Hemodialysis and Peritoneal Dialysis: Comparison of Adjusted Mortality Rates According to the Duration of Dialysis: Analysis of the Netherlands Cooperative Study on the Adequacy of Dialysis 2

Materials and Methods

Results

Discussion

We investigated the RR for HD patients, compared with PD patients, with an observational study design. In this analysis, no statistically significant overall difference in mortality rates between HD and PD patients during the first 2 yr of dialysis treatment could be established, provided that adjustments were made for important differences in patient characteristics at the 3-mo visit. During the subsequent years, statistically significant differences in adjusted mortality rates in favor of HD were observed. This tendency toward greater relative mortality rates for PD patients with longer duration of treatment was observed especially for elderly patients, irrespective of diabetic status, gender, or the presence of CVD. For patients <60 yr of age with diabetes mellitus, a significant difference in mortality rates in favor of PD during the first 2 yr was observed, with a decline in RR in favor of HD in the subsequent years.

The most suitable study design for evaluation of differences in outcomes between treatment modalities would be a prospective, randomized, clinical trial. However, conducting such a study for ESRD could be regarded as extremely difficult, because of the special advantages and disadvantages associated with the modality choice for individual patients. Therefore, we are primarily left with the conscientious analysis of observational data from prospective studies in which special attention is paid to all sources of bias. The detailed follow-up data in the Netherlands Cooperative Study on the Adequacy of Dialysis database enabled us to assess the effects of patient characteristics, time since the initiation of dialysis, and modality switching on comparisons of mortality rates between HD and PD.

We observed a higher mortality rate for patients who were excluded from the analysis because of missing baseline data for important patient characteristics (residual renal function, nutritional parameters, or hemoglobin levels). Exclusion of patients with comparatively greater risks of death may yield biased results. However, the higher mortality rate associated with exclusion was observed for both HD and PD patients. Furthermore, similar results were observed in an analysis without adjustment for the aforementioned patient characteristics but with inclusion of the patients with missing data. Although the mortality rates in absolute terms were biased downward, the mortality rates for HD and PD patients relative to each other were not.

We performed analyses with both AT and ITT censoring strategies. In an AT analysis, the outcomes during actual treatment with HD are compared with the outcomes during actual treatment with PD; therefore, the estimated RR may be regarded as a pure reflection of the efficacy of one modality relative to the other. However, when switching from PD to HD (which occurred for many PD patients) is associated with a higher mortality rate because of deteriorating clinical conditions, then an AT analysis may yield a RR spuriously in favor of PD (5). To address this possibility in part, deaths that occurred within 60 d after switching of treatment modalities were attributed to the original treatment (11). In an ITT analysis, switches to the other treatment modality are disregarded and all deaths are allocated to the original treatment. The results of such an analysis may be especially useful for counseling regarding survival rates after the choice to start with HD or PD, but the treatment modalities are not analyzed as separate entities. This may be an advantage, for example when long-term survival rates are better with initial PD followed by transfer to HD after a number of months or years (21). The beneficial effect of such a treatment strategy could be determined with an ITT analysis. Because the two censoring strategies address different research questions and are associated with different analytical pitfalls, investigators should perform both types of analyses (14). We did not observe substantial differences between the results of our AT and ITT analyses. This indicates that no important bias associated with modality switching was introduced. The majority of patients did not switch modalities during the follow-up period and, for those nonswitching patients, the different censoring strategies are similar by definition.

The increase in mortality rates for PD patients with longer follow-up times and the resulting decrease in the RR for HD patients, compared with PD patients, are in accordance with the findings of other studies (5,6,11,14,21⇓⇓⇓⇓). For example, in the Canadian registry-based study of new dialysis patients by Fenton et al. (5), an increase in mortality rates for PD patients, which resulted in a trend toward a RR in favor of HD during a 4-yr follow-up period, was reported. Therefore, survival studies among new patients may yield results that are different from those of studies among chronically treated patients. In the study by Bloembergen et al. (4) of chronically treated patients registered in the United States Renal Data System (1987 to 1989), a survival benefit in favor of HD was observed. However, in the reanalysis by Vonesh and Moran (3), in which the same chronically treated and more recent new HD and PD patients (1989 to 1993) were included, no substantial overall differences in mortality rates between HD and PD could be established. This was probably because of the inclusion of the survival benefit of early PD patients in the latter analysis and the exclusion of the early PD experience in the former one. This phenomenon had already been suggested in the analysis by Bloembergen et al. (4), because the duration of dialysis seemed to be a substantial risk factor for PD patients.

