Fenglong Xie

Purpose: Few studies have reported the agreement between medication information derived from ambulatory EHR data compared to administrative claims for high‐cost specialty drugs. We used data from a national EHR‐enabled registry, the Rheumatology Informatics System for Effectiveness (RISE), with linked Medicare claims in a population of patients with rheumatoid arthritis (RA) to investigate variations in agreement for different biologic disease‐modifying agents (bDMARDs) between two data sources (RISE EHR data vs. Medicare claims), categorized by drug, route of administration, and patient insurance factors (dual eligibility).Methods: Patients ≥ 65 years old, with ≥ 2 visits in RISE with RA ICD codes ≥ 30 days apart, and continuous enrollment in Medicare Parts B and D in 2017–2018 were included. We classified patients as bDMARD users or nonusers in Medicare claims or EHR data in 2018, and we calculated sensitivity, specificity, positive predicted value (PPV), and negative predicted value (NPV) of EHR data for identifying bDMARD users, using Medicare as the reference standard. We also calculated these metrics after stratifying by clinic‐administered (Part B) versus. pharmacy‐dispensed (Part D) bDMARDs and by patient dual‐eligibility.Results: A total of 26 097 patients were included in the study. Using Medicare claims as the reference standard, EHR data had a sensitivity of 75.0%–90.8% for identifying patients with the same medication and route. PPV for Part B bDMARDs was higher compared with Part D bDMARDs (range 94.3%–97.3% vs. 51.0%–69.6%). We observed higher PPVs for Part D bDMARDs among patients who were dual‐eligible (range 82.4%–95.1%).Conclusion: The risk of misclassification of drug exposure based on EHR data sources alone is small for Medicare Part B bDMARDs but could be as high as 50% for Part D bDMARDs, in particular for patients who are not dually eligible for Medicare and Medicaid.

Background: Digital health studies using electronic patient-reported outcomes (ePROs) and wearables bring new challenges, including the need for participants to consistently provide trial data. Objective: This study aims to characterize the engagement, protocol adherence, and data completeness among participants with rheumatoid arthritis enrolled in the Digital Tracking of Arthritis Longitudinally (DIGITAL) study. Methods: Participants were invited to participate in this app-based study, which included a 14-day run-in and an 84-day main study. In the run-in period, data were collected via the ArthritisPower mobile app to increase app familiarity and identify the individuals who were motivated to participate. Successful completers of the run-in period were mailed a wearable smartwatch, and automated and manual prompts were sent to participants, reminding them to complete app input or regularly wear and synchronize devices, respectively, during the main study. Study coordinators monitored participant data and contacted participants via email, SMS text messaging, and phone to resolve adherence issues per a priori rules, in which consecutive spans of missing data triggered participant contact. Adherence to data collection during the main study period was defined as providing requested data for >70% of 84 days (daily ePRO, ≥80% daily smartwatch data) or at least 9 of 12 weeks (weekly ePRO). Results: Of the 470 participants expressing initial interest, 278 (59.1%) completed the run-in period and qualified for the main study. Over the 12-week main study period, 87.4% (243/278) of participants met the definition of adherence to protocol-specified data collection for weekly ePRO, and 57.2% (159/278) did so for daily ePRO. For smartwatch data, 81.7% (227/278) of the participants adhered to the protocol-specified data collection. In total, 52.9% (147/278) of the participants met composite adherence. Conclusions: Compared with other digital health rheumatoid arthritis studies, a short run-in period appears useful for identifying participants likely to engage in a study that collects data via a mobile app and wearables and gives participants time to acclimate to study requirements. Automated or manual prompts (ie, “It’s time to sync your smartwatch”) may be necessary to optimize adherence. Adherence varies by data collection type (eg, ePRO vs smartwatch data). International Registered Report Identifier (IRRID): RR2-10.2196/14665

Purpose: Inpatient mortality is an important variable in epidemiology studies using claims data. In 2016, MarketScan data began obscuring specific hospital discharge status types for patient privacy, including inpatient deaths, by setting the values to missing. We used a machine learning approach to correctly identify hospitalizations that resulted in inpatient death using data prior to 2016.Methods: All hospitalizations from 2011 to 2015 with discharge status of missing, died, or one of the other subsequently obscured values were identified and divided into a training set and two test sets. Predictor variables included age, sex, elapsed time from hospital discharge until last observed claim and until healthcare plan disenrollment, and absence of any discharge diagnoses. Four machine learning methods were used to train statistical models and assess sensitivity and positive predictive value (PPV) for inpatient mortality.Results: Overall 1 307 917 hospitalizations were included. All four machine learning approaches performed well in all datasets. Random forest performed best with 88% PPV and 93% sensitivity for the training set and both test sets. The two factors with the highest relative importance for identifying inpatient mortality were having no observed claims for the patient on days 2–91 following hospital discharge and patient disenrollment from the healthcare plan within 60 days following hospital discharge.Conclusion: We successfully developed machine learning algorithms to identify inpatient mortality. This approach can be applied to obscured data to accurately identify inpatient mortality among hospitalizations with missing discharge status.

