Development and Validation of a Claims-Based Model to Predict Categories of Obesity.

We developed and validated a claims-based algorithm that classifies patients into obesity categories. Using Medicare (2007-2017) and Medicaid (2000-2014) claims data linked to 2 electronic health record (EHR) systems in Boston, Massachusetts, we identified a cohort of patients with an EHR-based body mass index (BMI) measurement (calculated as weight (kg)/height (m)2). We used regularized regression to select from 137 variables and built generalized linear models to classify patients with BMIs of ≥25, ≥30, and ≥40. We developed the prediction model using EHR system 1 (training set) and validated it in EHR system 2 (validation set). The cohort contained 123,432 patients in the Medicare population and 40,736 patients in the Medicaid population. The model comprised 97 variables in the Medicare set and 95 in the Medicaid set, including BMI-related diagnosis codes, cardiovascular and antidiabetic drugs, and obesity-related comorbidities. The areas under the receiver-operating-characteristic curve in the validation set were 0.72, 0.75, and 0.83 (Medicare) and 0.66, 0.66, and 0.70 (Medicaid) for BMIs of ≥25, ≥30, and ≥40, respectively. The positive predictive values were 81.5%, 80.6%, and 64.7% (Medicare) and 81.6%, 77.5%, and 62.5% (Medicaid), for BMIs of ≥25, ≥30, and ≥40, respectively. The proposed model can identify obesity categories in claims databases when BMI measurements are missing and can be used for confounding adjustment, defining subgroups, or probabilistic bias analysis.

Improving the accuracy of automated gout flare ascertainment using natural language processing of electronic health records and linked Medicare claims data.

BACKGROUND: We aimed to determine whether integrating concepts from the notes from the electronic health record (EHR) data using natural language processing (NLP) could improve the identification of gout flares. METHODS: Using Medicare claims linked with EHR, we selected gout patients who initiated the urate-lowering therapy (ULT). Patients' 12-month baseline period and on-treatment follow-up were segmented into 1-month units. We retrieved EHR notes for months with gout diagnosis codes and processed notes for NLP concepts. We selected a random sample of 500 patients and reviewed each of their notes for the presence of a physician-documented gout flare. Months containing at least 1 note mentioning gout flares were considered months with events. We used 60% of patients to train predictive models with LASSO. We evaluated the models by the area under the curve (AUC) in the validation data and examined positive/negative predictive values (P/NPV). RESULTS: We extracted and labeled 839 months of follow-up (280 with gout flares). The claims-only model selected 20 variables (AUC = 0.69). The NLP concept-only model selected 15 (AUC = 0.69). The combined model selected 32 claims variables and 13 NLP concepts (AUC = 0.73). The claims-only model had a PPV of 0.64 [0.50, 0.77] and an NPV of 0.71 [0.65, 0.76], whereas the combined model had a PPV of 0.76 [0.61, 0.88] and an NPV of 0.71 [0.65, 0.76]. CONCLUSION: Adding NLP concept variables to claims variables resulted in a small improvement in the identification of gout flares. Our data-driven claims-only model and our combined claims/NLP-concept model outperformed existing rule-based claims algorithms reliant on medication use, diagnosis, and procedure codes.

Identifying Functional Status Impairment in People Living With Dementia Through Natural Language Processing of Clinical Documents: Cross-Sectional Study.

BACKGROUND: Assessment of activities of daily living (ADLs) and instrumental ADLs (iADLs) is key to determining the severity of dementia and care needs among older adults. However, such information is often only documented in free-text clinical notes within the electronic health record and can be challenging to find. OBJECTIVE: This study aims to develop and validate machine learning models to determine the status of ADL and iADL impairments based on clinical notes. METHODS: This cross-sectional study leveraged electronic health record clinical notes from Mass General Brigham's Research Patient Data Repository linked with Medicare fee-for-service claims data from 2007 to 2017 to identify individuals aged 65 years or older with at least 1 diagnosis of dementia. Notes for encounters both 180 days before and after the first date of dementia diagnosis were randomly sampled. Models were trained and validated using note sentences filtered by expert-curated keywords (filtered cohort) and further evaluated using unfiltered sentences (unfiltered cohort). The model's performance was compared using area under the receiver operating characteristic curve and area under the precision-recall curve (AUPRC). RESULTS: The study included 10,000 key-term-filtered sentences representing 441 people (n=283, 64.2% women; mean age 82.7, SD 7.9 years) and 1000 unfiltered sentences representing 80 people (n=56, 70% women; mean age 82.8, SD 7.5 years). Area under the receiver operating characteristic curve was high for the best-performing ADL and iADL models on both cohorts (>0.97). For ADL impairment identification, the random forest model achieved the best AUPRC (0.89, 95% CI 0.86-0.91) on the filtered cohort; the support vector machine model achieved the highest AUPRC (0.82, 95% CI 0.75-0.89) for the unfiltered cohort. For iADL impairment, the Bio+Clinical bidirectional encoder representations from transformers (BERT) model had the highest AUPRC (filtered: 0.76, 95% CI 0.68-0.82; unfiltered: 0.58, 95% CI 0.001-1.0). Compared with a keyword-search approach on the unfiltered cohort, machine learning reduced false-positive rates from 4.5% to 0.2% for ADL and 1.8% to 0.1% for iADL. CONCLUSIONS: In this study, we demonstrated the ability of machine learning models to accurately identify ADL and iADL impairment based on free-text clinical notes, which could be useful in determining the severity of dementia.

