Mortality is theoretically the simplest and the most important measure to evaluate the effect of our intervention in a patient population with chronic renal disease.1 There are many publications comparing survival between the different stages of the disease,2 etiologies, renal replacement therapy (RRT) techniques3–12 or between groups and countries.13,14 However, there is some controversy and disparity of results in publications, and this is partly due to the lack of homogeneity in case definitions, statistical analysis approaches, the management of censored patients censored and forms of description and presentation of results.15,16 Therefore, different records of patients with chronic kidney disease in RRT have tried to define a model that summarizes the various ways of measuring mortality.17, without having reached a consensus model. It would facilitate the comparison between different working groups on an important subject such as a patient who receives RRT, and more specifically, peritoneal dialysis (PD). Our analysis is intended for that purpose, which applies different methods on the same database (DB), allowing us to compare and establish a methodological discussion on the subject, for the purpose of advancing our knowledge and comprehend the information available in this field.

The Peritoneal Dialysis Center Group (GCDP) presents a descriptive DB analysis that we have generated from a prospective data collection since 2003, as described in previously published.18 We analyzed data from 1890 patient incidents in PD from 22 centers grouped in the territory of the communities of Madrid, Castilla-Leon, Castilla-La Mancha, Extremadura and Aragon.

The information is collected on an anonymous DB, with design, management and approved analysis by a scientific committee that was independent of companies that support the project analysis. Patients consent to inclusion on entering in PD. Statistical analysis was carried out by SPSS v. 15.0 and STATA v12 for CR analysis.

We conducted a first descriptive analysis of the population in line with what was previously published.18 The second part of the study focuses on the mortality calculation in the last 7 years (2007–2013) using the methodology referred in various groups and regional, national and international registers (see Table 1) to compare and evaluate the reasons for the disparity.

Next, it expounds the main methods for estimating mortality in a population, describing their advantages and disadvantages.

It included 1890 patients with an overall follow up time of 3.633 years, with an average follow up of 1.92 years/patient (range 1 month to 10.5 years). At the end of follow up there were 261 deaths (13.8%), 380 transfers to HD (20.1%), 682 receive a renal transplant (36.1%), 40 recover renal function (2.1%). Thirty four patients were transferred to other centers external to GCDP (1.8%). Finally, the remaining 493 patients continue in PD (26.1%).

The descriptive analysis of the population is summarized in Table 2; it compares the characteristics of different groups of patients classified according to their situation at the end of follow-up. Age, gender distribution and comorbidity associated with Charlson index, diabetes mellitus and cardiovascular events before initiation of PD and residual renal function are shown in Table 2.

It is analyzed an incident population since 2003, so that during the early years there is a very low mortality rate when compared to prevalent, obtaining biased mortality data. Therefore, it was decided to perform the analysis of mortality from 2007 to 2013, inclusive, when the proportions have been balanced.

In Table 3 it is shown number of deaths and prevalents at the end of the year, accumulated time and mortality rates and lethality, according to the different methods revised.

In the mortality analysis through KM method (Fig. 1A) the median survival is 6.4 years (95% CI: 5.9–7.0).

Fig. 1.

(A) Cumulative incidence calculated by the Kaplan–Meier method. (B) Incidence. Cumulative mortality estimated using the method of multiple decrements. It shows the number of patients at risk at the bottom of the figure. Cumulative mortality through both methods is shown in the table.

(0.23MB).

Fig. 1B shows the results obtained using the multiple decrement method, that calculates the mortality by evaluating competing risks (transfer to HD, transplantation, renal function recovery).

When comparing both methods (KM vs. CR), we observed that cumulative mortality is similar in the first year. In the KM analysis, this percentage increases more rapidly than in the CR model, thus overestimating mortality.

Our paper aims to highlight the disparity between results obtained with different mortality analysis methods in PD.25 Each method is correct in itself and tries to reflect mortality, but does not express the same and transmits a different message to the reader and hinders comparison between publications. Performing these analyses on the same DB, that has been proven to be reliable and proven to, helps to understand the methodological details of these differences. For example, in 2012 rates vary from 5.2 to 6.4% depending on the method; this represents a difference greater than 20%.

