| BackgroundDue to the rapid development of artificial intelligence and the explosion of medicine data,using machine learning algorithms to develop disease-aided diagnosis models had become more and more widely used in medical practice and had achieved fruitful achievements and has huge application potential.However,the present diagnosis models had certain limitations in their application because most of them are displayed in calculation formulas or graphs,which are limited to professional groups such as doctors,and cannot widely cover target patients because of their low degree of sharing.Especially for diseases with low incidence,insidious onset,few early screening methods,and high technical requirements,building auxiliary diagnosis models and establish self-service or semi-self-service screening mini-programs to achieve more target patients,is of practical significance in improving diagnosis of diseases,achieving early and correct diagnosis and early intervention and treatment.The prevalence of primary aldosteronism(PA)varied from 5%to 20%in patient with hypertension.Due to the high difficulty of sample collection for laboratory diagnosis,quality control and the cumbersome testing,it is easy to confused with essential hypertension and cannot be timely and accurately diagnosed.Therefore,early screening methods cannot be widely used in all hypertensive patients,resulting in a large underdiagnosed,which greatly hinders the prevention and control of the disease.Hypokalemia is the most common and typical clinical symptom of primary aldosteronism,and the related risk factors or primary aldosteronism-dependent hypokalemia are still unclear.Moreover,the detection of serum potassium in easily interfered by a variety of factors,and hypokalemia in easily masked.Early clinical identification of patients with hypokalemia is of great important.ObjectivesAn artificial intelligence-based online auxiliary diagnosis model was designed and constructed for the auxiliary diagnosis of primary aldosteronism,taking primary aldosteronism as the target disease.At the same time,an online risk assessment model for primary aldosteronism-dependent hypokalemia was designed and established based on the construction process and mothed of the above-mentioned online model.disease.Method1.Construction of online auxiliary diagnosis model:1314 patients with hypertension were firstly included in this study,including 824 patients with primary hypertension and 490 patients with primary aldosteronism;the clinical characteristics and laboratory test data of the study population were collected for modeling;patients were randomly divided into a training cohort(919 cases,70%)and an internal validation cohort(395 cases,30%)in a 7:3 ratio.Univariate logistic regression and multivariate logistic regression analyses were used in the training cohort to identify independent risk factors for primary aldosteronism and modeled using a logistic regression algorithm in machine learning algorithms;model validation was performed using internal validation cohort,in the meanwhile an external dataset(n=285)was used for model external validation.Sensitivity,specificity,accuracy,etc.were used to evaluated the performance of the model.2.Construction of online risk assessment model:490 patients with PA were enrolled in this study and randomly divided into the training and validation cohort in a ratio of 7:3.Clinical and laboratory characteristics were collected and predicted factors were identified by univariate and multivariate logistic analysis.Logistic regression algorithm in machine learning methods was applied to develop a prediction model.Model validation was performed using internal validation cohort.Sensitivity,specificity,accuracy,etc.were used to evaluated the performance of the model.3.Finally,the model was uploaded to the personal host server and cloud server and provide the corresponding access link for sharing(Online auxiliary diagnosis model:http://127.0.0.1:5341 and https://dxonline.deepwise.com/prediction/index.html?baseUrl=%2Fapi%2F&id=5244&topicName=undefined&from=share;Online risk assessment model:http://127.0.0.1:5213 and https://dxonline.deepwise.com/prediction/index.html?baseUrl=%2Fapi%2F&id=6304&topicName=undefined&from=share).Results1.Seven independent risk factors for predicting PA were identified,including age,sex,hypokalemia,serum Sodium,serum Sodium-to-Potassium ratio,Anion gap and alkaline urine.The prediction model comprised of all these seven predictors was developed and showed sufficient predictive accuracy,with an area under the receiver operating curve(AUC)of 0.839(95%CI:0.81-0.87)in the training cohort and an AUC of 0.814(95%CI:0.77-0.86)in the internal validation cohort and an AUC of 0.839(95%CI:0.79-0.89)in the external validation cohort.The calibration curves exhibited well agreement between the model predictive results and the actual risk.Finally,the model was uploaded to the personal host server and cloud server and provide the corresponding access link for sharing.2.Five risk predicted factors including diastolic blood pressure,serum sodium,chloride,calcium and alkaline urine were identified and prediction model was developed based on machine learning algorithm.The ROC curves showed high performances with an AUC of 0.835(95%CI:0.793-0.878)and 0.865(95%CI:0.800-0.920),and an accuracy of 0.75 and 0.78 in training and validation cohort respectively,and the calibration curve presented well agreement between the model and the actual observation results.Finally,the model was shared online through personal host server and cloud server.ConclusionsTaking primary aldosteronism as an example,this study designed and constructed a complete set of processes and methods for online auxiliary diagnosis models of the disease and online risk assessment model.The establishment and improvement of the processes and methods can be widely used in the development and application of online auxiliary diagnosis systems for various diseases,so as to cover more target groups and improve the detection rate of diseases.At the same time,the effective combination of laboratory diagnosis and machine learning algorithms will be more conducive to the realization of precision medicine. |
PDF Full Text Request