Emulation of Randomized Clinical Trials With Nonrandomized Database Analyses: Results of 32 Clinical Trials.

Importance: Nonrandomized studies using insurance claims databases can be analyzed to produce real-world evidence on the effectiveness of medical products. Given the lack of baseline randomization and measurement issues, concerns exist about whether such studies produce unbiased treatment effect estimates. Objective: To emulate the design of 30 completed and 2 ongoing randomized clinical trials (RCTs) of medications with database studies using observational analogues of the RCT design parameters (population, intervention, comparator, outcome, time [PICOT]) and to quantify agreement in RCT-database study pairs. Design, Setting, and Participants: New-user cohort studies with propensity score matching using 3 US claims databases (Optum Clinformatics, MarketScan, and Medicare). Inclusion-exclusion criteria for each database study were prespecified to emulate the corresponding RCT. RCTs were explicitly selected based on feasibility, including power, key confounders, and end points more likely to be emulated with real-world data. All 32 protocols were registered on ClinicalTrials.gov before conducting analyses. Emulations were conducted from 2017 through 2022. Exposures: Therapies for multiple clinical conditions were included. Main Outcomes and Measures: Database study emulations focused on the primary outcome of the corresponding RCT. Findings of database studies were compared with RCTs using predefined metrics, including Pearson correlation coefficients and binary metrics based on statistical significance agreement, estimate agreement, and standardized difference. Results: In these highly selected RCTs, the overall observed agreement between the RCT and the database emulation results was a Pearson correlation of 0.82 (95% CI, 0.64-0.91), with 75% meeting statistical significance, 66% estimate agreement, and 75% standardized difference agreement. In a post hoc analysis limited to 16 RCTs with closer emulation of trial design and measurements, concordance was higher (Pearson r, 0.93; 95% CI, 0.79-0.97; 94% meeting statistical significance, 88% estimate agreement, 88% standardized difference agreement). Weaker concordance occurred among 16 RCTs for which close emulation of certain design elements that define the research question (PICOT) with data from insurance claims was not possible (Pearson r, 0.53; 95% CI, 0.00-0.83; 56% meeting statistical significance, 50% estimate agreement, 69% standardized difference agreement). Conclusions and Relevance: Real-world evidence studies can reach similar conclusions as RCTs when design and measurements can be closely emulated, but this may be difficult to achieve. Concordance in results varied depending on the agreement metric. Emulation differences, chance, and residual confounding can contribute to divergence in results and are difficult to disentangle.

Emulating Randomized Clinical Trials With Nonrandomized Real-World Evidence Studies: First Results From the RCT DUPLICATE Initiative.

BACKGROUND: Regulators are evaluating the use of noninterventional real-world evidence (RWE) studies to assess the effectiveness of medical products. The RCT DUPLICATE initiative (Randomized, Controlled Trials Duplicated Using Prospective Longitudinal Insurance Claims: Applying Techniques of Epidemiology) uses a structured process to design RWE studies emulating randomized, controlled trials (RCTs) and compare results. We report findings of the first 10 trial emulations, evaluating cardiovascular outcomes of antidiabetic or antiplatelet medications. METHODS: We selected 3 active-controlled and 7 placebo-controlled RCTs for replication. Using patient-level claims data from US commercial and Medicare payers, we implemented inclusion and exclusion criteria, selected primary end points, and comparator populations to emulate those of each corresponding RCT. Within the trial-mimicking populations, we conducted propensity score matching to control for >120 preexposure confounders. All study measures were prospectively defined and protocols registered before hazard ratios and 95% CIs were computed. Success criteria for the primary analysis were prespecified for each replication. RESULTS: Despite attempts to emulate RCT design as closely as possible, differences between the RCT and corresponding RWE study populations remained. The regulatory conclusions were equivalent in 6 of 10. The RWE emulations achieved a hazard ratio estimate that was within the 95% CI from the corresponding RCT in 8 of 10 studies. In 9 of 10, either the regulatory or estimate agreement success criteria were fulfilled. The largest differences in effect estimates were found for RCTs where second-generation sulfonylureas were used as a proxy for placebo regarding cardiovascular effects. Nine of 10 replications had a standardized difference between effect estimates of <2, which suggests differences within expected random variation. CONCLUSIONS: Agreement between RCT and RWE findings varies depending on which agreement metric is used. Interim findings indicate that selection of active comparator therapies with similar indications and use patterns enhances the validity of RWE. Even in the context of active comparators, concordance between RCT and RWE findings is not guaranteed, partially because trials are not emulated exactly. More trial emulations are needed to understand how often and in what contexts RWE findings match RCTs. Registration: URL: https://www.clinicaltrials.gov; Unique identifiers: NCT03936049, NCT04215523, NCT04215536, NCT03936010, NCT03936036, NCT03936062, NCT03936023, NCT03648424, NCT04237935, NCT04237922.

