Recipient Criteria Predictive of Graft Failure in Kidney Transplantation

Comparison of artificial intelligence and human-based prediction and stratification of the risk of long-term kidney allograft failure

BACKGROUND

Clinical decisions are mainly driven by the ability of physicians to apply risk stratification to patients. However, this task is difficult as it requires complex integration of numerous parameters and is impacted by patient heterogeneity. We sought to evaluate the ability of transplant physicians to predict the risk of long-term allograft failure and compare them to a validated artificial intelligence (AI) prediction algorithm.

METHODS

We randomly selected 400 kidney transplant recipients from a qualified dataset of 4000 patients. For each patient, 44 features routinely collected during the first-year post-transplant were compiled in an electronic health record (EHR). We enrolled 9 transplant physicians at various career stages. At 1-year post-transplant, they blindly predicted the long-term graft survival with probabilities for each patient. Their predictions were compared with those of a validated prediction system (iBox). We assessed the determinants of each physician's prediction using a random forest survival model.

RESULTS

Among the 400 patients included, 84 graft failures occurred at 7 years post-evaluation. The iBox system demonstrates the best predictive performance with a discrimination of 0.79 and a median calibration error of 5.79%, while physicians tend to overestimate the risk of graft failure. Physicians' risk predictions show wide heterogeneity with a moderate intraclass correlation of 0.58. The determinants of physicians' prediction are disparate, with poor agreement regardless of their clinical experience.

CONCLUSIONS

This study shows the overall limited performance and consistency of physicians to predict the risk of long-term graft failure, demonstrated by the superior performances of the iBox. This study supports the use of a companion tool to help physicians in their prognostic judgement and decision-making in clinical care.

Changes in quality of life (QoL) and other patient-reported outcome measures (PROMs) in living-donor and deceased-donor kidney transplant recipients and those awaiting transplantation in the UK ATTOM programme: a longitudinal cohort questionnaire survey with additional qualitative interviews

OBJECTIVE

To examine quality of life (QoL) and other patient-reported outcome measures (PROMs) in kidney transplant recipients and those awaiting transplantation.

DESIGN

Longitudinal cohort questionnaire surveys and qualitative semi-structured interviews using thematic analysis with a pragmatic approach.

SETTING

Completion of generic and disease-specific PROMs at two time points, and telephone interviews with participants UK-wide.

PARTICIPANTS

101 incident deceased-donor (DD) and 94 incident living-donor (LD) kidney transplant recipients, together with 165 patients on the waiting list (WL) from 18 UK centres recruited to the Access to Transplantation and Transplant Outcome Measures (ATTOM) programme completed PROMs at recruitment (November 2011 to March 2013) and 1 year follow-up. Forty-one of the 165 patients on the WL received a DD transplant and 26 received a LD transplant during the study period, completing PROMs initially as patients on the WL, and again 1 year post-transplant. A subsample of 10 LD and 10 DD recipients participated in qualitative semi-structured interviews.

RESULTS

LD recipients were younger, had more educational qualifications and more often received a transplant before dialysis. Controlling for these and other factors, cross-sectional analyses at 12 months post-transplant suggested better QoL, renal-dependent QoL and treatment satisfaction for LD than DD recipients. Patients on the WL reported worse outcomes compared with both transplant groups. However, longitudinal analyses (controlling for pre-transplant differences) showed that LD and DD recipients reported similarly improved health status and renal-dependent QoL (p<0.01) pre-transplant to post-transplant. Patients on the WL had worsened health status but no change in QoL. Qualitative analyses revealed transplant recipients' expectations influenced their recovery and satisfaction with transplant.

CONCLUSIONS

While cross-sectional analyses suggested LD kidney transplantation leads to better QoL and treatment satisfaction, longitudinal assessment showed similar QoL improvements in PROMs for both transplant groups, with better outcomes than for those still wait-listed. Regardless of transplant type, clinicians need to be aware that managing expectations is important for facilitating patients' adjustment post-transplant.

A retrospective study of the relationship between postoperative urine output and one year transplanted kidney function

BACKGROUND

Kidney transplantation (KT) is the most obvious method of treating a patient with end-stage renal disease. In the early stages of KT, urine production is considered a marker of successful reperfusion of the kidney after anastomosis. However, there is no clear conclusion about the relationship between initial urine output after KT and 1-year renal function. Thus, we investigated the factors that affect 1-year kidney function after KT, including urine output.

METHODS

This retrospective study investigated the relationship between urine output in the 3 days after KT and transplanted kidney prognosis after 1-year. In total, 291 patients (129 living-donor and 162 deceased-donor transplant recipients) were analyzed; 24-h urine volume per body weight (in kilograms) was measured for 3 days postoperatively. The estimated glomerular filtration rate (eGFR), determined by the Modification of Diet in Renal Disease algorithm, was used as an index of renal function. Patients were grouped according to eGFR at 1-year after KT: a good residual function group, eGFR ≥60, and a poor residual function group, eGFR < 60.

RESULT

Recipients' factors affecting 1-year eGFR include height (P = 0.03), weight (P = 0.00), and body mass index (P = 0.00). Donor factors affecting 1-year eGFR include age (P = 0.00) and number of human leukocyte antigen (HLA) mismatches (P = 0.00). The urine output for 3 days after KT (postoperative day 1; 2 and 3) was associated with 1-year eGFR in deceased-donor (P = 0.00; P = 0.00 and P = 0.01). And, postoperative urine output was associated with the occurrence of delayed graft function (area under curve (AUC) = 0.913; AUC = 0.984 and AUC = 0.944).

CONCLUSION

Although postoperative urine output alone is not enough to predict 1-year GFR, the incidence of delayed graft function can be predicted. Also, the appropriate urine output after KT may differ depending on the type of the transplanted kidney.

TRIAL REGISTRATION

Clinical Research Information Service of the Korea National Institute of Health in the Republic of Korea (KCT0003571).

Using machine learning techniques to develop risk prediction models to predict graft failure following kidney transplantation: protocol for a retrospective cohort study

Background: A mechanism to predict graft failure before the actual kidney transplantation occurs is crucial to clinical management of chronic kidney disease patients.  Several kidney graft outcome prediction models, developed using machine learning methods, are available in the literature.  However, most of those models used small datasets and none of the machine learning-based prediction models available in the medical literature modelled time-to-event (survival) information, but instead used the binary outcome of failure or not. The objective of this study is to develop two separate machine learning-based predictive models to predict graft failure following live and deceased donor kidney transplant, using time-to-event data in a large national dataset from Australia.   Methods: The dataset provided by the Australia and New Zealand Dialysis and Transplant Registry will be used for the analysis. This retrospective dataset contains the cohort of patients who underwent a kidney transplant in Australia from January 1 st, 2007, to December 31 st, 2017.  This included 3,758 live donor transplants and 7,365 deceased donor transplants.  Three machine learning methods (survival tree, random survival forest and survival support vector machine) and one traditional regression method, Cox proportional regression, will be used to develop the two predictive models.  The best predictive model will be selected based on the model's performance. Discussion: This protocol describes the development of two separate machine learning-based predictive models to predict graft failure following live and deceased donor kidney transplant, using a large national dataset from Australia.   Furthermore, these two models will be the most comprehensive kidney graft failure predictive models that have used survival data to model using machine learning techniques.  Thus, these models are expected to provide valuable insight into the complex interactions between graft failure and donor and recipient characteristics.

Using machine learning techniques to develop risk prediction models to predict graft failure following kidney transplantation: protocol for a retrospective cohort study

Background: A mechanism to predict graft failure before the actual kidney transplantation occurs is crucial to clinical management of chronic kidney disease patients.  Several kidney graft outcome prediction models, developed using machine learning methods, are available in the literature.  However, most of those models used small datasets and none of the machine learning-based prediction models available in the medical literature modelled time-to-event (survival) information, but instead used the binary outcome of failure or not. The objective of this study is to develop two separate machine learning-based predictive models to predict graft failure following live and deceased donor kidney transplant, using time-to-event data in a large national dataset from Australia.   Methods: The dataset provided by the Australia and New Zealand Dialysis and Transplant Registry will be used for the analysis. This retrospective dataset contains the cohort of patients who underwent a kidney transplant in Australia from January 1 st, 2007, to December 31 st, 2017. This included 3,758 live donor transplants and 7,365 deceased donor transplants. Three machine learning methods (survival tree, random survival forest and survival support vector machine) and one traditional regression method, Cox proportional regression, will be used to develop the two predictive models (for live donor and deceased donor transplants). The best predictive model will be selected based on the model's performance. Discussion: This protocol describes the development of two separate machine learning-based predictive models to predict graft failure following live and deceased donor kidney transplant, using a large national dataset from Australia. Furthermore, these two models will be the most comprehensive kidney graft failure predictive models that have used survival data to model using machine learning techniques. Thus, these models are expected to provide valuable insight into the complex interactions between graft failure and donor and recipient characteristics.

Healthcare Resource Use, Cost, and Sick Leave Following Kidney Transplantation in Sweden: A Population-Based, 5-Year, Retrospective Study of Outcomes: COIN

Background

Improved understanding of the impact of kidney transplantation on healthcare resource use/costs and loss of productivity could aid decision making about funding allocation and resources needed for the treatment of chronic kidney disease in stage 5.

