Personalised perioperative opioids
Patient-relevant composite pain outcomes
Opioid minimal important differences
Motivation
300 million surgical procedures each year globally - most require opioids.
Individual opioid titration guidelines are non-existent.
30-60% of surgical patients experience pain or opioid adverse events.
Demographic, genetic and surgical predictors could help develop individualised opioid algorithms.
Research program
Designed in accordance with the highest methodological standards for developing prediction models for personalised medicine. Divided into six phases:
BACKGROUND:
World-wide, 300 million surgeries are performed each year. The majority need opioids as a part of immediate postoperative pain management. Opioid doses must be carefully balanced to provide adequate pain relief and minimise adverse events and long-term use. Currently, both pain and opioid-related adverse events occur in 20-30% of surgical patients.
AIM:
The aim of OPI•AID is to individualise perioperative opioid treatment to reduce pain, adverse events, complications, and overall opioid use, while facilitating effective and high-quality patient care.
NOVELTY:
Currently, opioid dosing during and after surgery is based on the clinician’s best estimate. If we succeed, OPI•AID will be the first algorithm for prediction of individualised perioperative opioid requirements.
METHOD:
- The OPI•AID Retrospective Cohort contains 742,095 surgeries, including detailed information on patient characteristics, surgical factors, perioperative analgesic treatment, pain, opioid-related adverse events, and morbidity. Using this database, a machine learning model will be trained to predict opioid requirements during and after surgery.
- The OPI•AID Prospective Cohort will validate the model and contribute new evidence on opioid minimal important differences, patient-centred pain outcomes and prognostic pharmacogenetics.
- Scoping reviews will allow incorporation of current evidence on opioid treatment in The OPI•AID Research Program.
- The OPI•AID ClusterRCT with 3,000 patients will validate and recalibrate the opioid dosing algorithm in a randomised setup against standard of care.
ORGANISATION:
The OPI•AID project is run from Bispebjerg and Frederiksberg University Hospital in collaboration with Zealand University Hospital Køge and Copenhagen Trial Unit at Rigshospitalet. The central organisation includes: expert in data management PhD Markus Harboe Olsen; expert in innovative solutions Professor Christian Meyhoff; experts in pain management Professor Ole Mathiesen and Associate Professor Troels H Lunn; and expert in trial methodology Professor Janus C Jakobsen. Ole Mathiesen and Janus C Jakobsen have driven some of the globally largest published multicentre RCTs on multimodal analgesia.2, 3, 28 Leading up to OPI•AID, the collaborators have conducted perioperative pain research that inform the research program (https://opiaid.dk/publications).29-35
Other collaborations:
FUNDING:
- The Capital Region of Denmark Research Fund
- Bispebjerg and Frederiksberg Hospital Research Fund
- Department of Anaesthesia and Intensive Care Medicine, Bispebjerg and Frederiksberg Hospital
USE OF MACHINE LEARNING IN THE OPI•AID RETROSPECTIVE COHORT:
Descriptive statistics will be used to summarize demographic and clinical characteristics. Simple data analyses will be used to understand trends and correlations. Exploratory Data Analysis (EDA) will identify patterns, correlations, and outliers, primarily through data visualization techniques. This phase will ensure a representative split for training and test/validation datasets and identify potential outliers for data point focus or elimination.
In data pre-processing, raw data will be transformed into a format suitable for model training. This includes handling missing values, encoding categorical variables, and normalizing variables to enhance data quality for modelling.
The machine learning models, including Logistic Regression, K-nearest Neighbor, Naïve Bayes, and others, will be developed and trained. Models will be adjusted and tuned through hyperparameter tuning using methods like Grid Search to find optimal configurations. The dataset will be split into training and test segments, ensuring a representative sample for model training.
Model evaluation based on whether the model addresses classification or regression problems will be used. Accuracy and Area Under Curve (AUC) for classification models and Root Mean Squared Error (RMSE) for regression models. The evaluation will include an examination of the models' loss-curve to assess overfitting or underfitting. This process will form the basis for determining the best fitting machine learning model for the dataset, supported by statistical analysis of various evaluation metrics.
The OPI•AID Project group is situated in Denmark at:
- Dept. of Anaesthesia and Intensive Care Medicine, Bispebjerg and Frederiksberg University Hospital
- Dept. of Anaesthesiology, Zealand University Hospital, Køge
- Copenhagen Trial Unit, Rigshospitalet
- The IT-University of Copenhagen
- In collaboration with CEPRA (cepra.nu)