Sparsity is often a desirable property of statistical models, and various feature selection methods exist so as to yield sparser and interpretable models. However, their application to biomedical text classification, particularly to mortality risk stratification among intensive care unit (ICU) patients, has not been thoroughly studied.To develop and characterize sparse classifiers based on the free text of nursing notes in order to predict ICU mortality risk and to discover text features most strongly associated with mortality.We selected nursing notes from the first 24h of ICU admission for 25,826 adult ICU patients from the MIMIC-II database. We then developed a pair of stochastic gradient descent-based classifiers with elastic-net regularization. We also studied the performance-sparsity tradeoffs of both classifiers as their regularization parameters were varied.The best-performing classifier achieved a 10-fold cross-validated AUC of 0.897 under the log loss function and full L2 regularization, while full L1 regularization used just 0.00025% of candidate input features and resulted in an AUC of 0.889. Using the log loss (range of AUCs 0.889-0.897) yielded better performance compared to the hinge loss (0.850-0.876), but the latter yielded even sparser models.Most features selected by both classifiers appear clinically relevant and correspond to predictors already present in existing ICU mortality models. The sparser classifiers were also able to discover a number of informative - albeit nonclinical - features.The elastic-net-regularized classifiers perform reasonably well and are capable of reducing the number of features required by over a thousandfold, with only a modest impact on performance.

Objectives

To determine whether initiation of cholinesterase inhibitors is associated with significant weight loss in a real-word clinical setting.

Design

Retrospective cohort study from 2007 to 2010 comparing weight loss in individuals with dementia newly prescribed cholinesterase inhibitors and those newly prescribed other chronic medications.

Setting

National Veterans Affairs data.

Participants

Individuals aged 65 and older with a diagnosis of dementia who received a new prescription for a cholinesterase inhibitor or other new chronic medication.

Measurements

The primary outcome was time to 10-pound weight loss over 12 months. Propensity score matching was used to control for the likelihood of receiving a cholinesterase inhibitor based on baseline characteristics. Data were analyzed in a priori defined subgroups according to age, comorbid burden, and initial weight.

Results

Of 6,504 individuals that met study criteria, 1,188 started on cholinesterase inhibitors were matched to 2,189 started on other medications. The propensity-matched cohorts were well balanced on baseline covariates. Participants initiated on cholinesterase inhibitors had a higher risk of weight loss than matched controls at 12 months (hazard ratio = 1.23, 95% confidence interval (CI) = 1.07-1.41). At 12 months, 29.3% of participants taking cholinesterase inhibitors had experienced weight loss, compared with 22.8% of nonusers, corresponding to a number needed to harm of 21.2 (95% CI = 12.5-71.4) over 1 year. There were no significant differences in the risk of weight loss within subgroups.

Conclusion

These results are consistent with the available data from randomized controlled trials. Clinicians should consider the risk of weight loss when prescribing cholinesterase inhibitors.

Background. Extraskeletal osteosarcoma (ESOS) is a rare subtype of osteosarcoma. We investigated patient characteristics, overall survival, and prognostic factors in ESOS. Methods. We identified cases of high-grade osteosarcoma with known tissue of origin in the Surveillance, Epidemiology, and End Results database from 1973 to 2009. Demographics were compared using univariate tests. Overall survival was compared with log-rank tests and multivariate analysis using Cox proportional hazards methods. Results. 256/4,173 (6%) patients with high-grade osteosarcoma had ESOS. Patients with ESOS were older, were more likely to have an axial tumor and regional lymph node involvement, and were female. Multivariate analysis showed ESOS to be favorable after controlling for stage, age, tumor site, gender, and year of diagnosis [hazard ratio 0.75 (95% CI 0.62 to 0.90); p = 0.002]. There was an interaction between age and tissue of origin such that older patients with ESOS had superior outcomes compared to older patients with skeletal osteosarcoma. Adverse prognostic factors in ESOS included metastatic disease, larger tumor size, older age, and axial tumor site. Conclusion. Patients with ESOS have distinct clinical features but similar prognostic factors compared to skeletal osteosarcoma. Older patients with ESOS have superior outcomes compared to older patients with skeletal osteosarcoma.

Background and objective

Quality prescribing for older adults involves multiple considerations. We evaluated multiple aspects of prescribing quality in older veterans to develop an integrated view of prescribing problems and to understand how the prevalence of these problems varies across clinically important subgroups of older adults.

Design and participants

Cross-sectional observational study of veterans age 65 years and older who received medications from Department of Veterans Affairs (VA) pharmacies in 2007.

Main measures

Using VA pharmacy data linked with encounter, laboratory and other data, we assessed five types of prescribing problems.

