Dataset Information

Data Mining for Adverse Drug Events With a Propensity Score-matched Tree-based Scan Statistic.

ABSTRACT: The tree-based scan statistic is a statistical data mining tool that has been used for signal detection with a self-controlled design in vaccine safety studies. This disproportionality statistic adjusts for multiple testing in evaluation of thousands of potential adverse events. However, many drug safety questions are not well suited for self-controlled analysis. We propose a method that combines tree-based scan statistics with propensity score-matched analysis of new initiator cohorts, a robust design for investigations of drug safety. We conducted plasmode simulations to evaluate performance. In multiple realistic scenarios, tree-based scan statistics in cohorts that were propensity score matched to adjust for confounding outperformed tree-based scan statistics in unmatched cohorts. In scenarios where confounding moved point estimates away from the null, adjusted analyses recovered the prespecified type 1 error while unadjusted analyses inflated type 1 error. In scenarios where confounding moved point estimates toward the null, adjusted analyses preserved power, whereas unadjusted analyses greatly reduced power. Although complete adjustment of true confounders had the best performance, matching on a moderately mis-specified propensity score substantially improved type 1 error and power compared with no adjustment. When there was true elevation in risk of an adverse event, there were often co-occurring signals for clinically related concepts. TreeScan with propensity score matching shows promise as a method for screening and prioritization of potential adverse events. It should be followed by clinical review and safety studies specifically designed to quantify the magnitude of effect, with confounding control targeted to the outcome of interest.

SUBMITTER: Wang SV

PROVIDER: S-EPMC6167151 | biostudies-literature | 2018 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Data Mining for Adverse Drug Events With a Propensity Score-matched Tree-based Scan Statistic.

Wang Shirley V SV Maro Judith C JC Baro Elande E Izem Rima R Dashevsky Inna I Rogers James R JR Nguyen Michael M Gagne Joshua J JJ Patorno Elisabetta E Huybrechts Krista F KF Major Jacqueline M JM Zhou Esther E Reidy Megan M Cosgrove Austin A Schneeweiss Sebastian S Kulldorff Martin M

Epidemiology (Cambridge, Mass.) 20181101 6

The tree-based scan statistic is a statistical data mining tool that has been used for signal detection with a self-controlled design in vaccine safety studies. This disproportionality statistic adjusts for multiple testing in evaluation of thousands of potential adverse events. However, many drug safety questions are not well suited for self-controlled analysis. We propose a method that combines tree-based scan statistics with propensity score-matched analysis of new initiator cohorts, a robust ...[more]

PMID: 30074538

Similar Datasets

Project description:ImportanceAlthough contact isolation has been widely recommended for multidrug-resistant organisms, contact isolation has raised some concerns that it may bring unintended patient harms.ObjectiveTo compare adverse events between a contact isolation group with vancomycin-resistant Enterococcus (VRE) and a matched comparison group using a relatively large data set from full electronic medical records (EMR) and propensity score-matching methods.Design, setting, and participantsThis retrospective, matched cohort study was conducted at Seoul National University Bundang Hospital (SNUBH) in Korea, a tertiary, university-affiliated hospital that has 1337 inpatient beds. Participants included a total of 98 529 hospitalized adult patients (aged ≥18 years) during 2015 to 2017.ExposuresContact isolation in a single or shared double room.Main outcomes and measuresAs adverse contact isolation-related outcomes, falls and pressure ulcers were included. All relevant EMR data were extracted from the SNUBH clinical data warehouse. Risk factors for adverse events were included in the propensity score model based on literature reviews, such as Braden scale score and Hendrich II fall risk score. A fine stratification and weighting (FSW) and a 1:10 nearest neighbor (NN) propensity score matching as a sensitivity analysis were adopted to compare adverse events between the 2 groups for the observation period from the study entry date and the exit date. Time-to-event analyses with a Cox proportional hazard model were conducted in December 2021.ResultsFor comparison of outcomes in wards, 177 patients (mean [SD] age, 67.38 [14.12] years; 98 [55.4%] female) with VRE and 93 022 patients (mean [SD] age, 56.44 [16.88] years; 49 462 [53.2%] female) without VRE were included and no difference was found in basic characteristics from the FSW (VRE contact isolation [n = 172] vs comparison [n = 69 434]) as well as from the 1:10 NN (VRE contact isolation [n = 168] vs comparison [n = 1650]). Among 177 patients with VRE contact isolation, 8 pressure ulcers and 3 falls occurred during their hospital stays; incidence rates of adverse events were 2.5 and 0.9 per 1000 patient-days, respectively (pressure ulcer incidence rate from the FSW: 2.53 per 1000 patient-days [95% CI, 1.09-4.99 per 1000 patient-days]; pressure ulcer incidence rate from the 1:10 NN: 2.54 per 1000 patient-days [95% CI, 1.10-5.01 per 1000 patient-days]; fall incidence rate from the FSW: 0.87 per 1000 patient-days [95% CI, 0.18-2.54 per 1000 patient-days]; fall incidence rate from the 1:10 NN: 0.87 per 1000 patient-days [95% CI, 0.18-2.55 per 1000 patient-days]). The hazard ratios for adverse events showed no statistically significant differences for both groups: 1.42 (95% CI, 0.67-2.99) for pressure ulcer and 0.66 (95% CI, 0.20-2.13) for fall from the FSW.Conclusions and relevanceIn this cohort study, no association was found between the likelihood of adverse events and contact isolation using propensity score-matching methods and closely related covariates for adverse events.

Project description:BackgroundTyrosine kinase inhibitors (TKIs) have shown long-term survival benefits in patients with chronic myeloid leukemia (CML). Nevertheless, significant concern has been raised regarding long-term TKI-associated vascular adverse events (VAEs). The objective of this retrospective cohort study was to investigate the incidence of VAEs in Taiwanese patients with CML treated with different TKIs (imatinib, nilotinib, and dasatinib) as well as potential risk factors.Materials and methodsWe conducted a retrospective cohort study using the Taiwan Cancer Registry Database and National Health Insurance Research Database. Adult patients diagnosed with CML from 2008 to 2016 were identified and categorized into three groups according to their first-line TKI treatment (imatinib, nilotinib, and dasatinib). Propensity score matching was performed to control for potential confounders. Cox regressions were used to estimate the hazard ratio (HR) of VAEs in different TKI groups.ResultsIn total, 1,111 patients with CML were included in our study. We found that the risk of VAEs in nilotinib users was significantly higher than that in imatinib users, with an HR of 3.13 (95% confidence interval (CI), 1.30-7.51), whereas dasatinib users also showed a nonsignificant trend for developing VAEs, with an HR of 1.71 (95% CI, 0.71-4.26). In multivariable logistic regression analysis, only nilotinib usage, older age, and history of cerebrovascular diseases were identified as significant risk factors. The annual incidence rate of VAEs was highest within the first year after the initiation of TKIs.ConclusionThese findings can support clinicians in making treatment decisions and monitoring VAEs in patients with CML in Taiwan.Implications for practiceThis study found that patients with chronic myeloid leukemia (CML) treated with nilotinib and dasatinib may be exposed to a higher risk of developing vascular adverse events (VAEs) compared with those treated with imatinib. Thus, this study suggests that patients with CML who are older or have a history of cerebrovascular diseases should be under close monitoring of VAEs, particularly within the first year after the initiation of tyrosine kinase inhibitors.

Dataset Information

Data Mining for Adverse Drug Events With a Propensity Score-matched Tree-based Scan Statistic.

Publications

Data Mining for Adverse Drug Events With a Propensity Score-matched Tree-based Scan Statistic.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets