Could artificial intelligence help people lose weight with less risks?
Welcome to Health Connect, the podcast for health professionals where we will share the latest news and information on science and technology in the medical industry. In this episode we will talk about bariatric surgery and how artificial intelligence can help physicians to better identify those patients who are at higher risk of adverse events if subjected to this kind of intervention.
Did you know that the history of bariatric surgery dates back to 1953 with the jejunoileal bypass surgery?1
Today we will talk about bariatric interventions, which are commonly used in patients with obesity to help them lose weight. These surgeries have evolved to being key in treating the worldwide pandemic of morbid obesity, however, it is still difficult to predict the short- and long-term complication rates and the degree of resulting weight loss. Using artificial intelligence could help in this matter, for example, by predicting whether a patient will show sufficient weight loss after the surgery, since currently insufficient weight loss occurs in approximately 10 to 30 percent of patients. However, predicting such outcomes becomes even more complex considering that the influencing factors are extremely diverse, varying from socio-economic factors to a specific type of microbiome.2
Artificial intelligence, or AI, could be applied not only to the prediction of weight loss, but also to the diagnosis of conditions that may impose per-operative technical challenges, like the presence of hiatal hernias when performing laparoscopic sleeve gastrectomy. Additionally, it could help predict postoperative quality of life and complications with accuracies of up to 98%, as well as other variables relevant for decision-making.2
In this context, during this episode, we will discuss two studies focusing on bariatric surgeries, one on sleeve gastrectomy and the other on duodenal switch. Sleeve gastrectomy, which we will refer to only as gastrectomy from now on, is the most commonly performed bariatric procedure across the world and its popularity is a result of the excellent outcome in terms of its safety, the associated weight loss and the improvement of comorbidities in patients who undergo this procedure. However, like any other surgery, gastrectomy can have many adverse events, including staple line leak and bleeding, gastric stenosis, volvulus, and the very frequent gastroesophageal reflux disease, or GERD.3
Lately, being able to predict whether patients will develop GERD after gastrectomy has become an utmost necessity to help guide medical decisions, such as the selection of the optimal bariatric procedure for every patient. In fact, recent evidence demonstrates that the onset of GERD could be predicted. For example, it has been suggested that higher preoperative BMI is less likely to be associated with de novo or worsening GERD, whereas more severe heartburn on standing can be predictive of increased risks or worsening of the symptoms after gastrectomy. However, there is conflicting evidence on this topic and plenty of additional investigations are still needed, for instance, some studies even suggest that GERD symptoms improve postoperatively, while others support a refluxogenic nature of this surgery.3
In this context, the research led by Sameh Emile and Waleed Ghareeb, from the Mansoura University Hospitals and the Suez Canal University Hospital in Egypt, published a study in the Obesity Surgery journal that aimed to develop a model based on artificial intelligence to predict the onset of GERD after gastrectomy. The idea behind it was that, if GERD can be predicted before the gastrectomy is performed, it could guide clinicians to either select another less refluxogenic procedure, such as Roux-en-Y gastric bypass, or to counsel the patients about the need to do frequent upper gastrointestinal endoscopy after the gastrectomy, if they still want to have the procedure.3
To test this hypothesis, the researchers used data from 441 adult patients with severe obesity who underwent laparoscopic gastrectomy. The patients were examined postoperatively with a gastrointestinal endoscopy at six or 12 months after the surgery, and GERD symptoms were assessed with the GERD-Q score to classify patients into those with zero, 50, 79 and 89 percent likelihood of GERD. Those with 50 percent or higher probability, or with a history of GERD were investigated with endoscopy to establish it as single or multiple erosions. The data from all the patients were collected, including demographics, comorbidities, details of the surgical procedure, the degree of weight loss at six and 12 months, and whether the patient presented GERD after the procedure.3
After the surgery, BMI decreased significantly from a baseline of about 50.7 plus/minus 7.7 to 34.1 plus/minus 4.6 at 12 months, resulting in a total weight loss of about 30.4 percent in one year. An important finding was that, among the 91 patients who had GERD before the surgery, 50 showed improvement or resolution of the GERD symptoms after the gastrectomy; while the remaining 41 patients had persistent or worsening symptoms of GERD. On the other hand, among the patients who didn’t have GERD before the procedure, 46 developed it de novo after the intervention. Together, the patients with new onset and persistent GERD accounted for 19.7 percent of all the patients in the study.3
