BY ALI RAZA
Introduction
Machine learning is an exciting field that has grown exponentially in the field of medicine in recent years. From personalised medicine to gene sequencing, the helping hand of machine learning models and tools has become massively beneficial in the field of medicine and medical research. Machine learning (ML) may be defined as a subset of artificial intelligence that involves the development of statistical models and algorithms that enable computers to perform tasks without explicit instructions. This can be achieved by feeding machine learning models large volumes of data and creating algorithms for ML models to recognise patterns, allowing the model to make predictions or decisions.1
With this in mind, how can we apply this in a surgical context to increase the quality of healthcare? Machine learning models can be trained using medical and genomic data to make predictions on various analytics that relate to surgical outcomes, allowing for more informed decisions that can aid doctors and surgeons to maximise the safety of patients.
What data can be used?
As we discussed previously, a machine learning model is dependent on high quality and relevant data to create a model that can make an accurate prediction. So what data could be used to get predictive analytics in surgery? It is dependent on a case by case basis for the surgery we want predictive analytics for, however, many data types can be
1 https://towardsdatascience.com/machine-learning-basics-part-1-a36d38c7916
used such as patient history, genomic data, imaging data and lab results from various tests.2
Patient history data includes metrics about age, gender, ethnicity, medical history and lifestyle. Genomic data is the sequence of genes in a person, this can be used to view any potential genetic factors that may have an influence on the outcome of the surgery. Imaging and lab results can be processed by machine learning models to identify abnormalities that may not be readily distinguishable to the human eye.
Example of predictive analytics
Predictive analysis can be defined as analysing data to forecast future outcomes. Machine learning models can use predictive analysis to answer the question of ‘What next?’. Predictive analytics refer to the result created by the machine learning model’s analysis.
In the context of surgical outcomes, predictive analytics come in different shapes and sizes. An example of one of these is risk stratification; risk stratification is grouping patients into categories of risks that could occur during surgery such as bleeding, adverse effect to anaesthesia, background health conditions etc. This can help surgeons to decide on the most appropriate and safe approach to treating a patient.3
Risk stratification also aids in postoperative care for patients as patients who were predicted to be more likely in having postoperative complications may be more closely monitored, enhancing patient care, reducing readmission rates and optimising hospital resources.4
2
https://www.cureus.com/articles/247197-unveiling-the-influence-of-ai-predictive-analytics-on-patient-outco mes-a-comprehensive-narrative-review#!/
3 Bellini V, Valente M, Bertorelli G, Pifferi B, Craca M, Mordonini M, Lombardo G, Bottani E, Del Rio P, Bignami E. Machine learning in perioperative medicine: a systematic review. J Anesth Analg Crit Care. 2022 Jan 15;2(1):2. doi: 10.1186/s44158-022-00033-y. PMID: 37386544; PMCID: PMC8761048. 4 van den Eijnden MAC, van der Stam JA, Bouwman RA, Mestrom EHJ, Verhaegh WFJ, van Riel NAW, Cox LGE. Machine Learning for Postoperative Continuous Recovery Scores of Oncology Patients in
Machine learning models can be used to improve surgery scheduling by more accurately predicting time spent in the operating room and allocating resources more effectively to operating rooms to ensure they are used optimally.5 Machine learning models are shown to be 13% more accurate than humans in predicting the time taken in the operating room and through the use of more accurate scheduling could potentially reduce labour expenses by $79,000 over a 4 month period.6
Types of Machine Learning Models used
Machine learning models themselves come in different forms which allow them to collect data and learn from data differently for different goals. There are 2 major types which are used for surgical outcome prediction: Supervised learning and unsupervised learning which use different algorithms.
Supervised learning is when a model is trained on a labelled dataset where each example of an input is paired with an output. This allows supervised learning models (SLMs) to map points between inputs and their correct outputs and find correlation between them. This allows them to be very useful at predicting outcomes.7In our surgical context, this allows SLMs to predict things such as blood loss, length of the hospital stay, etc. based on the previous data categories mentioned before through linear regression or classification models.
While SLMs are more suited for prediction tasks, they are less suited for discovery tasks. This makes unsupervised learning models (ULMs) more suited. ULMs are trained by unlabeled data which are not given correct outputs with inputs, instead, the algorithm tries to find patterns and relationships within the data. This makes it more suited for
Perioperative Care with Data from Wearables. Sensors. 2023; 23(9):4455.
https://doi.org/10.3390/s23094455
5 Edelman ER, van Kuijk SMJ, Hamaekers AEW, de Korte MJM, van Merode GG, Buhre WFFA. Improving the Prediction of Total Surgical Procedure Time Using Linear Regression Modeling. Front Med (Lausanne). 2017 Jun 19;4:85. doi: 10.3389/fmed.2017.00085. PMID: 28674693; PMCID: PMC5475434. 6 https://healthcare-in-europe.com/en/news/machine-learning-surgery-schedule.html 7 https://www.geeksforgeeks.org/types-of-machine-learning/
discovery tasks and within a surgical context can use clustering models for risk stratification and dimensionality reduction to simplify large medical datasets to identify the most important data to predict surgical outcomes.
Ethical considerations and limitations
While ML shows immense potential, there are still limitations and ethical concerns that must be taken into account. Bias is an issue that some ML models face due to unrepresentative training data. Another hurdle is ensuring privacy and anonymity of patient data making it potentially harder to access. Increased uses of ML models may lead to ethical concerns of accountability in cases where ML driven procedures lead to unsavoury outcomes. ML models have to be made to be able to be used by healthcare professionals and quality of data must be assured to prevent adverse effects of ML models entering hospitals.
Conclusion
Machine learning models for surgical predictive analysis show very exciting potential for the healthcare industry, having the potential to revolutionise and optimise many aspects of surgery, postoperative care and hospital management. As technology advances and
high quality data becomes more accessible, machine learning systems will evolve in predictive analysis for surgical outcomes, leading to a safer and more streamlined surgical process.
References
1. https://towardsdatascience.com/machine-learning-basics-part-1-a36d38c7916 2. https://www.cureus.com/articles/247197-unveiling-the-influence-of-ai-predictive-a nalytics-on-patient-outcomes-a-comprehensive-narrative-review#!/ 3. Bellini V, Valente M, Bertorelli G, Pifferi B, Craca M, Mordonini M, Lombardo G, Bottani E, Del Rio P, Bignami E. Machine learning in perioperative medicine: a systematic review. J Anesth Analg Crit Care. 2022 Jan 15;2(1):2. doi: 10.1186/s44158-022-00033-y. PMID: 37386544; PMCID: PMC8761048. 4. van den Eijnden MAC, van der Stam JA, Bouwman RA, Mestrom EHJ, Verhaegh WFJ, van Riel NAW, Cox LGE. Machine Learning for Postoperative Continuous Recovery Scores of Oncology Patients in Perioperative Care with Data from Wearables. Sensors. 2023; 23(9):4455. https://doi.org/10.3390/s23094455 5. Edelman ER, van Kuijk SMJ, Hamaekers AEW, de Korte MJM, van Merode GG, Buhre WFFA. Improving the Prediction of Total Surgical Procedure Time Using Linear Regression Modeling. Front Med (Lausanne). 2017 Jun 19;4:85. doi: 10.3389/fmed.2017.00085. PMID: 28674693; PMCID: PMC5475434. 6. https://healthcare-in-europe.com/en/news/machine-learning-surgery-schedule.ht ml
7. https://www.geeksforgeeks.org/types-of-machine-learning/