Advertisement
The Journal of Heart and Lung Transplantation
International Society for Heart and Lung Transplantation.

Determining the impact of ex-vivo lung perfusion on hospital costs for lung transplantation: a retrospective cohort study

  • JK Peel
    Affiliations
    Department of Anesthesiology, University Health Network, University of Toronto

    Toronto Lung Transplant Program, University Health Network

    Institute of Health Policy, Management and Evaluation, Dalla Lana School for Public Health, University of Toronto
    Search for articles by this author
  • S Keshavjee
    Affiliations
    Toronto Lung Transplant Program, University Health Network

    Division of Thoracic Surgery, Toronto General Hospital, University Health Network

    Toronto General Hospital Research Institute, University Health Network, Toronto, ON, Canada
    Search for articles by this author
  • D Naimark
    Affiliations
    Institute of Health Policy, Management and Evaluation, Dalla Lana School for Public Health, University of Toronto

    Division of Nephrology, Sunnybrook Health Sciences Centre
    Search for articles by this author
  • M Liu
    Affiliations
    Toronto Lung Transplant Program, University Health Network

    Division of Thoracic Surgery, Toronto General Hospital, University Health Network

    Toronto General Hospital Research Institute, University Health Network, Toronto, ON, Canada
    Search for articles by this author
  • L Del Sorbo
    Affiliations
    Toronto Lung Transplant Program, University Health Network

    Toronto General Hospital Research Institute, University Health Network, Toronto, ON, Canada

    Interdepartmental Division of Critical Care Medicine, University of Toronto
    Search for articles by this author
  • M Cypel
    Affiliations
    Toronto Lung Transplant Program, University Health Network

    Division of Thoracic Surgery, Toronto General Hospital, University Health Network

    Toronto General Hospital Research Institute, University Health Network, Toronto, ON, Canada
    Search for articles by this author
  • K Barrett
    Affiliations
    Institute of Health Policy, Management and Evaluation, Dalla Lana School for Public Health, University of Toronto

    Interdepartmental Division of Critical Care Medicine, University of Toronto
    Search for articles by this author
  • Author Footnotes
    # Joined senior author
    EM Pullenayegum
    Footnotes
    # Joined senior author
    Affiliations
    Institute of Health Policy, Management and Evaluation, Dalla Lana School for Public Health, University of Toronto

    Senior Scientist, The Hospital for Sick Children
    Search for articles by this author
  • B Sander
    Correspondence
    Corresponding Author: Beate Sander, PhD, THETA Collaborative, Toronto General Hospital, Eaton Building, 10th Floor, 200 Elizabeth Street, Toronto, ON, M5G 2C4
    Affiliations
    Institute of Health Policy, Management and Evaluation, Dalla Lana School for Public Health, University of Toronto

    Toronto General Hospital Research Institute, University Health Network, Toronto, ON, Canada

    Adjunct Scientist, ICES, ON, Canada

    Adjunct Scientist, Public Health Ontario, ON, Canada
    Search for articles by this author
  • Author Footnotes
    # Joined senior author
Open AccessPublished:November 02, 2022DOI:https://doi.org/10.1016/j.healun.2022.10.016

      ABSTRACT

      Introduction

      Ex-vivo lung perfusion (EVLP) has improved organ utilization for lung transplantation, but it is not yet known whether the benefits of this technology offset its additional costs. We compared the institutional costs of lung transplantation before versus after EVLP was available to identify predictors of costs and determine the health-economic impact of EVLP.

      Methods

      We performed a retrospective, before-after, propensity-score weighted cohort study of patients wait-listed for lung transplant at University Health Network (UHN) in Ontario, Canada, between January 2005 and December 2019 using institutional administrative data. We compared costs, in 2019 Canadian Dollars ($), between patients referred for transplant before EVLP was available (Pre-EVLP) to after (Modern EVLP). Cumulative costs were estimated using a novel application of multi-state survival models. Predictors of costs were identified using weighted log-gamma generalized linear regression.

      Results

      1,199 patients met inclusion criteria (352 Pre-EVLP; 847 Modern EVLP). Mean total costs for the transplant hospitalization were $111,878 ($94,123 - $130,767) in the Pre-EVLP era and $110,969 ($87,714 - $136,000) in the Modern EVLP era. Cumulative five-year costs since referral were $278,777 ($82,575–$298,135) in the Pre-EVLP era and $293,680 ($252,832–$317,599) in the Modern EVLP era. We observed faster progression to transplantation when EVLP was available. EVLP availability was not a predictor of waitlist (cost ratio [CR] 1.04 [0.81–1.37]; p=0.354) or transplant costs (CR 1.02 [0.80–1.29]; p=0.425) but was associated with lower costs during post-transplant years 1&2 (CR 0.75 [0.58–1.06]; p=0.05) and post-transplant years 3+ (CR 0.43 [0.26–0.74]; p=0.001).

      Conclusions

      At our centre, EVLP availability was associated with faster progression to transplantation at no significant marginal cost.

      Keywords

      INTRODUCTION

      Demand for lung transplantation exceeds organ availability.
      • Yeung JC
      • Keshavjee S.
      Overview of Clinical Lung Transplantation.
      • Valapour M
      • Lehr CJ
      • Skeans MA
      • et al.
      OPTN/SRTR 2016 Annual Data Report: Lung.
      Canadian Institute for Health Information /Institut canadien d'information sur la santé.
      This organ scarcity renders lung transplantation an interesting health economic case: transplant programs aim to maximize population health using limited available organs.
      • Peel JK
      • Keshavjee S
      • Krahn M
      • Sander B.
      Economic evaluations and costing studies of lung transplantation: A scoping review.
      Health economic evaluation measures the value added by a technology by assessing costs and outcomes at the population level. Economic evaluation helps to quantify the health gained by transplantation and related technologies to facilitate judicious resource allocation, but at present, insufficient evidence exists for this purpose.
      • Peel JK
      • Keshavjee S
      • Krahn M
      • Sander B.
      Economic evaluations and costing studies of lung transplantation: A scoping review.
      Ex-vivo lung perfusion (EVLP) is a technology which may help relieve resource constraint and improve organ availability by sustaining potentially-useable donor lungs outside the body prior to transplant.
      • Van Raemdonck D
      • Neyrinck A
      • Cypel M
      • Keshavjee S.
      Ex-vivo lung perfusion.
      • Andreasson ASI
      • Dark JH
      • Fisher AJ.
      Ex vivo lung perfusion in clinical lung transplantation–State of the art.
      • Wallinder A
      • Ricksten SE
      • Silverborn M
      • et al.
      Early results in transplantation of initially rejected donor lungs after ex vivo lung perfusion: a case-control study.
      • Divithotawela C
      • Cypel M
      • Martinu T
      • et al.
      Long-term Outcomes of Lung Transplant With Ex Vivo Lung Perfusion.
      • Yeung JC
      • Krueger T
      • Yasufuku K
      • et al.
      Outcomes after transplantation of lungs preserved for more than 12 h: a retrospective study.
      EVLP is a major change in the practice of lung transplantation: between 69% and 79% of conventionally unusable organs have been successfully transplanted after EVLP, corresponding to a 30% increase in transplants at these centres.
      • Van Raemdonck D
      • Neyrinck A
      • Cypel M
      • Keshavjee S.
      Ex-vivo lung perfusion.
      ,
      • Divithotawela C
      • Cypel M
      • Martinu T
      • et al.
      Long-term Outcomes of Lung Transplant With Ex Vivo Lung Perfusion.
      ,
      • Boffini M
      • Ricci D
      • Barbero C
      • et al.
      Ex vivo lung perfusion increases the pool of lung grafts: analysis of its potential and real impact on a lung transplant program.
      Clinical outcomes after transplantation of EVLP-treated organs are promising, with generally no difference in hospital length-of-stay, graft dysfunction, or mortality between patients receiving EVLP-treated versus conventionally-treated organs.
      • Cypel M
      • Yeung JC
      • Liu M
      • et al.
      Normothermic ex vivo lung perfusion in clinical lung transplantation.
      • Cypel M
      • Yeung JC
      • Machuca T
      • et al.
      Experience with the first 50 ex vivo lung perfusions in clinical transplantation.
      • Machuca TN
      • Mercier O
      • Collaud S
      • et al.
      Lung Transplantation With Donation After Circulatory Determination of Death Donors and the Impact of Ex VivoLung Perfusion.
      • Zych B
      • Popov AF
      • Stavri G
      • et al.
      Early outcomes of bilateral sequential single lung transplantation after ex-vivo lung evaluation and reconditioning.
      • Jawitz OK
      • Raman V
      • Becerra D
      • et al.
      Lung Transplantation After Ex Vivo Lung Perfusion.
      Health systems are facing challenging decisions about how to incorporate EVLP. Concerns about the expense of EVLP impede its uptake, but the benefit of adding EVLP to lung transplantation programs may offset the additional costs required. Economic evaluation of lung transplant with, versus without, EVLP is crucial for evidence-informed allocation of healthcare resources to this technology.

      Canadian Agency for Drugs and Technologies in …. Guidelines for the economic evaluation of health technologies: Canada. 2006.

      Jacobs P, Yim R.Using Canadian Administrative Databases to Derive Economic Data for Health Technology Assessments. 2009.

      • Goodman C.
      HTA 101 - INTRODUCTION TO HEALTH TECHNOLOGY ASSESSMENT.
      Our primary objective was to compare the costs of lung transplantation, from the hospital perspective, of adults who were referred for transplant in Ontario before, versus after, EVLP was clinically available. A secondary objective was to identify predictors of costs, focusing on EVLP-availability.

