Public and Population Health Informatics: The Bridging of Big Data to Benefit Communities

Gamache, Roland E.; Kharrazi, Hadi; Weiner, Jonathan P.

doi:10.1055/s-0038-1667081

Cited by 98 publications

(72 citation statements)

References 62 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This effect on total suicide rates may be even more pronounced in states with higher rates of overdose as a cause of death, such as Maryland, where current overdose rates far surpass those of Utah and the majority of such deaths are categorized as undetermined intent. Our findings of underreporting rate are similar to previous historical research estimating underreporting rates between 10% and 30% (Stone, Holland, Bartholow, Crosby et al, 2017;Stone, Holland, Bartholow, Logan et al, 2017), and the 30% underreporting rate documented from a previous pilot study conducted in Utah using a classification and regression tree technique (Donaldson et al, 2006). Our results also match the approximately 30% of poisoning suicides that were estimated to be missed when Rockett, Caine, Connery et al (2018) used the prevalence of suicide notes across methods to estimate underreporting of suicides in the NVDRS states (including Utah) 2011-2013, although national-level results might be biased due to heterogeneous data reporting and variations in medical investigation culture and practice across the United States.…”

Section: Conclusion/discussionsupporting

confidence: 90%

“…A full listing of variables with descriptions can be found in Table S3. We used common missing data approaches to retain all the data (Gelman & Hill, 2007). Out of 2,665 total cases, 406 missing values for alcohol level were imputed with its mean, and 46 missing values for education level were imputed with their mode.…”

Section: Data Preprocessingmentioning

confidence: 99%

“…We excluded indicators of other violent deaths such as weapon variables, suspect variables, intimate partner violence modules, or child fatality review modules, which are indicators of other types of violent deaths like homicides. The 111 remaining variables extracted from Utah's NVDRS provided us with enough candidates to make the final selection of variables used for the analysis, as they covered all variables used in prior research to achieve acceptable model accuracy (Donaldson et al, 2006). In addition to domain expert knowledge, we used the Random Forest feature importance analysis (James, Witten, Hastie, & Tibshirani, 2017) with 500 trees on the training set to rank the 111 variables.…”

Section: Feature Selection and Importance Analysismentioning

confidence: 99%

“…The current advancements in machine learning technologies provide a new opportunity to remedy 334 MACHINE LEARNING TO RECLASSIFY OVERDOSE INTENTION classification problems, when utilized as a supplement to recommended changes in death scene investigation practices (Stone, Holland, Bartholow, Crosby et al, 2017;Stone, Holland, Bartholow, Logan et al, 2017). Machine learning has already begun to be used to identify living cases at risk for suicide (Colic, Richardson, Reilly, & Hasey, 2018;Kessler et al, 2017;McKernan, Lenert, Crofford, & Walsh, 2018;Ryu, Lee, Lee, & Park, 2018), paving the way for the natural extension of its use for the retrospective prediction of missed suicides among the deceased (Esty et al, 2017). Utah has one of the highest rates of undetermined deaths in the United States, at double the national rate.…”

Section: Introductionmentioning

confidence: 99%

“…The majority of undetermined deaths in Utah are the result of poisoning or drug overdose, a high potential for misclassified suicide deaths (Centers for Disease Control and Prevention, & National Center for Health Statistics, 2018;Donaldson, Larsen, Fullerton-Gleason, & Olson, 2006). An earlier pilot study conducted in Utah classified poisoning deaths of undetermined intent as either suicide or unintentional using classification and regression tree analysis and found that suicide rates related to poisoning were estimated to be underreported by approximately 30% (Donaldson et al, 2006). This study, however, used old data with small sample sizes in each of the drug overdose death categories.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Discovering the Unclassified Suicide Cases Among Undetermined Drug Overdose Deaths Using Machine Learning Techniques

Liu,

Yu,

Duncan

et al. 2019

Suicide & Life Threat Behav

Self Cite

View full text Add to dashboard Cite

Objective The Centers for Disease Control and Prevention (CDC) monitor accidental and intentional deaths to answer questions that are critical for the development of effective prevention and resource allocation. CDC's National Violent Death Reporting System (NVDRS) is a major innovation in surveillance linking individual‐level data from multiple sources. However, suicide underreporting is common, particularly from drug overdose deaths. This study sought to assess machine learning (ML) techniques in quantifying drug overdose suicide underreporting rates. Methods Clinical, sociodemographic, toxicological, and proximal stressor data on overdose decedents (n = 2,665) were extracted from Utah's NVDRS from 2012 to 2015. The existing well‐determined cases were used to train and test our ML models. We assessed and compared multiple machine learning methods including Logistic Regression, Random Forest Classifier, Support Vector Machines, and Artificial Neural Networks. We applied a majority voting methodology to classify undetermined drug overdose deaths. Results Overdose suicide rates were estimated to be underreported by 33% across all years, increasing yearly from 29% in 2012 to 37% in 2015. The overall test accuracies for all models ranged from 92.3% to 94.6%. Conclusions This research identifies a cost‐effective, replicable, and expandable ML‐based methodology to estimate the true rates of suicide which may be partially masked during the opioid epidemic.

show abstract