Risk assessment for intra-abdominal injury following blunt trauma in children: Derivation and validation of a machine learning model. Pennell Christopher,Polet Conner,Arthur L Grier,Grewal Harsh,Aronoff Stephen The journal of trauma and acute care surgery BACKGROUND:Computed tomography is the criterion standard for diagnosing intra-abdominal injury (IAI) but is expensive and risks radiation exposure. The Pediatric Emergency Care Applied Research Network (PECARN) model identifies children at low risk of IAI requiring intervention (IAI-I) in whom computed tomography may be omitted but does not provide an individualized risk assessment to positively predict IAI-I. We sought to apply machine learning algorithms to the PECARN blunt abdominal trauma (BAT) data set experimentally to create models for predicting both the presence and absence of IAI-I for pediatric BAT victims. METHODS:Using the PECARN data set, we derived and validated predictive models for IAI-I. The data set was divided into derivation (n = 7,940) and validation (n = 4,089) subsets. Six algorithms were tested to create 2 models using 19 clinical variables including emesis, dyspnea, Glasgow Coma Scale score of <15, visible thoracic or abdominal trauma, seatbelt sign, abdominal distension, tenderness or rectal bleeding, peritoneal signs, absent bowel sounds, flank pain, pelvic pain or instability, sex, age, heart rate, and respiratory rate (RR). Five algorithms were fitted to predict the absence (low-risk model) or presence (high-risk model) of IAI-I. Models were validated using the test subset. RESULTS:For the low-risk model, four algorithms were significantly better than the baseline rate (2.28%) when validated using the test set. The random forest model identified 73% of children as low risk, having a predicted IAI-I rate of 0.54%. For the high-risk model, all six algorithms had added predictive power compared with the baseline rate with the highest reportable risk being 39.0%. By incorporating both models into a web application, child-specific risks of IAI-I can be estimated ranging from 0.28% to 39.0% CONCLUSION: We developed a tool that provides a child-specific risk estimate for IAI-I after BAT. This publically available model provides a powerful tool for clinicians triaging pediatric victims of blunt abdominal trauma. LEVEL OF EVIDENCE:Prognostic, Level II. 10.1097/TA.0000000000002717

Ontology-based venous thromboembolism risk assessment model developing from medical records. BMC medical informatics and decision making BACKGROUND:Padua linear model is widely used for the risk assessment of venous thromboembolism (VTE), a common but preventable complication for inpatients. However, genetic and environmental differences between Western and Chinese population limit the validity of Padua model in Chinese patients. Medical records which contain rich information about disease progression, are useful in mining new risk factors related to Chinese VTE patients. Furthermore, machine learning (ML) methods provide new opportunities to build precise risk prediction model by automatic selection of risk factors based on original medical records. METHODS:Medical records of 3,106 inpatients including 224 VTE patients were collected and various types of ontologies were integrated to parse unstructured text. A workflow of ontology-based VTE risk prediction model, that combines natural language processing (NLP) and machine learning (ML) technologies, was proposed. Firstly ontology terms were extracted from medical records, then sorted according to their calculated weights. Next importance of each term in the unit of section was evaluated and finally a ML model was built based on a subset of terms. Four ML methods were tested, and the best model was decided by comparing area under the receiver operating characteristic curve (AUROC). RESULTS:Medical records were first split into different sections and subsequently, terms from each section were sorted by their weights calculated by multiple types of information. Greedy selection algorithm was used to obtain significant sections and terms. Top terms in each section were selected to construct patients' distributed representations by word embedding technique. Using top 300 terms of two important sections, namely the 'Progress Note' section and 'Admitting Diagnosis' section, the model showed relatively better predictive performance. Then ML model which utilizes a subset of terms from two sections, about 110 terms, achieved the best AUC score, of 0.973 ± 0.006, which was significantly better compared to the Padua's performance of 0.791 ± 0.022. Terms found by the model showed their potential to help clinicians explore new risk factors. CONCLUSIONS:In this study, a new VTE risk assessment model based on ontologies extraction from raw medical records is developed and its performance is verified on real clinical dataset. Results of selected terms can help clinicians to discover meaningful risk factors. 10.1186/s12911-019-0856-2