      METHODS

      Study overview

      We performed a retrospective, before-after cohort study of patients wait-listed for first lung transplant in the Toronto Lung Transplant Program, University Health Network (UHN) in Ontario, Canada, between January 1, 2005 and December 1, 2019. UHN is the only lung transplant program in Ontario, one of four in Canada, and is the largest in the world. This study received institutional ethics approval (REB #20-5888). We adhered to the STROBE and RECORD reporting guidelines.
      • Elm von E
      • Altman DG
      • Egger M
      • et al.
      The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies.
      ,
      • Benchimol EI
      • Smeeth L
      • Guttmann A
      • et al.
      The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement.

      Data sources

      Clinical data was derived from the Organ Transplant Tracking Record, the Electronic Patient Record, and Toronto Lung Transplant Program Databases at UHN. Resource use and costs were accessed from the UHN's cost-coding department. Records were linked deterministically by medical record number.

      Population

      We included adults who were wait-listed for first lung transplant in Ontario during the study period. We excluded patients who were under 16 years old at referral, referred for combined multi-organ transplant or re-transplant, or referred but not wait-listed (incomplete or inappropriate referrals).

      Exposure

      Our exposure variable was EVLP availability at the time of referral. The introduction of EVLP into clinical practice represented a system-level change, impacting patients awaiting transplant as well as those undergoing the procedure. A comparison of recipients of EVLP-treated versus cold-static preservation alone-treated organs would not capture the system-level effect of this technology, nor would it allow for analysis of pre-transplant outcomes. We therefore stratified patients into “eras” based on referral date: pre-EVLP (2005-2008), early (2009-2012), and modern (2013-2019). The early era was thought to predominantly reflect the transition and consequences of EVLP implementation and was therefore excluded from analysis. To minimize misclassification, outcomes in the pre-EVLP era were censored on January 1, 2013. Where possible, we further stratified the modern era into recipients of lungs managed by cold-static preservation only (No EVLP) versus recipients of EVLP-treated lungs (EVLP). The study period was ended on December 1, 2019 because of the COVID-19 pandemic. All EVLP cases were performed using the Toronto EVLP protocol.
      • Cypel M
      • Yeung JC
      • Hirayama S
      • et al.
      Technique for Prolonged Normothermic Ex Vivo Lung Perfusion.

      Phases of care

      Observations were pieced into clinically meaningful, sequential “phases”, defined by event dates: referral, wait-list, transplant hospitalization, post-transplant years 1&2, and post-transplant years 3+.(Figure 1) The post-transplant period was divided into two phases (years 1&2, years 3+) because post-transplant surveillance at our centre involves frequent evaluation for the first two postoperative years, with decreased intensity of follow-up thereafter.
      Figure 1
      Figure 1Schematic of multi-state model
      For each patient in the study, observations were pieced into clinically meaningful phases: referral, wait-list, transplant hospitalization, post-transplant years 1&2, and post-transplant years 3+. At any point during the study, patients could be transitioned to end-of-life care and die. These phases of care exist along a continuum, and transition from one to the next occurred on specific event dates. We modelled this continuum as a k-progressive multi-state model, with the structure shown in this figure. Transition intensity (λ) indicates the instantaneous hazard of progression between states.
      To account for censoring between phases and higher costs around end-of-life, we organized costs accrued during the final days-of-life into separate perimortem phases. Perimortem phase durations were determined empirically by segmented regression of log-adjusted costs for the final year-of-life, stratified by EVLP era and the phase occupied at the time of death.
      • Bennell MC
      • Qiu F
      • Micieli A
      • et al.
      Identifying predictors of cumulative healthcare costs in incident atrial fibrillation: a population-based study.
      • Kim HJ
      • Fay MP
      • Feuer EJ
      • Midthune DN.
      Permutation tests for joinpoint regression with applications to cancer rates.
      • Wijeysundera HC
      • Wang Tomlinson
      • Krahn Ko D
      Techniques for estimating health care costs with censored data: an overview for the health services researcher.

      Chan BC-F, Cadarette SM, Wodchis WP, Krahn MD, Mittmann N. The lifetime cost of spinal cord injury in Ontario, Canada: A population-based study from the perspective of the public health care payer. February 2019:1-11. doi:10.1080/10790268.2018.1486622.

      (Supplemental Figure 3)

      Propensity score adjustment

      To account for baseline covariates differences across eras, we performed inverse probability of treatment weighted (IPTW) analysis. Weights were calculated using propensity scores derived from a generalized boosted model accounting for age at referral, sex, ABO blood-group, smoking history, hypertension, diabetes, estimated glomerular filtration rate (eGFR), transplantation from home versus hospital, and pre-transplant life-support requirement.
      • McCaffrey DF
      • Griffin BA
      • Almirall D
      • Slaughter ME
      • Ramchand R
      • Burgette LF.
      A tutorial on propensity score estimation for multiple treatments using generalized boosted models.
      • Austin PC
      • Stuart EA.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      • Austin PC.
      The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments.
      • Schulte PJ
      • Mascha EJ.
      Propensity Score Methods: Theory and Practice for Anesthesia Research.

      Olmos A, Govindasamy P. A practical guide for using propensity score weighting in R. Practical Assessment, Research, and Evaluation. 2015;20(13). doi:DOI: https://doi.org/10.7275/jjtm-r398.

      • Austin PC.
      Variance estimation when using inverse probability of treatment weighting (IPTW) with survival analysis.
      • Ali MS
      • Prieto-Alhambra D
      • Lopes LC
      • et al.
      Propensity Score Methods in Health Technology Assessment: Principles, Extended Applications, and Recent Advances.
      • Stuart EA.
      Matching methods for causal inference: A review and a look forward.
      Variables were selected a priori for their known or anticipated association with transplant outcomes. We did not adjust for donor organ characteristics because they are along the causal pathway for the associations of interest. Balance was assessed using density plots of propensity scores before versus after weighting, and standardized mean difference and Kolmogorov-Smirnov statistics for pairwise comparisons.
      • McCaffrey DF
      • Griffin BA
      • Almirall D
      • Slaughter ME
      • Ramchand R
      • Burgette LF.
      A tutorial on propensity score estimation for multiple treatments using generalized boosted models.
      ,
      • Zhou T
      • Tong G
      • Li F
      • Thomas LE
      • PSweight Li F.
      An R Package for Propensity Score Weighting Analysis.
      (Supplemental Figures 1 & 2) Weighted analyses were performed using the Average Treatment effect on the Treated (ATT) estimand, which allowed us to draw inferences about the consequence of withholding EVLP availability from those in the modern era, thus better accounting for selection-maturation bias.
      • Schulte PJ
      • Mascha EJ.
      Propensity Score Methods: Theory and Practice for Anesthesia Research.
      ,
      • Austin PC.
      Variance estimation when using inverse probability of treatment weighting (IPTW) with survival analysis.
      (Supplemental Methods)

      Outcomes

      Phase-based analysis of hospital costs

      Our primary outcomes were the phase-specific and cumulative costs for lung transplantation before versus after EVLP was available. Economic analyses were conducted from the hospital perspective. Consistent with this perspective, we included direct patient and program costs borne by the transplant institution, including labor and supply costs pertaining to EVLP. These costs did not include provincially-covered physician costs, costs for hospitalization at outside centres, costs for provincial transplant coordination, nor out-of-pocket and indirect costs borne by patients. The rationale for a hospital perspective was threefold: (i) lung transplantation is a centralized service in Ontario, with the majority of adult lung transplantation-related health resource use coordinated by, and accessed at, a single institution, so the institutional perspective would most directly capture these transplantation-specific costs; (ii) EVLP implementation is primarily the decision of institutional stakeholders and administrators, so the hospital perspective would provide these decision-makers with directly relevant information; (iii) UHN receives a high volume of referrals from jurisdictions outside of Ontario, representing patients who are not covered under provincial health insurance coverage, so the hospital perspective would capture the costs for these patients that would not be captured by analysis from the provincial perspective. All costs were inflated to 2019 Canadian Dollars ($) using the Consumer Price Index for healthcare.

      Government of Canada, Statistics Canada. Consumer Price Index, annual average, not seasonally adjusted. 2020. doi:https://doi.org/10.25318/1810000501-eng.

      Intensive care unit (ICU) costs represented expenditure paid out of the hospital ICU's budget; costs accrued by ICU patients but paid from other budgets were captured in estimates for total, but not ICU, costs. The cost of EVLP was corrected to account for the cost of EVLP cases that were declined for transplant. IPTW-weighted means for phase-specific total and component costs were calculated, with 95% confidence intervals (95% CI) derived from 1,000 bootstrap iterations.
      • Austin PC
      • Stuart EA.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      ,
      • Zhou T
      • Tong G
      • Li F
      • Thomas LE
      • PSweight Li F.
      An R Package for Propensity Score Weighting Analysis.
      ,
      • Manca A
      • Austin PC.
      Using propensity score methods to analyse individual patient level cost effectiveness data from observational studies.
      We developed a novel approach to estimate cumulative costs from referral to a five-year time-horizon. Typically, cumulative cost estimation using phase-based costs involves projection of lifetime survival duration using parametric survival regression, over which cumulative costs are summed. Phase durations in conventional analyses are determined empirically at the population-level using piecewise linear regression.

      Chan BC-F, Cadarette SM, Wodchis WP, Krahn MD, Mittmann N. The lifetime cost of spinal cord injury in Ontario, Canada: A population-based study from the perspective of the public health care payer. February 2019:1-11. doi:10.1080/10790268.2018.1486622.

