In this study, we described the development and validation of a simple comorbidity score that summarizes in one index several risk factors to predict the one-year mortality in patients with ESRD. Our results suggest that the inclusion of comorbidities recorded at dialysis start in the REIN database, and not depending on dialysis parameters, is sufficient to construct a score to predict the one-year mortality risk. Indeed, additional information from the medico-administrative database did not increase the score ability to predict mortality. Nevertheless, diagnoses derived from hospital stays in the two years before dialysis initiation (extracted from the SNDS database) allowed us to verify and complete REIN data. Moreover, the new Rennes score is a good predictor of mortality and outperformed previous scores (Charlson and Wright comorbidity indexes). This score let clinicians to identify patients having a high risk of one-year-mortality before dialysis initiation and could help them to improve the patients’ personalized management regarding to dialysis initiation.
Our new score is simple to use because it has been established using only five comorbidities, one laboratory parameter, and age at dialysis start. This score could be calculated even before dialysis start because no dialysis-dependent item was retained. We observed that the first dialysis condition (in emergency or as a planned procedure) did not significantly modify the Rennes score.
Despite the use of only seven variables, the Rennes score outperformed recent scores (not directly compared in this study) developed using a large European cohort of patients on hemodialysis9 and data from the United States Renal Data System10. Floege’s score included many factors, but the observed AUC (0.73) was “acceptable, but not excellent”9. This score is not easy to use because it requires collecting several biological parameters (e.g., ferritin, LDL-cholesterol…). Moreover, it is not generalizable to all patients with ESRD because it was constructed specifically for people on hemodialysis. Liu et al., established a score based on 11 comorbid conditions in addition to the primary renal disease. Liu’s score outperforms Charlson comorbidity index, but its ability to predict mortality is low (AUC = 0.669). Moreover, it was developed using data from patients dialyzed in the early 2000s10, and patients’ medical conditions at dialysis start and dialysis practices might have changed in the last years.
In our study, we compared our score with the Wright comorbidity index, developed in the early 1990s and adapted specifically to a small population of patients with ESRD8, and also to the well-known Charlson comorbidity index3, developed in the general population in the 1980s. The Wright comorbidity index, as defined by the author, did not allow categorizing our population in three homogeneous subgroups and could not predict mortality (AUC = 0.631), as recently observed by McArthur et al.20. Wright’s index was developed using data from a small population of patients with ESRD dialyzed in the same unit between 1984 and 1988. Moreover, it was based on literature data of that time suggesting that early survival on dialysis was limited mainly by age and presence of diabetes or coronary artery disease. Nevertheless, in our study, diabetes was not significantly associated with the risk of death, and this variable was not included in the Rennes score. This result could be explained by the fact that diabetes treatment has changed in these last decades and this condition is currently not considered as a major risk of death for dialyzed patients. Consequently, due to the changes in the management of dialyzed patients, the Wright comorbidity index cannot adequately predict the survival of patients with ESRD and should be updated.
We then compared our score to the Charlson comorbidity indexes. First, we used the score that included 15 comorbidities (leukemia, lymphoma and solid tumors were grouped in one variable, and none of our patients had HIV/AIDS). Renal disease was not considered in the score because all dialyzed patients had ESRD. In our cohort, the original Charlson comorbidity index had a low ability to predict the 1-year mortality (AUC = 0.622). This improved when the patient’s age was included in the score (AUC = 0.703). Indeed, without the age variable, a large percentage of elderly patients with few comorbid conditions were grouped in the low-risk group. After the inclusion of age, the score distribution was more parsimonious and elderly patients were included in the higher-risk group. These results confirmed the value of age in a comorbidity score and its association with survival as previously observed9,25.
Inclusion of data from hospital stays that occurred two years before dialysis initiation (SNDS database) did not improve the prediction ability of our score compared with the model based only on risk factors from the REIN database. This indicates that data from the REIN registry are sufficient to develop a strong score; however, data from the SNDS allowed completing missing data because comorbidities are not mandatory items in the REIN registry. For instance, if a patient was hospitalized for a cancer two years before dialysis initiation, but the item was missing or filled as absence of cancer in the REIN registry, we could modify the cancer status of this patient. Our approach was to complete data from the REIN database using diagnoses from the SNDS database, but not to assess the quality of the registry, as performed earlier in Canada26,27, Australia and New Zealand28 and also in the United States29. Indeed in our study, 33.2% of included patients did not have any hospital stay during the two years before dialysis initiation, and therefore this complementary analysis to complete/confirm their comorbidity list could not be done for all patients.
The strengths of our study are that we established a simple mortality risk score based on few variables that are easy to collect. We developed an open access application in English and French to easily calculate the Rennes score (https://apladys.shinyapps.io/Rennes_score/). We tested and cross-checked two models to identify the contribution of a medico-administrative database to the establishment of our comorbidity score. Finally, we used only data from the REIN registry to develop the Rennes score that displays a good ability to predict the one-year mortality in dialyzed patients. Moreover, we showed that the Rennes score outperforms the widely used Charlson comorbidity index and also the Wright score developed for dialyzed patients. In addition, thanks to a linkage procedure established by our team, we could link SNDS data to the REIN registry for the first time in France.
Our study has also several limitations. If patients were not hospitalized during the two years before dialysis initiation, their Charlson score could not be calculated. Consequently, we might not have the full comorbidity picture of all our patients. Moreover, we validated our scores using incident French patients, whereas an external validation population could have been more suitable.