Estimating the Causal Effect of Low Levels of Fine Particulate Matter on Hospitalization.

BACKGROUND: In 2012, the EPA enacted more stringent National Ambient Air Quality Standards (NAAQS) for fine particulate matter (PM2.5). Few studies have characterized the health effects of air pollution levels lower than the most recent NAAQS for long-term exposure to PM2.5 (now 12 µg/m). METHODS: We constructed a cohort of 32,119 Medicare beneficiaries residing in 5138 US ZIP codes who were interviewed as part of the Medicare Current Beneficiary Survey (MCBS) between 2002 and 2010 and had 1 year of follow-up. We considered four outcomes: all-cause hospitalizations, hospitalizations for circulatory diseases and respiratory diseases, and death. RESULTS: We found that increasing exposure to PM2.5 from levels lower than 12 µg/m to levels higher than 12 µg/m is associated with increases in all-cause admission rates of 7% (95% CI = 3%, 10%) and in circulatory admission hazard rates of 6% (95% CI = 2%, 9%). When we restricted analysis to enrollees with exposure always lower than 12 µg/m, we found that increasing exposure from levels lower than 8 µg/m to levels higher than 8 µg/m increased all-cause admission hazard rates by 15% (95% CI = 8%, 23%), circulatory by 18% (95% CI = 10%, 27%), and respiratory by 21% (95% CI = 9%, 34%). CONCLUSIONS: In a nationally representative sample of Medicare enrollees, changes in exposure to PM2.5, even at levels consistently below standards, are associated with increases in hospital admissions for all causes and cardiovascular and respiratory diseases. The robustness of our results to inclusion of many additional individual level potential confounders adds validity to studies of air pollution that rely entirely on administrative data.

Emergency care in 59 low- and middle-income countries: a systematic review.

OBJECTIVE: To conduct a systematic review of emergency care in low- and middle-income countries (LMICs). METHODS: We searched PubMed, CINAHL and World Health Organization (WHO) databases for reports describing facility-based emergency care and obtained unpublished data from a network of clinicians and researchers. We screened articles for inclusion based on their titles and abstracts in English or French. We extracted data on patient outcomes and demographics as well as facility and provider characteristics. Analyses were restricted to reports published from 1990 onwards. FINDINGS: We identified 195 reports concerning 192 facilities in 59 countries. Most were academically-affiliated hospitals in urban areas. The median mortality within emergency departments was 1.8% (interquartile range, IQR: 0.2-5.1%). Mortality was relatively high in paediatric facilities (median: 4.8%; IQR: 2.3-8.4%) and in sub-Saharan Africa (median: 3.4%; IQR: 0.5-6.3%). The median number of patients was 30 000 per year (IQR: 10 296-60 000), most of whom were young (median age: 35 years; IQR: 6.9-41.0) and male (median: 55.7%; IQR: 50.0-59.2%). Most facilities were staffed either by physicians-in-training or by physicians whose level of training was unspecified. Very few of these providers had specialist training in emergency care. CONCLUSION: Available data on emergency care in LMICs indicate high patient loads and mortality, particularly in sub-Saharan Africa, where a substantial proportion of all deaths may occur in emergency departments. The combination of high volume and the urgency of treatment make emergency care an important area of focus for interventions aimed at reducing mortality in these settings.

Association between the Medicare hospice benefit and health care utilization and costs for patients with poor-prognosis cancer.