      ,
      • Bhattacharya K
      • Bentley JP
      • Ramachandran S
      • et al.
      Phase-Specific and Lifetime Costs of Multiple Myeloma Among Older Adults in the US.
      A limitation of this approach is that phases are not bounded by an individual's event dates (i.e. a phase may be “days 0-30 from referral” rather than the period between referral and wait-list entry dates). To overcome this limitation, our novel phase-based costing methodology employs multi-state modelling. First, we constructed an IPTW-weighted multi-state model of the phases between referral and death using individual-level data.
      • Gran JM
      • Lie SA
      • Yeflaten I
      • Borgan R
      • Aalen OO
      Causal inference in multi-state models - sickness absence and work for 1145 participants after work rehabilitation.
      • Hoff R.
      The impact of completing upper secondary education - a multi-state model for work, education and health in young men.
      • Putter H
      • Fiocco M
      • Geskus RB.
      Tutorial in biostatistics: competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      mstate: an R package for the analysis of competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      The mstate package for estimation and prediction in non- and semi-parametric multi-state and competing risks models.
      (Figure 1) Using this model, we estimated the expected prevalence in each phase (state-occupancy probabilities) at 10-day intervals up to a 5-year time-horizon using the Landmark Aalen-Johansen estimator.
      • Putter H
      • Fiocco M
      • Geskus RB.
      Tutorial in biostatistics: competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      mstate: an R package for the analysis of competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      The mstate package for estimation and prediction in non- and semi-parametric multi-state and competing risks models.
      • Meira-Machado L
      • de Uña-Alvarez J
      • Cadarso-Suárez C
      • Andersen PK.
      Multi-state models for the analysis of time-to-event data.
      • Putter H.
      Tutorial in biostatistics: Competing risks and multi-state models Analyses using the mstate package.
      • Jackson C.
      Cumulative costs were then calculated by multiplying these state-occupancy probabilities by the mean 10-day cost for each phase, and then summing those 10-day costs up to the time-horizon;

      Chan BC-F, Cadarette SM, Wodchis WP, Krahn MD, Mittmann N. The lifetime cost of spinal cord injury in Ontario, Canada: A population-based study from the perspective of the public health care payer. February 2019:1-11. doi:10.1080/10790268.2018.1486622.

      ,
      • Bhattacharya K
      • Bentley JP
      • Ramachandran S
      • et al.
      Phase-Specific and Lifetime Costs of Multiple Myeloma Among Older Adults in the US.
      95% CI were derived from 2,000 bootstrap iterations.
      To confirm that differences in costs between eras were due to EVLP availability, and not maturation or history biases, we performed an interrupted time-series analysis.
      • Ho AMH
      • Phelan R
      • Mizubuti GB
      • et al.
      Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
      • Mascha EJ
      • Sessler DI.
      Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
      • Linden A
      • Adams JL.
      Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in programme evaluation.
      Log-adjusted costs were regressed by referral date and EVLP era using a piecewise IPTW-weighted linear model with fixed break-points.
      • Ho AMH
      • Phelan R
      • Mizubuti GB
      • et al.
      Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
      • Mascha EJ
      • Sessler DI.
      Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
      • Linden A
      • Adams JL.
      Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in programme evaluation.

      Identifying predictors of cost

      A secondary objective of this study was to identify predictors of costs. We constructed multivariable, IPTW-weighted log-gamma generalized linear regressions to estimate the effect of EVLP era, primary diagnosis (Supplemental Table 1), single versus bilateral transplant, donor type (donation after circulatory death (DCD) versus donation after neurologically defined death (NDD)), extended versus standard-criteria donor, wait-list duration (per-day), post-transplant mechanical ventilation >72hrs, and chronic lung allograft dysfunction (CLAD) on phase-specific costs.
      • Jones AM
      • Lomas J
      • Rice N.
      Healthcare Cost Regressions: Going Beyond the Mean to Estimate the Full Distribution.
      ,
      • Jones AM.
      Models for health care.
      95% CI were constructed using 1,000 bootstrap replicates. One-tailed non-parametric p-values were estimated as the probability of sampling bootstrapped regression coefficients equal to, or more extreme than, the nil-effect value.

      Statistical Analysis

      Analysis of variance (ANOVA) and chi-square tests were used for unadjusted descriptive analyses of continuous and categorical variables, respectively. All analyses were performed in R (R-project, ver. 4.0.1), with a two-sided level of significance of 0.05 unless otherwise specified.
      R: A language and environment for statistical computing.
      We performed sensitivity analyses comparing recipients of EVLP-treated organs to recipients of cold-static preserved (non-EVLP) organs (Supplemental Figures 4-6 & Supplemental Table 2).

      RESULTS

      Patient characteristics

      1,199 patients met the inclusion criteria: 352 were referred during the pre-EVLP era and 847 were referred during the modern era. In the modern era, 755 patients received an organ, 263 (35%) of whom received an EVLP-treated lung. Fewer patients in the modern era experienced post-transplant mechanical ventilation >72hrs (287 [38%] versus 133 [47%]; p=0.029).
      In unadjusted analysis, patients in the modern era tended to be older (56 ±14 years versus 51 ±14 years; p<0.001), and greater proportions of patients in the modern era had a smoking history (498 [59%] versus 172 [49%]; p=0.002) or required pre-transplant life-support (198 [24%] versus 30 [9%], p<0.001). A larger proportion of patients in the modern era were status 3 (high) urgency (100 [12%] versus 3 [1%]; p<0.001). We observed relatively more patients with restrictive lung disease, and fewer with suppurative disease, in the modern era. However, because the overall number of patients in the modern era increased, this observation may be due to increased referral of patients with restrictive disease rather than a relative decrease in enrollment of patients with suppurative disease. These results suggest that patient selection in the modern era entertained wait-list entry of more complex patients than in earlier eras. (Table 1)
      Table 1Description of study cohort
      WaitlistTransplantPost-transplant
      Pre-EVLPModern Era (No EVLP)Modern Era (EVLP)p-valuePre-EVLPModern Era (No EVLP)Modern Era (EVLP)p-valuePre-EVLPModern Era (No EVLP)Modern Era (EVLP)p-value
      Number of patients(N=352)(N=584)(N=263)(N=285)(N=492)(N=263)(N=279)(N=475)(N=258)
      Age51.4 (13.5)55.1 (14.0)57.4 (12.1)<0.00150.1 (13.8)54.5 (14.2)57.4 (12.1)<0.00150.0 (13.7)54.5 (14.2)57.3 (12.2)<0.001
      Female149 (42.3%)262 (44.9%)88 (33.5%)0.007113 (39.6%)212 (43.1%)88 (33.5%)0.036110 (39.4%)207 (43.6%)84 (32.6%)0.014
      Weight (kg)68.6 (15.9)69.4 (16.3)73.4 (17.1)<0.00168.4 (16.4)69.5 (16.5)73.4 (17.1)<0.00168.5 (16.4)69.3 (16.5)73.6 (17.1)<0.001
      Height (cm)167 (9.30)166 (10.1)169 (9.58)<0.001168 (9.20)167 (10.1)169 (9.58)<0.001168 (9.16)167 (10.1)169 (9.55)<0.001
      Underweight BMI37 (10.5%)49 (8.4%)16 (6.1%)0.14935 (12.3%)40 (8.1%)16 (6.1%)0.0335 (12.5%)40 (8.4%)16 (6.2%)0.031
      Healthy BMI151 (42.9%)253 (43.3%)97 (36.9%)0.188128 (44.9%)224 (45.5%)97 (36.9%)0.058124 (44.4%)216 (45.5%)95 (36.8%)0.066
      Overweight BMI104 (29.5%)178 (30.5%)100 (38.0%)0.0576 (26.7%)145 (29.5%)100 (38.0%)0.0175 (26.9%)140 (29.5%)97 (37.6%)0.018
      Obese BMI54 (15.3%)101 (17.3%)50 (19.0%)0.48146 (16.1%)82 (16.7%)50 (19.0%)0.62945 (16.1%)78 (16.4%)50 (19.4%)0.526
      ABO A148 (42.0%)219 (37.5%)100 (38.0%)0.362120 (42.1%)185 (37.6%)100 (38.0%)0.435117 (41.9%)180 (37.9%)99 (38.4%)0.525
      ABO AB11 (3.1%)21 (3.6%)9 (3.4%)0.9299 (3.2%)19 (3.9%)9 (3.4%)0.879 (3.2%)18 (3.8%)9 (3.5%)0.92
      ABO B44 (12.5%)76 (13.0%)21 (8.0%)0.09635 (12.3%)70 (14.2%)21 (8.0%)0.04335 (12.5%)69 (14.5%)21 (8.1%)0.043
      ABO O149 (42.3%)268 (45.9%)133 (50.6%)0.128121 (42.5%)218 (44.3%)133 (50.6%)0.131118 (42.3%)208 (43.8%)129 (50.0%)0.156
      Obstructive Disease103 (29.3%)140 (24.0%)81 (30.8%)0.06295 (33.3%)126 (25.6%)81 (30.8%)0.05692 (33.0%)122 (25.7%)79 (30.6%)0.082
      Restrictive Disease160 (45.5%)316 (54.1%)146 (55.5%)0.015114 (40.0%)258 (52.4%)146 (55.5%)<0.001112 (40.1%)249 (52.4%)144 (55.8%)<0.001
      Vascular Disease23 (6.5%)42 (7.2%)10 (3.8%)0.16415 (5.3%)32 (6.5%)10 (3.8%)0.29415 (5.4%)31 (6.5%)10 (3.9%)0.322
      Suppurative Disease60 (17.0%)78 (13.4%)24 (9.1%)0.01757 (20.0%)69 (14.0%)24 (9.1%)0.00156 (20.1%)67 (14.1%)23 (8.9%)0.001
      Other diagnosis6 (1.7%)8 (1.4%)NR0.598NR7 (1.4%)NR0.71NR6 (1.3%)NR0.763
      Total Lung Capacity (L)5.54 (2.33)5.00 (2.29)5.10 (2.31)<0.0015.54 (2.33)5.00 (2.29)5.10 (2.31)<0.0015.54 (2.33)5.00 (2.31)5.11 (2.31)<0.001
      Urgency Status 1168 (47.7%)200 (34.2%)129 (49.0%)<0.001168 (58.9%)200 (40.7%)129 (49.0%)<0.001164 (58.8%)191 (40.2%)124 (48.1%)<0.001
      Urgency Status 2114 (32.4%)216 (37.0%)108 (41.1%)0.082114 (40.0%)216 (43.9%)108 (41.1%)0.526112 (40.1%)210 (44.2%)108 (41.9%)0.536
      Urgency Status 3NR74 (12.7%)26 (9.9%)<0.001NR74 (15.0%)26 (9.9%)<0.001NR73 (15.4%)26 (10.1%)<0.001
      Hypertension239 (67.9%)401 (68.7%)215 (81.7%)<0.001239 (83.9%)401 (81.5%)215 (81.7%)0.692238 (85.3%)397 (83.6%)211 (81.8%)0.545
      Smoker172 (48.9%)316 (54.1%)182 (69.2%)<0.001172 (60.4%)316 (64.2%)182 (69.2%)0.096170 (60.9%)308 (64.8%)178 (69.0%)0.148
      Diabetic55 (15.6%)92 (15.8%)52 (19.8%)0.29355 (19.3%)92 (18.7%)52 (19.8%)0.93554 (19.4%)91 (19.2%)52 (20.2%)0.947
      eGFR94.0 (25.8)90.7 (24.4)85.8 (22.9)<0.00194.0 (25.8)90.7 (24.4)85.8 (22.9)<0.00193.8 (25.8)90.7 (24.4)85.6 (23.0)<0.001
      Transplant from hospital61 (17.3%)112 (19.2%)43 (16.3%)0.11261 (21.4%)112 (22.8%)43 (16.3%)0.11260 (21.5%)112 (23.6%)43 (16.7%)0.091
      Pre-transplant life support30 (8.5%)133 (22.8%)65 (24.7%)<0.00130 (10.5%)133 (27.0%)65 (24.7%)<0.00130 (10.8%)123 (25.9%)61 (23.6%)<0.001
      Mechanical ventilation >72hrs----133 (46.7%)182 (37.0%)105 (39.9%)0.029130 (46.6%)176 (37.1%)102 (39.5%)0.034
      CLAD--------137 (49.1%)121 (25.5%)52 (20.2%)<0.001
      Each phase of care represents a unique sub cohort because of the censoring that occurs when some patients proceed to the next steps of care while others do not. For this reason, we present the results separately for each phase of care. The Modern Era (EVLP) group in the waitlist phase represents only those patients who survived the waitlist phase and underwent transplantation. The data in this table are the unweighted values describing the study participants. Comparisons between groups are unadjusted analyses, using ANOVA and chi-square tests for continuous and categorical variables, respectively. Overall, in unadjusted analysis before propensity score adjustment, we observed that patients in the Modern era tended to be older and with higher frequency of comorbid conditions. Fewer patients in the Modern EVLP era required mechanical ventilation >72hrs immediately post-transplant, and fewer patients experienced Chronic Lung Allograft Dysfunction in the Modern EVLP era. Table cells representing data from fewer than 6 patients were not reported (NR) for publication to minimize the risk of re-identification. (ANOVA = analysis of variance; BMI = body mass index; CLAD = chronic lung allograft dysfunction; eGFR = estimated glomerular filtration rate; EVLP = ex-vivo lung perfusion)