Comparative effectiveness of Empagliflozin in reducing the burden of recurrent cardiovascular hospitalizations among older adults with diabetes in routine clinical care.

BACKGROUND: The effect of sodium glucose cotransporter 2 inhibitors (SGLT2i) on the total (first and recurrent) burden of cardiovascular (CV) hospitalizations, including hospitalization for heart failure, myocardial infarction, and stroke, is poorly understood. OBJECTIVE: To assess the effect of empagliflozin, an SGLT2i, on total CV hospitalizations among older adults with T2D. METHODS: Using data from Medicare fee-for-service (08/2014-09/2017), we identified 1:1 propensity score-matched cohorts of patients with T2D initiating empagliflozin versus sitagliptin or empagliflozin versus glucagon-like peptide-1 receptor agonists (GLP-1RA), balancing >140 baseline covariates. We compared the risk of first and recurrent hospitalizations with any CV condition as the primary discharge diagnosis (ICD-9: 390-459; ICD-10: I00-I99), hospitalizations for heart failure (HHF), and myocardial infarctions (MI) or stroke. We estimated treatment effects based on the Ghosh-Lin semiparametric model for recurrent events as primary and joint frailty model as secondary analysis. RESULTS: We included 11,429 matched-pairs of empagliflozin and sitagliptin initiators and 17,502 matched-pairs of empagliflozin and GLP1-RA initiators with an average age of 72 years. Empagliflozin was associated with a reduced risk of total CV hospitalizations (0.80 [0.69-0.93] vs sitagliptin; 0.88 [0.77-1.00] vs GLP-1RA) and total HHF (0.70 [0.51-0.98] vs sitagliptin; 0.76 [0.56-1.03] vs GLP1-RA) over a mean follow up of 6.3 months. No differences between treatments were observed for MI or stroke. Results were consistent for joint frailty models. CONCLUSION: Empagliflozin, compared to sitagliptin or to a lesser extent GLP1-RA, was associated with a reduction in the burden of total CV hospitalizations and HHF in older patients with T2D.

Effectiveness and safety of empagliflozin in routine care patients: Results from the EMPagliflozin compaRative effectIveness and SafEty (EMPRISE) study.

AIM: To investigate effectiveness and safety outcomes among patients with type 2 diabetes (T2D) initiating empagliflozin versus dipeptidyl peptidase-4 (DPP-4) inhibitor treatment across the broad spectrum of cardiovascular risk. METHODS: In a population-based cohort study we identified 39 072 pairs of 1:1 propensity score-matched adult patients with T2D initiating empagliflozin or DPP-4 inhibitors, using data from 2 US commercial insurance databases and Medicare between August 2014 and September 2017. The primary outcomes were a composite of myocardial infarction (MI)/stroke, and hospitalization for heart failure (HHF). Safety outcomes were bone fractures, lower-limb amputations (LLAs), diabetic ketoacidosis (DKA), and acute kidney injury (AKI). We estimated pooled hazard ratios (HRs) and 95% confidence intervals (CIs) adjusting for more than 140 baseline covariates. RESULTS: Study participants had a mean age of 60 years and only 28% had established cardiovascular disease. Compared to DPP-4 inhibitors, empagliflozin was associated with similar risk of MI/stroke (HR 0.99 [95% CI 0.81-1.21]), and lower risk of HHF (HR 0.48 [95% CI 0.35-0.67] and 0.63 [95% CI 0.54-0.74], based on a primary and any heart failure discharge diagnosis, respectively). The HR was 0.52 (95% CI 0.38-0.72) for all-cause mortality (ACM) and 0.83 (95% CI 0.70-0.98) for a composite of MI/stroke/ACM. Empagliflozin was associated with a similar risk of LLA and fractures, an increased risk of DKA (HR 1.71 [95% CI 1.08-2.71]) and a decreased risk of AKI (HR 0.60 [95% CI 0.43-0.85]). CONCLUSIONS: In clinical practice, the initiation of empagliflozin versus a DPP-4 inhibitor was associated with a lower risk of HHF, ACM and MI/stroke/ACM, a similar risk of MI/stroke, and a safety profile consistent with documented information.

Accuracy of identifying diagnosis of moderate to severe chronic kidney disease in administrative claims data.