Material/Methods

This was a retrospective study utilizing data from Swedish national health registers of patients undergoing kidney transplantation. Primary outcomes were renal disease-related healthcare resource utilization and costs during the 5 years after transplantation. Secondary outcomes included total costs and loss of productivity. Regression analysis identified factors that influenced resource use, costs, and loss of productivity.

Results

During the first year after transplantation, patients (N=3120) spent a mean of 25.7 days in hospital and made 21.6 outpatient visits; mean renal disease-related total cost was €66,014. During the next 4 years, resource use was approximately 70% (outpatient) to 80% (inpatient) lower, and costs were 75% lower. Before transplantation, 62.8% were on long-term sick leave, compared with 47.4% 2 years later. Higher resource use and costs were associated with age <10 years, female sex, graft from a deceased donor, prior hemodialysis, receipt of a previous transplant, and presence of comorbidities. Higher levels of sick leave were associated with female sex, history of hemodialysis, and type 1 diabetes. Overall 5-year graft survival was 86.7% (95% CI 85.3–88.2%).

Conclusions

After the first year following transplantation, resource use and related costs decreased, remaining stable for the next 4 years. Demographic and clinical factors, including age <10 years, female sex, and type 1 diabetes were associated with higher costs and resource use.

Does the Kidney Donor Profile Index (KDPI) predict graft and patient survival in a Spanish population?

BACKGROUND AND OBJECTIVE

The Kidney Donor Profile Index (KDPI), together with other donor and recipient variables, can optimise the organ allocation process. This study aims to check the feasibility of the KDPI for a Spanish population and its predictive ability of graft and patient survival.

MATERIALS AND METHODS

Data from 2,734 kidney transplants carried out in Andalusia between January 2006 and December 2015 were studied. Cases were grouped by recipient age, categorised by KDPI quartile and both graft and patient survival were compared among groups.

RESULTS

The KDPI accurately discriminated optimal organs from suboptimal or marginal ones. For adult recipients (aged: 18-59years) it presents a hazard ratio of 1.013 (P<.001) for death-censored graft survival and of 1.013 (P=.007) for patient survival. For elderly recipients (aged: 60+years), KDPI presented a hazard ratio of 1.016 (P=.001) for death-censored graft survival and of 1.011 (P=.007) for patient survival. A multivariate analysis identified the KDPI, donor age, donation after circulatory death, recipient age and gender as predictive factors of graft survival.

CONCLUSIONS

The results obtained show that the KDPI makes it possible to relate the donor's characteristics with the greater or lesser survival of the graft and the patient in the Spanish population. However, due to certain limitations, a new index for Spain based on Spanish or European data should be created. In this study, some predictive factors of graft survival are identified that may serve as a first step in this path.

Assessment of risk factors for increased resource utilization in kidney transplantation

BACKGROUND

There are only a limited number of studies that have sought to identify patients at high risk for medication errors and subsequent adverse clinical outcomes. This study sought to identify risk factors for increased health care resource utilization in kidney transplant recipients based on drug-related problems and self-administered surveys.

METHODS

In this prospective observational study, adult kidney transplant recipients seen in the transplant clinic between September and November 2015 were surveyed for self-reported demographics, medication adherence, and health status/outlook. Subsequently, patients were assessed for associations between survey results, pharmacist-derived drug-related problems, and health resource utilization over a minimum 6-mo follow-up period. Based on univariate associations, two risk cohorts were identified and compared for health care utilization using multivariable Poisson regression.

RESULTS

A total of 237 patients were included, with a mean follow-up of 8 mo. From the patient survey data, Medicaid insured or self-rated poor health status were identified as a significant risk cohort. From pharmacist assessments, those who received incorrect medication or lacked appropriate follow-up medication monitoring were identified as a significant risk cohort (pharmacy errors). The Medicaid insured or self-rated poor health status cohort experienced 43% more total health care encounters (incident rate ratios [IRR] 1.43, 1.01-2.02) and 35% more transplant clinic visits (IRR 1.35, 1.03-1.77). The pharmacy errors cohort experienced 4.2 times the rate of total health care encounters (IRR 4.17, 1.55-11.2), 4.1 times the rate of hospital readmissions (IRR 4.09, 1.58-10.6), and 2.3 times the rate of transplant clinic visits (IRR 2.31, 1.04-5.11).

CONCLUSIONS

Medicaid insurance, self-rated poor health status, and errors in the medication regimen or monitoring were significant risk factors for increased health care utilization in kidney transplant recipients. Further research is warranted to validate these potential risk factors, determine the long-term impact on graft/patient survival, and assess the mutability of these risks through prospective identification and intervention.