ABSTRACT
PURPOSE: Pharmacy chains can differ with respect to the characteristics of their patient populations as well as their nonprescription products, services, and practices, and thus may serve as a surrogate for potential unmeasured confounding in observational studies of prescription drugs. This study evaluates whether a single-source drug can have different patient outcomes based on the dispensing pharmacy chain. METHODS: Separate analyses for two anticoagulant drugs, rivaroxaban and apixaban, were conducted using Medicare Fee-for-Service claims evaluating the association between dispensing pharmacy chain and outcomes of acute myocardial infarction, ischemic stroke, intracranial hemorrhage, gastrointestinal (GI) bleeding, all-cause mortality, and major GI bleeding. Inverse probability of treatment weighting (IPTW) was used to balance baseline covariates across pharmacy chain cohorts, and outcome association was assessed with a Cox Proportional Hazards model. RESULTS: We observed no differences in outcomes across pharmacy chains for apixaban recipients. Rivaroxaban recipients from pharmacy chain C, however, had lower rates of GI bleeding (adjusted HR 0.83; 95% CI 0.69-1.00) and ischemic stroke (adjusted HR 0.57; 95% CI 0.38-0.87) as compared to chain A in primary analyses with a 3-day grace period. The results moved closer to the null when 14- and 30-day grace periods were implemented. CONCLUSIONS: These results suggest that dispensing pharmacy chains may have the potential to act as a confounder of associations between drug exposure and outcome in some observational studies. Additional studies of potential confounding by pharmacy chain are needed. Further evaluation of potential pharmacy chain effects on safe use would be of value.
Subject(s)
Atrial Fibrillation , Ischemic Stroke , Stroke , Aged , Humans , United States , Anticoagulants/adverse effects , Rivaroxaban/adverse effects , Stroke/epidemiology , Stroke/prevention & control , Stroke/complications , Dabigatran/therapeutic use , Atrial Fibrillation/drug therapy , Medicare , Gastrointestinal Hemorrhage/chemically induced , Gastrointestinal Hemorrhage/epidemiology , Ischemic Stroke/drug therapy , Pyridones/therapeutic use , Retrospective StudiesABSTRACT
Neutralizing antibodies to the SARS CoV-2 spike proteins have been issued Emergency Use Authorizations and are a likely mechanism of vaccines to prevent COVID-19. However, benefit of treatment with monoclonal antibodies has only been observed in clinical trials in outpatients with mild to moderate COVID-19 but not in patients who are hospitalized and/or have advanced disease. To address this observation, we evaluated the timing of anti SARS-CoV-2 antibody production in hospitalized patients with the use of a highly sensitive multiplexed bead-based immunoassay allowing for early detection of antibodies to SARS-CoV-2. We found significantly lower levels of antibodies to the SARS-CoV-2 spike protein in the first week after symptom onset in patients who expired as compared to patients who were discharged. We also developed a model to characterize the relationship between each patient's individual antibody level trajectory and eventual COVID 19 outcome which can be adapted into a prediction model with more data.
Subject(s)
Antibodies, Viral/blood , COVID-19/immunology , COVID-19/mortality , SARS-CoV-2/immunology , Adult , Aged , Aged, 80 and over , Antibody Specificity , Antigens, Viral/immunology , Female , Humans , Immunoglobulin G/blood , Linear Models , Male , Middle Aged , Models, Immunological , Pandemics , Prognosis , Spike Glycoprotein, Coronavirus/immunology , Time Factors , United States/epidemiologyABSTRACT
A major concern in any observational study is unmeasured confounding of the relationship between a treatment and outcome of interest. Instrumental variable (IV) analysis methods are able to control for unmeasured confounding. However, IV analysis methods developed for censored time-to-event data tend to rely on assumptions that may not be reasonable in many practical applications, making them unsuitable for use in observational studies. In this report, we develop weighted estimators of the complier average causal effect (CACE) on the restricted mean survival time in the overall population as well as in an evenly matchable population (CACE-m). Our method is able to accommodate instrument-outcome confounding and adjust for covariate-dependent censoring, making it particularly suited for causal inference from observational studies. We establish the asymptotic properties and derive easily implementable asymptotic variance estimators for the proposed estimators. Through simulation studies, we show that the proposed estimators tend to be more efficient than instrument propensity score matching-based estimators or IPIW estimators. We apply our method to compare dialytic modality-specific survival for end stage renal disease patients using data from the U.S. Renal Data System.
