Increasing availability of electronic health databases capturing real-world experiences with medical products has garnered much interest in their use for pharmacoepidemiologic and pharmacovigilance studies. The traditional practice of having numerous groups use single databases to accomplish similar tasks and address common questions about medical products can be made more efficient through well-coordinated multi-database studies, greatly facilitated through distributed data network (DDN) architectures. Access to larger amounts of electronic health data within DDNs has created a growing interest in using data-adaptive machine learning (ML) techniques that can automatically model complex associations in high-dimensional data with minimal human guidance. However, the siloed storage and diverse nature of the databases in DDNs create unique challenges for using ML. In this paper, we discuss opportunities, challenges, and considerations for applying ML in DDNs for pharmacoepidemiologic and pharmacovigilance studies. We first discuss major types of activities performed by DDNs and how ML may be used. Next, we discuss practical data-related factors influencing how DDNs work in practice. We then combine these discussions and jointly consider how opportunities for ML are affected by practical data-related factors for DDNs, leading to several challenges. We present different approaches for addressing these challenges and highlight efforts that real-world DDNs have taken or are currently taking to help mitigate them. Despite these challenges, the time is ripe for the emerging interest to use ML in DDNs, and the utility of these data-adaptive modeling techniques in pharmacoepidemiologic and pharmacovigilance studies will likely continue to increase in the coming years.
OBJECTIVE: While infliximab combined to thiopurines is more effective than infliximab monotherapy in patients with Crohn's disease (CD) and UC, the impact of adding thiopurines to vedolizumab remains controversial. We emulated two target trials comparing the effectiveness of combination therapy versus vedolizumab monotherapy in CD and UC.
DESIGN: Based on two US and the French nationwide healthcare databases, patients with CD and UC who initiated vedolizumab were identified. The study methodology, including confounding adjustment and outcome definitions, were previously validated in successful emulations of the SONIC and SUCCESS trials. Risk ratios for treatment failure based on hospitalisation or surgery related to disease activity, treatment switch, or prolonged corticosteroids use, were estimated after 1:1 propensity score (PS) matching.
RESULTS: Among a total of 10 299 vedolizumab users, 804 CD and 1088 UC pairs of combination therapy versus vedolizumab monotherapy users were PS matched. Treatment failure occurred at week 26 in 236 (29.3%) and 376 (34.3%) patients with CD and at week 16 in 236 (21.7%) and 263 (24.2%) patients with UC initiating combination therapy and vedolizumab monotherapy, respectively. The risk of treatment failure was decreased with combination therapy compared with vedolizumab monotherapy in CD (RR 0.85, 95% CI: 0.74 to 0.98) and to a lesser extent in UC (RR 0.90, 95% CI: 0.77 to 1.05). Findings were consistent across databases.
CONCLUSION: Using validated methodologies, combination therapy with vedolizumab and thiopurines was associated with lower treatment failure compared with vedolizumab monotherapy in CD but not UC across the USA and France.
OBJECTIVES: To characterise the incidence rate of skin cancer associated with methotrexate and hydroxychloroquine in older adults with rheumatoid arthritis (RA).
METHODS: RA patients aged ≥65 years who initiated methotrexate or hydroxychloroquine as their first disease modifying antirheumatic drugs (DMARDs). The primary outcome was new occurrence of any skin cancer (i.e. malignant melanoma or non-melanoma skin cancer; NMSC) based on validated algorithms (positive predictive value >83%). Secondary outcomes were malignant melanoma, NMSC, basal cell carcinoma (BCC), and squamous cell carcinoma (SCC). We estimated the incidence rates (IRs) and hazard ratios (HRs) for each outcome in the 1:1 propensity score (PS)-matched methotrexate and hydroxychloroquine groups.
RESULTS: We included 24,577 PS-matched pairs of methotrexate and hydroxychloroquine initiators. Compared with hydroxychloroquine (IR 25.20/1,000 person-years), methotrexate initiators (IR 26.21/1,000 person-years) had a similar risk of any skin cancer [HR 1.03 -(95%CI 0.92, 1.14)] over a mean follow-up of 388 days. The HR (95%CI) associated with methotrexate was 1.39 (0.87, 2.21) for malignant melanoma, 1.01(0.90, 1.12) for NMSC, 1.37 (1.13, 1.66) for BCC, and 0.79 (0.63, 0.99) for SCC compared with hydroxychloroquine.