IMPORTANCE: More patients with cancer use hospice currently than ever before, but there are indications that care intensity outside of hospice is increasing, and length of hospice stay decreasing. Uncertainties regarding how hospice affects health care utilization and costs have hampered efforts to promote it. OBJECTIVE: To compare utilization and costs of health care for patients with poor-prognosis cancers enrolled in hospice vs similar patients without hospice care. DESIGN, SETTING, AND PARTICIPANTS: Matched cohort study of patients in hospice and nonhospice care using a nationally representative 20% sample of Medicare fee-for-service beneficiaries who died in 2011. Patients with poor-prognosis cancers (eg, brain, pancreatic, metastatic malignancies) enrolled in hospice before death were matched to similar patients who died without hospice care. EXPOSURES: Period between hospice enrollment and death for hospice beneficiaries, and the equivalent period of nonhospice care before death for matched nonhospice patients. MAIN OUTCOMES AND MEASURES: Health care utilization including hospitalizations and procedures, place of death, cost trajectories before and after hospice start, and cumulative costs, all during the last year of life. RESULTS: Among 86,851 patients with poor-prognosis cancers, median time from first poor-prognosis diagnosis to death was 13 months (interquartile range [IQR], 3-34), and 51,924 patients (60%) entered hospice before death. Matching yielded a cohort balanced on age, sex, region, time from poor-prognosis diagnosis to death, and baseline care utilization, with 18,165 patients in the hospice group and 18,165 in the nonhospice group. After matching, 11% of nonhospice and 1% of hospice beneficiaries who had cancer-directed therapy after exposure were excluded. Median hospice duration was 11 days. After exposure, nonhospice beneficiaries had significantly more hospitalizations (65% [95% CI, 64%-66%], vs hospice with 42% [95% CI, 42%-43%]; risk ratio, 1.5 [95% CI, 1.5-1.6]), intensive care (36% [95% CI, 35%-37%], vs hospice with 15% [95% CI, 14%-15%]; risk ratio, 2.4 [95% CI, 2.3-2.5]), and invasive procedures (51% [95% CI, 50%-52%], vs hospice with 27% [95% CI, 26%-27%]; risk ratio, 1.9 [95% CI, 1.9-2.0]), largely for acute conditions not directly related to cancer; and 74% (95% CI, 74%-75%) of nonhospice beneficiaries died in hospitals and nursing facilities compared with 14% (95% CI, 14%-15%) of hospice beneficiaries. Costs for hospice and nonhospice beneficiaries were not significantly different at baseline, but diverged after hospice start. Total costs over the last year of life were $71,517 (95% CI, $70,543-72,490) for nonhospice and $62,819 (95% CI, $62,082-63,557) for hospice, a statistically significant difference of $8697 (95% CI, $7560-$9835). CONCLUSIONS AND RELEVANCE: In this sample of Medicare fee-for-service beneficiaries with poor-prognosis cancer, those receiving hospice care vs not (control), had significantly lower rates of hospitalization, intensive care unit admission, and invasive procedures at the end of life, along with significantly lower total costs during the last year of life.

Prospective evaluation of data-driven models to predict daily risk of Clostridioides difficile infection at 2 large academic health centers.

Many data-driven patient risk stratification models have not been evaluated prospectively. We performed and compared the prospective and retrospective evaluations of 2 Clostridioides difficile infection (CDI) risk-prediction models at 2 large academic health centers, and we discuss the models' robustness to data-set shifts.

Learning Concept Credible Models for Mitigating Shortcuts.

During training, models can exploit spurious correlations as shortcuts, resulting in poor generalization performance when shortcuts do not persist. In this work, assuming access to a representation based on domain knowledge (i.e., known concepts) that is invariant to shortcuts, we aim to learn robust and accurate models from biased training data. In contrast to previous work, we do not rely solely on known concepts, but allow the model to also learn unknown concepts. We propose two approaches for mitigating shortcuts that incorporate domain knowledge, while accounting for potentially important yet unknown concepts. The first approach is two-staged. After fitting a model using known concepts, it accounts for the residual using unknown concepts. While flexible, we show that this approach is vulnerable when shortcuts are correlated with the unknown concepts. This limitation is addressed by our second approach that extends a recently proposed regularization penalty. Applied to two real-world datasets, we demonstrate that both approaches can successfully mitigate shortcut learning.

Impact of a dedicated palliative radiation oncology service on the use of single fraction and hypofractionated radiation therapy among patients with bone metastases.

