Effect of a Novel Multicomponent Intervention to Improve Patient Access to Kidney Transplant and Living Kidney Donation

Importance Patients with advanced chronic kidney disease (CKD) have the best chance for a longer and healthier life if they receive a kidney transplant. However, many barriers prevent patients from receiving a transplant. Objectives To evaluate the effect of a multicomponent intervention designed to target several barriers that prevent eligible patients from completing key steps toward receiving a kidney transplant. Design, Setting, and Participants This pragmatic, 2-arm, parallel-group, open-label, registry-based, superiority, cluster randomized clinical trial included all 26 CKD programs in Ontario, Canada, from November 1, 2017, to December 31, 2021. These programs provide care for patients with advanced CKD (patients approaching the need for dialysis or receiving maintenance dialysis). Interventions Using stratified, covariate-constrained randomization, allocation of the CKD programs at a 1:1 ratio was used to compare the multicomponent intervention vs usual care for 4.2 years. The intervention had 4 main components, (1) administrative support to establish local quality improvement teams; (2) transplant educational resources; (3) an initiative for transplant recipients and living donors to share stories and experiences; and (4) program-level performance reports and oversight by administrative leaders. Main Outcomes and Measures The primary outcome was the rate of steps completed toward receiving a kidney transplant. Each patient could complete up to 4 steps: step 1, referred to a transplant center for evaluation; step 2, had a potential living donor contact a transplant center for evaluation; step 3, added to the deceased donor waitlist; and step 4, received a transplant from a living or deceased donor. Results The 26 CKD programs (13 intervention, 13 usual care) during the trial period included 20 375 potentially transplant-eligible patients with advanced CKD (intervention group [n = 9780 patients], usual-care group [n = 10 595 patients]). Despite evidence of intervention uptake, the step completion rate did not significantly differ between the intervention vs usual-care groups: 5334 vs 5638 steps; 24.8 vs 24.1 steps per 100 patient-years; adjusted hazard ratio, 1.00 (95% CI, 0.87-1.15). Conclusions and Relevance This novel multicomponent intervention did not significantly increase the rate of completed steps toward receiving a kidney transplant. Improving access to transplantation remains a global priority that requires substantial effort. Trial Registration ClinicalTrials.gov Identifier: NCT03329521

eTable 17.Effect of the intervention when restricted to patients who entered the trial approaching the need for dialysis eTable 18.Effect of the intervention when restricted to patients who were receiving maintenance dialysis when they entered the trial eTable 19.Rates of kidney transplantation (total, living, deceased) in intervention and usualcare groups in pre-trial, trial, and pre-COVID-19 pandemic periods eFigure 1. Multistate models for the intervention and usual-care groups eFigure 2. Forest plot of the effect of the intervention on the primary composite outcome in multiple subgroups This supplemental material has been provided by the authors to give readers additional information about their work.
Published Study Process Evaluation Protocol: Background (The Need for a Process Evaluation of the EnAKT LKD Trial), Methods (Participants) (Data Collection) Protecting vulnerable participants Clusters may contain vulnerable participants.In these circumstances, researchers and research ethics committees must consider whether additional protections are needed.
The Research Ethics Board considered vulnerable patients with advanced chronic kidney disease or patients receiving maintenance dialysis in this cluster trial.The Research Ethics Board agreed that the trial met the criteria for a waiver of patient consent for participation.
Manuscript, section: Methods (Study Design) Published Study Protocol: Methods (Ethical Considerations), Supplemental Appendix 13 When individual informed consent is required and there are individuals who may be less able to choose participation freely because of their position in a cluster or organizational hierarchy, research ethics committees should pay special attention to recruitment, privacy, and consent procedures for those participants.eTable 2. CONSORT statement for cluster-randomized trials, pragmatic trials and trials using routinely collected data a,b,c,d

Participant flow
For each group, the numbers of participants who were randomly assigned, received intended treatment, and were analysed for the primary outcome The number of participants or units approached to take part in the trial, the number which were eligible, and reasons for nonparticipation should be reported

No
All individual participants (patients) approaching the need for dialysis or receiving maintenance dialysis at one of the CKD programs in the intervention group would have participated in the trial (i.e., intervention was designed to impact entire CKD programs).Therefore, the selection of individual participants would not be affected by knowledge of the intervention assigned to the cluster.1b.3 Were there baseline imbalances that suggest differential identification or recruitment of individual participants between intervention groups?