BACKGROUND: Prior validation studies of claims-based definitions of chronic kidney disease (CKD) using ICD-9 codes reported overall low sensitivity, high specificity, and variable but reasonable PPV. No studies to date have evaluated the accuracy of ICD-10 codes to identify a US patient population with CKD. METHODS: We assessed the accuracy of claims-based algorithms to identify adults with CKD Stages 3-5 compared with laboratory values in a subset (~40%) of a US commercial insurance claims database (Optum's de-identified Clinformatics® Data Mart Database). We calculated the positive predictive value (PPV) of one or two ICD-9 (2012-2014) or ICD-10 (2016-2018) codes for CKD compared with a lab-based estimated glomerular filtration rate (eGFR) occurring within prespecified windows (±90 days, ±180 days, ±365 days) of the ICD-based CKD code(s). RESULTS: The study population ranged between 104 774 and 161 305 patients (ICD-9 cohorts) and between 285 520 and 373 220 patients (ICD-10 cohorts). The mean age was 74.4 years (ICD-9) and 75.6 years (ICD-10) and the median eGFR was 48 ml/min/1.73 m2 . The algorithm of two CKD codes compared with a lab value ±90 days of the first code achieved the highest PPV (PPV 86.36% [ICD-9] and 86.07% [ICD-10]). Overall, ICD-10 based codes had comparable PPVs to ICD-9 based codes and all ICD-10 based algorithms had PPVs >80%. The algorithm of one CKD code compared with laboratory value ±180 days maintained the PPV above 80% but still retained a large number of patients (PPV 80.32% [ICD-9] and 81.56% [ICD-10]). CONCLUSION: An ICD-10-based definition of CKD identified with sufficient accuracy a patient population with CKD Stages 3-5. Our findings suggest that claims databases could be used for future real-world research studies in patients with CKD Stages 3-5.

Frailty and Clinical Outcomes of Direct Oral Anticoagulants Versus Warfarin in Older Adults With Atrial Fibrillation : A Cohort Study.

BACKGROUND: The role of differing levels of frailty in the choice of oral anticoagulants for older adults with atrial fibrillation (AF) is unclear. OBJECTIVE: To examine the outcomes of direct oral anticoagulants (DOACs) versus warfarin by frailty levels. DESIGN: 1:1 propensity score-matched analysis of Medicare data, 2010 to 2017. SETTING: Community. PATIENTS: Medicare beneficiaries with AF who initiated use of dabigatran, rivaroxaban, apixaban, or warfarin. MEASUREMENTS: Composite end point of death, ischemic stroke, or major bleeding by frailty levels, defined by a claims-based frailty index. RESULTS: In the dabigatran-warfarin cohort (n = 158 730; median follow-up, 72 days), the event rate per 1000 person-years was 63.5 for dabigatran initiators and 65.6 for warfarin initiators (hazard ratio [HR], 0.98 [95% CI, 0.92 to 1.05]; rate difference [RD], -2.2 [CI, -6.5 to 2.1]). For nonfrail, prefrail, and frail persons, HRs were 0.81 (CI, 0.68 to 0.97), 0.98 (CI, 0.90 to 1.08), and 1.09 (CI, 0.96 to 1.23), respectively. In the rivaroxaban-warfarin cohort (n = 275 944; median follow-up, 82 days), the event rate per 1000 person-years was 77.8 for rivaroxaban initiators and 83.7 for warfarin initiators (HR, 0.98 [CI, 0.94 to 1.02]; RD, -5.9 [CI, -9.4 to -2.4]). For nonfrail, prefrail, and frail persons, HRs were 0.88 (CI, 0.77 to 0.99), 1.04 (CI, 0.98 to 1.10), and 0.96 (CI, 0.89 to 1.04), respectively. In the apixaban-warfarin cohort (n = 218 738; median follow-up, 84 days), the event rate per 1000 person-years was 60.1 for apixaban initiators and 92.3 for warfarin initiators (HR, 0.68 [CI, 0.65 to 0.72]; RD, -32.2 [CI, -36.1 to -28.3]). For nonfrail, prefrail, and frail persons, HRs were 0.61 (CI, 0.52 to 0.71), 0.66 (CI, 0.61 to 0.70), and 0.73 (CI, 0.67 to 0.80), respectively. LIMITATIONS: Residual confounding and lack of clinical frailty assessment. CONCLUSION: For older adults with AF, apixaban was associated with lower rates of adverse events across all frailty levels. Dabigatran and rivaroxaban were associated with lower event rates only among nonfrail patients. PRIMARY FUNDING SOURCE: National Institute on Aging.

Sodium-Glucose Cotransporter-2 Inhibitors Versus Glucagon-like Peptide-1 Receptor Agonists and the Risk for Cardiovascular Outcomes in Routine Care Patients With Diabetes Across Categories of Cardiovascular Disease.

