Optimizing β-lactam antibiotic exposure in critically ill patients with hospital-acquired pneumonia (HAP) remains a challenge due to significant pharmacokinetic variability, particularly in the setting of renal dysfunction and replacement therapies. Continuous infusion (CI) of piperacillin/tazobactam aims to improve pharmacodynamic target attainment, though both subtherapeutic and potentially toxic concentrations have been reported in practice. We developed a population pharmacokinetic model of piperacillin using 162 plasma samples from 35 intensive care unit (ICU) patients with HAP, including those receiving continuous renal replacement therapy (CRRT). Piperacillin concentrations were quantified using a validated LC-MS method. A one-compartment model parameterized with renal and non-renal clearance was implemented in Monolix, incorporating creatinine clearance (CrCL), CRRT effluent flow rate, and intermittent hemodialysis as key covariates. Monte Carlo simulations in Simulx evaluated steady-state drug exposures following renal dose-adjusted CI regimens. Simulation showed that renally adjusted lower doses administered via CI (3-9 g/day) achieved target concentrations in 74-82% of patients with CrCL ≤75 mL/min. Higher doses (6-12 g/day) resulted in >20% of patients exceeding 96 mg/L across all renal strata. Among CRRT patients, lower doses provided a 100% probability of maintaining targeted piperacillin concentrations. In patients with supra-normal renal function (i.e., CrCL = 150 mL/min), low-dose CI regimens yielded a 6.1% probability of underexposure, compared to 2.7% with high-dose. CI PIP dosing based on CrCL results in variable exposures among ICU patients. Individualized dosing of PIP may be required to optimize efficacy and minimize toxicity in ICU patients treated with CI dosing.

Abstract Background The impact of extracorporeal membrane oxygenation (EMCO) on cefepime (FEP) pharmacokinetics (PK) in the absence of renal replacement therapy (RRT) is unclear. Figure 1. Observed versus predicted cefepime plasma concentrations in the population (A) and for each individual (B) Methods We evaluated FEP PK in patients with hospital-acquired pneumonia from June 2018 to May 2023, with/without ECMO in the absence of RRT, in a single-center study nested within the Successful Clinical Response In Pneumonia Treatment (SCRIPT) study. Patient data were extracted from electronic health records. FEP dosing was according to institutional protocols based on renal function and indication. Plasma concentrations were quantified by validated LC-MS/MS assay. We used Pmetrics 2.1.1 for R to estimate population model parameter values [e.g., volume (V) and clearance (CL)] and individual target attainment rates, assuming a free (f) fraction of 80%. Targets of 100% fT>1xMIC and 100% fT>4xMIC were evaluated vs. the susceptible breakpoint MICs of Enterobacterales (2 mcg/mL) and P. aeruginosa (8 mcg/mL) and stratified by concurrent ECMO. Figure 2. Pharmacokinetic parameters (CL and V) standardized to body weight (kg) stratified by ECMO status. Results Sixty-two SCRIPT patients (63% male) contributed 95 plasma FEP samples (1-13 per patient). A total of 9 patients (one female) required ECMO. Patients had a mean±SD CRCL of 123±84 mL/min and a mean±SD body weight of 86±24 kg. The median/max first 24-hr FEP dose was 4/8 grams. A two-compartment PK model fitted best (Fig 1). Body weight and creatinine clearance were related to V and CL, respectively (Table 1). ECMO was not significantly associated with FEP CL (P=0.2) but was associated with V (P< 0.005) (Fig 2). ECMO patients had a median 3.5-fold greater Vd vs. non-ECMO patients. For targets of fT >1xMIC and fT >4xMIC, median individual plasma attainment rates were 100% and 100% against Enterobacterales and 100% and 49% against P. aeruginosa, respectively. Attainment rates were lowest in patients requiring ECMO vs. P. aeruginosa at a target of 100% fT >4xMIC, (Fig 3). Figure 3. Individual predicted cefepime PK/PD target attainment stratified by organism breakpoint and concurrent ECMO status for fT>1xMIC and fT>4xMIC. Conclusion We found that ECMO increased FEP V but not CL suggesting that ECMO patients may require loading doses and prolonged infusions to improve target attainment. Protocolized short infusions of FEP in ECMO patients resulted in suboptimal target attainment for a PK/PD goal of 100% fT>4xMIC against P. aeruginosa. More work is needed to identify the impact of FEP target attainment on clinical outcomes in populations such as these. Table 1. Population Median Parameters and 95% Credible Intervals for Cefepime Patients with and without the Requirement of Extracorporeal Membrane Oxygenation Support Disclosures Marc H. Scheetz, PharmD, MSc, Abbvie: Advisor/Consultant|Basilea: Advisor/Consultant|Cidara: Advisor/Consultant|DoseMe: Advisor/Consultant|Entasis: Advisor/Consultant|F2G: Advisor/Consultant|GSK: Advisor/Consultant|Lykos: Advisor/Consultant|Roche: Advisor/Consultant|Third Pole Therapeutics: Advisor/Consultant|Xelia: Advisor/Consultant Nathaniel J. Rhodes, PharmD MS, Apothecademy, LLC: Advisor/Consultant

ABSTRACT It is unclear whether plasma is a reliable surrogate for target attainment in the epithelial lining fluid (ELF). The objective of this study was to characterize meropenem target attainment in plasma and ELF using prospective samples. The first 24-hour T >MIC was evaluated vs 1xMIC and 4xMIC targets at the patient (i.e., fixed MIC of 2 mg/L) and population [i.e., cumulative fraction of response (CFR) according to EUCAST MIC distributions] levels for both plasma and ELF. Among 65 patients receiving ≥24 hours of treatment, 40% of patients failed to achieve >50% T >4xMIC in plasma and ELF, and 30% of patients who achieved >50% T >4xMIC in plasma had <50% T >4xMIC in ELF. At 1xMIC and 4xMIC targets, 3% and 25% of patients with >95% T >MIC in plasma had <50% T >MIC in ELF, respectively. Those with a CRCL >115 mL/min were less likely to achieve >50%T >4xMIC in ELF ( P < 0.025). In the population, CFR for Escherichia coli at 1xMIC and 4xMIC was >97%. For Pseudomonas aeruginosa , CFR was ≥90% in plasma and ranged 80%–85% in ELF at 1xMIC when a loading dose was applied. CFR was reduced in plasma (range: 75%–81%) and ELF (range: 44%–60%) in the absence of a loading dose at 1xMIC. At 4xMIC, CFR for P. aeruginosa was 60%–86% with a loading dose and 18%–62% without a loading dose. We found that plasma overestimated ELF target attainment inup to 30% of meropenem-treated patients, CRCL >115 mL/min decreased target attainment in ELF, and loading doses increased CFR in the population.