[Risk factor analysis on anastomotic leakage after laparoscopic surgery in rectal cancer patient with neoadjuvant therapy and establishment of a nomogram prediction model]. Jiang W,Feng M Y,Dong X Y,Dong S M,Zheng J X,Liu X M,Liu W J,Yan J Zhonghua wei chang wai ke za zhi = Chinese journal of gastrointestinal surgery To investigate the risk factors of anastomotic leakage (AL) after laparoscopic surgery in rectal cancer patient with neoadjuvant therapy and construct a nomogram prediction model. This study was a retrospective case-control study that collected and reviewed the clinicopathological data of 359 patients who underwent laparoscopic surgery from January 2012 to January 2018, including 202 patients from the Department of General Surgery, Nanfang Hospital of Southern Medical University and 157 patients from the Department of Gastrointestinal Surgery of Fujian Provincial Cancer Hospital. Inclusion criteria: (1) age ≥ 18 years old; (2) diagnosis as rectal cancer by biopsy before treatment; (3) distance from tumor to anus within 12 cm; (4) locally advanced stage (T3-T4 or N+) diagnosed by imaging (CT, MRI, PET or ultrasound); (5) standardized neoadjuvant therapy followed by laparoscopic radical operation. Exclusion criteria: (1) previous history of colorectal cancer surgery; (2) short-term or incomplete standardized neoadjuvant therapy; (3) Miles, Hartmann, emergency surgery, palliative resection; (4) conversion to open surgery. Clinicopathological data, including age, gender, body mass index (BMI), preoperative albumin, distance from tumor to anus, operation hospital, American Society of Anesthesiologists score (ASA score), operation time, T stage, N stage, M stage, TNM stage, pathological complete response (pCR) were analyzed with univariate analysis to identify predictors for AL after laparoscopic surgery in rectal cancer patient with neoadjuvant therapy. Then, incorporated predictors of AL, which were screened by multivariate logistic regression, were plotted by the "rms" package in R software to establish a nomogram model. According to the scale of the nomogram of each risk factor, the total score could be obtained by adding each single score, then the corresponding probability of postoperative AL could be acquired. The area under ROC curve (AUC) was used to evaluate the predictive ability of each risk factor and nomogram on model. AUC > 0.75 indicated that the model had good predictive ability. The Bootstrap method (1000 bootstrapping resamples) was applied as internal verification to show the robustness of the model. The discrimination of the nomogram was determined by calculating the average consistency index (C-index) whose rage was 0.5 to 1.0. Higher C-index indicated better consistency with actual risk. The calibration curve was used to assess the calibration of prediction model. The Hosmer-Lemeshow test yielding a non-significant statistic (>0.05) suggested no departure from the perfect fit. Of 359 cases, 224 were male, 135 were female, 189 were ≥ 55 years old, 98 had a BMI > 24 kg/m(2), 176 had preoperative albumin ≤ 40 g/L, 128 had distance from tumor to anus ≤ 5 cm, 257 were TNM 0-II stage, 102 were TNM III-IV stage, and 84 achieved pCR after neoadjuvant therapy. The incidence of postoperative AL was 9.5% (34/359). Univariate analysis showed that gender, preoperative albumin and distance from tumor to the anus were associated with postoperative AL (All <0.05). Multivariate logistic regression analysis revealed that male (OR=2.480, 95% CI: 1.012-6.077, =0.047), preoperative albumin ≤40 g/L (OR=5.319, 95% CI: 2.106-13.433, <0.001) and distance from tumor to anus ≤ 5 cm (OR=4.339, 95% CI: 1.990-9.458, <0.001) were significant independent risk factors for postoperative AL. According to these results, a nomogram prediction model was constructed. The male was for 55 points, the preoperative albumin ≤ 40 g/L was for 100 points, and the distance from tumor to the anus ≤ 5 cm was for 88 points. Adding all the points of each risk factor, the corresponding probability of total score would indicated the morbidity of postoperative AL predicted by this nomogram modal. The AUC of the nomogram was 0.792 (95% CI: 0.729-0.856), and the C-index was 0.792 after internal verification. The calibration curve showed that the predictive results were well correlated with the actual results (=0.562). Male, preoperative albumin ≤ 40 g/L and distance from tumor to the anus ≤ 5 cm are independent risk factors for AL after laparoscopic surgery in rectal cancer patient with neoadjuvant therapy. The nomogram prediction model is helpful to predict the probability of AL after surgery. 10.3760/cma.j.issn.1671-0274.2019.08.009