BACKGROUND Gastric cancer(GC)is the fourth leading cause of cancer-related deaths worldwide.Diagnosis relies on histopathology and the number of endoscopies is increasing.Helicobacter pylori(H.pylori)infection is a ma...BACKGROUND Gastric cancer(GC)is the fourth leading cause of cancer-related deaths worldwide.Diagnosis relies on histopathology and the number of endoscopies is increasing.Helicobacter pylori(H.pylori)infection is a major risk factor.AIM To develop an in-silico GC prediction model to reduce the number of diagnostic surgical procedures.The meta-data of patients with gastroduodenal symptoms,risk factors associated with GC,and H.pylori infection status from Holy Family Hospital Rawalpindi,Pakistan,were used with machine learning.METHODS A cohort of 341 patients was divided into three groups[normal gastric mucosa(NGM),gastroduodenal diseases(GDD),and GC].Information associated with socioeconomic and demographic conditions and GC risk factors was collected using a questionnaire.H.pylori infection status was determined based on urea breath test.The association of these factors and histopathological grades was assessed statistically.K-Nearest Neighbors and Random Forest(RF)machine learning models were tested.RESULTS This study reported an overall frequency of 64.2%(219/341)of H.pylori infection among enrolled subjects.It was higher in GC(74.2%,23/31)as compared to NGM and GDD and higher in males(54.3%,119/219)as compared to females.More abdominal pain(72.4%,247/341)was observed than other clinical symptoms including vomiting,bloating,acid reflux and heartburn.The majority of the GC patients experienced symptoms of vomiting(91%,20/22)with abdominal pain(100%,22/22).The multinomial logistic regression model was statistically significant and correctly classified 80%of the GDD/GC cases.Age,income level,vomiting,bloating and medication had significant association with GDD and GC.A dynamic RF GC-predictive model was developed,which achieved>80%test accuracy.CONCLUSION GC risk factors were incorporated into a computer model to predict the likelihood of developing GC with high sensitivity and specificity.The model is dynamic and will be further improved and validated by including new data in future research studies.Its use may reduce unnecessary endoscopic procedures.It is freely available.展开更多
文摘BACKGROUND Gastric cancer(GC)is the fourth leading cause of cancer-related deaths worldwide.Diagnosis relies on histopathology and the number of endoscopies is increasing.Helicobacter pylori(H.pylori)infection is a major risk factor.AIM To develop an in-silico GC prediction model to reduce the number of diagnostic surgical procedures.The meta-data of patients with gastroduodenal symptoms,risk factors associated with GC,and H.pylori infection status from Holy Family Hospital Rawalpindi,Pakistan,were used with machine learning.METHODS A cohort of 341 patients was divided into three groups[normal gastric mucosa(NGM),gastroduodenal diseases(GDD),and GC].Information associated with socioeconomic and demographic conditions and GC risk factors was collected using a questionnaire.H.pylori infection status was determined based on urea breath test.The association of these factors and histopathological grades was assessed statistically.K-Nearest Neighbors and Random Forest(RF)machine learning models were tested.RESULTS This study reported an overall frequency of 64.2%(219/341)of H.pylori infection among enrolled subjects.It was higher in GC(74.2%,23/31)as compared to NGM and GDD and higher in males(54.3%,119/219)as compared to females.More abdominal pain(72.4%,247/341)was observed than other clinical symptoms including vomiting,bloating,acid reflux and heartburn.The majority of the GC patients experienced symptoms of vomiting(91%,20/22)with abdominal pain(100%,22/22).The multinomial logistic regression model was statistically significant and correctly classified 80%of the GDD/GC cases.Age,income level,vomiting,bloating and medication had significant association with GDD and GC.A dynamic RF GC-predictive model was developed,which achieved>80%test accuracy.CONCLUSION GC risk factors were incorporated into a computer model to predict the likelihood of developing GC with high sensitivity and specificity.The model is dynamic and will be further improved and validated by including new data in future research studies.Its use may reduce unnecessary endoscopic procedures.It is freely available.