Abstract Background Whether pharmacokinetic (PK)/pharmacodynamic (PD) target attainment for meropenem measured in plasma is a clinically reliable surrogate for target attainment in the epithelial lining fluid (ELF) in pneumonia is unclear. The objective of the current study was to characterize target attainment in plasma and ELF using prospectively collected samples and to define predictors of optimal target attainment. Methods Individual Bayesian meropenem plasma and ELF profiles were generated using Pmetrics for the first 24 hours of treatment. First 24-hr PK/PD target attainment was evaluated vs. PK/PD goals of 100% T>1xMIC and 100% T>4xMIC for a susceptible MIC of 2 mg/L. Results Sixty-seven patients contributed ELF and plasma PK data (range: 1-3 samples/patient for each matrix). Mean population CL, Vplasma, and VELF estimates were 6.2 L/hr, 45.7, and 28.3 L. For the 100% T>4xMIC goal the following target attainment groups were identified: optimal (≥95%) in plasma and ELF (n=27/67, 40%), near-optimal ( >50% and < 95%) in both plasma and ELF (n=11/67, 16.4%), suboptimal (< 50%) in ELF only (n=15/67, 22.4%), suboptimal (< 50%) in plasma only (n=1/67, 1.5%), and suboptimal (< 50%) in both ELF and plasma (n=13, 19.4%). Loading doses significantly improved the likelihood of optimal vs. suboptimal target attainment in ELF and plasma (96.2% vs. 14.3%; p< 0.0001) but did not reliably discriminate other groups. Individual observed meropenem (ordinate) and model predicted meropenem (abscissa) concentrations in plasma (A) and ELF (B) in patients with pneumonia.Panel A.) grouping according to a 1xMIC target. A total of 53 patients had optimal (green) attainment in both plasma and ELF, 9 patients had near optimal attainment in plasma and ELF (gold), 3 had suboptimal attainment in ELF only (red), and 2 had suboptimal attainment in both plasma and ELF (purple). Panel B.) grouping according to a 4xMIC target. A total of 27 patients had optimal (green) attainment in both plasma and ELF, 11 patients had near optimal attainment in plasma and ELF (gold), 1 patient had suboptimal attainment in plasma only (blue), 15 had suboptimal attainment in ELF only (red), and 13 had suboptimal attainment in both plasma and ELF (purple). Conclusion Optimal target attainment was achieved in only 40% of patients while nearly 20% of patients had suboptimal attainment in both plasma and ELF. Use of plasma as a surrogate for ELF would miss up to 20% of patients with suboptimal ELF attainment, suggesting that therapeutic monitoring of ELF may be required in some patients to optimize PK/PD attainment. Disclosures Marc H. Scheetz, PharmD, MSc, Abbvie: Advisor/Consultant|ASHP: Honoraria|Chambless, Higdon, Richardson, Katz & Griggs, LLP: Expert Testimony|Cidara: Advisor/Consultant|Entasis: Advisor/Consultant|F2G: Advisor/Consultant|GSK: Advisor/Consultant|Guidepoint Global: Honoraria|Hall, Booth, Smith, P.C.: Expert Testimony|Merck: Advisor/Consultant|Reminger Co., L.P.A: Expert Testimony|Spero: Advisor/Consultant|Takeda: Advisor/Consultant|Taylor, English, Duma, LLP: Expert Testimony|Third Pole Therapeutics: Advisor/Consultant Michael N. Neely, MD, Astellas Pharma Global Development, Inc.: Advisor/Consultant|Astellas Pharma Global Development, Inc.: Support for the present publication Richard G. Wunderink, MD, bioMerieux: Honoraria|Kariius: Clinical Evaluation Committee|LaJolla: Advisor/Consultant|Pfizer, Inc: Clinical Evaluation Committee|Shionogi: Advisor/Consultant Nathaniel J. Rhodes, PharmD MS, Third Pole Therapeutics: Advisor/Consultant

Logistic regression is a statistical tool of paramount significance in the field of epidemiology1 and ranks as one of the most frequently published multivariable analyses for designs involving a single binary dependent variable and one or more independent variables in the fields of public health2,3 and medical4 research.

Methicillin-resistant Staphylococcus aureus (MRSA) is an uncommon but serious cause of community-acquired pneumonia (CAP). A lack of validated MRSA CAP risk factors can result in overuse of empirical broad-spectrum antibiotics. Methicillin-resistant Staphylococcus aureus (MRSA) is an uncommon but serious cause of community-acquired pneumonia (CAP). A lack of validated MRSA CAP risk factors can result in overuse of empirical broad-spectrum antibiotics. We sought to develop robust models predicting the risk of MRSA CAP using machine learning using a population-based sample of hospitalized patients with CAP admitted to either a tertiary academic center or a community teaching hospital. Data were evaluated using a machine learning approach. Cases were CAP patients with MRSA isolated from blood or respiratory cultures within 72 h of admission; controls did not have MRSA CAP. The Classification Tree Analysis algorithm was used for model development. Model predictions were evaluated in sensitivity analyses. A total of 21 of 1,823 patients (1.2%) developed MRSA within 72 h of admission. MRSA risk was higher among patients admitted to the intensive care unit (ICU) in the first 24 h who required mechanical ventilation than among ICU patients who did not require ventilatory support (odds ratio [OR], 8.3; 95% confidence interval [CI], 2.4 to 32). MRSA risk was lower among patients admitted to ward units than among those admitted to the ICU (OR, 0.21; 95% CI, 0.07 to 0.56) and lower among ICU patients without a history of antibiotic use in the last 90 days than among ICU patients with antibiotic use in the last 90 days (OR, 0.03; 95% CI, 0.002 to 0.59). The final machine learning model was highly accurate (receiver operating characteristic [ROC] area = 0.775) in training and jackknife validity analyses. We identified a relatively simple machine learning model that predicted MRSA risk in hospitalized patients with CAP within 72 h postadmission.

