Search

BACKGROUND: The demographics of heart failure are changing. The rate of growth of the “older” heart failure population, specifically those ≥ 75, has outpaced that of any other age group. These older patients were underrepresented in the early beta-blocker trials. There are several reasons, including a decreased potential for mortality benefit and increased risk of side effects, why the risk/benefit tradeoff may be different in this population. OBJECTIVE: We aimed to determine the association between receipt of a beta-blocker after heart failure discharge and early mortality and readmission rates among patients with heart failure and reduced ejection fraction (HFrEF), specifically patients aged 75+. DESIGN AND PARTICIPANTS: We used 100% Medicare Parts A and B and a random 40% sample of Part D to create a cohort of beneficiaries with ≥ 1 hospitalization for HFrEF between 2008 and 2016 to run an instrumental variable analysis. MAIN MEASURE: The primary measure was 90-day, all-cause mortality; the secondary measure was 90-day, all-cause readmission. KEY RESULTS: Using the two-stage least squared methodology, among all HFrEF patients, receipt of a beta-blocker within 30-day of discharge was associated with a − 4.35% (95% CI − 6.27 to − 2.42%, p < 0.001) decrease in 90-day mortality and a − 4.66% (95% CI − 7.40 to − 1.91%, p = 0.001) decrease in 90-day readmission rates. Even among patients ≥ 75 years old, receipt of a beta-blocker at discharge was also associated with a significant decrease in 90-day mortality, − 4.78% (95% CI − 7.19 to − 2.40%, p < 0.001) and 90-day readmissions, − 4.67% (95% CI − 7.89 to − 1.45%, p < 0.001). CONCLUSION: Patients aged ≥ 75 years who receive a beta-blocker after HFrEF hospitalization have significantly lower 90-day mortality and readmission rates. The magnitude of benefit does not appear to wane with age. Absent a strong contraindication, all patients with HFrEF should attempt beta-blocker therapy at/after hospital discharge, regardless of age. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s11606-021-06901-7

The target of inference in microbiome analyses is usually relative abundance (RA) because RA in a sample (e.g., stool) can be considered as an approximation of RA in an entire ecosystem (e.g., gut). However, inference on RA suffers from the fact that RA are calculated by dividing absolute abundances (AA) over the common denominator (CD), the summation of all AA (i.e., library size). Because of that, perturbation in one taxon will result in a change in the CD and thus cause false changes in RA of all other taxa, and those false changes could lead to false positive/negative findings. We propose a novel analysis approach (IFAA) to make robust inference on AA of an ecosystem that can circumvent the issues induced by the CD problem and compositional structure of RA. IFAA can also address the confounding effect of library size and handle zero-inflated data structures. IFAA identifies microbial taxa associated with the covariates in Phase one and estimates the association parameters by employing an independent reference taxon in Phase two. Two real data applications are presented and extensive simulations show that IFAA outperforms other established existing approaches by a big margin in the presence of confounding effect of library size., Corresponding email: zhigang.li@ufl.edu

Purpose We examined self‐reported financial toxicity and out‐of‐pocket expenses among adult women with breast cancer. Methods Patients spoke English, Spanish, or Mandarin Chinese, were aged 18+ years, had stage I–IIIA breast cancer, and were eligible for breast‐conserving and mastectomy surgery. Participants completed surveys about out‐of‐pocket costs and financial toxicity at 1 week, 12 weeks, and 1 year postsurgery. Results Three hundred ninety‐five of 448 eligible patients (88.2%) from the parent trial completed surveys. Excluding those reporting zero costs, crude mean ± SD out‐of‐pocket costs were $1,512 ± $2,074 at 1 week, $2,609 ± $6,369 at 12 weeks, and $3,308 ± $5,000 at 1 year postsurgery. Controlling for surgery, cancer stage, and demographics with surgeon and clinic as random effects, higher out‐of‐pocket costs were associated with higher financial toxicity 1 week and 12 weeks postsurgery (p < .001). Lower socioeconomic status (SES) was associated with lower out‐of‐pocket costs at each time point (p = .002–.013). One week postsurgery, participants with lower SES reported financial toxicity scores 1.02 points higher than participants with higher SES (95% confidence interval [CI], 0.08–1.95). Black and non‐White/non‐Black participants reported financial toxicity scores 1.91 (95% CI, 0.46–3.37) and 2.55 (95% CI, 1.11–3.99) points higher than White participants. Older (65+ years) participants reported financial toxicity scores 2.58 points lower than younger (, The term “financial toxicity” describes both the financial hardship and the psychological distress associated with health care costs. This article examines financial toxicity of women diagnosed with breast cancer at three time points up to 1 year post‐surgery, assessing demographic and clinical variables associated with higher financial toxicity and higher out‐of‐pocket expenses.

Network centrality measures assign importance to influential or key nodes in a network based on the topological structure of the underlying adjacency matrix. In this work, we define the importance of a node in a network as being dependent on whether it is the only one of its kind among its neighbors’ ties. We introduce linchpin score, a measure of local uniqueness used to identify important nodes by assessing both network structure and a node attribute. We explore linchpin score by attribute type and examine relationships between linchpin score and other established network centrality measures (degree, betweenness, closeness, and eigenvector centrality). To assess the utility of this measure in a real-world application, we measured the linchpin score of physicians in patient-sharing networks to identify and characterize important physicians based on being locally unique for their specialty. We hypothesized that linchpin score would identify indispensable physicians who would not be easily replaced by another physician of their specialty type if they were to be removed from the network. We explored differences in rural and urban physicians by linchpin score compared with other network centrality measures in patient-sharing networks representing the 306 hospital referral regions in the United States. We show that linchpin score is uniquely able to make the distinction that rural specialists, but not rural general practitioners, are indispensable for rural patient care. Linchpin score reveals a novel aspect of network importance that can provide important insight into the vulnerability of health care provider networks. More broadly, applications of linchpin score may be relevant for the analysis of social networks where interdisciplinary collaboration is important.

Background Conversation aids can facilitate shared decision-making and improve patient-centered outcomes. However, few examples exist of sustained use of conversation aids in routine care due to numerous barriers at clinical and organizational levels. We explored factors that will promote the sustained use of two early-stage breast cancer conversation aids. We examined differences in opinions between the two conversation aids and across socioeconomic strata. Methods We nested this study within a randomized controlled trial that demonstrated the effectiveness of two early-stage breast cancer surgery conversation aids, one text-based and one picture-based. These conversation aids facilitated more shared decision-making and improved the decision process, among other outcomes, across four health systems with socioeconomically diverse patient populations. We conducted semi-structured interviews with a purposive sample of patient participants across conversation aid assignment and socioeconomic status (SES) and collected observations and field notes. We interviewed trial surgeons and other stakeholders. Two independent coders conducted framework analysis using the NOrmalization MeAsure Development through Normalization Process Theory. We also conducted an inductive analysis. We conducted additional sub-analyses based on conversation aid assignment and patient SES. Results We conducted 73 semi-structured interviews with 43 patients, 16 surgeons, and 14 stakeholders like nurses, cancer center directors, and electronic health record (EHR) experts. Patients and surgeons felt the conversation aids should be used in breast cancer care in the future and were open to various methods of giving and receiving the conversation aid (EHR, email, patient portal, before consultation). Patients of higher SES were more likely to note the conversation aids influenced their treatment discussion, while patients of lower SES noted more influence on their decision-making. Intervention surgeons reported using the conversation aids did not lengthen their typical consultation time. Most intervention surgeons felt using the conversation aids enhanced their usual care after using it a few times, and most patients felt it appeared part of their normal routine. Conclusions Key factors that will guide the future sustained implementation of the conversation aids include adapting to existing clinical workflows, flexibility of use, patient characteristics, and communication preferences. Trial registration ClinicalTrials.gov Identifier: NCT03136367, registered on May 2, 2017

Patient characteristics have been linked to prevalence and quality of shared decision-making (SDM) behaviors across diverse studies of varied size and focus. We aim to evaluate the extent to which patient characteristics are associated with patient-rated SDM scores as measured by collaboRATE and whether or not collaboRATE varies at the provider group level. We used the 2017 California Patient Assessment Survey data set, which included adult patients of 153 California-based medical groups receiving services between January and October 2016. Mixed-effects logistic regression evaluated relationships between collaboRATE scores and patient characteristics. We analyzed 31 265 total survey responses. Among included covariates, patients’ health status, race, primary language, and mode of survey response were significantly associated with collaboRATE scores. Case-mix adjustment is common in healthcare quality measurement and can be useful in pay-for-performance systems. For those use cases, we recommend adjusting collaboRATE scores by patients’ age, health status, gender, race, and language spoken at home, and survey response mode. However, when case-mix adjustment is not required, we suggest highlighting observed disparities across diverse patient populations to improve attention to inequities in patient experience.

