Living Kidney Donors

Living Kidney Donors#

+ Expand

Personalized Risk Estimates

The distinction between static and dynamic data presentation in scientific literature is stark: nearly all peer-reviewed clinical and public health research findings rely on static visualizations of pre-determined cohorts. If personalized medicine is to fulfill its promise, dynamic visualization—such as the interactive model above—must become the standard for conveying information to patients, care providers, and public health practitioners.

“Give him a mask, and he will tell you the truth.”
—Oscar Wilde, The Truth of Masks

Introduction
In my April 2nd talk, I framed the discussion around five nested layers—tactical, informational, strategic, operational, and existential—to situate my thesis within a broader clinical, ethical, and epistemological context, for the benefit of those who are new to the forum.

At the tactical level, I concentrated on living kidney donation as a superior therapeutic option compared to dialysis and deceased donation for patients with end-stage renal disease (ESRD). Unlike the 5–15 year wait often associated with deceased donor transplants—or the indefinite burden of dialysis—living donation can frequently be achieved within a year of referral, reducing both morbidity and mortality. It is a form of intervention that is at once intimate, invasive, and logistically streamlined.

From there, I moved to the informational layer: the imperative of truly informed consent—grounded not in generic population-level probabilities, but in personalized risk derived from multivariable modeling. With beta coefficient vectors and variance-covariance matrices in hand, we can project individualized survival curves that respect the donor’s unique profile. This redefines consent: no longer a ritualistic disclosure of average outcomes, but an ethically precise dialogue, situating the donor not merely as a generous volunteer but as a fully aware agent within a biologically entangled, morally charged, and socially situated decision. Though referred on behalf of the patient, the donor arrives bearing their own lifeworlds, expectations, fears, and futures. In this frame, informed consent is not a checkbox—it is a philosophical commitment to epistemic clarity and emotional resonance, calibrated by statistical truth yet sensitive to narrative texture.

The demographic shift toward older kidney donors necessitates improved characterization of risks to enhance consent processes and outcomes.
— Al Ammary, Muzaale, et al.

Next, at the strategic layer, I addressed the challenge of characterizing long-term donor risk—especially relative to the elusive, counterfactual scenario of what would have happened had the donor never donated. This comparative risk framework is not just a statistical exercise; it shapes how we counsel, regulate, and frame the ethics of living donation. Risk is never absolute—it is narrated, estimated, and situated. Thus, our strategic models must be transparent about uncertainty while grounded in rigor.

Fig. 1 An 84-year-old man became one of the oldest living kidney donors in the United States in May 2019. The consent process for such advanced-age donors often relies on data extrapolated from significantly younger cohorts, typically with a median age of around 40 years. As the living donor population ages, this extrapolative gap exposes both epistemic and clinical blind spots—underscoring the need to better characterize existing datasets and surface where data is missing or underpowered. One such blind spot is frailty, a critical determinant of perioperative risk, yet a variable absent from most donor registries. Its omission complicates both the consent process and postoperative care planning for older candidates. Photo and case source: Houston Methodist#

I then gestured toward the operational demands of long-term donor follow-up. This is the unglamorous but indispensable work of systems maintenance. Tracking donor health over years—decades, ideally—requires infrastructure, coordination, funding, and the sustained commitment of institutions whose priorities often shift. The federally mandated 2-year follow-up is a start, but it falls drastically short of what is needed to capture delayed harms such as nephrectomy-attributable ESRD. In truth, no existing cohort or database offers the scale, fidelity, and longitudinal reach required to draw confident conclusions. Yet without this scaffolding of care, we cannot promise donors that their sacrifice is met with institutional memory, let alone meaningful stewardship.

Finally, at the existential level, I returned to the hard outcomes that make this field ethically weighty: perioperative mortality, long-term all-cause mortality, and the possibility—however rare—of ESRD in the donor themselves. They represent the heavy price that a few may pay so that others may live. To engage with these truths is not to undermine donation, but to dignify it—by refusing euphemism, honoring risk, and insisting on reverence over rhetoric.