BackgroundAdverse drug/device reactions (ADRs) can result in severe patient harm. We define very serious ADRs as being associated with severe toxicity, as measured on the Common Toxicity Criteria Adverse Events (CTCAE)) scale, following use of drugs or devices with large sales, large financial settlements, and large numbers of injured persons. We report on impacts on patients, clinicians, and manufacturers following very serious ADR reporting.MethodsWe reviewed clinician identified very serious ADRs published between 1997 and 2019. Drugs and devices associated with reports of very serious ADRs were identified. Included drugs or devices had market removal discussed at Food and Drug Advisory (FDA) Advisory Committee meetings, were published by clinicians, had sales > $1 billion, were associated with CTCAE Grade 4 or 5 toxicity effects, and had either >$1 billion in settlements or >1,000 injured patients. Data sources included journals, Congressional transcripts, and news reports. We reviewed data on: 1) timing of ADR reports, Boxed warnings, and product withdrawals, and 2) patient, clinician, and manufacturer impacts. Binomial analysis was used to compare sales pre- and post-FDA Advisory Committee meetings.FindingsTwenty very serious ADRs involved fifteen drugs and one device. Legal settlements totaled $38.4 billion for 753,900 injured persons. Eleven of 18 clinicians (61%) reported harms, including verbal threats from manufacturer (five) and loss of a faculty position (one). Annual sales decreased 94% from $29.1 billion pre-FDA meeting to $4.9 billion afterwards (p<0.0018). Manufacturers of four drugs paid $1.7 billion total in criminal fines for failing to inform the FDA and physicians about very serious ADRs. Following FDA approval, the median time to ADR reporting was 7.5 years (Interquartile range 3,13 years). Twelve drugs received Box warnings and one drug received a warning (median, 7.5 years following ADR reporting (IQR 5,11 years). Six drugs and 1 device were withdrawn from marketing (median, 5 years after ADR reporting (IQR 4,6 years)).InterpretationBecause very serious ADRs impacts are so large, policy makers should consider developing independently funded pharmacovigilance centers of excellence to assist with clinician investigations.FundingThis work received support from the National Cancer Institute (1R01 CA102713 (CLB), https://www.nih.gov/about-nih/what-we-do/nih-almanac/national-cancer-institute-nci; and two Pilot Project grants from the American Cancer Society's Institutional Grant Award to the University of South Carolina (IRG-13–043–01) https://www.cancer.org/ (SH; BS).

Objective: To survey potential users and pilot a 24/7 concierge pharmacy service within a private practice and quantify the number of clinically actionable examples of pharmacogenomics in an ambulatory care population.Methods: Eligible subjects included those 181 years old and patientsofsupportingphysiciansattheclinic.Afterconsenting, subjects completed a survey regarding pharmacy services and willingness to pay.Pharmacogenomic testing (n534) included collecting DNA from buccal swabs and genotyping by RT-PCR.A subset of participants (n56) piloted a free 24/7 concierge pharmacy service for one month.Results: Pre-survey findings highlighted that the majority of subjects would utilize an in-person pharmacist service to review medications 4 times a year (mean score58.6 out of 10).The average score dropped to 4.6/10 if insurance did not cover costs.Use of an on-call pharmacy service as well as if insurance covered 50% of total costs averaged 5/10.Nonetheless, 23 subjects (67%) indicated willingness to pay $25-$50 per month for individual pharmacy services, although 6 subjects (18%) would not pay anything out of pocket.In the concierge service, two subjects (33%) utilized the on-call pharmacist.Preliminary pharmacogenomic results identified a range of 6-34% of subjects who have the potential for clinical changes based on ultra-rapid or poor metabolism.Implications: Subjects prefer to sit down with a pharmacist for medication services, and the primary barrier is likely cost.Commercial pharmacogenomic testing is currently underway for the pilot, and these results will be reported in charts and used for clinical decision-making as warranted.

To address appropriateness of antibiotic use, we implemented an electronic framework to evaluate antibiotic “never events” (NEs) at 2 medical centers. Patient-level vancomycin administration records were classified as NEs or non-NEs. The objective framework allowed capture of true-positive vancomycin NEs in one-third of patients identified by the electronic strategy.