No
Covariate-constrained randomization was used to help ensure key baseline characteristics were balanced between intervention groups.

Risk of bias judgement Low Bias due to deviations from intended interventions
2.1a Were participants aware that they were in a trial?

Partial Yes
This trial was delivered as part of the Ontario Renal Network's (part of Ontario Health) provincial quality improvement strategy to improve access to kidney transplant.Individual patient consent is not required to execute these strategies.Since our unit of randomization was at the cluster and the intervention was at the cluster level it made it impractical to let each participant know about the trial given much of the intervention involved components that may not directly affect them (e.g., administrative support provided to the program and quality improvement teams reviewing transplant metrics).The participants could also opt out of many of the intervention components.For example, patients did not have to participate in the Transplant Ambassador Program if it was not of interest to them and they could decide not to receive education on transplantation.
Patients could not opt out of trial data collection with the de-identified baseline and outcome data coming from administrative healthcare databases at ICES, which allows for the collection of personal health information without consent for the purpose of analysis or compiling of statical information with respect to the management of, evaluation or monitoring of, the allocation of resources to or planning for all or part of the health system.
The Ontario Renal Network will share the results with all the 26 Chronic Kidney Disease programs in Ontario who can then share this information with staff and patients.
Since the intervention was at the cluster-level all patients within the CKD programs would have received the intervention and all staff (e.g., clinicians and healthcare professionals) within the CKD programs were aware of the intervention.2.3 If Y/PY/NI to 2.1b or 2.2: Were there deviations from the intended intervention that arose because of the trial context?

Yes
The onset of the COVID-19 pandemic, 2.4 years into the 4.2-year trial period, substantially impacted intervention delivery for at least a year.Transplant activity ceased temporarily, local quality improvement teams met less often, there was a pause on the provincial rounds, healthcare staff retired or were re-deployed, and transplant ambassadors transitioned from in-person to virtual meetings.
Effects of the pandemic on intervention delivery will be explored in the process evaluation.
In summary, the trial context did not result in deviations from the intended intervention but rather the delivery of the intervention was affected by extenuating circumstances.unblinded outcome data were available for analysis?
Any changes were publicly documented in the table of protocol updates located on clincialtrialsl.govand these changed occurred before unblinded outcome data was available.5.2 ... multiple eligible outcome measurements (e.g.scales, definitions, time points) within the outcome domain?

