Cox regression analysis of compositional covariates related to death of the kidney transplant-recipient in northeastern Brazil : modeling of covariates associated with renal allograft failure

Introduction: Kidney transplant (KT) has the highest survival rate amongst kidney replacement therapies (KRT). Objective: Analyze the incidence density of all-cause mortality in chronic kidney disease transplant-recipients and to identify covariables associated with higher risk of death. Methodology: Cohort study using medical records of 605 KT patients with seven years follow-up (2011-2018). Records with insufficient data or from patients with incomplete treatment were excluded. The variables analyzed were demographic, clinical and laboratory data, duration of KRT, type of donor, immunological compatibility, panel-reactive HLA-antibody, infections, and use of hypothermic perfusion machine (HPPM). Hazard ratio (HR) and incidence density of all-cause deaths were estimated. Results: 15 of 553 KTrecipients died during the follow-up. The survival in the first year post-KT was 98.0% and in the fifth year was 93.2%. The incidence density of deaths is 10/1,000 person-years. Variables preand post-KT related with higher death risk were allograft pyelonephritis ≥6-months and delayed graft function >4 weeks. Survival among KT-recipients with loss >5 mL/min/1.73m/year in the estimated glomerular filtration rate (eGFR) were lower than the others (88% vs. 97%). Covariates associated with mortality post-transplant included pre-KT obesity, HPPM, allograft pyelonephritis, and new-onset diabetes after transplantation. Conclusion: The mortality post-KT is low in these population. Cox's modelling demonstrated that the decline in eGFR >5 mL/min/1.73m2/year, allograft pyelonephritis ≥6-months, pre-KT obesity, fasting blood glucose ≥126 mg/dL presented worst probability of survival. Rapid decline in eGFR reduces substantially the survival probability in these population. Research, Society and Development, v. 9, n. 11, e84791110276, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i11.10276 3

Keywords: Kidney transplantation; Progression of kidney function; Survival analysis; Cox regression model.

Introduction
Kidney transplantation (KT) is the kidney replacement therapy (KRT) with the longest survival rate amongst all the therapeutic options available. The survival of chronic kidney disease transplant recipients (CKD-T) has been increasing in recent decades, possibly due to advances in immunopharmacology and improved care (Wekerle et al., 2017), together with reduction of infections, early identification of acute T-cell mediated rejections (aTCMR) or acute antibody-mediated rejection (aABMR) (Ashby et al., 2017).
Despite improved graft and patient survival in the first year after KT due to more appropriate clinical control and diagnosis, together with improved treatment of bacterial and viral infections, and rejection, each subsequent year increases the risk of the recipient returning to KRT or evolving to death even with a functioning graft. Long-term CKD-T survival may involve processes linked to malignancy, infections related to immunosuppression, infarction and stroke (Van Loon et al., 2020). The analysis of survival rates and the components of mortality-associated covariates are crucial in the planning and evaluation of CKD-T care (Djamali et al., 2006). Development, v. 9, n. 11, e84791110276, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i11.10276 5 Therefore, the primary purpose of this study is to analyze the cumulative incidence of all-cause mortality in CKD-T from a reference center in Brazil. The research also aims to model a Cox regression capable of determining the composition of pre-and post-KT covariates which are associated with the mortality post-transplant.

Immunosuppressive protocol at the RHP/PE (IP-IMS/RHP-PE)
All CKD-T underwent the IP-IMS/RHP-PE immunosuppressive protocol. This protocol used 15 mg/kg of methylprednisolone (MP) up to two hours before induction with maximum dose of one gram.
Recipients-sensitized patients were induced using Thymo (6mg/kg), administered at doses of (2mg/kg) on days zero, three and six post-KT. The IP-IMS/RHP-PE maintenance was administered using TAC + MPS + PRED. PRED was administered throughout maintenance at doses of 0.5 mg/kg with maximum dose of 30 mg/day (Van Loon et al., 2020;Wekerle et al., 2017).
The post-KT estimate glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula (Levey et al., 2009).

