Reinforcement learning informs optimal treatment strategies to limit antibiotic resistance.

Antimicrobial resistance was estimated to be associated with 4.95 million deaths worldwide in 2019. It is possible to frame the antimicrobial resistance problem as a feedback-control problem. If we could optimize this feedback-control problem and translate our findings to the clinic, we could slow, prevent, or reverse the development of high-level drug resistance. Prior work on this topic has relied on systems where the exact dynamics and parameters were known a priori. In this study, we extend this work using a reinforcement learning (RL) approach capable of learning effective drug cycling policies in a system defined by empirically measured fitness landscapes. Crucially, we show that it is possible to learn effective drug cycling policies despite the problems of noisy, limited, or delayed measurement. Given access to a panel of 15 [Formula: see text]-lactam antibiotics with which to treat the simulated Escherichia coli population, we demonstrate that RL agents outperform two naive treatment paradigms at minimizing the population fitness over time. We also show that RL agents approach the performance of the optimal drug cycling policy. Even when stochastic noise is introduced to the measurements of population fitness, we show that RL agents are capable of maintaining evolving populations at lower growth rates compared to controls. We further tested our approach in arbitrary fitness landscapes of up to 1,024 genotypes. We show that minimization of population fitness using drug cycles is not limited by increasing genome size. Our work represents a proof-of-concept for using AI to control complex evolutionary processes.

Evolution-Informed Strategies for Combating Drug Resistance in Cancer.

The ever-changing nature of cancer poses the most difficult challenge oncologists face today. Cancer's remarkable adaptability has inspired many to work toward understanding the evolutionary dynamics that underlie this disease in hopes of learning new ways to fight it. Eco-evolutionary dynamics of a tumor are not accounted for in most standard treatment regimens, but exploiting them would help us combat treatment-resistant effectively. Here, we outline several notable efforts to exploit these dynamics and circumvent drug resistance in cancer.

Crosstalkr: An open-source R package to facilitate drug target identification.

In the last few decades, interest in graph-based analysis of biological networks has grown substantially. Protein-protein interaction networks are one of the most common biological networks, and represent the molecular relationships between every known protein and every other known protein. Integration of these interactomic data into bioinformatic pipelines may increase the translational potential of discoveries made through analysis of multi-omic datasets. Crosstalkr provides a unified toolkit for drug target and disease subnetwork identification, two of the most common uses of protein protein interaction networks. First, crosstalkr enables users to download and leverage high-quality protein-protein interaction networks from online repositories. Users can then filter these large networks into manageable subnetworks using a variety of methods. For example, network filtration can be done using random walks with restarts, starting at the user-provided seed proteins. Affinity scores from a given random walk with restarts are compared to a bootstrapped null distribution to assess statistical significance. Random walks are implemented using sparse matrix multiplication to facilitate fast execution. Next, users can perform in-silico repression experiments to assess the relative importance of nodes in their network. At this step, users can supply protein or gene expression data to make node rankings more meaningful. The default behavior evaluates the human interactome. However, users can evaluate more than 1000 non-human protein-protein interaction networks as a result of integration with StringDB. It is a free, open-source R package designed to allow users to integrate functional analysis using the protein-protein interaction network into existing bioinformatic pipelines. A beta version of crosstalkr available on CRAN (https://cran.rstudio.com/web/packages/crosstalkr/index.html).

Reinforcement Learning informs optimal treatment strategies to limit antibiotic resistance.

Antimicrobial resistance was estimated to be associated with 4.95 million deaths worldwide in 2019. It is possible to frame the antimicrobial resistance problem as a feedback-control problem. If we could optimize this feedback-control problem and translate our findings to the clinic, we could slow, prevent or reverse the development of high-level drug resistance. Prior work on this topic has relied on systems where the exact dynamics and parameters were known a priori. In this study, we extend this work using a reinforcement learning (RL) approach capable of learning effective drug cycling policies in a system defined by empirically measured fitness landscapes. Crucially, we show that is possible to learn effective drug cycling policies despite the problems of noisy, limited, or delayed measurement. Given access to a panel of 15 ß-lactam antibiotics with which to treat the simulated E. coli population, we demonstrate that RL agents outperform two naive treatment paradigms at minimizing the population fitness over time. We also show that RL agents approach the performance of the optimal drug cycling policy. Even when stochastic noise is introduced to the measurements of population fitness, we show that RL agents are capable of maintaining evolving populations at lower growth rates compared to controls. We further tested our approach in arbitrary fitness landscapes of up to 1024 genotypes. We show that minimization of population fitness using drug cycles is not limited by increasing genome size. Our work represents a proof-of-concept for using AI to control complex evolutionary processes.

