Page 135 - ITU KALEIDOSCOPE, ATLANTA 2019

A HEALTHCARE COST CALCULATOR FOR OLDER PATIENTS OVER THE FIRST
YEAR AFTER RENAL TRANSPLANTATION

1
1
Rui Fu ; Nicholas Mitsakakis ; Peter C. Coyte 1
1 Institute of Health Policy, Management and Evaluation, University of Toronto, Canada

ABSTRACT Multiple clinical decision aids have been developed to help
inform patients, families and the health system about
Forecasting tools that accurately predict post- possible outcomes once patients have received a transplant.
transplantation healthcare use of older end-stage renal Patzer et al. [8] from Atlanta, Georgia, US have developed
disease (ESRD) patients are needed at the time of iChoose Kidney (http://ichoosekidney.emory.edu/), a risk
transplantation in order to ensure smooth care delivery in calculator that provides individualized one and three-year
the post-transplant period. We addressed this need by mortality estimates for dialysis or transplantation. The
developing a machine-learning-based calculator that calculator is based on conventional multivariate logistic
predicts the cost of healthcare for older recipients of a regression models applied to large person-level records
deceased-donor kidney over the first year following extracted from the US Renal Disease System (USRDS) and
transplantation. Regression tree and regularized linear has been validated outside of the US (in Ontario, Canada [9])
regression methods, including ridge regression, lasso with acceptable performance.
regression and elastic net regression were explored on all
cases of deceased-donor renal transplants performed for For patients undergoing transplantation, surprisingly, little
patients aged over 60 in Ontario, Canada between March 31, effort has been made to understand their imminent use of
2002 and April 31, 2013 (N=1328), The optimal model healthcare services after transplantation, especially during
(lasso) identified age, membership of one of 14 regionalized the first post-transplant year, a high-risk period marked by
Local Health Integration Networks, blood type, sensitization, elevated rates of hospitalization and intensive use of care [7].
having diabetes as the primary case of ESRD, total This is an urgent gap in healthcare planning since renal
healthcare costs in the 12-month pre-workup period and the transplant recipients, especially those aged over 60, require
6-month workup period to be inputs to the cost calculator. care planning immediately after surgery to ensure optimal
This cost calculator, in conjunction with clinical outcome outcomes. Hence, predictive tools are required to foresee
information, will aid health system planning and their use of healthcare after transplantation using information
performance to ensure better management of recipients of available at the time of transplant.
scarce kidneys.
Recent years have witnessed exponential growth of machine-
Keywords –healthcare costs, health economics, machine learning applications in healthcare [10]–[13], including
learning, renal transplant studies of organ transplantation [10], [12], [13]. Haddad et al.
[10] used machine-learning methods to predict the total
1. INTRODUCTION hospital cost of liver transplant recipients at one-year after
transplant, and found comorbidities to be the most significant
End-stage renal disease (ESRD) is the terminal stage of drivers of high cost. However, their study is limited by the
chronic kidney disease [1]. At this stage, patients have lost at hospital setting and short observational window (from 2011
least 85% of renal function and require immediate initiation to 2012). Furthermore, their use of the Charlson Comorbidity
of renal replacement therapy, whether by lifelong dialysis or Index and van Walraven Score, both ordinal measures of
renal transplantation, to sustain life [1]. Compared with comorbidities, impeded their ability to unveil potential
dialysis, transplantation is preferred for most patients since associations between the use of healthcare and specific types
it improves health outcomes in the long run at substantially of comorbid conditions. Additionally, the optimal model
lower costs [2], [3]. However, unless patients can identify a estimated in their study (Support Vector Machine with linear
medically compatible living donor at the time of diagnosis, kernel) does not directly identify a set of cost predictors.
they must wait for a kidney from a deceased donor, a scarce
healthcare resource. Average waiting time in the US and In the present study, we used machine-learning methods to
Canada has exceeded five and four years, respectively, in develop a calculator that predicts the healthcare costs of
recent years [4], [5]. Meanwhile, transplantation has deceased-donor renal transplant recipients aged over 60
extremely high upfront costs to the healthcare system, which during the first year after transplant using patient-level
may not pay off until decades after transplantation, characteristics known at the time of transplantation. To the
especially for older recipients [6], [7]. best of our knowledge, this is the first investigation that uses

978-92-61-28401-5/CFP1968P-ART @ ITU 2019 – 115 – Kaleidoscope