Kidney Dialysis Survival Depends on Statistical Approach

[okarol]

Kidney Dialysis Survival Depends on Statistical Approach

By Chris Emery, Contributing Writer, MedPage Today
Published: July 30, 2009
Reviewed by Zalman S. Agus, MD; Emeritus Professor
University of Pennsylvania School of Medicine and
Dorothy Caputo, MA, RN, BC-ADM, CDE, Nurse Planner

PRINCETON, N.J., July 30 -- Using different statistical models to analyze the same data on kidney disease patients produced different results concerning survival during dialysis, a new study found.
Action Points 

    * Note that small studies on hemodialysis that use a proportional hazard model may lack statistical power.


    * Note that the authors cautioned against using their findings on the relationship between kidney dialysis and survival in interpreting existing national and international guidelines concerning an adequate dialysis dose.

When using a conventional statistical method known as a "proportional hazards model," the researchers found no relationship between how much dialysis patients underwent (their "dose") and how long they survived (95% CI 0.45 to 1.14), according to a report published online in the July 30 Journal of the American Society of Nephrology.

In contrast, a newer method called an "accelerated failure time model" found that each small increase in dialysis (0.1-U increment of Kt/V) improved adjusted median patient survival by 3.5% (95% CI 0.20 to 7.08).

"The association of hemodialysis dosage with patient survival is controversial," Christos Argyropoulos, MD, PhD, of the University of Pittsburgh Medical Center, and colleagues wrote.

"Here, we tested the hypothesis that methods for survival analysis may influence conclusions regarding dialysis dosage and mortality."

The authors said conflicting conclusions from previous studies may relate to which statistical methods researchers used, particularly in small to moderately sized clinical trials.

They suggested that future studies use accelerated failure time models in addition to more conventional proportional hazards models.

To compare the different statistical models, the researchers used data from 766 hemodialysis patients enrolled in the Choices for Healthy Outcomes in Caring for End Stage Renal Disease (CHOICE) study.

That effort monitored the health of patients from 81 dialysis clinics across the U.S. The data included information about age, gender, race, cause of kidney disease, diagnoses of heart failure and comorbidities.

The patients were followed for nine years, during which time 315 participants died.

The researchers analyzed the data with the Cox proportional hazards model (PHM), a commonly used statistical tool by British statistician Sir David Cox, and with an accelerated failure time model (AFTM), sometimes used as an alternative to proportional hazards models.

They used the models to determine the relationship between the amount of kidney dialysis the patients received and all-cause mortality.

Both models predicted that age, race, heart failure, physical functioning, and comorbidity scores were important predictors of survival.

Beyond that, the findings from the two models diverged. In addition to different conclusions about the relationship between dialysis dose and survival, the PHM yielded less accurate estimates for median survival than the AFTM.

The authors said they tried out the AFTM because they thought the model might better capture what happens when kidneys cease working and toxins start building up in a patient's bloodstream.

Because the PHM doesn't take into account the cumulative damage caused by these toxins, the authors wrote, the models might lose predictive power in studies lacking large numbers of patients.

"This phenomenon may partly explain the discrepant findings between the large observational studies and the much smaller randomized trials relating dialysis dosage to mortality," they wrote.

"Conversely," they continued, "alternatives to the PHM, such as the accelerated failure time model (AFTM) can model cumulative exposures and, in fact, may be the preferred method to study survival in the context of additive damages."

They pointed out several limitations of the study, including the fact that "the CHOICE study was an observational one, not a randomized clinical trial designed to test the effects of dialysis dosage on survival."

Additionally, they wrote, "approximately one third of enrolled patients had missing covariate data."

Dr. Argyropoulos and his colleagues also noted that their findings do not speak to the validity of existing national and international guidelines concerning an adequate dialysis dose.

The study was funded by the Renal Discoveries-Baxter Extramural Grant Program.

The researchers reported no financial conflicts of interest.

Primary source: Journal of the American Society of Nephrology
Source reference:
Argyropoulos C, et al "Considerations in the statistical analysis of hemodialysis patient survival" J Am Soc Nephrol 2009; DOI: 10.1681/ASN.2008050551.

http://www.medpagetoday.com/Nephrology/ESRD/15321