N.K.Y., P.C., & P.R.Y. contributed equally to this study Introduction: Many studies have concluded that active cancer patients infected with SARS-CoV-2 have a more complicated infection course and worse outcomes compared to the general patient population hospitalized with COVID-19. However, little evidence exists whether having a history of cancer plays a significant role in these observations. Patients with hematologic malignancy (HM) might have worse prognosis among all cancer patients but the reason remains unclear. Our objective is to evaluate outcomes and severity of COVID-19 in patients with Hematological Malignancy (HM) versus Solid-tumors (ST) in different clinical settings and also compare these outcomes within the group of patients with hematological malignancies. Methods: This retrospective study examines risk factors and outcomes of COVID-19 in patients with a history of cancer and laboratory-confirmed COVID-19 diagnosis between March 1 st, 2020, and December 31 st, 2020, at Rush University Medical Center, one of the largest COVID-19 tertiary care hospitals in Chicago. Baseline characteristics, malignancy type and types of cancer treatment within the last 30 days were recorded. Measures of COVID-19 severity included hospital admission versus outpatient care, use of oxygen, intensive care unit (ICU) admission, and mechanical ventilation. The primary outcome was death. Statistical analysis was conducted using optimal discriminant analysis, a non-parametric exact machine-learning algorithm which identifies the relationship between independent and dependent variables that maximizes model predictive accuracy adjusted to remove the effect of chance. Analysis was performed separately for each attribute using the entire sample (“training” analysis), then one-sample jackknife analysis was conducted to estimate cross-generalizability of findings using the model to classify an independent random sample. Results: 378 total patients with a history of cancer tested positive for COVID-19 within the time frame of the study. Of these, 294 (78%) patients had ST malignancy and 84 (22%) patients had HM. Characteristics and outcomes are summarized in Table 1. ST patients were marginally older than HM patients (p<0.025). A significantly greater proportion of HM patients were male (p<0.0023). HM and ST patients did not differ with respect to percentage receiving active cancer treatment (p<0.81). Compared to ST patients, more HM patients had received corticosteroids in the 30 days prior to COVID-19 diagnosis (p<0.017), had higher rates of hospitalization (p<0.0013) and ICU requirement (p<0.0001) with a significantly longer length of ICU stay (p<0.0036). Compared to ST patients, HM patients also required oxygen (p<0.002) and mechanical ventilation (p<0.0005) more often and had a 3.88-fold statistically higher death rate (OR 3.88 [95% CI 1.62-9.29] p<0.003). Patients with HM are categorized by disease subtype and summarized in Table 2. The case fatality rate from COVID-19 was 33.3% for patients with myeloproliferative neoplasms/myelodysplastic syndromes (MPN/MDS), 21.4% for patients with chronic lymphocytic leukemia (CLL), 13.6% for patients with non-Hodgkin lymphoma, 10.5% for patients with plasma cell neoplasms, and 4.5% for patients with acute leukemia. When looking at outcomes, CLL had the highest percentage of patients requiring hospital admission, oxygen, and ICU admission, and MPN/MDS had the highest percentage of patients requiring mechanical ventilation. Conclusions: Patients with hematologic malignancies had more severe COVID-19 illness and hospitalization rates and a 3.88-fold higher rate of death than patients with solid tumors. The comparable proportion of patients on anti-cancer therapy despite differences in survival suggests that being on anti-cancer therapy is less important than the underlying diagnosis of HM versus ST as a determinant of poor outcomes. Clinicians should closely monitor and initiate early COVID-19 treatments for all patients with HM and COVID-19. Because HM are highly heterogenous group of cancers, it is important to look at subtypes in greater detail. Numerous patient-level, disease-specific, and therapy-related factors may impact outcomes of COVID-19 among patients with HM, and we are currently analyzing additional data to better understand the factors which make this disease group more susceptible to severe infection. Figure 1 Disclosures Kuzel:  Sanofi-Genzyme Genomic Health Tempus laboratories Bristol Meyers Squibb: Honoraria; Genomic Health: Membership on an entity's Board of Directors or advisory committees; Exelixis: Membership on an entity's Board of Directors or advisory committees; Cardinal Health: Membership on an entity's Board of Directors or advisory committees; Abbvie: Other; Curio Science: Membership on an entity's Board of Directors or advisory committees; AmerisourceBergen Corp: Membership on an entity's Board of Directors or advisory committees; CVS: Membership on an entity's Board of Directors or advisory committees; Tempus Laboratories: Membership on an entity's Board of Directors or advisory committees; Bristol Meyers Squibb: Membership on an entity's Board of Directors or advisory committees; Merck: Other: Data Monitoring Committee Membership; Amgen: Other: Data Monitoring Committee Membership; SeaGen: Other: Data Monitoring Committee Membership; Medpace: Other: Data Monitoring Committee Membership.