Subject(s)
Survival Rate , Computer Simulation , Confounding Factors, Epidemiologic , Humans , Propensity ScoreABSTRACT
Case-crossover study designs are observational studies used to assess postmarket safety of medical products (eg, vaccines or drugs). As a case-crossover study is self-controlled, its advantages include better control for confounding because the design controls for any time-invariant measured and unmeasured confounding and potentially greater feasibility as only data from those experiencing an event (or cases) are required. However, self-matching also introduces correlation between case and control periods within a subject or matched unit. To estimate sample size in a case-crossover study, investigators currently use Dupont's formula (Biometrics 1988; 43:1157-1168), which was originally developed for a matched case-control study. This formula is relevant as it takes into account correlation in exposure between controls and cases, which are expected to be high in self-controlled studies. However, in our study, we show that Dupont's formula and other currently used methods to determine sample size for case-crossover studies may be inadequate. Specifically, these formulas tend to underestimate the true required sample size, determined through simulations, for a range of values in the parameter space. We present mathematical derivations to explain where some currently used methods fail and propose two new sample size estimation methods that provide a more accurate estimate of the true required sample size.
Subject(s)
Cross-Over Studies , Sample Size , Case-Control Studies , Humans , Models, Statistical , Observational Studies as Topic/methods , Proportional Hazards ModelsABSTRACT
It is often of interest to compare centers or healthcare providers on quality of care delivered. We consider the setting where evaluation of center performance on multiple competing events is of interest. We propose estimating center effects through cause-specific proportional hazards frailty models that allow correlation among a center's cause-specific effects. Estimation of our model proceeds via penalized partial likelihood and is implemented in R. To evaluate center performance, we also propose a directly standardized excess cumulative incidence (ECI) measure. Therefore, based on our proposed methods, practitioners can evaluate centers either through the cause-specific hazards or the cumulative incidence functions. We demonstrate, through simulations, the advantages of the proposed methods to detect outlying centers, by comparing the proposed methods and existing methods which assume uncorrelated random center effects. In addition, we develop a Correlation Score Test to test the null hypothesis that the competing event processes within a center are correlated. Using data from the Scientific Registry of Transplant Recipients, we apply our method to evaluate the performance of Organ Procurement Organizations on two competing risks: (i) receipt of a kidney transplant and (ii) death on the wait-list.
Subject(s)
Models, Statistical , Quality of Health Care/statistics & numerical data , Risk Factors , Humans , Kidney Failure, Chronic/mortality , Kidney Failure, Chronic/therapy , Kidney Transplantation/mortality , Waiting Lists/mortalityABSTRACT
BACKGROUND: Poor access to food among low-income adults has been recognized as a risk factor for chronic kidney disease (CKD), but there are no data for the impact of food insecurity on progression to end-stage renal disease (ESRD). We hypothesized that food insecurity would be independently associated with risk for ESRD among persons with and without earlier stages of CKD. STUDY DESIGN: Longitudinal cohort study. SETTING & PARTICIPANTS: 2,320 adults (aged ≥ 20 years) with CKD and 10,448 adults with no CKD enrolled in NHANES III (1988-1994) with household income ≤ 400% of the federal poverty level linked to the Medicare ESRD Registry for a median follow-up of 12 years. PREDICTOR: Food insecurity, defined as an affirmative response to the food-insecurity screening question. OUTCOME: Development of ESRD. MEASUREMENTS: Demographics, income, diabetes, hypertension, estimated glomerular filtration rate, and albuminuria. Dietary acid load was estimated from 24-hour dietary recall. We used a Fine-Gray competing-risk model to estimate the relative hazard (RH) for ESRD associated with food insecurity after adjusting for covariates. RESULTS: 4.5% of adults with CKD were food insecure. Food-insecure individuals were more likely to be younger and have diabetes (29.9%), hypertension (73.9%), or albuminuria (90.4%) as compared with their counterparts (P<0.05). Median dietary acid load in the food-secure versus food-insecure group was 51.2 mEq/d versus 55.6 mEq/d, respectively (P=0.05). Food-insecure adults were more likely to develop ESRD (RH, 1.38; 95% CI, 1.08-3.10) compared with food-secure adults after adjustment for demographics, income, diabetes, hypertension, estimated glomerular filtration rate, and albuminuria. In the non-CKD group, 5.7% were food insecure. We did not find a significant association between food insecurity and ESRD (RH, 0.77; 95% CI, 0.40-1.49). LIMITATIONS: Use of single 24-hour diet recall; lack of laboratory follow-up data and measure of changes in food insecurity over time; follow-up of cohort ended 10 years ago. CONCLUSIONS: Among adults with CKD, food insecurity was independently associated with a higher likelihood of developing ESRD. Innovative approaches to address food insecurity should be tested for their impact on CKD outcomes.
