Global Cancer Surgery: pragmatic solutions to improve cancer surgery outcomes worldwide.

The first Lancet Oncology Commission on Global Cancer Surgery was published in 2015 and serves as a landmark paper in the field of cancer surgery. The Commission highlighted the burden of cancer and the importance of cancer surgery, while documenting the many inadequacies in the ability to deliver safe, timely, and affordable cancer surgical care. This Commission builds on the first Commission by focusing on solutions and actions to improve access to cancer surgery globally, developed by drawing upon the expertise from cancer surgery leaders across the world. We present solution frameworks in nine domains that can improve access to cancer surgery. These nine domains were refined to identify solutions specific to the six WHO regions. On the basis of these solutions, we developed eight actions to propel essential improvements in the global capacity for cancer surgery. Our initiatives are broad in scope, pragmatic, affordable, and contextually applicable, and aimed at cancer surgeons as well as leaders, administrators, elected officials, and health policy advocates. We envision that the solutions and actions contained within the Commission will address inequities and promote safe, timely, and affordable cancer surgery for every patient, regardless of their socioeconomic status or geographic location.

Implementation of pharmacogenomics into inpatient general medicine.

Pharmacogenomics is a crucial piece of personalized medicine. Preemptive pharmacogenomic testing is only used sparsely in the inpatient setting and there are few models to date for fostering the adoption of pharmacogenomic treatment in the inpatient setting. We created a multi-institutional project in Chicago to enable the translation of pharmacogenomics into inpatient practice. We are reporting our implementation process and barriers we encountered with solutions. This study, 'Implementation of Point-of-Care Pharmacogenomic Decision Support Accounting for Minority Disparities', sought to implement pharmacogenomics into inpatient practice at three sites: The University of Chicago, Northwestern Memorial Hospital, and the University of Illinois at Chicago. This study involved enrolling African American adult patients for preemptive genotyping across a panel of actionable germline variants predicting drug response or toxicity risk. We report our approach to implementation and the barriers we encountered engaging hospitalists and general medical providers in the inpatient pharmacogenomic intervention. Our strategies included: a streamlined delivery system for pharmacogenomic information, attendance at hospital medicine section meetings, use of physician and pharmacist champions, focus on hospitalists' care and optimizing system function to fit their workflow, hand-offs, and dealing with hospitalists turnover. Our work provides insights into strategies for the initial engagement of inpatient general medicine providers that we hope will benefit other institutions seeking to implement pharmacogenomics in the inpatient setting.

Developing an Integrated Longitudinal Dataset for Patient-Centered Outcome Measures in Cost-Related Medication Nonadherence.

BACKGROUND: Cost-related medication nonadherence (CRN) is an important patient-centered outcome measure. Longitudinal follow-up of CRN is rare. OBJECTIVE: We propose to develop a novel integrated dataset to study CRN longitudinally. RESEARCH DESIGN: A dataset of 2000 Medicare beneficiaries at high risk of hospitalization surveyed quarterly on CRN and followed up individually for 8 quarters between 2013 and 2018 was linked to Medicare files. A metric of CRN categorizing persistent, intermittent, and transient CRN during the 8 quarters was developed. An ordered logit model and a logit model were developed to assess the factors influencing CRN overall and persistent CRN, respectively. RESULTS: A total of 1761 patients were included in the analysis, among whom 869 (49.3%) reported CRN at least once in the 8-quarter study period, 178 (10%) reported persistent CRN, 395 (22.4%) reported intermittent CRN, and 296 (16.8%) reported transient CRN. The conditional effect in the logit model for persistent CRN revealed that baseline dual eligibility was negatively associated (adjusted odds ratio = 0.45, P < 0.01) and depression positively associated (adjusted odds ratio = 1.55, P = 0.01) with persistent CRN. The marginal analysis in the ordered logit model revealed a clear pattern of higher probabilities of persistent and intermittent CRN at younger ages while transient CRN was flat. Among the 252 subjects who were deceased, 31 (12.3%) reported persistent CRN, compared with 147 (9.74%) who were alive (P = 0.21 by χ2 test). CONCLUSIONS: A significant number of patients reported persistent CRN, including those who were at the end of life. Research is critically needed to understand behavioral patterns among the younger Medicare population.