No
Our outcomes were pre-specified in our published trial protocol and statistical analysis plan.They were captured using administrative health databases.When outcomes could be assessed at multiple time points we pre-specified how the outcome would be captured.Our primary and secondary outcomes were routinely collected in our administrative healthcare databases.5.3 ... multiple eligible analyses of the data?No The published statistical analysis plan and trial protocol pre-specified multiple analyses of the data.These were all reported and for any outcomes that were not reported a rationale was provided.There were 4 steps in this trial.
Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.
Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.
Transplant center personnel are requested to accurately input this data into the TGLN database.TGLN performs data validation on some steps; if data discrepancies are identified, they are resolved by asking staff at the transplant centers to verify or correct the discrepancies.Despite these efforts, data errors may still occur.
We reviewed, quantified, and reconciled data discrepancies in the TGLN dataset before performing any outcome analyses.We explored the completeness of the TGLN data by looking at the total number of patients who completed each step by year and by the transplant center, and we checked whether all expected transplant steps were present and in the correct order.For example, all patients who received a deceased kidney transplant were referred to a transplant center and then placed on the deceased donor transplant waitlist.Similarly, before receiving a living donor transplant, patients need to have a referral and have a potential living donor contact a transplant center for evaluation (where the order of the two steps does not matter).Furthermore, activities for a current transplant had to occur after a prior kidney transplant, should one exist.
We also needed to ensure that patients would start in the correct state shown visually in eFigure 1a and 1b (patients were allowed to complete steps 1 to 3 before trial entry).Finally, we also needed to ensure we captured new steps after each patient entered the trial.
We used the following process to assess and correct suspected data entry errors using supplementary data sources, and to impute missing steps if there was sufficient evidence that the patient completed the step (for example if a patient received a kidney transplant but was missing a record of being referred to a transplant center).As described below, we imputed missing steps in the TGLN dataset before linking it to our trial cohort.
1. We linked the TGLN data to other data sources at ICES to identify any missing historic transplant activity.For example, we examined TGLN transplant data linked to the Ontario Health Insurance Plan (OHIP) database to identify patients with a prior kidney transplant and to differentiate new transplantrelated activities from historical activities in patients with a prior failed kidney transplant.
2. If information on when a potential living donor began their evaluation at a transplant center was missing, we used an additional TGLN date field based on the date the donor first contacted a transplant center (e.g., email or phone call to a transplant center).
3. The Canadian Organ Replacement Registry (CORR) database contains waitlist information for individuals who went on to receive a transplant.If the TGLN database did not contain a waitlist date for a patient who received a kidney transplant from a deceased donor, we obtained this information from CORR, if it was available.
4. Some patients with a record of receiving a kidney transplant had missing steps leading up to the transplant that we could not supplement using other data sources.In these cases, we used median imputation conditional on the 6 transplant centers, transplant type (deceased vs. living), and transplant history (i.e., whether there was evidence of at least one prior transplant).We did this to account for potential differences across centers, transplant types, and history.For example, we used median imputation for patients with a record of waitlisting who were missing a transplant center referral date, which was conditional on transplant history and the transplant center where the patient was listed.We calculated the median days between dates (i.e., time from referral to transplant, referral to a waitlist, waitlist to transplant, contact to transplant) for records where both date variables were available.
When two or more dates existed, we calculated the median proportion of time between the three dates to ensure that the order was appropriate with an expected alignment.
Imputation was done for all patients in the trial, using the same approach without considering whether they were in the intervention or usual-care group.Once the data were imputed, we proceeded with the trial outcome analyses.
Concerning the number of data errors and the amount of imputation performed for the 20 375 patients accrued into the trial:  33 patients (0.16%) had evidence of a waitlist date before their referral date;  275 patients (1.35%) had a missing referral date (i.e., the patient had evidence of a waitlist, deceased or living kidney transplant with no evidence of a referral date);  92 patients (0.45%) had a missing waitlist date (i.e., the patient had evidence of a deceased kidney donor transplant with no evidence of a waitlist date); and  68 patients (0.33%) had a missing date for when a living kidney donor began their evaluation (i.e., the patient had evidence of a living kidney donor transplant but no date for the start of the living kidney donor evaluation).
In summary, for outcome data, 2.1% of patients had evidence of at least one missing step in error, which we then imputed.
eTable 7. The intracluster correlation coefficient (ICC) a and coefficient of variation for primary and secondary outcomes Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.b Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.The methodology for deriving the ICC from multistate models is limited.For example, from the literature for survival anlaysis, Kalia et al. (2016) show negative biases when using the binary censoring indicator or the continuous event times to calculate the ICC.Nevertheless, we decided to use the binary censoring indicator as recommended by Campbell and Walters (2014) as we found more than minimal censoring (>5%).Furthermore, when there were multiple states, we created a binary indicator based on at least one event.Finally, we calculated ICCs from a Restricted Maximum Likelihood (REML) from a mixed model to naturally account for the variation in cluster sizes.

Variable
b Only the first potential donor was counted for a patient when there were multiple potential donors.
c As there was no intervention effect, we provided a single estimate for ICC and CV across the entire trial group.
Had we observed an intervention effect, we would have reported the ICC separately by the intervention and usualcare groups.
d As with the primary adjusted analysis, the following baseline characteristics were included when estimating the adjusted ICC: age, sex, Charlson Comorbidity Index, ≥1 intensive care unit admissions in the prior year, the frequency of hospital admissions in the prior year, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, home dialysis, or approaching the need for dialysis).
We included the historic transplant center referral pattern as an additional adjustment rather than inclusion as a stratification factor.This measure is provided to demonstrate that some of the variation can be explained by the characteristics of patients within clusters.For planning future trials, we recommend using the more conservative estimates.
e The CV is calculated based on Hayes and Moulton (2017).This method uses the cluster level rates (events per patient-years), where the events in the numerator are not limited to 1 per patient.
f Excludes patients who completed step 2 before entering the trial.
g Excludes patients who completed steps 1 or 2 before entering the trial.
h The follow-up time was censored if and when a patient started dialysis.
i Excludes patients who were receiving maintenance dialysis when they entered the trial.

