Time trends in liver cancer mortality rates in Montenegro from 1990 to 2018.

OBJECTIVE: Liver cancer is the third most common cause of cancer-related deaths worldwide. Hepatitis B and C infections are the main factors affecting mortality. During recent years, Montenegro conducted activities on eradication of viral hepatitis according to the global strategy for the primary prevention of liver cancer mortality. The objective of this study was to assess the liver cancer mortality trend in Montenegro for the period of 1990-2018 using regression techniques. METHODS: liver cancer mortality data in Montenegro from 1990 to 2018 were collected. Mortality rates were age standardized to the World Standard Population. The joinpoint, linear and Poisson regressions were used to assess liver cancer mortality trends both overall and gender specific. RESULTS: The mortality trend was constant, with no significant increase or decrease in mortality rates both at the overall level and by gender. The number of cases, however, increases significantly at the overall level by an average of 1.4% per year [average annual percentage change (AAPC) (95% confidence interval, CI): 1.4 (0.5-2.3); P  = 0.004] and in women by 1.9% per year [AAPC (95% CI): 1.9 (0.8-3.1); P  = 0.002]. In men, there was no change in the number of cases. The three age groups most burdened by mortality from liver cancer were 65-74 (34.9%), 75-84 (26.6%) and 55-64 (25.8%). CONCLUSION: The consistent implementation of prevention measures and hepatitis virus infection treatment has played a role in partially favorable liver cancer mortality trends in Montenegro. It is crucial to closely monitor guidelines for this cancer and give particular attention to the elderly population as the most affected.

Data-driven automated classification algorithms for acute health conditions: applying PheNorm to COVID-19 disease.

OBJECTIVES: Automated phenotyping algorithms can reduce development time and operator dependence compared to manually developed algorithms. One such approach, PheNorm, has performed well for identifying chronic health conditions, but its performance for acute conditions is largely unknown. Herein, we implement and evaluate PheNorm applied to symptomatic COVID-19 disease to investigate its potential feasibility for rapid phenotyping of acute health conditions. MATERIALS AND METHODS: PheNorm is a general-purpose automated approach to creating computable phenotype algorithms based on natural language processing, machine learning, and (low cost) silver-standard training labels. We applied PheNorm to cohorts of potential COVID-19 patients from 2 institutions and used gold-standard manual chart review data to investigate the impact on performance of alternative feature engineering options and implementing externally trained models without local retraining. RESULTS: Models at each institution achieved AUC, sensitivity, and positive predictive value of 0.853, 0.879, 0.851 and 0.804, 0.976, and 0.885, respectively, at quantiles of model-predicted risk that maximize F1. We report performance metrics for all combinations of silver labels, feature engineering options, and models trained internally versus externally. DISCUSSION: Phenotyping algorithms developed using PheNorm performed well at both institutions. Performance varied with different silver-standard labels and feature engineering options. Models developed locally at one site also worked well when implemented externally at the other site. CONCLUSION: PheNorm models successfully identified an acute health condition, symptomatic COVID-19. The simplicity of the PheNorm approach allows it to be applied at multiple study sites with substantially reduced overhead compared to traditional approaches.

Scalable Incident Detection via Natural Language Processing and Probabilistic Language Models.

Post marketing safety surveillance depends in part on the ability to detect concerning clinical events at scale. Spontaneous reporting might be an effective component of safety surveillance, but it requires awareness and understanding among healthcare professionals to achieve its potential. Reliance on readily available structured data such as diagnostic codes risk under-coding and imprecision. Clinical textual data might bridge these gaps, and natural language processing (NLP) has been shown to aid in scalable phenotyping across healthcare records in multiple clinical domains. In this study, we developed and validated a novel incident phenotyping approach using unstructured clinical textual data agnostic to Electronic Health Record (EHR) and note type. It's based on a published, validated approach (PheRe) used to ascertain social determinants of health and suicidality across entire healthcare records. To demonstrate generalizability, we validated this approach on two separate phenotypes that share common challenges with respect to accurate ascertainment: 1) suicide attempt; 2) sleep-related behaviors. With samples of 89,428 records and 35,863 records for suicide attempt and sleep-related behaviors, respectively, we conducted silver standard (diagnostic coding) and gold standard (manual chart review) validation. We showed Area Under the Precision-Recall Curve of ∼ 0.77 (95% CI 0.75-0.78) for suicide attempt and AUPR ∼ 0.31 (95% CI 0.28-0.34) for sleep-related behaviors. We also evaluated performance by coded race and demonstrated differences in performance by race were dissimilar across phenotypes and require algorithmovigilance and debiasing prior to implementation.