The next step was to identify the most important variables in the prediction of GERD after the surgery, using a model based on machine learning algorithms. To do this, first, all the variables were ranked and queued based on their importance to make the prediction. This process was intended to minimize the amount of data used in the model by choosing only the most important predictors or variables that give the best accuracy for the prediction process. After doing this, it was clear that the top five predictors or variables were: preoperative GERD, distance from the pylorus, orogastric tube size, intraoperative complication, and age; but other variables were also deemed important, and thus none was excluded.3
Next, the researchers used these variables to evaluate five models with 15 different algorithms. One of the models tested was the “ensemble” model, which was trained by three different algorithms, one of them, the “bagged” algorithm outperformed the other models. The ensemble model achieved an area under the ROC curve of 0.97 during the validation set, and a value of 0.93 on the test set; a sensitivity of 79.2 percent, and a specificity of 86.1 percent, when trying to predict the risk of developing GERD in patients who underwent gastrectomy.3
The next step was to determine the best cutoff values for the top variables. The researchers focused on the orogastric tube size and distance from the pylorus, because these were two modifiable predictors significantly correlated to the risk of developing GERD. They found that two models, ensemble and support vector machine, had a high accuracy to determine which orogastric tube size and distance from the pylorus are associated with the greatest risk of GERD after gastrectomy. In general, the cutoff values that better determined the risk of GERD after surgery were an age older than 42 years, weight greater than 140 kg, body mass index greater than 52 kg/m2, orogastric tube larger than 38 French units, and distance from the pylorus shorter than 3 cm.3
Considering these results, we can say that one of the main contributions of this study, compared to previous ones with similar goals that used statistical models; was that this one was based on machine learning and had an excellent accuracy. The top five variables that helped to predict GERD were the presence of baseline GERD, age, weight, size of the orogastric tube, and the distance of the first stapler firing from the pylorus. According to the researchers, these variables, when present, would predictably impact the likelihood of postoperative GERD substantially. This is relevant considering that there is still controversy about the risk factors linked with GERD after gastrectomy.3
Finally, these findings also provide cutoff values for age, weight, bougie size, and distance from the pylorus that may help surgeons tailor the modifiable technical factors to reduce the risk of GERD after gastrectomy in patients with non-modifiable risk factors, such as older age and heavier weight.3
CÁPSULA
In bariatric surgery, currently six procedures are commonly performed: the jejunoileal bypass, Roux-en-Y gastric bypass, vertical banded gastroplasty, biliopancreatic diversion with or without duodenal switch, adjustable gastric banding, and sleeve gastrectomy. Regardless of the surgery performed, several changes and metabolic benefits can be observed in patients undergoing bariatric surgery.1
Among the most important changes after these interventions are the eating habits, which are inevitable because patients are unable to overeat. There is also a strong effect on hormone signaling pathways, such as insulin, glucagon, ghrelin, and many others. Another immense benefit, and the most intuitive change after bariatric surgery, is the weight loss. In fact, it has been suggested that different degrees of weight loss impact gradually the metabolic function and the adipose tissue biology.1
Furthermore, bariatric surgery also decreases the rates of type 2 diabetes mellitus by targeting multiple aspects of the disease, such as improvement of glucose control, and can even lead to a remission of diabetes in 95 to 100 percent of patients. The cause was initially thought to be associated with the decrease in body weight, thus, theoretically, those who have the greatest decrease were considered to have the greatest improvement. However, more studies found that modifications in metabolism and hormones played a greater role in fighting diabetes after the intervention.1
Astonishingly, many patients with pre-existing cardiovascular disease can reduce or completely discontinue their cardiovascular medications after undergoing bariatric surgery. Gastrectomy can also control blood pressure in obese patients with a remission rate of hypertension between 60 to 70 percent in the year after the surgery and may even reach 90 percent in long-term follow-up.1
FIN DE CÁPSULA
Welcome again. In the previous section, we talked about artificial intelligence to predict the presence of GERD after sleeve gastrectomy. Now we will talk about the prediction of morbidity and mortality after duodenal switch, another intervention used in patients with obesity.4
Contrary to the sleeve gastrectomy that we discussed in the previous section, the duodenal switch is far less commonly performed as a weight loss procedure. Since it involves significant intestinal rearrangement, it can lead to more weight loss and even resolution of type 2 diabetes mellitus. However, duodenal switch is associated with more short- and long-term morbidities compared to the other procedures, including postoperative leak, malnutrition, chronic diarrhea, and vitamin deficiencies.4
