Patients with Idiopathic Membranous Nephropathy: A Real-World Clinical and Economic Analysis of U.S. Claims Data

IDA-MIL: Classification of Glomerular with Spike-like Projections via Multiple Instance Learning with Instance-level Data Augmentation

BACKGROUND AND OBJECTIVE

Tiny spike-like projections on the basement membrane of glomeruli are the main pathological feature of membranous nephropathy at stage II (MN II), which is the most significant stage for the diagnosis and treatment of renal disease. Pathological technology is the gold standard in the diagnosis of spike-like and other MNs, and automatic classification of spike-like projection is a crucial step in assisting pathologists in their diagnosis. However, owing to hard-to-label spile-like projections and the scarcity of patient data, classification of glomeruli with spike-like projections based on supervised learning methods is a challenging task.

METHOD

To overcome the aforementioned problems, the idea of integrating weakly-supervised learning and data augmentation methods is utilized in designing the classification framework. Specifically, a multiple instance learning with instance-level data augmentation (IDA-MIL) method for the classification of glomeruli with spike-like projections is established in this paper. The proposed classification framework first trains the MIL model on a dataset with image-level labels, and the well-trained MIL model is used to extract instances that include spike-like projections in the whole glomerular image. Then, rather than using an image-level generative adversarial network (ImgGAN), an instance-level generative adversarial network (InsGAN) based on the StyleGAN2-ADA model is trained on the spike-like instances obtained by the MIL model and synthesizes new spike-like projection instances. Finally, the synthesized spike-like instances are extended to the training dataset to further improve the classification performance of MIL.

RESULTS

The designed IDA-MIL model is verified and evaluated from two aspects based on the in-house dataset. On the one hand, the performance comparisons between InsGAN and ImgGAN on five metrics, which involve FID, KID, Precision, Recall and IS, show that InsGAN obtains a better score and can synthesize effective spike-like projections. However, the proposed IDA-MIL model achieves the best classification performance with an accuracy of 0.9405. Then, to make nephrologists believe the inference result of the proposed model, we use heatmap technology to visualize the basis of the model inferences and show the top 4 probability spike-like instances per glomerulus. Furthermore, we analyze the correlation between the disease and the proportion of spike-like instances in bags from historical cases.

CONCLUSION

Compared with the ImgGAN, the InsGAN can synthesize natural and varied spike-like projections, which results in the classification performance of the MIL model achieving great improvement by adding synthesized instance samples into the training dataset. The heatmap of spike-like inferences and the proportion of spike-like instances can help nephrologists to make a preliminary reliable diagnosis in clinical practice. This work provides a valuable reference for medical image classification with limited data and small-scale lesions based on deep learning.

Acute kidney injury in idiopathic membranous nephropathy with nephrotic syndrome

BACKGROUND AND OBJECTIVES

The impact of acute kidney injury (AKI) on the progression of renal function in idiopathic membranous nephropathy (iMN) with nephrotic syndrome (NS) patients have not yet been reported, we sought to investigate the incidence, clinical features and prognosis of AKI in iMN with NS patients and determine clinical predictors for progression from AKI to advanced chronic kidney disease (CKD) stage.

METHODS

We analyzed clinical and pathological data of iMN with NS patients retrospectively collected from Jan 2012 to Dec 2018. The primary renal endpoint was defined as persistent eGFR <45ml/min per 1.73 m² more than 3 months. Comparisons of survival without primary renal endpoint were performed by Kaplan-Meier curves and log-rank test. Univariate and multivariate Cox proportional hazard models were constructed to determine independent variables associated with primary renal endpoint .

RESULTS

434 iMN with NS patients were enrolled. The incidence of AKI 1 stage, AKI 2 stage and AKI 3 stage was 23.1, 4.8 and 0.7% respectively. 66 (53.2%) patients with AKI had complete renal function recovery and 42 (33.9%) patients with AKI reached primary renal endpoint. Survival without primary renal endpoint was worse in AKI patients than No AKI patients (67.1 ± 5.3 and 43.7 ± 7.3% vs 99.5 ± 0.5 and 92.5 ± 4.2% at 2 and 4 years,p < 0.001). AKI was independently associated with primary renal endpoint, with an adjusted hazard ratio(HR) of 25.1 (95%CI 7.7-82.1, p < 0.001).

CONCLUSIONS

AKI was usually mild and overlooked in iMN patients with NS, but it had a strong association with poor clinical outcomes and was an independent risk factor for CKD progression.

Identifying Patients with Rare Disease Using Electronic Health Record Data: The Kaiser Permanente Southern California Membranous Nephropathy Cohort

INTRODUCTION

Developing a reliable means to identify and study real-world populations of patients with membranous nephropathy (MN) using electronic health records (EHRs) would help advance glomerular disease research. Identifying MN cases using EHRs is limited by the need for manual reviews of biopsy reports.

OBJECTIVE

To evaluate the accuracy of identifying patients with biopsy-proven MN using the EHR in a large, diverse population of an integrated health system.

METHODS

A retrospective cohort study was performed between June 28, 1999, and June 25, 2015, among patients with kidney biopsy results (N = 4723), which were manually reviewed and designated as MN or non-MN. The sensitivity, specificity, and positive predictive value (PPV) of International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes were determined using 2 approaches: 1) clinical (MN-specific codes 581.1, 582.1, or 583.1) and 2) agnostic/data-derived (codes selected from supervised learning at the highest predictive performance).

RESULTS

One year after biopsy, the sensitivity and specificity of an MN diagnosis were 86% and 76%, respectively, but the PPV was 26%. The data-driven approach detected that using only 2 codes (581.1 or 583.1) improved specificity to 94% and PPV to 58%, with a small decrease in sensitivity to 83%. When any code was reported at least 3 times, specificity was 98%; PPV, 78%; and sensitivity, 64%.

DISCUSSION

Our findings suggest that ICD-9 diagnosis codes might be a convenient tool to identify patients with MN using EHR and/or administrative claims information. Codes selected from supervised learning achieved better overall performance, suggesting the potential of developing data-driven methods.