BACKGROUND: Both sodium-glucose cotransporter-2 (SGLT2) inhibitors and glucagon-like peptide-1 receptor agonists (GLP-1 RAs) have shown cardiovascular benefits in placebo-controlled trials of patients with type 2 diabetes (T2D) and established cardiovascular disease (CVD). OBJECTIVE: To evaluate whether SGLT2 inhibitors and GLP-1 RAs are associated with differential cardiovascular benefit among T2D patients with and without CVD. DESIGN: Population-based cohort study. SETTING: Medicare and 2 U.S. commercial claims data sets (April 2013 to December 2017). PARTICIPANTS: 1:1 propensity score-matched adult T2D patients with and without CVD (52 901 and 133 139 matched pairs) initiating SGLT2 inhibitor versus GLP-1 RA therapy. MEASUREMENTS: Primary outcomes were myocardial infarction (MI) or stroke hospitalization and hospitalization for heart failure (HHF). Pooled hazard ratios (HRs) and rate differences (RDs) per 1000 person-years were estimated, with 95% CIs, controlling for 138 preexposure covariates. RESULTS: The initiation of SGLT2 inhibitor versus GLP-1 RA therapy was associated with a slightly lower risk for MI or stroke in patients with CVD (HR, 0.90 [95% CI, 0.82 to 0.98]; RD, -2.47 [CI, -4.45 to -0.50]) but similar risk in those without CVD (HR, 1.07 [CI, 0.97 to 1.18]; RD, 0.38 [CI, -0.30 to 1.07]). The initiation of SGLT2 inhibitor versus GLP-1 RA therapy was associated with reductions in HHF risk regardless of baseline CVD in patients with CVD (HR, 0.71 [CI, 0.64 to 0.79]; RD, -4.97 [CI, -6.55 to -3.39]) and in those without CVD (HR, 0.69 [CI, 0.56 to 0.85]; RD, -0.58 [CI, -0.91 to -0.25]). LIMITATION: Treatment selection was not randomized. CONCLUSION: Use of SGLT2 inhibitors versus GLP-1 RAs was associated with consistent reductions in HHF risk among T2D patients with and without CVD, although the absolute benefit was greater in patients with CVD. There were no large differences in risk for MI or stroke among T2D patients with and without CVD. PRIMARY FUNDING SOURCE: Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School.

The associated press stylebook changes and the use of addiction-related stigmatizing terms in news media.

BACKGROUND: The May 2017 publication of the updated Associated Press (AP) Stylebook offered guidance that advised against stigmatizing. The objective of this study was to assess the frequency of stigmatizing terms describing substance use and disorder in news articles before and after the update of the AP Stylebook.Methods: We reviewed articles containing terms "opioid" or "addiction" from three major news outlets. We counted the number of AP Stylebook proscribed terms found in each article and compared the proportions of articles from each outlet with proscribed terms before and after AP Stylebook publication.Results: In 2016, 56-94% of articles from each of the three news outlets contained a proscribed term. The use of proscribed terms in articles identified by searching "opioid" published in the New York Times decreased (72% vs. 94%, p = 0.01) after the AP Stylebook change. For other news outlets, there were no significant differences, though all proportions were lower after publication.Conclusions: Evidence for a decrease in the use of stigmatizing terminology for substance use and disorders in news articles after publication of guidance was limited. Additional efforts should address use of such terminology to maximize implementation of effective addiction health policies and practices.

Empagliflozin and the Risk of Heart Failure Hospitalization in Routine Clinical Care.

BACKGROUND: The EMPA-REG OUTCOME trial (Empagliflozin Cardiovascular Outcome Event Trial in Type 2 diabetes Mellitus Patients) showed that empagliflozin, a sodium-glucose cotransporter-2 inhibitor, reduces the risk of hospitalization for heart failure (HHF) by 35%, on top of standard of care in patients with type 2 diabetes mellitus (T2D) and established cardiovascular disease. The EMPRISE (Empagliflozin Comparative Effectiveness and Safety) study aims to assess empagliflozin's effectiveness, safety, and healthcare utilization in routine care from August 2014 through September 2019. In this first interim analysis, we investigated the risk of HHF among T2D patients initiating empagliflozin versus sitagliptin, a dipeptidyl peptidase-4 inhibitor. METHODS: Within 2 commercial and 1 federal (Medicare) claims data sources in the United States, we identified a 1:1 propensity score-matched cohort of T2D patients ≥18 years old initiating empagliflozin or sitagliptin from August 2014 through September 2016. The HHF outcome was defined as a HF discharge diagnosis in the primary position (HHF-specific); a broader definition was based on a HF discharge diagnosis in any position (HHF-broad). Hazard ratios (HRs) and 95% CIs were estimated controlling for over 140 baseline characteristics in each data source and pooled by fixed-effects meta-analysis. RESULTS: After propensity-score matching, we identified 16,443 patient pairs who initiated empagliflozin or sitagliptin. Average age was approximately 59 years, almost 54% of the participants were males, and approximately 25% had records of existing cardiovascular disease. Compared with sitagliptin, the initiation of empagliflozin decreased the risk of HHF-specific by 50% (HR, 0.50; 95% CI, 0.28-0.91), and the risk of HHF-broad by 49% (HR, 0.51;95% CI, 0.39-0.68), over a mean follow-up of 5.3 months. The results were consistent in patients with and without baseline cardiovascular disease, and for empagliflozin at both the 10- and 25-mg daily doses; analyses comparing empagliflozin versus the dipeptidyl peptidase-4 inhibitor class, and comparing sodium-glucose cotransporter-2 inhibitor versus dipeptidyl peptidase-4 inhibitor classes also produced consistent findings. CONCLUSIONS: The first interim analysis from EMPRISE showed that compared with sitagliptin, the initiation of empagliflozin was associated with a decreased risk of HHF among patients with T2D as treated in routine care, with and without a history of cardiovascular disease. CLINICAL TRIAL REGISTRATION: URL: https://www.clinicaltrials.gov . Unique identifier: NCT03363464.