Summary Two-stage instrumental variable methods are commonly used for estimating average causal effects in the presence of an unmeasured confounder. In the context of the proportional hazard Cox regression models, this problem has recently received attention with several methods being proposed. Previously, we developed an improved estimator under the incumbent two-stage residual inclusion procedure called ‘2SRI’ by adding a Gaussian frailty in the second stage. We now consider the more complex situation in which the treatment and the unmeasured confounders can have time varying effects, illustrating the method with the case of a step function with one prespecified change point. We prove that, in situations where the effects of the unmeasured confounder or the treatment change during the follow-up, the first stage of the 2SRI algorithm induces a frailty with time varying coefficients in the second stage, which enables incumbent methods and our previously developed procedure to be improved on. A Monte Carlo simulation study demonstrates the superior performance of the proposed extension of 2SRI that we develop. We apply the new procedure to estimate the effect of endarterectomy versus carotid artery stenting on the time to death of patients suffering from carotid artery disease by using linked vascular quality initiative registry–Medicare data.

BACKGROUND/OBJECTIVES Neurohormonal therapy, which includes beta-blockers and angiotensin-converting enzyme inhibitor/angiotensin receptor blockers (ACEi/ARBs), is the cornerstone of heart failure with reduced ejection fraction (HFrEF) treatment. While neurohormonal therapies have demonstrated efficacy in randomized clinical trials, older patients, which now comprise the majority of HFrEF patients, were underrepresented in those original trials. This study aimed to determine the association between short- (30 day) and long-term (1 year) mortality and the use of neurohormonal therapy in HFrEF patients, across the age spectrum. DESIGN/SETTING/PARTICIPANTS This is a population-based, retrospective, cohort study between January 2008 and December 2015. We used 100% Medicare Parts A and B and a random 40% sample of Part D to create a cohort of 295,494 fee-for-service beneficiaries with at least one hospitalization for HFrEF between 2008 and 2015. All analyses were performed between May 2019 and July 2020. EXPOSURE We used Part D data to determine exposure to beta-blocker and ACEi and ARB therapy. RESULTS We found that in 295,494 patients admitted for HFrEF between 2008 and 2015, the average age was 80 years, 54% were female and 17% were non-white. The baseline mortality rate was higher among those aged ≥85, but the mortality benefits of neurohormonal therapy were preserved across the age spectrum. Among those ≥85 years old, the hazard ratio for death within 30 days was 0.59 (95% confidence interval [CI] 0.56-0.62; p

BackgroundCosts of care are important to patients making cancer treatment decisions, but clinicians often do not feel prepared to discuss treatment costs. We aim to (1) assess the impact of a conversation-based decision aid (Option Grid) containing cost information about slow-growing prostate cancer management options, combined with urologic surgeon training, on the frequency and quality of patient-urologic surgeon cost conversations, and (2) examine the impact of the decision aid and surgeon training on decision quality.MethodsWe will conduct a stepped-wedge cluster randomized trial in outpatient urology practices affiliated with a large academic medical center in the USA. We will randomize five urologic surgeons to four intervention sequences and enroll their patients with a first-time diagnosis of slow-growing prostate cancer independently at each period. Primary outcomes include frequency of cost conversations, initiator of cost conversations, and whether or not a referral is made to address costs. These outcomes will be collected by patient report (post-visit survey) and by observation (audio-recorded clinic visits) with consent. Other outcomes include the following: patient-reported decisional conflict post-visit and at 3-month follow-up, decision regret at 3-month follow-up, shared decision-making post-visit, communication post-visit, and financial toxicity post-visit and at 3-month follow-up; clinician-reported attitudes about shared decision-making before and after the study, and feasibility of sustained intervention use. We will use hierarchical regression analysis to assess patient-level outcomes, including urologic surgeon as a random effect to account for clustering of patient participants.DiscussionThis study evaluates a two-part intervention to improve cost discussions between urologic surgeons and patients when deciding how to manage slow-growing prostate cancer. Establishing the effectiveness of the strategy under study will allow for its replication in other clinical decision contexts.Trial registrationClinicalTrials.govNCT04397016. Registered on 21 May 2020

IMPORTANCE: Knowledge of health outcomes among opioid-exposed infants is limited, particularly for those not diagnosed with neonatal opioid withdrawal syndrome (NOWS). OBJECTIVES: To describe infant mortality among opioid-exposed infants and identify how mortality risk differs in opioid-exposed infants with and without a diagnosis of NOWS compared with infants without opioid exposure. DESIGN, SETTING, AND PARTICIPANTS: A retrospective cohort study of maternal-infant dyads was conducted, linking health care claims with vital records for births from January 1, 2010, to December 31, 2014, with follow-up of infants until age 1 year (through 2015). Maternal-infant dyads were included if the infant was born in Texas at 22 to 43 weeks’ gestational age to a woman aged 15 to 44 years insured by Texas Medicaid. Data analysis was performed from May 2019 to October 2020. EXPOSURE: The primary exposure was prenatal opioid exposure, with infants stratified by the presence or absence of a diagnosis of NOWS during the birth hospitalization. MAIN OUTCOMES AND MEASURES: Risk of infant mortality (death at age

Background Estimation that employs instrumental variables (IV) can reduce or eliminate bias due to confounding. In observational studies, instruments result from natural experiments such as the effect of clinician preference or geographic distance on treatment selection. In randomized studies the randomization indicator is typically a valid instrument, especially if the study is blinded, e.g. no placebo effect. Estimation via instruments is a highly developed field for linear models but the use of instruments in time-to-event analysis is far from established. Various IV-based estimators of the hazard ratio (HR) from Cox’s regression models have been proposed. Methods We extend IV based estimation of Cox’s model beyond proportionality of hazards, and address estimation of a log-linear time dependent hazard ratio and a piecewise constant HR. We estimate the marginal time-dependent hazard ratio unlike other approaches that estimate the hazard ratio conditional on the omitted covariates. We use estimating equations motivated by Martingale representations that resemble the partial likelihood score statistic. We conducted simulations that include the use of copulas to generate potential times-to-event that have a given marginal structural time dependent hazard ratio but are dependent on omitted covariates. We compare our approach to the partial likelihood estimator, and two other IV based approaches. We apply it to estimation of the time dependent hazard ratio for two vascular interventions. Results The method performs well in simulations of a stepwise time-dependent hazard ratio, but illustrates some bias that increases as the hazard ratio moves away from unity (the value that typically underlies the null hypothesis). It compares well to other approaches when the hazard ratio is stepwise constant. It also performs well for estimation of a log-linear hazard ratio where no other instrumental variable approaches exist. Conclusion The estimating equations we propose for estimating a time-dependent hazard ratio using an IV perform well in simulations. We encourage the use of our procedure for time-dependent hazard ratio estimation when unmeasured confounding is a concern and a suitable instrumental variable exists.

Background Few studies have examined the best way to convey the probability of serious events occurring in the future (i.e., risk of stroke or death) to persons with low numeracy or graph literacy proficiency. To address this gap, we developed and user-tested a bar graph and compared it to icon arrays to assess its impact on understanding and preference for viewing risk information. Objectives To determine the: (i) formats’ impact on participants’ understanding of risk information; (ii) formats’ impact on understanding and format preference across numeracy and graph literacy subgroups; (iii) rationale supporting participants’ preference for each graphical display format. Methods An online sample (evenly made up of participants with high and low objective numeracy and graph literacy) was randomized to view either the icon array or the bar graph. Each format conveyed the risk of major stroke and death five years after choosing surgery, a stent, or medication to treat carotid artery stenosis. Participants answered questions to assess their understanding of the risk information. Lastly, both formats were presented in parallel, and participants were asked to identify their preferred format to view risk information and explain their preference. Results Of the 407 participants, 197 were assigned the icon array and 210 the bar graph. Understanding of risk information and format preference did not differ significantly between the two trial arms, irrespective of numeracy and graph literacy proficiency. High numeracy and graph literacy proficiency was associated with high understanding (p Conclusion We found no evidence to demonstrate the superiority of one format over another on understanding. The majority of participants preferred viewing the risk information using the bar graph format.