Framing the Challenges for the Doctoral Seminar
My 30-minute presentation focused mostly on the challenges inherent in designing, executing, and interpreting perioperative and long-term risks following living kidney donation—particularly for older donors—and integrating that research into meaningful tools for informed consent and risk communication. I presented five interlocking layers of challenge: data access, infrastructure, automation, epistemic missingness, and demographic shifts in donor profiles.

1. Challenge of Data Access#

The two critical datasets for my work—SRTR (for actual donor outcomes) and NHANES (for counterfactual healthy non-donor trajectories)—are both federal registries with restricted access. This is not merely a technical hurdle but a political one: recent federal government layoffs have disrupted continuity, delaying data pulls and complicating permissions. I’ve been sharing analytic scripts with federal employees to keep backend development of my risk visualization webApp on track. However, updating these analyses will be increasingly challenging.

2. Challenge of Integrative Infrastructure#

Living kidney donation sits at the intersection of public health, nephrology, surgery, and bioethics—but the infrastructure to unify these domains remains fragmented. The challenge here is not a lack of knowledge per se, but the absence of systems that coherently integrate donor identification, surgical risk stratification, long-term follow-up, and ethical reflection into a single analytic continuum. My thesis lays the conceptual groundwork for this integration, even if the full technical implementation lies outside its immediate scope.

3. Challenge of Inefficiency and the Need for Automation#

Current processes—from risk estimation to consent documentation to post-donation tracking—are labor-intensive, redundant, and fragmented across institutional silos. They are also vulnerable to political disruption. While automation is not the central deliverable of this dissertation, the project lays a modular foundation for future automation: harmonized variable definitions, interpretable Cox-based risk outputs, and a patient-facing web platform for risk visualization. Here, efficiency does not mean cutting corners—it means scaffolding complexity into reproducible, transparent pipelines.

4. Challenge of Missingness and Interpretive Gaps#

There are two core interpretive challenges here:

Operational Missingness: Long-term follow-up of donors is inconsistent. This isn’t just a data problem—it reflects deeper failures in care continuity, institutional memory, and ethical responsibility.
Statistical Epistemology: The precision of our risk estimates varies dramatically with donor age. For example, older donors (e.g., 84 years old) often have much wider confidence intervals around mortality and ESRD estimates than donors under 40, even when derived from the same Cox regression framework. This isn’t a flaw—it’s a truth about how sparse data limits inference. But it demands a rethinking of what “informed consent” really means when the information is itself unstable.

Summary#

In short, the challenges I presented are not barriers to my thesis—they are its reason for being. By mapping where data is fragmented, where consent falters, and where older donors are medically underserved, my work invites both infrastructure and interpretation. The layered structure of the talk helped emphasize that these challenges are not isolated—they form a continuum from the tactical act of donation to the existential implications of long-term health.

This thesis focuses on addressing these gaps by prioritizing the patient’s central question, “Can I safely donate my kidney?” over population-level metrics. The scientific aims include: (1) quantifying donation-attributable risks of perioperative death, (2) long-term risks of ESRD and (3) mortality; (4) describing hospitalization prevalence in older donors compared to non-donors; and (5) implementing a risk calculator for perioperative death, long-term ESRD and mortality risks, and sentinel hospitalization in older donors.

This thesis will frame its work in the context of various areas of inquiry, including kidney disease, aging research, and data science. The results of all four aims are delivered in an innovative tool—an app—to enhance provider-patient dialogue about kidney donation risks (aim 5). But the architecture of the app offers so much more that will be discussed in the appendix.

This thesis addresses the critical need to improve understanding of perioperative and long-term risks for older living kidney donors. By identifying resiliency markers and individualizing risk assessment, the study aims to empower older adults and providers with evidence-based tools to make informed decisions in a growing and vulnerable donor population.

