Try a new search

Format these results:

Searched for:

person:mankom01

in-biosketch:yes

Total Results:

36


Medical large language models are vulnerable to data-poisoning attacks

Alber, Daniel Alexander; Yang, Zihao; Alyakin, Anton; Yang, Eunice; Rai, Sumedha; Valliani, Aly A; Zhang, Jeff; Rosenbaum, Gabriel R; Amend-Thomas, Ashley K; Kurland, David B; Kremer, Caroline M; Eremiev, Alexander; Negash, Bruck; Wiggan, Daniel D; Nakatsuka, Michelle A; Sangwon, Karl L; Neifert, Sean N; Khan, Hammad A; Save, Akshay Vinod; Palla, Adhith; Grin, Eric A; Hedman, Monika; Nasir-Moin, Mustafa; Liu, Xujin Chris; Jiang, Lavender Yao; Mankowski, Michal A; Segev, Dorry L; Aphinyanaphongs, Yindalon; Riina, Howard A; Golfinos, John G; Orringer, Daniel A; Kondziolka, Douglas; Oermann, Eric Karl
The adoption of large language models (LLMs) in healthcare demands a careful analysis of their potential to spread false medical knowledge. Because LLMs ingest massive volumes of data from the open Internet during training, they are potentially exposed to unverified medical knowledge that may include deliberately planted misinformation. Here, we perform a threat assessment that simulates a data-poisoning attack against The Pile, a popular dataset used for LLM development. We find that replacement of just 0.001% of training tokens with medical misinformation results in harmful models more likely to propagate medical errors. Furthermore, we discover that corrupted models match the performance of their corruption-free counterparts on open-source benchmarks routinely used to evaluate medical LLMs. Using biomedical knowledge graphs to screen medical LLM outputs, we propose a harm mitigation strategy that captures 91.9% of harmful content (F1 = 85.7%). Our algorithm provides a unique method to validate stochastically generated LLM outputs against hard-coded relationships in knowledge graphs. In view of current calls for improved data provenance and transparent LLM development, we hope to raise awareness of emergent risks from LLMs trained indiscriminately on web-scraped data, particularly in healthcare where misinformation can potentially compromise patient safety.
PMID: 39779928
ISSN: 1546-170x
CID: 5782182

Targeted Broader Sharing for Liver Continuous Distribution

Mankowski, Michal A; Wood, Nicholas L; Massie, Allan B; Segev, Dorry L; Trichakis, Nikolaos; Gentry, Sommer E
BACKGROUND:In recent years, changes to US organ allocation have aimed to improve equity and accessibility across regions. The Organ Procurement and Transplantation Network plans to adopt continuous liver distribution, prioritizing candidates based on a weighted composite allocation score (CAS) incorporating proximity, ABO types, medical urgency, and pediatric priority. The Liver Committee has requested research on CAS variations that account for geographical heterogenicity. METHODS:We describe a method for designing a geographically heterogeneous CAS with targeted broader sharing (CAS-TBS) to balance the highly variable geographic distributions of liver transplant listings and liver donations. CAS-TBS assigns each donor hospital to either broader sharing or nearby sharing, adjusting donor-candidate distance allocation points accordingly. RESULTS:We found that to reduce geographic disparity in the median Model for End-stage Liver Disease at transplant (MMaT), >75% of livers recovered in regions 2 and 10 should be distributed with broader sharing, whereas 95% of livers recovered in regions 5 and 1 should be distributed with nearby sharing. In a 3-y simulation of liver allocation, CAS-TBS decreased MMaT by 2.1 points in high-MMaT areas such as region 5 while increasing MMaT only by 0.65 points in low-MMaT areas such as region 3. CAS-TBS significantly decreased median transport distance from 202 to 167 nautical miles under acuity circles and decreased waitlist deaths. CONCLUSIONS:Our CAS-TBS design methodology could be applied to design geographically heterogeneous allocation scores that reflect transplant community values and priorities within the continuous distribution project of the Organ Procurement and Transplantation Network. In our simulations, the incremental benefit of CAS-TBS over CAS was modest.
PMID: 39245819
ISSN: 1534-6080
CID: 5689942

Balancing equity and human leukocyte antigen matching in deceased-donor kidney allocation with eplet mismatch

