also offer a more reliable tool for administrative and regulatory uses. Therefore, the development of such a machine learning tool presents a significant opportunity to advance both the science and practice of perioperative care. Incorporating medication use into the algorithm could further enhance its predictive power, as it is closely linked to systemic disease. This addition could help refine the ASA-PS classification, making it an even more valuable tool in the clinical setting.
The American Society of Anesthesiologists Physical Status (ASA-PS) classification system is a widely used tool for assessing surgical fitness and other clinical contexts. However, its inherent subjectivity and heavy reliance on clinician judgment can lead to inconsistencies in patient risk stratification, a critical component of perioperative care. Furthermore, the ASA-PS system has been adopted for various administrative and regulatory purposes beyond its original intent, such as quality assessment by the Dutch Health and Youth Care Inspectorate (IGJ), compensation decisions by private payers in the USA, patient triage, and determining suitability for certain types of surgery.
Given the broad and critical applications of the ASA-PS system, enhancing its precision and objectivity is of paramount importance. One way to achieve this is through the development of a machine learning algorithm that predicts ASA-PS based on preoperative variables. Anesthesiologists base the ASA-PS score on the presence of systemic diseases, which can be inferred from medication use. By leveraging data such as Anatomical Therapeutic Chemical (ATC) codes, BMI, sex, age, routinely collected preoperative health data, and medication use, this algorithm could provide a more consistent and objective measure of ASA-PS.
This would not only improve clinical decision-making in patient risk stratification but also offer a more reliable tool for administrative and regulatory uses. Therefore, the development of such a machine learning tool presents a significant opportunity to advance both the science and practice of perioperative care. Incorporating medication use into the algorithm could further enhance its predictive power, as it is closely linked to systemic disease. This addition could help refine the ASA-PS classification, making it an even more valuable tool in the clinical setting.
Erasmus Medical Center is the lead sponsor of 466 studies on the registry; 179 are open to participants now.
Counted across the registry records on this site, refreshed daily.
All patients who underwent a surgical, diagnostic or therapeutic procedure within the surgical suite of the Erasmus MC since 2018 (introduction new digital Hospital Information System) and who had a ASA-PS class recorded.
Exclusion Criteria:
The American Society of Anesthesiologists physical status (ASA-PS) class
The dependent response variable will be the ASA-PS class, both as a four-level variable (ASA-PS I, II, III and IV) and a two-level variable (ASA-PS I and II versus ASA-PS III and IV). The ASA-PS class was assigned to the patient and recorded in the patients file in the EMR by an anesthesiologist of resident anesthesiology as a part of the routinely performed preoperative anesthesiological screening in preparation for a procedure.
Time frame: Day 0
Performance metrics: accuracy
The correct classification of the ASA-PS score will be evaluated using performance metrics of the machine learning algorithms. Common performance metrics include: Accuracy (the proportion of correctly predicted instances).
Time frame: day 0
Performance metrics: precision
The correct classification of the ASA-PS score will be evaluated using performance metrics of the machine learning algorithms. Common performance metrics include: precision (the ratio of true positive predictions to the total positive predictions)
Time frame: day 0
Performance metrics:recall
The correct classification of the ASA-PS score will be evaluated using performance metrics of the machine learning algorithms. Common performance metrics include:ecall (sensitivity or the ratio of true positive predictions to the actual positive instances)
Time frame: day 0
Performance metrics: F1-score
The correct classification of the ASA-PS score will be evaluated using performance metrics of the machine learning algorithms. Common performance metrics include: F1-score (the harmonic mean of precision and recall).
Time frame: day 0
Performance metrics: Area Under the Receiver Operating Characteristic Curve
The correct classification of the ASA-PS score will be evaluated using performance metrics of the machine learning algorithms. Common performance metrics include:the Area Under the Receiver Operating Characteristic Curve (AUC-ROC, Measures the model's ability to discriminate between positive and negative instances).
Time frame: day 0
Calibration
Calibration plots will be used to assess the agreement between predictions and the event rate (i.e. correct classification).
Time frame: Day 0
Misclassification of the ASA-PS score
A manual review of a selection of misclassifications will be performed by two anesthesiologists to qualitatively assess the cause of the misclassification.
Time frame: Day 0
Explainability of the prediction model:Shapley additive explanations (SHAP)
Shapley additive explanations (SHAP) if applicable, as it can offer insights into the contribution of each feature to the prediction of individual instances.
Time frame: day 0
Explainability of the prediction model:Local interpretable model-agnostic explanations (LIME)
Local interpretable model-agnostic explanations (LIME) can offer insights into the contribution of each feature to the prediction of individual instances.
Time frame: day 0
Optimal sample size
Analysis of the learning curves to determine if additional data would likely improve the model's performance or if the current dataset is sufficient.
Time frame: day 0
Documents are hosted by the registry — open the source record to download them.
Plan to share: Yes — All (Underlying) pseudonymised data will be made available alongside with the publication to execute the training and validation of the models. Data will be uploaded in dataverse.
Supporting information: Study protocol, Sap, Csr
This study is completed, as verified in May 2025. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.
Erasmus Medical Center