Perioperative Mortality

The 90-day risk of death following living kidney donation has been estimated at 3 per 10,000. Notably, some of these deaths were due to homicide, an outcome unlikely to reflect a perioperative risk directly attributable to nephrectomy.

To contextualize these risks, we introduce a control population to quantify the baseline 90-day mortality risk faced by healthy eligible donors who do not proceed with donation. Using data from the Organ Procurement and Transplantation Network (OPTN) and the National Health and Nutrition Examination Survey (NHANES), we estimate the 90-day risk of death for both donors and their healthy nondonor counterparts through registry linkages to the National Death Index. Cumulative incidence estimates derived from Cox regression are presented in an accompanying app (see App Features by clicking on + Expand under Living Kidney Donors).

For white female nondonors aged 40 years or younger who self-reported “Very Good” health at the time of survey, the 90-day risk of death was estimated at 1.3 per 10,000 (95% CI: 0.5-3.5 per 10,000). Among comparable donors, the 90-day risk of death was 2.2 per 10,000 (95% CI: 1.0-4.7 per 10,000). Among nondonors aged over 50 years, the 90-day risk of death was no higher than that observed in nondonors with similar demographic and health characteristics.

The nephrectomy-attributable risk for a typical young donor is approximately 0.9 per 10,000, substantially lower than the risk currently cited during the informed consent process. For older donors, the nephrectomy-attributable risk may be even lower, though it remains insufficiently quantified due to the limited availability of data.

Fig. 2 Static vs. Dynamic Presentation. Authors typically have to commit to what data they will present to colleagues, reviewers, and the public. But clinicians and policy makers typically need to explore a combination of factors to make a personalized decision tailored to their situation. Our platform demonstrates this for the person who is considering donation, but it is applicable to any decision-making situation. We also critically consider the value of a control population for accurate and meaningful inferences. Click on + Expand under -> Living Kidney Donors to review output for a 40-year-old white female in very good health (the most prevalent phenotype in the).#

Long-term Risk of ESRD
Donation-attributable risk of end-stage renal disease (ESRD) has been estimated at 26.9 per 10,000 at 15 years. Subgroup estimates for white, black, and Hispanic donors have also been reported. 4 However, personalized risk estimates that take into account age and other associated risk factors remain unknown.

We compare the risk of ESRD in kidney donors (using OPTN data) with that of a cohort of nondonors (using NHANES data). See Perioperative Mortality above for more information. Cumulative incidence estimates derived from Cox regression are presented in an accompanying app (see App Features below).

The maximum follow-up period was 15 years, with a median follow-up of 7.6 years for kidney donors (interquartile range [IQR], 3.9–11.5) and 20 years for matched healthy nondonors (IQR, 18–35). The estimated risk of ESRD at 6.5 years for the typical 40 year old white female with SBP 120 mmHg, DBP 80mmHg, BMI 24 kg/\(m^2\), eGFR 95 ml/min/1.73, urine albumin/creatinine ratio of 4mg/g, with a college level education, “Very Good” self-rated health status, no history of diabetes, hypertension, and smoking was significantly higher in kidney donors (5.6 per 10,000; 95% CI, 1.5–20.1) compared to matched healthy nondonors (1.3 per 10,000; 95% CI, 0.8–2.8). Donation by older donors such as the one seen above: 84 year old white male, SBP 140 mmHg, BMI 24/\(m^2\), eGFR 80ml/min/1.73, no history of diabetes or smoking, but has well-controlled hypertension. He is a college graduate. We estimate his risk if his 6.5 year risk at (34.6 per 10,000; 95% CI, 2.2–552.0) compared to the counterfactual scenario in which he had no donated (9.2 per 10,000; 95% CI, 5.2–16.1).

Our personalized donation-attributable risk sets a standard for informed consent among the diverse range of those who seek to donate a kidney. But uncertainty of our estimates for such a rare donor phenotype highlights the challenges that remain.