Anesthesia providers as stakeholders to adoption of pharmacogenomic information in perioperative care.

OBJECTIVES: Integration of pharmacogenomics into clinical care is being studied in multiple disciplines. We hypothesized that understanding attitudes and perceptions of anesthesiologists, critical care and pain medicine providers would uncover unique considerations for future implementation within perioperative care. METHODS: A survey (multiple choice and Likert-scale) was administered to providers within our Department of Anesthesia and Critical Care prior to initiation of a department-wide prospective pharmacogenomics implementation program. The survey addressed knowledge, perceptions, experiences, resources and barriers. RESULTS: Of 153 providers contacted, 149 (97%) completed the survey. Almost all providers (92%) said that genetic results influence drug therapy, and few (22%) were skeptical about the usefulness of pharmacogenomics. Despite this enthusiasm, 87% said their awareness about pharmacogenomic information is lacking. Feeling well-informed about pharmacogenomics was directly related to years in practice/experience: only 38% of trainees reported being well-informed, compared to 46% of those with 1-10 years of experience, and nearly two-thirds with 11+ years (P < 0.05). Regarding barriers, providers reported uncertainty about availability of testing, turnaround time and whether testing is worth financial costs. CONCLUSIONS: Anesthesiology, critical care and pain medicine providers are optimistic about the potential clinical utility of pharmacogenomics, but are uncertain about practical aspects of testing and desire clear guidelines on the use of results. These findings may inform future institutional efforts toward greater integration of genomic results to improve medication-related outcomes.

Pilot Findings of Pharmacogenomics in Perioperative Care: Initial Results From the First Phase of the ImPreSS Trial.

BACKGROUND: Pharmacogenomics, which offers a potential means by which to inform prescribing and avoid adverse drug reactions, has gained increasing consideration in other medical settings but has not been broadly evaluated during perioperative care. METHODS: The Implementation of Pharmacogenomic Decision Support in Surgery (ImPreSS) Trial is a prospective, single-center study consisting of a prerandomization pilot and a subsequent randomized phase. We describe findings from the pilot period. Patients planning elective surgeries were genotyped with pharmacogenomic results, and decision support was made available to anesthesia providers in advance of surgery. Pharmacogenomic result access and prescribing records were analyzed. Surveys (Likert-scale) were administered to providers to understand utilization barriers. RESULTS: Of eligible anesthesiology providers, 166 of 211 (79%) enrolled. A total of 71 patients underwent genotyping and surgery (median, 62 years; 55% female; average American Society of Anesthesiologists (ASA) score, 2.6; 58 inpatients and 13 ambulatories). No patients required postoperative intensive care or pain consultations. At least 1 provider accessed pharmacogenomic results before or during 41 of 71 surgeries (58%). Faculty were more likely to access results (78%) compared to house staff (41%; P = .003) and midlevel practitioners (15%) ( P < .0001). Notably, all administered intraoperative medications had favorable genomic results with the exception of succinylcholine administration to 1 patient with genomically increased risk for prolonged apnea (without adverse outcome). Considering composite prescribing in preoperative, recovery, throughout hospitalization, and at discharge, each patient was prescribed a median of 35 (range 15-83) total medications, 7 (range 1-22) of which had annotated pharmacogenomic results. Of 2371 prescribing events, 5 genomically high-risk medications were administered (all tramadol or omeprazole; with 2 of 5 pharmacogenomic results accessed), and 100 genomically cautionary mediations were administered (hydralazine, oxycodone, and pantoprazole; 61% rate of accessing results). Providers reported that although results were generally easy to access and understand, the most common reason for not considering results was because remembering to access pharmacogenomic information was not yet a part of their normal clinical workflow. CONCLUSIONS: Our pilot data for result access rates suggest interest in pharmacogenomics by anesthesia providers, even if opportunities to alter prescribing in response to high-risk genotypes were infrequent. This pilot phase has also uncovered unique considerations for implementing pharmacogenomic information in the perioperative care setting, and new strategies including adding the involvement of surgery teams, targeting patients likely to need intensive care and dedicated pain care, and embedding pharmacists within rounding models will be incorporated in the follow-on randomized phase to increase engagement and likelihood of affecting prescribing decisions and clinical outcomes.

Improving natural product research translation: From source to clinical trial.