Diagnosis and management of cytomegalovirus disease (CMV)
Polymerase chain reaction (PCR) was used for the diagnosis of CMV. For the management of CMV, either prophylaxis with oral ganciclovir/valganciclovir, according to availability, or preemptive treatment with intravenous ganciclovir was employed, using PCR for diagnosis of CMV biweekly, resulting in a number of copies > 2,000/mL, collected from 15 days post-KT.
The protocol was carried out for three months in CKD-T with negative serology with positive donors, and in recipients who received induction with Thymo (6mg/kg) or maintenance with MPS. CKD-T using TAC + EVR or SRL + PRED did not undergo the preemptive strategy.
The same conduct was also carried out for three months, in all patients who were treated for rejection or were converted from MPA to SRL or EVL 10. All CKD-T showing signs or symptoms compatible with CMV syndrome, with laboratory confirmation, were diagnosed with CMV disease. In some cases, a histopathological study was necessary to diagnose invasive CMV disease or in the tissue (CMV nephritis). As a cure criterion, negative PCR for CMV was used. A CMV event was defined as any treatment registered in medical records preemptive or CMV disease treatment with antivirus therapy with ganciclovir registered in (Levin et al, 2013).

Bacterial allograft pyelonephritis (ICU) in CKD-T
ICU was determined by a value >105 colony-forming unit per milliliter in the urine culture, and fever associated with one of the following signs or symptoms: painful allograft, worsening of serum Cr or biopsy detecting bacterial neutrophilic nephritis.

Statistical analysis
Statistical analysis was conducted using Stata version 14.0 (StataCorpLLC™, College Station, Texas, USA). Absolute and relative frequencies were used to describe the clinical characteristics associated with kidney disease.
Quantitative variables were classified into categories and measurements were standardized. Mortality rates (MR) were expressed as incidence density of deaths with a 95% confidence interval (CI95%) and the number of cases standardized per 100 or 1,000 personyears of observation time. Pearson's chi-square (X 2 ) or Fisher's exact tests were used to assess associations of clinical factors related to CKD-T pre-and post-KT. The measures of the association of hazard ratio (HR) were also estimated for each covariate.
Survival functions were presented by the Kaplan-Meier graph and the differences were tested using the logrank test (Mantel-Cox) (Chuang, Parikh & Langone, 2005). In the survival analysis, the outcome variable was death and covariate dependent on time until death or the end of the cohort follow-up. Cases of loss to follow-up or abandonment of treatment were censored. The risk of death was estimated using the Cox-model and the measure of association of covariates was HR with CI95%. The Schoenfeld residue test was used to determine the proportional hazards assumption (Armitage, Berry & Matthews, 2013).
The HR of the association of mortality with clinical and CKD-T related pre-and post-KT factors was adjusted by stepwise-forward modelling. For input criterion, P < 0.20 was used in the bivariate stage while P < 0.10 was used as exit criterion. A bilateral P value < 0.05 was considered statistically significant. Biological plausibility served as a basis for interpreting concomitance at the stages (Schoenfeld, 1982). The quality of the goodness-of-fit adjustment was performed after modelling. Harrel's probability (Harrel's C index) was used to assess the discriminatory power and predictive accuracy of the mortality model. Gönen & Heller's K concordance coefficient was used to correct censoring interference and intensify the modelling power (Gönen & Heller, 2005).

Description of the general characteristics of the population
From 605 CKD-T recruited between 2011 and 2018, 553 were selected for analysis.
The general profile of the recipient population was composed of 78.5% of individuals < 60 years old, 61.7% of males and 72% multiracial (Table 1). When CKD-T-fast (n=126) and CKD-T-slow (n=427) progressors were compared, 77.2% of patients had a CKD-T-slow behavior with a CKD-T-slow/fast ratio estimated of 3.4.
Elderly patients were a minority in both groups. The most frequent etiologies were DM and SAH. SDD predominated in both groups and the recipients had < 5 years of KRT (Table 2).