Hospitalization Risks
Hospitalization may be a sentinel event signaling risk for adverse outcomes, including end-stage renal disease (ESRD) and mortality, among living kidney donors (LKDs). However, only two years of follow-up are mandated for LKDs, limiting long-term risk characterization. To address this, we used a multicenter retrospective cohort study of LKDs with over 40 years of follow-up to identify factors associated with patient-reported all-cause hospitalization.

Patient factors were captured from self-report and supplemented by linkage to the SRTR database. Among an cohort of 3,084 LKDs, 2,251 (73%) donors who donated between May 1968 and December 2019 responded to the survey.

Overall, 1575 (70%) reported any hospitalization a median (interquartile range) of 13 (10-20+) years post-nephrectomy, with surgery/procedure as the most common cause (57%). The cumulative incidence of hospitalization was 47.2% (95%CI: 45.5-48.9) at 20 years post-donation for a 40 year old white female donor with BMI of 25, SBP of 120 mmHg, DBP of 80 mmHg, eGFR of 95 ml/min/1.73 (see App Features above). In a parsimonious model, age at donation (aHR 1.01–1.13–1.27, p=0.04), female sex (aHR 1.07–1.39–1.80, p=0.01), and post-donation diabetes/hypertension (aHR 1.22–1.52–1.88, p<0.01) were associated with hospitalization (see appendix).

While self-reported, the frequency of hospitalization among LKDs beyond two years post-nephrectomy suggests longer follow-up may be warranted for this population. Furthermore, surveillance and follow-up should be emphasized in populations at higher risk of developing diabetes and hypertension after nephrectomy. The absence of a control population with ascertainment of hospitalization remains a critical limitation of this study.

Show code cell source Hide code cell source

import numpy as np
import matplotlib.pyplot as plt
import networkx as nx

# Define the neural network fractal
def define_layers():
    return {
        'Suis': ['Patient Access/Care', 'Hospital Records', 'Disclosure Risk', 'Collaborators', 'Databases', 'Analytic Scripts', ], # Static
        'Voir': ['Information'],  
        'Choisis': ['Baseline', 'Decision'],  
        'Deviens': ['Adverse Event Markers', 'Comorbidity/ICD Codes', 'Temporal Changes'],  
        "M'èléve": ['Mortality Rate', 'Organ Failure',  'Hospitalization', 'Dependency', 'Physical Frailty']  
    }

# Assign colors to nodes
def assign_colors():
    color_map = { # Dynamic
        'yellow': ['Information'],  
        'paleturquoise': ['Analytic Scripts', 'Decision', 'Temporal Changes', 'Physical Frailty'],  
        'lightgreen': ['Databases', 'Comorbidity/ICD Codes', 'Organ Failure', 'Dependency', 'Hospitalization'],  
        'lightsalmon': [
            'Disclosure Risk', 'Collaborators', 'Baseline',  
            'Adverse Event Markers', 'Mortality Rate'
        ],
    }
    return {node: color for color, nodes in color_map.items() for node in nodes}

# Calculate positions for nodes
def calculate_positions(layer, x_offset):
    y_positions = np.linspace(-len(layer) / 2, len(layer) / 2, len(layer))
    return [(x_offset, y) for y in y_positions]

# Create and visualize the neural network graph
def visualize_nn():
    layers = define_layers()
    colors = assign_colors()
    G = nx.DiGraph()
    pos = {}
    node_colors = []

    # Add nodes and assign positions
    for i, (layer_name, nodes) in enumerate(layers.items()):
        positions = calculate_positions(nodes, x_offset=i * 2)
        for node, position in zip(nodes, positions):
            G.add_node(node, layer=layer_name)
            pos[node] = position
            node_colors.append(colors.get(node, 'lightgray'))   

