As a home-care therapy, PD requires patients to self-monitor and self-manage their treatment [8], [9], [10], abilities that are closely associated with socioeconomic status (SES) [11], [12]. SES has been investigated for its association with the outcomes of treatment in the general population and in patients with chronic kidney disease at the individual [13], [14], [15], [16], [17] and environmental [13], [15], [16], [17], [18] level. A few large-scale multi-center cohort studies have also explored this issue in PD patients and shown inconsistencies in the impact of individual [19], [20], [21], [22], [23] and environmental [21], [22], [24], [25], [26], [27] SES on PD outcome. In China, we have experienced great economic development in the last 20 years, with an increase in gross domestic product (GDP) of around 10% per year, but there are imbalances between regions. Individual income and education levels also vary markedly within regions and probably lead to diverse availability of medical facilities and health-care services. Given this situation, a specific model is needed to determine whether variations in individual or environmental SES influence PD outcome. Moreover, the interaction between individual and environmental SES on PD outcome, which has never been studied in a dialysis population, should be explored.

Consistent with former studies [19], [33], [34], [35], lower individual income emerged as an independently significant risk indicator for death and initial peritonitis. Although poorer patients were younger and had less comorbidity, they bore a heavier burden of medical expenses and had access to fewer medical resources, which strongly supports increasing the personal medical coverage rate for PD patients in China. Of note, poorer patients were prone to be anemic and malnourished at baseline, probably due to inadequate health care and late referral to nephrologists before the development of ESRD. Given that the standard chronic kidney disease program is helpful in retarding progression to ESRD and improving complications such as anemia, malnutrition and bone mineral disease [36], [37], the timing of and strategies used for medical support before dialysis in China need to be urgently investigated. We note that effective medical support aids the establishment of successful PD programs for disadvantaged minorities [38], [39].

By contrast, no impact of regional GDP per capita on PD outcome was observed, the opposite of previous findings showing an independently negative effect of regional SES on outcome in chronic disease [15], [16], [17]. One potential reason for our finding is that, in the undeveloped regions in our study, the patients were younger, there were fewer cases of diabetes, and baseline serum albumin levels were higher than in the developed regions. Given that age, diabetes and serum albumin have been recognized as the strongest predictors by most studies [40], [41], [42], these favorable individual factors in our study possibly offset the disadvantages of regional SES. A similar finding was obtained in a study from the USA [24], in which the risk of technique failure was significantly lower in remote-dwelling patients than in those living closer to the hospital, because the former were younger and had fewer complications from diabetes. However, no association between regional SES and cardiovascular or all-cause death or initial peritonitis was observed, even after adjustment for age, proportion of patients with diabetes, serum albumin, and so on. We hypothesize that unknown confounding factors related to regional SES are also associated with PD outcome. In addition, only baseline characteristics, and not the change trend in clinical variables, were analyzed for the prediction of outcome during follow-up. This choice may have influenced the reliability of our results.

Our analysis indicates that education level had no effect on risk of initial peritonitis. This result is consistent with the analysis of a US regional ESRD registry in which 1595 new PD patients were observed over 2 years [23], but is contrary to recently published data from Brazil and Canada [21], [22]. One possible explanation is that our selected centers had professional PD doctors/nurses and well-constructed training programs. Patients and their home-care helpers were often trained simultaneously, which probably led to stronger family support [43]. Whether better compliance in Asian people [44], [45] plays a role in this phenomemon is unclear. However, lower education level still significantly predicted death in undeveloped regions. The causes for this finding are unknown, but it has been shown that PD networks linking developed and developing units might be a means of improving the quality of therapy [46]. Whether the establishment of a standardized PD program in undeveloped regions will benefit less-educated patients needs to be investigated. More trials are also needed to explore new strategies for improving the efficacy of training for less-educated patients.

Our large-scale multicenter PD cohort gave us a valuable opportunity to explore the association of SES and outcome at both the individual and the environmental level. Our results will be helpful for PD clinicians and health policy-makers in generating appropriate strategies to improve the use of PD in developing countries such as China. Compared with former studies, this study has the advantage of more detailed information on individual SES. The enrollment of representative centers in a rapidly developing country with huge diversity in regional economic development is also a merit. The interactive effect between individual and regional SES has never been investigated in a dialysis population.

This study has some limitations. First, based on estimates of the size of the dialysis population (no registry data available as yet), only about 10% of PD patients in China were enrolled in the study and thus we cannot confirm the generalizability of our results. However, all incident patients were enrolled from ‘core’ PD centers of medical school-affiliated hospitals and came from provinces and counties with varied level of SES and penetration rate of PD therapy. Therefore, it is reasonable to conclude that the range of SES of our study population reflects the overall situation to a certain degree. A national dialysis registry has been initiated recently, so more representative data will be obtained in future years. We Hope our study will provide useful cues for the analysis of national data in the future. Second, individual SES is a general index and many SES-related physiological and non-physiological factors were not measured. We cannot verify whether worse nutritional status, deficient pre-dialysis care, less access to standardized training courses, or poor compliance contributed to the worse outcome observed in poorer patients. Regional SES-related information such as local hygiene status, availability of medical services and social support systems was not assessed in detail. Third, we should be aware of the possibility of ascertainment bias (4.1% of eligible patients were not included), vintage bias (different centers created their databases at different times), and residual confounding and recall bias because of the retrospective nature of this study. Furthermore, an observational study cannot demonstrate cause–effect relationships.