Development and Validation of an Intracranial Hemorrhage Risk Score in Older Adults with Atrial Fibrillation Treated with Oral Anticoagulant.

Background: High risk of intracranial hemorrhage (ICH) is a leading reason for withholding anticoagulation in patients with atrial fibrillation (AF). We aimed to develop a claims-based ICH risk prediction model in older adults with AF initiating oral anticoagulation (OAC). Methods: We used US Medicare claims data to identify new users of OAC aged ≥65 years with AF in 2010-2017. We used regularized Cox regression to select predictors of ICH. We compared our AF ICH risk score with the HAS-BLED bleed risk and Homer fall risk scores by area under the receiver operating characteristic curve (AUC) and assessed net reclassification improvement (NRI) when predicting 1-year risk of ICH. Results: Our study cohort comprised 840,020 patients (mean [SD] age 77.5 [7.4] years and female 52.2%) split geographically into training (3963 ICH events [0.6%] in 629,804 patients) and validation (1397 ICH events [0.7%] in 210,216 patients) sets. Our AF ICH risk score, including 50 predictors, had superior AUCs of 0.653 and 0.650 in the training and validation sets than the HAS-BLED score of 0.580 and 0.567 (p<0.001) and the Homer score of 0.624 and 0.623 (p<0.001). In the validation set, our AF ICH risk score reclassified 57.8%, 42.5%, and 43.9% of low, intermediate, and high-risk patients, respectively, by HAS-BLED score (NRI: 15.3%, p<0.001). Similarly, it reclassified 0.0, 44.1, and 19.4% of low, intermediate, and high-risk patients, respectively, by the Homer score (NRI: 21.9%, p<0.001). Conclusion: Our novel claims-based ICH risk prediction model outperformed the standard HAS-BLED score and can inform OAC prescribing decisions.

The impact of using reinforcement learning to personalize communication on medication adherence: findings from the REINFORCE trial.

Text messaging can promote healthy behaviors, like adherence to medication, yet its effectiveness remains modest, in part because message content is rarely personalized. Reinforcement learning has been used in consumer technology to personalize content but with limited application in healthcare. We tested a reinforcement learning program that identifies individual responsiveness ("adherence") to text message content and personalizes messaging accordingly. We randomized 60 individuals with diabetes and glycated hemoglobin A1c [HbA1c] ≥ 7.5% to reinforcement learning intervention or control (no messages). Both arms received electronic pill bottles to measure adherence. The intervention improved absolute adjusted adherence by 13.6% (95%CI: 1.7%-27.1%) versus control and was more effective in patients with HbA1c 7.5- < 9.0% (36.6%, 95%CI: 25.1%-48.2%, interaction p < 0.001). We also explored whether individual patient characteristics were associated with differential response to tested behavioral factors and unique clusters of responsiveness. Reinforcement learning may be a promising approach to improve adherence and personalize communication at scale.

The impact of electronic health record discontinuity on prediction modeling.