    # Add edges (automated for consecutive layers)
    layer_names = list(layers.keys())
    for i in range(len(layer_names) - 1):
        source_layer, target_layer = layer_names[i], layer_names[i + 1]
        for source in layers[source_layer]:
            for target in layers[target_layer]:
                G.add_edge(source, target)

    # Draw the graph
    plt.figure(figsize=(12, 8))
    nx.draw(
        G, pos, with_labels=True, node_color=node_colors, edge_color='gray',
        node_size=3000, font_size=15, connectionstyle="arc3,rad=0.2"
    )
    plt.title("Ecosystem Integration", fontsize=23)    
    # ✅ Save the actual image *after* drawing it
    plt.savefig("figures/ecosystem-integration.jpeg", dpi=300, bbox_inches='tight')
    # plt.show()

# Run the visualization
visualize_nn()

_images/2d581caf96d04d3a12b95ca354fd3ce835ccde4b68d1ddff3625b6b798a043b1.png

Fig. 3 The Scientific Enterprise: Ecosystem, Navigation, Decisions, Tracking, Outcomes. The Ecosystem encapsulates the intricate flow of information, decision-making, and outcomes within clinical medicine and public health. It functions as a multi-layered neural network, where patients, students, researchers, and clinicians navigate a complex landscape filled with both opportunities and inefficiencies. At its front-end (yellow node), the web app serves as a transparent, personalized tool, guiding prospective kidney donors through risks, benefits, and post-donation considerations. But beyond this immediate utility lies a deeper infrastructure—a structured backend where information, data, and decision-making processes interconnect dynamically. Students exploring this system may uncover the technical mechanics of data flow, decision nodes, and analytical scripts that underpin clinical research. Meanwhile, principal investigators and research labs act as key agents, shaping the ecosystem through their decisions on disclosure risks, collaboration, and methodological rigor. As stakeholders engage with the platform, they are not merely passive participants but active contributors to the evolution of an adaptive, decentralized, and highly efficient medical knowledge network. This ecosystem is not just a digital tool—it is an epistemic revolution, inviting users at every level to move beyond static knowledge and actively shape the future of patient care, research, and public health decision-making#

This thesis invites the reader to consider the philosophical and practical shift from static to dynamic presentation of clinical risk. Traditional tools tend to summarize population-level averages—offering a narrow window into lived decision-making—but the application developed alongside this work proposes a more flexible, exploratory approach. Rather than rely solely on aggregate risk estimates, users can manipulate key variables—age, baseline GFR, comorbid conditions—to generate personalized, real-time projections of risk. In doing so, the app does not merely inform—it engages. It allows patients, clinicians, and researchers to see how the risk landscape shifts under different assumptions and across different bodies.

This dynamic interface marks a meaningful evolution in the ethics and pragmatics of informed consent. By visualizing the uncertainty that surrounds long-term donor outcomes—especially for older adults or those outside conventional risk bins—the platform offers a more transparent depiction of what we know, what we don’t, and how confidence intervals behave in edge-case scenarios. These elements are often lost in the language of “standard of care,” yet they are essential to the dignity of the donor. In this sense, the tool is not just technological—it is ethical. It dignifies variability, contextualizes uncertainty, and respects autonomy by rooting consent in deeper epistemic clarity.

Crucially, the development of this platform aligns with broader movements in open science and computational transparency. The backend of the application leverages robust national datasets and reproducible statistical code, and the thesis’s supplemental section details how the architecture—built on openly available packages and hosted with free-tier cloud tools—can be adapted, forked, or scaled. By contributing to the open-source ecosystem, this work asserts that clinical tools should not be locked behind paywalls or proprietary codebases. Instead, it embraces a model of translational research where methods, risks, and tools are legible to all stakeholders—from patients to policymakers.

