The Dialysis Outcomes and Practice Pattern Study (DOPPS) study conducted in 11 countries showed that the highest mortality of HD patients was observed in the first month after dialysis [16]. It is well documented that the mortality of HD patients is higher within three to six months after initiation of dialysis. According to the United States Renal Data System (USRDS) report [1], all-cause mortality peaked about two months after dialysis initiation in HD patients. Therefore, the high mortality rate of dialysis patients in the early stage of HD should not be ignored. In the present study, we developed and validated a model for predicting all-cause mortality risk among incident HD patients in the first 6-months using five easily available baseline variables.
The five predictors were: age, temporary dialysis catheter, intradialytic hypotension, use of ACEi or ARB, and use of loop diuretics. Traditional risk factors for death and dialysis-related factors were included. The easy and calculable score described here was designed to identify HD patients who were at high risk of death during the first six months after initiation of dialysis. This model can not only identify patient risk factors for early death, but also help health care workers to implement targeted treatment measures for patients. Identifying death risk factors for dialysis patients in early stage can help initiate early interventions for those at risk. Among the risk factors include hypertension and hypotension management, choice of the dialysis access, and strategies for the use of ACEi or ARB or diuretics in different populations.
Multiple studies, including ours, have reported several clinical models for predicting all-cause mortality in HD patients in the past years [17,18,19,20,21]. a numerous mortality scores for dialysis patients have been established on the basis of various comorbidities and laboratory data, but only a few can predict the short-term survival. Therefore, few data are available for developing tools for predicting the risk of early death in HD patients.
Although scoring systems for elderly HD patients have been reported in multiple countries [19,20,21], these clinical models do not include Asian populations. A prognostic score was developed to predict the 6-month prognosis of elderly patients in French HD patients [19]. In the study, 9 risk factors were identified. Among them, unplanned dialysis overlapped with the temporary dialysis catheter identified in the present study. Other factors were unique to the clinical model (e.g., congestive heart failure, peripheral vascular disease, cancer, and serious functional limitations, BMI, diabetes, arrhythmia, and severe behavioral disorders). However, they found that age was not an independent risk factor of mortality, which differs from our model and other clinical models. Therefore, the application of their model to Chinese HD patients may be limited given the important differences in practice patterns. Thamer et al. [20] developed a clinical score to predict mortality in the first 3 and 6-months based on US Renal Data System comprising 7 predictors. of which “age” is the similar factor to our model. However, the other 6 predictors were not available in our data. Wick et al. [21] utilized a big population-based data source in outpatient settings to develop a score for elderly dialysis patients. Their model for predicting the 6-month mortality included 7 predictors: age (≥ 80 years), increased eGFR, hospitalization in the prior 6 months, atrial fibrillation, congestive heart failure, lymphoma and metastatic cancer, none of those variables except older age were strongly predictive in our model. These three studies could be related to differences in the populations from which they were included (Chinese as opposed to Canadian or American or French). In addition, some factors, such as “cancer” were not included in our inclusion criteria. The risk scores reported by Thamer et al., (AUC = 0.69–0.72), Couchoud et al., (0.68–0.74) and Wick et al., (c-statistic = 0.72) showed fair performance in predicting the risk of early death in HD patients. Compared with these three models, the tool established in the present study showed good discrimination (c-statistic = 0.775).
In another prediction model for predicting early mortality in United Kingdom HD patients reported by Wagner [22] et al., several clinical variables were identified among which two risk factors were used in the presented study (i.e., age and dialysis modality). Our risk prediction model included 3 variables that were not included in previous tools, namely intradialytic hypotension, use of ACEi or ARB, and use of loop diuretics. In this study we found that patients with intradialytic hypotension had a higher mortality compared with those with normal or hypertension in the first six months after initiating dialysis. It has previously been reported that intradialytic hypotension is a common complication of HD patients, which may be associated with decreased blood volume, autonomic nervous dysfunction, cardiac dysfunction, and vascular dysfunction during dialysis [23]. Young et al. found that ACEi and ARB have different efficacy in regulating hemodynamics, cardiovascular remodeling, cardiovascular events, and all-cause death in HD patients [24]. However, in this study, we found that HD patients using ACEi or ARB had a lower 6-month survival rate, which differs from previous studies and may be related to hyperkalemia. Movilli et al. reported that ACEi/ARB treatment increased the risk of hyperkalemia in anuric HD patients suggesting that great caution should be applied in the wider utilization of this class of drugs in anuric HD patients [25]. In a 3-year study of 74,000 HD patients, Sanghavi et al. found that a pre-dialysis serum potassium concentration of more than 6 mEq/l was associated with 50% higher risk of cardiovascular mortality and all-cause mortality [26]. These results suggest that hyperkalemia caused by ACEi/ARB or other factors may be risk factors of death in HD patients. However, the effect of ACEi/ARB on HD patients is still controversial which need to be clarified in future clinical studies. It has been reported that continued use of loop diuretics during the first year of dialysis is associated with lower hospitalization rates, lower intradialytic hypotension rates, and lower interdialysis weight gain, but does not affect mortality [27]. Herein, the results showed that use of loop diuretics before dialysis initiation reduced the risk of death within the first six months. In addition to increasing urine output, loop diuretics improve sodium excretion by about 20% and is unaffected by the levels of eGFR in different types of kidney disease [28], similarly, use of diuretics was shown to increase urine volume, sodium and potassium excretion in dialysis patients [29] and Bragg-Gresham et al. reported that volume managed with diuretics had a 7% lower all-cause mortality risk and 14% lower cardiac-specific mortality risk in HD patients which is similar to our finding [30].
Our clinical model has a few strengths. Firstly, to our knowledge, this is the first prognostic score for predicting early death (within 6 months) in HD patients that is developed and externally validated in a Chinese population. In some previous studies, the models were only internally validated. Second, the constructed nomogram is simple, practical, and robust, all the variables can be collected easily and the risk of early death can be calculated within a short time. Finally, in most current clinical prediction models, drug factors have not been sufficiently considered in development of clinical prediction models, which may limit the application of their findings. Our model incorporates drug factors making it more comprehensive and accurate.
Nonetheless, this study had some limitations. First, the sample size was small, which may increase the possibility of type II errors. In addition, only variables subjected to univariate analysis with P < 0.05 were selected for Cox analysis, which to some extent eliminated some risk factors that affect death. Therefore, a study with larger sample size should be conducted to confirm our findings. Second, although the nomogram was subjected to extensive internal validation using bootstrap testing, its performance in other HD patients remain to be clarified. Thus, external assessment should be conducted in wider HD populations. Finally, the eGFR at HD initiation are significantly different according to the used various eGFR equations [31], in this study, the GFR was estimated using the CKD EPI-equation which may result in a lower eGFR value than Cockcroft-Gault equation and MDRD equation at the HD initiation [31]. In addition, the use of calculated GFR instead of measured GFR by radionuclide imaging may not reflect true GFR levels in these patients, therefore the assessment of baseline GFR at the HD initiation may be biased to some extent, and whether baseline GFR is one of the factors affecting the 6-months survival rate in HD patients requires further clinical verification in the future.