References:
Campbell The semiparametric multistate model for the primary outcome was assessed using a constrained intervention effect with cluster-level bootstrap standard errors (to account for clustering on outcomes within chronic kidney disease [CKD] programs) and with a t-distribution to adjust the degrees of freedom as a small sample correction when calculating the margin of error in the 95% CI because there were less than 40 clusters in the trial.
The resulting estimand is an individual-averaged cause-specific hazard ratio for the total effect of the intervention on the key steps towards transplantation.
We evaluated the model assumptions in the multistate model.Decisions on how to manage any violation in assumptions were made prior to knowledge of the results and considered (1) impact to the statistical interpretation of the intervention effect, (2) overall model fit (i.e., model improvement), (3) rules of parsimony between two statisticians (Dr.Dixon and Dr. Luo), and (4) clinical interpretation through discussions with the clinical lead (Dr.Garg).
The following are details on the model assumptions.
1. Markov: The Markov assumption in multistate models assumes that transition intensities only depend on the information in the current state at a particular time.To test this assumption, we incorporated event history variables in the model in a few ways (i.e., time spent since completing a prior step at time of state entry, or time to complete a prior step in history; each was also further assessed as time during the trial, prior to trial start, and overall).Regardless of the way in which the history was defined, there was an indication that the Markov assumption was violated.As such, our final selected models included the historic variables in the model.It is worth noting that when specifying the history in more detail, some of the historical variables were not significant.Although the model showed improvements in overall fit with a more detailed information of history, the intervention effect remained unchanged regardless of how the event history was defined.We chose to keep the simplest summary of history (adding only 3 additional terms to the model based on time spent since completing the prior 3 steps at the time of state entry) because there was no impact to the main findings and due to rules of parsimony.

Linearity:
We had 4 continuous variables in the model: age, the historic rate of kidney transplant, the Charlson Comorbidity Index, and the frequency of hospital admissions in the prior year.We tested the assumption of linearity for each variable using restricted cubic splines and Bsplines.We assessed the impact to the model (i.e., change in intervention effect, significance and overall model fit).We found violations in linearity for age and historic rate of kidney transplant.However, we chose to not include the splines in our final model because the interpretation of the intervention effect remained unchanged, and because of rules of parsimony.

Proportional Hazards:
Our multistate model used a semiparametric framework where no assumptions are made about the baseline hazard (i.e., transition intensity).This approach assumes proportional hazards (PH) in our model variables.We tested the PH assumption in the variables included in our model using the Schoenfeld residuals [Reference: Fox and Weisberg (2023) a ].The PH assumption was satisfied for the intervention effect in models for the primary and secondary outcomes.In the primary model, we did find PH violations for age (p-value < 0.001), baseline cluster historic rate of kidney transplant (p-value < 0.001), CKD treatment modality at time of trial entry (p-value 0.002), Charlson Comorbidity Index (p-value < 0.001), and whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs) (p-value 0.034).We did not include time-interactions to the adjusted factors that violated the proportional hazard assumption because our estimated intervention effect was robust regardless of changes to the model.

Constrained intervention effect:
Our primary model reported the constrained intervention effect.We assumed an overall effect of the intervention across all transitions between states (rather than the specific effects from individual transitions between states).We present results of the unconstrained model in eTable 11.The unconstrained model has better model fit than the constrained model.