One of the more sobering realities during the course of this project has been the fragility of access to federal registries, particularly the Scientific Registry of Transplant Recipients (SRTR) and NHANES. While these datasets underpin the statistical integrity of the project, recent federal layoffs and staffing shortages have disrupted communication pipelines and delayed data approvals. At several points, progress depended on personally sharing analytic scripts with federal employees to manually generate outputs for app updates. This workaround, while functional, is neither scalable nor sustainable.

The long-term vision anticipates replacing such fragile arrangements with secure API access, which would enable continuous updates to the web application as new data becomes available—similar to how software updates now occur seamlessly for computers and mobile devices. This transition to automated interoperability will reduce human bottlenecks and increase transparency, but it is not without political and institutional sensitivity. The same shifts that promise efficiency also imply displacement. The reality is that some of the very delays I’ve experienced—particularly in responding to thesis committee feedback—are downstream of job losses among federal collaborators. This project thus resides in a tense but necessary space: advocating for open systems and automation while also recognizing the emotional and structural upheaval such changes produce.

From a data perspective, this thesis draws on more than three decades of longitudinal outcomes from the SRTR and contrasts those with counterfactual trajectories derived from NHANES. Perioperative mortality, 30-year mortality, and ESRD risk estimates were built using verified death and dialysis data obtained via interagency partnerships between OPTN, CMS, and the National Death Index. Donor mortality was stringently verified, including required reporting within 72 hours for deaths occurring within two years postdonation. ESRD endpoints were defined with clinical precision: dialysis initiation, waitlisting, or transplant receipt—whichever came first.

The counterfactual control cohort—comprising 73,000 NHANES participants—was selected to mirror the clinical baseline of healthy nondonors. These data were carefully linked to federal death and dialysis registries to ensure robust longitudinal comparisons. In this way, the app does more than offer visualizations: it embodies a scientifically grounded, methodologically transparent rethinking of how comparative risk can—and should—be communicated.

Ultimately, this project underscores that technology in medicine should do more than predict—it should illuminate. It should make visible the contours of risk, the ethical texture of choice, and the systems that govern survival and sacrifice. The app is thus both artifact and argument: it argues that meaningful consent cannot be divorced from meaningful representation. And it stakes a claim that scientific infrastructure, thoughtfully built and equitably updated, can help bridge the epistemic gap between what we know and what we can responsibly ask of others. Source: Earlier Presentation

🔥 Traditional:
“I originally structured this thesis around four classic empirical aims—perioperative mortality, long-term ES RD , mortality, and hospitalization risk. But in light of the April 2 seminar, it’s now clear that Aim 5—implementing the interactive risk app—is not the delivery mechanism. It’s the thesis itself. The others? They’re scaffolding. This isn’t a pivot. It’s a re-centering.”

🌐 Data-Science:
“This app isn’t static output—it’s a dynamic, living interface. It allows patients and providers to simulate their unique clinical profile and see real-time risk projections. But more than that: it invites interaction, iteration, and trust. It hands the epistemic lens to the patient—not as a token gesture, but as a redefinition of informed consent for the 21st century. Consent isn’t just a signature anymore. It’s navigation through risk, guided by transparent data, personalized projections, and the dignity of agency.”

🌀 Hybrid:
“Because the app is code, not prose, it’s built to evolve—just like the science behind it. As new data becomes available, as models are refined, the app can and should improve. This is a scientific instrument in perpetual calibration—an interface where the user is not a passive recipient but an active participant in their own risk calculus. That’s not just modern—it’s necessary.”

📍 User-Centric:

“Aim 1 through 4 are rigorous, peer-reviewed, and necessary. But they orbit Aim 5.”
“The app is the thesis—it unites methodology, ethics, and interface into a living epistemology.”
“This is what translational science should be in 2025: iterative, interactive, and intelligible—to the very people whose lives it seeks to inform.”

Hopefully this is not just writing a dissertation. It’s architecting a 21st-century epistemic scaffold—where truth isn’t static, consent isn’t bureaucratic, and patients aren’t props. They’re protagonists.

Lets let them feel that. Then let’s walk them through it.