Key results

Among 462,405 patients age 65 and older, mean age was 75 years, 98 % were male, and patients were prescribed a median of five medications. Half of patients (50 %) had one or more prescribing problems, including 12 % taking one or more medications at an inappropriately high dose, 30 % with drug-drug interactions, 3 % with drug-disease interactions, and 26 % taking one or more Beers criteria drugs. In addition, 16 % were taking a high-risk drug (warfarin, insulin, and/or digoxin). On multivariable analysis, age was not strongly associated with four of the five types of prescribing issues assessed (relative risk < 1.3 across age groups), and comorbid burden conferred substantially increased risk only for drug-disease interactions and use of high-risk drugs. In contrast, the number of drugs used was consistently the strongest predictor of prescribing problems. Patients in the highest quartile of medication use had 6.6-fold to12.5-fold greater risk of each type of prescribing problem compared to patients in the lowest quartile (P < 0.001 for each).

Conclusions

The number of medications used is by far the strongest risk factor for each of five types of prescribing problems. Efforts to improve prescribing should especially target patients taking multiple medications.

Background and objective

Most methods for developing clinical prognostic models focus on identifying parsimonious and accurate models to predict a single outcome; however, patients and providers often want to predict multiple outcomes simultaneously. As an example, for older adults one is often interested in predicting nursing home admission as well as mortality. We propose and evaluate a novel predictor-selection computing method for multiple outcomes and provide the code for its implementation.

Methods

Our proposed algorithm selected the best subset of common predictors based on the minimum average normalized Bayesian Information Criterion (BIC) across outcomes: the Best Average BIC (baBIC) method. We compared the predictive accuracy (Harrell's C-statistic) and parsimony (number of predictors) of the model obtained using the baBIC method with: 1) a subset of common predictors obtained from the union of optimal models for each outcome (Union method), 2) a subset obtained from the intersection of optimal models for each outcome (Intersection method), and 3) a model with no variable selection (Full method). We used a case-study data from the Health and Retirement Study (HRS) to demonstrate our method and conducted a simulation study to investigate performance.

Results

In the case-study data and simulations, the average Harrell's C-statistics across outcomes of the models obtained with the baBIC and Union methods were comparable. Despite the similar discrimination, the baBIC method produced more parsimonious models than the Union method. In contrast, the models selected with the Intersection method were the most parsimonious, but with worst predictive accuracy, and the opposite was true in the Full method. In the simulations, the baBIC method performed well by identifying many of the predictors selected in the baBIC model of the case-study data most of the time and excluding those not selected in the majority of the simulations.

Conclusions

Our method identified a common subset of variables to predict multiple clinical outcomes with superior balance between parsimony and predictive accuracy to current methods.

Objective

To determine the predictors of elevated transaminases in an incident user cohort of older adult patients with rheumatic diseases receiving methotrexate (MTX) using elements derived from an electronic health record.

Methods

Using a national, administrative database of patients seen through the Veterans Health Administration that included pharmacy and laboratory data, we performed an observational cohort study of veterans ages ≥65 years who were new users of MTX to identify risk factors for elevated transaminases.

Results

We studied 659 incident users of MTX. We found a 6% incidence of moderate (≥1.5 × the upper limit of normal) elevations in aspartate aminotransferase or alanine aminotransferase over a mean followup period of 7 months. We identified predictors of moderate transaminase elevations to include obesity (per body mass index ≥30 kg/m(2) ), total cholesterol >240 mg/dl, pre-MTX liver function test (LFT) elevations, use of biologic agents, and lack of folic acid supplementation. A patient with these characteristics and >3 comorbid conditions would be predicted to have a 90% chance of developing a moderate transaminase elevation in the 7 months after starting MTX.

Conclusion

Moderate LFT abnormalities were uncommon in the first 7 months of MTX use, but were more likely to occur in patients with obesity, untreated high cholesterol, pre-MTX LFT elevations, biologic agent use, and lack of folic acid supplementation. Future work should aim to develop a robust, automated prediction rule for identifying patients at high risk for MTX-related liver toxicity.

Background

Between 0 and 48% of normal HIV-uninfected individuals score below threshold neuropsychological test scores for HIV-associated neurocognitive disorders (HAND) or are false positives. There has been little effort to understand the effect of varied interpretations of research criteria for HAND on false-positive frequencies, prevalence and analytic estimates.

Methods

The proportion of normal individuals scoring below Z score thresholds drawn from research criteria for HAND, or false-positive frequencies, was estimated in a normal Kenyan population and a simulated normal population using varied interpretations of research criteria for HAND. We calculated the impact of false-positive frequencies on prevalence estimates and statistical power.

Results

False-positive frequencies of 2-74% were observed for asymptomatic neurocognitive impairment/mild neurocognitive disorder and 0-8% for HIV-associated dementia. False-positive frequencies depended on the definition of an abnormal cognitive domain, Z score thresholds and neuropsychological battery size. Misclassification led to clinically important overestimation of prevalence and dramatic decreases in power.

Conclusions

Minimizing false-positive frequencies is critical to decrease bias in prevalence estimates and minimize reductions in power in studies of association, particularly for mild forms of HAND. We recommend changing the Z score threshold to ≤-1.5 for mild impairment, limiting analysis to 3-5 cognitive domains and using the average Z score to define an abnormal domain.

Background

Solid organ transplant recipients (OTR) are at an increased risk of developing squamous cell carcinoma (SCC) of the skin after transplantation. In predominantly white cohorts, Fitzpatrick skin type (FST) has been reported to be a risk factor for developing posttransplantation skin cancers.

Objective

Our goal was to determine if FST is a statistically significant risk factor for the development of SCC after solid organ transplantation in a diverse US population of OTR.

Methods

A cohort of OTR completed a questionnaire of demographic factors, transplant type, FST, and skin cancer history. Univariate and multivariate analyses were performed to determine the risk factors for development of SCC after transplantation.

Results

As expected, male subjects had an increased risk for SCC compared with female subjects (P = .02), and those aged 50 years and older at the time of transplantation were more likely to develop SCC compared with those younger than 50 years (P < .001). The risk of SCC increased with each incremental decrease in FST, from FST VI to FST I (linear test for trend P < .001).

Limitations

Our questionnaire did not ask specifically about immunosuppressive medications; instead, organ transplant category was used as a proxy for level of immunosuppression.

Conclusions

FST, a patient-reported variable, is an independent risk factor for the development of SCC in OTR, and should be elicited from patients who have gone or will undergo organ transplantation.

Objectives

To understand where older adults spend time (at home, in the hospital, or in a nursing home) in the year following high-risk cancer surgery.

Design

Retrospective cohort study.

Setting

Medicare beneficiaries using data from Medicare Inpatient claims to ascertain hospital days and the Minimum Data Set to ascertain nursing home days.

Participants

Beneficiaries who underwent high-risk cancer surgery (cystectomy, pancreaticoduodenectomy, gastrectomy, or esophagectomy) were identified to determine cumulative time spent away from home in the year following surgery.

Measurements

Adjusted percentages of time spent away from home (ie, days in a hospital or nursing home) were modeled for the year following surgery.

Results

A total of 37 748 beneficiaries underwent high-risk cancer surgery during the study period, and 28.3% died within 1 year. Overall, beneficiaries spent 13.9 ± 26.2 days in the hospital (over 1.5 ± 2.0 hospital readmissions) and 37.2 ± 50.6 days in the nursing home (over 1.5 ± 1.0 admissions) in the year following surgery. Among beneficiaries who were alive and dead at 1 year, 18.5% and 30.1% of time was spent away from home, respectively. Beneficiaries who were initially discharged to a facility following surgery and died within 1 year spent 44.4% of their final year away from home.

Conclusion

Time spent away from home in the hospital and/or nursing home in the year following high-risk cancer surgery is substantial among Medicare beneficiaries. This information is crucial in counseling patients on postoperative expectations and may additionally influence preoperative decision making. J Am Geriatr Soc 68:505-510, 2020.

Objectives

To determine the rate of unplanned PICU readmissions, examine the characteristics of index admissions associated with readmission, and compare outcomes of readmissions versus index admissions.

Design

Retrospective cohort analysis.

Setting

Ninety North American PICUs that participated in the Virtual Pediatric Intensive Care Unit Systems.

Patients

One hundred five thousand four hundred thirty-seven admissions between July 2009 and March 2011.

Interventions

None.

Measurements and main results

Unplanned PICU readmission within 48 hours of index discharge was the primary outcome. Summary statistics, bivariate analyses, and mixed-effects logistic regression model with random effects for each hospital were performed.There were 1,161 readmissions (1.2%). The readmission rate varied among PICUs (0-3.3%), and acute respiratory (56%), infectious (35%), neurological (28%), and cardiovascular (20%) diagnoses were often present on readmission. Readmission risk increased in patients with two or more complex chronic conditions (adjusted odds ratio, 1.72; p < 0.001), unscheduled index admission (adjusted odds ratio, 1.37; p < 0.001), and transfer to an intermediate unit (adjusted odds ratio, 1.29; p = 0.004, compared with ward). Trauma patients had a decreased risk of readmission (adjusted odds ratio, 0.67; p = 0.003). Gender, race, insurance, age more than 6 months, perioperative status, and nighttime transfer were not associated with readmission. Compared with index admissions, readmissions had longer median PICU length of stay (3.1 vs 1.7 d, p < 0.001) and higher mortality (4% vs 2.5%, p = 0.002).

Conclusions

Unplanned PICU readmissions were relatively uncommon, but were associated with worse outcomes. Several patient and admission characteristics were associated with readmission. These data help identify high-risk patient groups and inform risk-adjustment for standardized readmission rates.