Premium
A simple prediction score for kidney disease in the Korean population
Author(s) -
KWON KEUNSANG,
BANG HEEJUNG,
BOMBACK ANDREW S,
KOH DAIHA,
YUM JUNGHO,
LEE JUHYUNG,
LEE SIK,
PARK SUNG K,
YOO KEUNYOUNG,
PARK SUE K,
CHANG SOUNGHOON,
LIM HYUNSUL,
CHOI JOONG MYUNG,
KSHIRSAGAR ABHIJIT V
Publication year - 2012
Publication title -
nephrology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 61
eISSN - 1440-1797
pISSN - 1320-5358
DOI - 10.1111/j.1440-1797.2011.01552.x
Subject(s) - medicine , kidney disease , population , diabetes mellitus , national health and nutrition examination survey , disease , renal function , youden's j statistic , proteinuria , demography , kidney , predictive value , environmental health , endocrinology , sociology
Aim: Screening algorithms for chronic kidney disease have been developed and validated in American populations. Given the worldwide burden of kidney disease, developing algorithms for populations outside the USA is needed. Methods: Using simple, non‐invasive questions, we developed a prediction model for chronic kidney disease from national population samples in Korea. The Korean National Health and Nutrition Examination Survey ( n = 6565) was used for model development while validation was performed in two independent population samples, internal ( n = 2921) and external datasets ( n = 8166). Chronic kidney disease was defined as glomerular filtration rate < 60 mL/min per 1.73 m 2 . Results: Seven factors – age, female gender, anaemia, hypertension, diabetes mellitus, cardiovascular disease and proteinuria – were significantly associated with prevalent chronic kidney disease. Integer scores were assigned to variables based on the magnitude of associations: 2 for age 50–59 years, 3 for age 60–69 years and 4 for age 70 years or older, and 1 for female gender, anaemia, hypertension, diabetes, proteinuria and cardiovascular disease. Based on the Youden index, a value of 4 or greater defined a high risk population with sensitivity 89%, specificity 71%, and positive predictive value 19%, and negative predictive value 99%. The area under the curve was 0.83 for the development set, and 0.87 and 0.78 in the two validation datasets. Conclusion: This prediction algorithm, weighted towards common non‐invasive variables, had good performance characteristics in an Asian population, and provides new evidence of the similarity of the algorithms for Western and Eastern populations.