| AA | Aortic aneurysm |
| AD | Aortic dissection |
| CDM | Common Data Model |
| OMOP | Observational Medical Outcomes Partnership |
| OHDSI | Observational Health Data Science and Informatics |
| RxNorm | US-specific terminology that contains all medications available on the US market |
| SNOMED | Systematized Nomenclature of Medicine |
| UTI | Urinary Tract Infection |
| Investigator | Institution/Affiliation |
|---|---|
| Jack Janetzki* | University of South Australia |
| Nicole Pratt* | University of South Australia |
| Seng Chan You* | Yonsei University |
| Seonji Kim | Yonsei University |
| Jung Ho Kim* | Yonsei University |
| Jung Ah Lee | Yonsei University |
| Patrick Ryan* | Columbia University, Janssen |
| * Principal Investigator |
This study is undertaken within Observational Health Data Sciences and Informatics (OHDSI), an open collaboration.
Question Are fluoroquinolones linked to increased risk of aortic aneurysm or dissection?
This distributed network analysis will describe the incidence and time-to event of aortic aneurysm and aortic dissection following exposure to fluoroquinolone antibiotics. The study will also compare the risk of AA/AD between individual fluoroquinolone antibiotics with other individual antibiotics used to treat the same infection (head-to-head comparisons). We will also compare the risk of AA/AD between fluoroquinolone antibiotics as a class with other individual antibiotics used to treat the same infection.
Background and Significance:
Fluoroquinolone antibiotics are a broad-spectrum class of antibiotics which are used to treat urinary tract infections, pneumonia, gastroenteritis, epididmyo-orchitis, prostatitis or bone and joint infections. Fluoroquinolone use is rising internationally; approximately 7.81 billion doses of fluoroquinolones have been administered over the last decade. Whilst fluoroquinolone antibiotics are generally well tolerated, neurological and cardiovascular adverse events have been identified in post-market surveillance studies. Additionally in rare cases (0.1-1%), use of fluoroquinolones has been associated with risk of aortic aneurysm or dissection. The pharmacological mechanism between fluoroquinolone use and aortic aneurysm or dissection occurrence is not fully understood. However, the pathogenesis of this adverse event appears to involve extracellular matrix degradation in the aorta mediated by matrix metalloproteinases, dysfunction and apoptosis of vascular smooth muscle cells in the aorta and also increased concentrations of pro-inflammatory cytokines at the site contributing to remodelling of the aorta.
Subsequent to post-market surveillance, international medicine regulators have issued safety warnings regarding the risk of aortic aneurysm or dissection (AA/AD) with fluoroquinolone antibiotics (FQ) antibiotics. Whilst there have been numerous warnings from regulators about the association of fluoroquinolone antibiotics with risk of aortic aneurysms or dissections, the quality of the evidence underpinning this association is moderate. The most recent meta-analysis of four observational studies found that people who used fluoroquinolone antibiotics were at an increased risk of aortic diseases compared to those who used other antibiotics (adjusted odds ratio 2.10; 95% CI 1.65-2.68)(1) and other studies have shown that risk of aortic dissection appears highest within 30 days of taking a fluoroquinolone and the risk remains elevated for a year(2). Prior studies supporting the risk are limited by inconsistencies in study design, approaches to identifying exposures and outcomes, and methods used to address confounding.
*Study Aims/Objectives:
This study is a multinational cohort study which will estimate the risk of AA/AD following FQ exposure compared to other comparable antibiotics for urinary tract infection (UTI).
More specifically we will:
A. Characterizs the incidence and time-to event of aortic events following fluoroquinolone exposure
B. Estimate the comparative safety of fluoroquinolones versus other antibiotics indicated for urinary tract infections
*Study Description:
A large-scale distributed network study will be undertaken via the OHDSI network. The selection of data sources will be based on their comprehensive population coverage, extensive historical data, diverse data capture processes, voluntary participation, and the availability of verified outcome information, ensuring robust and reliable research.
*Population:
Participants will be patients 35 years or older initiating antibiotic treatment for UTI
People taking systemic fluoroquinolones for treatment of urinary tract infections (UTI)
Cox proportional hazards models will be used to estimate the risk of each outcome after 1:1 propensity score (PS) matching. All results will be calibrated based on the results from negative control outcomes. A series of pre-defined diagnostics will be employed to evaluate the analytical method performance. Only results of analyses passing all diagnostics will be reported. Hazard ratios (HRs) will be pooled across databases using Bayesian random effects meta-analysis.
