Establishment Of Prognosis Prediction Model And Evaluation Of Immunosuppression Treatment Pattern For Patients With IgA Nephropathy Based On Machine Learning

IgA nephropathy(IgAN)is the most common primary glomerulonephritis and an important cause of end-stage kidney disease(ESKD)worldwide,especially in Asian regions.Patients with IgAN have a wide range of clinical presentations,a variety of histologic lesions,and highly heterogeneous phenotypes.The prognosis for IgAN is poor,with up to 30% to 40% of patients developing ESKD within 10 to 25 years.Stratifying the risk for disease progression according to clinical and pathologic manifestations of patients with IgAN and predicting the longterm kidney prognosis accurately are of great clinical significance for guiding not only individualized treatment and management of patients,but also related clinical trials.Although IgAN is considered an immune-mediated disease,which IgAN patients will benefit from immunosuppression treatment and how to identify these patients remain controversial.The establishment of accurate prediction models and the identification of patient subgroups with differential responses to treatment through machine learning has begun to be applied in medicine.Our study aimed to use machine learning to analyze the multi-center baseline and long-term follow-up data of IgAN patients from China to build a prognostic prediction and risk stratification model that combines clinical and pathologic variables to assist physicians in predicting kidney prognosis quickly and accurately.Meanwhile,the study identified and externally validated subgroups of patients benefitting from immunosuppression therapy based on a broad spectrum of clinical and pathological features in IgAN patients using machine learning method, provides basis for individualized therapy in IgAN patients.Part One Establishment of prognosis prediction model for patients with IgA nephropathy based on machine learningAim: Establishing prediction model to predict long-term outcomes and stratify risk of IgAN patients based on machine learning for clinical decision making and future clinical trials designing.Methods: Multicenter cohort of 2047 patients with biopsy-proven IgAN from 18 renal centers in China was retrospectively analyzed.Derivation cohort was composed of 1022 patients from a single center.Validation cohort was composed of 1025 patients from 18 centers.A total of 36 characteristics,including demographic,clinical,and pathologic variables were included.The study outcome was defined as combined outcome of ESKD or 50% reduction in estimated glomerular filtration rate(e GFR)within 5 years after diagnostic kidney biopsy.A gradient tree boosting method implemented in the e Xtreme Gradient Boosting(XGBoost)system was used to select the 10 most important variables from 36 candidate variables.Stepwise Cox regression analysis was used to derive a simplified scoring scale model(SSM)based on these 10 variables.Model discrimination and calibration were assessed using the C statistic and Hosmer-Lemeshow test.Risk stratification of the SSM was evaluated using Kaplan-Meier analysis.SHapley Additive ex Planations(SHAP)method was used to explain the XGBoost prediction results.Results: 1.In the derivation and validation cohorts,74 and 114 patients reached the combined outcome,respectively.2.A variety of machine learning and statistical models were established.XGBoost predicted the outcome with C statistics of 0.89(95% confidence interval [CI],0.87-0.94)and 0.84(95% CI,0.80-0.88)for the derivation and the validation cohort,respectively.XGBoost reached the highest C statistic for the validation cohort among machine learning and statistical models,and was selected as the final accurate prediction model.3.XGBoost provided the importance score of each variable,the top 10 important variables were tubular atrophy/interstitial fibrosis(importance score =0.156),serum albumin(0.125),global sclerosis(0.109),hypertension(0.078),serum uric acid(0.063),microscopic hematuria(0.063),age at biopsy(0.063),urine protein(0.063),mean mesangial score(0.047),and serum creatinine(0.047).4.In order to further improve the practicability of the model,the SSM was built,which included 3 variables: urine protein excretion,global sclerosis,and tubular atrophy/interstitial fibrosis(T).The cut-off values were obtained by variable discretization and the risk scores of variables were derived from the coefficient of the regression model: T1,global sclerosis>25% and urine protein>1 g/24 h is 1 point and T2 is 2 point.Using Kaplan-Meier analysis,the SSM identified significant risk stratification(P < 0.001).5.We incorporated the XGBoost prediction model and SSM into the Nanjing IgAN risk stratification system which was posted on a web page,and the model can be accessed using a web-based calculator.6.The impact of the predictive variables on the outcome were obtained from the SHAP transformation results.Conclusions: A prediction system,the Nanjing IgAN risk stratification system,including an accurate XGBoost prediction model and a simplified SSM for risk stratification,using routinely available characteristics and based on the combination of a machine learning algorithm and survival analysis was built in the study.An online calculator permits easy implementation of this model.Through the Nanjing IgAN risk stratification system,we can accurately predict the long-term renal prognosis of IgAN patients,provide a basis for screening specific patient groups in clinical trials,and assist stratified prognosis management and graded diagnosis and treatment,so as to improve the level of individualized treatment and management of IgAN patients.Part Two Identification and validation of IgA nephropathy patients benefiting from immunosuppression therapy based on machine learningAim: This study employed machine learning methods to identify and validate subgroups of IgAN patients who may benefit from immunosuppression therapy.Methods: Clinical and pathological data from 4047 biopsy-proven IgAN patients from 24 renal centers in China were retrospectively included.The derivation and validation cohorts were composed of 2058 patients from 18 renal centers and 1989 patients from 7 centers,respectively.Model-based recursive partitioning,a machine learning approach,was performed to partition patients in the derivation cohort into subgroups with different immunosuppression long-term benefits,associated with time to ESKD,measured by adjusted hazard ratio(HR)using Cox regression and adjusted KaplanMeier estimator.Results: 1.In the derivation cohort,immunosuppression-treated patients had higher proteinuria,heavier renal histological active lesions compared with those without immunosuppression.More immunosuppression-treated patients received antihypertensive drugs and statins during follow-up,with slower e GFR decline,while did not show lower ESKD incidence.No difference in kidney survival curves was observed between treated and untreated groups.2.Based on the partition decision tree,the long-term benefits of immunosuppression therapy in patients with internal nodes and leaf nodes were evaluated,and confounding factors were adjusted.Three identified subgroups obtained significant immunosuppression benefits with HRs ? 1.In patients with serum creatinine ?1.437 mg/d L,the benefits of immunosuppression were observed in those with proteinuria >1.525 g/24h(node 6;HR =0.50;95% CI,0.29-0.89;P =0.02),especially in those with proteinuria > 2.480 g/24h(node 8;HR =0.23;95% CI,0.11-0.50;P <0.001).In patients with serum creatinine >1.437 mg/d L,those with high proteinuria and crescents benefitted from immunosuppression(node 12;HR =0.29;95% CI,0.09-0.94;P =0.04).3.The long-term benefits were also internally validated using a bootstrap analysis,where the same immunosuppression treatment benefits were observed in each of the benefit nodes.4.Sensitivity analyses were performed to assess the robustness of the findings using 4 different imputation methods,and the results from the benefit nodes were consistent with our findings.5.In the three benefit subgroups,immunosuppression treatment reduced risk of all three secondary outcomes(30%,40% and 50% decline in e GFR),providing indirect evidence to our conclusions.6.In the validation cohort,the long-term treatment benefits were externally validated in every subgroup.Conclusion: Comprehensive consideration of renal function,proteinuria and renal histological characteristics would serve as indicators for the selection of immunosuppression therapy in IgAN patients.The identification of subgroups benefitting from immunosuppression promotes decision-making,assists targeted clinical trial design,and provides basis for individualized treatment of IgAN patients,so as to shed light on precision medicine for kidney diseases.