Despite these morbidities, it is generally recommended for bariatric patients with a body-mass index in excess of 50 kg/m2, but the decision to proceed with the intervention is taken very seriously by the bariatric surgical team and the patient. Because of this, the potential identification of the patients who may be prone to these morbidities or mortality after duodenal switch is a difficult but highly desirable task.4
In this context, Eric Wise and colleagues, from the University of Minnesota in the United States, published in the Surgical Endoscopy journal a study that addressed this issue. They used a large dataset that contained information about the 30-day postoperative morbidity and mortality of patients who underwent duodenal switch. The dataset is called MBSAQIP, which collects clinical information from more than 800 bariatric centers in the USA.4
They examined the dataset to assess information from 2 907 patients undergoing the procedure, including age, BMI, comorbidities, and critical 30-day outcome measures, like the reoperation, readmission, reintervention, and death rates. Of the total population, 229 patients, corresponding to the 7.9 percent, presented at least one of the critical 30-day outcomes. A detailed analysis of these patients showed that of that 7.9 percent, about 3 percent corresponded to reoperations, 5.6 percent to readmissions, 2 percent to re-interventions and 0.4 percent to mortality, which was mainly caused by intra-abdominal sepsis, bleeding, and pulmonary embolism.4
Next, the researchers performed bivariate and multivariate analyses to identify factors associated with an occurrence of the composite endpoint. In the bivariate analysis, they found seven associated risk factors , which were advanced age, non-white race, any cardiac history, severe hypertension, previous obesity or foregut surgery, obstructive sleep apnea, decreased albumin, and increased creatinine. Of these factors, only severe hypertension, previous obesity or foregut surgery, and obstructive sleep apnea were significantly associated in the multivariate analysis with an area under the ROC curve of 0.619.4
Next, the seven significantly associated variables found in the bivariate analysis were used to test the artificial neural network model to evaluate their predictive ability in the identification of patients at risk of morbidity and mortality after surgery. The artificial intelligence model was generated using 80 percent of the patients randomly chosen for the training set, and the other 20 percent withheld for the validation set. After training, testing, and validating the artificial neural network model, the training set showed an area under the ROC curve of 0.656, and 0.685 for validation set. Proving that the use of this model optimized the interactions among these risk factors with good accuracy.4
In general, these results demonstrate that artificial intelligence has higher potential in predicting patient surgical outcomes compared to the traditional multivariate analysis, which is exemplified by their corresponding area under the ROC curves of 0.685 and 0.619, respectively. Furthermore, the artificial intelligence approach reveals the critical risk factors for a highly relevant endpoint to bariatricians and patients alike. Thus, the artificial intelligence approach can improve patient selection for the duodenal switch, a major operation with lifelong physiologic sequelae that should not be underestimated.4
In summary, these are some promising predictive capabilities of machine learning in bariatric surgery. However, to improve the predictive ability of these methods, there is still a need for additional data derived from large patient databases, laparoscopic surgery, or robotic surgeries.2
Thanks for joining us on this episode of Health Connect. Don’t miss out on our next episode. Don't forget to subscribe to discover the latest medical news.
Referências:
- 1. Ji Y, Lee H, Kaura S, Yip J, Sun H, Guan L, et al. Effect of Bariatric Surgery on Metabolic Diseases and Underlying Mechanisms. Biomolecules. 2021;11:1582. Available at: https://doi.org/10.3390/biom11111582
- 2. Bektaş M, Reiber BMM, Costa-Pereira J, Burchell GL, van der Peet DL. Artificial Intelligence in Bariatric Surgery: Current Status and Future Perspectives. Obes Surg. 2022;32:2772-2783. Available at: https://doi.org/10.1007/s11695-022-06146-1
- 3. Emile SH, Ghareeb W, Elfeki H, Sorogy ME, Fouad A, Elrefai M. Development and Validation of an Artificial Intelligence-Based Model to Predict Gastroesophageal Reflux Disease After Sleeve Gastrectomy. Obes Surg. 2022; 32:2537–2547. Available at: https://doi.org/10.1007/s11695-022-06112-x
- 4. Wise E, Leslie D, Amateau S, Hocking K, Scott A, Dutta N, et al. Prediction of thirty-day morbidity and mortality after duodenal switch using an artificial neural network. Surg Endosc. 2022. Available at: https://doi.org/10.1007/s00464-022-09378-5
Os links para todos os sites de terceiros são oferecidos como um serviço aos nossos visitantes e não implicam endosso, indicação ou recomendação do Health Connect. Os artigos vinculados são fornecidos apenas para fins informativos e não visam implicar uma atribuição pelo autor e/ou editor. O Health Connect se isenta de qualquer responsabilidade pelo conteúdo ou pelos serviços de outros sites. Recomendamos que você analise as políticas e condições de todos os sites que escolher acessar.
NON-2022-14821
NON-2023-2512