IntroductionFewer than half of all people in the USA have a documented advance care plan (ACP). Hospitalisation offers an opportunity for physicians to initiate ACP conversations. Despite expert recommendations, hospital-based physicians (hospitalists) do not routinely engage in these conversations, reserving them for the critically ill.The objective of this study is to test the effect of a novel behavioural intervention on the incidence of ACP conversations by hospitalists practicing at a stratified random sample of hospitals drawn from 220 US acute care hospitals staffed by a large, nationwide acute care physician practice with an ongoing ACP quality improvement initiative.Methods and analysisWe developedHopewell Hospitalist, a theory-based adventure video game, to modify physicians' attitudes towards ACP conversations and to increase their motivation for engaging in them. The planned study is a pragmatic stepped-wedge crossover phase III trial, testing the efficacy ofHopewell Hospitalistfor increasing ACP conversations. We will randomise 40 hospitals to the month (step) in which they receive the intervention. We aim to recruit 30 hospitalists from up to eight hospitals each step to complete the intervention, playingHopewell Hospitalistfor at least 2 hours. The primary outcome is ACP billing for patients aged 65 and older managed by participating hospitalists. We hypothesise that the intervention will increase ACP billing in the quarter after dissemination, and have 80% power to detect a 1% absolute increase and 99% power to detect a 3.5% absolute increase.Ethics and disseminationDartmouth’s Committee for the Protection of Human Subjects has approved the study protocol, which is registered on clinicaltrials.gov. We will disseminate the results through manuscripts and the trials website.Hopewell Hospitalistwill be made available on the iOS Application Store for download, free of cost, at the conclusion of the trial.Trial registration numberNCT04557930.

Objective Building on the original taxonomy of hospital-based health systems from 20 years ago, we develop a new taxonomy to inform emerging public policy and practice developments. Data sources The 2016 American Hospital Association's (AHA) Annual Survey; the 2016 IQVIA Healthcare Organizations and Systems (HCOS) database; and the 2017-2018 National Survey of Healthcare Organizations and Systems (NSHOS). Study design Cluster analysis of the 2016 AHA Annual Survey data to derive measures of differentiation, centralization, and integration to create categories or types of hospital-based health systems. Data collection Principal components factor analysis with varimax rotation generating the factors used in the cluster algorithms. Principal findings Among the four cluster types, 54% (N = 202) of systems are decentralized (-0.35) and relatively less differentiated (-0.37); 23% of systems (N = 85) are highly differentiated (1.28) but relatively decentralized (-0.29); 15% (N = 57) are highly centralized (2.04) and highly differentiated (0.65); and approximately 9 percent (N = 33) are least differentiated (-1.35) and most decentralized (-0.64). Despite differences in calculation, the Highly Centralized, Highly Differentiated System Cluster and the Undifferentiated, Decentralized System Cluster were similar to those identified 20 years ago. The other two system clusters contained similarities as well as differences from those 20 years ago. Overall, 82 percent of the systems remain relatively decentralized suggesting they operate largely as holding companies allowing autonomy to individual hospitals operating within the system. Conclusions The new taxonomy of hospital-based health systems bears similarities as well as differences from 20 years ago. Important applications of the taxonomy for addressing current challenges facing the healthcare system, such as the transition to value-based payment models, continued consolidation, and the growing importance of the social determinants of health, are highlighted.

Study objective Rates of admission from the emergency department (ED) vary widely across regions of the country, hospitals within regions, and physicians within hospitals. Our objective was to determine the extent to which variation in admission decisions was described by differences in admission rates at these 3 levels. This understanding will serve to better target interventions to modify rates of admission where appropriate. Methods In this cross-sectional observational cohort study, we analyzed Medicare fee-for-service claims for ED visits from 2012 to 2015 in a 20% random sample of beneficiaries. We first estimated the total regional-, hospital-, and physician-level variations in rates of admission and their proportions of the total variation after adjusting for patient and each level's covariates. We then estimated the extent to which each level's characteristics accounted for variation at that respective level. Results Our study sample included 5,778,218 visits with 45,491 physicians at 3,480 EDs across 306 hospital referral regions. The mean rate of admission was 38.9% and ranged from 21.4% to 53.0% for physicians at the 10th and 90th percentile of the distribution, respectively. The residual (unexplained) variations at the regional, hospital, and physician levels were 13.3% (95% confidence interval [CI], 11.2 to 15.5%), 60.1% (57.1 to 62.9%), and 26.7% (26.4 to 26.9%), respectively. Regional, hospital, and physician characteristics accounted for 9.1% (95% CI, -5.6 to 23.8%), 51.1% (48.8 to 53.5%), and 2.7% (1.3 to 4.1%), respectively, of the explained variation at their respective levels. Conclusion Within-area variation, both across hospitals within a region and across physicians within a hospital, is a more substantial component of observed variation in admission rates from the ED than regional level variation. These findings suggest that variation in admission rates is at least in part related to institutional norms and cultures as well as heterogeneity of physician decisionmaking within hospitals, both of which could be targets of interventions to modify rates of admission.

BACKGROUND: Fewer than half of the US population has an advance healthcare directive. Hospitalizations offer a key opportunity for clinicians to engage patients in advance care planning (ACP) conversations. Guidelines suggest screening for the presence of “serious illness” but do not further specify how to prioritize the 12.4 million patients hospitalized each year. OBJECTIVE: To establish a normative standard for prioritizing hospitalized patients for ACP conversations. DESIGN AND SETTING: A modified Delphi study, with three iterative rounds of online surveys. PARTICIPANTS: Multi-disciplinary group of US-based clinicians with research and practical expertise in ACP. MAIN MEASURES: Indirect and direct elicitation of short-term and 1-year risk of mortality that prompt experts to prioritize ACP conversations for hospitalized adults. MAIN RESULTS: Fifty-seven of 108 (52%) candidate panelists completed round 1, and 47 completed rounds 2 and 3. Panelists were primarily physicians (84%), with significant experience (mean years 23 [SD 9.8]), who either taught (55%) and/or performed research about ACP (55%). In round 1, > 70% of panelists agreed that all hospitalized adults ≥ 65 years should have an ACP conversation before discharge, but disagreed about the timing and content of the conversation. By round 3, > 70% of participants agreed that patients with either high (> 10%) short-term or high (≥ 34%) 1-year risk of mortality should have a goals of care conversation (i.e., focused on preferences for near-term treatment), while patients with low (≤ 10%) short-term and low (< 19%) 1-year risk of mortality warranted an ACP conversation (i.e., focused on preferences for future care) before discharge. LIMITATIONS: Use of case vignettes to elicit clinician judgment; response rate. CONCLUSIONS: Panelists agreed that clinicians should have an ACP conversation with all hospitalized adults over 65 years in an ACP conversation, adjusting the content and timing of the conversation conditional on the patient’s risk of short-term and 1-year mortality. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (10.1007/s11606-020-06150-0) contains supplementary material, which is available to authorized users.

In the United States the number of health systems that own practices or hospitals have increased in number and complexity leading to interest in assessing the relationship between health organization factors and health outcomes. However, the existence of multiple types of organizations combined with the nesting of some hospitals and practices within health systems and the nesting of some health systems within larger health systems generates numerous analytic objectives and complicates the construction of optimal survey designs. An objective function that explicitly weighs all objectives is theoretically appealing but becomes unwieldy and increasingly ad hoc as the number of objectives increases. To overcome this problem, we develop an alternative approach based on constraining the sampling design to satisfy desired statistical properties. For example, to support evaluations of the comparative importance of factors measured in different surveys on health system performance, a constraint that requires at least one organization of each type (corporate owner, hospital, practice) to be sampled whenever any component of a system is sampled may be enforced. Multiple such constraints define a nonlinear system of equations that "couples" the survey sampling designs whose solution yields the sample inclusion probabilities for each organization in each survey. A Monte Carlo algorithm is developed to solve the simultaneous system of equations to determine the sampling probabilities and extract the samples for each survey. We illustrate the new sampling methodology by developing the constraints and solving the ensuing systems of equations to obtain the sampling design for the National Surveys of United States Health Care Systems, Hospitals and Practices. We illustrate the virtues of "coupled sampling" by comparing the proportion of eligible systems for whom the corporate owner and both a hospital and a practice that are expected to be sampled to that expected under alternative sampling designs. Comparative and descriptive analyses that illustrate features of the sampling design are also presented.

