Using machine learning to estimate survival curves for patients receiving an increased risk for disease transmission heart, liver, or lung versus waiting for a standard organ.

INTRODUCTION: Over 19% of deceased organ donors are labeled increased risk for disease transmission (IRD) for viral blood-borne disease transmission. Many potential organ recipients need to decide between accepting an IRD organ offer and waiting for a non-IRD organ. METHODS: Using machine learning and simulation, we built transplant and waitlist survival models and compared the survival for patients accepting IRD organ offers or waiting for non-IRD organs for the heart, liver, and lung. The simulation consisted of generating 20 000 different scenarios of a recipient either receiving an IRD organ or waiting and receiving a non-IRD organ. RESULTS: In the simulations, the 5-year survival probabilities of heart, liver, and lung recipients who accepted IRD organ offers increased on average by 10.2%, 12.7%, and 7.2%, respectively, compared with receiving a non-IRD organ after average wait times (190, 228, and 223 days, respectively). When the estimated waitlist time was at least 5 days for the liver, and 1 day for the heart and lung, 50% or more of the simulations resulted in a higher chance of 5-year survival when the patient received an IRD organ versus when the patient remained on the waitlist. We also developed a simple equation to estimate the benefits, in terms of 5-year survival probabilities, of receiving an IRD organ versus waiting for a non-IRD organ, for a particular set of recipient/donor characteristics. CONCLUSION: For all three organs, the majority of patients are predicted to have higher 5-year survival accepting an IRD organ offer compared with waiting for a non-IRD organ.

A mathematical model to describe survival among liver recipients from deceased donors with risk of transmitting infectious encephalitis pathogens.

BACKGROUND: Between 2002 and 2013, the organs of 13 deceased donors with infectious encephalitis were transplanted, causing infections in 23 recipients. As a consequence, organs from donors showing symptoms of encephalitis (increased probability of infectious encephalitis (IPIE) organs) might be declined. We had previously characterized the risk of IPIE organs using data available to most transplant teams and not requiring special diagnostic tests. If the probability of infection is low, the benefits of a transplant from a donor with suspected infectious encephalitis might outweigh the risk and could be lifesaving for some transplant candidates. METHODS: Using organ transplant data and Cox Proportional Hazards models, we determined liver donor and recipient characteristics predictive of post-transplant or waitlist survival and generated 5-year survival probability curves. We also calculated expected waiting times for an organ offer based on transplant candidate characteristics. Using a limited set of actual cases of infectious encephalitis transmission via transplant, we estimated post-transplant survival curves given an organ from an IPIE donor. RESULTS: 54% (1256) of patients registered from 2002-2006 who died or were removed from the waiting list because of deteriorated condition within 1 year could have had an at least marginal estimated benefit by accepting an IPIE liver with some probability of infection, with the odds increasing to 86% of patients if the probability of infection was low (5% or less). Additionally, 54% (1252) were removed from the waiting list prior to their estimated waiting time for a non-IPIE liver and could have benefited from an IPIE liver. CONCLUSION: Improved allocation and utilization of IPIE livers could be achieved by evaluating the patient-specific trade-offs between (a) accepting an IPIE liver and (b) remaining on the waitlist and accepting a non-IPIE liver after the estimated waiting time.

Assessment of risk for transplant-transmissible infectious encephalitis among deceased organ donors.

BACKGROUND: There were 13 documented clusters of infectious encephalitis transmission via organ transplant from deceased donors to recipients during 2002-2013. Hence, organs from donors diagnosed with encephalitis are often declined because of concerns about the possibility of infection, given that there is no quick and simple test to detect causes of infectious encephalitis. METHODS: We constructed a database containing cases of infectious and non-infectious encephalitis. Using statistical imputation, cross-validation, and regression techniques, we determined deceased organ donor characteristics, including demographics, signs, symptoms, physical exam, and laboratory findings, predictive of infectious vs non-infectious encephalitis, and developed a calculator which assesses the risk of infection. RESULTS: Using up to 12 predictive patient characteristics (with a minimum of 3, depending on what information is available), the calculator provides the probability that a donor may have infectious vs non-infectious encephalitis, improving the prediction accuracy over current practices. These characteristics include gender, fever, immunocompromised state (other than HIV), cerebrospinal fluid elevation, altered mental status, psychiatric features, cranial nerve abnormality, meningeal signs, focal motor weakness, Babinski's sign, movement disorder, and sensory abnormalities. CONCLUSION: In the absence of definitive diagnostic testing in a potential organ donor, infectious encephalitis can be predicted with a risk score. The risk calculator presented in this paper represents a prototype, establishing a framework that can be expanded to other infectious diseases transmissible through solid organ transplantation.

New approach for optimal electricity planning and dispatching with hourly time-scale air quality and health considerations.