The study will be a retrospective cohort study.
| Number | Date | Section of study protocol | Amendment or update | Reason |
|---|---|---|---|---|
| The SOS challenge tutorial schedule is described below: |
|---|
| • Mar 28th 2023: Initiate network study, identifying data partners |
| • Apr 4th 2023: Data diagnostics (data partners to share data diagnostics results) |
| • Apr 11th 2023: Phenotype development |
| • Apr 18th 2023: Phenotype evaluation (complete candidate phenotypes and data partners to share phenotype diagnostic results) |
| • Apr 25th 2023: Create analysis specifications (finalize analytic package) |
| • May 2nd 2023: Network execution (Data partners to run analytic package) |
| • May 9th 2023: Study diagnostics (data partners to share study diagnostic results) |
| • May 14 2023: Evidence synthesis |
| • May 23 2023: Interpretation of results by investigators |
| • May 2023 to June 204: Manuscript drafting and writing |
Fluoroquinolone antibiotics (FQ) are broad-spectrum antibiotics that treat a variety of infections, including urinary tract infections (UTIs). Although FQs are generally well tolerated, a recent meta-analysis based on five observational post-market studies concluded that FQ use doubled the risk of incident aortic diseases. While aortic aneurysm (AA) and aortic dissections (AD) are rare, they are fatal in 65-90% of cases, especially when an AA ruptures.
In response to safety concerns raised by observational studies and adverse events reported to regulatory bodies internationally, warnings were issued regarding the risk of AA/AD with FQ antibiotics. Studies ahve shown that the use of outpatient FQs has decreased following these safety warnings in some but not all settings, with the United Kingdom Medicines and Healthcare products Regulatory Agency recently issuing further restrictions on prescribing. Patients have also shared concerns about the risk profile of FQs in public hearings.
Recent studies have suggested that the previous studies reporting associations between FQs and some adverse drug reactions (ADRs) may have been affected by indication bias or surveillance bias. A meta-analysis of the four studies listed in the U.S. Federal Drug Administration (FDA) Drug Safety Communication regarding the association of FQs with AA/AD noted that confounding by indication may have influenced results and only one study included an active comparator.
Given the conflicting results of previous studies, the recognised efficacy of FQs and the serious nature of AA/AD complications, we will conduct a large-scale distributed network study to estimate the risk of AA/AD after FQ exposure for treatment of UTI compared to other antibiotics. We will employ best-practice methodology to minimize potential systematic bias and use objective diagnostics to evaluate the analytical method performance.
A new-user cohort study will be conducted to estimate the comparative risk of AA/AD with new use of FQs compared to comparator antibiotics.
We will conduct a distributed network analysis across the Observational Health Data Science and Informatics (OHDSI) Data Network. This study will utilise routinely-collected health care data which has been mapped to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM).
The study will be conducted using data from real world data sources that have been mapped to the OMOP Common Data Model in collaboration with the Observational Health Data Sciences and Informatics (OHDSI) and European Health Data and Evidence Network (EHDEN) initiatives. The OMOP Common Data Model (https://ohdsi.github.io/CommonDataModel/) includes a standard representation of health care trajectories (such as information related to drug utilization and condition occurrence), as well as common vocabularies for coding clinical concepts, and enables consistent application of analyses across multiple disparate data sources (Voss et al., 2015).
We intend to study data with different data source provenance (e.g., electronic medical records, insurance claims) as well data representing different populations (privately insured employees or patients with limited income) and geographies. This study will be run on datasets that have been converted to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM)
We intend to study data with different data source provenance (e.g., electronic medical records, insurance claims) as well data representing different populations (privately insured employees or patients with limited income) and geographies. This study will be run on datasets that have been converted to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM)
All participating sites and data partners will obtain approval or exemption from their institutional review boards prior to participating in the study.
The study will utilize data from 1 January 2010 to 31 December 2019. The study period will be restricted to end of 2019 to avoid the potential impact of COVID-19 pandemic on patterns in antibiotic use.
Study population: all subjects in the database will be included who meet the following criteria (note the index date is the start of the first exposure to fluoroquinolone or active comparator):
Patients will be included if at index date they are aged 35 years or older, have at least 365 days of prior observation in the database, have a recorded condition occurrence of UTI on or within 7 days, and were not hospitalized for any reason within 7 days prior to the index date. Patients could have no diagnosis of aortic aneurysm or dissection preceding the index date.
