Page 136 - ITU KALEIDOSCOPE, ATLANTA 2019

2019 ITU Kaleidoscope Academic Conference

machine-learning methods to predict patient-level healthcare 2.4.2 Socioeconomic status
costs following renal transplantation.
We included each patient’s membership of one of 14
2. MATERIAL AND VARIABLES regionally based Local Health Integration Networks
(LHINs), which govern and coordinate health and social
2.1 Patient population services in Ontario, Canada [18].

We used all cases of kidney-only transplantation from a 2.4.3 Comorbidities
deceased donor performed in Ontario, Canada between
March 31, 2002 and April 1, 2013. Recipients aged over 60 We considered each patient’s status of the 11 Collapsed
were followed until death or to April 1, 2016 (N=1425). We Aggregated Disease Groups (CADG) as defined in Johns
excluded the small number of patients who died within one Hopkins’ Adjusted Clinical Group (ACG® [19]) case-mix
year after transplant (N=70, 4.9%) in order to have our final system, a well-validated method of categorizing
model generalize to the more homogeneous group of comorbidities [20]. We excluded CADG 12 (pregnancy)
transplant recipients who survived for at least a year. We since no patients in our cohort were pregnant at the time of
further excluded patients with missing data on healthcare use transplantation. CADGs were established using
(N=27, 1.9%), most of whom were transplanted during the administrative records, including diagnostic codes in the
earliest year of our study period, 2002 (N=19 of the 27 format of the International Statistical Classification of
patients with unknown healthcare use). These exclusions Disease and Related Health Problems, 10 version, Canada
th
resulted in a total of 1328 transplant recipients in the cohort. (ICD-10-CA), from the Discharge Abstract Database (DAD),
a database that includes acute care inpatient hospitalization;
2.2 Data sources physician billing codes from the Ontario Health Insurance
Plan (OHIP); and records from the National Ambulatory
We used a multicenter, population-based dataset derived by Care Reporting System (NACRS). ICD-10-CA codes were
the Institute for Clinical Evaluative Sciences (ICES) in first assigned to one of the 32 Aggregated Diagnosis Groups
Toronto, Ontario, Canada. Person-level records submitted by (ADG) based on five clinical dimensions, including the
all (six) transplantation centers in Ontario to the Canadian duration of the condition, severity, diagnostic certainty,
Organ Replacement Registry (CORR), a national database etiology and involvement of special care. ADGs were then
that tracks end-stage organ failure patients, were linked to collapsed into 12 CADGs based on the likelihood that the
various health service utilization and administrative condition would persist or recur, severity and the type of
databases using a validated unique patient identifier [14]. healthcare services required [19].

2.3 Outcome 2.4.4 Clinical characteristics

We focused on total healthcare costs of transplant recipients We included sensitization indicated by level of peak panel
during the first year after transplantation. In Ontario, Canada, reactive antibodies (PRA) of 0% (not sensitized) or > 0%
renal transplantation is covered for all residents by universal (sensitized), primary cause of ESRD (glomerulonephritis /
public health insurance. Costs were calculated at the patient autoimmune, diabetes, renal vascular, cystic / genetic, or
level across healthcare sectors using a validated, micro- others), and blood type (O, A, B, or AB).
costing, algorithm [15]. We reported costs in Canadian
Dollars (CAD) that were adjusted to 2019 (April) values 2.4.5 Transplant information
using the monthly Consumer Price Index [16], where $1.00
CAD = $0.75 USD [17]. We considered dialysis vintage (pre-emptive transplant or
transplantation without initiating dialysis, or transplant
2.4 Predictors following dialysis duration of <6 months, 6-12 months or >
12 months) and graft number (first graft or re-graft).
We considered a similar set of patient-level predictors as
those examined by Patzer et al. [8] and Tan et al. [9] in their 2.4.6 Pre-transplant healthcare use
respective iChoose Kidney models. Five categories of patient
attributes were collected at transplantation that we have For each patient, we calculated the total healthcare costs for
listed below. during a 6-month workup period before transplant and for
the 12-month period before the start of workup (i.e., pre-
2.4.1 Demographics workup). Costs were measured in 2019 (April) CAD.

We included patient sex (female or male), age (61-70, 71-80, There were missing values found in our dataset for race
or 81+) and race (Caucasian, African American, Asian or (N=98, 7.4%), sensitization (N=237, 17.8%) and primary
Pacific Islanders, or others). cause of ESRD (N=296, 22.3%). In our primary regression
analysis, we imputed Caucasian for those with missing race,
not sensitized (peak PRA = 0%) for those with missing
sensitization, and glomerulonephritis / autoimmune for

– 116 –