Professional Documents
Culture Documents
Cutoff 4-5
Cutoff 4-5
American Journal of
Key Words value, the probability of death for an MHD patient whose
Hemodialysis ⴢ Malnutrition-inflammation complex MIS was 3, 4, and 5 is 10, 40, and 80%, respectively. Conclu-
syndrome ⴢ Malnutrition-inflammation score sions: Our study shows that MHD patients with MIS score of
more than 4–5 had a significant risk of 1-year mortality. Ad-
ditional risk factors associated with short-term mortality be-
Abstract sides malnutrition-inflammation complex syndrome were
Background: Malnutrition-inflammation score (MIS) is a anemia and renal osteodystrophy. This study proves that MIS
comprehensive and quantitative system to assess malnutri- is a useful tool to risk-stratify Asian MHD patients and to
tion-inflammation complex syndrome, and a strong correla- identify those at risk of short-term death. Nutritional inter-
tion between MIS and morbidity/mortality in maintenance ventions that can improve the MIS may also improve surviv-
hemodialysis (MHD) patients had been demonstrated. How- al, but this hypothesis needs to be verified in interventional
ever, there is no cut-off value of MIS to categorize patients studies. Copyright © 2008 S. Karger AG, Basel
into high risk or low risk patients. Methods: A total of 257
chronic stable and ambulatory adult MHD patients from Far
Eastern Memorial Hospital were enrolled for the study. The
MIS of each patient was recorded at the initiation of study Introduction
and the study population was followed up as a 12-month
prospective cohort to evaluate mortality as the primary out- More and more reports disclose the association be-
come. Results: Twelve patients died in the 12-month obser- tween protein-energy malnutrition, inflammation, and
vational period. Both multiple logistic regression analyses mortality in maintenance hemodialysis (MHD) patients
and Cox proportional hazards model denoted MIS, alkaline [1, 2]. The term malnutrition-inflammation complex
phosphatase, transferrin saturation, ferritin, and total iron syndrome (MICS) was created to denote the important
binding capacity as significant predictors of 1-year mortali- contribution of malnutrition and inflammation to the
ty. The conditional effect plot of MIS on 1-year mortality re-
vealed that when fixing the alkaline phosphatase, transferrin
saturation, ferritin, and total iron binding capacity at a mean L.H. and H.-H.W. contributed equally to this work.
130.239.20.174 - 4/7/2015 1:04:28 AM
© 2008 S. Karger AG, Basel Chih-Kang Chiang, MD, PhD, No. 7 Chung-Shan South Road
0250–8095/08/0285–0840$24.50/0 Department of Diagnostics and Therapeutics and Internal Medicine
Fax +41 61 306 12 34 National Taiwan University Hospital, Taipei, Taiwan (ROC)
E-Mail karger@karger.ch Accessible online at: Tel. +886 2 2312 3456, ext. 980 2381, Fax +886 2 2322 2955
www.karger.com www.karger.com/ajn E-Mail ckchiang@ntu.edu.tw
Umea University
Downloaded by:
clinical outcome of end-stage renal disease (ESRD) pa- wise specified. Comparisons between patients were performed
tients. Many markers, such as tumor necrosis factor-␣, using Student’s t test (two tailed) for normally distributed data,
Mann-Whitney U test for median data, 2 test for categorical data.
albumin, C-reactive protein, interleukin-6, normalized The association between MIS and other variables was estimated
protein nitrogen appearance, and adipocytokines had by Spearman’s rank correlation.
been used to represent the severity of MICS, but none of Multiple logistic regression models and multivariate Cox pro-
them had sufficient predictive power. Malnutrition-in- portional hazards models were fitted to the collected data for
flammation score (MIS), originally developed by Kalan- modeling the probability of 1-year mortality [4]. Model-fitting
techniques for logistic regression analysis including variable se-
tar-Zadeh et al. [3], is a comprehensive and quantitative lection, assessment of the goodness of fit, and regression diagnos-
system to assess MICS, and a strong correlation between tics (e.g. residual analysis, detection of influential cases, and
MIS and morbidity/mortality in MHD patients had been check for multicollinearity) were used to assure the quality of
demonstrated. However, in their work they did not pro- analysis results [4]. In variable selection, the stepwise procedure
vide a cut-off value of MIS to categorize patients into was applied to logistic regression analysis with both significance
level for entry and significance level for stay set to 0.15. The good-
high-risk or low-risk groups. Besides, the utility of MIS ness of fit (GOF) measures (e.g. the percentage of concordant
in Asian MHD patients was not investigated. This study pairs and the adjusted generalized coefficient of determination)
was conducted to define the predictability of MIS for and tests (e.g. deviance, Pearson 2 GOF test, and the Hosmer-
clinical outcome in Asian MHD patients and to search Lemeshow GOF test) for logistic regression analysis were com-
for a cut-off point which best predicts mortality in MHD puted [4]. In all statistical testing, the type I error ␣ = 0.05 was
taken, and thus p ! 0.05 indicated a statistically significant result.
patients. The variance inflation factors (VIF) were evaluated to uncover
possible collinearity. Parameters in the regression model with
VIF 15 were discarded to avoid biased estimations.
