Survival Percentile and Predictors of Difference in Survival among Hemodialysis Patients and Their Additive Interaction Using Laplace Regression

Background: Identifying survival modifiable factors and additive interaction between them could help in prioritizing the clinical care of Hemodialysis (HD) patients. We aimed to examine the survival rate and its predictors in HD patients; and explore the additive interaction between survival modifiable factors. Study design: A retrospective cohort study. Methods: The present study was performed on 1142 HD patients in Hamadan Province, western Iran from 2007 to 2017. Data were collected through a researcher-made checklist on hospital records. Laplace regression was used to evaluate differences in 40th survival percentiles in different levels of predictors as well as exploring the pairwise additive interactions between variables. Results: We observed significantly higher survival in nonsmoker patients (40th percentile difference = 5.34 months, 95% CI: 2.06, 8.61). Survival was shorter by more than 3 years in CRP positive patients (40th percentile difference=36.9 months, 95% CI: 32.37, 41.42). Patients with normal albumin (40th percentile difference =24.92, 95% CI: 18.04, 31.80) and hemoglobin (40th percentile difference = 18.65, 95% CI: 12.43, 24.86) had significantly higher survival (P<0.001). There was super-additive interaction between being CRP negative and nonsmoker (β3 = 9.42 months, 95% CI: 3.35, 15.49 (P=0.002)). Conclusion: High CRP and low serum albumin and hemoglobin were associated with the increased risk of death in HD patients. The results of this study support the presence of super-additive interaction between CRP status with serum hemoglobin and also CRP status with smoking, resulting in excess survival in HD patients.


Introduction
lobally, chronic kidney disease (CKD) is a major health concern and an excessive cost of health care finances devoted to this problem 1 . The survival rate of these patients is lower than the general population, and it was not seen any improvement in their survival over recent years 2 .
Hemodialysis (HD) in Iran is generally used as the main choice for renal replacement therapy in end-stage renal disease (ESRD) patients and is provided free of charge in Iran 3 . Until now, limited studies have been performed in Iran regarding the survival of HD patients, indicating a low survival rate for these patients 4,5 .
According to litterateurs, age, ethnic background, level of serum albumin and hemoglobin, adequacy of dialysis, mean duration of dialysis per treatment session, renal replacement therapy method, body mass index (BMI), causes of kidney failure, and comorbidities with some diseases such as heart failure and cancer are considered as predictors of death in the HD patients 3, 6-9 .
The interaction generally occurs as a result of the dependence of the influence of one risk factor to the presence of another risk factor. When there is additive interaction between the two risk factors that the patients with both risk factors simultaneously have higher risk of death than those expected based on summing the separate effects of the abovementioned two risk factors 10 . Assessing interaction provides a better vision into the mechanisms for the occurrence of outcome and identify more beneficiary subgroups to be intervened when resources are limited. Therefore it is one of G the stimulants for assessing additive interaction 10,11 . A strikingly increase showed in mortality risk among dialysis patients which was due to interaction between protein-energy wasting (PEW), cardiovascular diseases (CVD) and inflammation in patients 12 .
Identification of the additive interaction between survival modifiable factors could help in prioritizing the clinical care of HD patients. To our knowledge, there was no previous study done on the additive interaction between survival modifiable factors in hemodialysis patients, moreover, evidence about the survival of hemodialysis patients and its related factors in developing countries is limitedd.
Therefore, the purposes of the present study were (1) evaluating the survival rate and its predictors in HD patients in Hamadan province, and (2) investigating the additive interaction between the survival modifiable factors in HD patients.