Use of Electronic Health Record Data for Drug Safety Signal Identification: A Scoping Review.

INTRODUCTION: Pharmacovigilance programs protect patient health and safety by identifying adverse event signals through postmarketing surveillance of claims data and spontaneous reports. Electronic health records (EHRs) provide new opportunities to address limitations of traditional approaches and promote discovery-oriented pharmacovigilance. METHODS: To evaluate the current state of EHR-based medication safety signal identification, we conducted a scoping literature review of studies aimed at identifying safety signals from routinely collected patient-level EHR data. We extracted information on study design, EHR data elements utilized, analytic methods employed, drugs and outcomes evaluated, and key statistical and data analysis choices. RESULTS: We identified 81 eligible studies. Disproportionality methods were the predominant analytic approach, followed by data mining and regression. Variability in study design makes direct comparisons difficult. Studies varied widely in terms of data, confounding adjustment, and statistical considerations. CONCLUSION: Despite broad interest in utilizing EHRs for safety signal identification, current efforts fail to leverage the full breadth and depth of available data or to rigorously control for confounding. The development of best practices and application of common data models would promote the expansion of EHR-based pharmacovigilance.

Novel methods for pregnancy drug safety surveillance in the FDA Sentinel System.

PURPOSE: It is a priority of the US Food and Drug Administration (FDA) to monitor the safety of medications used during pregnancy. Pregnancy exposure registries and cohort studies utilizing electronic health record data are primary sources of information but are limited by small sample sizes and limited outcome assessment. TreeScan™, a statistical data mining tool, can be applied within the FDA Sentinel System to simultaneously identify multiple potential adverse neonatal and infant outcomes after maternal medication exposure. METHODS: We implemented TreeScan using the Sentinel analytic tools in a cohort of linked live birth deliveries and infants nested in the IBM MarketScan® Research Database. As a case study, we compared first trimester fluoroquinolone use and cephalosporin use. We used the Bernoulli and Poisson TreeScan statistics with compatible propensity score-based study designs for confounding control (matching and stratification) and used multiple propensity score models with various strategies for confounding control to inform best practices. We developed a hierarchical outcome tree including major congenital malformations and outcomes of gestational length and birth weight. RESULTS: A total of 1791 fluoroquinolone-exposed and 8739 cephalosporin-exposed mother-infant pairs were eligible for analysis. Both TreeScan analysis methods resulted in single alerts that were deemed to be due to uncontrolled confounding or otherwise not warranting follow-up. CONCLUSIONS: In this implementation of TreeScan using Sentinel analytic tools, we did not observe any new safety signals for fluoroquinolone use in the first trimester. TreeScan, with tailored or high-dimensional propensity scores for confounding control, is a valuable tool in addition to current safety surveillance methods for medications used during pregnancy.

Improving Methods of Identifying Anaphylaxis for Medical Product Safety Surveillance Using Natural Language Processing and Machine Learning.

We sought to determine whether machine learning and natural language processing (NLP) applied to electronic medical records could improve performance of automated health-care claims-based algorithms to identify anaphylaxis events using data on 516 patients with outpatient, emergency department, or inpatient anaphylaxis diagnosis codes during 2015-2019 in 2 integrated health-care institutions in the Northwest United States. We used one site's manually reviewed gold-standard outcomes data for model development and the other's for external validation based on cross-validated area under the receiver operating characteristic curve (AUC), positive predictive value (PPV), and sensitivity. In the development site 154 (64%) of 239 potential events met adjudication criteria for anaphylaxis compared with 180 (65%) of 277 in the validation site. Logistic regression models using only structured claims data achieved a cross-validated AUC of 0.58 (95% CI: 0.54, 0.63). Machine learning improved cross-validated AUC to 0.62 (0.58, 0.66); incorporating NLP-derived covariates further increased cross-validated AUCs to 0.70 (0.66, 0.75) in development and 0.67 (0.63, 0.71) in external validation data. A classification threshold with cross-validated PPV of 79% and cross-validated sensitivity of 66% in development data had cross-validated PPV of 78% and cross-validated sensitivity of 56% in external data. Machine learning and NLP-derived data improved identification of validated anaphylaxis events.