Background Machine learning approaches have become increasingly popular modeling techniques, relying on data-driven heuristics to arrive at its solutions. Recent comparisons between these algorithms and traditional statistical modeling techniques have largely ignored the superiority gained by the former approaches due to involvement of model-building search algorithms. This has led to alignment of statistical and machine learning approaches with different types of problems and the under-development of procedures that combine their attributes. In this context, we hoped to understand the domains of applicability for each approach and to identify areas where a marriage between the two approaches is warranted. We then sought to develop a hybrid statistical-machine learning procedure with the best attributes of each. Methods We present three simple examples to illustrate when to use each modeling approach and posit a general framework for combining them into an enhanced logistic regression model building procedure that aids interpretation. We study 556 benchmark machine learning datasets to uncover when machine learning techniques outperformed rudimentary logistic regression models and so are potentially well-equipped to enhance them. We illustrate a software package, InteractionTransformer, which embeds logistic regression with advanced model building capacity by using machine learning algorithms to extract candidate interaction features from a random forest model for inclusion in the model. Finally, we apply our enhanced logistic regression analysis to two real-word biomedical examples, one where predictors vary linearly with the outcome and another with extensive second-order interactions. Results Preliminary statistical analysis demonstrated that across 556 benchmark datasets, the random forest approach significantly outperformed the logistic regression approach. We found a statistically significant increase in predictive performance when using hybrid procedures and greater clarity in the association with the outcome of terms acquired compared to directly interpreting the random forest output. Conclusions When a random forest model is closer to the true model, hybrid statistical-machine learning procedures can substantially enhance the performance of statistical procedures in an automated manner while preserving easy interpretation of the results. Such hybrid methods may help facilitate widespread adoption of machine learning techniques in the biomedical setting.

Background Women of lower socioeconomic status (SES) with early‐stage breast cancer are more likely to report poorer physician‐patient communication, lower satisfaction with surgery, lower involvement in decision making, and higher decision regret compared to women of higher SES. The objective of this study was to understand how to support women across socioeconomic strata in making breast cancer surgery choices. Methods We conducted a 3‐arm (Option Grid, Picture Option Grid, and usual care), multisite, randomized controlled superiority trial with surgeon‐level randomization. The Option Grid (text only) and Picture Option Grid (pictures plus text) conversation aids were evidence‐based summaries of available breast cancer surgery options on paper. Decision quality (primary outcome), treatment choice, treatment intention, shared decision making (SDM), anxiety, quality of life, decision regret, and coordination of care were measured from T0 (pre‐consultation) to T5 (1‐year after surgery. Results Sixteen surgeons saw 571 of 622 consented patients. Patients in the Picture Option Grid arm (n = 248) had higher knowledge (immediately after the visit [T2] and 1 week after surgery or within 2 weeks of the first postoperative visit [T3]), an improved decision process (T2 and T3), lower decision regret (T3), and more SDM (observed and self‐reported) compared to usual care (n = 257). Patients in the Option Grid arm (n = 66) had higher decision process scores (T2 and T3), better coordination of care (12 weeks after surgery or within 2 weeks of the second postoperative visit [T4]), and more observed SDM (during the surgical visit [T1]) compared to usual care arm. Subgroup analyses suggested that the Picture Option Grid had more impact among women of lower SES and health literacy. Neither intervention affected concordance, treatment choice, or anxiety. Conclusions Paper‐based conversation aids improved key outcomes over usual care. The Picture Option Grid had more impact among disadvantaged patients. Lay Summary The objective of this study was to understand how to help women with lower incomes or less formal education to make breast cancer surgery choices.Compared with usual care, a conversation aid with pictures and text led to higher knowledge. It improved the decision process and shared decision making (SDM) and lowered decision regret. A text‐only conversation aid led to an improved decision process, more coordinated care, and higher SDM compared to usual care. The conversation aid with pictures was more helpful for women with lower income or less formal education.Conversation aids with pictures and text helped women make better breast cancer surgery choices., A paper‐based pictorial conversation aid (pictures plus text) is beneficial to all patients with early‐stage breast cancer and particularly to disadvantaged patients. Between‐surgeon variation suggests that the maximal impact of such interventions requires standardized physician training combined with these interventions.

Background The coronavirus disease (COVID-19) has led to dramatic changes worldwide in people’s everyday lives. To combat the pandemic, many governments have implemented social distancing, quarantine, and stay-at-home orders. There is limited research on the impact of such extreme measures on mental health. Objective The goal of this study was to examine whether stay-at-home orders produced differential changes in mental health symptoms using internet search queries on a national scale. Methods In the United States, individual states vary in their adoption of measures to reduce the spread of COVID-19; as of March 23, 2020, 11 of the 50 states had issued stay-at-home orders. The staggered rollout of stay-at-home measures across the United States allows us to investigate whether these measures impact mental health by exploring variations in mental health search queries across the states. This paper examines the changes in mental health search queries on Google between March 16-23, 2020, across each state and Washington, DC. Specifically, this paper examines differential changes in mental health searches based on patterns of search activity following issuance of stay-at-home orders in these states compared to all other states. The participants were all the people who searched mental health terms in Google between March 16-23. Between March 16-23, 11 states underwent stay-at-home orders to prevent the transmission of COVID-19. Outcomes included search terms measuring anxiety, depression, obsessive-compulsive, negative thoughts, irritability, fatigue, anhedonia, concentration, insomnia, and suicidal ideation. Results Analyzing over 10 million search queries using generalized additive mixed models, the results suggested that the implementation of stay-at-home orders are associated with a significant flattening of the curve for searches for suicidal ideation, anxiety, negative thoughts, and sleep disturbances, with the most prominent flattening associated with suicidal ideation and anxiety. Conclusions These results suggest that, despite decreased social contact, mental health search queries increased rapidly prior to the issuance of stay-at-home orders, and these changes dissipated following the announcement and enactment of these orders. Although more research is needed to examine sustained effects, these results suggest mental health symptoms were associated with an immediate leveling off following the issuance of stay-at-home orders.

BackgroundWhile recent research efforts to reduce pressure ulcers in the clinical context have focused on key retrospective characteristics, little work has focused on creating real-time predictive models to prevent this avoidable hospital-acquired injury. Furthermore, existing machine learning heuristics often fail to surpass traditional statistical models or provide individual-level risk assessments with explanations for each patient. Thus, we sought to compare the predictive performance of five machine learning and traditional statistical modeling techniques to predict the occurrence of Hospital Acquired Pressure Injuries (HAPI).MethodsElectronic Medical Record (EMR) information was collected from 57,227 hospitalizations, containing 241 positive HAPI cases, acquired from Dartmouth Hitchcock Medical Center from April 2011 to December 2016. The five classifiers were trained to predict HAPI incidence and performance was assessed using the C-statistic or Area Under the Receiver Operating Curve (AUC).ResultsLogistic Regression was the best modeling approach (AUC=0.91±0.034). We report discordance between predictors deemed important by the machine learning models compared to traditional statistical model. We provide means to visually assess factors important to every patient’s prediction, regardless of the modeling approach, through Shapley Additive Explanations.ConclusionsMachine learning models will continue to inform decision making processes but should be compared to traditional modeling approaches to ensure proper utilization. Disagreements between important predictors found by traditional and machine learning modeling approaches can potentially confuse clinicians and as such need to be reconciled. Future efforts to analyze time-stamped, prospective medical record data will be enhanced by patient-specific details. These developments represent important steps forward in developing real-time predictive models that can be integrated and readily deployed in electronic medical record systems to reduce unnecessary harm.

To determine if two encounter conversation aids for early-stage breast cancer surgery increased observed and patient-reported shared decision making (SDM) compared with usual care and if observed and patient-reported SDM were associated.Surgeons in a cluster randomized trial at four cancer centers were randomized to use an Option Grid, Picture Option Grid, or usual care. We used bivariate statistics, linear regression, and multilevel models to evaluate the influence of trial arm, patient socioeconomic status and health literacy on observed SDM (via OPTION-5) and patient-reported SDM (via collaboRATE).From 311 recordings, OPTION-5 scores were 73/100 for Option Grid (n = 40), 56.3/100 for Picture Option Grid (n = 144), and 41.0/100 for usual care (n = 127; p0.0001). Top collaboRATE scores were 81.6 % for Option Grid, 80.0 % for Picture Option Grid, and 56.4 % for usual care (p0.001). Top collaboRATE scores correlated with an 8.60 point (95 %CI 0.66, 13.7) higher OPTION-5 score (p = 0.008) with no correlation in the multilevel analysis. Patients of lower socioeconomic status had lower OPTION-5 scores before accounting for clustering.Both conversation aids led to meaningfully higher observed and patient-reported SDM. Observed and patient-reported SDM were not strongly correlated.Healthcare providers could implement these conversation aids in real-world settings.