CONCLUSIONS: In this large cohort of older RA patients initiating methotrexate or hydroxychloroquine as their first DMARD, we found no difference in the risk of skin cancer including malignant melanoma and NMSC. However, for specific components of NMSC, methotrexate initiators had higher risk of BCC but lower risk of SCC compared with hydroxychloroquine initiators.
Little is known about the impact of dose, duration, and timing of prenatal prescription opioid exposure on the risk of neonatal opioid withdrawal syndrome (NOWS). Using a cohort of 18,869 prepregnancy chronic opioid users nested within the 2000-2014 Medicaid Analytic eXtract, we assessed average opioid dosage within biweekly gestational age intervals, created group-based trajectory models, and evaluated the association between trajectory groups and NOWS risk. Women were grouped into 6 distinct opioid use trajectories which, based on observed patterns, were categorized as 1) continuous very low-dose use, 2) continuous low-dose use, 3) initial moderate-dose use with a gradual decrease to very low-dose/no use, 4) initial high-dose use with a gradual decrease to very low-dose use, 5) continuous moderate-dose use, and 6) continuous high-dose use. Absolute risk of NOWS per 1,000 infants was 7.7 for group 1 (reference group), 28.8 for group 2 (relative risk (RR) = 3.7, 95% confidence interval (CI): 2.8, 5.0), 16.5 for group 3 (RR = 2.1, 95% CI: 1.5, 3.1), 64.9 for group 4 (RR = 8.4, 95% CI: 5.6, 12.6), 77.3 for group 5 (RR = 10.0, 95% CI: 7.5, 13.5), and 172.4 for group 6 (RR = 22.4, 95% CI: 16.1, 31.2). Trajectory models-which capture information on dose, duration, and timing of exposure-are useful for gaining insight into clinically relevant groupings to evaluate the risk of prenatal opioid exposure.
OBJECTIVES: Recent results from 'ORAL Surveillance' trial have raised concerns regarding the cardiovascular safety of tofacitinib in patients with rheumatoid arthritis (RA). We further examined this safety concern in the real-world setting.
METHODS: We created two cohorts of patients with RA initiating treatment with tofacitinib or tumour necrosis factor inhibitors (TNFI) using deidentified data from Optum Clinformatics (2012-2020), IBM MarketScan (2012-2018) and Medicare (parts A, B and D, 2012-2017) claims databases: (1) A 'real-world evidence (RWE) cohort' consisting of routine care patients and (2) A 'randomised controlled trial (RCT)-duplicate cohort' mimicking inclusion and exclusion criteria of the ORAL surveillance trial to calibrate results against the trial findings. Cox proportional hazards models with propensity score fine stratification weighting were used to estimate HR and 95% CIs for composite outcome of myocardial infarction and stroke and accounting for 76 potential confounders. Database-specific effect estimates were pooled using fixed effects models with inverse-variance weighting.
RESULTS: In the RWE cohort, 102 263 patients were identified of whom 12 852 (12.6%) initiated tofacitinib. The pooled weighted HR (95% CI) comparing tofacitinib with TNFI was 1.01 (0.83 to 1.23) in RWE cohort and 1.24 (0.90 to 1.69) in RCT-duplicate cohort which aligned closely with ORAL-surveillance results (HR: 1.33, 95% CI 0.91 to 1.94).
CONCLUSIONS: We did not find evidence for an increased risk of cardiovascular outcomes with tofacitinib in patients with RA treated in the real-world setting; however, tofacitinib was associated with an increased risk of cardiovascular outcomes, although statistically non-significant, in patients with RA with cardiovascular risk factors.
TRIAL REGISTRATION NUMBER: NCT04772248.
