만성 신부전증의 진행예측에 관한 연구

Abstract

[한글]

만성 신부전증(chronic renal failure)은 원인 신질환에 관계없이 지속적으로 신기능이 감소하는 특성을 가지고 있으며, 그간 만성 신부전증의 진행 예측에 관하여 여러 가지 수학적인 방법들이 제시되어 왔다. 만성 신부전증 환자에서 혈청 크레아티닌을 연속적으

로 관찰하여 혈청 크레아티닌의 역수치(1/Scr) 혹은 Log치(Log Scr)와 관찰 시간과의 관계에서 일직선이 수립된다는 사실이 알려져 있으며, 이러한 수치의 회기방정식(regression equation)을 이용하여 예후 방정식(prognostic equation)으로도 사용할 수 있다. 오늘날 이러한 방법들을 이용하여 여러가지 신질환의 진행과 치료 수단의 평가에 이용되고 있으나 그 정확도의 비교에 대한 보고나 성별, 연령, 원인 신질환등에 따른 차이에 대한 연구는 많지 않다.

이에 저자는 변형된 혈청 크레아티닌의 연속적인 측정으로 만성신부전의 진행을 정확히 예측할 수 있는지 알아보고저 만성 신부전증 환자 61예 에서 변형된 혈청 크레아티닌치와 시간과의 관계에서 예측오차(prediction error)를 구하고 또한 성별, 연령별, 원인 신질환별로 구분 분석하여 어떠한 요소가 만성 신부전증의 진행 예측에 영향을 주는지 관찰하여 다음과 같은 결론을 얻었다.

1. 1/Scr의 예측오차 중앙치(평균±표준오차)는 2.1(3.2±13.1)개월로서, Scr의 47(64.7±62.8)개월, Log Scr의 15.7(21±23.9)개월 보다 유의하게 적었다. Scr, 1/Scr, Log Sc

r의 상관계수는 각각 0.910(0.895±0.090), -0.918(-0.889±0.101), 0.843(0,896±0.093)으로 세방법 모두상관계수가 높았다.

2. Scr, 1/Scr, Log Scr의 예측 오차를 성별로 비교하면 각각 남자(40예)의 경우 50.3(78.3±67.4)개월, 5.0(5.7±10.7)개월, 20.4(26.0±22.2)개월이었고 여자(21예)의 경우 33.8(39.0±43.8)개월, -1.2(-1.7±16.7)개월, 5.6(11.4±24.1)개월로서 세방법 모두 여자에서 통계학적으로 유의하게 적었다(p<0.05).

3. Scr, 1/Scr, Log Scr의 예측 오차를 연령별로 비교하면 각각 45세미만(23예)의 경우 52.8(86.6±71 4)개월, 4.5(6.4±11.7)개월, 20.1(28.1±23.9)개월이였고, 45세 이상 65세 미만(28예)의 경우 41.5(54.8±54.8)개월, 1.2(2.6±11.4)개월, 15.2(18.3±21.1)개월

이였고, 65세 이상(10예)의 경우 16.0(42.4±52.3)개월, -2.4(-2.9±18.6)개월, 3.0(12.1±20.6)개월로 어느 방법에서나 연령에 따른 예측오차의 차이는 없었다.

4. 원인 신질환별로 Scr, 1/Scr, Log Scr의 예측 오차를 비교해 보면 만성 사구체 신염(22예)의 경우 51.8(77.2±60.4)개월, 5.5(4.6±14.3)개월, 20.4(24.8±23.6)개월, 당뇨병성 신증(17예)의 경우 12.7(28.6±39.9)개월, -1.2(-0.2±9.9)개월, 4.6(8.8±17.1)개

월, 고혈압성 신증(7예)의 경우 56.2(62.7±56.8)개월, -0.2(-1.6±13.8)개월, 26.1(9.6±22.3)개월로 어는 방법에서나 원인 신질환에 따른 예측오차의 차이는 없었다.

이상의 결과로 만성 신부전증에서의 예측오차(prediction error)는 Scr, Log Scr 보다 1/Scr에서 유의하게 적었으며 1/Scr을 사용하는 경우 예측의 정확도가 가장 높았다. 또한 예측 오차는 여자가 남자보다 적어 성별로는 유의한 차이가 있었으나, 연령별, 원인 신

질환 별로는 차이를 보이지 않았다.

A study on the predictability of progression of chronic renal failure

Dong Hun Cha

Department of Medical Science Graduate Schoo Yonsei University

(Directed by Professor Dae Suk Han, M. D.)

To predict the progression of chronic renal failure by means of mathematical

models, the serum creatinine data of patients were linearized by reciprocal

model(1/Scr) and logarithmic model (LogScr). Thereafter, regression equations of

the transformed data were calculated and used as prognostic equations. When these

models were applied to an individual case, the prediction of the progression of

disease was claimed to be independent of the underlying renal disease, sex, age,

and other factors influencing the progression of chronic renal failure.

Today, there is a widespread use of these models in the evaluation of therapeutic

trials and in the monitoring of various kidney disorders. However, only a few

attempts have been made to assess the accuracy of these models. A high degree of

accuracy is assumed exclusively because of the high correlation coefficients of the

transformed serum creatinine data. The models may also be criticized on the basis

that data outside the range of original data set were predicted. It is the aim of

my study to reevaluate the accuracy of the prediction by application of the

reciprocal and logarithmic model according to sex, age and the underlying renal

disease. The results obtained in this study were as follows:

1. The summary of the 61 cases of chronic renal failure

The mean age of all patients was 51.1 years old and the ratio of male:female was

40:21. The number of determinations from first Scr to last Scr was 6.5±3.4. The

mean first Scr was 2.4±0.7mg/dl, the mean last Scr was 10.6±2.9mg/dl. The

prediction interval was 13.1±10.4 months. The mean duration of observation period

from first Scr to last Scr was 33.3±22.7 months.