Study Design
Retrospective cohort study

Settings and Participants
We examined data obtained from 1142 hemodialysis patients in Hamadan Province, western Iran in the period of 11 years from Mar 2007 to Mar 2017. Hamadan Province is located in the west of Iran and has an area of 19,493 square km in extent and 1,758,268 population according to the national census by the Statistical Center of Iran in 2011. Information was obtained from the eight hospitals of the province with dialysis centers including Alimoradian, Besat, Vali-asr, Ghaem, Imam Hossein, Valiasr, Shahid-Beheshti and Imam Reza in Nahavand, Hamadan, Tuyserkan, Asadabad, Malayer, Razan, Hamadan and Kabudarahang city, respectively.
Patients undergoing HD due to acute renal failure, patients undergoing peritoneal dialysis, patients on transient hemodialysis and patients who had incomplete medical records were not included in the study and were considered as exclusion criteria.

Clinical and Demographic Measures
Data were collected using a checklist on hospital records of all HD patients hospitalized in provincial hospitals. The checklist used in this study included characteristics related to demographic profiles (age, gender, marriage status, Body mass index (BMI), location, educational level, previous history of smoking or substance abuse), and patient biochemical and clinical information (Hemoglobin level, C-reactive protein (CRP) status (+/-), blood urea nitrogen (BUN), creatinine, urea reduction ratio (URR), sodium, phosphor, calcium, albumin and etiology of ESRD). Clinical and biochemical data at the time of diagnosis and before onset of the first dialysis were gathered for each patient as well and considered as baseline data. To minimize measurement variability, we averaged both baseline measures for each patient. These records were gathered by assessing patients' medical records in the dialysis ward.

Outcomes
We considered death due to renal failure as the endpoint of the study. Patients with renal transplantation, withdrawal of dialysis, lost-to-follow-up, competing risks (patients who died due injury, accident, or other causes unrelated to renal failure) and those transferred to another dialysis facility out of province were treated as censored cases.

Analytical Methods
Laplace regression was used to evaluate differences in survival percentiles according to the levels of predictors and adjusting for potential confounders. Laplace regression as a flexible method can be used for computing the conditional percentiles of the time-to-event variables 13 . The time that the specific percentage of the investigated cases have experienced the outcome of interest can be considered as survival percentile for that time.
To avoid extrapolation, we examined the 40 th percentile of survival time given that during the study period, death occurred for 43% of participants. Therefore, using Laplace regression, we estimated differences in the time duration by which the first 40% of the HD patients died according to the levels of predictors. We also assessed pairwise, additive interactions between predictors.
To assess the effect of the two binary predictors (e.g. G and E) and their additive interaction on the 40 th survival percentile (p(40)), we fitted the following Laplace model: (40)). . " According to the above equation the measure of additive interaction between two predictors is the parameter β3(p(40)). If β3(p(40)) >0, superadditive interaction between two predictors exists and If β3(p(40)) <0, interaction is subadditive 14 .
We analyzed the data using Stata software version 12 (Stata Corp LP, College Station, Texas) at less than 5% significant level.

Ethical approval
The Ethics Committee of Tehran University of Medical Sciences (TUMS.SPH.REC.1395.1300) approved our study. To keep confidentiality, all patients identifier were removed.

Results
The baseline characteristics of HD patients are shown in

The interaction between smoking and CRP status
The additive interaction between smoking (0 = smoker; 1 = never smoker) and CRP (0 = CRP positive; 1 = CRP negative) in predicting overall mortality were assessed. Predicted values of the 40 th survival percentile for each of the four subgroups formed, calculated by combining the obtained coefficients estimates, have been shown in Table 2 and Figure  1a. The Laplace regression model indicated superadditive interaction in predicting the death between being nonsmoker and CRP negative: 9.42 months' excess in 40 th survival percentile for the effect of one predictor when the other predictor changes from 0 to 1. (β3 = 9.42 months, 95% CI: 3.35, 15.49 (P=0.002)).