While great interest in health effects of natural product (NP) including dietary supplements and foods persists, promising preclinical NP research is not consistently translating into actionable clinical trial (CT) outcomes. Generally considered the gold standard for assessing safety and efficacy, CTs, especially phase III CTs, are costly and require rigorous planning to optimize the value of the information obtained. More effective bridging from NP research to CT was the goal of a September, 2018 transdisciplinary workshop. Participants emphasized that replicability and likelihood of successful translation depend on rigor in experimental design, interpretation, and reporting across the continuum of NP research. Discussions spanned good practices for NP characterization and quality control; use and interpretation of models (computational through in vivo) with strong clinical predictive validity; controls for experimental artefacts, especially for in vitro interrogation of bioactivity and mechanisms of action; rigorous assessment and interpretation of prior research; transparency in all reporting; and prioritization of research questions. Natural product clinical trials prioritized based on rigorous, convergent supporting data and current public health needs are most likely to be informative and ultimately affect public health. Thoughtful, coordinated implementation of these practices should enhance the knowledge gained from future NP research.

Variability in comorbidites and health services use across homeless typologies: multicenter data linkage between healthcare and homeless systems.

BACKGROUND: Homelessness is associated with substantial morbidity. Data linkages between homeless and health systems are important to understand unique needs across homeless populations, identify homeless individuals not registered in homeless databases, quantify the impact of housing services on health-system use, and motivate health systems and payers to contribute to housing solutions. METHODS: We performed a cross-sectional survey including six health systems and two Homeless Management Information Systems (HMIS) in Cook County, Illinois. We performed privacy-preserving record linkage to identify homelessness through HMIS or ICD-10 codes captured in electronic medical records. We measured the prevalence of health conditions and health-services use across the following typologies: housing-service utilizers stratified by service provided (stable, stable plus unstable, unstable) and non-utilizers (i.e., homelessness identified through diagnosis codes-without receipt of housing services). RESULTS: Among 11,447 homeless recipients of healthcare, nearly 1 in 5 were identified by ICD10 code alone without recorded homeless services (n = 2177; 19%). Almost half received homeless services that did not include stable housing (n = 5444; 48%), followed by stable housing (n = 3017; 26%), then receipt of both stable and unstable services (n = 809; 7%). Setting stable housing recipients as the referent group, we found a stepwise increase in behavioral-health conditions from stable housing to those known as homeless solely by health systems. Compared to those in stable housing, prevalence rate ratios (PRR) for those without homeless services were as follows: depression (PRR = 2.2; 95% CI 1.9 to 2.5), anxiety (PRR = 2.5; 95% CI 2.1 to 3.0), schizophrenia (PRR = 3.3; 95% CI 2.7 to 4.0), and alcohol-use disorder (PRR = 4.4; 95% CI 3.6 to 5.3). Homeless individuals who had not received housing services relied on emergency departments for healthcare-nearly 3 of 4 visited at least one and many (24%) visited multiple. CONCLUSIONS: Differences in behavioral-health conditions and health-system use across homeless typologies highlight the particularly high burden among homeless who are disconnected from homeless services. Fragmented and high use of emergency departments for care should motivate health systems and payers to promote housing solutions, especially those that incorporate substance use and mental health treatment.

Pharmacogenomic genotypes define genetic ancestry in patients and enable population-specific genomic implementation.

The importance of genetic ancestry characterization is increasing in genomic implementation efforts, and clinical pharmacogenomic guidelines are being published that include population-specific recommendations. Our aim was to test the ability of focused clinical pharmacogenomic SNP panels to estimate individual genetic ancestry (IGA) and implement population-specific pharmacogenomic clinical decision-support (CDS) tools. Principle components and STRUCTURE were utilized to assess differences in genetic composition and estimate IGA among 1572 individuals from 1000 Genomes, two independent cohorts of Caucasians and African Americans (AAs), plus a real-world validation population of patients undergoing pharmacogenomic genotyping. We found that clinical pharmacogenomic SNP panels accurately estimate IGA compared to genome-wide genotyping and identify AAs with ≥70 African ancestry (sensitivity >82%, specificity >80%, PPV >95%, NPV >47%). We also validated a new AA-specific warfarin dosing algorithm for patients with ≥70% African ancestry and implemented it at our institution as a novel CDS tool. Consideration of IGA to develop an institutional CDS tool was accomplished to enable population-specific pharmacogenomic guidance at the point-of-care. These capabilities were immediately applied for guidance of warfarin dosing in AAs versus Caucasians, but also provide a real-world model that can be extended to other populations and drugs as actionable genomic evidence accumulates.

