Characterization of the Pathogenesis of Lymphangioleiomyomatosis (LAM)

National Heart, Lung, And Blood Institute

Investigators

Linked publications & trials

Abstract

To obtain a better understanding of the factors that may have a role in the progression of patients with LAM, we used our longitudinal study of patients with LAM or LAM in the setting of TSC. 721 female LAM patients were seen at the NIH for 5556 visits, over more than 25 years (protocols 95-H-0186 and 96-H-0100). We derived clusters of patients via machine learning using easily obtainable baseline clinical characteristics. Clustering was performed by k-means and hierarchical methods with the following factors: age at diagnosis; mTORi use, BMI, menopause status, percent-predicted FEV1, FVC, and DLCO at first visit; and history of TSC, lymphatic involvement, lymphadenopathy, AML, and pneumothorax. The optimal number of clusters was three, based on within-cluster sums of squares and gap statistics, but the size of Cluster 1 was significantly smaller than the other two (Cluster 1: 27 patients; Cluster 2: 306 patients; Cluster 3: 388 patients). Because of the small size of Cluster 1 and since all the patients of Cluster 1 would map to Cluster 2 if only two clusters were determined, analysis was performed comparing characteristics of combined Clusters 1 and 2 versus those of Cluster 3. Cluster 1/2 patients were older, with sporadic LAM (S-LAM) or LAM/TSC and more frequent bilateral AMLs than Cluster 3 patients, who were premenopausal at baseline, with S-LAM and more frequent LLMs. Once cluster assignment was performed using first visit data, pulmonary function outcomes were analyzed prospectively, using mixed effects models adjusted for initial pulmonary function, follow-up time, and/or mTORi use. Without mTORi treatment, Cluster 3 patients declined faster in FEV1 than Cluster 1/2 patients, while during mTORi treatment, Cluster 3 patients declined faster in both FEV1 and DLCO than Cluster 1/2 patients. Importantly, Cluster 3 patients continued declining in pulmonary function even with mTORi treatment, albeit at a significantly slower rate. Cluster 3 postmenopausal patients showed reduced effectiveness of mTORi treatment. We suggest that two types of S-LAM patients exist, with the Cluster 1/2 S-LAM patients displaying an intermediate phenotype between the Cluster 1/2 LAM/TSC patients and the Cluster 3 S-LAM patients. More Cluster 3 S-LAM patients had a history of pneumothorax than Cluster 1/2 S-LAM patients and were more likely to be premenopausal at baseline. Interestingly, Cluster 1/2 S-LAM patients had slower rates of decline of FEV1 than Cluster 3 S-LAM patients. The rates of decline of DLCO between Cluster1/2 S-LAM patients and Cluster 3 S-LAM patients were not significantly different for patients not taking mTORi, but were significantly faster for Cluster 3 S-LAM patients versus Cluster 1/2 S-LAM patients on mTORi. The Cluster 1/2 S-LAM patients may have clinical mosaic LAM/TSC. Using baseline clinical characteristics from our large retrospective study of LAM patients, we derived patient clusters that differed significantly in pulmonary function decline and mTORi response. Our study is the first to cluster patients leading to groups showing differences in pulmonary function decline and mTORi response. These findings highlight the importance of baseline clinical characteristics in understanding LAM progression and treatment, with the goal of optimizing and personalizing treatment based on a profile which could be established in any clinical setting. While it may be difficult to sort an individual patient into a cluster currently, further comparison of the patients of Cluster 1/2 and Cluster 3 with regard to differences in other clinical tests or molecular differences may give better understanding of the pathophysiology of LAM,and eventually lead to individualized LAM treatment. In this study, machine learning was employed using features from the patients’ first visits. Once clusters were determined, these features and others were compared between the clusters to describe them. Pulmonary function and survival of the clusters were analyzed prospectively from the first visit. The analyses were exploratory in nature as we attempt to identify differences in the clusters that would suggest future studies. This study highlights the potential for phenotype clustering to guide clinical decisions, suggesting that mTORi treatment may not be effective for all patient groups and identifying the need for tailored surveillance for those with clinical mosaic LAM/TSC. mTORi treatment for pulmonary function decline in postmenopausal Cluster 3 patients may not be effective and may lead to harmful side effects.