1. Mayo Clinic College of Medicine
Enhancement of MELD
W. Ray Kim, MD
Mayo Clinic College of Medicine
Rochester, MN
Plummer Building, Rochester, MN

2. Agenda How to make MELD better
As an indicator of pretransplant mortality
Add more variable(s): MELDNa
Optimize model: Refit MELD
As an indicator of posttransplant outcome
Limitations

3. Serum Sodium and Mortality
UNOS Registrants, 2005
N=6769
Risk of 3-Month Mortality
(after adjusting for MELD)
Na (mEq/L)
Kim NEJM 2008;1018

4. MELDNa: Incorporating Na to MELD
MELDNa = MELD - Na *MELD*(140-Na) + 140
125
130
135
140
MELD
Na (mEq/L)
Kim NEJM 2008;1018

5. Validation: MELDNa vs MELD
< <
MELD
Observed/Expected
Based on MELD
Based on MELDNa
Kim NEJM 2008;1018

6. Refitting Coefficients
Methods
Waitlist data obtained from OPTN from :
Model derivation set (05-06): n=14,214
Model validation set (07-08): n=13,945
All adult, primary LTx candidates with end stage liver disease included (HCC and status 1 excluded)
The proportional hazards regression analysis predicting mortality within 90 days of listing
Define lower and upper bounds
Optimize coefficients
Bilirubin, Creatinine, INR, and Na

7. Bilirubin
05-06 data

8. Bilirubin
05-06 data
Current MELD
LB=1.0

9. Bilirubin
05-06 data
Refit
LB=1.0

10. Creatinine
05-06 data

11. Creatinine
05-06 data
Current MELD
LB=1.0
UB=4.0

12. Creatinine
05-06 data
Refit
LB=0.8
UB=3.0

13. INR
05-06 data

14. INR
05-06 data
Current MELD
LB=1.0

15. INR
05-06 data
Refit
LB=1.0
UB=3.0

16. Sodium
05-06 data

17. Sodium
05-06 data
Refit
LB=125
UB=140

18. Validation: Bilirubin
07-08 data *
*Sharma et al. Gastro. 2008;1575

19. Validation: Creatinine
07-08 data *
*Sharma et al. Gastro. 2008;1575

20. Validation: INR
07-08 data *
*Sharma et al. Gastro. 2008;1575

21. Validation: MELD 07-08 data
LB (6) and UB (40) ignored for comparison purpose

22. Validation: SRTR MELD
07-08 data

23. Validation: Refit MELD(Na)
07-08 data

24. The Bottom Line Validation Set (2007-2008) Model Chi-Square
Concordance (SE)
MELD
2263
(0.0062)
SRTR MELD
2055 (-8)
(0.0063)
MELDNa
2370 (+7)
(0.0059)
Refit MELDNa
2374 (+11)
(0.0058)

25. ‘Kitchen Sink’ ≠ Crystal Ball
‘Unsupervised’ incorporation of all potentially relevant variables
C-statistics
* includes: refit MELDNa, age, BMI, albumin, HCV, diabetes, prior malignancies
** includes refit MELDNa, age, sex, race, albumin, HCV, HBV, diabetes, prior malignancies, life support, encephalopathy, ascites, previous abdominal surgery, portal vein thrombosis

26. MELD and Resource Utilization
pRBC
LOS 29 26 25 18 15
Kim ATC. 2004

27. Transplant Benefit Survival after LTx C-statistics 0.57 0.88
90 Days Survival
Survival
w/o LTx
Mean MELD
Refit MELDNa Deciles

28. Conclusions Ways to improve prediction of waitlist mortality
Add more variable: MELD Na
Represents a meaningful improvement
Big change in mortality in a subgroup of patients
Optimize model coeffcients: Refit MELDNa
New upper and lower bounds:
Creatinine:
INR:
Limit to predictability
MELD and post-transplant outcome
Statistically significant impact on survival and resource utilization
Limited accuracy