Hierarchical modeling is the preferred approach of modeling neighborhood effects. When both residential and workplace neighborhoods are known, a bivariate (residential-workplace) neighborhood random effect that quantifies the extent that a neighborhood's residential and workplace effects are correlated may be modeled. However, standard statistical software for hierarchical models does not easily allow correlations between the random effects of distinct clustering variables to be incorporated. To overcome this challenge, we develop a Bayesian model and an accompanying estimation procedure that allows for correlated bivariate neighborhood effects and allows individuals to reside or work in multiple neighborhoods, cross-sectional and longitudinal heterogeneity between individuals, and serial correlation between repeated observations over time. Simulation studies that vary key model parameters evaluate how well each aspect of the model is identified by the data. We apply the model to the motivating Framingham Heart Study linked food establishment data to examine whether (i) proximity to fast-food establishments is associated with body mass index, (ii) workplace neighborhood exposure associations are larger than those for residential neighborhood exposure, and (iii) residential neighborhood exposure associations correlate with workplace neighborhood exposure. Comparisons of the full model to models with restricted versions of the covariance structure illustrate the impact of including each feature of the covariance structure.

A property and criticism of instrumental variable (IV) estimators are that they estimate the exposure effect in the compliers; individuals whose exposure is influenced by the instrument. It is conceivable that the exposure effect is different in individuals whose exposure is not influenced by the instrument. For that reason, it is useful to know who the compliers are. We present methods for reporting averages and other statistics, and comparing them between the compliers and non-compliers, which are applicable to scenarios in which the instrument and exposure are fixed in time and dichotomous. The methods are illustrated in a comparison of outcomes between teaching and nonteaching hospitals in patients undergoing brain or spinal surgery.

At the 2018 International Conference on Health Policy Statistics (ICHPS) held in Charleston, South Carolina, Anirban Basu was awarded the Mid-Career Excellence Award from the American Statistical Association Section on Health Policy Statistics (HPSS). Anirban was exceptionally and uniquely qualified for this award. Highlights include his providing outstanding service to the HPSS, advancing statistical methodology, advancing methodology in other domains of health policy, and performing extensive and highly impactful applied work in medicine and health care. In this interview, we trace Anirban’s upbringing, schooling, early career, and mid-career phases to gain insights into his success. We also sought his opinions on salient topics or issues.

Objective To compare standard methods for constructing physician networks from patient-physician encounter data with a new method based on clinical episodes of care. Data source We used data on 100% of traditional Medicare beneficiaries from 51 nationally representative geographical regions for the years 2005–2010. Study design We constructed networks of physicians based on their shared patients. In the fixed-threshold networks and adaptive-threshold networks, we included data on all patient-physician encounters to form the physician-physician ties, and then subsequently thresholded some proportion of the strongest ties. In contrast, in the episode-based approach, only those patient-physician encounters that occurred within shared clinical episodes treating specific conditions contributed towards physician-physician ties. Data collection/extraction methods We extracted clinical episodes in the Medicare data and investigated structural properties of the patient-sharing networks of physicians, temporal dynamics of their ties, and temporal stability of network communities across the two approaches. Principal findings The episode-based networks accentuated ties between primary care physicians (PCPs) and medical specialists, had ties that were more likely to reappear in the future, and appeared to have more fluid community structure. Conclusions Constructing physician networks around shared episodes of care is a clinically sound alternative to previous approaches to network construction that does not require arbitrary decisions about thresholding. The resulting networks capture somewhat different aspects of patient-physician encounters. Electronic supplementary material The online version of this article (10.1007/s41109-018-0084-1) contains supplementary material, which is available to authorized users.

In this paper, we analyze the millions of referral paths of patients’ interactions with the healthcare system for each year in the 2006-2011 time period and relate them to U.S. cardiovascular treatment records. For a patient, a “referral path” records the chronological sequence of physicians encountered by a patient (subject to certain constraints on the times between encounters). It provides a basic unit of analysis in a broader referral network that encodes the flow of patients and information between physicians in a healthcare system. We consider referral networks defined over a range of interactions as well as the characteristics of referral paths, producing a characterization of the various networks as well as the physicians they comprise. We further relate these metrics and findings to outcomes in the specific area of cardiovascular care. In particular, we match a referral path to occurrences of Acute Myocardial Infarction (AMI) and use the summary measures of the referral path to predict the treatment a patient receives and medical outcomes following treatment. Some referral path features are more significant with respect to their ability to boost a tree-based predictive model, and have stronger correlations with numerical treatment outcome variables. The patterns of referral paths and the derived informative features illustrate the potential for using network science to optimize patient referrals in healthcare systems for improved treatment outcomes and more efficient utilization of medical resources.

While higher lead surgeon volume has been associated with lower mortality following open abdominal aortic aneurysm (AAA) repair, little is known about the impact of using an attending surgeon as assistant surgeon. The aim of this study was to determine whether the presence of an assistant surgeon, particularly a high volume assistant, mitigates the relationship between lead surgeon volume and outcomes.All Medicare beneficiaries who underwent intact, open AAA repair between 2003 and 2008 were evaluated and nested regression models were constructed to evaluate the relationship between surgeon and assistant volume and peri-operative mortality, adjusting for comorbid conditions and hospital volume.In total 28,590 repairs were studied, of which 19,284 (67.5%) were performed by a single surgeon and 9306 (32.5%) included an assistant surgeon. Of cases with an assistant, 12.3% included a high volume assistant surgeon. Lower volume surgeons more frequently used an assistant (lead surgeon Q1 volume: 40%; Q2: 36%; Q3: 34%; Q4: 29%; Q5: 27% [p .01]). In cases with no assistant, adjusted peri-operative mortality varied monotonically with surgeon volume (Q1: 4.7%; Q2: 4.4%; Q3: 4.1%; Q4: 3.3%; Q5: 3.2%). However, the use of a high or a low volume assistant surgeon, compared with no attending surgeon as assistant, was not associated with lower peri-operative mortality in any lead surgeon volume quintile, even among those operations performed by the lowest volume lead surgeons.Employing an assistant surgeon does not improve outcomes amongst any quintile of volume of the lead surgeon. As surgeons perform fewer open AAA repairs in the modern era, these data imply that even the help of a high volume assistant surgeon may not mitigate the detrimental effect of a lower volume surgeon.

Evidence-based smoking cessation treatment is safe and effective for smokers with serious mental illness (SMI); however, quit rates remain low in this population. We explored how social networks influence smoking outcomes among people with SMI who participated in smoking cessation treatment at community mental health centers. We conducted egocentric social network interviews with 41 individuals with SMI who participated in a statewide Medicaid demonstration project of smoking cessation treatment. We estimated the proportions of current smokers, former smokers, and never smokers in participants' (i.e., egos') networks and examined support for quitting from alters, defined as family, friends, peers, or significant others. We used logistic regression and mixed-effect logistic regression to explore the relationship between social network variables and ego's smoking status following cessation treatment. Egos reported an average of 5.9 ± 2.2 alters; 52% of alters were current smokers; 18% were former smokers; and 30% never smoked. The majority (57%) of alters helped an ego quit smoking. The presence of former smokers in the network was associated with decreased odds that the ego was smoking post-treatment. Egos whose friends were highly interconnected were less likely to smoke after treatment. Former smokers in the network may be a valuable resource for quitting particularly for vulnerable groups where there is a high prevalence and acceptability of smoking. Our initial findings suggest that a highly interconnected friendship network structure may be beneficial for people with SMI who are trying to quit smoking.

