Introduction

Tacrolimus is one of the most widely used immunosuppressants following solid organ transplantation (SOT). It has a narrow therapeutic range between efficacy and toxicity and tacrolimus therapeutic drug monitoring is standard of care (SOC) in the transplant population [1]. Therapeutic drug monitoring measures levels of tacrolimus in the blood at certain intervals to optimize individual dosing regimens and maintain drug levels within a specified range [2]. Therapeutic goal ranges vary between type of SOT and generally decrease over time following transplant, as the risk of rejection decreases [3, 4].

Tacrolimus is primarily metabolized in the liver and small intestine by the cytochrome P450 (CYP) 3A5 metabolizing enzyme, and to a lesser extent by CYP3A4 [5]. CYP3A5 and CYP3A4 genetic variants, such as CYP3A5*3 and CYP3A4*22, have been shown to contribute to inter-patient variability in tacrolimus pharmacokinetics [6,7,8,9]. The Clinical Pharmacogenetics Implementation Consortium (CPIC) has published guidelines with recommendations for tacrolimus dosing based on CYP3A5 genotype [10]. For individuals carrying two nonfunctional CYP3A5 alleles (i.e., *3, *6, or *7), also known as CYP3A5 nonexpressers, CPIC recommends initiating tacrolimus at the standard recommended dose. For patients carrying at least one CYP3A5*1 allele (i.e., CYP3A5 expressers), CPIC recommends increasing the starting tacrolimus dose by 1.5-2 times the standard dose [10].

Clinical trials have examined the effects of CYP3A5 genotype-guided (based on the CYP3A5*3 variant) versus non-CYP3A5 genotype-guided (i.e., SOC) tacrolimus dosing on pharmacokinetic and clinical outcomes in kidney transplantation [11,12,13]. CYP3A5 genotype-guided dosing was associated with improved achievement of target tacrolimus trough concentrations compared to SOC tacrolimus dosing [11, 13]. While these investigations did not report a difference in clinical outcomes such as biopsy-proven acute rejection or acute tacrolimus nephrotoxicity in relation to CYP3A5 genotype-guided dosing, they are limited by small sample sizes and inclusion of a single transplant type [11,12,13]. Nonetheless, given the associations with pharmacokinetic outcomes in these clinical trials, as well as suggested differences in resource utilization based on CYP3A5 expresser status in observational studies [14], several institutions are implementing CYP3A5 genotype-guided tacrolimus dosing in clinical practice [15,16,17,18,19,20].

Understanding the cost-effectiveness of CYP3A5 genotype-guided tacrolimus dosing is key to large-scale implementation of pharmacogenetic testing in the transplant setting; however, to our knowledge, no studies have reported on the cost-effectiveness of pre-SOT pharmacogenetic screening in relation to tacrolimus therapy. Therefore, the objective of this study was to estimate the cost-effectiveness of CYP3A5 genotype-guided tacrolimus dosing in kidney, liver, heart, and lung transplant as compared to SOC dosing, from a US healthcare payer perspective.

Methods

Data sources and base-case population

To conduct this analysis for the base-case (i.e., average) scenario, we used a combination of real-world data and literature estimates. We estimated real-world values for tacrolimus exposure and clinical inputs from electronic health record (EHR) data from UCHealth in Aurora, Colorado, USA. We obtained EHR data from Health Data Compass (HDC), an enterprise data warehouse (https://www.healthdatacompass.org). The HDC data warehouse includes data for all patients seen at UCHealth for any reason who have an EHR in the UCHealth system.

For decision tree models, we included kidney, liver, heart, and lung transplant recipients transplanted at UCHealth between February 1, 2011 and August 9, 2020 who met the following criteria: ≥18 years of age at the time of transplant; ≥2 recorded tacrolimus levels in the first six months post-transplant; pre-transplant medication, vitals, and laboratory data available; identified as a race for which there are published CYP3A5*3 frequency data (Non-Hispanic American Indian/Alaska Native, Non-Hispanic Asian, Hispanic Black/African American, Non-Hispanic Black/African American, Hispanic White/Caucasian, Non-Hispanic White/Caucasian, or Hispanic and multiple race); did not have multiple or combined organ transplantation (e.g., kidney-heart); and not re-transplanted in the first six months for liver, heart, and lung recipients. The study included re-transplanted kidney recipients if their original transplanted organ was removed with a tacrolimus washout period of ≥1 month. This study was approved as secondary use of clinical data with a HIPAA waiver of consent by the Colorado Multiple Institutional Review Board.