Network potential identifies therapeutic miRNA cocktails in Ewing sarcoma.

MicroRNA (miRNA)-based therapies are an emerging class of targeted therapeutics with many potential applications. Ewing Sarcoma patients could benefit dramatically from personalized miRNA therapy due to inter-patient heterogeneity and a lack of druggable (to this point) targets. However, because of the broad effects miRNAs may have on different cells and tissues, trials of miRNA therapies have struggled due to severe toxicity and unanticipated immune response. In order to overcome this hurdle, a network science-based approach is well-equipped to evaluate and identify miRNA candidates and combinations of candidates for the repression of key oncogenic targets while avoiding repression of essential housekeeping genes. We first characterized 6 Ewing sarcoma cell lines using mRNA sequencing. We then estimated a measure of tumor state, which we term network potential, based on both the mRNA gene expression and the underlying protein-protein interaction network in the tumor. Next, we ranked mRNA targets based on their contribution to network potential. We then identified miRNAs and combinations of miRNAs that preferentially act to repress mRNA targets with the greatest influence on network potential. Our analysis identified TRIM25, APP, ELAV1, RNF4, and HNRNPL as ideal mRNA targets for Ewing sarcoma therapy. Using predicted miRNA-mRNA target mappings, we identified miR-3613-3p, let-7a-3p, miR-300, miR-424-5p, and let-7b-3p as candidate optimal miRNAs for preferential repression of these targets. Ultimately, our work, as exemplified in the case of Ewing sarcoma, describes a novel pipeline by which personalized miRNA cocktails can be designed to maximally perturb gene networks contributing to cancer progression.

UV decontamination of personal protective equipment with idle laboratory biosafety cabinets during the COVID-19 pandemic.

Personal protective equipment (PPE) is crucially important to the safety of both patients and medical personnel, particularly in the event of an infectious pandemic. As the incidence of Coronavirus Disease 2019 (COVID-19) increases exponentially in the United States and many parts of the world, healthcare provider demand for these necessities is currently outpacing supply. In the midst of the current pandemic, there has been a concerted effort to identify viable ways to conserve PPE, including decontamination after use. In this study, we outline a procedure by which PPE may be decontaminated using ultraviolet (UV) radiation in biosafety cabinets (BSCs), a common element of many academic, public health, and hospital laboratories. According to the literature, effective decontamination of N95 respirator masks or surgical masks requires UV-C doses of greater than 1 Jcm-2, which was achieved after 4.3 hours per side when placing the N95 at the bottom of the BSCs tested in this study. We then demonstrated complete inactivation of the human coronavirus NL63 on N95 mask material after 15 minutes of UV-C exposure at 61 cm (232 µWcm-2). Our results provide support to healthcare organizations looking for methods to extend their reserves of PPE.

UV-C tower for point-of-care decontamination of filtering facepiece respirators.

BACKGROUND: Filtering facepiece respirators (FFR) are critical for protecting essential personnel and limiting the spread of disease. Due to the current COVID-19 pandemic, FFR supplies are dwindling in many health systems, necessitating re-use of potentially contaminated FFR. Multiple decontamination solutions have been developed to meet this pressing need, including systems designed for bulk decontamination of FFR using vaporous hydrogen peroxide or ultraviolet-C (UV-C) radiation. However, the large scale on which these devices operate may not be logistically practical for small or rural health care settings or for ad hoc use at points-of-care. METHODS: Here, we present the Synchronous UV Decontamination System, a novel device for rapidly deployable, point-of-care decontamination using UV-C germicidal irradiation. We designed a compact, easy-to-use device capable of delivering over 2 J cm2 of UV-C radiation in one minute. RESULTS: We experimentally tested Synchronous UV Decontamination System' microbicidal capacity and found that it eliminates near all virus from the surface of tested FFRs, with less efficacy against pathogens embedded in the inner layers of the masks. CONCLUSIONS: This short decontamination time should enable care-providers to incorporate decontamination of FFR into a normal donning and doffing routine following patient encounters.

Potential survival benefits from optimized chemotherapy implementation in advanced ovarian cancer: Projections from a microsimulation model.