Independence:
There is a natural correlation of outcomes within CKD programs induced by the clustered design which violates the independence assumption.To accommodate, variance inflation techniques are required to avoid spurious statistical significance.The inferences in the paper (i.e., p-values and confidence intervals) were calculated using a cluster-level bootstrap standard error and a tdistribution to adjust the degrees of freedom from the small number of clusters included in the trial (note: this approach has not been formally tested and may deviate from the nominal levels; i.e., wider or narrower intervals than expected).The bootstrap approach inflates the variance beyond a model that assumes independence or a model that applies a robust standard error without a small sample correction.
6. Non-informative censoring: There were competing events in our follow-up period that could impact the probability of observing our events of interest (i.e., steps toward transplantation).The main competing events were death and recorded contradiction to transplant, neither of which differed between the arms of the trial.We choose to censor at these events, and therefore, the model produced an estimand of the individualaveraged cause-specific hazard ratio for the total effect of the intervention on the key steps towards transplantation.We acknowledge that the estimate could be mediated by the competing events and that the total intervention effect does not provide information about whether the intervention effect on key steps is partially driven by the intervention effect on the competing risks.However, we explored the impact of the intervention on the competing risks and did not find any associations.The primary analysis used an intention-to-treat approach.All outcomes were attributed to the program where a patient entered the trial, regardless of whether they transferred to another program in follow-up.As the primary outcome was ascertained using provincial healthcare databases, the only reason for lost follow-up was emigration from Ontario.
* We estimated the intervention on each of the competing risks and did not find an association in either the unadjusted or adjusted analyses.The adjusted hazard ratios (95% confidence intervals) are as follows: emigration: 1.00 (0.81, 1.25); recovered kidney function: 0.89 (0.57,1.39); developed contraindication to receiving a transplant: 1.00 (0.87, 1.14); death: 0.96 (0.87, 1.07).We also looked at a composite outcome of all competing events other than death: 0.99 (0.89, 1.10).The proportional hazard assumption was not violated for the intervention effect in any of these analyses.a Developed a contraindication to receiving a transplant that was recorded in our data sources; for example, a diagnosis of dementia, use home oxygen [a sign of serious lung disease], transfer to a long-term care home, or developed a comorbidity likely to preclude transplantation.
b The states listed here are shown visually in eFigure 1a and 1b.The average time is the mean patient time spent in a certain state until transitioning into another state or being censored during the trial period.For example, patients receiving the intervention have an average time spent in the 'no steps' state of 1.8 years.We did not provide the mean sojourn time as it is an estimate relying on strict parametric assumptions, and our main model used a semi-parametric framework.
© a For the hazard ratio, the referent group is usual care.We used a stratified, constrained, multistate model accounting for the order in which steps were completed and the clustered design.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error in the 95% CI because our trial included <40 clusters.
b The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, ≥1 intensive care unit admissions in the prior year, the frequency of hospital admissions in the prior year, the historic rate of transplant, whether the transplant center was co-located with a CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, home dialysis, or approaching the need for dialysis).
c Besides the baseline characteristics used for adjustment listed in the primary analysis b , two additional baseline characteristics were added based on their distribution in Table 1 in the main paper: one-way distance from residence to a transplant center (as a continuous variable), and higher residential instability (as a binary variable).
eTable 11.Effect of the intervention on each transition (shown visually in eFigure 1a and 1b); these estimates are from an unconstrained model Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.
Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.
Patients available for this transition, No.
Transitions completed during the trial, No. a Only the first potential donor was counted for a patient when there were multiple potential donors.
b For the adjusted hazard ratio, the referent group is usual care.We used a stratified, multistate model accounting for the order in which steps were completed and the clustered design.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error in the 95% CI because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, ≥1 intensive care unit admissions in the prior year, the frequency of hospital admissions in the prior year, the historic rate of transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, home dialysis, or approaching the need for dialysis).
eTable 12.Effect of the intervention on steps not specified as primary or secondary outcomes Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.a Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.a Only the first potential donor was counted for a patient when there were multiple potential donors.

Outcome
b Excludes patients who completed step 1 before entering the trial.
c Excludes patients who completed step 3 before entering the trial.
d For the adjusted hazard ratio, the referent group is usual care.Outcomes in this trial were analyzed at the patient level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The multistate model for the outcomes in this table reduced to a classic Cox proportional hazards model (i.e., a single endpoint).The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error in the 95% CI because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, other forms of dialysis, or approaching the need for dialysis).
eTable 13.Effect of the intervention when follow-up was truncated to March 16, 2020, the start of the COVID-19 pandemic in Ontario.
Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.a Step 3: Added to deceased donor waitlist.