Purpose: To assess accuracy of administrative claims prescription fill‐based estimates of glucocorticoid use and dose, and approximate bias from glucocorticoid exposure misclassification.Methods: We identified adults with rheumatoid arthritis with linked Medicare and CorEvitas registry data. An algorithm identifying glucocorticoid use and average dose over 90 days from Medicare prescription fills was compared to physician‐reported measures from a CorEvitas visit during the same period, using weighted kappa to compare doses (none, ≤5 mg, 5–10 mg, >10 mg/day). A deterministic sensitivity analysis examined the effect of exposure misclassification on estimated glucocorticoid‐associated infection risk from a prior study.Results: We identified 621 observations among 494 patients. Prescription fills identified glucocorticoid use in 41.9% of observations versus 31.1% identified by CorEvitas physician‐report. For glucocorticoid use (yes/no), prescription fills had sensitivity 88.1% (95% CI 82.7–92.3), specificity 79.0% (74.8–82.7), PPV 65.4% (59.3–71.2), NPV 93.6% (90.6–95.9), and 81.8% agreement with CorEvitas, with kappa 0.61 (moderate to substantial agreement). There was 89.5% agreement between prescription fills and physician‐reported doses, with weighted kappa 0.56 (moderate agreement). Applying these results to a prior Medicare study evaluating glucocorticoid‐associated infection risk [risk ratio 1.44 (95% CI 1.41–1.48)] led to an externally adjusted risk ratio of 1.74 when accounting for exposure misclassification, representing −17% bias in infection risk estimate.Conclusions: This study supports the use of claims data to estimate glucocorticoid use and dose, but investigators should account for exposure misclassification, which may lead to underestimates of glucocorticoid risks. Our results could be applied to adjust risk estimates in other studies that use prescription fills to estimate glucocorticoid use.

Background: The increasing use of activity trackers in mobile health studies to passively collect physical data has shown promise in lessening participation burden to provide actively contributed patient-reported outcome (PRO) information. Objective: The aim of this study was to develop machine learning models to classify and predict PRO scores using Fitbit data from a cohort of patients with rheumatoid arthritis. Methods: Two different models were built to classify PRO scores: a random forest classifier model that treated each week of observations independently when making weekly predictions of PRO scores, and a hidden Markov model that additionally took correlations between successive weeks into account. Analyses compared model evaluation metrics for (1) a binary task of distinguishing a normal PRO score from a severe PRO score and (2) a multiclass task of classifying a PRO score state for a given week. Results: For both the binary and multiclass tasks, the hidden Markov model significantly (P<.05) outperformed the random forest model for all PRO scores, and the highest area under the curve, Pearson correlation coefficient, and Cohen κ coefficient were 0.750, 0.479, and 0.471, respectively. Conclusions: While further validation of our results and evaluation in a real-world setting remains, this study demonstrates the ability of physical activity tracker data to classify health status over time in patients with rheumatoid arthritis and enables the possibility of scheduling preventive clinical interventions as needed. If patient outcomes can be monitored in real time, there is potential to improve clinical care for patients with other chronic conditions.

Purpose: We assessed the suitability of pooled electronic health record (EHR) data from clinical research networks (CRNs) of the patient‐centered outcomes research network to conduct studies of the association between tumor necrosis factor inhibitors (TNFi) and infections.Methods: EHR data from patients with one of seven autoimmune diseases were obtained from three CRNs and pooled. Person‐level linkage of CRN data and Centers for Medicare and Medicaid Services (CMS) fee‐for‐service claims data was performed where possible. Using filled prescriptions from CMS claims data as the gold standard, we assessed the misclassification of EHR‐based new (incident) user definitions. Among new users of TNFi, we assessed subsequent rates of hospitalized infection in EHR and CMS data.Results: The study included 45 483 new users of TNFi, of whom 1416 were successfully linked to their CMS claims. Overall, 44% of new EHR TNFi prescriptions were not associated with medication claims. Our most specific new user definition had a misclassification rate of 3.5%–16.4% for prevalent use, depending on the medication. Greater than 80% of CRN prescriptions had either zero refills or missing refill data. Compared to using EHR data alone, there was a 2‐ to 8‐fold increase in hospitalized infection rates when CMS claims data were added to the analysis.Conclusions: EHR data substantially misclassified TNFi exposure and underestimated the incidence of hospitalized infections compared to claims data. EHR‐based new user definitions were reasonably accurate. Overall, using CRN data for pharmacoepidemiology studies is challenging, especially for biologics, and would benefit from supplementation by other sources.