Our study indicates that differences in important patient characteristics, either at baseline or during the follow-up period, do not explain the mortality rate difference between HD and PD after 2 yr of follow-up monitoring. At baseline, residual renal function, hemoglobin concentrations, and SGA scores were higher for PD patients; Davies comorbidity scores and the prevalence of CVD were lower for PD patients. A decline in residual renal function during the follow-up period was observed for both HD and PD patients. However, the differences in residual renal function and in the other parameters in favor of PD patients remained during the follow-up period. A comparatively high mortality rate for PD patients after 2 yr of follow-up monitoring was also observed when the analysis was restricted to patients who survived the first 24 mo of dialysis treatment and adjustments were made for comorbidity and other patient characteristics at the 24-mo visit. These observations indicate that a possible confounding effect attributable to the selection of patients with more health problems at later stages of PD treatment is less likely. This was supported by the finding that the results were not substantially modified when eligibility for transplantation was taken into account. Eligibility for transplantation may be regarded as a subtle marker of health status and the absence of severe comorbidity, suggesting selective dropout at the time of censoring because of transplantation. Unobserved differences associated with the choice of dialysis modality might have biased the outcomes in favor of one or the other treatment modality. However, these findings suggest that the increase in RR with long-term PD is possibly associated with the actual effects of this treatment modality. It is known that residual urine production is an important determinant of survival among dialysis patients (22,23⇓). Decreases in urine production among PD patients necessitate higher peritoneal ultrafiltration rates, which require the use of high-glucose dialysis solutions. Exposure to such dialysis fluids, which also contain high concentrations of glucose degradation products, is likely to induce morphologic and functional abnormalities of the peritoneum (24,25⇓). Loss of peritoneal ultrafiltration capacity is the most common clinical manifestation of peritoneal membrane alterations (26). The resulting overhydration is likely to be partly responsible for the high cardiovascular mortality rates for PD patients. Our finding of decreases in serum albumin levels for PD patients from the 24-mo visit onward might be attributable to this overhydration, because long-term PD is not associated with increases in peritoneal protein loss (27,28⇓). The high cardiovascular mortality rate may also be attributable to peritoneal absorption of glucose degradation products that enhance the formation of advanced glycation end products, thus contributing to the progression of atherosclerosis (29). On the basis of these considerations, it can be speculated that reductions in exposure to glucose and glucose degradation products, in combination with the use of glucose polymers as osmotic agents, might improve the long-term results for PD.

A difference in mortality rates in favor of PD was observed in the first 1 yr of dialysis treatment. This difference (44% in the analysis with AT censoring) did not reach the level of statistical significance, which might be attributable to a lack of statistical power. A real mortality difference in favor of PD in the early stages of RRT may exist and may have clinical relevance. It could be argued that the intermittent nature of HD is associated with greater fluctuations in BP, with deterioration of residual renal function, and greater fluctuations in toxic solute and blood glucose levels. For example, in a study by Berlanga et al. (30), it was observed that PD was associated with a beneficial effect on residual renal function; in a study by Jansen et al. (31), it was observed that HD was associated with a further deterioration of residual renal function shortly after the initiation of RRT.

An approach in which HD and PD are regarded as complementary modalities and initiation with PD is followed by timely transfer to HD might be advocated, as was done by van Biesen et al. (21). The slightly higher adjusted RR for HD versus PD for time periods of >36 mo in our ITT analysis, in comparison with the AT analysis (RR, 0.42 versus 0.29) (Table 6), may indeed indicate that there was a survival benefit for a number of long-term PD patients after switching to HD. Further in-depth analyses are required to substantiate the roles of small-solute clearance (Kt/V_urea) and other important characteristics of dialysis (such as fluid removal, BP control, and clearance of other solutes) in long-term mortality rate differences and to establish the potential survival benefits of the complementary modality approach (21).

The tendency toward a mortality rate difference in favor of HD with longer follow-up times was observed particularly for elderly patients, with diabetes mellitus or not. This is partly in accordance with other reports. Held et al. (11) observed a higher risk of death for older diabetic patients treated with PD. However, the effects of age, diabetes mellitus, and follow-up times were not examined in one multivariate model, and it is difficult to compare those results with our findings. Vonesh and Moran (3) also observed an increased risk of death associated with PD among older diabetic patients, especially female patients, but no adjustments for comorbidities were performed. In a recent study by Winkelmayer et al. (12), an increase in mortality risk during the first 1 yr of PD treatment for patients ≥65 yr of age was observed.

We observed a statistically significant RR in favor of PD treatment in the first 2 yr among younger diabetic patients. This finding must be interpreted with caution, because unobserved patient selection might have contributed to the large mortality rate difference, and the number of young patients with diabetes mellitus was relatively small. It might be argued that the greater fluctuations in volume status and BP with HD are harmful, especially among patients with diabetes mellitus, because of the preexisting vascular damage and failing adaptive physiologic responses to sudden fluid removal associated with neuropathy. Furthermore, beneficial effects of continuous ambulatory PD on metabolic control have been reported (32). In later stages, the possible disadvantages associated with HD may be outweighed by problems associated with long-term PD. A RR in favor of PD among young diabetic patients with ESRD was also reported by Nelson et al. (15). The later decrease in the RR in favor of HD, as observed for elderly patients, may indicate that timely transfer to the other treatment modality may also be a useful treatment strategy for this subgroup of patients. For younger patients without diabetes mellitus, the two treatment modalities seemed to be equivalent during both the early and later stages of dialysis treatment.

In conclusion, the results of this analysis indicate that long-term use of PD, especially among elderly patients, is associated with increases in mortality rates. Further analyses are required to determine the potential role of dialysis adequacy in the observed long-term differences in mortality rates between HD and PD patients and to establish the possible survival benefit for PD patients who switch to HD in time.