Dalton et al. recently wrote that classification and regression tree (CART) models increase the type 1 error rate [[1]Dalton B.R. Krishnan A. Stewart J.J. Jorgensen S.C. Limitations of classification and regression tree analysis in vancomycin exposure-response relationship studies: insights from data simulation.Clin Microbiol Infect. 2021; https://doi.org/10.1016/j.cmi.2021.07.028Abstract Full Text Full Text PDF Scopus (1) Google Scholar]. They question the validity of studies cited within the updated vancomycin therapeutic monitoring guidelines [[2]Rybak M.J. Le J. Lodise T.P. Levine D.P. Bradley J.S. Liu C. et al.Therapeutic monitoring of vancomycin for serious methicillin-resistant Staphylococcus aureus infections: a revised consensus guideline and review by the American Society of Health-System Pharmacists, the Infectious Diseases Society of America, the Pediatric Infectious Diseases Society, and the Society of Infectious Diseases Pharmacists.Am J Health Syst Pharm. 2020; 77: 835-864PubMed Google Scholar], which report that greater exposures (vis-à-vis area under the curve) are associated with different clinical outcomes, and they suggest that classification rules derived using CART models are invalid [[1]Dalton B.R. Krishnan A. Stewart J.J. Jorgensen S.C. Limitations of classification and regression tree analysis in vancomycin exposure-response relationship studies: insights from data simulation.Clin Microbiol Infect. 2021; https://doi.org/10.1016/j.cmi.2021.07.028Abstract Full Text Full Text PDF Scopus (1) Google Scholar]. Using simulation to demonstrate the risk of type 1 error with CART analysis, they generated 30 data sets of randomly sampled areas under the curve and randomly assigned outcome classes (e.g. death or survival) and analyzed the results using CART and linear regression [[1]Dalton B.R. Krishnan A. Stewart J.J. Jorgensen S.C. Limitations of classification and regression tree analysis in vancomycin exposure-response relationship studies: insights from data simulation.Clin Microbiol Infect. 2021; https://doi.org/10.1016/j.cmi.2021.07.028Abstract Full Text Full Text PDF Scopus (1) Google Scholar]. Herein we describe the serious mathematical and methodological flaws in the work of Dalton et al. [[1]Dalton B.R. Krishnan A. Stewart J.J. Jorgensen S.C. Limitations of classification and regression tree analysis in vancomycin exposure-response relationship studies: insights from data simulation.Clin Microbiol Infect. 2021; https://doi.org/10.1016/j.cmi.2021.07.028Abstract Full Text Full Text PDF Scopus (1) Google Scholar], and we offer an alternative data-driven interpretation of the existing literature. First, Dalton et al. incorrectly state that type 1 error is increased with CART because of an iterative search across independent variables for the threshold value, which best classifies the sample according to a statistical criterion. Their supposition is incorrect because type 1 error increases with each hypothesis test, not with each analytical step. The result of any specific analysis (machine learning or otherwise), and the number of steps needed to satisfy a defined objective criterion for that analysis, are clearly distinct entities. That is, each completed specific analysis constitutes a separate, singular hypothesis test. Second, Dalton et al. simulated sample sizes of 50, 100 and 200, but they failed to provide an a priori assessment of the sample size needed to detect a given effect or to specify an a priori target effect size—which together constitute statistical power analysis, a crucial antecedent for testing any hypothesis. Statistical power analysis is mandatory in obtaining institutional review board approval for conducting research studies and for obtaining research funding. Third, Dalton et al. incorrectly concluded that 50% of their hypothesis tests were statistically significant, each having per-comparison p < 0.05. This conclusion is incorrect because they failed to control the type 1 error rate for their study. That is, for their study involving n = 30 analyses the chance of making a type 1 error at an a priori per-comparison α of 0.05 is 1 – (1 – 0.05)30 = 78%. Current statistical multiple comparisons methodology requires a Bonferroni-type α adjustment for the specific number of comparisons made. In the case of 30 comparisons, the adjusted threshold p value is 1 – (1 – 0.05)(1/30) = 0.0017. When this threshold is applied to their simulations not a single comparison is significant, which is expected for randomly generated data. Fourth, Dalton et al. failed to provide a transparent methodology for their simulation analysis because the seed number and area under the curve distribution they generated were undefined—which makes the work unreproducible by definition. To ensure reproducibility across analytical platforms and laboratories, it is essential to specify the seed number used in analyses involving random sampling. For example, we have shown that the degree of overlap between the distribution of resampled predictions based on a given model attribute versus the resampled class data after Fisher randomization (i.e. class scrambling) varies according to the random seed number used [[3]Rhodes N.J. Assessing reproducibility of novometric bootstrap confidence interval analysis using multiple seed numbers (invited).Optimal Data Anal. 2020; 9: 190-194Google Scholar]. Fifth, Dalton et al. failed to evaluate the validity of their own models. Validity could have been assessed using hold-out (train-test), jackknife (e.g. leave-one-out) or bootstrap-resampling. Yet none of these were attempted. The importance of ensuring model validity cannot be understated. Linear and non-linear multivariate methods and machine learning algorithms have been shown to find effects in random data, whereas methods such as classification tree analysis have not [[4]Linden A. Bryant F.B. Yarnold P.R. Logistic discriminant analysis and structural equation modeling both identify effects in random data.Optimal Data Anal. 2019; 8: 97-102Google Scholar,[5]Linden A. Yarnold P.R. Multi-layer perceptron neural net model identifies effect in random data.Optimal Data Anal. 2019; 8: 94-96Google Scholar]. Classification tree analysis is a tree-based machine-learning algorithm that maximizes predictive accuracy as the objective function [[6]Yarnold P.R. Soltysik R.C. Maximizing predictive accuracy. ODA Books, Chicago, IL2016Google Scholar]. Use of validity assessments (e.g. using leave-one-out jackknife) can help to prevent the selection of trees with chaotic instability. Hence, evaluation of model validity is essential to avoid selecting sub-optimal solutions and accepting false-positive results. When interpreting any study, including those cited by the vancomycin guidelines, one must consider the population, intervention and comparator groups, outcome and time context. Prospective trials have demonstrated the existence of a vancomycin exposure–toxicity relationship [[7]Lodise T.P. Rosenkranz S.L. Finnemeyer M. Evans S. Sims M. Zervos M.J. et al.The emperor's new clothes: PRospective observational evaluation of the association between initial VancomycIn exposure and failure rates among ADult HospitalizEd patients with methicillin-resistant Staphylococcus aureus bloodstream infections (PROVIDE).Clin Infect Dis. 2020; 70: 1536-1545Crossref PubMed Scopus (42) Google Scholar]. Rigorous studies have found that higher exposures are associated with a greater risk of acute kidney injury [[8]Scheetz M.H. Pais G. Lodise T.P. Tong S.Y.C. Davis J.S. O'Donnell J.N. et al.Of rats and men, a translational model to understand vancomycin pharmacokinetic/toxicodynamic relationships.Antimicrob Agents Chemother. 2021; https://doi.org/10.1128/AAC.01060-21Crossref Scopus (0) Google Scholar], and contemporary clinical evidence confirms that vancomycin-induced acute kidney injury is associated with a greater risk of mortality [[9]Poston-Blahnik A. Moenster R. Association between vancomycin area under the curve and nephrotoxicity: a single center, retrospective cohort study in a veteran population.Open Forum Infect Dis. 2021; 8https://doi.org/10.1093/ofid/ofab094Crossref PubMed Google Scholar]. Given the evolving nature of vancomycin use over time, it is unsurprising that various vancomycin exposure–response thresholds have been identified because the populations (e.g. methicillin-resistant Staphylococcus aureus infective endocarditis, bloodstream infection or pneumonia), interventions (monotherapy or combination therapy [[10]Tong S.Y.C. Lye D.C. Yahav D. Sud A. Robinson J.O. Nelson J. et al.Australasian Society for Infectious Diseases Clinical Research, Effect of vancomycin or daptomycin with vs without an antistaphylococcal beta-lactam on mortality, bacteremia, relapse, or treatment failure in patients with MRSA bacteremia: a randomized clinical trial.JAMA. 2020; 323: 527-537Crossref PubMed Scopus (62) Google Scholar]), outcome measures (attributable mortality, clinical failure, recurrence, readmission or a composite of these), and time-context (high-versus low-trough goal periods) are heterogeneous across studies. What remains unchanged is that vancomycin continues to be a nephrotoxic but highly important therapeutic agent in the management of invasive methicillin-resistant Staphylococcus aureus infections. The focus of clinician scientists must continue to be on how to maximize the safety and efficacy of this agent in practice. The authors declare that they have no relevant conflicts of interest. NJR acknowledges receipt of research funds from Paratek and the American Association of Colleges of Pharmacy , outside the current manuscript and PRY reports no relevant conflicts and serves as the President and Chief Scientist at Optimal Data Analysis, LLC. No financial conflicts whatsoever exist because Dr. Yarnold has made the copyrighted ODA software freely available without any financial remuneration, and technical manuals and guidance on how to use these programs are published to the ODA Journal, which Dr. Yarnold makes freely available at his own expense. Additionally, Dr. Rhodes has created an R-based front-end interface to the ODA program, which he has made freely available via GitHub without remuneration [https://github.com/njrhodes/ODA]. ‘Limitations of classification and regression tree analysis in vancomycin exposure-response relationship studies’ – Author's replyClinical Microbiology and InfectionVol. 27Issue 12PreviewWe thank Rhodes and Yarnold [1] for their comments on our analyses [2]. Herein we address what they describe as serious mathematical and methodological flaws in our work. Full-Text PDF