BACKGROUND: To determine the impact of electronic health record (EHR)-discontinuity on the performance of prediction models. METHODS: The study population consisted of patients with a history of cardiovascular (CV) comorbidities identified using US Medicare claims data from 2007 to 2017, linked to EHR from two networks (used as model training and validation set, respectively). We built models predicting one-year risk of mortality, major CV events, and major bleeding events, stratified by high vs. low algorithm-predicted EHR-continuity. The best-performing models for each outcome were chosen among 5 commonly used machine-learning models. We compared model performance by Area under the ROC curve (AUROC) and Area under the precision-recall curve (AUPRC). RESULTS: Based on 180,950 in the training and 103,061 in the validation set, we found EHR captured only 21.0-28.1% of all the non-fatal outcomes in the low EHR-continuity cohort but 55.4-66.1% of that in the high EHR-continuity cohort. In the validation set, the best-performing model developed among high EHR-continuity patients had consistently higher AUROC than that based on low-continuity patients: AUROC was 0.849 vs. 0.743 when predicting mortality; AUROC was 0.802 vs. 0.659 predicting the CV events; AUROC was 0.635 vs. 0.567 predicting major bleeding. We observed a similar pattern when using AUPRC as the outcome metric. CONCLUSIONS: Among patients with CV comorbidities, when predicting mortality, major CV events, and bleeding outcomes, the prediction models developed in datasets with low EHR-continuity consistently had worse performance compared to models developed with high EHR-continuity.

Scalable Feature Engineering from Electronic Free Text Notes to Supplement Confounding Adjustment of Claims-Based Pharmacoepidemiologic Studies.

Natural language processing (NLP) tools turn free-text notes (FTNs) from electronic health records (EHRs) into data features that can supplement confounding adjustment in pharmacoepidemiologic studies. However, current applications are difficult to scale. We used unsupervised NLP to generate high-dimensional feature spaces from FTNs to improve prediction of drug exposure and outcomes compared with claims-based analyses. We linked Medicare claims with EHR data to generate three cohort studies comparing different classes of medications on the risk of various clinical outcomes. We used "bag-of-words" to generate features for the top 20,000 most prevalent terms from FTNs. We compared machine learning (ML) prediction algorithms using different sets of candidate predictors: Set1 (39 researcher-specified variables), Set2 (Set1 + ML-selected claims codes), and Set3 (Set1 + ML-selected NLP-generated features), vs. Set4 (Set1 + 2 + 3). When modeling treatment choice, we observed a consistent pattern across the examples: ML models utilizing Set4 performed best followed by Set2, Set3, then Set1. When modeling the outcome risk, there was little to no improvement beyond models based on Set1. Supplementing claims data with NLP-generated features from free text notes improved prediction of prescribing choices but had little or no improvement on clinical risk prediction. These findings have implications for strategies to improve confounding using EHR data in pharmacoepidemiologic studies.

Development and Validation of a Novel Tool to Predict Model for End-Stage Liver Disease (MELD) Scores in Cirrhosis, Using Administrative Datasets.

Background: The Model for End-Stage Liver Disease (MELD) score predicts disease severity and mortality in cirrhosis. To improve cirrhosis phenotyping in administrative databases lacking laboratory data, we aimed to develop and externally validate claims-based MELD prediction models, using claims data linked to electronic health records (EHR). Methods: We included adults with established cirrhosis in two Medicare-linked EHR networks (training and internal validation; 2007-2017), and a Medicaid-linked EHR network (external validation; 2000-2014). Using least absolute shrinkage and selection operator (LASSO) with 5-fold cross-validation, we selected among 146 investigator-specified variables to develop models for predicting continuous MELD and relevant MELD categories (MELD<10, MELD≥15 and MELD≥20), with observed MELD calculated from laboratory data. Regression coefficients for each model were applied to the validation sets to predict patient-level MELD and assess model performance. Results: We identified 4501 patients in the Medicare training set (mean age 75.1 years, 18.5% female, mean MELD=13.0), and 2435 patients in the Medicare validation set (mean age: 74.3 years, 31.7% female, mean MELD=12.3). Our final model for predicting continuous MELD included 112 variables, explaining 58% of observed MELD variability; in the Medicare validation set, the area-under-the-receiver operating characteristic curves (AUC) for MELD<10 and MELD≥15 were 0.84 and 0.90, respectively; the AUC for the model predicting MELD≥20 (using 27 variables) was 0.93. Overall, these models correctly classified 77% of patients with MELD<10 (95% CI=0.75-0.78), 85% of patients with MELD≥15 (95% CI=0.84-0.87), and 87% of patients with MELD≥20 (95% CI=0.86-0.88). Results were consistent in the external validation set (n=2240). Conclusion: Our MELD prediction tools can be used to improve cirrhosis phenotyping in administrative datasets lacking laboratory data.

Comparative Effectiveness and Safety of Oral Anticoagulants by Dementia Status in Older Patients With Atrial Fibrillation.

