Applying Transportability to Strengthen Pharma Evidence in the Nordics
Transportability and generalizability refer to the ability to apply evidence from a single study to different patient populations, clinical settings, or geographical regions.
Smoking clearly increases the risk of postoperative complications
A smoker’s risk for any postoperative complications is significantly higher than non-smokers’. A machine learning-based big data study undisputedly confirms the risks of smoking for recovery.
The material of the register study published in March included all operations completed in HUS in 2015-2019. The extensive study reveals important information on the effect of smoking on postoperative complications but is also a concrete demonstration of the wide possibilities of machine learning and hospital data lakes in research.
According to Porvoo Hospital Head Physician for Lung Diseases Heikki Ekroos, who led the study, it undisputedly confirmed the negative effects of smoking on surgery patients.
The original material covered approximately million operations
The original material covered approximately million operations. The patient records of surgery patients were analysed, and all smoking-related sentences were collected.
“After this, clinicians read the sentences and analysed which describe smokers, and which former smokers or non-smokers. When the sentences were read and the definitions made, the machine learning algorithm was trained to find smokers”, Medaffcon’s Senior Data Scientist Juhani Aakko tells.
The algorithm was able to differentiate smokers and former smokers as well as non-smokers from the patient records. The physicians had also pre-defined complications that were searched.
The study excluded patients that were under the age of 16 years, those with unclear smoking status as well as those whose ASA class was unknown.
ASA is a physical status classification system of the American Society of Anesthesiologists Classification, describing the physical status of a patient having an operation. In the 1-5 category classification system, 1 is a healthy patient under 65 years and 5 is a terminally ill patient whose estimated survival is no more than 24 hours without an operation.
Finally, the screening resulted in approximately 160,000 operations
In the end, the AI analysed a total of 158,638 operations in the study. According to the results, both current and former smokers have a clearly increased risk of having postoperative complications.
“The material was very extensive, but we had to exclude hundreds of thousands of operations, because smoking status was not included in the patient record”, states Ekroos.
Sometimes, the information was so unclearly recorded that it could not be utilised.
The study is part of the doctoral dissertation of specialising physician Helene Gräsbeck, supervised by Ekroos together with Professor Tuula Vasankari. Medaffcon’s Juhani Aakko and Olivia Hölsä were in charge of the data analysis, the latter of which created the algorithm as part of her master’s thesis.
The study will continue by assessing the costs related to postoperative treatment of smoking patients and how much larger they are compared to the costs of non-smoking patients. The costs are impacted, for example, by the duration of hospitalisation and the number of emergency room visits.
Read more:
Text mining helped analyse a million entries on smoking
Medaffcon’s Machine Learning Algorithm Adapts to Various Research Needs
Unstructured healthcare data contributes to solving new issues
Unstructured data and Machine learning in Real-World Evidence studies
