Please use this identifier to cite or link to this item: http://localhost/handle/Hannan/732634
Title: Natural Language Processing in Perinatal Epidemiology
Authors: Williams, Michelle;Avillach, Paul;Cai, Tianxi;Karlson, Elizabeth;Smoller, Jordan;Zhong, Qiuyue
subject: Health Sciences, Epidemiology;Health Sciences, Mental Health;Health Sciences, Obstetrics and Gynecology
Year: 2018
Description: Suicide is one of the leading causes of maternal deaths. Early detection of pregnant women with suicidal behavior presents an important opportunity for directing suicide prevention efforts to those at high risk for suicide and, therefore, can help to prevent maternal mortality. The increasing utilization of electronic medical records (EMRs) has provided unprecedented opportunities for identifying pregnant women with suicidal behavior. However, suicidal behavior is often under-coded in EMRs. Studies have illustrated the clinical relevance of applying natural language processing (NLP) to identify patients with suicidal behavior; and this inspired our effort to develop algorithms specific to pregnant women, a population that is understudied when it comes to understanding the determinants and sequelae of suicidal behavior. We first examined the comparative performance of structured data using diagnostic codes vs. NLP of unstructured text for screening suicidal behavior among pregnant women in EMRs. We found that the use of NLP substantially improves the sensitivity of screening for suicidal behavior. However, the chart review confirmation rate for suicidal behavior was lower among women who did not have diagnostic codes for suicidal behavior but screened positive by NLP, as compared with women screened positive by diagnostic codes. We then developed classification algorithms to identify pregnant women with suicidal behavior using information extracted from clinical notes by NLP in EMRs. Using expert-curated features, our algorithm achieved an area under the receiver operating characteristic curve of 0.83. By setting a PPV comparable to the PPV of diagnostic codes related to suicidal behavior (0.71), we obtained an estimated sensitivity of 0.34, a specificity of 0.96, and a NPV of 0.83. The algorithm resulted in a 11-fold increase in the number of pregnant women identified with suicidal behavior. Lastly, using the suicidal behavior algorithm, we examined the associations of antepartum suicidal behavior and psychiatric disorders, alone and in combination, with adverse outcomes in EMRs. We found that independent of psychiatric disorders, antepartum suicidal behavior was associated with increased odds of several adverse obstetric outcomes, but not adverse infant outcomes. The combination of suicidal behavior and psychiatric disorders showed higher odds of adverse infant outcomes and some adverse obstetric outcomes than either suicidal behavior or psychiatric disorders alone.  
text
natural language processing; electronic medical records; suicidal behavior; pregnancy;
URI: http://nrs.harvard.edu/urn-3:HUL.InstRepos:37945648
http://localhost/handle/Hannan/15778
http://localhost/handle/Hannan/732634
Appears in Collections:SPH Theses and Dissertations

Files in This Item:
Click on the URI links for accessing contents.
Title: Natural Language Processing in Perinatal Epidemiology
Authors: Williams, Michelle;Avillach, Paul;Cai, Tianxi;Karlson, Elizabeth;Smoller, Jordan;Zhong, Qiuyue
subject: Health Sciences, Epidemiology;Health Sciences, Mental Health;Health Sciences, Obstetrics and Gynecology
Year: 2018
Description: Suicide is one of the leading causes of maternal deaths. Early detection of pregnant women with suicidal behavior presents an important opportunity for directing suicide prevention efforts to those at high risk for suicide and, therefore, can help to prevent maternal mortality. The increasing utilization of electronic medical records (EMRs) has provided unprecedented opportunities for identifying pregnant women with suicidal behavior. However, suicidal behavior is often under-coded in EMRs. Studies have illustrated the clinical relevance of applying natural language processing (NLP) to identify patients with suicidal behavior; and this inspired our effort to develop algorithms specific to pregnant women, a population that is understudied when it comes to understanding the determinants and sequelae of suicidal behavior. We first examined the comparative performance of structured data using diagnostic codes vs. NLP of unstructured text for screening suicidal behavior among pregnant women in EMRs. We found that the use of NLP substantially improves the sensitivity of screening for suicidal behavior. However, the chart review confirmation rate for suicidal behavior was lower among women who did not have diagnostic codes for suicidal behavior but screened positive by NLP, as compared with women screened positive by diagnostic codes. We then developed classification algorithms to identify pregnant women with suicidal behavior using information extracted from clinical notes by NLP in EMRs. Using expert-curated features, our algorithm achieved an area under the receiver operating characteristic curve of 0.83. By setting a PPV comparable to the PPV of diagnostic codes related to suicidal behavior (0.71), we obtained an estimated sensitivity of 0.34, a specificity of 0.96, and a NPV of 0.83. The algorithm resulted in a 11-fold increase in the number of pregnant women identified with suicidal behavior. Lastly, using the suicidal behavior algorithm, we examined the associations of antepartum suicidal behavior and psychiatric disorders, alone and in combination, with adverse outcomes in EMRs. We found that independent of psychiatric disorders, antepartum suicidal behavior was associated with increased odds of several adverse obstetric outcomes, but not adverse infant outcomes. The combination of suicidal behavior and psychiatric disorders showed higher odds of adverse infant outcomes and some adverse obstetric outcomes than either suicidal behavior or psychiatric disorders alone.  
text
natural language processing; electronic medical records; suicidal behavior; pregnancy;
URI: http://nrs.harvard.edu/urn-3:HUL.InstRepos:37945648
http://localhost/handle/Hannan/15778
http://localhost/handle/Hannan/732634
Appears in Collections:SPH Theses and Dissertations