The study population will be defined as:
All people included in this analysis must be diagnosed with urinary tract infection. For all antibiotics, we restrict to people with a diagnosis of Urinary Tract Infection, a major indication for the treatment of interest. To minimize confounding by indication we will restrict analysis to patients treated in an outpatient setting for UTIs, which are commonly localised, well-characterized infections with a simliar severity profile across presentations.
Three exposure cohorts will be generated: (1) FQ exposure, (2) TMP exposure, or (3) CPH exposure.
Only systemic exposure from medications administered orally or via injection will be included in the exposure definitions. TMP and CPH were selected as comparator antibiotics as they are generally recommended in treatment guidelines for UTI.
| Concept ID | Concept name |
| 1721543 | norfloxacin |
| 35198003 | pazufloxacin mesilate |
| 1747032 | grepafloxacin |
| 19041153 | temafloxacin |
| 1592954 | delafloxacin |
| 1716721 | gemifloxacin |
| 923081 | ofloxacin |
| 19050750 | fleroxacin |
| 1712549 | trovafloxacin |
| 35197938 | garenoxacin mesilate hydrate |
| 36878831 | nadifloxacin |
| 40161662 | besifloxacin |
| 43009030 | tosufloxacin tosylate hydrate |
| 1743222 | enoxacin |
| 35834909 | lascufloxacin hydrochloride |
| 35198165 | sitafloxacin hydrate |
| 1789276 | gatifloxacin |
| 1797513 | ciprofloxacin |
| 1742253 | levofloxacin |
| 1716903 | moxifloxacin |
| 35197897 | prulifloxacin |
| 19027679 | pefloxacin |
| 1733765 | sparfloxacin |
| 1707800 | lomefloxacin |
Concept ID Concept Name 40081374 sulfamethoxazole / trimethoprim Injectable Solution 36882762 Sulfamethoxazole / Trimethoprim Injectable Suspension 40220482 sulfamethoxazole / trimethoprim Injection 35153537 Sulfamethoxazole / Trimethoprim Oral Granules 40147374 sulfamethoxazole / trimethoprim Oral Solution 40081379 sulfamethoxazole / trimethoprim Oral Suspension 40081388 sulfamethoxazole / trimethoprim Oral Tablet 1705674 trimethoprim
Concept ID Concept Name 40798709 Cefacetrile 1796435 cefixime 40798704 Cefmenoxime 19072255 cefmetazole 43008993 cefminox sodium 19028286 cefodizime 19072857 cefonicid 1773402 cefoperazone 19028288 ceforanide 1774470 cefotaxime 1774932 cefotetan 19051271 cefotiam 1775741 cefoxitin 43009045 cefpiramide sodium 1749008 cefpodoxime 1738366 cefprozil 43009083 cefroxadine 19051345 cefsulodin 1776684 ceftazidime 35198137 cefteram pivoxil 43008994 ceftezole sodium 1749083 ceftibuten 1777254 ceftizoxime 1777806 ceftriaxone 1778162 cefuroxime 1786621 cephalexin 19052683 cephaloridine 19086759 cephalothin 19086790 cephapirin 1786842 cephradine 43009087 flomoxef sodium 1708100 loracarbef 19126622 moxalactam 1768849 cefaclor 1769535 cefadroxil 19070174 cefamandole 19070680 cefatrizine 40798700 Cefazedone 1771162 cefazolin 43009082 cefbuperazone sodium 43009044 cefcapene pivoxil hydrochloride hydrate 1796458 cefdinir 1747005 cefditoren 19028241 cefetamet
The primary outcome will be diagnosis of either AA or AD during a hospital or emergency department visit in the 60 days after index date. The secondary endpoints included the occurrence of each outcome separately. Patients will be excluded if they have a recorded outcome event in the 365 days prior to index date. A time at risk window of 60 days after treatment initiation will be employed as the primary analysis as this is consistent with previous studies. We will also vary the time at risk window as: 30, 60, 90 and 365 days as sensitivity analyses.