      Phase-specific cost estimates

      IPTW-weighted mean total, component, and perimortem costs for each phase are summarized in Table 2 and Figure 2. In the pre-EVLP era, mean total costs per-month for the referral and waitlist phases were $1,634 ($694–$2,851) and $14,904 ($11,776 - $18,622) respectively. Modern era monthly costs for the same pre-transplant phases were $2,022 ($1,607 - $2,518) and $26,507 ($23,632 - $29,465) respectively. Mean total costs for the transplant hospitalization were $111,878 ($94,123 - $130,767) in the pre-EVLP era. In the modern era, transplantation with No EVLP cost $122,008 ($91,457 - $163,954), while transplantation with EVLP cost $90,923 ($75,788 - $109,428), though the median costs for transplantation in this era appeared higher with EVLP versus without. Overall, costs for transplant hospitalization appear similar before versus after EVLP implementation. Costs for ICU care during the transplant hospitalization were $45,461 ($35,990 - $56,333) in the pre-EVLP era, $59,890 ($36,061 - $91,399) in the modern era with No EVLP, and $24,839 ($18,377 - $32,201) in the modern era with EVLP. Post-transplant costs were lower in the modern era. Mean monthly total costs for the first two years post-transplant were $3,132 ($2,339 - $4,148) versus $2,183 ($1,919 - $2,523) in the pre-EVLP and modern eras, respectively. Mean total costs for each month beyond three years from discharge were $701 ($476 - $1,025) in the pre-EVLP era versus $320 ($227 - $457) in the modern era.
      Table 2Phase-specific total and component costs
      Referral PhaseWaitlist PhaseTransplant PhasePost-Transplant Years 1&2Post-Transplant Years 3+
      Pre-EVLP EraModern Era (No EVLP)Modern Era (EVLP)Modern EraPre-EVLP EraModern Era (No EVLP)Modern Era (EVLP)Modern EraPre-EVLP EraModern Era (No EVLP)Modern Era (EVLP)Modern EraPre-EVLP EraModern Era (No EVLP)Modern Era (EVLP)Modern EraPre-EVLP EraModern Era (No EVLP)Modern Era (EVLP)Modern Era
      Total Cost$25,498 ($10,821 - $44,468)$26,681 ($20,239 - $33,500)$24,771 ($14,880 - $37,673)$26,081 ($20,731 - $32,482)$89,919 ($71,051 - $112,353)$93,006 ($82,211 - $105,304)$95,026 ($77,048 - $117,292)$93,658 ($83,501 - $104,110)$111,878 ($94,123 - $130,767)$122,008 ($91,457 - $163,954)$90,923 ($75,788 - $109,428)$110,969 ($87,714 - $136,000)$69,105 ($51,620 - $91,543)$49,412 ($41,673 - $58,135)$48,587 ($39,838 - $58,889)$49,121 ($43,177 - $56,778)$43,435 ($29,462 - $63,498)$16,082 ($9,806 - $25,224)$15,062 ($9,438 - $21,976)$15,734 ($11,194 - $22,494)
      Phase duration [days]468 (406 - 539)372 (347 - 398)419 (382 - 456)387 (366 - 408)181 (157 - 207)104 (93 - 116)110 (93 - 129)106 (97 - 116)32 (29 - 36)45 (38 - 52)37 (31 - 44)42 (37 - 47)662 (634 - 686)677 (662 - 690)672 (652 - 690)675 (663 - 686)1,858 (1,712 - 1,996)1,492 (1,437 - 1,546)1,450 (1,380 - 1,521)1,477 (1,432 - 1,523)
      Monthly Cost$1,634 ($694 - $2,851)$2,152 ($1,632 - $2,702)$1,774 ($1,065 - $2,697)$2,022 ($1,607 - $2,518)$14,904 ($11,776 - $18,622)$26,829 ($23,715 - $30,376)$25,916 ($21,013 - $31,989)$26,507 ($23,632 - $29,465)$104,886 ($88,240 - $122,594)$81,339 ($60,971 - $109,303)$73,721 ($61,450 - $88,725)$79,264 ($62,653 - $97,143)$3,132 ($2,339 - $4,148)$2,190 ($1,847 - $2,576)$2,169 ($1,778 - $2,629)$2,183 ($1,919 - $2,523)$701 ($476 - $1,025)$323 ($197 - $507)$312 ($195 - $455)$320 ($227 - $457)
      Labour Costs$13,024 ($4,997 - $24,508)$14,166 ($10,623 - $18,093)$13,157 ($7,505 - $19,821)$13,849 ($10,737 - $17,544)$47,119 ($36,686 - $59,796)$50,776 ($44,104 - $57,202)$52,081 ($42,022 - $64,733)$51,197 ($45,724 - $57,222)$57,153 ($47,777 - $66,543)$67,395 ($48,128 - $92,308)$40,062 ($31,962 - $48,906)$57,689 ($46,181 - $73,457)$34,158 ($24,564 - $46,419)$26,616 ($22,255 - $31,639)$26,188 ($21,053 - $31,623)$26,465 ($23,216 - $30,135)$23,984 ($15,604 - $35,958)$9,073 ($5,336 - $14,624)$8,261 ($5,239 - $12,553)$8,796 ($6,105 - $12,912)
      Supply Costs$4,158 ($1,918 - $6,885)$4,590 ($3,518 - $5,914)$4,311 ($2,431 - $6,359)$4,502 ($3,546 - $5,623)$15,492 ($11,907 - $19,757)$15,896 ($13,926 - $17,800)$15,921 ($13,110 - $19,206)$15,904 ($14,300 - $17,542)$20,164 ($17,119 - $23,356)$20,527 ($16,101 - $25,997)$13,624 ($11,162 - $17,170)$18,076 ($14,727 - $21,785)$9,817 ($7,224 - $13,008)$7,150 ($5,856 - $8,626)$6,838 ($5,438 - $8,561)$7,040 ($6,003 - $8,101)$6,064 ($4,013 - $8,582)$2,050 ($1,287 - $3,256)$2,068 ($1,291 - $3,091)$2,056 ($1,427 - $2,819)
      Inpatient Costs$13,863 ($5,220 - $25,500)$18,531 ($13,305 - $23,989)$24,655 ($12,977 - $40,146)$20,524 ($15,691 - $26,071)$72,511 ($55,831 - $90,506)$77,512 ($68,840 - $86,668)$75,994 ($61,587 - $93,434)$74,879 ($67,549 - $82,677)NANANANA$38,175 ($24,107 - $55,232)$26,105 ($20,471 - $33,293)$24,784 ($18,461 - $32,061)$25,622 ($21,088 - $30,709)$9,898 ($6,018 - $15,240)$3,893 ($2,219 - $6,656)$3,894 ($2,192 - $5,899)$5,439 ($3,534 - $8,097)
      Outpatient Costs$2,967 ($2,639 - $3,281)$4,777 ($4,299 - $5,403)$4,871 ($4,466 - $5,354)$4,117 ($3,816 - $4,525)$5,060 ($4,314 - $5,714)$4,193 ($3,810 - $4,642)$4,077 ($3,526 - $4,634)$3,826 ($3,516 - $4,137)NANANANA$17,823 ($16,438 - $19,249)$13,424 ($12,702 - $14,157)$12,774 ($11,786 - $13,818)$13,195 ($12,627 - $13,778)$3,654 ($2,907 - $4,551)$1,248 ($929 - $1,692)$1,313 ($1,054 - $1,619)$1,513 ($1,226 - $1,853)
      ICU Costs$10,938 ($2,047 - $21,336)$6,620 ($4,312 - $9,480)$8,997 ($3,631 - $15,821)$5,671 ($3,961 - $7,792)$31,100 ($21,853 - $42,034)$25,379 ($21,085 - $30,341)$24,610 ($17,076 - $33,917)$22,942 ($19,298 - $27,223)$45,461 ($35,990 - $56,333)$59,890 ($36,061 - $91,399)$24,839 ($18,377 - $32,201)$47,286 ($32,631 - $69,210)$8,973 ($3,903 - $18,001)$5,076 ($2,221 - $8,610)$3,746 ($1,661 - $6,735)$4,768 ($2,696 - $7,667)$1,983 ($930 - $3,273)$1,247 ($169 - $2,780)$725 ($61 - $1,548)$1,980 ($612 - $4,292)
      ICU Duration [days]NANANANANANANANA12 (10 - 15)10 (9 - 12)9 (7 - 11)10 (8 - 11)NANANANANANANANA
      Perimortem 10-day CostsNANANANA$10,652 ($7,370 - $14,113)NANA$9,374 ($6,869 - $12,839)$48,254 ($6,373 - $64,092)$36,742 ($15,277 - $57,835)$30,090 ($61 - $60,120)$34,842 ($11,661 - $53,523)$18,596 ($12,030 - $25,227)$12,968 ($8,216 - $18,592)$4,469 ($2,136 - $7,015)$8,713 ($5,990 - $12,596)$5,589 ($2,710 - $9,931)$6,625 ($2,574 - $10,953)$16,710 ($16,710 - $16,710)$7,893 ($3,477 - $12,622)
      Total and component costs per patient for each phase of care were estimated as the IPTW-weighted average with 95% confidence intervals (in parentheses) derived using 1,000 bootstrap replicates. Inpatient, outpatient, and ICU costs in the pre-transplant and post-transplant phases were adjusted by the proportion of patients who experienced these costs. (i.e., since some patients on the wait-list did not accrue inpatient costs, the estimate for inpatient costs was corrected to account for this). Since the labour and supply costs for EVLP were not estimated separately, the specific cost of EVLP was included in the transplant phase, under total costs only. ICU costs reflected all costs paid by the hospital's ICU operating budget and may therefore include ICU outreach activities in hospital as well as true ICU admission. For the transplant operation, ICU length-of-stay is reported as the duration between the transplant operation to first discharge from ICU; readmission to the ICU was not included in this measure. Transplant hospitalization phase length indicates hospital length-of-stay. (EVLP = Ex-vivo lung perfusion; ICU = Intensive care unit; IPTW = Inverse-probability of treatment weight)
      Figure 2
      Figure 2Boxplot of phase-specific total cost estimates
      This figure depicts the median and IQR for total cost estimates in each phase, stratified by EVLP era and EVLP use. All estimates were IPTW-weighted and inflated to 2019 CAD. Total phase-specific costs were similar across each phase of care. (CAD = Canadian Dollars; EVLP = Ex-vivo lung perfusion; IPTW = Inverse probability of treatment weight; IQR = interquartile range)
      In IPTW-weighted interrupted time-series analysis comparing log-adjusted costs from the modern era against projections from the pre-EVLP era, we observed no significant change in trend occurring at the start of the modern era for the referral (p=0.444), wait-list (p=0.148), and transplant (p=0.285) phases.
      • Ho AMH
      • Phelan R
      • Mizubuti GB
      • et al.
      Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
      ,
      • Mascha EJ
      • Sessler DI.
      Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
      (Figure 3) Conversely, we observed significant reductions in post-transplant log-adjusted costs occurring in the modern era (p<0.001 for postoperative years 1&2; p=0.003 for postoperative years 3+), and significant changes in trend for postoperative years 1&2 (p<0.001) and postoperative years 3+ (p=0.003), suggesting that postoperative costs may decline after EVLP becomes available.
      Figure 3
      Figure 3Time-series plots
      Among the greatest threats to validity of before-after studies are the risks of maturation and history biases.
      • Ho AMH
      • Phelan R
      • Mizubuti GB
      • et al.
      Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
      • Mascha EJ
      • Sessler DI.
      Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
      • Linden A
      • Adams JL.
      Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in programme evaluation.
      To overcome this, we performed an interrupted time-series analysis using segmented regression. This figure shows the relationship of total phase-specific costs to referral date. The Pre-EVLP era trend (shown in red) is continued with the dashed line, and the Modern EVLP era trend is shown in blue. The p-value at the right of each plot indicates whether the trend in the Modern EVLP era is significantly different from the trend on the Pre-EVLP era. The p-value shown in the middle of the figure (at the breakpoint) indicates whether the mean log costs after the interruption (EVLP implementation) was significantly different from the mean log costs before.