Decision model

We developed decision tree models using RStudio v 3.6.0 to estimate the short-term cost-effectiveness of CYP3A5 genotype-guided tacrolimus dosing relative to SOC in kidney, liver, heart, and lung transplant recipients (Fig. 1). The decision trees, structured for a time horizon of six months post-transplant, generated estimates for clinical outcomes (i.e., cumulative incidence) and economic outcomes (i.e., costs, incremental cost per avoided event). CPIC guidelines provide CYP3A5 genotype-guided recommendations for individualizing initial tacrolimus dosing [10]. Therefore, we evaluated incremental cost per avoided event over a short-term time horizon, which is more applicable in this scenario than long-term economic outcomes, such as cost per quality-adjusted life year gained (QALY).

Fig. 1: Decision tree model for CYP3A5 genotype-guided vs SOC tacrolimus dosing.
figure 1

SOC standard of care, TAC TTR-Rosendaal, tacrolimus time in therapeutic range calculated with the Rosendaal algorithm.

The lower branch of the decision tree (Fig. 1) represents costs and outcomes for SOC tacrolimus dosing. SOT recipients generally begin on a standard tacrolimus dose that differs by transplant population. For example, kidney transplant recipients typically begin with 0.2 mg/kg/day of tacrolimus when taken with azathioprine or 0.1 mg/kg/day when taken with mycophenolate; liver transplant recipients typically begin with 0.1-0.15 mg/kg/day of tacrolimus; heart and lung recipients typically begin with 0.075 mg/kg/day of tacrolimus [21]. For all organs, doses are then titrated to maintain drug levels within therapeutic goal ranges. The upper branch of the decision tree reflects costs and outcomes for tacrolimus dosing based on CYP3A5 genotype results. This branch follows dosing recommendations provided by CPIC, specifically: CYP3A5 nonexpressers (i.e., patients with a CYP3A5*3/*3 genotype) should initiate tacrolimus with the standard recommended dose, while CYP3A5 expressers (i.e., patients with at least one copy of the CYP3A5*1 allele) should initiate tacrolimus with 1.5-2x the standard dose [10]. CPIC guidelines are written for kidney, heart, and lung transplant patients, as well as liver transplant patients in whom the donor and recipient genotypes are identical (i.e., concordant). Therefore, for the purposes of the model, we assumed a genotype test was performed on both liver recipients and their donors, and that the recipient and donor CYP3A5 genotypes were concordant.

Decision tree model inputs

Allowing for symmetry in the model (Fig. 1), the main branches (i.e., CYP3A5 genotype-guided tacrolimus dosing and SOC tacrolimus dosing) include inputs for CYP3A5 phenotype (i.e., CYP3A5 expresser or CYP3A5 nonexpresser); the decision to follow recommendations based on CYP3A5 genotype results; tacrolimus exposure (i.e., time spent in therapeutic goal range); and the occurrence of acute tacrolimus nephrotoxicity, acute cellular rejection, a combination of acute tacrolimus nephrotoxicity and acute cellular rejection, death, or none of these outcomes. Inputs used in the decision tree models are shown in Table 1. The cost-effectiveness of CYP3A5 genotype-guided compared to SOC tacrolimus dosing is defined as the incremental cost per avoided event and is calculated as the difference in costs in USD between CYP3A5 genotype-guided and SOC tacrolimus dosing divided by the difference in incidence of all non-fatal outcomes between CYP3A5 genotype-guided and SOC tacrolimus dosing.

Table 1 Key model parameters for cost-effectiveness of CYP3A5 genotype-guided vs SOC tacrolimus dosing in SOT recipients.