Patients
a Only the first potential donor was counted for a patient when there were multiple potential donors.
b For the adjusted hazard ratio, the referent group is usual care.The primary outcome was analyzed at the patient-level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.
To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error in the 95% CI because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, home dialysis, or approaching the need for dialysis).The secondary outcomes were analyzed using the same multistate model framework and evaluation process as for the primary outcome.For outcomes with a single endpoint, the model reduced to a classic Cox proportional hazards model.c The multistate model for this outcome reduced to a classic Cox proportional hazards model.
d Excludes patients who completed step 2 before entering the trial.
e Excludes patients who completed steps 1 or 2 before entering the trial.
f The follow-up time was censored if and when a patient started dialysis.
g Excludes patients who were receiving maintenance dialysis when they entered the trial.
© All subgroups were pre-specified unless indicated otherwise.
The widths of the confidence intervals (CI) for these outcomes were not adjusted for multiplicity, so the CIs should not be used to infer definitive treatment effects for these outcomes.
a For the adjusted hazard ratio, the referent group is usual care.Outcomes in this trial were analyzed at the patient-level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.
To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error in the 95% CI because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, other forms of dialysis, or approaching the need for dialysis.b Income quintile: measured by neighborhood-level median income c Residential instability refers to the area-level concentration of people with high housing instability or family instability d Material deprivation refers to individuals and communities not being able to access and realize basic material needs.
e Ethnic diversity refers to area-level concentrations of residents who are recent immigrants and/or those who self-identify as a visible minority.
f Dependency refers to the area-level concentrations of people who do not have employment income, including seniors and children.The following measures were listed in the protocol but were not analyzed because either the data were of insufficient quality or there was concern that the measure would not provide meaningful information: (1) Time from nephrologist consultation to donor nephrectomy.
(2) Proportion of potential living kidney donors who began their evaluation who completed all the following: a nephrology consultation, a surgeon consultation, and an abdominal computed tomography angiogram.( 3) Time from when the potential kidney donor began their evaluation to abdominal CT angiogram, restricted to those who had a donor nephrectomy during the trial period.(4) Time from referral to waitlisting for a deceased donor transplant for patients who received a living kidney donor transplant during the trial period.(5) Time from consulting with a transplant nephrologist (after referral to a transplant center) to waitlisting (this measure could also be restricted to patients who received a living kidney donor transplant during the trial period).( 6) Time from referral to a transplant center to consulting with a transplant nephrologist (this measure could also be restricted to patients who received a living kidney donor transplant during the trial period).( 7) Proportion of referrals to a transplant center that were incomplete.(8) Rate of living kidney donor transplants assessed in patients waitlisted for a deceased donor transplant.The published trial protocol indicated some of measures in this table would be reported during the trial period, while others would be assessed as a change from historical norms pre-dating the trial.For reasons of time, cost, and feasibility we focused on reporting these measures by group only during the trial period.
b Assessed in patients who entered the trial receiving maintenance dialysis after November 1, 2017 or those who entered the trial approaching the need for dialysis and started maintenance dialysis during the trial period.In both cases we only considered patients who received a referral after they began maintenance dialysis.c Assessed in patients who had a transplant referral and living kidney donor transplant during the trial period.
d Assessed in patients who had a transplant referral and a living or deceased donor transplant during the trial period.
e Assessed in patients who had a potential living kidney donor begin their evaluation and who received a living kidney donor transplant during the trial period.
f Assessed in patients who had a potential living kidney donor who began their evaluation, and the donor had an abdominal CT angiogram during the trial period.There are 88.5% (1913/2162) of recipients who had a donor who began their evaluation and at least one of the donors had a valid Ontario health card number (i.e., this analysis excludes potential out of province donors and donors with invalid health card numbers since they could not be linked to our other data sources).This corresponds to 88.4% (956/1081) and 88.5% (957/1081) in the intervention and control groups respectively.Only 53% overall had evidence of a CT angiograph during follow up (1014/1913), corresponding to 54.3% (519/956) and 51.7% (495/957) in the intervention and control groups respectively.g Assessed in patients who had a living donor transplant, a referral and a potential living kidney donor begin their evaluation during the trial period.Potential living kidney donors who began their evaluation before the referral were given a value of 0 months.h Assessed in patients who had a transplant referral and were waitlisted for a deceased donor transplant during the trial period.
i Referral declined indicates a patient no longer moves forward with the kidney transplant assessment process (i.e., will not be scheduled to meet with a transplant specialist).
Reasons for a declined status, include 1) patient did not meet referral criteria; 2) patient referred to another program; 3) patient was too sick to receive a transplant; or 4) patient declined to consult with a transplant specialist.In accordance with ICES policy of suppressing cell sizes <6, numbers are presented as ranges.j Referral accepted indicates the patient was approved to proceed with their kidney transplant assessment.
k Referral deferred indicates that a consultation with a transplant specialist was not made at the time but could be made in the future (within 12 months).
l Assessed in patients who were referred to a transplant center during the trial period and had at least a year of follow-up when no transplant or no waitlist was observed within a year of the referral date during the trial.m Assessed in those patients with advanced chronic kidney disease who were approaching the need for dialysis when they entered the trial and were not on dialysis when they received a living kidney donor transplant during the trial period.
eTable 16.Effect of the intervention when restricted to patients who completed no steps toward receiving a transplant before trial entry Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.a Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.a Only the first potential donor was counted for a patient when there were multiple potential donors.

