PS-C17-3: ONE-YEAR TIME-SERIES OBSERVATIONS OF VARIABLE COMPONENTS OF URINE PROTEIN AND EGFR PREDICT WORSENING RENAL FUNCTION IN TYPE 2 DIABETES. (January 2023)

• Record Type: Journal Article
• Authors: Watanabe, Mari; Meguro, Shu; Kimura, Kaiken; Furukoshi, Michiaki; Masuda, Tsuyoshi; Enomoto, Makoto; Itoh, Hiroshi
• Is Part Of: Journal of hypertension. Volume 41(2023)Supplement 1
• Page Start: e381
• Publication Date: 2023-01
• Subjects: Hypertension -- Periodicals; Hypertension -- Periodicals; 616.132005
• DOI: 10.1097/01.hjh.0000916712.21929.4f
• Languages: English
• ISSNs: 1473-5598

Abstract

Abstract : Objective: Type 2 diabetes, along with hypertension, is a major cause of worsening renal function. It would be very meaningful if we could detect future deterioration of renal function. Design and method: Predictive models were created using machine learning techniques. We used the medical information database of 2533 patients who attended the outpatient clinic of our hospital between January 2003 and March 2015. The point at which the estimated glomerular filtration rate(eGFR) first falls below 60 mL/min/1.73 m 2 is used as the reference point. The input period was defined as the period from the reference point to one year before the reference point. Time series during the input period for eGFR, hematocrit, and urinary protein were extracted from blood and urine tests as features for predicting worsening renal function. The primary endpoint was a 50% reduction in eGFR during the evaluation period of three years since reference point from the mean value during the input period. The optimal combination was selected to maximize prediction performance using up to 4 features considering the number of endpoints reached. The prediction results were evaluated on AUC with 10 repeated 2-fold cross validation. Results: Among 2533 patients, 1409 patients had the reference point. Of those, 377 patients for whom a record existed for the input and evaluation periods but did not reach the primary endpoint, and 36 patients reached the primary endpoint. The mean eGFR during the input period was 67.9 ± 8.8 mL/min/1.73 m 2, the mean age was 64.9 ± 10.2, 41.6% was women. The mean AUC was 0.82 ± 0.033 when the maximum and minimum values of urinary protein, minimum of eGFR and the difference between maximum and minimum values of eGFR during the input period were used as features. Conclusions: In this study, we developed a prediction model for worsening renal function using time series measurements in patients with relatively preserved renal function(eGFR> 60 mL/min/1.73 m 2 ), and the results showed that it is very important to consider the variable components of urine protein and eGFR, which are laboratory values reflecting renal impairment. Recently, SGLT2 inhibitors have been used to treat kidney disease, but this study analyzed data before the release of SGLT2 inhibitors. We would like to examine whether our model can be applied to patients taking SGLT2 inhibitors.