Analysis of the survival function and characteristics related to death
Fifteen patients died during follow-up and the 5-year survival probability is 93.2%.
The standardized MR was 10 per 1,000 person-years, with death from CKD-T occurring within the first three years after transplant. Death was more prevalent in men aged ≥ 60 years with an MR of 1.55 (CI95%: 0.64-3.73) ( Table 3).   Arrow shows the critical point between kidney transplant-recipients slow progressors and fast progressor. Chronic kidney disease transplant-recipients with fast progressing behavior have average decline in annual glomerular filtration rate > 5mL/min/1.73m2/year. Source: Authors

Selection of the composition of covariates correlated to death
Multivariate modelling demonstrated that the pre-KT obesity, use of MPPH, ICU up to six months post-KT, fasting blood glycose ≥ 126 mg/dL (≥ 6.99 mmol/L) or NODAT, and CKD-T-fast behavior were associated with a higher risk of death (Table 7). However, only pre-KT obesity (P = 0.023) and the use of HPPM (P = 0.035) were independent risk factors.

Main causes of death in kidney recipients-transplant
A study conducted by Bicalho et al. (2019)  in the 10-year period among CKD-T with fast renal loss behavior. The CKD-T survival curve of the present study corroborates findings from well-reputed reference centers, such as the Hospital do Rim de São Paulo, which found a 93.2% in the five-year survival rate for CKD-T of LV and 79.6% for CKD-T of SDD (Jolissaint & Tullius, 2017).

CKD-T fast-progressor behavior
There is still no consensus over a clear characterization of CKD-T-fast patients. Renal function progression in these patients follows the general guidelines of KDIGO 2012 (Levin et al., 2013), considering a rate of decline in renal function >5 mL/min/1.73m 2 /year was a rapid decline in renal function (Stevens & Levin, 2013). The mechanisms underlying such rapid progression and the risk of death remain unclear.
The rate of decline in CKD may be a marker of subclinical atherosclerosis, endothelial dysfunction or oxidative stress. The Atherosclerosis Risk in Communities (ARIC) study, Development, v. 9, n. 11, e84791110276, 2020 (CC BY 4.0) | ISSN 2525-3409 | DOI: http://dx.doi.org/10.33448/rsd-v9i11.10276 18 which involved more than 13,000 patients, pointed to a risk of peripheral arterial disease and death in the group of patients with the greatest annual rate of decline in renal function (Matsushita et al., 2009). Shlipak et al. (2009) have demonstrated an association between accelerated rate of decline in eGFR and increased risk of heart failure, heart attack and death.
In a 34-month follow-up, Khan et al. (2006) also noticed an increase in mortality when associated with an accelerated rate of decline in renal function. In the study, a value >15 mL/min/1.73m 2 /year was a reference point for defining rapid progression, which was found in 12% of the 6,640 individuals studied. The increase in MR was not verified by the authors in subjects defined as CKD-T-slow.
In the present study, a higher risk of death was found for CKD-T fast progressors or patients with eGFR > 5 mL/min/1.73m 2 /year and this phenomenon was detected after the second year of post-KT follow-up ( Figure 2). We should like to emphasize the importance of analyzing the survival function curve, not only by the logrank, but by the configuration of its trajectory, since survival function is impacted by the censored group and by time of analysis of the probabilistic comparison pairs. A simple analysis may at first seem to invalidate the discriminatory analysis, but it is at the inflection point on the curve that care needs to be intensified. The critical point of the trajectory favors the determination of CKD-T-fast with a higher risk of rapid graft loss and leads to a greater likelihood of death (Park et al., 2013).