In analyses performed by our group over 10 years, we observed differences in estimation of the mortality rates in early years, because the incident patients prevail over the prevalent.20,21 This difference disappears from 2006, and rates remain stable when the initial effect of higher survival rate of incident patients is no longer present.20,21

There are some limitations in the way that REER calculates the mortality rate.26 This record assumes that the population at risk is constant throughout the follow up time and, therefore, is the same as the last day of the year on average. This may not be so in programs in growth or decrease, because it generates more errors when the sample is smaller and greater the proportion of deaths. It can also be significantly affected by high transplantation rates (as in the Autonomous Community of Madrid). Therefore, it does not seem to be an appropriate way to calculate mortality, especially in small samples such as a particular nephrology service, where there may be excessive fluctuations in consecutive years. Our recommendation is to calculate the actual time at risk for each patient and reduce the rate per patient-year.

The mortality analysis through KM is complete and descriptive, but we should bear in mind that it should meet the conditions for the application of the method. The death risk for our patients tends to be lower at the beginning, and the model requires that this risk is uniform over time in PD. Also, there is a crucial percentage of PD outputs with living patients. For example, in our study and after 10 years follow up, the proportion of patients considered lost (TX, transfer to HD, renal function recovery) quadruples patient deaths (1162 vs. 261). Most records report PD distribution outputs in three thirds, nearly the same as death, transplants and change to HD.25 However, in our group, the TX represent more than 50% of total PD program outputs. At the end of our follow up more than half of our patients are follow up losses, without reaching death event. Therefore, Kaplan–Meier analysis may not express an accurate impression.

CR model proposed here, analyses these situations differently and better reflects what is happening with our patients. In this model, these patients lost differ according to their output (TX, HD, renal function recovery) establishing each as an independent event that competes with the rest. Thus, it is expected that transplant patients had a less life-threatening risk than those who transfer to HD, which can thus explain the differences in the TX/HD relationship between analysis using KM or CR.

Despite being the best suited model from our point of view, there are also some limitations in the analysis carried out. The first is the lack of follow-up when the patient abandons the technique, unable to determine the evolution of each one of them. For example, there are situations where certain terminally ill patients are transferred to HD and die a few weeks or months later. Another limitation is the actual time in the technique that the patients provide in the analyzed data. The average duration of the procedure per patient is too short (approximately 2 years) and remains stable after 10 years of DB operation. As already published in other articles of our group, this short follow up time is a consequence of the high rate of departures to TX in a population that, by demographic characteristics and privileged clinical situation, has faster access to a kidney transplant as definitive TRS.27

Finally, we must consider other factors that influence other PD mortalities in each registry.28 Aspects such as the assignment/election policy of the first RRT can affect the different profile of patients in each technique. Mandatory allocation models (e.g. Hong Kong model) or priority allocation (model PD-first) behave differently than our Free election.29 In the Spanish model, it observes a PD inclusion rate around 15% and a positive selection of younger patients and less morbidity for this technique. We must also consider the general characteristics of the population, with different ages or prevalence of obesity and diabetes mellitus, which is very relevant when compared with records from the US. Finally, as the US records exclude patients who do not survive the first 90 days of treatment, making it impossible to compare with our results. The applicability of our environment analysis of other groups or records depends on these factors, which determine the external validity of the published work.

The GCDP DB is filled in prospectively by the clinics directly responsible for patient clinical follow up, which ensures greater accuracy than those administrative forms used in other records.

The actual mortality in a population might seem over – or underestimated depending on the chosen method to analyze and present. This analysis, using different methods on the same DB, helps to understand these differences. Finally, it seems reasonable, given the disparity of formulas, establish a shared method for all registries to standardize a unique system for comparing information from different international registries. CR model is appropriate in situations of high rate of losses in the follow up.

The authors declare no conflict of interest. Multicentre project funded by Baxter, Fresenius and Amgen through the Madrid Nephrology Foundation.