Physician and Nurse Practitioner Attitudes on Generic Prescribing of Oral Contraceptive Pills and Antidepressants.

IMPORTANCE: As prescription drug costs rise, it is important to understand attitudes among primary care physicians and nurse practitioners (NPs) towards generic drugs. OBJECTIVE: We aimed to examine the generic skepticism index (GSI) among primary care clinicians, and their willingness to discuss and prescribe generic antidepressants (ADs) and generic oral contraceptives (OCPs). DESIGN: We used a factorial vignette design survey to test 4 factors: message source, message, brand preference, and drug class. Participants were randomized to different combinations of factors. SETTING: This was a cross-sectional study. PARTICIPANTS: Physicians registered with the American College of Physicians (ACP) and NPs registered with the American Association of Nurse Practitioners (AANP) participated in the study. MAIN MEASURES: The primary outcomes were generic skepticism as measured using the generic skepticism index (GSI), and clinician willingness to discuss and prescribe generics. RESULTS: Surveys were completed by 56% of physicians (n = 369/661) and 60% of NPs (n = 493/819). Compared with physicians, NPs were younger (p < 0.001), predominantly female (p < 0.001), and differed in the race (p < 0.001). According to the GSI, 16% (n = 138/862) were identified as generic skeptics (18.5% of NPs and 12.7% of physicians, p = 0.023). Generic skeptics had lower odds of willingness to discuss switching (OR 0.22, 95% CI (0.14-0.35), p < 0.001) or prescribe (OR 0.18, 95% CI (0.11-0.28), p < 0.001) generic OCPs. Participants had lower odds of willingness to prescribe generic drugs to patients with brand preference compared with brand-neutral patients (OR 0.64, 95% CI 0.50-0.82, p < 0.001). CONCLUSIONS AND RELEVANCE: Generic skepticism was associated with lower willingness to discuss or prescribe generic drugs. Clinicians reported lower willingness to discuss switching or prescribe generics for OCPs than for ADs. Patient brand preference hindered generic prescribing. Message source and message type were not significantly associated with outcomes.

Simulation-based optimization to improve hospital patient assignment to physicians and clinical units.

A fundamental activity in hospital operations is patient assignment, which we define as the process of assigning hospital patients to specific physician services and clinical units based on their diagnosis. When the preferred assignment is not possible, typically due to capacity limits, hospitals often allow for overflow, which is the assignment of patients to other services and/or units. Overflow accelerates assignment, but can also reduce care quality and increase length of stay. This paper develops a discrete-event simulation model to evaluate different assignment strategies. Using a simulation-based optimization approach, we evaluate and heuristically optimize these strategies accounting for expected hospital and physician profit, care quality and patient waiting time. We apply the model using data from the University of Chicago Medical Center. We find that the strategies that use heuristically optimized designation of overflow services and units increase expected profit relative to the capacity-based strategy in which overflow patients are assigned to a service and unit with the most available capacity. We also find further improvement in the strategy that uses heuristically optimized overflow services and units as well as a holding unit that holds patients until a bed in their primary or secondary unit becomes available. Additionally, we demonstrate the effects of these strategies on other performance measures such as patient concentration, waiting time, and outcomes.

Analysis of comprehensive pharmacogenomic profiling to impact in-hospital prescribing.

