Secondary Logo

Journal Logo

Taking Kidneys for Granted? Time to Reflect on the Choices We Make

Wong, Germaine MD, PhD1,2; Howell, Martin PhD1; Patrick, Ellis PhD3; Yang, Jean PhD3

doi: 10.1097/TP.0000000000001850
Commentaries
Free

The authors analyze the complexity of organ allocation and acceptance of a specific kidney by a recipient. They emphasize the role of machine learning to improve the accuracy of prediction and thus of acceptance; although, the opinion of patients may vary according to many factors particularly with time making reality even more complex!

1 Sydney School of Public Health, University of Sydney, Australia.

2 Center for Transplant and Renal Research, Westmead Hospital, Westmead, Australia.

3 School of Mathematics and Statistics, University of Sydney, Australia.

Received 28 May 2017. Revision received 31 May 2017.

Accepted 8 June 2017.

The authors declare no funding or conflicts of interest.

G.W. conceived and wrote the article. M.H., E.P., and J.Y. participated in the writing and presentation of the article.

Correspondence: Germaine Wong, Centre for Transplant and Renal Research, Westmead Hospital, Hawkesbury Road, Westmead, 2145 N.S.W. Australia. (Germaine.wong@health.nsw.gov.au).

Kidney transplantation provides the best outcome for most people with end-stage kidney disease. Transplantation extends and improves quality of life1 and is cost-effective compared with continuing dialysis.2 The greatest impediment to transplantation is that the need far surpasses the availability of donated organs. Many patients will never have the chance of receiving a donor organ because their health deteriorates on the deceased-donor waiting list, and many die while waiting for a donor kidney. The annual mortality rate for dialysis patients on the waiting list exceeds more than 10%, particularly for the older candidates with comorbidities such as diabetes mellitus and vascular disease.3

To combat the issue of organ shortage, an approach taken internationally to increase the number of available organs is to include deceased donors who have previously not been accepted because of factors such as age and comorbidity. In Australia, the proportion of expanded-criteria kidneys (or higher Kidney Donor Profile Index [KDPI] kidneys) in the deceased-donor kidney pool has risen from 20.7% in 2005 to 33.6% in 2011, with a similar trend being observed in the United States and United Kingdom.4 Although inclusion of the higher KDPI donors increases the available organs, it also introduces additional complexity into decisions about allocation and acceptance of organs. Transplantation using higher KDPI donor kidneys may lead to survival advantage compared with dialysis, but the incremental gains in graft and patient survival are not universal for all recipients, and the benefits are largely dependent on recipient age and comorbid status.5 As such, the “1 kidney fits all” approach is no longer applicable. Internationally, several strategies have been developed with the prime objective of optimizing the usage and equity in the allocation of deceased-donor kidneys. In the United States, the New Kidney Allocation System (KAS), implemented in 2014, uses the KDPI and the Estimated Posttransplant Survival score for longevity matching of the top percentile of donors and recipients.5 In Europe, the Eurotransplant Senior Program considered age explicitly in giving priority allocation of grafts from donors 65 years or older to recipients in the same age range. At a population level, recent data suggested that implementation of the KAS have led to improved overall transplant volume, acceptance, and transplantation rates for the younger adults and the highly sensitized individuals, but offer acceptance rates have declined for the older candidates, and this may be partly attributed to KDPI differences.6

Once the kidneys are allocated, the decision to accept or reject a donor kidney lies at the discretion of the clinicians looking after the potential candidate. Acting as patients' advocates, clinicians have the primary objective of choosing a donor kidney that will ensure maximal gains in posttransplant survival for the individuals while balancing against the time spent waiting on dialysis for the “next” best kidney and the comorbidities the candidate may accrue over time. Whether this decision truly reflects the choice of the patients and whether this will lead to the best outcomes are uncertain. The work presented by Bertsimas et al7 in this issue of the Transplantation provides a unique opportunity to assist clinicians and patients in this very complex decision-making algorithm. Using statistical machine learning techniques, the authors examined one of the analytic tools, random forest, that predicts the probability of a potential candidate receiving a deceased-donor kidney based on some KDPI threshold within a specified time frame. Machine learning is an important feature in data science that focuses on the construction of models or algorithms to find association or make predictions by learning information and structures directly from observed data. Random forest is one of the many supervised statistical machine learning technique.8 Bertsimas et al7 used this method to build multiple decision trees on random subsets of the data and aggregated the information to create a stable predictor that can capture the complex relationships within the available data set. Based on information (donor, waitlisted candidates and transplant recipients) from the Organ Procurement and Transplantation Network, the model provides a test classification accuracy of 87%, comparable with 84% from traditional logistic regression modeling, suggesting that data analytics (including nonlinear models) can be applied to predictive modeling in organ allocation and transplantation. However, it is important to note that comparison between machine learning methods should be not only limited to prediction accuracy but also account for model interpretability and transparency.