Monitoring Drug Safety in Pregnancy with Scan Statistics: A Comparison of Two Study Designs.

BACKGROUND: Traditional surveillance of adverse infant outcomes following maternal medication exposures relies on pregnancy exposure registries, which are often underpowered. We characterize the statistical power of TreeScan, a data mining tool, to identify potential signals in the setting of perinatal medication exposures and infant outcomes. METHODS: We used empirical data to inform background incidence of major congenital malformations and other birth conditions. Statistical power was calculated using two probability models compatible with TreeScan, Bernoulli and Poisson, while varying the sample size, magnitude of the risk increase, and incidence of a specified outcome. We also simulated larger referent to exposure matching ratios when using the Bernoulli model in the setting of fixed N:1 propensity score matching. Finally, we assessed the impact of outcome misclassification on power. RESULTS: The Poisson model demonstrated greater power to detect signals than the Bernoulli model across all scenarios and suggested a sample size of 4,000 exposed pregnancies is needed to detect a twofold increase in risk of a common outcome (approximately 8 per 1,000) with 85% power. Increasing the fixed matching ratio with the Bernoulli model did not reliably increase power. An outcome definition with high sensitivity is expected to have somewhat greater power to detect signals than an outcome definition with high positive predictive value. CONCLUSIONS: Use of the Poisson model with an outcome definition that prioritizes sensitivity may be optimal for signal detection. TreeScan is a viable method for surveillance of adverse infant outcomes following maternal medication use.

Utility of fertility procedures and prenatal tests to estimate gestational age for live-births and stillbirths in electronic health plan databases.

PURPOSE: Current algorithms to evaluate gestational age (GA) during pregnancy rely on hospital coding at delivery and are not applicable to non-live births. We developed an algorithm using fertility procedures and fertility tests, without relying on delivery coding, to develop a novel GA algorithm in live-births and stillbirths. METHODS: Three pregnancy cohorts were identified from 16 health-plans in the Sentinel System: 1) hospital admissions for live-birth, 2) hospital admissions for stillbirth, and 3) medical chart-confirmed stillbirths. Fertility procedures and prenatal tests, recommended within specific GA windows were evaluated for inclusion in our GA algorithm. Our GA algorithm was developed against a validated delivery-based GA algorithm in live-births, implemented within a sample of chart-confirmed stillbirths, and compared to national estimates of GA at stillbirth. RESULTS: Our algorithm, including fertility procedures and 11 prenatal tests, assigned a GA at delivery to 97.9% of live-births and 92.6% of stillbirths. For live-births (n = 4 701 207), it estimated GA within 2 weeks of a reference delivery-based GA algorithm in 82.5% of pregnancies, with a mean difference of 3.7 days. In chart-confirmed stillbirths (n = 49), it estimated GA within 2 weeks of the clinically recorded GA at delivery for 80% of pregnancies, with a mean difference of 11.1 days. Implementation of the algorithm in a cohort of stillbirths (n = 40 484) had an increased percentage of deliveries after 36 weeks compared to national estimates. CONCLUSIONS: In a population of primarily commercially-insured pregnant women, fertility procedures and prenatal tests can estimate GA with sufficient sensitivity and accuracy for utility in pregnancy studies.

Prescription medication use and baseline health status of women with live-birth deliveries in a national data network.