INTRODUCTION: In-hospital adverse medication events result in increased morbidity and mortality. Many implicated drugs carry pharmacogenomic information. We hypothesized that comprehensive pre-emptive pharmacogenomic profiling could have high relevance for in-hospital prescribing. PATIENTS AND METHODS: We retrospectively analyzed the in-hospital medications of a genotyped outpatient cohort admitted at our institution from 2012 to 2015. The endpoints were medication changes (new medications initiated, dose adjustments, or medications discontinued) involving drugs with pharmacogenomic annotations from three sources: Clinical Pharmacogenetics Implementation Consortium guidance, Food and Drug Administration label information, and drugs with clinical decision supports in our institutional pharmacogenomic Genomic Prescribing System. RESULTS: Of 867 genotyped outpatients, 20 were hospitalized (mean: 78.2 years, 65% male). This hospitalized cohort was significantly older (78.2 vs. 61.3 years, P<0.0001) and took more medications (8.9 vs. 5.0 medications, P<0.0001). Out of 159 medication changes made, most (67.9%) were new medications (average: 2.5/hospitalization) with one-third of these having clinically annotated pharmacogenomic information. Half of all hospitalizations involved at least one pharmacogenomic medication. Over half (55%) of the hospitalized cohort was newly prescribed at least one of eight key pharmacogenomic medications, including high-risk drugs such as clopidogrel, codeine, and warfarin. CONCLUSION: Our study suggested that older patients and those with polypharmacy were at increased risk for hospitalizations, where many new prescriptions included frequently used pharmacogenomic drugs. Targeting this group for pre-emptive genotyping would facilitate the delivery of highly relevant information to inform inpatient prescribing.

Assessment of provider-perceived barriers to clinical use of pharmacogenomics during participation in an institutional implementation study.

OBJECTIVE: The objective of this study was to study provider attitudes of and perceived barriers to the clinical use of pharmacogenomics before and during participation in an implementation program. PARTICIPANTS AND METHODS: From 2012 to 2017, providers were recruited. After completing semistructured interviews (SSIs) about pharmacogenomics, providers received training on and access to a clinical decision support tool housing patient-specific pharmacogenomic results. Thematic analysis of SSI was conducted (inter-rater reliability κ≥0.75). Providers also completed surveys before and during study participation, and provider-perceived barriers to pharmacogenomic implementation were analyzed. RESULTS: Seven themes emerged from the SSI (listed from most frequent to least): decision-making, concerns with pharmacogenomic adoption, outcome expectancy, provider knowledge of pharmacogenomics, patient attitudes, individualized treatment, and provider interest in pharmacogenomics. Although there was prestudy enthusiasm among all providers, concerns with clinical utility, time, results accession, and knowledge of pharmacogenomics were frequently stated at baseline. Despite this, adoption of pharmacogenomics was robust, as patient-specific results were accessed at 64% of visits, and medication changes were influenced by provided pharmacogenomic information 42% of the time. Providers reported they had enough time to evaluate the information and the results were easily understood on 74 and 98% of surveys, respectively. Nevertheless, providers consistently felt there was insufficient pharmacogenomic information for most drugs they prescribed and clear guidelines for using pharmacogenomic information were lacking. CONCLUSION: Despite initial concerns about adequate time and knowledge for adoption, providers frequently utilized pharmacogenomic results. Provider-perceived barriers to wider use included lack of clear guidelines and evidence for most drugs, highlighting important considerations for the field of pharmacogenomics.

Patient-provider communications about pharmacogenomic results increase patient recall of medication changes.

Effective doctor-patient communication is critical for disease management, especially when considering genetic information. We studied patient-provider communications after implementing a point-of-care pharmacogenomic results delivery system to understand whether pharmacogenomic results are discussed and whether medication recall is impacted. Outpatients undergoing preemptive pharmacogenomic testing (cases), non-genotyped controls, and study providers were surveyed from October 2012-May 2017. Patient responses were compared between visits where pharmacogenomic results guided prescribing versus visits where pharmacogenomics did not guide prescribing. Provider knowledge of pharmacogenomics, before and during study participation, was also analyzed. Both providers and case patients frequently reported discussions of genetic results after visits where pharmacogenomic information guided prescribing. Importantly, medication changes from visits where pharmacogenomics influenced prescribing were more often recalled than non-pharmacogenomic guided medication changes (OR = 3.3 [1.6-6.7], p = 0.001). Case patients who had separate visits where pharmacogenomics did and did not, respectively, influence prescribing more often remembered medication changes from visits where genomic-based guidance was used (OR = 3.4 [1.2-9.3], p = 0.02). Providers also displayed dramatic increases in personal genomic understanding through program participation (94% felt at least somewhat informed about pharmacogenomics post-participation, compared to 61% at baseline, p = 0.04). Using genomic information during prescribing increases patient-provider communications, patient medication recall, and provider understanding of genomics, important ancillary benefits to clinical use of pharmacogenomics.