BACKGROUND: Ovarian cancer is often diagnosed in advanced stages, when survival is poor. Treatment advances have been made, but are inconsistently implemented. Our purpose was to project the maximum life expectancy gains that could be achieved in women with stage IIIC epithelial ovarian cancer if the implementation of available chemotherapy regimens could be optimized. METHODS: We used a microsimulation model to estimate life expectancy benefits associated with "optimized" implementation of four post-operative chemotherapy options: standard intravenous chemotherapy; intraperitoneal + intravenous chemotherapy; bevacizumab + intravenous chemotherapy; and hyperthermic intraperitoneal chemotherapy + intravenous chemotherapy. Optimized implementation was defined as follows. Patients triaged to primary cytoreductive surgery received intraperitoneal + intravenous chemotherapy if optimally or completely cytoreduced, and bevacizumab + intravenous chemotherapy if suboptimally cytoreduced. Patients triaged to neoadjuvant chemotherapy received hyperthermic intraperitoneal chemotherapy at interval cytoreductive surgery if optimally or completely cytoreduced, and standard IV chemotherapy if suboptimally cytoreduced. Life expectancy associated with optimized implementation was compared with that of current utilization practices, estimated using published literature and the National Cancer Database. Effects of model uncertainty were evaluated in sensitivity analyses. RESULTS: Life expectancy associated with optimized implementation vs. current practice was 76.7 vs. 64.5 months (life expectancy gain = 12.2 months). Providing intraperitoneal + intravenous chemotherapy to all eligible patients was the largest driver of life expectancy gains, due to both the potential benefit conferred by intraperitoneal + intravenous chemotherapy and the proportion of eligible women who do not receive intraperitoneal + intravenous chemotherapy in current practice. CONCLUSION: Population-level life expectancy in stage IIIC epithelial ovarian cancer could be substantially improved through greater uptake of available chemotherapy regimens.

Prevention of Prescription Opioid Misuse and Projected Overdose Deaths in the United States.

Importance: Deaths due to opioid overdose have tripled in the last decade. Efforts to curb this trend have focused on restricting the prescription opioid supply; however, the near-term effects of such efforts are unknown. Objective: To project effects of interventions to lower prescription opioid misuse on opioid overdose deaths from 2016 to 2025. Design, Setting, and Participants: This system dynamics (mathematical) model of the US opioid epidemic projected outcomes of simulated individuals who engage in nonmedical prescription or illicit opioid use from 2016 to 2025. The analysis was performed in 2018 by retrospectively calibrating the model from 2002 to 2015 data from the National Survey on Drug Use and Health and the Centers for Disease Control and Prevention. Interventions: Comparison of interventions that would lower the incidence of prescription opioid misuse from 2016 to 2025 based on historical trends (a 7.5% reduction per year) and 50% faster than historical trends (an 11.3% reduction per year), vs a circumstance in which the incidence of misuse remained constant after 2015. Main Outcomes and Measures: Opioid overdose deaths from prescription and illicit opioids from 2016 to 2025 under each intervention. Results: Under the status quo, the annual number of opioid overdose deaths is projected to increase from 33 100 in 2015 to 81 700 (95% uncertainty interval [UI], 63 600-101 700) in 2025 (a 147% increase from 2015). From 2016 to 2025, 700 400 (95% UI, 590 200-817 100) individuals in the United States are projected to die from opioid overdose, with 80% of the deaths attributable to illicit opioids. The number of individuals using illicit opioids is projected to increase by 61%-from 0.93 million (95% UI, 0.83-1.03 million) in 2015 to 1.50 million (95% UI, 0.98-2.22 million) by 2025. Across all interventions tested, further lowering the incidence of prescription opioid misuse from 2015 levels is projected to decrease overdose deaths by only 3.0% to 5.3%. Conclusions and Relevance: This study's findings suggest that interventions targeting prescription opioid misuse such as prescription monitoring programs may have a modest effect, at best, on the number of opioid overdose deaths in the near future. Additional policy interventions are urgently needed to change the course of the epidemic.

Testing for Verification Bias in Reported Malignancy Risks for Side-Branch Intraductal Papillary Mucinous Neoplasms: A Simulation Modeling Approach.

OBJECTIVE: The objective of our study was to test for the possibility that published malignancy risks for side-branch intraductal papillary mucinous neoplasms (IPMNs) are overestimates, likely due to verification bias. MATERIALS AND METHODS: We tested for possible verification bias using simulation modeling techniques. First, in age-defined hypothetical cohorts of 10 million persons, we projected the frequency of pancreatic ductal adenocarcinoma (PDAC) arising from side-branch IPMNs over 5 years using published estimates of their prevalence (4.4%) and rate of malignant transformation (1.9%). Second, we projected the total number of PDAC cases in corresponding cohorts over the same time horizon using national cancer registry data. For each cohort, we determined whether the percentage of all PDAC cases that arose from side-branch IPMNs (i.e., side-branch IPMN-associated PDAC cases) was clinically plausible using an upper limit of 10% to define plausibility, as estimated from the literature. Model assumptions and parameter uncertainty were evaluated in sensitivity analysis. RESULTS: Across all cohorts, percentages of side-branch IPMN-associated PDACs greatly exceeded 10%. In the base case (mean age = 55.7 years), 80% of PDAC cases arose from side-branch IPMNs (7877/9786). In the oldest cohort evaluated (mean age = 75 years), this estimate was 76% (14,227/18,714). In a secondary analysis, we found that if an upper limit threshold of 10% for side-branch IPMN-associated PDAC was imposed, the model-predicted rate of malignancy for side-branch IPMNs would be less than 0.24% over a 5-year time horizon, substantially lower than most literature-based estimates. CONCLUSION: Our results suggest that reported malignancy risks associated with side-branch IPMNs are likely to be overestimates and imply the presence of verification bias.

Effect of PD-L1 testing on the cost-effectiveness and budget impact of pembrolizumab for advanced urothelial carcinoma of the bladder in the United States.

PURPOSE: Our purpose was to evaluate the effect of PD-L1 testing on the cost-effectiveness of pembrolizumab for second-line treatment of advanced urothelial carcinoma in the bladder from the U.S. societal perspective. MATERIALS AND METHODS: We developed a microsimulation model to compare 3 treatment strategies: (1) treat all patients with standard-of-care chemotherapy, (2) treat all patients with pembrolizumab, and (3) treat patients with PD-L1-positive tumors at a ≥1% expression threshold with pembrolizumab, and all others with standard-of-care chemotherapy. Additionally, we performed a budget impact analysis based on the projected number of urothelial carcinoma patients eligible for second-line pembrolizumab treatment. RESULTS: Treating all patients with chemotherapy resulted in a mean cost of $17,232 and mean effect of 0.43 quality-adjusted life-years. The PD-L1 test strategy was the most efficient strategy, with an incremental cost-effectiveness ratio of $122,933/quality-adjusted life-year. Treating all patients with pembrolizumab resulted in an incremental cost-effectiveness ratio of $197,383/quality-adjusted life-year compared to the PD-L1 test strategy. The PD-L1 test strategy would produce an incremental budget impact of $14.9 million in the first year of use compared to chemotherapy, increasing to $16.5 million in the fifth year of use. Treating all patients with pembrolizumab would produce an incremental budget impact of $19.6 million compared to the PD-L1 test strategy in its first year of use, increasing to $20.9 million by year 5. CONCLUSIONS: Pembrolizumab was not cost-effective in either strategy based on a $100,000/quality-adjusted life-year willingness-to-pay threshold. Using PD-L1 testing to select for patients who may have better associated outcomes may improve the affordability of pembrolizumab.

Progression to pancreatic ductal adenocarcinoma from pancreatic intraepithelial neoplasia: Results of a simulation model.

OBJECTIVES: To gain insight into the natural history and carcinogenesis pathway of Pancreatic Intraepithelial Neoplasia (PanIN) lesions by building a calibrated simulation model of PanIN progression to pancreatic ductal adenocarcinoma (PDAC) METHODS: We revised a previously validated simulation model of solid PDAC, calibrating the model to fit data from the National Cancer Institute's Surveillance, Epidemiology, and End Results program and published literature on PanIN prevalence by age. We estimated the likelihood of progression from PanIN states (1, 2, and 3) to PDAC and the time between PanIN onset and PDAC (dwell time). We evaluated a hypothetical intervention to test for and treat PanIN 3 lesions to estimate the potential benefits from PanIN detection. RESULTS: We estimated the lifetime probability of progressing from PanIN 1 to PDAC to be 1.5% (men), 1.3% (women). Progression from PanIN 1 to PDAC took 33.6 years and 35.3 years, respectively, and from PanIN 3 to PDAC took 11.3 years and 12.3 years. A hypothetical test for PanIN 3 detection and treatment could provide a maximum, average life expectancy gain of 40 days. CONCLUSIONS: Our modeling analysis estimates PanINs have a relatively indolent course to PDAC, supporting the feasibility of potential future early detection strategies.