Subjects and Methods Model-fitting techniques for multivariate Cox proportional
hazards models also included variable selection, assessment of the
Patients goodness of fit, and regression diagnostics [4]. The variable selec-
In June, 2004, a total of 257 chronic stable and ambulatory tion approaches included both Akaike Information Criterion and
adult MHD patients from Far Eastern Memorial Hospital were the stepwise procedure with both significance level for entry and
enrolled in the study. All the patients had been on maintenance significance level for stay set to 0.15. The adjusted R square for
dialysis for more than 3 months and did not have infection, car- Cox proportional hazards model was computed or conducted. Re-
diovascular events, hospitalization, or surgery in the previous 2 gression diagnosis included residual analysis, influence analysis,
months. The demographic and laboratory data were obtained af- and check for multicollinearity. Residual analysis was conducted
ter the Far Eastern Memorial Hospital Institutional Review Board by checking Martingale residuals, scaled Schoenfeld residuals,
had approved exemption from written consent. The MIS of each and deviance residuals. Influence diagnosis was made by calcu-
patient was recorded at the initiation of study and the study pop- lating DFBETAS of each tested subject. Parameters with VIF !5
ulation was followed up as a 12-month prospective cohort to eval- were discarded due to remarkable collinearity. We also used c sta-
uate mortality as the primary outcome. tistic to evaluate the prediction power of multiple logistic regres-
sion models and multivariate Cox proportional hazards models.
Malnutrition-Inflammation Score
The MIS consists of four main parts: patient’s related medical
history, physical examination, body mass index (BMI), and labo-
ratory parameters [3]. Patient’s medical history includes weight Results
changes, dietary intake, gastrointestinal symptoms, functional
capacity, and co-morbidity including number of years on dialysis.
Physical examination is to detect loss of subcutaneous fat and The characteristics of enrolled patients are listed in
signs of muscle wasting. Laboratory parameters are serum albu- table 1. Twelve patients died in the 12-month observa-
min and serum total iron binding capacity (TIBC) levels. Each of tional period. Since the MIS did not show normal distri-
the 10 components has four levels of severity, from 0 (normal) to bution, its data are presented as median and quartiles.
3 (severely abnormal). The clinical data of age, sex, duration on Ages ranged from 24 to 87 years (mean 58.8 8 13.5 years),
dialysis, body weight and height, underlying renal condition and
presence of co-morbid conditions were obtained by chart review. and dialysis vintage varied from 3.1 to 265.7 months
One of the investigators (C.-K.C.) interviewed all the patients to (mean 48.9 8 51.1 months). One hundred and nine pa-
obtain physical morbidity and ability of self-care and to assess the tients (42.4%) had diabetes mellitus, 42 patients (16.4%)
subcutaneous muscle tissue and muscle mass. had coronary artery disease, 25 patients (9.7%) had con-
gestive heart failure, 12 patients (4.7%) had peripheral ar-
Statistical Analysis
The data were analyzed using SAS/STAT쏐 software, v8.0 (SAS terial obstructive disease, and 10 patients (3.9%) had ma-
Institute Inc., Cary, N.C., USA). All the results of descriptive anal- lignancy. Among patients with malignancy, only 3 were
ysis were expressed in mean 8 standard deviation unless other- metastatic. The deceased patients had higher MIS score,
130.239.20.174 - 4/7/2015 1:04:28 AM
Calcium level had been adjusted by albumin. Kt/V = Dialysis urea clearance; DM = diabetes mellitus;
CAD = coronary artery disease; CHF = congestive heart failure; PAOD = peripheral artery obstructive disease;
Hb = hemoglobin.
higher alkaline phosphatase (ALK-P), and lower albumin tion with predialysis creatinine (r = –0.466, p ! 0001) and
levels. Otherwise, there were no significant differences in hemoglobin (r = –0.315, p ! 0.001). Normalized protein
age, duration of dialysis, BMI, comorbidity, dialysis urea catabolism rate, plasma concentration of potassium, cho-
clearance (Kt/V), plasma concentration of creatinine, lesterol, iron, and uric acid correlated negatively with
cholesterol, TIBC, ferritin, transferring saturation, para- MIS but their correlation coefficient was low (r ranged
thyroid hormone (PTH), calcium, phosphate and hemo- from 0 to –0.3). It is not surprising that albumin, TIBC,
globin level between the survived and the deceased pa- and BMI had a stronger correlation since they are com-
tients. ponents of MIS.
