Dataset Information

Automatically determining cause of death from verbal autopsy narratives.

ABSTRACT: BACKGROUND:A verbal autopsy (VA) is a post-hoc written interview report of the symptoms preceding a person's death in cases where no official cause of death (CoD) was determined by a physician. Current leading automated VA coding methods primarily use structured data from VAs to assign a CoD category. We present a method to automatically determine CoD categories from VA free-text narratives alone. METHODS:After preprocessing and spelling correction, our method extracts word frequency counts from the narratives and uses them as input to four different machine learning classifiers: naïve Bayes, random forest, support vector machines, and a neural network. RESULTS:For individual CoD classification, our best classifier achieves a sensitivity of.770 for adult deaths for 15 CoD categories (as compared to the current best reported sensitivity of.57), and.662 with 48 WHO categories. When predicting the CoD distribution at the population level, our best classifier achieves.962 cause-specific mortality fraction accuracy for 15 categories and.908 for 48 categories, which is on par with leading CoD distribution estimation methods. CONCLUSIONS:Our narrative-based machine learning classifier performs as well as classifiers based on structured data at the individual level. Moreover, our method demonstrates that VA narratives provide important information that can be used by a machine learning system for automated CoD classification. Unlike the structured questionnaire-based methods, this method can be applied to any verbal autopsy dataset, regardless of the collection process or country of origin.

SUBMITTER: Jeblee S

PROVIDER: S-EPMC6617656 | biostudies-literature | 2019 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Automatically determining cause of death from verbal autopsy narratives.

Jeblee Serena S Gomes Mireille M Jha Prabhat P Rudzicz Frank F Hirst Graeme G

BMC medical informatics and decision making 20190709 1

<h4>Background</h4>A verbal autopsy (VA) is a post-hoc written interview report of the symptoms preceding a person's death in cases where no official cause of death (CoD) was determined by a physician. Current leading automated VA coding methods primarily use structured data from VAs to assign a CoD category. We present a method to automatically determine CoD categories from VA free-text narratives alone.<h4>Methods</h4>After preprocessing and spelling correction, our method extracts word freque ...[more]

PMID: 31288814

Similar Datasets

Project description:BackgroundThe South African national cause of death validation (NCODV 2017/18) project collected a national sample of verbal autopsies (VA) with cause of death (COD) assignment by physician-coded VA (PCVA) and computer-coded VA (CCVA).ObjectiveThe performance of three CCVA algorithms (InterVA-5, InSilicoVA and Tariff 2.0) in assigning a COD was compared with PCVA (reference standard).MethodsSeven performance metrics assessed individual and population level agreement of COD assignment by age, sex and place of death subgroups. Positive predictive value (PPV), sensitivity, overall agreement, kappa, and chance corrected concordance (CCC) assessed individual level agreement. Cause-specific mortality fraction (CSMF) accuracy and Spearman's rank correlation assessed population level agreement.ResultsA total of 5386 VA records were analysed. PCVA and CCVAs all identified HIV/AIDS as the leading COD. CCVA PPV and sensitivity, based on confidence intervals, were comparable except for HIV/AIDS, TB, maternal, diabetes mellitus, other cancers, and some injuries. CCVAs performed well for identifying perinatal deaths, road traffic accidents, suicide and homicide but poorly for pneumonia, other infectious diseases and renal failure. Overall agreement between CCVAs and PCVA for the top single cause (48.2-51.6) indicated comparable weak agreement between methods. Overall agreement, for the top three causes showed moderate agreement for InterVA (70.9) and InSilicoVA (73.8). Agreement based on kappa (-0.05-0.49)and CCC (0.06-0.43) was weak to none for all algorithms and groups. CCVAs had moderate to strong agreement for CSMF accuracy, with InterVA-5 highest for neonates (0.90), Tariff 2.0 highest for adults (0.89) and males (0.84), and InSilicoVA highest for females (0.88), elders (0.83) and out-of-facility deaths (0.85). Rank correlation indicated moderate agreement for adults (0.75-0.79).ConclusionsWhilst CCVAs identified HIV/AIDS as the leading COD, consistent with PCVA, there is scope for improving the algorithms for use in South Africa.