Development and testing of a web module to IMPROVE generic prescribing of oral contraceptives among primary care physicians.

WHAT IS KNOWN AND OBJECTIVE: The use of generic oral contraceptives (OCPs) can improve adherence and reduce healthcare costs, yet scepticism of generic drugs remains a barrier to generic OCP discussion and prescription. An educational web module was developed to reduce generic scepticism related to OCPs, improve knowledge of generic drugs and increase physician willingness to discuss and prescribe generic OCPs. METHODS: A needs assessment was completed using in-person focus groups at American College of Physicians (ACP) Annual Meeting and a survey targeting baseline generic scepticism. Insights gained were used to build an educational web module detailing barriers and benefits of generic OCP prescription. The module was disseminated via email to an ACP research panel who completed our baseline survey. Post-module evaluation measured learner reaction, knowledge and intention to change behaviour along with generic scepticism. RESULTS AND DISCUSSION: The module had a response rate of 56% (n = 208/369). Individuals defined as generic sceptics at baseline were significantly less likely to complete our module compared to non-sceptics (responders 9.6% vs non-responders 16.8%, P = 0.04). The majority (85%, n = 17/20) of baseline sceptics were converted to non-sceptics (P < 0.01) following completion of the module. Compared to non-sceptics, post-module generic sceptics reported less willingness to discuss (sceptic 33.3% vs non-sceptic 71.5%, P < 0.01), but not less willingness to prescribe generic OCPs (sceptic 53.3% vs non-sceptic 67.9%, P = 0.25). Non-white physicians and international medical graduates (IMG) were more likely to be generic sceptics at baseline (non-white 86.9% vs white 69.9%, P = 0.01, IMG 13.0% vs USMG 5.0% vs unknown 18.2%, P = 0.03) but were also more likely to report intention to prescribe generic OCPs as a result of the module (non-white 78.7% vs white 57.3%, P < 0.01, IMG 76.1% vs USMG 50.3% vs unknown 77.3%, P = 0.03). WHAT IS NEW AND CONCLUSION: A brief educational web module can be used to promote prescribing of generic OCPs and reduce generic scepticism.

Risk of Social Isolation among Older Patients: What Factors Affect the Availability of Family, Friends, and Neighbors upon Hospitalization?

OBJECTIVES: To examine the prevalence of social isolation among older patients admitted to a hospital, and the effects of sociodemographic and health-related factors on the availability of their family, friends, and neighbor networks. METHODS: Analyses are based on interviews with a sample of 2,449 older patients admitted to an urban academic medical center in the United States. A nine-item version of Lubben's Social Network Scale was developed and used to assess the availability of different social networks. RESULTS: About 47% of the sample was at risk of social isolation. The oldest old and non-White older adults showed greater risk. The availability of family networks was associated with age, sex, marital status, and prior hospitalization; friend networks with age, race, education, prior hospitalization, and functional limitations; neighbor networks with race, education, marital status, and functional limitations. CONCLUSIONS: The risk of social isolation and the availability of social support for hospitalized older adults varies by both patient and network characteristics. Health professionals should attend to this risk and the factors associated with such risk. CLINICAL IMPLICATIONS: By assessing the availability of various types and frequency of support among older patients, health professionals can better identify those who may need additional support after discharge. Such information should be used in discharge planning to help prevent unnecessary complications and potential readmission.

The Relative Ability of Comorbidity Ascertainment Methodologies to Predict In-Hospital Mortality Among Hospitalized Community-acquired Pneumonia Patients.