New-onset diabetes after transplantation and pre-kidney transplant obesity
NODAT is a growing problem in KT recipients, since this is a serious complication during follow-up (Ducloux et al., 2005). The NODAT continually increases after KT, as demonstrated by the cumulative incidence of cardiovascular events and increased risk of death associated with FBG higher levels (Cia et al., 2016). The present study corroborates these findings, since FBG levels ≥126 mg/dL (>6.99 mmol/L) were also associated with a higher risk of death. In the literature, NODAT has been found to increase the risk of death from cardiovascular diseases by 1.5 to 3 times (Srinivasan et al., 2019).
The modern global epidemic of obesity is yet another serious risk factor for loss of transplanted kidney and death. Although the pathophysiology of obesity and the risk of death are multifactorial,24 the present study found that obesity was the post-KT comorbidity associated with greatest MR. As shown in other studies, obesity in CKD-T increases the risk of SAH, DM, dyslipidemia and coronary artery diseases. Post-KT obesity is an independent risk factor for allograft dysfunction and death from cardiovascular events (Ducloux et al., 2005).

Allograft pyelonephritis (ICU) up to six months after six months post-KT
In the present cohort, the ICU represented a 3.65 times greater risk and was present in 53% of the CKD-T who died. Infectious complications represent a challenge to KT and are associated with longer hospital stays and increased health costs, with CKD-T patients presenting increased risk for ICU. This condition is the major of source of bacteremia and infectious complications, increasing the death risk for CKD-T patients (Ojo, 2005).

Hypothermic pulsatile perfusion preservation machine
In our study, HPPM was used only in selected cases of higher risk of DGF, including ECD kidney-recipients, and only after a period under static preservation, i.e., hybrid preservation. Thus, the value found in the Cox model may be an indirect measure of the quality of the received kidney, and not a linear risk factor for death.
In a multicenter randomized study, involving a kidney donation after circulatory death (DCD-donor), Watson et al. (2010) have found no difference in the incidence of DGF between the kidneys assigned for HPPM or static preservation, with percentages of 58% and 56% respectively. Kidney function at three and twelve months was similar between groups, as well as graft and patient survival.
Apparently, there is a benefit for selected groups regarding reductions in DGF, length of hospital stays and renal function after the first year post-KT (Sandal et al., 2018;Watson et al., 2010).

Effectiveness of the prediction model
Regarding the covariables associated with fatal outcomes, the Harrell's C index of the model was calculated at 85% and Gönen & Heller's K concordance coefficient at 74.0% (Armitage, Berry & Matthews, 2013;Gönen & Heller, 2005;Schoenfeld, 1982). The combination of the two indices demonstrates that the model is both solid and consistent for adjusting the prediction for risk of death in the pre-and post-KT assessment up to one year.
The survival trajectory curve and the survival coefficient of the CKD-T showed a smooth slope in five years. This fact generated a discrete injector function in the codomain (CKD-T who died). As a result, there was both a reduction in MR and in the absolute number of deaths. However, Cox modelling combined with Harrell's C index and Gönen & Heller's K-concordance coefficient helped to correct biases related to deaths and counterbalance the limitations of the Kaplan-Meyer curve analysis.

Final Considerations
A five-year probability of survival rate of 93% was identified among chronic kidney disease transplant recipients. The composition of covariates related to worst probability of survival was demonstrated using Cox's modelling: decline in eGFR >5 mL/min/1.73m 2 /year, ICU up to six months post-KT, pre-KT obesity, FBG ≥ 126 mg/dL (≥ 6.99 mmol/L) or NODAT. Rapid decline in eGFR reduces substantially the survival probability in these population.
In addition, to improving the knowledge of transplantation, we will also have the introduction of techniques and artificial intelligence and data validation using machine learning techniques. Several studies are using regression robots and techniques with complex algorithms to predict kidney allograft graft failure and the patient's risk of death. The tendency is for the biomedical context to appropriate the new knowledge of "Big Data" and computer science for the benefit of kidney-recipients.