BACKGROUND: The US Food and Drug Administration increasingly uses administrative databases to conduct surveillance of medications used during pregnancy. To assess adverse fetal effects, administrative records must be linked between the mother and infant. The Sentinel Initiative's Mother-Infant Linkage Table provides a large cohort of linked deliveries from national, regional, and public insurance claims in the United States for the US Food and Drug Administration to conduct surveillance. OBJECTIVE: This study aimed to describe baseline health conditions and prescription medication use during pregnancy in cohorts of women with a live-birth delivery linked and not linked to an infant. STUDY DESIGN: Live-birth deliveries in women aged 10 to 54 years with at least 391 days of medical and drug coverage before delivery were identified in the Sentinel Mother-Infant Linkage Table from 2000 to 2019. Two cohorts were created for analysis: deliveries linked to infant records (linked deliveries, n=2,320,805) and deliveries unable to be linked to an infant (not-linked deliveries, n=504,785). Baseline health conditions, pregnancy history, healthcare utilization, and pregnancy conditions were defined using International Classification of Diseases, Ninth Revision, and International Classification of Diseases, Tenth Revision, diagnosis and procedure codes. Medication exposure was identified in a 90-day prepregnancy period and in each trimester. RESULTS: Few notable differences were observed between the linked and not-linked deliveries except for maternal age and preterm birth; the not-linked cohort was 3.4 years younger on average and more likely to have a preterm delivery. At baseline among the linked deliveries, the most common conditions were depression and anxiety (5.2% each), acquired hypothyroidism (5.0%), pain conditions (3.9%), and asthma (2.8%). Among linked deliveries, 6.9% had evidence of gestational diabetes mellitus, 3.9% had gestational hypertension, and 4.5% had preeclampsia or eclampsia. The most commonly used medications during pregnancy in the linked deliveries were antibacterials (41.6%) and antiemetics (21.5%); similar medication use patterns were observed in the not-linked cohort. Age trends were observed for some medication groups; anti-infectives, pain medications, and antiemetics were more common in younger mothers, whereas endocrine medications were more common in older mothers. CONCLUSION: Similarities between the linked and not-linked cohorts suggested that the ability to link mother and infant records is not influenced by maternal health status. In the linked cohort, the prevalence of specific pregnancy complications and medication use were similar to previously reported national estimates. Some baseline comorbidities, such as obesity and smoking, may be underestimated in the Sentinel Mother-Infant Linkage.

Validation of an ICD-10-based algorithm to identify stillbirth in the Sentinel System.

PURPOSE: To develop and validate an International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM)-based algorithm to identify cases of stillbirth using electronic healthcare data. METHODS: We conducted a retrospective study using claims data from three Data Partners (healthcare systems and insurers) in the Sentinel Distributed Database. Algorithms were developed using ICD-10-CM diagnosis codes to identify potential stillbirths among females aged 12-55 years between July 2016 and June 2018. A random sample of medical charts (N = 169) was identified for chart abstraction and adjudication. Two physician adjudicators reviewed potential cases to determine whether a stillbirth event was definite/probable, the date of the event, and the gestational age at delivery. Positive predictive values (PPVs) were calculated for the algorithms. Among confirmed cases, agreement between the claims data and medical charts was determined for the outcome date and gestational age at stillbirth. RESULTS: Of the 110 potential cases identified, adjudicators determined that 54 were stillbirth events. Criteria for the algorithm with the highest PPV (82.5%; 95% CI, 70.9%-91.0%) included the presence of a diagnosis code indicating gestational age ≥20 weeks and occurrence of either >1 stillbirth-related code or no other pregnancy outcome code (i.e., livebirth, spontaneous abortion, induced abortion) recorded on the index date. We found ≥90% agreement within 7 days between the claims data and medical charts for both the outcome date and gestational age at stillbirth. CONCLUSIONS: Our results suggest that electronic healthcare data may be useful for signal detection of medical product exposures potentially associated with stillbirth.

A General Propensity Score for Signal Identification Using Tree-Based Scan Statistics.

The tree-based scan statistic (TreeScan; Martin Kulldorff, Harvard Medical School, Boston, Massachusetts) is a data-mining method that adjusts for multiple testing of correlated hypotheses when screening thousands of potential adverse events for signal identification. Simulation has demonstrated the promise of TreeScan with a propensity score (PS)-matched cohort design. However, it is unclear which variables to include in a PS for applied signal identification studies to simultaneously adjust for confounding across potential outcomes. We selected 4 pairs of medications with well-understood safety profiles. For each pair, we evaluated 5 candidate PSs with different combinations of 1) predefined general covariates (comorbidity, frailty, utilization), 2) empirically selected (data-driven) covariates, and 3) covariates tailored to the drug pair. For each pair, statistical alerting patterns were similar with alternative PSs (≤11 alerts in 7,996 outcomes scanned). Inclusion of covariates tailored to exposure did not appreciably affect screening results. Inclusion of empirically selected covariates can provide better proxy coverage for confounders but can also decrease statistical power. Unlike tailored covariates, empirical and predefined general covariates can be applied "out of the box" for signal identification. The choice of PS depends on the level of concern about residual confounding versus loss of power. Potential signals should be followed by pharmacoepidemiologic assessment where confounding control is tailored to the specific outcome(s) under investigation.