2. Prediction error and correlation coefficient of total 61 patients

The prediction error of Scr, 1/Scr, Log Scr was 64.7±62.8 months, 3.2±13.1

months and 21.0±23.9 months, respectively. The prediction error of the 1/Scr was

smallest and this means that the accuracy of predictability of 1/Scr was highest

among three models.

The mean correlation coefficient of Scr, 1/Scr, Log Scr was 0.910, -0.918, 0.843,

respectively.

3. Prediction error according to sex.

The prediction error of male and female patients was 78.3±67.4 months and

39.0±43.8 months in Scr, 5.7±10.0 months and -1.7±16.7 months in 1/Scr,

26.0±22.4 months and 11.4±24.1 months in Log Scr.

Significant differences were noted in all three models between male and female.

4. Prediction error according to age

The prediction error of less than 45 years old, 45-65 years old more than 65

years old was 86.6±71.4 months, 54.5±54.8 months,42.4±52.3 months in Scr and

6.4±11.7 months, 2.6±11.4 months,-2.9±18.6 months in 1/Scr and 28.1±23.9

months, 18.3±21.1 months, 12.1±20,6 months in Log Scr. There was no significant

difference between three different age subgroups regardless of models used.

5. Prediction error according to underlying renal disease

The prediction errors of chronic glomerulonephritis, diabetic nephropathy,

hypertensive golmerulosclerosis were 77.2± 60.4 months, 28.6±39.9 months,

62.7±56.8 months in Scr and 4.6±14.3 months, -0.2±9.9 months, 1.6±13.8 months

in 1/Scr and 24.8±23.6 months,8.8±17.1 Months, 19.7±22.3 months in Log Scr.

There was no significant difference between three different etiologic subgroups

regardless of models used.

In conclusion, the prediction error of the chronic renal failure was smallest in

1/Scr model than two other models indicating that the accuracy of the prediction

was highest in 1/Scr model than two other models. There was significant difference

of prediction error according to sex, but there was no significant difference of

prediction error according to age and the underlying renal disease.

[영문]

To predict the progression of chronic renal failure by means of mathematical models, the serum creatinine data of patients were linearized by reciprocal model(1/Scr) and logarithmic model (LogScr). Thereafter, regression equations of the transformed data were calculated and used as prognostic equations. When these

models were applied to an individual case, the prediction of the progression of disease was claimed to be independent of the underlying renal disease, sex, age, and other factors influencing the progression of chronic renal failure.

Today, there is a widespread use of these models in the evaluation of therapeutic trials and in the monitoring of various kidney disorders. However, only a few attempts have been made to assess the accuracy of these models. A high degree of accuracy is assumed exclusively because of the high correlation coefficients of the transformed serum creatinine data. The models may also be criticized on the basis that data outside the range of original data set were predicted. It is the aim of my study to reevaluate the accuracy of the prediction by application of the reciprocal and logarithmic model according to sex, age and the underlying renal disease. The results obtained in this study were as follows:

1. The summary of the 61 cases of chronic renal failure

The mean age of all patients was 51.1 years old and the ratio of male:female was 40:21. The number of determinations from first Scr to last Scr was 6.5±3.4. The mean first Scr was 2.4±0.7mg/dl, the mean last Scr was 10.6±2.9mg/dl. The

prediction interval was 13.1±10.4 months. The mean duration of observation period from first Scr to last Scr was 33.3±22.7 months.

2. Prediction error and correlation coefficient of total 61 patients

The prediction error of Scr, 1/Scr, Log Scr was 64.7±62.8 months, 3.2±13.1 months and 21.0±23.9 months, respectively. The prediction error of the 1/Scr was smallest and this means that the accuracy of predictability of 1/Scr was highest among three models.

The mean correlation coefficient of Scr, 1/Scr, Log Scr was 0.910, -0.918, 0.843, respectively.

3. Prediction error according to sex.

The prediction error of male and female patients was 78.3±67.4 months and 39.0±43.8 months in Scr, 5.7±10.0 months and -1.7±16.7 months in 1/Scr, 26.0±22.4 months and 11.4±24.1 months in Log Scr.

Significant differences were noted in all three models between male and female.

4. Prediction error according to age

The prediction error of less than 45 years old, 45-65 years old more than 65 years old was 86.6±71.4 months, 54.5±54.8 months,42.4±52.3 months in Scr and 6.4±11.7 months, 2.6±11.4 months,-2.9±18.6 months in 1/Scr and 28.1±23.9

months, 18.3±21.1 months, 12.1±20,6 months in Log Scr. There was no significant difference between three different age subgroups regardless of models used.

5. Prediction error according to underlying renal disease

The prediction errors of chronic glomerulonephritis, diabetic nephropathy, hypertensive golmerulosclerosis were 77.2± 60.4 months, 28.6±39.9 months, 62.7±56.8 months in Scr and 4.6±14.3 months, -0.2±9.9 months, 1.6±13.8 months in 1/Scr and 24.8±23.6 months,8.8±17.1 Months, 19.7±22.3 months in Log Scr.

There was no significant difference between three different etiologic subgroups regardless of models used.

In conclusion, the prediction error of the chronic renal failure was smallest in 1/Scr model than two other models indicating that the accuracy of the prediction was highest in 1/Scr model than two other models. There was significant difference of prediction error according to sex, but there was no significant difference of prediction error according to age and the underlying renal disease.