An algorithm which can predict how long a patient might spend in hospital if they're diagnosed with bowel cancer could save the NHS millions of pounds and help patients feel better prepared, according to researchers from the University of Portsmouth and the Portsmouth Hospitals University NHS Trust. The intelligent model will allow healthcare providers to design the best patient care and prioritize resources.
In their study, ‘Data analytics and artificial intelligence in predicting length of stay, readmission, and mortality: a population-based study of surgical management of colorectal cancer’, published in Discover Oncology, they used artificial intelligence and data analytics to predict the length of hospital stay for bowel cancer patients, whether they will be readmitted after surgery, and their likelihood of death over a one or three-month period.
"It is estimated that by 2035 there will be around 2.4 million new cases of bowel cancer annually worldwide,” said Professor of Intelligent Systems, Adrian Hopgood, from the University of Portsmouth, is one of the lead authors on the new paper. “This is a staggering figure and one that can't be ignored. We need to act now to improve patient outcomes. This technology can give patients insight into what they're likely to experience. They can not only be given a good indication of what their longer-term prognosis is, but also what to expect in the shorter term.”
The study used data taken from a database of over 4,000 bowel cancer patients who underwent surgery between 2003 and 2019. It looked at 47 different variables including age, weight, fitness, surgical approaches, and mortality.
A prospectively maintained colorectal cancer database was used, covering 4,336 patients who underwent colorectal cancer surgery between 2003 and 2019. The 47 patient parameters included demographics, peri- and post-operative outcomes, surgical approaches, complications and mortality. Data analytics were used to compare the importance of each variable and AI prediction models were built for length of stay (LOS), readmission, and mortality. Accuracies of at least 80% have been achieved. The significant predictors of LOS were age, ASA grade, operative time, presence or absence of a stoma, robotic or laparoscopic approach to surgery, and complications.
The insights of consultant surgeon Jim Khan and his colleagues Samuel Stefan and Karen Flashman were complemented by the analytical expertise of Dr Shamsul Masum, under Professor Hopgood's direction.
Professor Hopgood said: "We used a full set of data that included the 47 variables, but also predicted outcomes with just some of the most significant ones and found the two approaches showed very little difference. This is useful in itself because it shows that the algorithm is just as effective using a streamlined set of variables."
The significant predictors of readmission were age, laparoscopic procedure, stoma performed, preoperative nodal (N) stage, operation time, operation mode, previous surgery type, LOS, and the specific procedure. A BI-LSTM model predicted readmission with 87.5% accuracy, 84% sensitivity, and 90% specificity. The significant predictors of mortality were age, ASA grade, BMI, the formation of a stoma, preoperative TNM staging, neoadjuvant chemotherapy, curative resection and LOS. Classification predictive modelling predicted three different colorectal cancer mortality measures (overall mortality, and 31- and 91-days mortality) with 80–96% accuracy, 84–93% sensitivity, and 75–100% specificity.
The technology could be rolled out straightaway in principle, but would need to be approved for use in a clinical setting. However, Professor Hopgood is keen to work with an even bigger dataset to improve the accuracy of predictions, which is already above 80 percent.
"If we could attract funding, we would love to get together with other bowel cancer centres so we have access to even bigger datasets. With machine learning, the simple rule is the more data the better," he said. "Everyone I've spoken to in the health domain thinks that artificial intelligence will help them do a better job and we hope this research will do exactly that by providing more accurate predictions, the health service can allocate the best resources to each patient and improve patient care."
To access this paper, please click here