Complementary Use of U.S. FDA's Adverse Event Reporting System and Sentinel System to Characterize Direct Oral Anticoagulants-Associated Cutaneous Small Vessel Vasculitis.

BACKGROUND: Cutaneous small vessel vasculitis (CSVV) has been reported after exposure to direct oral anticoagulants (DOACs), such as dabigatran, rivaroxaban, apixaban, and edoxaban. OBJECTIVE: We used the U.S. Food and Drug Administration Adverse Event Reporting System (FAERS) to describe clinical characteristics associated with CSVV among DOAC-exposed patients. Furthermore, we characterized this signal in the Sentinel System to relate the clinical data from the individual FAERS cases to population-based electronic healthcare data. METHODS: We queried FAERS for all cases of CSVV associated with DOACs from U.S. approval date of each DOAC through March 16, 2018. Within the Sentinel System, we identified incident CSVV cases using ICD-9 and ICD-10 diagnosis codes among adults aged ≥ 30 years who received a DOAC in the prior 90 days between January 1, 2010, and June 30, 2018. We excluded patients with evidence of select autoimmune diagnoses in the 183 days prior to their CSVV diagnoses and reported patient characteristics in the 183-day period prior to CSVV diagnoses. RESULTS: In FAERS, we identified 50 cases of CSVV reported with rivaroxaban (n=26), apixaban (n=14), dabigatran (n=9), and edoxaban (n=1). Approximately 50% of the cases reported time to onset within 10 days after DOAC exposure. When specified, the predominant type of CSVV reported was leukocytoclastic vasculitis (n=31), followed by Henoch-Schonlein purpura (n=4). Hospitalization occurred in most of the cases (n=37). Switching of the offending agent after the development of CSVV was reported (n=26). Three rivaroxaban (n=3) cases and one dabigatran case (n=1) reported positive rechallenge. In the Sentinel system, we identified 3659 CSVV cases with prior DOAC exposure, with 85% of events occurring within 10 days. CONCLUSIONS: The assessment of FAERS cases, combined with the temporal clustering of the Sentinel System cases suggest a possible causal relationship of DOACs and CSVV. Future efforts should characterize the risk of CSVV among the various DOAC users.

Soluble Inflammatory Markers and Risk of Incident Fractures in Older Adults: The Cardiovascular Health Study.

Several in vitro and animal studies have showed that inflammatory markers play a role in bone remodeling and pathogenesis of osteoporosis. Additionally, some human longitudinal studies showed suggestive associations between elevated inflammatory markers and increased risk of nontraumatic fractures. We examined several inflammatory markers and multiple fracture types in a single study of older individuals with extensive follow-up. We assessed the association of four inflammatory markers with the risk of incident hip fractures among 5265 participants of the Cardiovascular Health Study (CHS) and a composite endpoint of incident fractures of the hip, pelvis, humerus, or proximal forearm in 4477 participants. Among CHS participants followed between 1992 and 2009, we observed 480 incident hip fractures during a median follow-up of 11 years. In the composite fracture analysis cohort of 4477 participants, we observed 711 fractures during a median follow-up of 7 years. Adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for hip fracture associated with doubling of IL-6 were HR 1.15 (95% CI, 1.02 to 1.30) overall and HR 1.17 (95% CI, 1.01 to 1.35) in women. We also observed a positive association between each unit increase in white blood cell (WBC) count and risk of hip fracture: HR 1.04 (95% CI, 1.01 to 1.06) overall and HR 1.06 (95% CI, 0.95 to 1.20) in women. We observed no significant associations between any of the four inflammatory markers and a composite fracture endpoint. Our findings suggest that chronic inflammatory and immune processes may be related to higher rates of incident hip fractures. © 2017 American Society for Bone and Mineral Research.