Importance: The development of an optimal stroke prevention strategy, including the use of oral anticoagulant (OAC) therapy, is particularly important for patients with atrial fibrillation (AF) who are living with dementia, a condition that increases the risk of adverse outcomes. However, data on the role of dementia in the safety and effectiveness of OACs are limited. Objective: To assess the comparative safety and effectiveness of specific OACs by dementia status among older patients with AF. Design, Setting, and Participants: This retrospective comparative effectiveness study used 1:1 propensity score matching among 1 160 462 patients 65 years or older with AF. Data were obtained from the Optum Clinformatics Data Mart (January 1, 2013, to June 30, 2021), IBM MarketScan Research Database (January 1, 2013, to December 31, 2020), and Medicare claims databases maintained by the Centers for Medicare & Medicaid Services (inpatient, outpatient, and pharmacy; January 1, 2013, to December 31, 2017). Data analysis was performed from September 1, 2021, to May 24, 2022. Exposures: Apixaban, dabigatran, rivaroxaban, or warfarin. Main Outcomes and Measures: Composite end point of ischemic stroke or major bleeding events over the 6-month period after OAC initiation, pooled across databases using random-effects meta-analyses. Results: Among 1 160 462 patients with AF, the mean (SD) age was 77.4 (7.2) years; 50.2% were male, 80.5% were White, and 7.9% had dementia. Three comparative new-user cohorts were established: warfarin vs apixaban (501 990 patients; mean [SD] age, 78.1 [7.4] years; 50.2% female), dabigatran vs apixaban (126 718 patients; mean [SD] age, 76.5 [7.1] years; 52.0% male), and rivaroxaban vs apixaban (531 754 patients; mean [SD] age, 76.9 [7.2] years; 50.2% male). Among patients with dementia, compared with apixaban users, a higher rate of the composite end point was observed in warfarin users (95.7 events per 1000 person-years [PYs] vs 64.2 events per 1000 PYs; adjusted hazard ratio [aHR], 1.5; 95% CI, 1.3-1.7), dabigatran users (84.5 events per 1000 PYs vs 54.9 events per 1000 PYs; aHR, 1.5; 95% CI, 1.2-2.0), and rivaroxaban users (87.4 events per 1000 PYs vs 68.5 events per 1000 PYs; aHR, 1.3; 95% CI, 1.1-1.5). In all 3 comparisons, the magnitude of the benefits associated with apixaban was similar regardless of dementia diagnosis on the HR scale but differed substantially on the rate difference (RD) scale. The adjusted RD of the composite outcome per 1000 PYs for warfarin vs apixaban users was 29.8 (95% CI, 18.4-41.1) events in patients with dementia vs 16.0 (95% CI, 13.6-18.4) events in patients without dementia. The corresponding adjusted RD estimates of the composite outcome were 29.6 (95% CI, 11.6-47.6) events per 1000 PYs in patients with dementia vs 5.8 (95% CI, 1.1-10.4) events per 1000 PYs in patients without dementia for dabigatran vs apixaban users and 20.5 (95% CI, 9.9-31.1) events per 1000 PYs in patients with dementia vs 15.9 (95% CI, 11.4-20.3) events per 1000 PYs in patients without dementia for rivaroxaban vs apixaban users. The pattern was more distinct for major bleeding than for ischemic stroke. Conclusions and Relevance: In this comparative effectiveness study, apixaban was associated with lower rates of major bleeding and ischemic stroke compared with other OACs. The increased absolute risks associated with other OACs compared with apixaban were greater among patients with dementia than those without dementia, particularly for major bleeding. These findings support the use of apixaban for anticoagulation therapy in patients living with dementia who have AF.

Predicting Treatment Effects of a New-to-Market Drug in Clinical Practice Based on Phase III Randomized Trial Results.

Trial results may not be generalizable to target populations treated in clinical practice with different distributions of baseline characteristics that modify the treatment effect. We used outcome models developed with trial data to predict treatment effects in Medicare populations. We used data from the Randomized Evaluation of Long-Term Anticoagulation Therapy trial (RE-LY), which investigated the effect of dabigatran vs. warfarin on stroke or systemic embolism (stroke/SE) among patients with atrial fibrillation. We developed outcome models by fitting proportional hazards models in trial data. Target populations were trial-eligible Medicare beneficiaries who initiated dabigatran or warfarin in 2010-2011 ("early") and 2010-2017 ("extended"). We predicted 2-year risk ratios (RRs) and risk differences (RDs) for stroke/SE, major bleeding, and all-cause death in the Medicare populations using the observed baseline characteristics. The trial and early target populations had similar mean (SD) CHADS2 scores (2.15 (SD 1.13) vs. 2.15 (SD 0.91)) but different mean ages (71 vs. 79 years). Compared with RE-LY, the early Medicare population had similar predicted benefit of dabigatran vs. warfarin for stroke/SE (trial RR = 0.63, 95% confidence interval (CI) = 0.50 to 0.76 and RD = -1.37%, -1.96% to -0.77%, Medicare RR = 0.73, 0.65 to 0.82 and RD = -0.92%, -1.26% to -0.59%) and risks for major bleeding and all-cause death. The time-extended target population showed similar results. Outcome model-based prediction facilitates estimating the average treatment effects of a drug in different target populations when treatment and outcome data are unreliable or unavailable. The predicted effects may inform payers' coverage decisions for patients, especially shortly after a drug's launch when observational data are scarce.