      Predictors of phase-specific costs

      A separate IPTW-weighted log-gamma generalized linear model was constructed to identify predictors of total cost in each phase. (Table 3) In the referral and wait-list phases, we did not observe an association between EVLP era and costs (cost ratio [CR] 1.12 [0.66–2.31]; p=0.346 and 1.04 [0.81–1.37]; p=0.354, respectively). Restrictive and pulmonary-vascular diseases were associated with pre-transplant costs. (Table 3) Multivariable regression of total costs for the transplant hospitalization showed that EVLP era was not associated with costs (1.02 [0.80–1.29]; p=0.425) whereas mechanical ventilation >72hrs immediately post-operatively was associated with higher costs (CR 2.16 [1.71–2.73]; p <0.001). The modern era was associated with significantly lower post-transplant years 1&2 costs (CR 0.75 [0.58–1.06]; p=0.05). Accounting for EVLP era and patient characteristics, transplant with an extended-criteria organ was associated with higher post-transplant years 1&2 costs (CR 1.29 [1.04–1.64]; p=0.02). Similarly, the modern era was associated with significantly lower post-transplant years 3+ costs (CR 0.43 [0.26–0.74]; p=0.001), and CLAD was associated with larger costs in this phase (CR 1.7 [1.0–2.71]; p=0.026).
      Table 3Predictors of total cost in each phase of care
      ReferralWaitlistTransplantPost-Transplant Years 1&2Post-Transplant Years 3+
      Modern EVLP Era1.13 (0.66 - 2.31)p=0.3461.04 (0.81 - 1.37)p=0.3541.02 (0.8 - 1.29)p=0.4250.75 (0.58 - 1.06)p=0.050.43 (0.26 - 0.74)p=0.001
      Pre-EVLP EraReferenceReferenceReferenceReferenceReference
      Restrictive lung disease2.81 (1.65 - 4.49)p=0.0011.57 (1.19 - 2.06)p <0.0010.97 (0.74 - 1.21)p=0.3830.85 (0.65 - 1.19)p=0.181.09 (0.63 - 1.89)p=0.394
      Pulmonary vascular disease3.18 (1.39 - 6.45)p=0.0042.16 (1.18 - 3.64)p=0.0031.13 (0.77 - 1.62)p=0.2651.43 (0.79 - 2.56)p=0.1190.58 (0.22 - 1.49)p=0.126
      Suppurative lung disease2.08 (0.7 - 4.45)p=0.0761.51 (1.04 - 2.16)p=0.011.28 (0.92 - 1.75)p=0.0771.56 (0.88 - 2.53)p=0.0630.57 (0.3 - 1.19)p=0.066
      Other diagnoses4.3 (0.41 - 13.71)p=0.3513.01 (0.29 - 6.1)p=0.0643.3 (0.95 - 6.3)p=0.0311.07 (0.59 - 1.87)p=0.4211.11 (0.08 - 3.16)p=0.432
      Obstructive lung diseaseReferenceReferenceReferenceReferenceReference
      DCD Donor0.88 (0.72 - 1.11)p=0.1540.95 (0.71 - 1.31)p=0.3360.82 (0.41 - 1.53)p=0.252
      NDD DonorReferenceReferenceReference
      Single lung transplant0.94 (0.76 - 1.19)p=0.3061.02 (0.79 - 1.3)p=0.4260.91 (0.5 - 1.62)p=0.406
      Double lung transplantReferenceReferenceReference
      Extended criteria donor1.15 (0.93 - 1.41)p=0.0931.29 (1.04 - 1.64)p=0.0171.16 (0.71 - 2.04)p=0.304
      Standard criteria donorReferenceReferenceReference
      Waitlist duration (days)1 (1 - 1)p=0.2051 (0.999 - 1)p=0.4441 (0.998 - 1.001)p=0.429
      Mechanical ventilation >72hrs2.16 (1.71 - 2.73)p <0.0011.24 (0.96 - 1.63)p=0.060.82 (0.49 - 1.39)p=0.282
      Mechanical ventilation <72hrsReferenceReferenceReference
      CLAD1.24 (0.95 - 1.65)p=0.0631.7 (1 - 2.71)p=0.026
      No CLADReferenceReference
      A separate log-gamma generalized linear regression was constructed to identify predictors of total costs in each phase. Results are presented as cost ratios, on the multiplicative scale, with values greater than 1 indicating higher costs. For instance, in the referral phase, restrictive lung disease was associated with costs 2.81 (1.65–4.49) times higher than with obstructive lung disease. 95% Confidence intervals and one-tailed p-values were derived by non-parametric bootstraps. (CLAD = Chronic Lung Allograft Dysfunction; EVLP = Ex-Vivo Lung Perfusion; DCD = Donation after circulatory death; NDD = Neurologically Defined Donor)

      Cumulative cost estimates

      Compared with the pre-EVLP era, patients in the modern era were more likely to proceed to the post-transplant phases, and appeared to do so in a shorter time from referral (Figure 4). Accordingly, we observed wait-list occupancy at any one time to be lower since EVLP became available. The multi-state model suggests shorter time-to-transplant when EVLP was available.
      Figure 4
      Figure 4Multi-state occupancy probabilities
      This figure depicts the probability of an individual occupying each phase of care over time since referral in the Pre-EVLP era and the Modern EVLP era. Probabilities were derived using an IPTW-weighted landmark Aalen-Johansen estimator.
      • Gran JM
      • Lie SA
      • Yeflaten I
      • Borgan R
      • Aalen OO
      Causal inference in multi-state models - sickness absence and work for 1145 participants after work rehabilitation.
      ,
      • Hoff R.
      The impact of completing upper secondary education - a multi-state model for work, education and health in young men.
      ,
      • Maltzahn N
      • Hoff R
      • Aalen OO.
      A hybrid landmark Aalen-Johansen estimator for transition probabilities in partially non-Markov multi-state models.
      Because the Markov assumption was violated for all transitions, the landmark Aalen-Johansen was the preferred estimator, as it does not depend on this assumption.
      • Putter H
      • Fiocco M
      • Geskus RB.
      Tutorial in biostatistics: competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      mstate: an R package for the analysis of competing risks and multi-state models.
      • de Wreede LC
      • Fiocco M
      • Putter H.
      The mstate package for estimation and prediction in non- and semi-parametric multi-state and competing risks models.
      • Meira-Machado L
      • de Uña-Alvarez J
      • Cadarso-Suárez C
      • Andersen PK.
      Multi-state models for the analysis of time-to-event data.
      • Putter H.
      Tutorial in biostatistics: Competing risks and multi-state models Analyses using the mstate package.
      • Jackson C.
      Compared with the Pre-EVLP, the probability of an individual entering the post-transplant states (red solid [years 1&2] and dashed [years 3+]) was higher and occurred earlier in the Modern EVLP era. Accordingly, the probability of being in the wait-list state (green) is lower in Modern EVLP era, indicating that occupancy in the wait list at any one time is lower since EVLP became available. In addition, the overall probability of death (black) is modestly lower in the Modern EVLP era versus the Pre-EVLP era. Together, these suggest shorter time to transplant and more efficient progress through the transplant program when EVLP is available. (EVLP = Ex-vivo lung perfusion; IPTW = Inverse probability of treatment weight)
      We estimated cumulative costs at five-years since referral to be $278,777 ($82,575–$298,135) in the pre-EVLP era and $293,680 ($252,832–$317,599) in the modern era. The absolute difference between means was $14,903, reflecting an approximate 5% relative increase.

      DISCUSSION

      This study estimated costs, from the hospital perspective, for lung transplantation before versus after EVLP was available. Our cost estimates of $110,959 ($87,739–$137,875) for the transplant hospitalization with EVLP available and $108,422 ($93,140–$124,918) without EVLP available are commensurate to the evidence from other jurisdictions. We previously reported that the institutional cost of lung transplantation ranged between $16,748 and $361,959.
      • Peel JK
      • Keshavjee S
      • Krahn M
      • Sander B.
      Economic evaluations and costing studies of lung transplantation: A scoping review.
      Two studies in that review evaluated the costs of EVLP, using data from the DEVELOP-UK study.
      • Peel JK
      • Keshavjee S
      • Krahn M
      • Sander B.
      Economic evaluations and costing studies of lung transplantation: A scoping review.
      ,
      • McMeekin N
      • Chrysos AE
      • Vale L
      • Fisher AJ.
      Incorporating ex-vivo lung perfusion into the UK adult lung transplant service: an economic evaluation and decision analytic model.
      ,
      • Fisher A
      • Andreasson A
      • Chrysos A
      • et al.
      An observational study of Donor Ex Vivo Lung Perfusion in UK lung transplantation: DEVELOP-UK.
      Fisher et al. (2016) estimated one-year costs, from the United Kingdom healthcare-payer perspective, for transplant with EVLP as $219,518±$92,990, versus $94,082±$67,338 without EVLP over the same time-horizon.
      • Fisher A
      • Andreasson A
      • Chrysos A
      • et al.
      An observational study of Donor Ex Vivo Lung Perfusion in UK lung transplantation: DEVELOP-UK.
      However, the DEVELOP-UK study was limited because patient enrollment was stopped early due to excess mortality and because their EVLP protocol deviated from accepted protocols. A study from the United States, published after our review, found higher costs, from the institutional perspective, for the transplant hospitalization with EVLP transplants costing $200,404 (interquartile range IQR $161,966–$362,143) versus cold-static preserved standard criteria organs $154,709 (IQR $137,890–$218,180]).
      • Halpern SE
      • Kesseli SJ
      • Au S
      • et al.
      Lung transplantation after ex vivo lung perfusion versus static cold storage: An institutional cost analysis.
      However, Halpern et al. (2021) reported that this difference was not statistically significant, and that the use of EVLP at their centre was overall profitable.
      • Halpern SE
      • Kesseli SJ
      • Au S
      • et al.
      Lung transplantation after ex vivo lung perfusion versus static cold storage: An institutional cost analysis.
      Although that study did not adjust for potential confounders and ultimately concluded that further investigation was required, these results appear to complement those of our study. Neither of these studies considered pre-transplant or post-transplant costs beyond 1-year; inclusion of these is an added strength of our study.
      • Fisher A
      • Andreasson A
      • Chrysos A
      • et al.
      An observational study of Donor Ex Vivo Lung Perfusion in UK lung transplantation: DEVELOP-UK.
      ,
      • Halpern SE
      • Kesseli SJ
      • Au S
      • et al.
      Lung transplantation after ex vivo lung perfusion versus static cold storage: An institutional cost analysis.
      We observed more complex patient characteristics in the modern era in unadjusted analyses (Table 1), and higher IPTW-weighted mean waitlist costs (Table 2). These findings suggest that EVLP availability has allowed for more complex patients to be considered for transplant, when previously these patients would be considered too high risk. Further, while we observed that post-transplant mechanical ventilation >72hrs was associated with higher costs, significantly fewer patients in the modern era required prolonged ventilation. Taken together, these findings demonstrate that EVLP has increased access to transplantation while improving postoperative outcomes – findings consistent with the evidence that EVLP transforms clinical practice for pre-transplant assessment, organ allocation, and postoperative care.
      • Van Raemdonck D
      • Neyrinck A
      • Cypel M
      • Keshavjee S.
      Ex-vivo lung perfusion.
      ,
      • Divithotawela C
      • Cypel M
      • Martinu T
      • et al.
      Long-term Outcomes of Lung Transplant With Ex Vivo Lung Perfusion.
      ,
      • Boffini M
      • Ricci D
      • Barbero C
      • et al.
      Ex vivo lung perfusion increases the pool of lung grafts: analysis of its potential and real impact on a lung transplant program.
      Alongside our finding that EVLP era was not a significant predictor of pre-transplant or transplant hospitalization costs (i.e: that EVLP implementation did not significantly increase phase-specific costs)(Table 4), these results demonstrate important clinical improvements from EVLP at no significant marginal cost.
      Overall, we interpret our results to indicate that the cost of EVLP represents a relatively small contribution to the total costs of transplantation since the addition of EVLP to our lung transplantation program was not associated with higher overall costs. While we observed slightly higher pre-transplant costs in the modern era, perhaps attributable to increased care requirements and novel medical therapies for waitlist patients, the shorter waitlist duration afforded by EVLP may be offsetting these costs. Our observation of shorter pre-transplant durations in the modern era suggests that the clinical benefit of EVLP may be offsetting some of the costs. Since we observed shorter ICU stay and lower ICU costs among recipients of EVLP-treated lungs, this cost-savings might be offsetting the costs of EVLP in the modern era. Lower post-transplant costs could be a function of truly decreased post-transplant care requirements resulting from improved organ quality at transplantation due to EVLP, but this may also reflect greater comfort at our centre in having post-transplant patients evaluated by their community respirologists. Future evaluation of provincial healthcare spending could help determine if post-transplant costs in the modern era are lower, or simply transferred from the transplant institution to the broader healthcare system. While it's possible that transplant hospitalization costs were declining over time, perhaps related to practice changes unrelated to EVLP, and that the implementation of EVLP maintained those costs, we find this explanation unlikely based on our time-series analysis. Interrupted time-series analysis showed no significant increase in pre-transplant or transplant phase costs at the time of EVLP implementation, nor did it demonstrate a significantly different slope for changing costs over time. (Figure 3) While the purpose of our study was to measure costs and identify clinical predictors, our study did not produce sufficient evidence to confirm these interpretations: exploration of the mechanisms underlying the economic differences in the pre-EVLP and modern eras should be the focus of future work.
      While observational research allows for generalization outside of the clinical trial setting, as it reflects real-world experiences (and expenditure), there are limitations to this methodology. Incomplete data and censoring are inevitable with retrospective studies using routinely-collected data. We mitigated the effect of this by analyzing each phase of care separately, which overcomes this common limitation of administrative data.
      • Wijeysundera HC
      • Wang Tomlinson
      • Krahn Ko D
      Techniques for estimating health care costs with censored data: an overview for the health services researcher.
      Our creation of perimortem phases further reduces bias from censoring due to death. To further reduce the risk of bias by censoring, we constrained the observation window of the pre-EVLP era so that measurements in each EVLP era were collected for similar duration. Despite this, shorter observation times in the modern era could contribute to the observed lower CLAD incidence and post-transplant years 3+ costs: these observations should be considered circumspectly. A limitation of this study was that EVLP costs were provided from our institutional accounting department as a single value per case: we could not discern from this value what portion of EVLP expenditure represented the specific costs for supplies versus labour. Costs for organ procurement were not included in this analysis, and these costs may be high. However, this is consistent with the perspective of the transplant institution, because organ procurement in Ontario is coordinated by an independent government agency (Trillium Gift of Life Network). This institutional perspective is narrow, which may be considered a limitation, but is nevertheless important: transplant-specific care is provided by a single institution, so institutional economic analysis is directly relevant to transplant program administrators. Evaluation from broader health system perspectives is the focus of ongoing work by our group. Our recent review of economic evaluations and costing studies revealed that a dearth exists for health economic analysis of lung transplantation from both perspectives.
      • Peel JK
      • Keshavjee S
      • Krahn M
      • Sander B.
      Economic evaluations and costing studies of lung transplantation: A scoping review.
      A limitation in the before-after study design is the risk of selection-maturation bias, where baseline covariates and exposure status are a function of the passage of time. We mitigated the risk of this by conducting an interrupted time-series analysis.
      • Ho AMH
      • Phelan R
      • Mizubuti GB
      • et al.
      Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
      • Mascha EJ
      • Sessler DI.
      Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
      • Linden A
      • Adams JL.
      Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in programme evaluation.
      We additionally accounted for covariate differences using IPTW, though the risk of unmeasured confounding may persist. While the benefits of IPTW outweigh its drawbacks, this technique is sensitive to model misspecification – particularly with manually-specified models, which tend to be too simple; we therefore used a machine-learning tool, generalized boosted models, to estimate propensity scores.
      • McCaffrey DF
      • Griffin BA
      • Almirall D
      • Slaughter ME
      • Ramchand R
      • Burgette LF.
      A tutorial on propensity score estimation for multiple treatments using generalized boosted models.
      • Austin PC
      • Stuart EA.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      • Austin PC.
      The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments.
      • Schulte PJ
      • Mascha EJ.
      Propensity Score Methods: Theory and Practice for Anesthesia Research.

      Olmos A, Govindasamy P. A practical guide for using propensity score weighting in R. Practical Assessment, Research, and Evaluation. 2015;20(13). doi:DOI: https://doi.org/10.7275/jjtm-r398.

      • Austin PC.
      Variance estimation when using inverse probability of treatment weighting (IPTW) with survival analysis.
      • Ali MS
      • Prieto-Alhambra D
      • Lopes LC
      • et al.
      Propensity Score Methods in Health Technology Assessment: Principles, Extended Applications, and Recent Advances.
      • Stuart EA.
      Matching methods for causal inference: A review and a look forward.
      Nevertheless, we did not achieve balance for urgency status; the best overall balance was achieved by the present propensity score selection model which did not include this variable. Our study is limited by this imperfect covariate balance. We considered this acceptable since including the urgency status variable resulted in overall worse covariate balance, and because it is a broadly-defined variable. We gave preference to more specific covariates for our selection model that capture the same fundamental idea as the urgency status variable (pre-transplant life support, transplant from home-versus-hospital).
      • Austin PC
      • Stuart EA.
      Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
      Further, since we anticipate that higher urgency patients would accrue higher healthcare costs and we observed a higher proportion of high urgency patients in the modern era, the lack of balance achieved for this covariate should yield higher costs in the modern era. Our finding that the modern era was not associated with significantly higher costs than the pre-EVLP era is thereby a conservative one. A further limitation was that primary graft dysfunction was not reliably available in our dataset, so we could not explicitly incorporate this important variable into the analysis. While the cost estimates for the transplant phase and ICU costs implicitly incorporate the costs associated with primary graft dysfunction for those patients who experienced this outcome, this limitation of our dataset prevented us from incorporating primary graft dysfunction into the analyses more explicitly. We did, however, incorporate graft dysfunction in the form of CLAD, and by the surrogate variable mechanical ventilation >72hours post-transplant in our analysis of post-transplant costs. Lastly, EVLP costs may change as new devices come to market and EVLP-use increases; generalization of our results to outside jurisdictions with different EVLP devices and protocols should be performed circumspectly.
      We estimated cumulative costs from referral using a novel phase-based costing method which employs an IPTW-weighted multi-state model to more precisely assign costs to their correct clinical phase and account for confounding and selection biases.
      • Gran JM
      • Lie SA
      • Yeflaten I
      • Borgan R
      • Aalen OO
      Causal inference in multi-state models - sickness absence and work for 1145 participants after work rehabilitation.
      ,
      • Hoff R.
      The impact of completing upper secondary education - a multi-state model for work, education and health in young men.
      This method is an improvement over conventional phase-based costs-assessments, which require predefined population-level phase durations, and for which uncertainty and subgroup analyses are not straightforward. An advantage of our approach is that phase-specific costs measured in this way map directly onto the health states of a decision-analytic model. An additional strength is that it yields state-occupancy probabilities, from which we may draw inferences about the transplant program.

      CONCLUSIONS

      This study estimated costs, from the institutional perspective, for the transplant hospitalization as $111,878 ($94,123 - $130,767) without EVLP available versus $122,008 ($91,457 - $163,954) for a non-EVLP transplant in the modern era and $90,923 ($75,788 - $109,428 for a transplant of EVLP-treated lungs in the modern era. Post-transplant mechanical ventilation >72hrs is a significant predictor of costs, and significantly fewer patients required prolonged ventilation after EVLP was available. Despite waitlist enrolment of significantly older and more complex patients, access to lung transplantation was shorter in the EVLP era, with no significantly higher IPTW-adjusted phase-specific or cumulative costs: important clinical improvements were achieved with EVLP at no significant marginal cost.

      AUTHOR CONTRIBUTIONS

      JKP was involved in study design, data analysis, writing of manuscript, and acquisition of funding. EP and DN were involved in study design, and data analysis. SK and BS were involved in study design, data analysis, and acquisition of funding. KB, LDS, MC, and ML were involved in data analysis. All authors participated in revision of the manuscript and gave approval for publication. All listed authors contributed sufficiently to qualify for authorship.

      FINANCIAL DISCLOSURE STATEMENTS

      Dr. Peel has been supported by the Canadian Institutes of Health Research (CIHR) CGS-M award, CIHR Fellowship, Ontario Graduate Scholarship, UofT Graduate Fellowship, Dr. Brian Kavanagh scholarship from UHN, and PSI Foundation Resident Research grant.
      Dr. Keshavjee is a founder of Perfusix and XOR Laboratories, and has received consulting fees from Lung Bioengineering, outside the submitted work.

      ACKNOWLEDGEMENTS

      We would like to thank Dr. Murray Krahn, Dr. Lianne Singer, the Toronto Lung Transplant Program research team, and the team at the Toronto Health Economics and Technology Assessment (THETA) collaborative for their mentorship and contribution to the study conceptualization. Privacy legislation prohibits release of our raw data. Please contact the author for example programming code.

      REFERENCES

        • Yeung JC
        • Keshavjee S.
        Overview of Clinical Lung Transplantation.
        Cold Spring Harbor Perspectives in Medicine. 2014; 4 (-a015628)a015628https://doi.org/10.1101/cshperspect.a015628
        • Valapour M
        • Lehr CJ
        • Skeans MA
        • et al.
        OPTN/SRTR 2016 Annual Data Report: Lung.
        Am J Transplant. 2018; 18 (Suppl): 363-433https://doi.org/10.1111/ajt.14562
      1. Canadian Institute for Health Information /Institut canadien d'information sur la santé.
        Annual Statistics on Organ Replacement in Canada: Dialysis, Transplantation and Donation, 2010 to 2019. November 2020; : 1-6
        • Peel JK
        • Keshavjee S
        • Krahn M
        • Sander B.
        Economic evaluations and costing studies of lung transplantation: A scoping review.
        J Heart Lung Transplant. 2021; 40: 1625-1640https://doi.org/10.1016/j.healun.2021.08.007
        • Van Raemdonck D
        • Neyrinck A
        • Cypel M
        • Keshavjee S.
        Ex-vivo lung perfusion.
        Transpl Int. 2014; 28: 643-656https://doi.org/10.1111/tri.12317
        • Andreasson ASI
        • Dark JH
        • Fisher AJ.
        Ex vivo lung perfusion in clinical lung transplantation–State of the art.
        Eur J Cardiothorac Surg. 2014; 46: 779-788https://doi.org/10.1093/ejcts/ezu228
        • Wallinder A
        • Ricksten SE
        • Silverborn M
        • et al.
        Early results in transplantation of initially rejected donor lungs after ex vivo lung perfusion: a case-control study.
        Eur J Cardiothorac Surg. 2013; 45: 40-45https://doi.org/10.1093/ejcts/ezt250
        • Divithotawela C
        • Cypel M
        • Martinu T
        • et al.
        Long-term Outcomes of Lung Transplant With Ex Vivo Lung Perfusion.
        JAMA Surg. October 2019; : 1-8https://doi.org/10.1001/jamasurg.2019.4079
        • Yeung JC
        • Krueger T
        • Yasufuku K
        • et al.
        Outcomes after transplantation of lungs preserved for more than 12 h: a retrospective study.
        Lancet Respir Med. 2017; 5: 119-124https://doi.org/10.1016/S2213-2600(16)30323-X
        • Boffini M
        • Ricci D
        • Barbero C
        • et al.
        Ex vivo lung perfusion increases the pool of lung grafts: analysis of its potential and real impact on a lung transplant program.
        Transplant Proc. 2013; 45: 2624-2626https://doi.org/10.1016/j.transproceed.2013.08.004
        • Cypel M
        • Yeung JC
        • Liu M
        • et al.
        Normothermic ex vivo lung perfusion in clinical lung transplantation.
        N Engl J Med. 2011; 364: 1431-1440https://doi.org/10.1056/NEJMoa1014597
        • Cypel M
        • Yeung JC
        • Machuca T
        • et al.
        Experience with the first 50 ex vivo lung perfusions in clinical transplantation.
        J Thorac Cardiovasc Surg. 2012; 144: 1200-1206https://doi.org/10.1016/j.jtcvs.2012.08.009
        • Machuca TN
        • Mercier O
        • Collaud S
        • et al.
        Lung Transplantation With Donation After Circulatory Determination of Death Donors and the Impact of Ex VivoLung Perfusion.
        American Journal of Transplantation. 2015; 15: 993-1002https://doi.org/10.1111/ajt.13124
        • Zych B
        • Popov AF
        • Stavri G
        • et al.
        Early outcomes of bilateral sequential single lung transplantation after ex-vivo lung evaluation and reconditioning.
        J Heart Lung Transplant. 2012; 31: 274-281https://doi.org/10.1016/j.healun.2011.10.008
        • Jawitz OK
        • Raman V
        • Becerra D
        • et al.
        Lung Transplantation After Ex Vivo Lung Perfusion.
        Annals of Surgery. 2020; (Publish Ahead of Print): 1-7https://doi.org/10.1097/SLA.0000000000004233
      2. Canadian Agency for Drugs and Technologies in …. Guidelines for the economic evaluation of health technologies: Canada. 2006.

      3. Jacobs P, Yim R.Using Canadian Administrative Databases to Derive Economic Data for Health Technology Assessments. 2009.

        • Goodman C.
        HTA 101 - INTRODUCTION TO HEALTH TECHNOLOGY ASSESSMENT.
        National Library of Medicine US. May 2014; : 1-218
        • Elm von E
        • Altman DG
        • Egger M
        • et al.
        The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies.
        Lancet. 2007; 370: 1453-1457https://doi.org/10.1016/S0140-6736(07)61602-X
        • Benchimol EI
        • Smeeth L
        • Guttmann A
        • et al.
        The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement.
        PLoS Medicine. 2015; 12e1001885https://doi.org/10.1371/journal.pmed.1001885
        • Cypel M
        • Yeung JC
        • Hirayama S
        • et al.
        Technique for Prolonged Normothermic Ex Vivo Lung Perfusion.
        Journal of Heart and Lung Transplantation. 2008; 27: 1319-1325https://doi.org/10.1016/j.healun.2008.09.003
        • Bennell MC
        • Qiu F
        • Micieli A
        • et al.
        Identifying predictors of cumulative healthcare costs in incident atrial fibrillation: a population-based study.
        JAHA. 2015; 4https://doi.org/10.1161/JAHA.114.001684
        • Kim HJ
        • Fay MP
        • Feuer EJ
        • Midthune DN.
        Permutation tests for joinpoint regression with applications to cancer rates.
        Stat Med. 2000; 19: 335-351https://doi.org/10.1002/(sici)1097-0258(20000215)19:3<335::aid-sim336>3.0.co;2-z
        • Wijeysundera HC
        • Wang Tomlinson
        • Krahn Ko D
        Techniques for estimating health care costs with censored data: an overview for the health services researcher.
        CEOR. June 2012; (-11): 145https://doi.org/10.2147/CEOR.S31552
      4. Chan BC-F, Cadarette SM, Wodchis WP, Krahn MD, Mittmann N. The lifetime cost of spinal cord injury in Ontario, Canada: A population-based study from the perspective of the public health care payer. February 2019:1-11. doi:10.1080/10790268.2018.1486622.

        • McCaffrey DF
        • Griffin BA
        • Almirall D
        • Slaughter ME
        • Ramchand R
        • Burgette LF.
        A tutorial on propensity score estimation for multiple treatments using generalized boosted models.
        Stat Med. 2013; 32: 3388-3414https://doi.org/10.1002/sim.5753
        • Austin PC
        • Stuart EA.
        Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies.
        Stat Med. 2015; 34: 3661-3679https://doi.org/10.1002/sim.6607
        • Austin PC.
        The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments.
        Stat Med. 2014; 33: 1242-1258https://doi.org/10.1002/sim.5984
        • Schulte PJ
        • Mascha EJ.
        Propensity Score Methods: Theory and Practice for Anesthesia Research.
        Anesth Analg. 2018; 127: 1074-1084https://doi.org/10.1213/ANE.0000000000002920
      5. Olmos A, Govindasamy P. A practical guide for using propensity score weighting in R. Practical Assessment, Research, and Evaluation. 2015;20(13). doi:DOI: https://doi.org/10.7275/jjtm-r398.

        • Austin PC.
        Variance estimation when using inverse probability of treatment weighting (IPTW) with survival analysis.
        Stat Med. 2016; 35: 5642-5655https://doi.org/10.1002/sim.7084
        • Ali MS
        • Prieto-Alhambra D
        • Lopes LC
        • et al.
        Propensity Score Methods in Health Technology Assessment: Principles, Extended Applications, and Recent Advances.
        Front Pharmacol. 2019; 10: 973https://doi.org/10.3389/fphar.2019.00973
        • Stuart EA.
        Matching methods for causal inference: A review and a look forward.
        Stat Sci. 2010; 25: 1-21https://doi.org/10.1214/09-STS313
        • Zhou T
        • Tong G
        • Li F
        • Thomas LE
        • PSweight Li F.
        An R Package for Propensity Score Weighting Analysis.
        arXiv preprint arXiv:200700974. 2021; VV: 1-35https://doi.org/10.18637/jss.v000.i00
      6. Government of Canada, Statistics Canada. Consumer Price Index, annual average, not seasonally adjusted. 2020. doi:https://doi.org/10.25318/1810000501-eng.

        • Manca A
        • Austin PC.
        Using propensity score methods to analyse individual patient level cost effectiveness data from observational studies.
        yorkacuk. 2008;
        • Bhattacharya K
        • Bentley JP
        • Ramachandran S
        • et al.
        Phase-Specific and Lifetime Costs of Multiple Myeloma Among Older Adults in the US.
        JAMA Netw Open. 2021; 4e2116357https://doi.org/10.1001/jamanetworkopen.2021.16357
        • Gran JM
        • Lie SA
        • Yeflaten I
        • Borgan R
        • Aalen OO
        Causal inference in multi-state models - sickness absence and work for 1145 participants after work rehabilitation.
        BMC Public Health. 2015; (October): 1-16https://doi.org/10.1186/s12889-015-2408-8
        • Hoff R.
        The impact of completing upper secondary education - a multi-state model for work, education and health in young men.
        BMC Public Health. 2018; (April): 1-16https://doi.org/10.1186/s12889-018-5420-y
        • Putter H
        • Fiocco M
        • Geskus RB.
        Tutorial in biostatistics: competing risks and multi-state models.
        Stat Med. 2007; 26: 2389-2430https://doi.org/10.1002/sim.2712
        • de Wreede LC
        • Fiocco M
        • Putter H.
        mstate: an R package for the analysis of competing risks and multi-state models.
        J Stat Soft. 2011; 38: 1-30
        • de Wreede LC
        • Fiocco M
        • Putter H.
        The mstate package for estimation and prediction in non- and semi-parametric multi-state and competing risks models.
        Computer Methods and Programs in Biomedicine. 2010; 99: 261-274https://doi.org/10.1016/j.cmpb.2010.01.001
        • Meira-Machado L
        • de Uña-Alvarez J
        • Cadarso-Suárez C
        • Andersen PK.
        Multi-state models for the analysis of time-to-event data.
        Stat Methods Med Res. 2009; 18: 195-222https://doi.org/10.1177/0962280208092301
        • Putter H.
        Tutorial in biostatistics: Competing risks and multi-state models Analyses using the mstate package.
        cranrapporternet. December 2020;
        • Jackson C.
        Multi-state modelling with R: the msm package. December 2019; : 1-57
        • Ho AMH
        • Phelan R
        • Mizubuti GB
        • et al.
        Bias in Before-After Studies: Narrative Overview for Anesthesiologists.
        Anesth Analg. 2018; 126: 1755-1762https://doi.org/10.1213/ANE.0000000000002705
        • Mascha EJ
        • Sessler DI.
        Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care.
        Anesth Analg. 2019; 129: 618-633https://doi.org/10.1213/ANE.0000000000004153
        • Linden A
        • Adams JL.
        Applying a propensity score-based weighting model to interrupted time series data: improving causal inference in programme evaluation.
        J Eval Clin Pract. 2011; 17: 1231-1238https://doi.org/10.1111/j.1365-2753.2010.01504.x
        • Jones AM
        • Lomas J
        • Rice N.
        Healthcare Cost Regressions: Going Beyond the Mean to Estimate the Full Distribution.
        Health Econ. 2015; 24: 1192-1212https://doi.org/10.1002/hec.3178
        • Jones AM.
        Models for health care.
        in: Hendry D Clements M Oxford Handbook of Economic Forecasting. 2010
      7. R: A language and environment for statistical computing.
        R Core Team, 2020
        • McMeekin N
        • Chrysos AE
        • Vale L
        • Fisher AJ.
        Incorporating ex-vivo lung perfusion into the UK adult lung transplant service: an economic evaluation and decision analytic model.
        BMC Health Serv Res. 2019; 19: 1-14https://doi.org/10.1186/s12913-019-4154-6
        • Fisher A
        • Andreasson A
        • Chrysos A
        • et al.
        An observational study of Donor Ex Vivo Lung Perfusion in UK lung transplantation: DEVELOP-UK.
        Health Technol Assess. 2016; 20: 1-276https://doi.org/10.3310/hta20850
        • Halpern SE
        • Kesseli SJ
        • Au S
        • et al.
        Lung transplantation after ex vivo lung perfusion versus static cold storage: An institutional cost analysis.
        Am J Transplant. 2022; 22: 552-564https://doi.org/10.1111/ajt.16794
        • Maltzahn N
        • Hoff R
        • Aalen OO.
        A hybrid landmark Aalen-Johansen estimator for transition probabilities in partially non-Markov multi-state models.
        arXiv preprint arXiv:200700974. August 2020;