Real-world evidence (RWE) on the effectiveness of treatments in Crohn's disease (CD) derived from clinical practice data will help fill many evidence gaps left by randomized controlled trials (RCTs). Emulating RCTs with healthcare database studies may calibrate RWE studies in CD. We aimed to emulate the SONIC trial on the effectiveness of infliximab in patients with CD using US and French healthcare claims data. SONIC had shown improved remission with combination therapy (i.e., infliximab plus thiopurines) compared with infliximab monotherapy. Using claims data (2004-2019) from commercially insured patients in the United States (IBM MarketScan and Optum) and France (Système National des Données de Santé (National Healthcare Data System) (SNDS)), we conducted a cohort study of patients with CD who initiated combination therapy and compared them with patients who initiated infliximab alone. The primary outcome was a composite end point of treatment failure including hospitalization or surgery related to CD, treatment switch, or continuation of corticosteroids 26 weeks after infliximab initiation. Risk ratios (RRs) with 95% confidence intervals (CIs) were estimated in propensity score (PS)-matched cohorts. We identified 1,437 PS-matched pairs of combination therapy vs. infliximab monotherapy users. As in SONIC, the risk of treatment failure was decreased with combination therapy in the overall cohort (RR, 0.71; 95% CI, 0.62-0.82; RR, 0.78; 95% CI, 0.62-0.97 in SONIC). Findings were consistent across MarketScan, Optum, and SNDS databases: RR (95% CI), 0.83 (0.63-1.10), 0.66 (0.46-0.93), and 0.68 (0.57-0.82), as well as component end points. These robust findings highlight opportunities in RWE analysis for studying treatment effectiveness in patients with CD in clinical practice.
BACKGROUND: New bone-directed therapies, including denosumab, abaloparatide, and romosozumab, emerged during the past decade, and recent trends in use of these therapies are unknown.
OBJECTIVE: To examine temporal trends in bone-directed therapies.
DESIGN: An open cohort study in a US commercial insurance database, January 2009 to March 2020.
PARTICIPANTS/INTERVENTIONS: All-users of bone-directed therapies age >50 years, users with osteoporosis, users with malignancies, and patients with recent (within 180 days) fractures at key osteoporotic sites.
MAIN MEASURES: The percentage of each cohort with prescription dispensing or medication administration claims for each bone-directed therapy during each quarter of the study period.
KEY RESULTS: We analyzed 15.48 million prescription dispensings or medication administration claims from 1.46 million unique individuals (89% women, mean age 69 years). Among all users of bone-directed therapies, alendronate, and zoledronic acid use increased modestly (49 to 63% and 2 to 4%, respectively, during the study period). In contrast, denosumab use increased rapidly after approval in 2010, overtaking use of all other medications except alendronate by 2017 and reaching 16% of users by March 2020. Similar trends were seen in cohorts of osteoporosis, malignancy, and recent fractures. Importantly, use of any bone-directed therapy after fractures was low and declined from 15 to 8%.
CONCLUSIONS: Rates of denosumab use outpaced growth of all other bone-directed therapies over the past decade. Treatment rates after osteoporotic fractures were low and declined over time, highlighting major failings in osteoporosis treatment in the US.
BACKGROUND & AIMS: The risk of serious infections associated with vedolizumab in patients with inflammatory bowel disease (IBD) is uncertain. We assessed the risk of serious infections associated with use of vedolizumab versus anti-TNF in patients with IBD, according to IBD subtype and previous exposure to anti-TNF.
METHODS: Based on two U.S. nationwide commercial insurance databases and the French nationwide health insurance database, anti-TNF naïve and experienced patients diagnosed with Crohn's disease (CD) and ulcerative colitis (UC) aged 18 years or older who initiated vedolizumab or an anti-TNF agent after 2010 were identified. Hazard ratios for serious infections comparing vedolizumab and anti-TNF were estimated in propensity score matched cohorts.
RESULTS: Among 8768 vedolizumab and 26,656 anti-TNF initiators included after 1:4 variable ratio propensity score matching, 893 serious infections occurred during 37,725 person-years of follow-up. The risk of serious infections was not different between vedolizumab and anti-TNF in the overall IBD cohort (HR, 0.95; 95% CI, 0·79-1.13), while the risk was decreased for vedolizumab users in patients with UC (HR, 0.68; 95% CI, 0.50-0.93), but not CD (HR, 1.10; 95% CI, 0.87-1.38). In patients with UC, vedolizumab was consistently associated with lower risk of serious infections after exclusion of gastrointestinal infections (HR, 0.59; 95% CI, 0.39-0.90).
CONCLUSIONS: While the risk of serious infections associated with vedolizumab was not different compared to anti-TNF in the overall group of patients with IBD, the risk varied according to IBD subtype, by decreasing in patients with UC, but not CD. These findings may help to clarify the optimal position of vedolizumab in the therapeutic management of IBD.