CYP3A5 phenotype inputs

Due to the difference in tacrolimus dosing guidance between CYP3A5 phenotypes, we included the proportion of patients expressing the CYP3A5 phenotype (Table 1) as a branch in the model. We calculated weighted averages of CYP3A5 phenotype frequencies based on the race/ethnicity data of the kidney, liver, heart, and lung transplant patients in our UCHealth transplant population. Although we had a small subpopulation of patients (19.6%) with CYP3A5 genotype data available, we chose to estimate CYP3A5 phenotype frequencies from multiple sources, including the Allele Frequency Aggregator project [22], Genome Aggregation Database [23], and three observational studies that characterized CYP3A genetic variation in American Indian and Alaska Native communities [24,25,26]. Calculations for CYP3A5 phenotype frequencies are provided in the Supplemental Material.

Dosing, exposure, and clinical inputs

Previously, we surveyed transplant physicians and found that 37.8% reported ordering a pharmacogenetic test at least once in the previous 12 months, of which only 29% reported ordering a CYP3A5 genetic test [27]. In addition, a prior study of antiplatelet therapy showed that as few as 23% of prescribing physicians follow the recommendations accompanying pharmacogenetic test results [28]. Therefore, we included physician use of genetic test results to inform clinical care as a parameter in our models [28, 29]. We did not have detailed test usage and physician clinical action data from our previous survey; therefore, we derived the percentage of physician use of test results to inform clinical care from published estimates on the selection of alternative antiplatelet therapy based on CYP2C19 genotype (e.g., percentage of patients with an actionable vs non-actionable genotype prescribed alternative therapy) (Table 1) [28]. For the SOC tacrolimus dosing arm, we assumed the percentage following CYP3A5 genotype-guided dosing for CYP3A5 expressers and nonexpressers was null. Owing to the uncertainty surrounding this parameter, we examined it further in the univariate sensitivity analysis.

Due to a high level of inter- and intra-patient variability in tacrolimus pharmacokinetics observed in SOT patients, contemporary work has focused on exposures such as time in therapeutic range (TAC TTR). TAC TTR can be calculated multiple ways, e.g., number of days within range divided by total number of days. A retrospective study by Calabrese and colleagues performed in 321 US lung transplant patients found that CYP3A5 expressers had decreased TAC TTR compared to nonexpressers (mean TAC TTR: 3.5% vs 18.5%, respectively; p < 0.001) during the first 14 days post-transplant [9]. Similarly, Salah et al. demonstrated in 69 heart transplant recipients that CYP3A5 expressers had less TAC TTR vs CYP3A5 nonexpressers [mean TAC TTR: 24.0% vs 33.3%, respectively; p = 0.01] during the first year post-transplant [30].

Another way to calculate TAC-TTR is via the Rosendaal, or linear interpolation, method (hereafter referred to as TAC TTR-Rosendaal) [31,32,33,34,35,36,37]. This method, developed for use in anticoagulant therapy, calculates the percent of time a patient is within therapeutic goal range using an algorithm that assumes the difference between consecutive levels is linear [38]. Leino and coworkers published a study in pediatric kidney and heart transplant recipients demonstrating that CYP3A5 nonexpressers had a 1.4-fold higher TAC TTR-Rosendaal compared to CYP3A5 expressers in the first three months post-transplant [36]. Recent work in kidney, kidney-pancreas, heart, and lung transplant recipients has suggested an association between TAC TTR-Rosendaal categorized into high vs low groups (e.g., ≥40% vs <40%) and acute cellular or antibody-mediated rejection in the first 6-12 months post-transplant [31,32,33,34,35, 37]. As such, in our decision tree models, we defined tacrolimus exposure as high vs low TAC TTR-Rosendaal (in an organ-specific manner, using our own preliminary data, see Supplemental Material), and tacrolimus-associated outcomes as acute inpatient tacrolimus nephrotoxicity, acute cellular rejection, or death, in the first six months post-transplant. For model inputs, we estimated the probability of experiencing high vs low TAC TTR-Rosendaal for CYP3A5 expressers and nonexpressers, and the rates of clinical outcomes (i.e., acute tacrolimus nephrotoxicity, acute cellular rejection, a combination of acute tacrolimus nephrotoxicity and acute cellular rejection, and death) for patients with high and low TAC TTR-Rosendaal from our UCHealth transplant populations (see Supplemental Material).

Consistent with other recent literature, we used the median TAC TTR-Rosendaal in each cohort to categorize patients into high and low TAC TTR-Rosendaal in our transplant populations: ≥50% vs <50% for kidney; ≥33% vs <33% for liver; ≥33% vs <33% for heart; and ≥16% vs <16% for lung transplant patients [35, 39]. We defined acute inpatient tacrolimus nephrotoxicity as an increase in serum creatinine by 50% or 0.3 mg/dL from baseline leading to a decrease in tacrolimus dose or discontinuation of tacrolimus followed by a decrease in serum creatinine, during or followed by an inpatient stay [40, 41]. In the kidney transplant cohort, baseline serum creatinine was the highest recorded level in the first three days post-transplant; and in the liver, heart, and lung transplant cohorts, baseline serum creatinine was the last level reported prior to the start of tacrolimus [42].

Acute cellular rejection was defined as any biopsy-proven acute cellular rejection of any severity (borderline, indeterminate, minimal, mild, moderate, or severe), diagnosed using the respective scoring systems recommended for each organ – Banff criteria for kidney and liver transplant and International Society for Heart and Lung Transplantation criteria for heart and lung transplant; or empirically-treated (i.e., administration of methylprednisolone, thymoglobulin, basiliximab, or an increase in tacrolimus, prednisone, or mycophenolate dose) suspected acute cellular rejection episode [43,44,45,46].

Cost inputs

We estimated all costs in 2022 US dollars (USD, Table 1). In 2022, through phone and email communications, three commercial laboratory companies provided the cost for single-gene CYP3A5 genotyping [47,48,49]. We used the mean cost among the three laboratories for the base-case scenario ($305) and included the lowest ($160) and highest ($420) costs in one-way sensitivity analyses. We assumed costs for acute inpatient tacrolimus nephrotoxicity for each cohort to be the average cost for an inpatient stay for kidney, liver, heart, and lung transplant patients experiencing transplant complications, as there is no SOC treatment for this outcome besides a decrease in tacrolimus dosing [50]. We estimated costs for acute cellular rejection as weighted averages for the cost of a biopsy, acute cellular rejection treatments (e.g., intravenous methylprednisolone, thymoglobulin, everolimus, and/or increase in prednisone, mycophenolate, tacrolimus), inpatient stays, and outpatient visits based on the proportion of patients utilizing these treatments and services observed in our transplant populations. The breakdown of costs is provided in Supplemental Material. We derived the average cost per day for hospitalizations related to kidney, liver, heart, and lung transplant complications (i.e., ICD-9 codes 996.81, 996.82, 996.83, and 996.84, respectively) from the National Inpatient Sample, Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality [50]. We derived medication costs from Medicare Part B average sales prices for January-March 2022 [51], and biopsy and outpatient care costs from the Medicare Physician Fee schedules for 2022 [52]. We converted applicable costs to 2022 costs using the Bureau of Labor Statistics conversion calculator (https://data.bls.gov/cgi-bin/cpicalc.pl).

Sensitivity analyses

We performed univariate and multivariate probabilistic sensitivity analyses to address uncertainty in the decision tree models. Univariate sensitivity analysis altered one variable at a time and held all others constant, focusing on CYP3A5 phenotype probabilities, physician use of genotype test results to inform clinical care, proportion of patients with high vs low TAC TTR-Rosendaal for CYP3A5 genotype-guided tacrolimus dosing, and CYP3A5 genotyping costs. For CYP3A5 genotyping costs, we used the lowest ($160) and highest ($420) costs provided to us from the three laboratories. For proportion of CYP3A5 expressers, we assumed 0% for the lowest possible proportion and 75% for the highest proportion. This is slightly higher than the estimated proportion of CYP3A5 expressers in a population of all Black/African American patients [23]. We used evidence-based probability distributions for the other variables (Table 1). The multivariate sensitivity analysis varied all genotype, clinical, and cost inputs simultaneously, using Monte Carlo simulation programmed into RStudio v 3.6.0. Total cost and cost per avoided event were re-calculated 10,000 times with inputs randomly drawn using evidence-based probability distributions, provided in Table 1, and presented in a scatterplot.

Results

Base-case cohort

Characteristics of the final cohorts included as the base-case population in the cost-effectiveness analysis are shown in Table 2. A total of 2299 transplant patients met inclusion criteria [1364 kidney recipients (60.8% men, 73.6% White/Caucasian, 70.5% Non-Hispanic, and mean age 50 ± 15 years at the time of transplant); 645 liver recipients (67.0% men, 82.3% White/Caucasian, 74.4% Non-Hispanic, and mean age 52 ± 12 years at the time of transplant); 175 heart recipients (70.3% men, 74.9% White/Caucasian, 84.0% Non-Hispanic, and mean age 53 ± 13 years at the time of transplant); and 115 lung recipients (57.4% men, 92.2% White/Caucasian, 93.9% Non-Hispanic, and mean age 57 ± 12 years at the time of transplant)].

Table 2 Characteristics of study population.

Base-case scenario

Results for the base-case scenario are presented in Table 3. The incremental cost per avoided event in the first six months post-transplant for CYP3A5 genotype-guided vs SOC tacrolimus therapy was $176,667 for kidney recipients, $364,000 for liver recipients, $12,982 for heart recipients, and $93,333 for lung recipients.

Table 3 Base-case results from cost-effectiveness analysis of CYP3A5 genotype-guided vs SOC tacrolimus dosing.

Univariate sensitivity analyses

Results from the univariate analyses are provided in Fig. 2. Findings from these analyses showed that the interpretation of these models was most sensitive to physician use of pharmacogenetic test results to guide clinical care and the proportion of patients with high TAC TTR-Rosendaal following CYP3A5 genotype-guided tacrolimus dosing. The incremental cost per avoided event ranged from $463,343 (kidney), $631,035 (liver), $74,523 (heart), and $154,378 (lung) when physician response was close to 0%. The incremental cost per avoided event ranged from $78,006 (kidney), $194,651 (liver), -$1084 (i.e., dominant) (heart), and $52,010 (lung) when physician response was close to 100%. An increase in the proportion of patients with high TAC TTR-Rosendaal following CYP3A5 genotype-guided tacrolimus dosing resulted in incremental costs per avoided event of $14,424 for kidney transplant patients, $28,597 for liver transplant patients, and dominant incremental costs per avoided event (i.e., lower costs and less adverse events) for heart and lung patients. In contrast, a decrease in the proportion of patients with high TAC TTR-Rosendaal after CYP3A5 genotype-guided tacrolimus dosing resulted in a dominated incremental cost per avoided event for each transplant cohort, with higher costs and increased rates of adverse events for CYP3A5 genotype-guided vs SOC tacrolimus dosing. Assuming a genotyping cost of $160, the incremental cost per avoided event ranged from $88,174 (kidney), $181,065 (liver), $1332 (heart), to $43,836 (lung). Assuming a genotyping cost of $420, the incremental cost per avoided event ranged from $264,329 (kidney), $521,120 (liver), $29,247 (heart), to $141,010 (lung). Finally, as the frequency of CYP3A5 expressers reached 75%, incremental costs per avoided event for CYP3A5 genotype-guided tacrolimus dosing were $99,960 (kidney), $220,058 (liver), $1854 (heart), and $53,313 (lung); as the frequency of CYP3A5 expressers decreased to almost 0%, incremental costs per avoided event were $323,120 (kidney), $479,333 (liver), $48,091 (heart), and $121,998 (lung).

Fig. 2: Univariate sensitivity analysis.
figure 2

Tornado diagram demonstrating univariate sensitivity analysis results for cost-effectiveness of CYP3A5 genotype-guided vs SOC tacrolimus dosing in (a) kidney, (b) liver, (c) heart, and (d) lung transplant recipients. Along the y-axis are the parameters altered in the univariate sensitivity analysis, and the x-axis is the incremental cost per avoided event. The vertical line is the base-case incremental cost per avoided event. The horizontal bars represent the upper and lower bounds of the incremental cost per avoided event for each parameter while the other parameters remain constant. TAC TTR-Rosendaal, tacrolimus time in therapeutic range calculated with the Rosendaal algorithm.

Probabilistic sensitivity analyses

Figure 3 shows the results from the probabilistic sensitivity analyses. After simultaneously adjusting all probabilities and cost values, most of the points remained robust to the base-case scenarios. However, for each transplant cohort, there were points indicating that CYP3A5 genotype-guided dosing may be either dominated (i.e., less effective and more costly) or dominant (i.e., more effective and less costly). The proportion of parameter combinations that favored CYP3A5 genotype-guided vs SOC tacrolimus dosing as a cost-saving strategy was 19.8% in kidney, 32.3% in liver, 51.8% in heart, and 54.1% in lung transplant recipients.

Fig. 3: Scatterplot of probabilistic sensitivity analysis.
figure 3

Results are shown for cost-effectiveness of CYP3A5 genotype-guided vs SOC tacrolimus dosing in (a) kidney, (b) liver, (c) heart, and (d) lung transplant recipients. Each point represents one Monte Carlo simulation. Incremental effectiveness is the difference in the rate of adverse events. The x-axis is incremental effectiveness (i.e., rate of avoided events), and the y-axis is incremental costs. Points on the plot represent incremental cost per avoided event values corresponding to the result from one Monte Carlo simulation. Points that lie in the upper left-hand quadrant indicate that SOC tacrolimus dosing is economically preferable to CYP3A5 genotype-guided tacrolimus dosing. Points in the upper-right quadrant indicate that CYP3A5 genotype-guided tacrolimus dosing is more costly, but more effective than SOC dosing. Points in the lower-left quadrant indicate that that CYP3A5 genotype-guided tacrolimus dosing is cost-saving but less effective than SOC dosing. Points in the lower-right quadrant indicate that CYP3A5 genotype-guided tacrolimus dosing is economically preferable to SOC tacrolimus dosing. SOC standard of care, USD US dollars.

Discussion

This is the first study to evaluate the cost-effectiveness of CYP3A5 genotype-guided tacrolimus dosing in SOT recipients. Our findings indicate that CYP3A5 genetic testing prior to initiation of tacrolimus resulted in reduced rates of adverse events with higher cost compared to SOC tacrolimus dosing. The lowest incremental cost per avoided event was $15,806 for heart transplant patients and the highest was $217,917 for liver transplant patients. Changes to factors including CYP3A5 expresser phenotype frequency and the proportion of patients with high vs low TAC TTR-Rosendaal following CYP3A5 genotype-guided tacrolimus dosing altered the cost-effectiveness of CYP3A5 genotype-guided tacrolimus therapy in these transplant cohorts.

To date, the cost-effectiveness of preemptive pharmacogenetic testing has not been well-studied in SOT; consequently, no direct comparisons can be made regarding these findings. However, recent work has investigated the relationship between CYP3A5 genetics and resource utilization in various transplant populations [14, 18, 53]. Using data collected from a clinical trial on CYP3A5 genotype-guided tacrolimus dosing in Thailand, a study of 125 kidney transplant recipients showed that patients with the CYP3A5*1/*1 or CYP3A5*1/*3 genotype (i.e., CYP3A5 expressers) had higher costs for tacrolimus dosing, therapeutic drug monitoring, and hospitalizations during transplant admission compared to patients with the CYP3A5*3/*3 genotype (i.e., CYP3A5 nonexpressers) [53]. The authors suggested that part of the cost difference was due to higher dose requirements among CYP3A5 expressers as well as a greater proportion of CYP3A5 expressers requiring dialysis within one-week post-transplant compared to CYP3A5 nonexpressers [53]. Another investigation showed an association between CYP3A5 expresser status and increased hospital length of stays and total costs up to 12 months post-transplant in pediatric heart recipients [18]. While we did not measure differences in healthcare utilization between CYP3A5 expressers and nonexpressers, our one-way sensitivity analyses altered the proportion of patients with CYP3A5 expresser status. Results showed that maximizing the proportion of patients with CYP3A5 expresser status resulted in a decreased incremental cost per avoided event for CYP3A5 genotype-guided relative to SOC tacrolimus dosing in all cohorts, with the heart transplant cohort demonstrating a dominant strategy (i.e., less costly and greater benefit). Taken together, these data suggest that the economic value of CYP3A5 genotype-guided tacrolimus dosing may be greater in cohorts with a higher frequency of CYP3A5 expressers. Further studies are needed to validate our sensitivity analyses and investigate the clinical applicability in populations with higher proportions of patients with the CYP3A5 expresser phenotype.