Subject(s)
Food Supply/statistics & numerical data , Kidney Failure, Chronic/epidemiology , Renal Insufficiency, Chronic/epidemiology , Adult , Aged , Cohort Studies , Disease Progression , Female , Humans , Kidney Failure, Chronic/etiology , Longitudinal Studies , Male , Middle Aged , Renal Insufficiency, Chronic/etiology , Risk Factors , United States/epidemiology , Young AdultABSTRACT
BACKGROUND: Multiple criteria decision analysis (MCDA) and stochastic multi-criteria acceptability analysis (SMAA) in their current implementation cannot incorporate prior or external information on benefits and risks. We demonstrate how to incorporate prior data using a Bayesian mixture model approach while conducting quantitative benefit-risk assessments (qBRA) for medical products. METHODS: We implemented MCDA and SMAA in a Bayesian framework. To incorporate information from a prior study, we use mixture priors on each benefit and risk attribute that mixes information from a previous study with a vague prior distribution. The degree of borrowing is varied using a mixing proportion parameter. RESULTS: A demonstration case study for qBRA using the supplementary New Drug Application (sNDA) filing for Rivaroxaban for the indication of reduction in the risk of major thrombotic vascular events in patients with peripheral artery disease (PAD) was used to illustrate the method. Net utility scores, obtained from the randomized controlled trial data to support the sNDA, from the MCDA for Rivaraxoban and comparator were 0.48 and 0.56, respectively, with Rivaroxaban being the preferred alternative only 33% of the time. We show that with only 30% borrowing from a previous RCT, the MCDA and SMAA results are favorable for Rivaroxaban, accounting for the seemingly aberrant results on all-cause death in the trial data used to support the sNDA. CONCLUSION: Our method to formally incorporate prior data in MCDA and SMAA is easy to use and interpret. Software in the form of an RShiny App is available here: https://sai-dharmarajan.shinyapps.io/BayesianMCDA_SMAA/ .
Subject(s)
Bayes Theorem , Rivaroxaban , Humans , Risk Assessment , Rivaroxaban/therapeutic use , Rivaroxaban/adverse effects , Decision Support Techniques , Randomized Controlled Trials as Topic , Peripheral Arterial Disease/drug therapy , Factor Xa Inhibitors/therapeutic use , Factor Xa Inhibitors/adverse effects , Factor Xa Inhibitors/administration & dosageABSTRACT
OBJECTIVE: Quantitative bias analysis (QBA) methods evaluate the impact of biases arising from systematic errors on observational study results. This systematic review aimed to summarize the range and characteristics of quantitative bias analysis (QBA) methods for summary level data published in the peer-reviewed literature. STUDY DESIGN AND SETTING: We searched MEDLINE, Embase, Scopus, and Web of Science for English-language articles describing QBA methods. For each QBA method, we recorded key characteristics, including applicable study designs, bias(es) addressed; bias parameters, and publicly available software. The study protocol was pre-registered on the Open Science Framework (https://osf.io/ue6vm/). RESULTS: Our search identified 10,249 records, of which 53 were articles describing 57 QBA methods for summary level data. Of the 57 QBA methods, 53 (93%) were explicitly designed for observational studies, and 4 (7%) for meta-analyses. There were 29 (51%) QBA methods that addressed unmeasured confounding, 19 (33%) misclassification bias, 6 (11%) selection bias, and 3 (5%) multiple biases. 38 (67%) QBA methods were designed to generate bias-adjusted effect estimates and 18 (32%) were designed to describe how bias could explain away observed findings. 22 (39%) articles provided code or online tools to implement the QBA methods. CONCLUSION: In this systematic review, we identified a total of 57 QBA methods for summary level epidemiologic data published in the peer-reviewed literature. Future investigators can use this systematic review to identify different QBA methods for summary level epidemiologic data.