BACKGROUND Digital health studies using electronic patient-reported outcomes (ePROs) and wearables bring new challenges, including the need for participants to consistently provide trial data. OBJECTIVE This study aims to characterize the engagement, protocol adherence, and data completeness among participants with rheumatoid arthritis enrolled in the Digital Tracking of Arthritis Longitudinally (DIGITAL) study. METHODS Participants were invited to participate in this app-based study, which included a 14-day run-in and an 84-day main study. In the run-in period, data were collected via the ArthritisPower mobile app to increase app familiarity and identify the individuals who were motivated to participate. Successful completers of the run-in period were mailed a wearable smartwatch, and automated and manual prompts were sent to participants, reminding them to complete app input or regularly wear and synchronize devices, respectively, during the main study. Study coordinators monitored participant data and contacted participants via email, SMS text messaging, and phone to resolve adherence issues per a priori rules, in which consecutive spans of missing data triggered participant contact. Adherence to data collection during the main study period was defined as providing requested data for >70% of 84 days (daily ePRO, ≥80% daily smartwatch data) or at least 9 of 12 weeks (weekly ePRO). RESULTS Of the 470 participants expressing initial interest, 278 (59.1%) completed the run-in period and qualified for the main study. Over the 12-week main study period, 87.4% (243/278) of participants met the definition of adherence to protocol-specified data collection for weekly ePRO, and 57.2% (159/278) did so for daily ePRO. For smartwatch data, 81.7% (227/278) of the participants adhered to the protocol-specified data collection. In total, 52.9% (147/278) of the participants met composite adherence. CONCLUSIONS Compared with other digital health rheumatoid arthritis studies, a short run-in period appears useful for identifying participants likely to engage in a study that collects data via a mobile app and wearables and gives participants time to acclimate to study requirements. Automated or manual prompts (ie, “It’s time to sync your smartwatch”) may be necessary to optimize adherence. Adherence varies by data collection type (eg, ePRO vs smartwatch data). INTERNATIONAL REGISTERED REPORT RR2-10.2196/14665

BACKGROUND The increasing use of activity trackers in mobile health studies to passively collect physical data has shown promise in lessening participation burden to provide actively contributed patient-reported outcome (PRO) information. OBJECTIVE The aim of this study was to develop machine learning models to classify and predict PRO scores using Fitbit data from a cohort of patients with rheumatoid arthritis. METHODS Two different models were built to classify PRO scores: a random forest classifier model that treated each week of observations independently when making weekly predictions of PRO scores, and a hidden Markov model that additionally took correlations between successive weeks into account. Analyses compared model evaluation metrics for (1) a binary task of distinguishing a normal PRO score from a severe PRO score and (2) a multiclass task of classifying a PRO score state for a given week. RESULTS For both the binary and multiclass tasks, the hidden Markov model significantly (P<.05) outperformed the random forest model for all PRO scores, and the highest area under the curve, Pearson correlation coefficient, and Cohen κ coefficient were 0.750, 0.479, and 0.471, respectively. CONCLUSIONS While further validation of our results and evaluation in a real-world setting remains, this study demonstrates the ability of physical activity tracker data to classify health status over time in patients with rheumatoid arthritis and enables the possibility of scheduling preventive clinical interventions as needed. If patient outcomes can be monitored in real time, there is potential to improve clinical care for patients with other chronic conditions.

Objective: Persons with immune‐mediated inflammatory diseases (IMIDs) are at an increased risk of herpes zoster (HZ). In 2018, the Centers for Disease Control and Prevention recommended a highly efficacious vaccine, recombinant zoster vaccine (RZV), for prevention of HZ in immunocompetent patients ≥50 years of age. This study was undertaken to estimate RZV vaccination among adults ages ≥50 years with IMIDs during 2018–2019 and to examine possible vaccine‐related flares following RZV.Methods: We identified a cohort of IMID patients using medical claims data from the IBM MarketScan (ages 50–64 years) and Centers for Medicare and Medicaid Services Medicare (ages ≥65 years) databases. Presumed flares were defined as hospitalization/emergency department visit for their respective IMIDs, or steroid treatment with a short‐acting oral glucocorticoid or parenteral glucocorticoid injection. We conducted a self‐controlled case series (SCCS) analysis to examine a temporal association between RZV and flares.Results: Among enrollees with IMIDs, 14.8% of 55,654 MarketScan enrollees and 43.2% of 160,545 Medicare enrollees received ≥1 dose of RZV in 2018–2019. Two‐dose series completion rates were 76.6% in MarketScan enrollees and 85.4% in Medicare enrollees. In the SCCS analysis, 10% and 13% developed flares in the control window, compared to 9% and 11–12% in the risk window following 1 or 2 doses of RZV among MarketScan and Medicare enrollees, respectively. We found no statistically significant increase in flares following RZV administration for any IMID in either age group following RZV dose 1 or dose 2.Conclusion: We did not find an increase in presumed flares following RZV vaccination. Among adults ages ≥50 years with IMIDs, a substantial proportion received RZV compared to general zoster coverage estimates, and series completion rates were high.