Increased renal complications have been suggested with suprarenal stent grafts, but long-term analyses have been limited. Therefore, the purpose of this study was to evaluate the effect of endograft choice on perioperative and long-term outcomes.We compared Medicare beneficiaries undergoing endovascular abdominal aortic aneurysms repair from 2005 to 2008 with endografts with infrarenal fixation and a single docking limb (AneuRx, Excluder) to those with suprarenal fixation and 2 docking limbs (Zenith), or a unibody configuration (Powerlink). Propensity score weighting accounted for differences in patient characteristics among the different graft formations, and perioperative mortality, complications, and length of stay and 4-year rates of survival, rupture, and reintervention were compared.Forty-six thousand one hundred seventy-one Medicare beneficiaries were identified including 11,002 (24%) with suprarenal fixation, 32,909 (71%) with infrarenal fixation, and 2,260 (5%) with a unibody graft. After propensity score weighting, there were no significant differences in patients' baseline clinical and demographic characteristics. The suprarenal fixation patients had higher rates of perioperative mortality (1.7% vs. 1.3%, P 0.01), renal failure (6.0% vs. 4.7%, P 0.001), and mesenteric ischemia (0.7% vs. 0.4%, P 0.01) and longer length of stay (3.4 days vs. 3.0 days, P 0.001) compared with patients with infrarenal fixation. Unibody grafts had higher rates of renal failure (5.9% vs. 4.7%, P 0.001), mesenteric ischemia (1.0% vs. 0.4%, P 0.001), and conversion to open repair (0.7% vs. 0.1%, P 0.001) compared to those with infrarenal fixation and single docking limbs. At 4 years, mortality remained slightly higher with suprarenal compared to infrarenal fixation (30% vs. 29%, P = 0.047), although these patients had fewer conversions to open repair (0.6% vs. 0.9%, P = 0.03) and aneurysm-related reinterventions (10% vs. 12%, P 0.01). At 4 years, unibody grafts had more aneurysm-related interventions compared to infrarenal fixation grafts (15% vs. 12%, P 0.01) but fewer conversions to open repair (0.4% vs. 0.9%, P = 0.02). Late rupture did not differ among the groups.Compared to infrarenal fixation devices, patients who underwent EVAR with suprarenal fixation had higher perioperative mortality and renal complications but fewer reinterventions including conversion, while the unibody graft had more perioperative complications and aneurysm-related reinterventions, but fewer conversions to open repair. Although these differences could be explained by selection bias, these data suggest that further comparative effectiveness analyses should be performed to understand the outcomes following EVAR with suprarenal fixation and unibody grafts.

In this paper, we analyze the US Patient Referral Network (also called the Shared Patient Network) and various subnetworks for the years 2009 to 2015. In these networks, two physicians are linked if a patient encounters both of them within a specified time interval, according to the data made available by the Centers for Medicare and Medicaid Services. We find power law distributions on most state-level data as well as a core-periphery structure. On a national and state level, we discover a so-called small-world structure as well as a "gravity law" of the type found in some large-scale economic networks. Some physicians play the role of hubs for interstate referral. Strong correlations between certain network statistics with health care system statistics at both the state and national levels are discovered. The patterns in the referral network evinced using several statistical analyses involving key metrics derived from the network illustrate the potential for using network analysis to provide new insights into the health care system and opportunities or mechanisms for catalyzing improvements.

Background Research has explored associations between diet, body weight, and the food environment; however, few studies have examined historical trends in food environments. Methods In the Framingham Heart Study Offspring (N = 3321) and Omni (N = 447) cohorts, we created food environment metrics in four Massachusetts towns utilizing geocoded residential, workplace, and food establishment addresses from 1971 to 2008. We created multilevel models adjusted for age, sex, education, and census tract poverty to examine trends in home, workplace, and commuting food environments. Results Proximity to and density of supermarkets, fast-food, full service restaurants, convenience stores, and bakeries increased over time for residential, workplace, and commuting environments; exposure to grocery stores decreased. The greatest increase in access was for supermarkets, with residential distance to the closest supermarket 1406 m closer (95% CI 1303 m, 1508 m) by 2005–2008 than in 1971–1975. Although poorer census tracts had higher access to fast-food restaurants consistently across follow-up, this disparity dissipated over time, due to larger increases in proximity to fast-food in wealthier neighborhoods. Conclusions Access to most food establishment types increased over time, with similar trends across home, workplace, and commuter environments. Electronic supplementary material The online version of this article (doi:10.1186/s12966-017-0537-4) contains supplementary material, which is available to authorized users.

Higher hospital and surgeon volumes are independently associated with improved mortality after open repair of abdominal aortic aneurysms (AAAs) in the era before endovascular AAA repair (EVAR). The effects of both surgeon and hospital volume on mortality after EVAR and open repair in the current era are less well defined.We studied Medicare beneficiaries who underwent elective AAA repair from 2001 to 2008. Volume was measured by procedure type during the 1-year period preceding each procedure and was further categorized into quintiles of volume for surgeon and hospital. Multilevel logistic regression models were used to evaluate the effect of surgeon volume, accounting for hospital volume, on mortality after adjusting for patient demographic and comorbid conditions as well as the analogous effect of hospital volume adjusting for surgeon volume. The multilevel models included random effects for surgeon and hospital to account for the clustering of multiple patients within the same surgeon and within the same hospital.We studied 122,495 patients who underwent AAA repair (open: 45,451; EVAR: 77,044). After EVAR, perioperative mortality did not differ by surgeon volume (quintile 1 [0-6 EVARs]: 1.8%; quintile 5 [28-151 EVARs]: 1.6%; P = .29), but decreased with greater hospital volume (quintile 1 [0-9 EVARs]: 1.9%; quintile 5 [49-198 EVARs]: 1.4%; P .01). After open repair, perioperative mortality decreased with both higher surgeon volume (quintile 1 [0-3 open repairs]: 6.4%; quintile 5 [14-62 open repairs]: 3.8%; P .01) and hospital volume (quintile 1 [0-5 open repairs]: 6.3%; quintile 5 [14-62 open repairs]: 3.8%; P .01). After adjustment for other predictors, surgeon volume was not associated with perioperative mortality after EVAR (odds ratio [OR], 0.9; 95% confidence interval [CI], 0.7-1.1); however, hospital volume was associated with higher perioperative mortality (quintile 1: OR, 1.5; 95% CI, 1.2-1.9; quintile 2: OR, 1.3; 95% CI, 1.02-1.6; and quintile 3: OR, 1.2; 95% CI, 1.01-1.5, compared with 5). After open repair, higher surgeon volume was also associated with lower mortality (quintile 1: OR, 1.5; 95% CI, 1.3-1.8; quintile 2: OR, 1.3; 95% CI, 1.1-1.6; and quintile 3: OR, 1.2; 95% CI, 1.1-1.4, compared with 5). Risk of mortality also was higher for patients treated at lower-volume hospitals (quintile 1: OR, 1.3; 95% CI, 1.1-1.5; quintile 2: OR, 1.3; 95% CI, 1.1-1.5; and quintile 3: OR, 1.2; 95% CI, 1.1-1.4, compared with 5).After EVAR, hospital volume is minimally associated with perioperative mortality, with no such association for surgeon volume. After open AAA repair, surgeon and hospital volume are both strongly associated with mortality. These findings suggest that open surgery should be concentrated in hospitals and surgeons with high volume.

OBJECTIVE The comparative effectiveness of the 2 treatment options—surgical clipping and endovascular coiling—for unruptured cerebral aneurysms remains an issue of debate and has not been studied in clinical trials. The authors investigated the association between treatment method for unruptured cerebral aneurysms and outcomes in elderly patients. METHODS The authors performed a cohort study of 100% of Medicare fee-for-service claims data for elderly patients who had treatment for unruptured cerebral aneurysms between 2007 and 2012. To control for measured confounding, the authors used propensity score conditioning and inverse probability weighting with mixed effects to account for clustering at the level of the hospital referral region (HRR). An instrumental variable (regional rates of coiling) analysis was used to control for unmeasured confounding and to create pseudo-randomization on the treatment method. RESULTS During the study period, 8705 patients underwent treatment for unruptured cerebral aneurysms and met the study inclusion criteria. Of these patients, 2585 (29.7%) had surgical clipping and 6120 (70.3%) had endovascular coiling. Instrumental variable analysis demonstrated no difference between coiling and clipping in 1-year postoperative mortality (OR 1.25, 95% CI 0.68–2.31) or 90-day readmission rate (OR 1.04, 95% CI 0.66–1.62). However, clipping was associated with a greater likelihood of discharge to rehabilitation (OR 6.39, 95% CI 3.85–10.59) and 3.6 days longer length of stay (LOS; 95% CI 2.90–4.71). The same associations were present in propensity score–adjusted and inverse probability–weighted models. CONCLUSIONS In a cohort of Medicare patients, there was no difference in mortality and the readmission rate between clipping and coiling of unruptured cerebral aneurysms. Clipping was associated with a higher rate of discharge to a rehabilitation facility and a longer LOS.

Shared decision-making (SDM) between clinicians and patients is a key component of patient experience, but measurement efforts have been hampered by a lack of valid and reliable measures that are feasible for routine use. In this study, we aim to investigate collaboRATE’s reliability, calculate required sample sizes for reliable measurement, and compare Consumer Assessment of Healthcare Providers and Systems (CAHPS) patient experience survey items to collaboRATE. CollaboRATE’s provider group-level reliability reached acceptable reliability at 190 patient reports, while the CAHPS SDM measure demonstrated similar reliability at a sample size of 124. The CAHPS communication measure reached acceptable reliability with 55 patient reports. A strong correlation was observed between collaboRATE and CAHPS communication measures ( r = 0.83). As a reliable measure of SDM, collaboRATE may be useful for both building payment models that support shared clinical decision-making and encouraging data transparency with regard to provider group performance.