Mankowski, Michal A; Gragert, Loren; Keating, Brendan; Lonze, Bonnie E; Segev, Dorry L; Montgomery, Robert; Gentry, Sommer E; Mangiola, Massimo
Human leukocyte antigen-level matching in US kidney allocation has been deemphasized due to its role in elevating racial disparities. Molecular matching based on eplets might improve risk stratification compared to antigen matching, but the magnitude of racial disparities in molecular matching is not known. To assign eplets unambiguously, we utilized a cohort of 5193 individuals with high-resolution allele-level human leukocyte antigen genotypes from the National Kidney Registry. Using repeated random sampling to simulate donor-recipient genotype pairings based on the ethnic composition of the historical US deceased-donor pool, we profiled the percentage of well-matched donors available for candidates by ethnicity. The prevalence of well-matched donors with 0-DR/DQ eplet mismatch was 3-fold less racially disparate for Black and Asian candidates and 2-fold less for Latino candidates compared to 0-ABDR antigen mismatches. Compared to 0-DR antigen mismatch, 0-DR eplet mismatch was 1.33-fold more racially disparate for Asian and 1.28-fold more for Latino, with similar disparity for Black candidates, whereas 0-DQ eplet mismatch reduced disparities, showing 1.26-fold less disparity for Black, 1.14-fold less for Latino, but 1.26-fold higher for Asian candidates. The prevalence of well-matched donors for candidates of different ethnicities varied according to which molecules were chosen to define a low-risk match.
PMID: 39631566
ISSN: 1600-6143
CID: 5781742

HLA EPLET Frequencies Are Similar in Six Population Groups and Are Expressed by the Most Common HLA Alleles

Mangiola, Massimo; Ellison, Mitchell; Marrari, Marilyn; Xu, Qingyong; Mankowski, Michal; Sese, Doreen; Lonze, Bonnie E; Montgomery, Robert A; Zeevi, Adriana
The degree of immunological compatibility between donors and recipients greatly impacts allograft survival. In the United States kidney allocation system, HLA antigen-level matching has been shown to cause ethnic disparities and thus, has been de-emphasised. However, priority points are still awarded for antigen-level zero-ABDR matching, zero-DR matching and one-DR matching. Recently, the degree of HLA molecular (eplet) mismatch has emerged as a more accurate measure of immunological risk, and eplet mismatch load has gained attention as a possible biomarker to improve HLA compatibility. However, little is known about the frequency of eplets in population groups, which is a necessary step to ensure that candidates from any ethnical background can have similar chances at a well-matched organ. Eplet frequencies were estimated using HLA alleles in the Common, Intermediate and Well-Documented (CIWD) 3.0.0 catalogue for six population groups: African-American (AFA), Asian-Pacific Islander (API), European/European descent (EURO), Middle East/North Coast of Africa (MENA), Hispanic/Latino (HIS) and Native-American (NAM). We determined that 98.6% (484 out of 491) of HLA eplets are expressed by the common HLA alleles in all population groups. Of the seven eplets that were expressed by less common HLA alleles, six were Class I eplets and one was expressed by HLA-DQB1 alleles and most were expressed by HLA alleles that were more commonly observed in European/European descent populations. Our observations indicate that HLA eplets will not cause any significant disparity if applied to HLA molecular compatibility, regardless of the ethnic origin of both recipients and donors.
PMCID:11670879
PMID: 39711219
ISSN: 2059-2310
CID: 5767162

Generalizability of Kidney Transplant Data in Electronic Health Records - The Epic Cosmos Database versus the Scientific Registry of Transplant Recipients

Mankowski, Michal A; Bae, Sunjae; Strauss, Alexandra T; Lonze, Bonnie E; Orandi, Babak J; Stewart, Darren; Massie, Allan B; McAdams-DeMarco, Mara A; Oermann, Eric K; Habal, Marlena; Iturrate, Eduardo; Gentry, Sommer E; Segev, Dorry L; Axelrod, David
Developing real-world evidence from electronic health records (EHR) is vital to advance kidney transplantation (KT). We assessed the feasibility of studying KT using the Epic Cosmos aggregated EHR dataset, which includes 274 million unique individuals cared for in 238 U.S. health systems, by comparing it with the Scientific Registry of Transplant Recipients (SRTR). We identified 69,418 KT recipients transplanted between January 2014 and December 2022 in Cosmos (39.4% of all US KT transplants during this period). Demographics and clinical characteristics of recipients captured in Cosmos were consistent with the overall SRTR cohort. Survival estimates were generally comparable, although there were some differences in long-term survival. At 7 years post-transplant, patient survival was 80.4% in Cosmos and 77.8% in SRTR. Multivariable Cox regression showed consistent associations between clinical factors and mortality in both cohorts, with minor discrepancies in the associations between death and both age and race. In summary, Cosmos provides a reliable platform for KT research, allowing EHR-level clinical granularity not available with either the transplant registry or healthcare claims. Consequently, Cosmos will enable novel analyses to improve our understanding of KT management on a national scale.
PMID: 39550008
ISSN: 1600-6143
CID: 5754062

Hospitalization and Hospitalized Delirium Are Associated With Decreased Access to Kidney Transplantation and Increased Risk of Waitlist Mortality

Long, Jane J; Hong, Jingyao; Liu, Yi; Nalatwad, Akanksha; Li, Yiting; Ghildayal, Nidhi; Johnston, Emily A; Schwartzberg, Jordan; Ali, Nicole; Oermann, Eric; Mankowski, Michal; Gelb, Bruce E; Chanan, Emily L; Chodosh, Joshua L; Mathur, Aarti; Segev, Dorry L; McAdams-DeMarco, Mara A
BACKGROUND:Kidney transplant (KT) candidates often experience hospitalizations, increasing their delirium risk. Hospitalizations and delirium are associated with worse post-KT outcomes, yet their relationship with pre-KT outcomes is less clear. Pre-KT delirium may worsen access to KT due to its negative impact on cognition and ability to maintain overall health. METHODS:Using a prospective cohort of 2374 KT candidates evaluated at a single center (2009-2020), we abstracted hospitalizations and associated delirium records after listing via chart review. We evaluated associations between waitlist mortality and likelihood of KT with hospitalizations and hospitalized delirium using competing risk models and tested whether associations differed by gerontologic factors. RESULTS: < 0.001), with those aged ≥65 having a 61% lower likelihood of KT. CONCLUSION/CONCLUSIONS:Hospitalization and delirium are associated with worse pre-KT outcomes and have serious implications on candidates' access to KT. Providers should work to reduce preventable instances of delirium.
PMID: 39498973
ISSN: 1399-0012
CID: 5766752

Trials and Tribulations: Responses of ChatGPT to Patient Questions About Kidney Transplantation

Xu, Jingzhi; Mankowski, Michal; Vanterpool, Karen B; Strauss, Alexandra T; Lonze, Bonnie E; Orandi, Babak J; Stewart, Darren; Bae, Sunjae; Ali, Nicole; Stern, Jeffrey; Mattoo, Aprajita; Robalino, Ryan; Soomro, Irfana; Weldon, Elaina; Oermann, Eric K; Aphinyanaphongs, Yin; Sidoti, Carolyn; McAdams-DeMarco, Mara; Massie, Allan B; Gentry, Sommer E; Segev, Dorry L; Levan, Macey L
PMID: 39477825
ISSN: 1534-6080
CID: 5747132

ChatGPT Solving Complex Kidney Transplant Cases: A Comparative Study With Human Respondents

Mankowski, Michal A; Jaffe, Ian S; Xu, Jingzhi; Bae, Sunjae; Oermann, Eric K; Aphinyanaphongs, Yindalon; McAdams-DeMarco, Mara A; Lonze, Bonnie E; Orandi, Babak J; Stewart, Darren; Levan, Macey; Massie, Allan; Gentry, Sommer; Segev, Dorry L
INTRODUCTION/BACKGROUND:ChatGPT has shown the ability to answer clinical questions in general medicine but may be constrained by the specialized nature of kidney transplantation. Thus, it is important to explore how ChatGPT can be used in kidney transplantation and how its knowledge compares to human respondents. METHODS:We prompted ChatGPT versions 3.5, 4, and 4 Visual (4 V) with 12 multiple-choice questions related to six kidney transplant cases from 2013 to 2015 American Society of Nephrology (ASN) fellowship program quizzes. We compared the performance of ChatGPT with US nephrology fellowship program directors, nephrology fellows, and the audience of the ASN's annual Kidney Week meeting. RESULTS:Overall, ChatGPT 4 V correctly answered 10 out of 12 questions, showing a performance level comparable to nephrology fellows (group majority correctly answered 9 of 12 questions) and training program directors (11 of 12). This surpassed ChatGPT 4 (7 of 12 correct) and 3.5 (5 of 12). All three ChatGPT versions failed to correctly answer questions where the consensus among human respondents was low. CONCLUSION/CONCLUSIONS:Each iterative version of ChatGPT performed better than the prior version, with version 4 V achieving performance on par with nephrology fellows and training program directors. While it shows promise in understanding and answering kidney transplantation questions, ChatGPT should be seen as a complementary tool to human expertise rather than a replacement.
PMCID:11441623
PMID: 39329220
ISSN: 1399-0012
CID: 5714092