Project description:Mortality statistics by age, sex and cause are the foundation of basic health data required for health status assessment, epidemiological research and formation of health policy. Close to half the deaths in Malaysia occur outside a health facility, are not attended by medical personnel, and are given a lay opinion as to the cause of death, leading to poor quality of data from vital registration. Verbal autopsy (VA) is a very useful tool in diagnosing broad causes of deaths for events that occur outside health facilities. This article reports the development of the VA methods and our principal finding from a validation study.A cross sectional study on nationally representative sample deaths that occurred in Malaysia during 2013 was used. A VA questionnaire suitable for local use was developed. Trained field interviewers visited the family members of the deceased at their homes and conducted face to face interviews with the next of kin. Completed questionnaires were reviewed by trained physicians who assigned multiple and underlying causes. Reference diagnoses for validation were obtained from review of medical records (MR) available for a sample of the overall study deaths.Corresponding MR diagnosis with matched sample of the VA diagnosis were available in 2172 cases for the validation study. Sensitivity scores were good (>75%) for transport accidents and certain cancers. Moderate sensitivity (50% - 75%) was obtained for ischaemic heart disease (64%) and cerebrovascular disease (72%). The validation sample for deaths due to major causes such as ischaemic heart disease, pneumonia, breast cancer and transport accidents show low cause-specific mortality fraction (CSMF) changes. The scores obtained for the top 10 leading site-specific cancers ranged from average to good.We can conclude that VA is suitable for implementation for deaths outside the health facilities in Malaysia. This would reduce ill-defined mortality causes in vital registration data, and yield more accurate national mortality statistics.

Project description:BackgroundGood quality cause of death (COD) information is fundamental for formulating and evaluating public health policy; yet most deaths in developing countries, including the Solomon Islands, occur at home without medical certification of cause of death (MCCOD). As a result, COD data in such contexts are often of limited use for policy and planning. Verbal autopsies (VAs) are a cost-effective way of generating reliable COD information in populations lacking comprehensive MCCOD coverage, but this method has not previously been applied in the Solomon Islands. This study describes the establishment of a VA system to estimate the cause specific mortality fractions (CSMFs) for community deaths that are not medically certified in the Solomon Islands.MethodsAutomated VA methods (SmartVA) were introduced into the Solomon Islands in 2016. Trained data collectors (nurses) conducted VAs on eligible deaths to December 2020 using electronic tablet devices and VA responses were analysed using the Tariff 2.0 automated diagnostic algorithm. CSMFs were generated for both non-inpatient deaths in hospitals (i.e. 'dead on/by arrival') and community deaths.ResultsVA was applied to 914 adolescent-and-adult deaths with a median (IQR) age of 62 (45-75) years, 61% of whom were males. A specific COD could be diagnosed for more than 85% of deaths. The leading causes of death for both sexes combined were: ischemic heart disease (16.3%), stroke (13.5%), diabetes (8.1%), pneumonia (5.7%) and chronic-respiratory disease (4.8%). Stroke was the top-ranked cause for females, and ischaemic heart disease the leading cause for males. The CSMFs from the VAs were similar to Global Burden of Disease (GBD) estimates. Overall, non-communicable diseases (NCDs) accounted for 73% of adult deaths; communicable, maternal and nutritional conditions 15%, and injuries 12%. Six of the ten leading causes reported for facility deaths in the Solomon Islands were also identified as leading causes of community deaths based on the VA diagnoses.ConclusionsNCDs are the leading cause of adult deaths in the Solomon Islands. Automated VA methods are an effective means of generating reliable COD information for community deaths in the Solomon Islands and should be routinely incorporated into the national mortality surveillance system.

Dataset Information

Automatically determining cause of death from verbal autopsy narratives.

Publications

Automatically determining cause of death from verbal autopsy narratives.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets