A very frequent commentary about the NCDS is that the slightly increased mortality with short treatment time was highly significantly, clinically, falsely interpreted as not significant because p = 0.06, and therefore the ne-phrology community was misled about the importance

Fig. 8. Reconciliation of observational studies with randomized trials.

120 110 100 90 80 2 70

60 50 40 30 20

spKt/V

0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80

NPCR

spKt/V

0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80

NPCR

Typical observational analysis r = 0.74

RRM in the two groups is equal

Dose-targeting bias 0.86 observed in HEMO after stratification

RRM in the two groups is equal

1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80 1.90 2.00

Standard arm spKt/V HD arm spKt/V

The NCDS design | ||

The design goal was | ||

not achieved with |
Groups I, III | |

respect to t and Kt/V | ||

Median | ||

Groups II, IV |

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5

eKt/V

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5

eKt/V

Fig. 10. The NCDS design called for identical distributions of spKt/V at treatment time (t) = 2.5-3.5 and t = 4.5-5.5 h as illustrated in figure 9 and the spKt/V distribution for mean t = 3.2 h was substantially lower than for mean t = 4.5 h. This part of the design was not achieved.

Fig. 9. The NCDS design called for identical distributions of spKt/V at treatment time (t) = 2.5-3.5 and t = 4.5-5.5 h as illustrated here. In this plot values for short t are plotted as virtually identical to those with long t. This part of the design was not achieved.

Fig. 10. The NCDS design called for identical distributions of spKt/V at treatment time (t) = 2.5-3.5 and t = 4.5-5.5 h as illustrated in figure 9 and the spKt/V distribution for mean t = 3.2 h was substantially lower than for mean t = 4.5 h. This part of the design was not achieved.

For t = 3.2, 52% patient had 56% probability of failure

For t = 3.2, 52% patient had 56% probability of failure

eKt/V

Fig. 11. The marginal p = 0.06 in the NCDS for RRF = /(t) was clearly of no significance because of the highly significant correlation of Kt/V to treatment time.

eKt/V

Fig. 11. The marginal p = 0.06 in the NCDS for RRF = /(t) was clearly of no significance because of the highly significant correlation of Kt/V to treatment time.

Dialysis time (h)

Dialysis time (h)

5.00

Statistical melting pot

0 —I—I—I—I—I—I—I—I—I—I—I—I—I—I—I—I—I—I—I—

Dialysis time (h)

Fig. 12. Mortality rate as function of treatment time in DOPPS data. A The regression of mortality rate on treatment time is extremely variable across different geographic regions. These curves are not convincing evidence of a generalizable inverse dependence of mortality on treatment time. B The DOPPS data after statistical analysis showed that mortality rate decreases 7%/ 30-min increase in treatment time.

of treatment time in the dialysis prescription [3, 4, 9, 10]. In fact, the contrary argument might be more reasonable: The marginal effect seen actually may suggest that short time was beneficial in the study because the study design goal of separating Kt/V from treatment time was not achieved and Kt/V was significantly lower in the short time groups compared to long time groups. The study design is schematically illustrated in figure 9 where it can be seen that median spKt/V 0.90 and identical distributions should have been prescribed and delivered in both short and long time groups. The actual distributions are shown in figure 10 where the Kt/V distribution is substantially lower in the short time patients compared to long time patients with median 0.75 vs 0.91. The profound effect of this can be seen in figure 11 where 52% of the short time patients were in the Kt/V region of 56% probability of failure while only 18% of the long time patients fell into the high probability of failure region. It is surprising that there was not a more significant relationship between the probability of failure and treatment time in view of the 5-fold increase in risk of failure in the low Kt/V group. The effect of Kt/V was never tested in the initial analysis of the NCDS [1] but rather a surrogate, tacrolimus urea, which is a very poor substitute for Kt/V since it cannot distinguish the effect of protein catabolic rate.

Was this article helpful?

## Post a comment