Balancing Equity and HLA Matching in Deceased-Donor Kidney Allocation with Eplet Mismatch

Mankowski, Michal A; Gragert, Loren; Segev, Dorry L; Montgomery, Robert; Gentry, Sommer E; Mangiola, Massimo
BACKGROUND/UNASSIGNED:Prioritization of HLA antigen-level matching in the US kidney allocation system intends to improve post-transplant survival but causes racial disparities and thus has been substantially de-emphasized. Recently, molecular matching based on eplets has been found to improve risk stratification compared to antigen matching. METHODS/UNASSIGNED:To assign eplets unambiguously, we utilized a cohort of 5193 individuals with high resolution allele-level HLA genotypes from the National Kidney Registry. Using repeated random sampling to simulate donor-recipient genotype pairings based on the ethnic composition of the historical US deceased donor pool, we profiled the percentage of well-matched donors for candidates by ethnicity. RESULTS/UNASSIGNED:The percentage of well-matched donors with zero-DR/DQ eplet mismatch was 3-fold less racially disparate for Black and Asian candidates than percentage of donors with zero-ABDR antigen mismatches, and 2-fold less racially disparate for Latino candidates. For other HLA antigen and eplet mismatch thresholds, the percentage of well-matched donors was more similar across candidate ethnic groups. CONCLUSIONS/UNASSIGNED:Compared to the current zero-ABDR antigen mismatch, prioritizing a zero-DR/DQ eplet mismatch in allocation would decrease racial disparities and increase the percentage of well-matched donors. High resolution HLA deceased donor genotyping would enable unambiguous assignment of eplets to operationalize molecular mismatch metrics in allocation. KEY POINTS/UNASSIGNED:
PMCID:11213093
PMID: 38947023
CID: 5738622

Removing geographic boundaries from liver allocation: A method for designing continuous distribution scores

Mankowski, Michal A; Wood, Nicholas L; Segev, Dorry L; Gentry, Sommer E
BACKGROUND:The Organ Procurement and Transplantation Network (OPTN) is eliminating geographic boundaries in liver allocation, in favor of continuous distribution. Continuous distribution allocates organs via a composite allocation score (CAS): a weighted sum of attributes like medical urgency, candidate biology, and placement efficiency. The opportunity this change represents, to include new variables and features for prioritizing candidates, will require lengthy and contentious discussions to establish community consensus. Continuous distribution could instead be implemented rapidly by computationally translating the allocation priorities for pediatric, status 1, and O/B blood type liver candidates that are presently implemented via geographic boundaries into points and weights in a CAS. METHODS:Using simulation with optimization, we designed a CAS that is minimally disruptive to existing prioritizations, and that eliminates geographic boundaries and minimizes waitlist deaths without harming vulnerable populations. RESULTS:Compared with Acuity Circles (AC) in a 3-year simulation, our optimized CAS decreased deaths from 7771.2 to 7678.8 while decreasing average (272.66 NM vs. 264.30 NM) and median (201.14 NM vs. 186.49 NM) travel distances. Our CAS increased travel only for high MELD and status 1 candidates (423.24 NM vs. 298.74 NM), and reduced travel for other candidates (198.98 NM vs. 250.09 NM); overall travel burden decreased. CONCLUSION/CONCLUSIONS:Our CAS reduced waitlist deaths by sending livers for high-MELD and status 1 candidates farther, while keeping livers for lower MELD candidates nearby. This advanced computational method can be applied again after wider discussions of adding new priorities conclude; our method designs score weightings to achieve any specified feasible allocation outcomes.
PMID: 37204074
ISSN: 1399-0012
CID: 5486532