ABSTRACT
OBJECTIVE: To compare the risk of intentional self-harm (ISH) and suicide in older men using 5-α reductase inhibitors (5-ARIs) and alpha-blockers for benign prostatic hyperplasia (BPH). Observational research of older men with BPH suggested an increase in ISH with 5-ARI use compared with nonuse; we aimed to address potential confounding by indication with an active comparator reference group. METHODS: Using Medicare data linked to the National Death Index (NDI) from 2007-2016, we implemented a retrospective cohort design in males aged ≥65 years who initiated 5-ARI or alpha-blocker use for BPH. ISH was identified using ICD-9-CM and ICD-10-CM diagnosis codes. Suicides were identified through cause-of-death information from the NDI. We used inverse probability of treatment weighted Cox proportional hazards regression to compare time-to-event between treatment groups, with robust variance estimation. RESULTS: The event rates for ISH and suicide, respectively, were 0.314 and 0.308 per 1000 person-years (PY) among 5-ARI users (n = 181,675), and 0.364 and 0.382 per 1000PY among alpha-blocker users (n = 850,476). For 5-ARI use relative to alpha-blocker use, hazard ratios (HRs) for ISH and suicide, respectively, were 0.88 (95% CI:0.62-1.25) and 0.82 (95% CI:0.54-1.24); for the composite outcome (non-fatal ISH or suicide), the HR was 0.88 (95% CI:0.66-1.16). Subgroup and sensitivity analyses supported these results. CONCLUSION: 5-ARI use was not associated with an increased risk for ISH or suicide compared to alpha-blocker use in older men with BPH. Study limitations included low event rates and potentially low sensitivity for ISH events.
ABSTRACT
Suicide risk prediction models can identify individuals for targeted intervention. Discussions of transparency, explainability, and transportability in machine learning presume complex prediction models with many variables outperform simpler models. We compared random forest, artificial neural network, and ensemble models with 1500 temporally defined predictors to logistic regression models. Data from 25,800,888 mental health visits made by 3,081,420 individuals in 7 health systems were used to train and evaluate suicidal behavior prediction models. Model performance was compared across several measures. All models performed well (area under the receiver operating curve [AUC]: 0.794-0.858). Ensemble models performed best, but improvements over a regression model with 100 predictors were minimal (AUC improvements: 0.006-0.020). Results are consistent across performance metrics and subgroups defined by race, ethnicity, and sex. Our results suggest simpler parametric models, which are easier to implement as part of routine clinical practice, perform comparably to more complex machine learning methods.
ABSTRACT
Post marketing safety surveillance depends in part on the ability to detect concerning clinical events at scale. Spontaneous reporting might be an effective component of safety surveillance, but it requires awareness and understanding among healthcare professionals to achieve its potential. Reliance on readily available structured data such as diagnostic codes risk under-coding and imprecision. Clinical textual data might bridge these gaps, and natural language processing (NLP) has been shown to aid in scalable phenotyping across healthcare records in multiple clinical domains. In this study, we developed and validated a novel incident phenotyping approach using unstructured clinical textual data agnostic to Electronic Health Record (EHR) and note type. It's based on a published, validated approach (PheRe) used to ascertain social determinants of health and suicidality across entire healthcare records. To demonstrate generalizability, we validated this approach on two separate phenotypes that share common challenges with respect to accurate ascertainment: 1) suicide attempt; 2) sleep-related behaviors. With samples of 89,428 records and 35,863 records for suicide attempt and sleep-related behaviors, respectively, we conducted silver standard (diagnostic coding) and gold standard (manual chart review) validation. We showed Area Under the Precision-Recall Curve of â¼ 0.77 (95% CI 0.75-0.78) for suicide attempt and AUPR â¼ 0.31 (95% CI 0.28-0.34) for sleep-related behaviors. We also evaluated performance by coded race and demonstrated differences in performance by race were dissimilar across phenotypes and require algorithmovigilance and debiasing prior to implementation.
