Home Dialysis in the Prospective Payment System Era

Abstract

The ESRD Prospective Payment System introduced two incentives to increase home dialysis use: bundling injectable medications into a single payment for treatment and paying for home dialysis training. We evaluated the effects of the ESRD Prospective Payment System on home dialysis use by patients starting dialysis in the United States from January 1, 2006 to August 31, 2013. We analyzed data on dialysis modality, insurance type, and comorbidities from the United States Renal Data System. We estimated the effect of the policy on home dialysis use with multivariable logistic regression and compared the effect on Medicare Parts A/B beneficiaries with the effect on patients with other types of insurance. The ESRD Prospective Payment System associated with a 5.0% (95% confidence interval [95% CI], 4.0% to 6.0%) increase in home dialysis use by the end of the study period. Home dialysis use increased by 5.8% (95% CI, 4.3% to 6.9%) among Medicare beneficiaries and 4.1% (95% CI, 2.3% to 5.4%) among patients covered by other forms of health insurance. The difference between these groups was not statistically significant (1.8%; 95% CI, -0.2% to 3.8%). Conversely, in both populations, the training add-on did not associate with increases in home dialysis use beyond the effect of the policy. The ESRD Prospective Payment System bundling, but not the training add-on, associated with substantial increases in home dialysis, which were identical for both Medicare and non-Medicare patients. These spill-over effects suggest that major payment changes in Medicare can affect all patients with ESRD.

Keywords: Economic Analysis; United States Renal Data System; chronic dialysis; end-stage renal disease; peritoneal dialysis.

Figure 1.

Assembly of the study cohort from the USRDS.

Figure 2.

Effect of the PPS on home dialysis use at day 90 stratified by type of insurance, Medicare Parts A/B (red) or other insurance (blue). (A) Depicts the unadjusted probability of home dialysis at day 90 over time for each of the populations. The scatter plot gives the average proportion of the population using home dialysis at day 90. In both populations, the predicted probability of home dialysis under the PPS (solid red and solid blue lines) increased substantially after passage of the MIPPA. We projected a similar increase in home dialysis use if the PPS had taken effect without the training add-on (dark dotted red and blue lines) and an overall decline in home dialysis without the PPS (light dotted red and blue lines). In (B), we show the association between the PPS and change in provider behavior in the Medicare Part A/B and the other insurance populations (dark red and blue lines respectively) with 95% CIs (light red and blue lines). The graph shows the difference between observed home dialysis use under the PPS and projected home dialysis use in the absence of the PPS, which represents the estimated change in provider behavior over time. The MIPPA was not associated with an immediate change in provider behavior, represented by the null effect in July 2008. However, the change in home dialysis use steadily increased afterward and continued after the PPS’s implementation. By the end of the study period, this difference in provider behavior resulted in an absolute increase in home dialysis use by 5.8% (95% CI, 4.3% to 6.9%) in the Medicare Parts A/B population and 4.1% (95% CI, 2.3% to 5.4%) in the non-Medicare subgroup. Although the PPS had a slightly higher effect in the Medicare Part A/B population relative to the non-Medicare population (difference of 1.8%), it was not statistically significant (P=0.08). (C) Shows the association between the training add-on component of the PPS and the nonsignificant change in provider behavior in both populations (dark red and blue lines) with 95% CIs (light red and blue lines). The graph shows the difference between observed home dialysis use under the full PPS and projected home dialysis use in the absence of the training add-on. We found that this difference was not statistically different from null during the entire study period.