Proportional hazard Cox regression models are frequently used to analyze the impact of different factors on time-to-event outcomes. Most practitioners are familiar with and interpret research results in terms of hazard ratios. Direct differences in survival curves are, however, easier to understand for the general population of users and to visualize graphically. Analyzing the difference among the survival curves for the population at risk allows easy interpretation of the impact of a therapy over the follow-up. When the available information is obtained from observational studies, the observed results are potentially subject to a plethora of measured and unmeasured confounders. Although there are procedures to adjust survival curves for measured covariates, the case of unmeasured confounders has not yet been considered in the literature. In this article we provide a semi-parametric procedure for adjusting survival curves for measured and unmeasured confounders. The method augments our novel instrumental variable estimation method for survival time data in the presence of unmeasured confounding with a procedure for mapping estimates onto the survival probability and the expected survival time scales.

Introduction. We aimed to conduct a multinational cross-sectional online survey of medical students’ attitudes toward, knowledge of, and experience with shared decision making (SDM). Methods. We conducted the survey from September 2016 until May 2017 using the following: 1) a convenience sample of students from four medical schools each in Canada, the United States, and the Netherlands ( n = 12), and 2) all medical schools in the United Kingdom through the British Medical School Council ( n = 32). We also distributed the survey through social media. Results. A total of 765 students read the information sheet and 619 completed the survey. Average age was 24, 69% were female. Mean SDM knowledge score was 83.6% (range = 18.8% to 100%; 95% confidence interval [CI] = 82.8% to 84.5%). US students had the highest knowledge scores (86.2%, 95% CI = 84.8% to 87.6%). The mean risk communication score was 57.4% (range = 0% to 100%; 95% CI = 57.4% to 60.1%). Knowledge did not vary with age, race, gender, school, or school year. Attitudes were positive, except 46% believed SDM could only be done with higher educated patients, and 80.9% disagreed that physician payment should be linked to SDM performance (increased with years in training, P < 0.05). Attitudes did not vary due to any tested variable. Students indicated they were more likely than experienced clinicians to practice SDM (72.1% v. 48.8%). A total of 74.7% reported prior SDM training and 82.8% were interested in learning more about SDM. Discussion. SDM knowledge is high among medical students in all four countries. Risk communication is less well understood. Attitudes indicate that further research is needed to understand how medical schools deliver and integrate SDM training into existing curricula.

Key Points Question Is the concurrent receipt of 2 fracture-associated drugs associated with increased risk of hip fracture? Findings In this cohort study, 11 million person-years of Medicare data revealed that the receipt of multiple fracture-associated drugs was common among older US residents. Concurrent receipt of 2 or more such drugs was associated with a more than 2-fold increase in hip fracture risk, with some specific combinations appearing especially hazardous, including commonly prescribed drugs, such as opioids, antidepressants, and sedatives. Meaning If confirmed, results of this cohort study suggest caution when combining fracture-associated medications, especially when use is discretionary, alternatives exist, or baseline fracture risk is high., This cohort study estimates hip fracture risk associated with concurrent exposure to multiple fracture-associated drugs., Importance Many prescription drugs increase fracture risk, which raises concern for patients receiving 2 or more such drugs concurrently. Logic suggests that risk will increase with each additional drug, but the risk of taking multiple fracture-associated drugs (FADs) is unknown. Objective To estimate hip fracture risk associated with concurrent exposure to multiple FADs. Design, Setting, and Participants This cohort study used a 20% random sample of Medicare fee-for-service administrative data for age-eligible Medicare beneficiaries from 2004 to 2014. Sex-stratified Cox regression models estimated hip fracture risk associated with current receipt of 1, 2, or 3 or more of 21 FADs and, separately, risk associated with each FAD and 2-way FAD combination vs no FADs. Models included sociodemographic characteristics, comorbidities, and use of non-FAD medications. Analyses began in November 2018 and were completed April 2019. Exposure Receipt of prescription FADs. Main Outcomes and Measures Hip fracture hospitalization. Results A total of 11.3 million person-years were observed, reflecting 2 646 255 individuals (mean [SD] age, 77.2 [7.3] years, 1 615 613 [61.1%] women, 2 136 585 [80.7%] white, and 219 579 [8.3%] black). Overall, 2 827 284 person-years (25.1%) involved receipt of 1 FAD; 1 322 296 (11.7%), 2 FADs; and 954 506 (8.5%), 3 or more FADs. In fully adjusted, sex-stratified models, an increase in hip fracture risk among women was associated with the receipt of 1, 2, or 3 or more FADs (1 FAD: hazard ratio [HR], 2.04; 95% CI, 1.99-2.11; P

Background The long‐term effectiveness of atherectomy treatment for peripheral arterial disease is unknown. We studied 5‐year clinical outcomes by endovascular treatment type among patients with peripheral arterial disease. Methods and Results We queried the Medicare‐linked VQI (Vascular Quality Initiative) registry for endovascular interventions from 2010 to 2015. The exposure was treatment type: atherectomy (with or without percutaneous transluminal angioplasty [ PTA] ), stent (with or without PTA ), or PTA alone. The outcomes were major amputation, any amputation, and major adverse limb event (major amputation or any reintervention). We used the center‐specific proportions of atherectomy procedures performed in the 12 months before a patient's procedure as the instruments to perform an instrumental‐variable Cox model analysis. Among 16 838 eligible patients (median follow‐up: 1.3–1.5 years), 11% underwent atherectomy, 40% received PTA alone, and 49% underwent stenting. Patients receiving atherectomy commonly underwent femoropopliteal artery treatment (atherectomy: 65%; PTA : 49%; stenting: 43%; P PTA : 68%; stenting: 67%; P PTA and 32% for stenting (log rank P PTA , except for a higher risk of any amputation (hazard ratio: 1.51; 95% CI , 1.08–2.13). However, compared with stenting, atherectomy patients had a higher risk of major amputation (hazard ratio: 3.66; 95% CI , 1.72–7.81), any amputation (hazard ratio: 2.73; 95% CI , 1.60–4.76), and major adverse limb event (hazard ratio: 1.61; 95% CI , 1.10–2.38). Conclusions Atherectomy is used to treat severe femoropopliteal and tibial peripheral arterial disease even though long‐term adverse outcomes occur more frequently after this treatment modality.

Rationale HIV-related stigma profoundly affects the physical and social wellbeing of people living with HIV, as well as the community’s engagement with testing, treatment, and prevention. Based on theories of stigma elaborating how it arises from the relationships between the stigmatized and the stigmatizer as well as within the general community, we hypothesized that social networks can shape HIV-related stigma. Objective To estimate social network correlates of HIV-related stigma. Methods During 2011-2012, we collected complete social network data from a community of 1669 adults (“egos”) in Mbarara, Uganda using six culturally-adapted name generators to elicit different types of social ties (“alters”). We measured HIV-related stigma using the 9-item AIDS-Related Stigma Scale. HIV serostatus was based on self-report. We fitted linear regression models that account for network autocorrelation to estimate the association between egos’ HIV-related stigma, alters’ HIV-related stigma and alters’ self-reported HIV serostatus, while adjusting for egos’ HIV serostatus, network centrality, village size, perceived HIV prevalence, and sociodemographic characteristics. Results The average AIDS-Related Stigma Score was 0.79 (Standard Deviation = 0.50). In the population 116 (7%) egos reported being HIV-positive, and 757 (46%) reported an HIV-positive alter. In the multivariable model, we found that egos’ own HIV-related stigma was positively correlated with their alters’ average stigma score (b=0.53; 95% confidence interval [CI] 0.42-0.63) and negatively correlated with having one or more HIV-positive alters (b=-0.05; 95% CI -0.10 to -0.003). Conclusion Stigma-reduction interventions should be targeted not only at the level of the individual but also at the level of the network. Directed and meaningful contact with people living with HIV may also reduce HIV-related stigma.