Biosimilars are biologic drug products that are highly similar to reference products in analytic features, pharmacokinetics and pharmacodynamics, immunogenicity, safety, and efficacy. Biosimilar epoetin received Food and Drug Administration (FDA) approval in 2018. The manufacturer received an FDA nonapproval letter in 2017, despite receiving a favorable review by FDA's Oncologic Drugs Advisory Committee (ODAC) and an FDA nonapproval letter in 2015 for an earlier formulation. We discuss the 2018 FDA approval, the 2017 FDA ODAC Committee review, and the FDA complete response letters in 2015 and 2017; review concepts of litigation, naming, labeling, substitution, interchangeability, and pharmacovigilance; review European and U.S. oncology experiences with biosimilar epoetin; and review the safety of erythropoiesis-stimulating agents. In 2020, policy statements from AETNA, United Health Care, and Humana indicated that new epoetin oncology starts must be for biosimilar epoetin unless medical need for other epoetins is documented. Empirical studies report that as of 2012, reference epoetin use decreased from 40%-60% of all patients with cancer with chemotherapy-induced anemia to <5% of such patients because of safety concerns. Between 2018 and 2020, biosimilar epoetin use varied, increasing to 81% among one private insurer's patients covered by Medicare whose cancer care is administered with Oncology Analytics and to 41% with the same private insurer's patients with cancer covered by commercial health insurance and administered by the private insurer, to 0% in several Veterans Administration Hospitals, increasing to 100% in one large county hospital in California, and with yet-to-be-reported data from most oncology settings. We conclude that biosimilar epoetin appears to have overcome some barriers since 2015, although current uptake in the U.S. is variable. Pricing and safety considerations for all erythropoiesis-stimulating agents are primary determinants of biosimilar epoetin oncology uptake. Implications for Practice Few oncologists understand substitution and interchangeability of biosimilars with reference drugs. Epoetin biosimilar is new to the market, and physician and patient understanding is limited. The development of epoetin biosimilar is not familiar to oncologists.

Hospitalized patients with community-acquired pneumonia (CAP) are at risk of developing Clostridioides difficile infection (CDI). We developed and tested clinical decision rules for identifying CDI risk in this patient population. The study was a single-center retrospective, case-control analysis of hospitalized adult patients empirically treated for CAP between 1 January 2014 and 3 March 2018. Differences between cases (CDI diagnosed within 180 days following admission) and controls (no test result indicating CDI during the study period) with respect to prehospitalization variables were modeled to generate propensity scores. Postadmission variables were used to predict case status on each postadmission day where (i) >= 1 additional case was identified and (ii) each model stratum contained >= 15 subjects. Models were developed and tested using optimal discriminant analysis and classification tree analysis. Forty-four cases and 181 controls were included. The median time to diagnosis was 50 days postadmission. After weighting, three models were identified (20, 117, and 165 days postadmission). The day 20 model yielded the greatest (weighted [w]) accuracy (weighted area under the receiver operating characteristic curve [wROC area] = 0.826) and the highest chance-corrected accuracy (weighted effect strength for sensitivity [wESS] = 65.3). Having a positive culture (odds, 1:4; P=0.001), receipt of ceftriaxone plus azithromycin for a defined infection (odds, 3:5; P=0.006), and continuation of empirical broad-spectrum antibiotics with activity against P. aeruginosa when no pathogen was identified (odds, 1:8; P=0.013) were associated with CDI on day 20. Three models were identified that accurately predicted CDI in hospitalized patients treated for CAP. Antibiotic use increased the risk of CDI in all models, underscoring the importance of antibiotic stewardship.

Personal health informatics have the potential to help patients discover personalized health management strategies that influence outcomes. Fibromyalgia (FM) is a complex chronic illness requiring individualized strategies that may be informed by analysis of personal health informatics data. An online health diary program with dynamic feedback was developed to assist patients with FM in identifying symptom management strategies that predict their personal outcomes, and found reduced symptom levels associated with program use. The aim of this study was to determine longitudinal associations between program use and functional impact of FM as measured by scores on a standardized assessment instrument, the Fibromyalgia Impact Questionnaire (FIQ). Participants were self-identified as diagnosed with FM and recruited via online FM advocacy websites. Participants used an online health diary program (“SMARTLog”) to report symptom ratings, behaviors, and management strategies used. Based on single-subject analysis of the accumulated data over time, individualized recommendations (“SMARTProfile”) were then provided by the automated feedback program. Indices of program use comprised of cumulative numbers of SMARTLogs completed and SMARTProfiles received. Participants included in this analysis met a priori criteria of sufficient program use to generate SMARTProfiles (ie, ≥22 SMARTLogs completed). Users completed the FIQ at baseline and again each subsequent month of program use as follow-up data for analysis. Kendall tau-b, a nonparametric statistic that measures both the strength and direction of an ordinal association between two repeated measured variables, was computed between all included FIQ scores and both indices of program use for each subject at the time of each completed FIQ. A total of 76 users met the a priori use criteria. The mean baseline FIQ score was 61.6 (SD 14.7). There were 342 FIQ scores generated for longitudinal analysis via Kendall tau-b. Statistically significant inverse associations were found over time between FIQ scores and (1) the cumulative number of SMARTLogs completed (tau-b=–0.135, P<.001); and (2) the cumulative number of SMARTProfiles received (tau-b=–0.133, P<.001). Users who completed 61 or more SMARTLogs had mean follow-up scores of 49.9 (n=25, 33% of the sample), an 18.9% drop in FM impact. Users who generated 11 or more new SMARTProfiles had mean follow-up scores of 51.8 (n=23, 30% of the sample), a 15.9% drop. Significant inverse associations were found between FIQ scores and both indices of program use, with FIQ scores declining as use increased. Based on established criteria for rating FM severity, the top one-third of users in terms of use had clinically significant reductions from “severe” to “moderate” FM impact. These findings underscore the value of self-management interventions with low burden, high usability, and high perceived relevance to the user. ClinicalTrials.gov NCT02515552; https://clinicaltrials.gov/ct2/show/NCT02515552

AbstractAdults hospitalized with community-acquired pneumonia (CAP) typically receive antibiotics and thus are at increased risk of developing Clostridioides difficile infection (CDI), a disease of significant morbidity. We developed and validated a CAP-specific clinical decision algorithm to facilitate optimal diagnostic stewardship of C. difficile polymerase chain reaction (PCR) testing. The study was a single-center retrospective, case-control analysis of hospitalized adult patients empirically treated for CAP between January 1, 2014 and May 29, 2018. A series of predictive models and validity assessments were used to evaluate demographic and post-admission patient-specific risk factors as predictors of CDI case status among patients with CAP. Thirty-two PCR confirmed CDI cases were identified and 232 randomly selected controls were drawn from the total CAP population. After propensity score weighting, hospital-onset (HO) CDI was significantly associated with broad-spectrum Gram-negative antibiotic use (P=0.002) as was subsequent community-onset (CO) CDI (P=0.005). Modified-APACHE II > 8.5 (P=0.003) and broad-spectrum Gram-negative antibiotic use (P=0.002) were associated with healthcare-associated CDI and were robust in multiple validity analyses. Patients with m-APACHE II ≤ 8.5 who received broad-spectrum Gram-negative antibiotics were more likely (odds=1:2) to experience healthcare-associated CDI compared to those who did not receive these broad-spectrum agents (odds=1:125) and compared to those with m-APACHE II > 8.5 irrespective of treatment (odds=5:27). We conclude that broad-spectrum Gram-negative antibiotic use was the common factor in development of CDI in patients with CAP in all settings. Prospective studies are needed to confirm the reproducibility and clinical utility of our model when used for diagnostic test stewardship.

Introduction SARS-CoV-2 is the beta-coronavirus responsible for COVID-19. Facemask use has been qualitatively associated with reduced COVID-19 cases, but no study has quantitatively assessed the impact of government mask mandates ( MM ) on new COVID-19 cases across multiple US States. Data and Methods We utilized a non-parametric machine-learning algorithm to test the a priori hypothesis that MM were associated with reductions in new COVID-19 cases. Publicly available data were used to analyze new COVID-19 cases from 37 States and the District of Columbia (i.e., “38 States”). We conducted confirmatory All-States and State-Wise analyses, validity analyses [e.g., leave-one-out (LOO) and bootstrap resampling], and covariate analyses. Results No statistically significant difference in the daily number of new COVID-19 infections was discernable in the All-States analysis. In State-Wise LOO validity analysis, 11 States exhibited reductions in new COVID-19 and the reductions in four of these States (AK, MA, MN, VA) were significant in bootstrap resampling. Only the Social Capital Index predicted MM success (training p <0.028 and LOO p <0.013). Conclusion Results obtained when studying the impact of MM on COVID-19 cases varies as a function of the heterogeneity of the sample being considered, providing clear evidence of Simpson’s Paradox and thus of confounded findings. As such, studies of MM effectiveness should be conducted on disaggregated data. Since transmissions occur at the individual rather than at the collective level, additional work is needed to identify optimal social, psychological, environmental, and educational factors which will reduce the spread of SARS-CoV-2 and facilitate MM effectiveness across diverse settings. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement The author(s) received no specific funding for this work. ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: Not human subjects research. Analysis of publicly available data. All necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines and uploaded the relevant EQUATOR Network research reporting checklist(s) and other pertinent material as supplementary files, if applicable. Yes All data are fully available without restriction and referenced as publicly available data.

Abstract Background Patients with community-acquired pneumonia (CAP) who are hospitalized and treated with antibiotics may carry an increased risk for developing Clostridioides difficile infection (CDI). Accurate risk estimation tools are needed to guide monitoring and CDI mitigation efforts. We aimed to identify patient-specific risk factors associated with CDI among hospitalized patients with CAP. Methods Design: retrospective case-control study of hospitalized patients who received CAP-directed antibiotic therapy between 1/1/2014 and 5/29/2018. Cases were hospitalized CAP patients who developed CDI post-admission. Control patients did not develop CDI and were selected at random from CAP patients hospitalized during this period. Variables: comorbidities, laboratory results, vital signs, severity of illness, prior hospitalization, and past antibiotic use. Propensity-score weights: identified via structural decomposition analysis of pre-treatment variables. Analysis: weighted classification tree models that predicted any CDI, hospital-onset CDI, and any healthcare-associated CDI according to CAP antibiotic treatment. Performance: percent accuracy in classification (PAC) and weighted positive (PPV) and negative predictive values (NPV). Modeling: completed using the ODA package (v1.0.1.3) for R (v3.5.1). Results A total of 32 cases and 232 controls were identified. Sixty pre-treatment variables were screened. Structural decomposition analysis, completed in two stages, identified prior hospitalization (OR 6.56, 95% CI: 3.01-14.31; PAC: 80.3%) and BUN greater than 29 mg/dL (OR 11.67, 95% CI: 2.41-56.5; PAC: 80.8%) as propensity-score weights. With respect to CDI, receipt of broad-spectrum anti-pseudomonal antibiotics was significantly (all P’s< 0.05) associated with any CDI (NPV: 90.29%, PPV: 27.94%), hospital-onset CDI (NPV: 97.53%, PPV: 26.86%), and healthcare-associated CDI (NPV: 92.89%, PPV: 27.94%). Conclusion We identified risk factors available at hospital admission and empiric use of broad-spectrum Gram-negative antibiotics as being associated with the development of CDI. Model PPVs were over two-fold greater than our sample base rate. Increased monitoring and avoidance of overly broad antibiotic use in high-risk patients appears warranted. Disclosures All Authors: No reported disclosures

Erythropoiesis‐stimulating agents (ESAs) are available to treat chemotherapy‐induced anemia (CIA). In 2007–2008, regulatory notifications advised of venous thromboembolism and mortality risks while the Center for Medicare and Medicaid Services' restricted ESA initiation to patients with hemoglobin <10 g/dl. In 2010, a Risk Evaluation and Mitigation Strategies required consent prior to administration. We evaluated ESA utilization from 2003 to 2012 and obtained private health insurer claims data for persons with lung, colorectal, or breast cancer from 2001 to 2012. ESA use for CIA was determined by an ESA claim after chemotherapy, up to 6 months after treatment. We identified 839,948 commercially insured patients, including 24,785 patients with ESA‐treated CIA (3.2%). Darbepoetin use increased 3.9‐fold from 2003 to 2007 (12.3% to 48.7%) and then decreased 95% to 2.6% by 2012. Epoetin use decreased 90% from 2003 to 2012 (30.3% to 3.1%). Between 2003 and 2012, mean epoetin dosing decreased 0.8‐fold (244,979 in 2003 vs. 196,216 units in 2012), but increased 1.8‐fold for darbepoetin‐treated CIA (262 in 2003 to 467 μg in 2012). Among CIA patients, transfusions were low (4.5%) in 2002–2007, then increased 2.2‐fold between 2008 and 2012. Safety initiatives between 2007 and 2010 facilitated reductions in ESA use combined with changes in coverage. These data show the efficacy of regulatory efforts, publication of adverse events and changes in reimbursement in reducing use of ESAs. Future studies are warranted to optimize deimplementation strategies to improve patient safety.

Antibiotic use is commonly tracked electronically by antimicrobial stewardship programs (ASPs). Traditionally, evaluating the appropriateness of antibiotic use requires time- and labor-intensive manual review of each drug order. A drug-specific “appropriateness” algorithm applied electronically would improve the efficiency of ASPs. We thus created an antibiotic “never event” (NE) algorithm to evaluate vancomycin use, and sought to determine the performance characteristics of the electronic data capture strategy. An antibiotic NE algorithm was developed to characterize vancomycin use (Figure) at a large academic institution (1/2016–8/2019). Patients were electronically classified according to the NE algorithm using data abstracted from their electronic health record. Type 1 NEs, defined as continued use of vancomycin after a vancomycin non-susceptible pathogen was identified, were the focus of this analysis. Type 1 NEs identified by automated data capture were reviewed manually for accuracy by either an infectious diseases (ID) physician or an ID pharmacist. The positive predictive value (PPV) of the electronic data capture was determined. Antibiotic Never Event (NE) Algorithm to Characterize Vancomycin Use A total of 38,774 unique cases of vancomycin use were available for screening. Of these, 0.6% (n=225) had a vancomycin non-susceptible pathogen identified, and 12.4% (28/225) were classified as a Type 1 NE by automated data capture. All 28 cases included vancomycin-resistant Enterococcus spp (VRE). Upon manual review, 11 cases were determined to be true positives resulting in a PPV of 39.3%. Reasons for the 17 false positives are given in Table 1. Asymptomatic bacteriuria (ASB) due to VRE in scenarios where vancomycin was being appropriately used to treat a concomitant vancomycin-susceptible infection was the most common reason for false positivity, accounting for 64.7% of false positive cases. After removing urine culture source (n=15) from the algorithm, PPV improved to 53.8%. An automated vancomycin NE algorithm identified 28 Type 1 NEs with a PPV of 39%. ASB was the most common cause of false positivity and removing urine culture as a source from the algorithm improved PPV. Future directions include evaluating Type 2 NEs (Figure) and prospective, real-time application of the algorithm. Marc H. Scheetz, PharmD, MSc, Merck and Co. (Grant/Research Support)

Oncology-associated adverse drug/device reactions can be fatal. Some clinicians who treat single patients with severe oncology-associated toxicities have researched case series and published this information. We investigated motivations and experiences of select individuals leading such efforts. Clinicians treating individual patients who developed oncology-associated serious adverse drug events were asked to participate. Inclusion criteria included having index patient information, reporting case series, and being collaborative with investigators from two National Institutes of Health funded pharmacovigilance networks. Thirty-minute interviews addressed investigational motivation, feedback from pharmaceutical manufacturers, FDA personnel, and academic leadership, and recommendations for improving pharmacovigilance. Responses were analyzed using constant comparative methods of qualitative analysis. Overall, 18 clinicians met inclusion criteria and 14 interviewees are included. Primary motivations were scientific curiosity, expressed by six clinicians. A less common theme was public health related (three clinicians). Six clinicians received feedback characterized as supportive from academic leaders, while four clinicians received feedback characterized as negative. Three clinicians reported that following the case series publication they were invited to speak at academic institutions worldwide. Responses from pharmaceutical manufacturers were characterized as negative by 12 clinicians. One clinician’s wife called the post-reporting time the “Maalox month,” while another clinician reported that the manufacturer collaboratively offered to identify additional cases of the toxicity. Responses from FDA employees were characterized as collaborative for two clinicians, neutral for five clinicians, unresponsive for negative by six clinicians. Three clinicians endorsed developing improved reporting mechanisms for individual physicians, while 11 clinicians endorsed safety activities that should be undertaken by persons other than a motivated clinician who personally treats a patient with a severe adverse drug/device reaction. Our study provides some of the first reports of clinician motivations and experiences with reporting serious or potentially fatal oncology-associated adverse drug or device reactions. Overall, it appears that negative feedback from pharmaceutical manufacturers and mixed feedback from the academic community and/or the FDA were reported. Big data, registries, Data Safety Monitoring Boards, and pharmacogenetic studies may facilitate improved pharmacovigilance efforts for oncology-associated adverse drug reactions. These initiatives overcome concerns related to complacency, indifference, ignorance, and system-level problems as barriers to documenting and reporting adverse drug events- barriers that have been previously reported for clinician reporting of serious adverse drug reactions.