Improvement of Abdominal Fat Pad Biopsy Technique in Systemic Light Chain Amyloidosis and Establishment of an Outcome Prediction Tool with Machine Learning

Background: Abdominal fat pad aspiration biopsy is a widely accepted tool for confirming the diagnosis of systemic light chain amyloidosis (AL), but not all physicians are experienced in managing its indication or guarantee the quality of sample collection. To assist clinical operation and decision making, our team wanted to improve the abdominal fat pad biopsy operation, and meanwhile establish a tool for predicting the biopsy results through this work.

Methods: Basing on established literatures and our practice experience, we chose 20ml-syringes (standard matched with 18G-needles in our center) as the biopsy tools. The aspirations were conducted at the sites approximately 10cm lateral to the umbilicus as described in other papers. During the aspiration, the needle was pointed to three directions: namely superior- lateral, lateral, and inferior- lateral; also, the needle was rotated during the puncture so that the entrance of needle could reach difference surfaces of the abdominal fat tissue, which might be beneficial for biopsy outcomes. It is known that adequate sample volume is helpful for increasing the accuracy of outcomes, so we set the target sample volume as 1 ml, of which 0.7 ml was for Congo red staining and 0.3 ml for electron microscopy examination. Positive outcomes of either examination would be treated as biopsy positive. Basing on the collected outcome data, a random forest machine learning method was used to select parameters with high predictive power, and predictive tools were established by ROC analysis.

Results: During 2018.9-2021.3, a total of 109 cases of abdominal fat pad aspiration biopsy were performed in our center using the improved procedures mentioned above, with a success rate of 100% within 10 minutes and without adverse events requiring clinical intervention. Larger negative pressure and wider sampling range made it easier to obtain sufficient samples without increase of risk of bleeding or infection. A total of 82 cases with complete information were included for further analysis, with a biopsy positive rate of 57.3%. The random forest model reached an area under curve (AUC) of 0.873. Referring to the importance values (Figure 1) and the ten-fold cross-validation results, we selected two parameters, namely the ratio between involved and uninvolved free light chain (RFLC) and serum cardiac troponin T (cTnT), to establish the predictive tool. Referring to the results of ROC analysis, either RFLC>3 or cTnT>60ng/L was counted as 1-score. Among the 82 cases mentioned above, the biopsy positivity rate was 0% among those with 0 score and 71.2% among those with non-0 score (59.2% for 1 score and 79.5% for 2 score respectively). The AUC of this 2-score predictive tool was 0.793 (p<0.001, Figure 2) in ROC analysis.

Conclusion: The improved operational procedure can highly guarantee the success of fat pad aspirations and the sufficiency of sample volume for pathological examinations. When encounter patients with suspicious AL, the predictive tool with RFLC plus cTnT can be helpful for clinical decision on abdominal fat pad aspiration biopsy. If neither of the parameters reach the cut-off points, a positive outcome of fat biopsy is unlikely to emerge, and other involved organs would be preferable biopsy sites.

No relevant conflicts of interest to declare.