BACKGROUND: Despite widespread use of comorbidities for population health descriptions and risk adjustment, the ideal method for ascertaining comorbidities is not known. We sought to compare the relative value of several methodologies by which comorbidities may be ascertained. METHODS: This is an observational study of 1596 patients admitted to the University of Chicago for community-acquired pneumonia from 1998 to 2012. We collected data via chart abstraction, administrative data, and patient report, then performed logistic regression analyses, specifying comorbidities as independent variables and in-hospital mortality as the dependent variable. Finally, we compared area under the curve (AUC) statistics to determine the relative ability of each method of comorbidity ascertainment to predict in-hospital mortality. RESULTS: Chart review (AUC, 0.72) and administrative data (Charlson AUC, 0.83; Elixhauser AUC, 0.84) predicted in-hospital mortality with greater fidelity than patient report (AUC, 0.61). However, multivariate logistic regression analyses demonstrated that individual comorbidity derivation via chart review had the strongest relationship with in-hospital mortality. This is consistent with prior literature suggesting that administrative data have inherent, paradoxical biases with important implications for risk adjustment based solely on administrative data. CONCLUSIONS: Although comorbidities derived through administrative data did produce an AUC greater than chart review, our analyses suggest a coding bias in several comorbidities with a paradoxically protective effect. Therefore, chart review, while labor and resource intensive, may be the ideal method for ascertainment of clinically relevant comorbidities.

The effect of red-blood-cell transfusion on fatigue in hospitalized patients with anaemia.

BACKGROUND AND OBJECTIVE: Guidelines suggest that red-blood-cell transfusion decisions for most hospitalized patients be based on haemoglobin (Hb) concentration and the presence of symptoms of anaemia, including fatigue. However, studies differ in whether transfusion is associated with improvements in fatigue. One explanation is that the benefit of transfusion varies by baseline fatigue levels, which existing studies have not examined. The objective of this study was to determine whether the association between transfusion during hospitalization and improvements in fatigue 30 days postdischarge varies by baseline fatigue level. METHODS: A prospective observational study of hospitalized general medicine patients with any Hb <9 g/dl. Patients with sickle cell anaemia and gastrointestinal bleeding were excluded since these diagnoses have alternative transfusion practices. Patients with depression were excluded because their fatigue is not primarily due to anaemia. Fatigue was measured during an in-person interview and a 30-day postdischarge phone interview. Hb values and receipt of a transfusion were collected from hospital administrative data. Linear regression was used to test associations between 'change in fatigue', Hb concentration and receipt of a transfusion. RESULTS: Transfusion interacted with nadir Hb was associated with reduced fatigue postdischarge for patients with higher baseline fatigue (20% most fatigued: ß = 12, P = 0·02; 10% most fatigued: ß = 17, P = 0·02). Patients <50 years old with high baseline fatigue had large reductions in fatigue from transfusion (20%: ß = 23, P = 0·02; 10%: ß = 29, P = 0·03). CONCLUSIONS: Transfusion during hospitalization is associated with reduced fatigue 30 days postdischarge in patients with higher levels of baseline fatigue.

Multicenter Comparison of Machine Learning Methods and Conventional Regression for Predicting Clinical Deterioration on the Wards.

OBJECTIVE: Machine learning methods are flexible prediction algorithms that may be more accurate than conventional regression. We compared the accuracy of different techniques for detecting clinical deterioration on the wards in a large, multicenter database. DESIGN: Observational cohort study. SETTING: Five hospitals, from November 2008 until January 2013. PATIENTS: Hospitalized ward patients INTERVENTIONS: None MEASUREMENTS AND MAIN RESULTS: Demographic variables, laboratory values, and vital signs were utilized in a discrete-time survival analysis framework to predict the combined outcome of cardiac arrest, intensive care unit transfer, or death. Two logistic regression models (one using linear predictor terms and a second utilizing restricted cubic splines) were compared to several different machine learning methods. The models were derived in the first 60% of the data by date and then validated in the next 40%. For model derivation, each event time window was matched to a non-event window. All models were compared to each other and to the Modified Early Warning score, a commonly cited early warning score, using the area under the receiver operating characteristic curve (AUC). A total of 269,999 patients were admitted, and 424 cardiac arrests, 13,188 intensive care unit transfers, and 2,840 deaths occurred in the study. In the validation dataset, the random forest model was the most accurate model (AUC, 0.80 [95% CI, 0.80-0.80]). The logistic regression model with spline predictors was more accurate than the model utilizing linear predictors (AUC, 0.77 vs 0.74; p < 0.01), and all models were more accurate than the MEWS (AUC, 0.70 [95% CI, 0.70-0.70]). CONCLUSIONS: In this multicenter study, we found that several machine learning methods more accurately predicted clinical deterioration than logistic regression. Use of detection algorithms derived from these techniques may result in improved identification of critically ill patients on the wards.