BACKGROUND: Radiation therapy (RT) is frequently used to palliate symptomatic bone metastases. While high quality literature has shown that for uncomplicated bone metastases, shorter radiotherapy courses are just as effective as longer courses for the treatment of pain, shorter courses remain under-utilized. We aimed to assess the impact of a dedicated palliative radiation oncology service on the frequency of single fraction RT (SF-RT) and hypofractionated radiation (hypo-RT) (≤5 fractions) among patients with bone metastases. METHODS: We identified 2,086 instances of palliative radiation (RT) for complicated and uncomplicated bone metastases between April 10, 2008 and September 17, 2014. We used multivariable logistic regression analysis (MVA) to estimate the association of the Supportive and Palliative Radiation Oncology (SPRO) service with the likelihood of receiving SF-RT or hypo-RT after controlling for age, sex, tumor type, and treatment site. RESULTS: Prior to SPRO's implementation on July 1, 2011, the proportion of SF-RT and hypo-RT for bone metastases was 6.4% and 27.6% respectively. After SPRO's implementation, the proportion of SF-RT and hypo-RT increased to 22.3% (P<0.001) and 53.5% (P<0.001) respectively. In MVA, patients were more likely to receive SF-RT [odds ratio (OR) =3.3, 95% confidence interval (CI) =2.4-4.7, P<0.001], and hypo-RT (OR =2.5, 95% CI =2.0-3.1, P<0.001) after SPRO's implementation. Compared to sites without a dedicated palliative service, patients receiving care at the SPRO affiliated department were more likely to receive SF-RT (OR =1.9, 95% CI =1.1-3.2, P=0.02) and hypo-RT (OR =1.5, 95% CI =1.1-2.0, P=0.004) for bone metastases. After SPRO's implementation, the average number of RT courses delivered for bone metastases increased from 17.4 to 25.6 per month, (+8.3, 95% CI =4.99-11.55, P<0.001). Despite greater SF-RT and hypo-RT, the average total fractions per month of palliative RT for bone metastases increased from 163.5 pre-SPRO to 166.8 post-SPRO, though not significantly (+3.22, P=NS). CONCLUSIONS: Implementation of a dedicated palliative radiation oncology service was associated with increased use of SF and hypo-RT and with greater courses of RT delivered for bone metastases.

A Generalizable, Data-Driven Approach to Predict Daily Risk of Clostridium difficile Infection at Two Large Academic Health Centers.

OBJECTIVE An estimated 293,300 healthcare-associated cases of Clostridium difficile infection (CDI) occur annually in the United States. To date, research has focused on developing risk prediction models for CDI that work well across institutions. However, this one-size-fits-all approach ignores important hospital-specific factors. We focus on a generalizable method for building facility-specific models. We demonstrate the applicability of the approach using electronic health records (EHR) from the University of Michigan Hospitals (UM) and the Massachusetts General Hospital (MGH). METHODS We utilized EHR data from 191,014 adult admissions to UM and 65,718 adult admissions to MGH. We extracted patient demographics, admission details, patient history, and daily hospitalization details, resulting in 4,836 features from patients at UM and 1,837 from patients at MGH. We used L2 regularized logistic regression to learn the models, and we measured the discriminative performance of the models on held-out data from each hospital. RESULTS Using the UM and MGH test data, the models achieved area under the receiver operating characteristic curve (AUROC) values of 0.82 (95% confidence interval [CI], 0.80-0.84) and 0.75 ( 95% CI, 0.73-0.78), respectively. Some predictive factors were shared between the 2 models, but many of the top predictive factors differed between facilities. CONCLUSION A data-driven approach to building models for estimating daily patient risk for CDI was used to build institution-specific models at 2 large hospitals with different patient populations and EHR systems. In contrast to traditional approaches that focus on developing models that apply across hospitals, our generalizable approach yields risk-stratification models tailored to an institution. These hospital-specific models allow for earlier and more accurate identification of high-risk patients and better targeting of infection prevention strategies. Infect Control Hosp Epidemiol 2018;39:425-433.

Emergency Care Use and the Medicare Hospice Benefit for Individuals with Cancer with a Poor Prognosis.

OBJECTIVES: To compare patterns of emergency department (ED) use and inpatient admission rates for elderly adults with cancer with a poor prognosis who enrolled in hospice to those of similar individuals who did not. DESIGN: Matched case-control study. SETTING: Nationally representative sample of Medicare fee-for-service beneficiaries with cancer with a poor prognosis who died in 2011. PARTICIPANTS: Beneficiaries in hospice matched to individuals not in hospice on time from diagnosis of cancer with a poor prognosis to death, region, age, and sex. MEASUREMENTS: Comparison of ED use and inpatient admission rates before and after hospice enrollment for beneficiaries in hospice and controls. RESULTS: Of 272,832 matched beneficiaries, 81% visited the ED in the last 6 months of life. At baseline, daily ED use and admission rates were not significantly different between beneficiaries in and not in hospice. By the week before death, nonhospice controls averaged 69.6 ED visits/1,000 beneficiary-days, versus 7.6 for beneficiaries in hospice (rate ratio (RR) = 9.7, 95% confidence interval (CI) = 9.3-10.0). Inpatient admission rates in the last week of life were 63% for nonhospice controls and 42% for beneficiaries in hospice (RR = 1.51, 95% CI = 1.45-1.57). Of all beneficiaries in hospice, 28% enrolled during inpatient stays originating in EDs; they accounted for 35.7% (95% CI = 35.4-36.0%) of all hospice stays of less than 1 month and 13.9% (95% CI = 13.6-14.2%) of stays longer than 1 month. CONCLUSION: Most Medicare beneficiaries with cancer with a poor prognosis visited EDs at the end of life. Hospice enrollment was associated with lower ED use and admission rates. Many individuals enrolled in hospice during inpatient stays that followed ED visits, a phenomenon linked to shorter hospice stays. These findings must be interpreted carefully given potential unmeasured confounders in matching.

Short-term Mortality Prediction for Elderly Patients Using Medicare Claims Data.

Risk prediction is central to both clinical medicine and public health. While many machine learning models have been developed to predict mortality, they are rarely applied in the clinical literature, where classification tasks typically rely on logistic regression. One reason for this is that existing machine learning models often seek to optimize predictions by incorporating features that are not present in the databases readily available to providers and policy makers, limiting generalizability and implementation. Here we tested a number of machine learning classifiers for prediction of six-month mortality in a population of elderly Medicare beneficiaries, using an administrative claims database of the kind available to the majority of health care payers and providers. We show that machine learning classifiers substantially outperform current widely-used methods of risk prediction-but only when used with an improved feature set incorporating insights from clinical medicine, developed for this study. Our work has applications to supporting patient and provider decision making at the end of life, as well as population health-oriented efforts to identify patients at high risk of poor outcomes.

Physician Characteristics Strongly Predict Patient Enrollment In Hospice.

Individual physicians are widely believed to play a large role in patients' decisions about end-of-life care, but little empirical evidence supports this view. We developed a novel method for measuring the relationship between physician characteristics and hospice enrollment, in a nationally representative sample of Medicare patients. We focused on patients who died with a diagnosis of poor-prognosis cancer in the period 2006-11, for whom palliative treatment and hospice would be considered the standard of care. We found that the proportion of a physician's patients who were enrolled in hospice was a strong predictor of whether or not that physician's other patients would enroll in hospice. The magnitude of this association was larger than that of other known predictors of hospice enrollment that we examined, including patients' medical comorbidity, age, race, and sex. Patients cared for by medical oncologists and those cared for in not-for-profit hospitals were significantly more likely than other patients to enroll in hospice. These findings suggest that physician characteristics are among the strongest predictors of whether a patient receives hospice care-which mounting evidence indicates can improve care quality and reduce costs. Interventions geared toward physicians, both by specialty and by previous history of patients' hospice enrollment, may help optimize appropriate hospice use.