Engaging natural supports may be a promising strategy to promote the use of evidence-based smoking cessation treatment for individuals with serious mental illness (SMI) who smoke. This qualitative study explored preferences for support for quitting from family and friends among individuals with SMI who participated in cessation treatment. Participants were 41 individuals with SMI enrolled in a Medicaid Demonstration Project of smoking cessation at community mental health centers. Open-ended questions asked during a social network interview explored participants' preferences for more support for quitting smoking from family and friends. The qualitative data was coded and common themes were identified across the dataset. Three primary preferences emerged for smoking cessation support from family members and friends: 1) more practical support for quitting (e.g., financial help with purchasing cessation medications); 2) more emotional support for quitting (e.g., encouraging progress toward quitting); and 3) changing their own smoking behaviors in the presence of participants (e.g., don't smoke around them or offer them cigarettes). Individuals with SMI who participated in smoking cessation treatment at community mental health centers indicated several ways that family members and friends could support their efforts to quit smoking. Understanding how people with SMI want support from family and friends to quit smoking will inform strategies to leverage these natural resources to promote the use of evidence-based smoking cessation treatment and support smoking abstinence for this population.

Objective:The influence of employment on subsequent psychiatric hospitalization for people with serious mental illness is unclear. This study examined whether unemployed people with serious mental illness were more or less likely to experience psychiatric hospitalization after gaining employment.Methods:A secondary analysis was conducted of data from the Mental Health Treatment Study. Two years of prospective employment and psychiatric hospital outcomes were examined for 2,055 adults with schizophrenia, bipolar disorder, or major depression. The analyses examined associations between employment and psychiatric hospitalization via multilevel regression by using time-lagged modeling.Results:Employment was associated with a lower subsequent three-month risk of psychiatric hospitalization (odds ratio=.65, 95% confidence interval=.50–.84) after the analysis adjusted for baseline characteristics, including previous psychiatric hospitalizations and self-reported physical health.Conclusions:Unemployed outpatients...

This article is the second installment of a two-part tutorial on the analysis of zero-modified count and semicontinuous data. Part 1, which appears as a companion piece in this issue of Statistics in Medicine, provides a general background and overview of the topic, with particular emphasis on applications to health services research. Here, we present three case studies highlighting various approaches for the analysis of zero-modified data. The first case study describes methods for analyzing zero-inflated longitudinal count data. Case study 2 considers the use of hurdle models for the analysis of spatiotemporal count data. The third case study discusses an application of marginalized two-part models to the analysis of semicontinuous health expenditure data. Copyright © 2016 John Wiley & Sons, Ltd.

Health services data often contain a high proportion of zeros. In studies examining patient hospitalization rates, for instance, many patients will have no hospitalizations, resulting in a count of zero. When the number of zeros is greater or less than expected under a standard count model, the data are said to be zero modified relative to the standard model. A similar phenomenon arises with semicontinuous data, which are characterized by a spike at zero followed by a continuous distribution with positive support. When analyzing zero-modified count and semicontinuous data, flexible mixture distributions are often needed to accommodate both the excess zeros and the typically skewed distribution of nonzero values. Various models have been introduced over the past three decades to accommodate such data, including hurdle models, zero-inflated models, and two-part semicontinuous models. This tutorial describes recent modeling strategies for zero-modified count and semicontinuous data and highlights their role in health services research studies. Part 1 of the tutorial, presented here, provides a general overview of the topic. Part 2, appearing as a companion piece in this issue of Statistics in Medicine, discusses three case studies illustrating applications of the methods to health services research. Copyright © 2016 John Wiley & Sons, Ltd.

Objective To evaluate the impact of the Nursing Home Value-Based Purchasing demonstration on quality of care and Medicare spending. Data Sources/Study Setting Administrative and qualitative data from Arizona, New York, and Wisconsin nursing homes over the base-year (2008–2009) and 3-year (2009–2012) demonstration period. Study Design Nursing homes were randomized to the intervention in New York, while the comparison facilities were constructed via propensity score matching in Arizona and Wisconsin. We used a difference-in-difference analysis to compare outcomes across the base-year relative to outcomes in each of the three demonstration years. To provide context and assist with interpretation of results, we also interviewed staff members at participating facilities. Principal Findings Medicare savings were observed in Arizona in the first year only and Wisconsin for the first 2 years; no savings were observed in New York. The demonstration did not systematically impact any of the quality measures. Discussions with nursing home administrators suggested that facilities made few, if any, changes in response to the demonstration, leading us to conclude that the observed savings likely reflected regression to the mean rather than true savings. Conclusion The Federal nursing home pay-for-performance demonstration had little impact on quality or Medicare spending.

BackgroundIn response to rising rates of opioid abuse and overdose, U.S. states enacted laws to restrict the prescribing and dispensing of controlled substances. The effect of these laws on opioid use is unclear. MethodsWe tested associations between prescription-opioid receipt and state controlled-substances laws. Using Medicare administrative data for fee-for-service disabled beneficiaries 21 to 64 years of age who were alive throughout the calendar year (8.7 million person-years from 2006 through 2012) and an original data set of laws (e.g., prescription-drug monitoring programs), we examined the annual prevalence of beneficiaries with four or more opioid prescribers, prescriptions yielding a daily morphine-equivalent dose (MED) of more than 120 mg, and treatment for nonfatal prescription-opioid overdose. We estimated how opioid outcomes varied according to eight types of laws. ResultsFrom 2006 through 2012, states added 81 controlled-substance laws. Opioid receipt and potentially hazardous prescriptio...

ObjectivesGraphic display formats are often used to enhance health information. Yet limited attention has been paid to graph literacy in people of lower education and lower socioeconomic status (SES). This study aimed to: 1) examine the relationship between graph literacy, numeracy, health literacy and sociodemographic characteristics in a Medicaid-eligible population 2) determine the impact of graph literacy on comprehension and preference for different visual formats.MethodsWe conducted a cross-sectional online survey among people in the US on Medicaid, and of presumed lower education and SES.ResultsThe mean graph literacy score among 436 participants was 1.47 (SD 1.05, range: 0 to 4). Only graph literacy was significantly associated with overall comprehension (p < .001). Mean comprehension scores were highest for the table format (1.91), closely followed by bar graph (1.85) and icon array (1.80). Information comprehension was aligned with preference scores.ConclusionsGraph literacy in a Medicaid-eligible population was lower than previous estimates in the US. Tables were better understood, with icon arrays yielding the lowest score. Preferences aligned with comprehension.Practice implicationsIt may be necessary to reconsider the use of graphic display formats when designing information for people with lower educational levels. Further research is needed.

Objective To user-test graphical display formats (icon arrays, pie, bar, and line charts) to identify preferred formats and metrics (‘probability of death or stroke’ or ‘proportion of time lived without death or stroke’) in order to display time-dependent risks of stroke or death for three carotid stenosis treatments: endarterectomy (surgery), stenting, and medical therapy. Methods Iterative cycles of semi-structured interviews with patients recruited from a Vascular Clinic. Results A total of 27 patients (mean age = 68; range: 50–85) were interviewed over four cycles. Patients strongly preferred the pie chart over icon arrays, and over bar or line graphs. The preference was based on patient recognition of the time-based increase in risk for stroke or death for treatment options. Patients preferred data presented as probabilities instead of the proportion of time lived. We did not assess patients’ understanding. Conclusion Patients preferred the pie chart formats and reported better realization that risks increase with time for each option and that tradeoffs exist when surgery has a higher short-term risk than medical therapy. Practice implications There remains debate on how best to convey time-dependent risk information to patients, especially where low literacy and numeracy might exist.

INTRODUCTION: Patients who undergo endovascular abdominal aortic aneurysm repair (EVR) remain at risk for reintervention and rupture. We sought to define the five-year rate of reintervention and rupture following EVR in the Vascular Quality Initiative (VQI). METHODS: We identified all patients in the VQI who underwent EVR from 2003 – 2015. We linked patients in the VQI to Medicare claims for long-term outcomes. We stratified patients on baseline clinical and procedural characteristics to identify those at risk for reintervention. Our primary outcomes were five-year rates of reintervention and late aneurysm rupture after EVR. We assessed these with Kaplan-Meier survival estimation. RESULTS: We studied 12,911 patients who underwent EVR. Mean age 75.5 years, 79.9% male, 3.9% black, and 89.1% of operations were performed in the elective setting. The five-year rate of reintervention for the entire cohort was 21%, and the five-year rate of late aneurysm rupture was 3%. Reintervention rates varied across categories of EVR urgency. Patients who underwent EVR in the elective setting had the lowest five-year rate of reintervention at 20%. Those who underwent surgery for symptomatic aneurysms had higher rates of reintervention at 25%. Patients undergoing EVR emergently for rupture had the highest rate of reintervention, 27% at four years (log-rank across the 3 groups, p