Details of the outcome definitions are provided below:
ICD-9-CM: 441;441.0; 441.1; 441.2; 441.3; 441.3; 441.4; 441.5; 441.6; 441.7; 441.9
ICD-10: I71.0; I71.1; I71.2; I71.3; I71.4; I71.5; I71.6; I71.8; I71.9
CPT-4: 001T; 0002T’ 0033T; 0034T; 0035T; 0036T; 0039T; 0078T; 0079T; 0080T; 0081T; 33720; 33877; 33880; 33881; 34800; 34802; 34803; 34804; 34805; 34813; 34830; 34831; 34832; 35081; 35082; 35091; 35092; 35102; 35103; 75952; 75953; 9001F; 9003F; 9004f
ICD-9 PROCEDURE: 39.71; 39.73
Positive Predictive Value was calculated for this outcome based on hospitalization or emergency room (ER) visits at any diagnosis position based on the ICD-10 code in the single Korean tertiary hospital, Severance Hospital (Yonsei University Heatlh System). PPV, % (n) = 97 (97/100)
ICD-9-CM: 441.0; 441.00; 441.01; 441.02; 441.03
ICD-10: I71.0
PPV= 79% (N= 79/100)
Negative control outcomes are known to have no association with the target (fluoroquinolone) or comparator (trimethoprim or cephalosporin) cohorts, such that we can assume the true relative risk between the two cohorts is 1. Negative control outcomes for this study will be selected based on a similar process to that outlined by Voss et al. Once potential negative control candidates are selected, a manual clinical review will be performed to exclude any pairs that may have a causal relationshipu or are similar to the study outcome.
The top 50 outcomes based on their prevalence will be selected.
After doing, so the final list of 50 negative outcomes to be used in the study are:
OMOP Concept ID Outcome Name
4170145 Absence of lung
4093531 Absence of toe
4092879 Absent kidney
42539582 Acquired absence of genital organ
434170 Atypical squamous cells of undetermined significance on cervical Papanicolaou smear
4067069 Callosity
4213540 Cervical somatic dysfunction
443570 Cervicovaginal cytology: Low grade squamous intraepithelial lesion
201613 Chronic nonalcoholic liver disease
43021250 Complication associated with orthopedic device
46269889 Complication due to Crohn’s disease
42537730 Coronary artery graft present
436233 Delayed milestone
438021 Disorder due to and following fracture of upper limb
192367 Dysplasia of cervix
4062791 Endocrine, nutritional and metabolic disease complicating pregnancy, childbirth and puerperium
200775 Endometrial hyperplasia
374358 Excess skin of eyelid
4059015 Falls
4264617 Foot-drop
4201388 Gastrostomy present
4166231 Genetic predisposition
4295287 Hypercoagulability state
196473 Hypertrophy of uterus
443447 Iatrogenic hypotension
4344500 Impingement syndrome of shoulder region
441417 Incoordination
4168222 Intra-abdominal and pelvic swelling, mass and lump
196168 Irregular periods
439795 Minimal cognitive impairment
40480893 Nonspecific tuberculin test reaction
438130 Opioid abuse
4022076 Patient dependence on care provider
4141640 Perimenopausal disorder
437092 Physiological development failure
4012231 Poor stream of urine
46286594 Problem related to lifestyle
443274 Psychostimulant dependence
436246 Reduced libido
43021237 Secondary erectile dysfunction
4052226 Sequelae of injuries of lower limb
4125590 Slurred speech
36713918 Somatic dysfunction of lumbar region
4002818 Spasm of back muscles
4008710 Stenosis due to any device, implant AND/OR graft
4201387 Tracheostomy present
42538119 Transplanted heart valve present
444074 Victim of vehicular AND/OR traffic accident
195603 Vulval and/or perineal noninflammatory disorders
4216670 Worried well
Characterizing the cohort:
All analyses will be performed using code developed for the OHDSI Methods library. The code for this study can be found at A diagnostic package, built off the OHDSI Cohort Diagnostics library, is included in the base package as a preliminary step to assess the fitness of use of phenotypes on your database. If a database passes cohort diagnostics, the full study package will be executed. Baseline covariates will be extracted using an optimized SQL extraction script based on principles of the Feature Extraction package () to quantify Demographics (Gender, Prior Observation Time, Age Group), Condition Group Eras and Drug Group Eras (at the above-listed time windows). Additional cohort-specific covariates will be constructed using OMOP standard vocabulary concepts.
Number and proportion of persons with feature variables during time-at-risk windows will be reported by target cohort and specific stratifications. Standardized mean differences (SMD) will be calculated when comparing characteristics of study cohorts, with plots comparing the mean values of characteristics for each of the features (with the color indicating the absolute value of the standardized difference of the mean).
Population-level estimation:
Proportional hazard models will be used to assess the hazard ratios between the two exposure cohorts for each condition of interest and for each setting.
Adjustment for baseline confounders will be done using propensity scores. First a propensity model will be fitted and used to create propensity scores (PS). These PS will be used ot match the treatment and comparator cohorts, and the proportional hazards outcome models will be conditioned on the matched sets of strata respectively.
The analysis will be performed as an intention to treat where exposure is considered from index date until end of the observation (until the occurrence of outcome, death or end of study).
Negative controls: as described above. The hazard ratios computed for these negative controls will be used to evaluate residual bias and compute calibrated p-values for the outcomes of interest.
Cox proportional hazards will be used to estimate the hazard ratio of each outcome for each time at risk window in each data source after propensity score (PS) matching. Propensity score models derivation is described below in ‘Covariate balance’.
A Bayesian random-effects meta-analysis will be conducted to combine each sites hazard ratio estimate into a single aggregated hazard ratio using non-normal likelihood approximations to avoid bias due to small or zero counts.
Analyses using observational data can produce misleading estimates as a result of study design and analytic choices. We will implement several study diagnostics using pre-specified decision thresholds to evaluate the reliability of our analyses. These study diagnostics will assist with understanding the risk of bias and generalizability of the estimates generated in each database across the network of databases involved in the study. Only results that passed the pre-specified thresholds set for each diagnostics will contribute to the meta-analytic estimates.
We will implement propensity score matching to account for confounding between treatment groups. Propensity score models will include covariates such as age, race, medications, medical conditions, procedure exposure, medical exposure, and laboratory values. The number of covariates to be included will be between 10,000 to 100,000. Covariate use behaviours over the 30 to 365 days prior to index date will be utliised for PS model calculation.
Large-scale regularized regression will be used to fit the propensity model. Separate PS models will be constructed for each comparator. Patients will be matched 1:1 between the target and comparator cohorts.
To determine whether the PS matching is sufficient to balance baseline patient characteristics we will calculate the standardised mean difference between treatment cohorts after propensity score matching. Covariate balance diagnostic will be achieved if all SMDs of predefined covariates below are less than 0.1.
Age group: Age was grouped in 5-year intervals
Gender
Race
Ethnicity
Medical history: Acute respiratory disease, Attention deficit hyperactivity disorder, Chronic liver disease, Chronic obstructive lung disease, Crohn’s disease, Dementia, Depressive disorder, Diabetes mellitus, Gastroesophageal reflux disease, Gastrointestinal hemorrhage, Human immunodeficiency virus infection, Hyperlipidemia,Hypertensive disorder, Lesion of liver, Obesity, Osteoarthritis, Pneumonia, Psoriasis, Renal impairment, Rheumatoid arthritis, Schizophrenia, Substance abuse, Ulcerative colitis, Viral hepatitis C, Visual system disorder, Atrial fibrillation, Cerebrovascular disease, Coronary arteriosclerosis, Heart disease, Heart failure, Ischemic heart disease, Peripheral vascular disease, Pulmonary embolism, Venous thrombosis, Hematologic neoplasm, Malignant lymphoma, Malignant neoplasm of anorectum, Malignant neoplastic disease, Malignant tumor of breast, Malignant tumor of colon, Malignant tumor of lung, Malignant tumor of urinary bladder, Primary malignant neoplasm of prostate
Medication use: Agents acting on the renin-angiotensin system, Antibacterials for systemic use, Antidepressants, Antiepileptics, Antiinflammatory and antirheumatic products, Antineoplastic agents, Antipsoriatics, Antithrombotic agents, Beta blocking agents, Calcium channel blockers, Diuretics, Drugs for acid related disorders, Drugs for obstructive airway diseases, Drugs used in diabetes, Immunosuppressants, Lipid modifying agents, Opioids, Psycholeptics, Psychostimulants, Agents used for ADHD and nootropics
Comorbidity index: Charlson Comorbidity Index, CHA2DS2-Vasc score, Diabetes Complications Severity Index
Empirical equipoise will be assessed by determining the overlap in preference score distribution between the target and comparator cohorts. The preference score distribution is a transformation of the propensity score. High overlap, and greater equipoise ensures that results will be generalisable back to the original cohort. Good equipoise means that even a large propensity score model could not discriminate between preference for two treatments. This is similar to randomised clinical trials where study participants have the same probability of receiving either intervention in a 1:1 randomised trial regardless of their characteristics. at the clinical equipoise diagnostic was achieved if at least 20% of matched patients had a preference score between 0.3 and 0.7.
To determine systematic error or residual bias (due to study design and analytic choices), we will estimate systematic error by employing negative controls. Negative controls (as described above) are determined as outcomes which are assumed to not be associated with either the target or comparator treatment. We will include 50 negative controls in our analysis. For each of these outcomes we will implement the analysis as per the primary analysis and we compare the estimated result of the analysis against the “known truth”, in this case we expect the true realative risk to be 1. Positive controls will not be utilised as there are no medicines which are known to increase risk of aortic aneurysm or dissection.
Overall systematic error will be calculated as the Expected Absolute Systematic Error (EASE) score. EASE will be calculated by first fitting the systematic error distribution across the set of negative control outcomes and then taking the aboslute expected value of the distribution.
If EASE score is low, then calibrated and uncalibrated estimates (hazard ratios, p-values, confidence intervals) will be similar indicating little to no systematic error or unmeasured confounding that is not being accounted for in the results of the study. The systematic bias diagnostic will be achieved if the EASE value was less than 0.25.
Study diagnostics: Overall an analysis was considered to have passed study diagnostics if SMDs of predefined covariates were less than 0.1, equipoise was greater than 0.2 and EASE was less than 0.25.
Sample size and study power will be calculated based on the number of patients included from the databases passing diagnostics and patients prescribed exposure drug (FQ) and comparators (CPH or TMP) for treatment of UTI.
This large-scale distributed network study wil’ use a rigorous design with objective diagnostics to reduce bias and confounding. The study design ensures transparency, reproducibility and reliability of results.
To minimize confounding by indication, we will restrict the analysis to patients treated in an outpatient setting for UTIs, which are commonly localized, well-characterized infections with a similar severity profile across presentations.
We will use this detailed protocol with standardized definitions of exposures, outcome measures, study design and approach to address confounding.
We will also use a standardized analytic package to generate the analytic results within each of the participating databases.
Our approach also mirrors that of a hypothetical randomized controlled trial by employing a new-user, active comparator design with clearly defined indications and we will implmement an outcome-agnostic, data-driven PS model, modelled using large-scale regularised regression.
We will also apply the suite of predefined objective diagnostics as described above to esnure the quality of evidence focusing on sufficient covaraiate balance, overlap in preference score distributions, and minimal systematic error as evaluated by a comprehensive set of negative control outcomes.
Only analyses that pass all objective diagnostic tests will be included in the meta-analysis.
The definition of AA/AD will rely on ICD-9-CM and ICD-10-CM codes rather than imaging modalities. We have however validated our specific diagnosis codes in one Korean tertiary hospital with high PPV for the outcomes of investigation.
Results may not be generalizable to other indications as we only examined FQs used to treat UTIs in the outpatient setting.
Antibiotics are usually used for a short period of time to treat UTIs, therefore, we are unable to investigate the effects of long-term cumulative use of FQs in this study.
Regulatory warnings about the use of FQs during the study may affect results, however the majority of the study period is before such warnings were made.
We are unable to consider the dose of FQ or comparator antibiotics in this study. As the critiera for antibiotic dosage and duration are well established in outpatient UTIs, it is likely that variation in dose would be small, however we are unable to determine whether there is a dose-response relationship.
The study uses only de-identified data. Confidentiality of patient records will be maintained at all times. Data custodians will remain in full control of executing the analysis and packaging results. There will be no transmission of patient-level data at any time during these analyses. Only aggregate statistics will be captured. Study packages will contain minimum cell count parameters to obscure any cells which fall below allowable reportable limits. All study reports will only contain aggregated data and will not identify individual patients or physicians.
N/A
Share results and study design at OHDSI conferences.
Share primary results in high-impact journal.
Dai XC, Yang XX, Ma L, Tang GM, Pan YY, Hu HL. Relationship between fluoroquinolones and the risk of aortic diseases: a meta-analysis of observational studies. BMC Cardiovasc Disord. 2020;20(1):49.
Yu X, Jiang DS, Wang J, Wang R, Chen T, Wang K, Cao S, Wei X. Fluoroquinolone Use and the Risk of Collagen-Associated Adverse Events: A Systematic Review and Meta-Analysis. Drug Saf. 2019;42(9):1025-1033.
Voss EA, Boyce RD, Ryan PB, van der Lei J, Rijnbeek PR, Schuemie MJ. Accuracy of an automated knowledge base for identifying drug adverse reactions. J Biomed Inform 2017; 66: 72–81.