ABSTRACT
BACKGROUND: The mechanism for anaphylaxis following mRNA COVID-19 vaccination has been widely debated; understanding this serious adverse event is important for future vaccines of similar design. A mechanism proposed is type I hypersensitivity (i.e., IgE-mediated mast cell degranulation) to polyethylene glycol (PEG). Using an assay that, uniquely, had been previously assessed in patients with anaphylaxis to PEG, our objective was to compare anti-PEG IgE in serum from mRNA COVID-19 vaccine anaphylaxis case-patients and persons vaccinated without allergic reactions. Secondarily, we compared anti-PEG IgG and IgM to assess alternative mechanisms. METHODS: Selected anaphylaxis case-patients reported to U.S. Vaccine Adverse Event Reporting System December 14, 2020-March 25, 2021 were invited to provide a serum sample. mRNA COVID-19 vaccine study participants with residual serum and no allergic reaction post-vaccination ("controls") were frequency matched to cases 3:1 on vaccine and dose number, sex and 10-year age category. Anti-PEG IgE was measured using a dual cytometric bead assay (DCBA). Anti-PEG IgG and IgM were measured using two different assays: DCBA and a PEGylated-polystyrene bead assay. Laboratorians were blinded to case/control status. RESULTS: All 20 case-patients were women; 17 had anaphylaxis after dose 1, 3 after dose 2. Thirteen (65â¯%) were hospitalized and 7 (35â¯%) were intubated. Time from vaccination to serum collection was longer for case-patients vs controls (post-dose 1: median 105 vs 21â¯days). Among Moderna recipients, anti-PEG IgE was detected in 1 of 10 (10â¯%) case-patients vs 8 of 30 (27â¯%) controls (pâ¯=â¯0.40); among Pfizer-BioNTech recipients, it was detected in 0 of 10 case-patients (0â¯%) vs 1 of 30 (3â¯%) controls (pâ¯>nâ¯0.99). Anti-PEG IgE quantitative signals followed this same pattern. Neither anti-PEG IgG nor IgM was associated with case status with both assay formats. CONCLUSION: Our results support that anti-PEG IgE is not a predominant mechanism for anaphylaxis post-mRNA COVID-19 vaccination.
Subject(s)
Anaphylaxis , COVID-19 Vaccines , COVID-19 , Female , Humans , Male , Anaphylaxis/etiology , COVID-19/prevention & control , COVID-19 Vaccines/adverse effects , Immunoglobulin E , Immunoglobulin G , Immunoglobulin M , Immunosuppressive Agents , Polyethylene Glycols/adverse effects , RNA, Messenger , Vaccination/adverse effectsABSTRACT
OBJECTIVE: Claims-based algorithms are used in the Food and Drug Administration Sentinel Active Risk Identification and Analysis System to identify occurrences of health outcomes of interest (HOIs) for medical product safety assessment. This project aimed to apply machine learning classification techniques to demonstrate the feasibility of developing a claims-based algorithm to predict an HOI in structured electronic health record (EHR) data. MATERIALS AND METHODS: We used the 2015-2019 IBM MarketScan Explorys Claims-EMR Data Set, linking administrative claims and EHR data at the patient level. We focused on a single HOI, rhabdomyolysis, defined by EHR laboratory test results. Using claims-based predictors, we applied machine learning techniques to predict the HOI: logistic regression, LASSO (least absolute shrinkage and selection operator), random forests, support vector machines, artificial neural nets, and an ensemble method (Super Learner). RESULTS: The study cohort included 32 956 patients and 39 499 encounters. Model performance (positive predictive value [PPV], sensitivity, specificity, area under the receiver-operating characteristic curve) varied considerably across techniques. The area under the receiver-operating characteristic curve exceeded 0.80 in most model variations. DISCUSSION: For the main Food and Drug Administration use case of assessing risk of rhabdomyolysis after drug use, a model with a high PPV is typically preferred. The Super Learner ensemble model without adjustment for class imbalance achieved a PPV of 75.6%, substantially better than a previously used human expert-developed model (PPV = 44.0%). CONCLUSIONS: It is feasible to use machine learning methods to predict an EHR-derived HOI with claims-based predictors. Modeling strategies can be adapted for intended uses, including surveillance, identification of cases for chart review, and outcomes research.
Subject(s)
Electronic Health Records , Machine Learning , Electronics , Humans , Outcome Assessment, Health Care , Pilot ProjectsABSTRACT
INTRODUCTION: This study examined state-level variation in chronic kidney disease (CKD) awareness using national estimates of disease awareness among adults in the U.S. with CKD. METHODS: Data on U.S. adults were obtained from two national, population-based surveys: (1) the Behavioral Risk Factor Surveillance System (BRFSS 2011; n=506,467), a state-level phone survey containing information on self-reported kidney disease; and (2) the National Health and Nutrition Examination Survey (NHANES 2005-2012; n=20,831), containing physical health examination, surveys containing data on self-reported kidney disease, risk factors, and laboratory values. CKD was defined as an estimated glomerular filtration rate of 15-59 mL/minute/1.73 m2 or urinary albumin-to-creatinine ratio >30 mg/g. As BRFSS does not include laboratory data, CKD status for each person was imputed (multiple) based on a logistic regression model predicting NHANES CKD status. CKD awareness in each state was estimated as the weighted proportion of BRFSS participants with imputed CKD who reported having kidney disease. RESULTS: Overall, estimated CKD awareness was 9.0% (95% CI=8.0%, 10.0%), ranging from 5.8% (95% CI=4.8%, 6.8%) in Iowa to 11.7% (95% CI=9.7%, 13.7%) in Arizona. Awareness was greater among adults with hypertension (12.0%) and diabetes (15.3%) than among adults without those conditions, and lower in Hispanics (6.0%) than in non-Hispanic whites (8.8%), non-Hispanic blacks (9.9%), and other racial/ethnic groups (12.7%). CONCLUSIONS: Among individuals with CKD, awareness of their condition was very low and varied approximately twofold among states. This is the first study to estimate awareness of kidney disease by state for the U.S. adult population.
Subject(s)
Diabetes Mellitus/epidemiology , Health Knowledge, Attitudes, Practice/ethnology , Hypertension/epidemiology , Renal Insufficiency, Chronic/epidemiology , Adult , Age Factors , Aged , Behavioral Risk Factor Surveillance System , Cross-Sectional Studies , Female , Glomerular Filtration Rate , Humans , Male , Middle Aged , Nutrition Surveys/statistics & numerical data , Prevalence , Renal Insufficiency, Chronic/prevention & control , Risk Factors , Self Report , United States/epidemiology , Young AdultABSTRACT
BACKGROUND: Pharmacies and pharmacists play an important role in the health care system, improving health outcomes and enhancing quality through better pharmaceutical care. Yet, little information is available to accurately evaluate pharmacy store quality and thereby encourage quality improvement at the pharmacy store level. OBJECTIVES: To (a) assess pharmacy performance in the area of medication adherence and (b) examine the impact of risk adjustment of performance scores on pharmacy rankings. METHODS: We used proportion of days covered (PDC) to compute pharmacy performance scores using the 2007 Mississippi Medicare administrative claims dataset. We calculated unadjusted and adjusted quality scores for 685 pharmacies serving 137,497 eligible Medicare beneficiaries. Risk-adjusted quality scores were computed using a hierarchical logistic regression model (Method 1) and the shrinkage estimators of the model (Method 2). Patient demographics, income subsidy status, and comorbidity burden were used as variables for risk adjustment. RESULTS: Unadjusted scores showed low levels of agreement (Cohen's kappa less than 0.45) with risk-adjusted scores in identifying statistical outliers based on 95% CIs. Unadjusted scores also failed to identify 39%-43% of the top 20% and bottom 20% of pharmacies and displayed moderate agreement (0.4 less than kappa less than 0.5) with risk-adjusted scores. Pharmacy classifications based on risk-adjusted scores obtained from different statistical methods showed high levels of agreement (0.79 less than kappa less than 0.98). CONCLUSIONS: In the risk-adjustment methods presented here, we account for many patient characteristics previously reported to be associated with medication adherence and available in our dataset. Risk-adjusted scores produced more robust indicators of pharmacy quality than unadjusted scores. Depending on the availability of important variables in the source data, the use of risk-adjusted quality indicators may lead to better evaluation of pharmacy quality and should be considered when providing public reports on pharmacy quality.