Integrating accurate air quality modeling with decision making is hampered by complex atmospheric physics and chemistry and its coupling with atmospheric transport. Existing approaches to model the physics and chemistry accurately lead to significant computational burdens in computing the response of atmospheric concentrations to changes in emissions profiles. By integrating a reduced form of a fully coupled atmospheric model within a unit commitment optimization model, we allow, for the first time to our knowledge, a fully dynamical approach toward electricity planning that accurately and rapidly minimizes both cost and health impacts. The reduced-form model captures the response of spatially resolved air pollutant concentrations to changes in electricity-generating plant emissions on an hourly basis with accuracy comparable to a comprehensive air quality model. The integrated model allows for the inclusion of human health impacts into cost-based decisions for power plant operation. We use the new capability in a case study of the state of Georgia over the years of 2004-2011, and show that a shift in utilization among existing power plants during selected hourly periods could have provided a health cost savings of $175.9 million dollars for an additional electricity generation cost of $83.6 million in 2007 US dollars (USD2007). The case study illustrates how air pollutant health impacts can be cost-effectively minimized by intelligently modulating power plant operations over multihour periods, without implementing additional emissions control technologies.

Split or whole liver transplantation? Utilization and posttransplant survival.

BACKGROUND: Split liver transplantation (SLT), where a single donor liver is divided for transplantation to 2 recipients, has the potential to increase the availability of size-matched livers for pediatric candidates and expand the supply of donor organs available for adult candidates. Although SLT is a well-established technique, the number of SLTs has remained flat during the past 2 decades, partly due to concerns about the posttransplant survival of SLT recipients compared with whole liver transplantation (WLT) recipients. Prior work on SLT versus WLT survival analysis had limitations because, for pediatric recipients, it did not consider the correlations between donor age/weight and the allograft type, and for adult recipients, it may have included records where the donor livers did not meet the split liver criteria (splittable). METHODS: Using the Organ Procurement and Transplantation Network's database (2003-2019), this study analyzes and compares (i) key characteristics of donors and recipients, (ii) donor-recipient match dynamics (organ offers and accept/decline decisions), and (iii) recipient posttransplant survival, for SLT and WLT. RESULTS AND CONCLUSIONS: The results in this study show that the posttransplant survival of SLT and WLT recipients is similar (controlling for other confounding factors that may impact posttransplant survival), highlighting the importance of SLT for increasing the liver supply and potential benefits for both pediatric and adult candidates.

Predicting a kidney transplant patient's pre-transplant functional status based on information from waitlist registration.

With over 100,000 patients on the kidney transplant waitlist in 2019, it is important to understand if and how the functional status of a patient may change while on the waitlist. Recorded both at registration and just prior to transplantation, the Karnofsky Performance Score measures a patient's functional status and takes on values ranging from 0 to 100 in increments of 10. Using machine learning techniques, we built a gradient boosting regression model to predict a patient's pre-transplant functional status based on information known at the time of waitlist registration. The model's predictions result in an average root mean squared error of 12.99 based on 5 rolling origin cross validations and 12.94 in a separate out-of-time test. In comparison, predicting that the pre-transplant functional status remains the same as the status at registration, results in average root mean squared errors of 14.50 and 14.11 respectively. The analysis is based on 118,401 transplant records from 2007 to 2019. To the best of our knowledge, there has been no previously published research on building a model to predict kidney pre-transplant functional status. We also find that functional status at registration and total serum albumin, have the most impact in predicting the pre-transplant functional status.

Flexible analytic model to inform multi-stakeholder pediatric vaccine scheduling decisions.

BACKGROUND: Pediatric immunization is important for preventing potentially life-threatening diseases in children. Over time, the number of recommended pediatric vaccines has increased and is likely to increase further as new vaccines are developed. Given the different number of doses for available vaccines and various constraints (e.g., the appropriate age for each dose of a vaccine or the time between doses), it is challenging to develop a recommended vaccination schedule or a catch-up schedule when a child falls behind on one or more vaccinations. METHODS: We developed an integer programming optimization model, enabled by Python programming and embedded into an Excel-based decision tool, to recommend childhood vaccination schedules or personalized catch-up schedules. The model recommends a vaccination schedule that balances the goal of being as close as possible to the clinically-indicated dosing schedules and the goal of minimizing clinic visits, and gives users the ability to trade off between these two goals. We illustrated the broad applicability of our proposed model with commonly-faced vaccine scheduling challenges in the United States. RESULTS: The illustrative computational case study confirms our model's ability to create personalized schedules based on each child's age and vaccination history, and to adjust appropriately when a new vaccine becomes available. CONCLUSIONS: The model presented in this paper fills the need for an easy-to-use tool to recommend vaccination schedules for de novo and catch-up purposes. It provides straightforward recommendations that can be easily used by physicians, is flexible to handle the requirements varying by region, and can be updated as new vaccines are approved for use.

Using machine learning and an ensemble of methods to predict kidney transplant survival.

We used an ensemble of statistical methods to build a model that predicts kidney transplant survival and identifies important predictive variables. The proposed model achieved better performance, measured by Harrell's concordance index, than the Estimated Post Transplant Survival model used in the kidney allocation system in the U.S., and other models published recently in the literature. The model has a five-year concordance index of 0.724 (in comparison, the concordance index is 0.697 for the Estimated Post Transplant Survival model, the state of the art currently in use). It combines predictions from random survival forests with a Cox proportional hazards model. The rankings of importance for the model's variables differ by transplant recipient age. Better survival predictions could eventually lead to more efficient allocation of kidneys and improve patient outcomes.