Files in This Item:
Click on the URI links for accessing contents.
Title: Natural Language Processing in Perinatal Epidemiology
Authors: Williams, Michelle;Avillach, Paul;Cai, Tianxi;Karlson, Elizabeth;Smoller, Jordan;Zhong, Qiuyue
subject: Health Sciences, Epidemiology;Health Sciences, Mental Health;Health Sciences, Obstetrics and Gynecology
Year: 2018
Description: Suicide is one of the leading causes of maternal deaths. Early detection of pregnant women with suicidal behavior presents an important opportunity for directing suicide prevention efforts to those at high risk for suicide and, therefore, can help to prevent maternal mortality. The increasing utilization of electronic medical records (EMRs) has provided unprecedented opportunities for identifying pregnant women with suicidal behavior. However, suicidal behavior is often under-coded in EMRs. Studies have illustrated the clinical relevance of applying natural language processing (NLP) to identify patients with suicidal behavior; and this inspired our effort to develop algorithms specific to pregnant women, a population that is understudied when it comes to understanding the determinants and sequelae of suicidal behavior. We first examined the comparative performance of structured data using diagnostic codes vs. NLP of unstructured text for screening suicidal behavior among pregnant women in EMRs. We found that the use of NLP substantially improves the sensitivity of screening for suicidal behavior. However, the chart review confirmation rate for suicidal behavior was lower among women who did not have diagnostic codes for suicidal behavior but screened positive by NLP, as compared with women screened positive by diagnostic codes. We then developed classification algorithms to identify pregnant women with suicidal behavior using information extracted from clinical notes by NLP in EMRs. Using expert-curated features, our algorithm achieved an area under the receiver operating characteristic curve of 0.83. By setting a PPV comparable to the PPV of diagnostic codes related to suicidal behavior (0.71), we obtained an estimated sensitivity of 0.34, a specificity of 0.96, and a NPV of 0.83. The algorithm resulted in a 11-fold increase in the number of pregnant women identified with suicidal behavior. Lastly, using the suicidal behavior algorithm, we examined the associations of antepartum suicidal behavior and psychiatric disorders, alone and in combination, with adverse outcomes in EMRs. We found that independent of psychiatric disorders, antepartum suicidal behavior was associated with increased odds of several adverse obstetric outcomes, but not adverse infant outcomes. The combination of suicidal behavior and psychiatric disorders showed higher odds of adverse infant outcomes and some adverse obstetric outcomes than either suicidal behavior or psychiatric disorders alone.  
text
natural language processing; electronic medical records; suicidal behavior; pregnancy;
URI: http://nrs.harvard.edu/urn-3:HUL.InstRepos:37945648
http://localhost/handle/Hannan/15778
http://localhost/handle/Hannan/732634
Appears in Collections:SPH Theses and Dissertations

Files in This Item:
Click on the URI links for accessing contents.