Table 2 lists correlation coefficient of MIS and relevant Multiple logistic regression analyses were conducted
variables. The MIS was positively and significantly cor- to find important risk factors of 1-year mortality. The re-
related with death (r = 0.231, p ! 0.001). It also had posi- sults are shown in table 3. The result of concordant pair
tive correlation with age, gender, dialysis vintage, alka- was 94.1% and the adjusted R 2 was 0.4747. According to
line phosphatase, ferritin, fasting blood sugar level, and the regression model, higher MIS, higher ALK-P, higher
Kt/V. Plasma phosphate correlated negatively with the transferrin saturation, lower ferritin, and higher TIBC
MIS, but the PTH and serum calcium did not signifi- were important risk factors of death. Although transfer-
cantly correlate with MIS. MIS had mild negative correla- rin (or TIBC) is one of the components of MIS, low VIF
130.239.20.174 - 4/7/2015 1:04:28 AM
Table 3. Hazard ratio for death according to multiple logistic re- Table 4. Hazard ratio for death according to multivariate Cox
gression analysis proportional hazards model
MIS (per one score increase) 1.51 (1.25–1.83) <0.0001 MIS (per one score increase) 1.437 (1.246–1.657) <0.0001
ALK-P (per 1 U/l increase) 1.01 (1.00–1.02) 0.047 ALK-P (per 1 U/l increase) 1.012 (1.003–1.020) 0.0068
Transferrin saturation Transferrin saturation
(per 1% increase) 1.07 (1.03–1.12) 0.0006 (per 1% increase) 1.069 (1.032–1.106) 0.0002
Ferritin (per 1 ng/ml increase) 0.99 (0.996–0.998) 0.0019 Ferritin (per 1 ng/ml increase) 0.996 (0.993–0.998) 0.0002
TIBC (per 1 g/dl increase) 1.01 (1.00–1.02) 0.0001 TIBC (per 1 g/dl increase) 1.017 (1.011–1.024) <0.0001
(!1.5) indicated no problem of multicollinearity. The c tiple logistic regression model. For every one unit in-
statistic of this multiple logistic regression model was crease in MIS, ALK-P, transferrin saturation, ferritin, and
0.943, better than MIS alone (c statistic = 0.815). Condi- TIBC, the hazard ratio for 1-year mortality was 1.51, 1.01,
tional effect plot of MIS on 1-year mortality was drawn 1.07, 0.99, and 1.01, respectively.
according to the multiple logistic regression model and
was shown as figure 1. When we fix the ALK-P, transfer-
rin saturation, ferritin, and TIBC at a mean value, the Discussion
probability of death for a MHD patient whose MIS was 3,
4, and 5 is 10, 40, and 80%, respectively. The MIS is a comprehensive scoring system with sig-
The results of multivariate Cox proportional hazards nificant strong correlations with prospective hospitaliza-
model for 1-year mortality are presented in table 4. The tion indices and mortality [2]. The easy and reproducible
adjusted R 2 for this model was 0.3817, which was good measure makes it a convenient tool to assess the severity
because the R 2 of Cox proportional hazards models tends of MICS. Our study showed that even though the mortal-
to be low. Although the influence of time had been con- ity rate was lower in Asian MHD patients than in white
sidered in Cox proportional hazards model, the risk fac- [5], the higher MIS remained a strong predictor of mor-
tors selected in the model were not different from mul- tality in Asian patients. Although the association between
130.239.20.174 - 4/7/2015 1:04:28 AM
References
1 Shinaberger CS, Kilpatrick RD, Regidor DL, 3 Kalantar-Zadeh K, Kopple JD, Block G, 5 Robinson BM, Joffe MM, Pisoni RL, Port FK,
McAllister CJ, Greenland S, Kopple JD, Ka- Humphreys MH: A malnutrition-inflam- Feldman HI: Revisiting survival differences
lantar-Zadeh K: Longitudinal associations mation score is correlated with morbidity by race and ethnicity among hemodialysis
between dietary protein intake and survival and mortality in maintenance hemodialysis patients: the Dialysis Outcomes and Practice
in hemodialysis patients. Am J Kidney Dis patients. Am J Kidney Dis 2001; 38: 1251– Patterns Study. J Am Soc Nephrol 2006; 17:
2006;48:37–49. 1263. 2910–2918.
2 Kalantar-Zadeh K, Ikizler TA, Block G, 4 Hosmer DW: Applied Logistic Regression, 6 Kalantar-Zadeh K, Kopple JD, Humphreys
Avram MM, Kopple JD: Malnutrition-in- ed 2. Hoboken, John Wiley & Sons, 2000. MH, Block G: Comparing outcome predict-
flammation complex syndrome in dialysis ability of markers of malnutrition-inflam-
patients: causes and consequences. Am J mation complex syndrome in haemodialysis
Kidney Dis 2003;42:864–881. patients. Nephrol Dial Transplant 2004; 19:
1507–1519.
130.239.20.174 - 4/7/2015 1:04:28 AM