Patients
b For the adjusted hazard ratio, the referent group was usual care.The primary outcome was analyzed at the patient-level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.
To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the CKD treatment modality at the time of trial entry (i.e., in-center hemodialysis, home dialysis, or approaching the need for dialysis).The secondary outcomes were analyzed using the same multistate model framework and evaluation process as for the primary outcome.eTable 17.Effect of the intervention when restricted to patients who entered the trial approaching the need for dialysis Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.a Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.a Only the first potential donor was counted for a patient when there were multiple potential donors.

Patients
b For the adjusted hazard ratio, the referent group was usual care.The primary outcome was analyzed at the patient-level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.
To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, and whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs).The secondary outcomes were analyzed using the same multistate model framework and evaluation process as for the primary outcome.For outcomes with a single transition, the model reduced to a classic Cox proportional hazards model.c The multistate model for this outcome reduced to a classic Cox proportional hazards model.
d Excludes patients who completed step 2 before entering the trial.
e Excludes patients who completed steps 1 or 2 before entering the trial.
f The follow-up time was censored if and when a patient started dialysis.
eTable 18.Effect of the intervention when restricted to patients who were receiving maintenance dialysis when they entered the trial Step 1: Referred to a transplant center for evaluation.
Step 2: Had a potential living donor contact a transplant center for evaluation.a Step 3: Added to deceased donor waitlist.
Step 4: Received a transplant from a living or deceased donor.a Only the first potential donor was counted for a patient when there were multiple potential donors.