Our investigation did not find substantial differences in the incidence of clinical outcomes including acute tacrolimus nephrotoxicity and acute cellular rejection for CYP3A5 genotype-guided vs SOC tacrolimus dosing in any cohort. Therefore, while there were minimal incremental costs for CYP3A5 genotype-guided vs SOC tacrolimus dosing ($265 for kidney, $546 for liver, $148 for heart, and $252 for lung recipients), the incremental effectiveness was also minor and, in the base-case scenarios, it may not be enough to justify the cost. Additionally, the probabilistic sensitivity analysis suggested cost-savings for CYP3A5 genotype-guided compared to SOC tacrolimus dosing in 20-54% of the simulations in each cohort. Recent clinical trials in kidney transplant recipients similarly reported no significant differences in clinical outcomes (i.e., patient survival, biopsy-proven acute rejection, renal function measured by estimated glomerular filtration rate, and acute tacrolimus nephrotoxicity) between CYP3A5 genotype-guided and SOC tacrolimus dosing groups, although these studies were not powered to detect differences in these outcomes [11,12,13]. Nevertheless, the authors of these studies showed that CYP3A5 genotype-guided tacrolimus dosing significantly affected the proportion of patients in therapeutic range at steady state, time to therapeutic range, and the number of dose modifications needed to achieve therapeutic goal ranges [11,12,13]. These outcomes may have time and cost implications, which we did not measure in our study, potentially decreasing the robustness of our results. Future cost studies should consider incorporating these outcomes as a measure of effectiveness in the analysis of CYP3A5 genotype-guided tacrolimus dosing, while prospective studies using real-world clinical data could also shed light on the cost-effectiveness of CYP3A5 genotype-guided treatment. Furthermore, while we used single-gene pharmacogenetic testing in this analysis, the cost of multigene, panel-based pharmacogenomic testing continues to decrease. Future analyses should assess multigene panels and the full cost-effectiveness that would be realized if results were used to inform dosing and selection of all pharmacogenetic-guided medications used in the transplant setting [54, 55].

Our univariate sensitivity analysis showed that the model was particularly sensitive to physician use of CYP3A5 genotype results to guide initial tacrolimus dosing. We made a conservative assumption regarding the rate of physician use of CYP3A5 genotype results, which was based on the results of our published transplant provider survey showing low adoption rates of CYP3A5 genotyping and uncertainty about the clinical value of pharmacogenetic testing in the transplant setting [29]. True rates of clinician usage of CYP3A5 genotype results to guide initial tacrolimus dosing may be markedly different in current real-world transplant populations. Our univariate sensitivity analysis suggests that when physician use of CYP3A5 genotype results increases to almost 100%, the incremental cost per avoided event decreases significantly and, in the heart transplant population, CYP3A5 genotype-guided tacrolimus dosing becomes the dominant option (i.e., less costly and more effective than SOC). These findings emphasize the need for provider education to maximize the potential clinical utility and cost-effectiveness of pharmacogenetic-guided therapy, particularly in the setting of SOT.