These findings are important messages to convey to our transplant candidate. Using a risk-based, personalized concept, the simulation model has the capacity to guide the development of a clinical decision support framework that assist decision-making in the choice of accepting or rejecting deceased-donor kidney offers through the provision of reliable estimates that may predict the accessibility of a donor organ with lower KDRI scores within a foreseeable time frame. More importantly, these findings have generated several important research questions. Although the data presented in the analyses were based on the previous allocation system, machine learning has the capacity to adapt and improve the performance with each new data sample and discover hidden patterns. With the availability of data from the New KAS, the model could be trained and applied using a similar concept. However, central to these decisions should be our patients. The ultimate decision to accept or reject a donor offer should also consider the preferences, priorities, and autonomy of the recipient. Patient preferences vary per their values, risk tolerance, age, experience, knowledge, and life circumstance and may also change with time.9-11 Although preferences are complex, it is possible to make quantitative evaluations of both preferences and the underpinning medical and nonmedical factors.12 Using an integrated approach, patient preference data can be incorporated in the machine learning model framework, and a predictive model that comprises of quantitative patient preferences and clinical data could be built. The current decision-making process relies on the “gut instincts” of the clinicians, which often are conglomerations of previous personal experiences and emotions. The model presented in this work has the capacity to generate evidence that will add to the clinician's own intuition and their role as patient advocate when making complex clinical decisions.

Back to Top | Article Outline

REFERENCES

1. Tonelli M, Wiebe N, Knoll G, et al. Systematic review: kidney transplantation compared with dialysis in clinically relevant outcomes. Am J Transplant. 2011;11:2093–2109.
2. Wong G, Howard K, Chapman JR, et al. Comparative survival and economic benefits of deceased donor kidney transplantation and dialysis in people with varying ages and co-morbidities. PLoS One. 2012;7:e29591.
3. Ma MK, Lim WH, Craig JC, et al. Mortality among younger and older recipients of kidney transplants from expanded criteria donors compared with standard criteria donors. Clin J Am Soc Nephrol. 2016;11:128–136.
4. Stratta RJ, Rohr MS, Sundberg AK, et al. Increased kidney transplantation utilizing expanded criteria deceased organ donors with results comparable to standard criteria donor transplant. Ann Surg. 2004;239:688–695. discussion 695–687.
5. Querard AH, Foucher Y, Combescure C, et al. Comparison of survival outcomes between expanded criteria donor and standard criteria donor kidney transplant recipients: a systematic review and meta-analysis. Transpl Int. 2016;29:403–415.
6. Hickey MJ, Zheng Y, Valenzuela N, et al. New priorities: analysis of the New Kidney Allocation System on UCLA patients transplanted from the deceased donor waitlist. Hum Immunol. 2017;78:41–48.
7. Bertsimas D, Kung J, Trichakis N, et al. Accept or decline? An analytics-based decision tool for kidney offer evaluation. Transplantation. 2017;101:2898–2904.
8. Breiman L. Random forests. Machine Learning. 2001;45:5–32.
9. Solomon DA, Rabidou N, Kulkarni S, et al. Accepting a donor kidney: an evaluation of patients' and transplant surgeons' priorities. Clin Transplant. 2011;25:786–793.
10. Reese PP, Tehrani T, Lim MA, et al. Determinants of the decision to accept a kidney from a donor at increased risk for blood-borne viral infection. Clin J Am Soc Nephrol. 2010;5:917–923.
11. Epstein RM, Peters E. Beyond information: exploring patients' preferences. JAMA. 2009;302:195–197.
12. Howell M, Wong G, Rose J, et al. Eliciting patient preferences, priorities and trade-offs for outcomes following kidney transplantation: a pilot best–worst scaling survey. BMJ Open. 2016;6:e008163. doi: 10.1136/bmjopen-2015-008163.
Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.