Patient Navigation to Improve Cancer Screening in Underserved Populations: Reported Experiences, Opportunities, and Challenges.

Our goal is to define patient navigation for an imaging audience, present a focused selection of published experiences with navigation programs for breast and colorectal cancer screening, and expose principal barriers to the success of such programs. Despite numerous advances in the early detection of cancers, many patients still present with advanced disease. A disproportionate number are low-income minority patients who experience worse health outcomes than their white or more financially stable counterparts. Patient navigation, which aims to assist the medically underserved by overcoming specific barriers to care, may represent one solution to narrowing disparities. Related research suggests that in general, patient navigation programs that have addressed breast or colorectal cancer screening have been successful in improving screening rates and timeliness of follow-up care. However, although beneficial, navigation is expensive and may present an unmanageable financial burden for many health care centers. To overcome this challenge, navigation efforts will likely need to target those patients that are most likely to benefit. Further research to identify such patients will be critically important for improving the sustainability of navigation programs, and, in turn, for realizing the benefits of such programs in reducing cancer disparities.

Modeling treatment outcomes for patients with advanced ovarian cancer: Projected benefits of a test to optimize treatment selection.

OBJECTIVE: For patients with advanced stage epithelial ovarian cancer (EOC), substantial emphasis has been placed on diagnostic tests that can discern which of two treatment options - primary cytoreductive surgery (PCS) or neoadjuvant chemotherapy followed by interval cytoreductive surgery (NACT+ICS) - optimizes patient-level outcomes. Our goal was to project potential life expectancy (LE) gains that could be achieved by use of such a test. METHODS: We developed a microsimulation model to project LE for patients with stage IIIC EOC. We compared: a "standard-of-care" strategy, in which patients were triaged to PCS vs. NACT+ICS based on current clinical practice; and a "test" strategy, in which patients were triaged based on results of a hypothetical test. We identified those test performance characteristics for which the test strategy outperformed the standard-of-care strategy, from a LE standpoint. Effects of parameter uncertainty were evaluated in sensitivity analysis. RESULTS: Even with a perfect test, the LE gain was modest (LE with test vs. standard-of-care strategy=67.6 vs. 66.4months; LE gain=1.2months). In order to outperform the standard-of-care, the test had to have a high probability of correctly identifying "resectable" patients at PCS (i.e. those for whom complete or optimal cytoreduction would be possible); this test property was more important than correct triage of unresectable patients to NACT+ICS. Results were sensitive to the proportion of patients whose underlying disease was resectable at PCS. CONCLUSION: Diagnostic tests that are designed to triage patients with advanced stage EOC will likely have only a modest effect on LE.

Imaging Follow-up of Low-Risk Incidental Pancreas and Kidney Findings: Effects of Patient Age and Comorbidity on Projected Life Expectancy.

Purpose To determine the effects of patient age and comorbidity level on life expectancy (LE) benefits associated with imaging follow-up of Bosniak IIF renal cysts and pancreatic side-branch (SB) intraductal papillary mucinous neoplasms (IPMNs). Materials and Methods A decision-analytic Markov model to evaluate LE benefits was developed. Hypothetical cohorts with varied age (60-80 years) and comorbidities (none, mild, moderate, or severe) were evaluated. For each finding, LE projections from two strategies were compared: imaging follow-up and no imaging follow-up. Under follow-up, it was assumed that cancers associated with the incidental finding were successfully treated before they spread. For patients without follow-up, mortality risks from Bosniak IIF cysts (renal cell carcinoma) and SBIPMNs (pancreatic ductal adenocarcinoma) were incorporated. Model assumptions and parameter uncertainty were evaluated in sensitivity analysis. Results In the youngest, healthiest cohorts (age, 60 years; no comorbidities), projected LE benefits from follow-up were as follows: Bosniak IIF cyst, 6.5 months (women) and 5.8 months (men); SBIPMN, 6.4 months (women) and 5.3 months (men). Follow-up of Bosniak IIF cysts in 60-year-old women with severe comorbidities yielded a LE benefit of 3.9 months; in 80-year-old women with no comorbidities, the benefit was 2.8 months, and with severe comorbidities the benefit was 1.5 months. Similar trends were observed in men and for SBIPMN. Results were sensitive to the performance of follow-up for cancer detection; malignancy risks; and stage at presentation of malignant, unfollowed Bosniak IIF cysts. Conclusion With progression of age and comorbidity level, follow-up of low-risk incidental findings yields increasingly limited benefits for patients. © RSNA, 2018 Online supplemental material is available for this article.