The interaction between Hemoglobin level and CRP status
The interaction between Hemoglobin level (0 = Hb<11 mg/dl; 1 = Hb ≥ 11 mg/dl) and CRP status (0 = CRP positive; 1 = CRP negative) in predicting overall mortality in 40 th survival percentile of HD patients has been shown in Table 3 and Figure 1b. The estimate of the product term β3, suggested 5.26 additional months of 40 th survival percentile for the effect of one variable when the level of the other one varies from 0 to 1 (β3 = 5.26 months, 95% CI: -3.31, 14.81 (P=0.21)). This excess in survival indicates the presence of superadditive interaction in predicting mortality between being normal hemoglobin level and CRP negative.

Discussion
Hemodialysis is a common treatment modality for ESRD patients in Iran. We investigated the predictors of survival in HD patients and explored the additive interaction between them in terms of survival. The findings of this study are useful as preliminary data for further studies in order to increase the quality care of HD patients. High CRP and low serum albumin and hemoglobin are independent predictors of mortality in HD patients. Being male, rural dweller and smoking were associated with a higher risk of mortality. There was an additive interaction between CRP and each of the variables serum hemoglobin and smoking on the (40th percentile of) survival.
In this regard, the low level of hemoglobin was associated with a higher risk of mortality in HD patients, which is consistent with our findings 15,16 . In a study, three-year survival rate of patients with Hb <9 g/dL was significantly lower than that for patients with Hb levels 10 to 11 g/dL (74.1% vs. 89.3%) 8 . The normal range of the hemoglobin level in HD patients is correlated with improving life quality 17 , cardiac and brain function 18,19 , decreasing hospitalization and treatment costs in these patients 20 .
In the present study, in agreement with several conducted observational studies, we found that low level of serum albumin was associated with poor survival 8,21,22 . Decreasing serum albumin levels with increasing time was associated with raising CVD related death 23 . In dialysis patients, hypoalbuminemia is applied as indicator of malnutrition and has a strong effect on mortality 24 .
Our findings revealed that the positive CRP was a strong predictor of death in HD patients. In consistence with our findings, several prospective studies have demonstrated CRP is an independent predictor for the future risk of death in these patients 25,26 . In general, inflammation is associated with wasting, oxidative stress, insulin resistance, endothelial dysfunction, and infections 17 . CRP can mediate processes involving in the development of atherosclerosis through plaque initiation, formation, and rupture, while it may not be merely a marker of inflammation 27 .
Our results showed a significant association between HD patients' smoking and their survival rate, which was inconsistent with the findings of another study, which showed there was no significant relationship between smoking and survival rate 28 . A meta-analysis study reported a same results as our findings 29 .
Smoking as a modifiable risk factor for kidney failure through some mechanisms like excessive generation of free radicals, promoting atherosclerosis in renal arteries, and intrarenal hemodynamic changes 30,31 .
Additive interaction scale as a more relevant public health measure helps detect subgroups with the highest benefits from treatment 10 . The findings of this study support the presence of the super additive interaction between CRP status and serum Hemoglobin also CRP status and smoking. This finding is important because HD patients with a high mortality risk can be identified through regular screening. More studies are required to determine that multiple pathophysiological pathways may underlie these interaction effects.
Our study has suffered from some key limitations which had been inherited from existing data. Firstly, the absence of a Kt/V as an accurate reference method to estimate dialysis adequacy of patients. Second, due to its retrospective design, it was not possible to control data quality. Third, the addiction and smoking status of patients was based on their self-report and were prone to measurement bias. Fourth, the retrospective and observational nature of our analyses allowed the detection of associations, not causation. However, moderately large sample size, comprehensive clinical and laboratory evaluations and examining the additive interaction between the predictors of HD patient's survival, can be considered as the strengths of our study.

Conclusion
High CRP and low serum albumin and hemoglobin are associated with the increased risk of death in HD patients, and male gender, rural dweller and smoking were significantly associated with a higher risk of mortality. The presence of super-additive interaction between CRP status and serum hemoglobin also CRP status and smoking, resulting in excess survival in HD patients. These findings can help screening programs to identify patients with a high mortality risk.