In the context of our study, literature related to the cost-effectiveness of pharmacogenetic-guided warfarin dosing may serve as a practical comparator due to its similarities to tacrolimus, including a narrow therapeutic range between efficacy and toxicity and the need for laboratory monitoring. Additionally, the Rosendaal algorithm, which is now used for calculating TAC TTR in the transplant setting, was developed for anticoagulant monitoring and has been included as an outcome in previous cost-effectiveness models of genotype-guided warfarin dosing. A cost-benefit analysis of genotype-guided warfarin dosing performed in 2008 reported incremental cost per avoided event for varying levels of adverse event severity [56]. Estimating a pharmacogenetic testing cost similar to ours ($350), the cost per avoided bleeding event (e.g., stroke, gastrointestinal hemorrhage) over 12 months ranged from $1198 for optimistic testing conditions (i.e., high baseline bleeding, large effect on stroke incidence), to $123,715 for pessimistic conditions (low baseline bleeding, no effect on stroke). This warfarin finding parallels our tacrolimus results showing divergent costs per avoided event between transplant cohorts, which likely correspond to differences in the rates of adverse events in our study. For example, among the kidney, heart and lung transplant cohorts who had a $295 testing cost, the incremental cost per avoided event was lowest in heart transplant patients ($12,982), which showed the most optimistic testing conditions among the cohorts (i.e., greatest difference in rates of adverse events between CYP3A5 genotype-guided and SOC tacrolimus dosing), and highest in kidney transplant patients ($176,667), which had the most pessimistic conditions (i.e., smallest difference in rates of adverse events between CYP3A5 genotype-guided and SOC tacrolimus dosing). Further evaluations of short-term economic outcomes of genotype-guided warfarin dosing have reported a cost per avoided event ranging from $5778 to $170,192 [57, 58]. This wide range of incremental costs per avoided event is likely due to different data sources such as clinical trials (higher cost per avoided event) compared to assumptions based on literature [58]. With the exception of liver transplant patients, for whom testing costs were doubled to account for the genotype of the donor liver as well as the recipient, the incremental cost per avoided events in our cohorts approximately fall within this range. Although results appear consistent, this comparison must be made with caution, as the adverse events analyzed in the warfarin studies are different than those analyzed in our transplant study.

Our cost-effectiveness analysis has limitations which deserve to be acknowledged. First, the sample sizes of the transplant cohorts were relatively small, especially the subset of patients for whom we had genetic data, limiting the power to detect a significant effect of CYP3A5 genotype-guided tacrolimus dosing on clinical outcomes, and potentially decreasing the observed cost-effectiveness. Therefore, the generalizability of our results is limited to populations like ours with respect to CYP3A5 expresser status. Further studies are needed in SOT populations with a higher proportion of CYP3A5 expressers to understand the full value of CYP3A5 genotype-guided tacrolimus dosing in SOT recipients. It is also important to acknowledge that self-reported race categories derived from the EHR may result in inaccurate estimations of sub-population differences, especially in countries with high genetic heterogeneity like the US [59,60,61]. Another limitation is that we included kidney, liver, heart, and lung recipients who were transplanted between 2011 and 2020, during which time changes in transplant standards of care may have influenced event rates over time. Next, we did not conduct this analysis from the societal perspective or assess long term outcomes, which may reduce the generalizability of the results [62]. Nonetheless, we used a healthcare perspective since it is more pertinent for informing the implementation of pharmacogenetic testing into the transplant clinic. Because we did not assess long term outcomes, we did not capture differences between expected life gains and life gains after surviving a transplant, nor produce QALYs, which would allow for comparison between diseases. However, a shorter time horizon is most relevant for these analyses as the early post-transplant time is when most of the variability in tacrolimus concentrations and dosing is typically observed. CYP3A5 genotype-guided tacrolimus dosing is not expected to provide value beyond initial dosing due to the routine use of therapeutic drug monitoring to maintain patients within goal range [1]. Assessing long-term outcomes such as QALYs would likely diminish the cost-effectiveness of CYP3A5 genotype-guided tacrolimus dosing (i.e., less effective, and more costly) due to the short-term nature of the outcomes. Finally, we assessed acute tacrolimus nephrotoxicity in the inpatient setting. This limitation may have led to underestimation of the incidence of acute tacrolimus nephrotoxicity and missed costs associated with outpatient management of this adverse event.

In summary, our cost-effectiveness analysis found that relative to SOC, CYP3A5 genotype-guided tacrolimus dosing showed higher cost for little additional benefit in each base-case scenario. Sensitivity analyses suggest that CYP3A5 genotype-guided tacrolimus dosing has a low likelihood of being cost-saving in any transplant cohort. The key parameters in each model were physician use of CYP3A5 genotype results and incidence of high TAC TTR following CYP3A5 genotype-guided tacrolimus dosing. Further economic evaluations that examine outcomes such as dose modifications in larger populations with higher frequencies of CYP3A5 expressers are needed to determine the full economic value of CYP3A5 genotype-guided tacrolimus dosing in SOT recipients.