Patients
b For the adjusted hazard ratio, the referent group was usual care.The primary outcome was analyzed at the patient-level using a stratified, constrained, multistate model accounting for the order in which steps were completed, the clustered design, and the covariates used in the randomization.The historic transplant center referral pattern was a stratification factor in both the randomization and our final model.We also stratified on the different transitions between steps to allow for separate baseline hazard functions.
To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction because our trial included <40 clusters.The following baseline characteristics were included in the model: age, sex, Charlson Comorbidity Index, the historic rate of kidney transplant, whether a transplant center was co-located with the CKD program (present in 6 of the 26 CKD programs), and the type of dialysis at the time of trial entry (in-center hemodialysis or home dialysis).The secondary outcomes were analyzed using the same multistate model framework and evaluation process as for the primary outcome.For outcomes with a single endpoint, the model reduced to a classic Cox proportional hazards model..5 (3.9 -5.1) 3.9 (3.0 -4.9) a The rate was calculated as the number of transplants over the total patient-years.To maintain valid inferences, we used cluster-level bootstrapping to obtain standard errors (accounting for the correlated outcomes within CKD programs) and a t-distribution as a small-sample correction when calculating the margin of error for the 95% CI because our trial included <40 clusters.b Historic transplant rates (pre-trial period) were calculated in a similar fashion as done during the trial period.Patients with advanced CKD (patients approaching the need for dialysis or receiving maintenance dialysis) were accrued on November 1, 2013 or during the period ending September 30, 2017 with a maximum follow up of October 31, 2017.Numbers along the arrows are the number of patients who transitioned from one state to another (and in the brackets is the unadjusted incidence rate per 100 patient-years).
For example, in the diagram for the usual-care group, the two arrows leaving the Referral box in Figure 1b (the usualcare group) show that during the trial (i) 734 patients with referrals transitioned to having a referral and being waitlisted at a rate of 14.9 events per 100 patient-years and (ii) 647 patients with referrals transitioned to having a referral and having a potential living donor start the evaluation process at a rate of 13.1 events per 100 patient-years.
The numbers within each state are: At index: Number of patients who were in the state at trial entry.Entered: Number of patients who entered the state during the trial.Left: Number of patients who left the state during the trial.
The total number of patients who occupied the state during the trial is the sum of 'at index' plus 'entered'.
The total number of patients in a state at the end of the trial is the sum of 'at index' plus 'entered' minus 'left'.
AX et al.JAMA Internal Medicine.
For outcomes with a single endpoint, the model reduced to a classic Cox proportional hazards model.c The multistate model for this outcome reduced to a classic Cox proportional hazards model.d The follow-up time was censored if and when a patient started dialysis.e Excludes patients who were receiving maintenance dialysis when they entered the trial.© 2023 Garg AX et al.JAMA Internal Medicine.

eFigure 1a and 1b .
Multistate model for the intervention and usual-care groups, respectively The arrows indicate transitions from one state to another.Possible states are: having no steps completed, a single step completed, or multiple steps completed.The different colored arrows indicate the steps completed in the transition:  An orange arrow indicates a patient was referred to a transplant center for evaluation. A blue arrow indicates a patient had a potential living kidney donor contact a transplant center for evaluation. A purple arrow indicates a patient was added to the deceased donor waitlist. A green arrow indicates a patient received a kidney transplant from a living or deceased donor.

eTable 3. Risk of bias a
CONSERVE-CONSORT (CONSORT Extension for Randomized Controlled Trials Revised in Extenuating Circumstance) Outcome ascertainment, data cleaning, and handling of missing data for the steps completed toward receiving a kidney transplant Information on steps completed toward receiving a kidney transplant by patients with advanced chronic kidney disease (CKD) in Ontario, Canada, was primarily obtained from the Trillium Gift of Life Network (TGLN) database housed at ICES.
a Sterne JAC, Savović J, Page MJ, et al.RoB 2: a revised tool for assessing risk of bias in randomised trials.BMJ.2019;366.©2023Garg AX et al.JAMA Internal Medicine.eTable4."mitigating strategy" and describe the changes in the trial manuscript or supplement.Check "no change" for items that are unaffected in the extenuating circumstance.Height, weight, and body mass index were missing for 1857 (9.1%) (intervention 1003 [10.3%], usual care 854 [8.1%]).No imputation was done, and only complete data are reported.All other variables were complete.© 2023 Garg AX et al.JAMA Internal Medicine.eTable 6.
© 2023 Garg AX et al.JAMA Internal Medicine.eTable 8. Multistate model assumptions Primary outcome: Primary outcome examined in two multistate models: the first unadjusted for baseline characteristics and the second adjusted for additional baseline characteristics 2023 Garg AX et al.JAMA Internal Medicine.eTable 10.
Effect of the intervention on the primary composite outcome in multiple subgroups.The adjusted hazard ratio results are also presented graphically in eFigure 2.
The time to complete steps and other measures in the intervention group and usual-care group.a Each result restricted to only those patients who completed all components for the measure during the trial period.
g Post hoc subgroups © 2023 Garg AX et al.JAMA Internal Medicine.eTable 15. a cThe multistate model for this outcome reduced to a classic Cox proportional hazards model.Excludes patients who completed steps 1 or 2 before entering the trial.Rates of kidney transplantation (total, living, deceased) in intervention and usual-care groups in pre-trial, trial, and pre-COVID-19 pandemic periods.
d Excludes patients who completed step 2 before entering the trial.e © 2023 Garg AX et al.JAMA Internal Medicine.eTable 19.