Automated Coding System Research Articles

The results of implementation of automated system of diagnosis encoding in multi-field hospital

The common coding of main diagnosis (so-called "manual coding") is accompanied by quite many errors associated with inobservance of classification algorithm and subjective causes that in statistical morbidity and mortality data corruption. The article presents the results of implementing automated system of diagnosis coding by the ICD-10 in the departments of the Medicosanitary Unit of MVD of Russia in City of Moscow. The functional and time losses and the error rate were studied under manual coding and automated coding system. In case of common coding of complete clinical diagnosis of one discharged patient per day, it takes from 5 to 15 minutes, for coding diagnoses of all discharged patients - about 1 hour. The number of errors in common coding made up to 14.1% (n=2472) and in case of automated coding it made up to 1.5% (n= 63, p>0.05). The application of automated coding system in medical organizations ensures exact correspondence of clinical and statistical diagnosis, unifies wording of clinical diagnosis, minimizes subjective deviations from algorithm of the ICD-10.

A Deep Learning Framework for Automated ICD-10 Coding.

The International Statistical Classification of Diseases and Related Health Problems (ICD) is one of the widely used classification system for diagnoses and procedures to assign diagnosis codes to Electronic Health Record (EHR) associated with a patient's stay. The aim of this paper is to propose an automated coding system to assist physicians in the assignment of ICD codes to EHR. For this purpose, we created a pipeline of Natural Language Processing (NLP) and Deep Learning (DL) models able to extract the useful information from French medical texts and to perform classification. After the evaluation phase, our approach was able to predict 346 diagnosis codes from heterogeneous medical units with an accuracy average of 83%. Our results were finally validated by physicians of the Medical Information Department (MID) in charge of coding hospital stays.

Automated Coding of Under-Studied Medical Concept Domains: Linking Physical Activity Reports to the International Classification of Functioning, Disability, and Health.

Linking clinical narratives to standardized vocabularies and coding systems is a key component of unlocking the information in medical text for analysis. However, many domains of medical concepts, such as functional outcomes and social determinants of health, lack well-developed terminologies that can support effective coding of medical text. We present a framework for developing natural language processing (NLP) technologies for automated coding of medical information in under-studied domains, and demonstrate its applicability through a case study on physical mobility function. Mobility function is a component of many health measures, from post-acute care and surgical outcomes to chronic frailty and disability, and is represented as one domain of human activity in the International Classification of Functioning, Disability, and Health (ICF). However, mobility and other types of functional activity remain under-studied in the medical informatics literature, and neither the ICF nor commonly-used medical terminologies capture functional status terminology in practice. We investigated two data-driven paradigms, classification and candidate selection, to link narrative observations of mobility status to standardized ICF codes, using a dataset of clinical narratives from physical therapy encounters. Recent advances in language modeling and word embedding were used as features for established machine learning models and a novel deep learning approach, achieving a macro-averaged F-1 score of 84% on linking mobility activity reports to ICF codes. Both classification and candidate selection approaches present distinct strengths for automated coding in under-studied domains, and we highlight that the combination of (i) a small annotated data set; (ii) expert definitions of codes of interest; and (iii) a representative text corpus is sufficient to produce high-performing automated coding systems. This research has implications for continued development of language technologies to analyze functional status information, and the ongoing growth of NLP tools for a variety of specialized applications in clinical care and research.

Impact of automated coding system implementation based on ICD-10 on mortality statistics in Taiwan: A bridge coding study

Impact of automated coding system implementation based on ICD-10 on mortality statistics in Taiwan: A bridge coding study

The Role of Dementia as Cause of Death: Certifiers’ Opinions versus Automated Coding

Background: Dementia is a major cause of death in many countries today. The way in which countries code causes of death determines the occurrence of dementia in statistics. The change over from manual to automated coding is accompanied by a 7–19% increase in the occurrence of dementia as the underlying cause of death. Because of this sudden change, researchers, physicians, policy makers, and press question the validity of the outcome of automated coding. Therefore, the role of dementia as a cause of death was investigated. Methods: A questionnaire was sent to a random sample of 700 certifiers who mentioned “dementia” on a death certificate in the second half of 2017. They were asked questions about the role of dementia as a cause of death. For each certificate, the opinion of the certifier was compared with the outcome of automated coding. Results: A response of 65% (n = 446) was obtained. The automated coding system selected dementia as the underlying cause of death 9.5% points (95% CI: 5.8–14.4%) more often than the certifier would do. This finding in the sample corresponded to an overestimation of dementia in the cause-of-death statistics with 22.7% (95% CI: 18–28%). Main reason for this overestimation was the selection of dementia as the underlying cause of death by the automated coding system, while it was noted as the contributory cause of death on part 2 of the death certificate by the certifier. Conclusion: For international comparisons of data on dementia as a cause of death, the outcome of automated coding can be used as the system adheres to international (ICD-10) guidelines and reduces coding variations in and between countries. However, for interpreting the local (national) impact of dementia as a cause of death, the opinion of the certifier should be taken into account.

Predicting Diagnosis Code from Medication List of an Electronic Medical Record Using Convolutional Neural Network.

Automated coding and classification systems play a role in healthcare for quality of care. Our objective was to predict diagnosis code from medication list of electronic medical record (EMR) using convolutional neural network (CNN). We collected the clinical note from outpatient department (OPD) of Wanfang hospital, Taiwan of 2016 and used three physicians from three departments. The dataset was split into two parts, 90% for training and 10% for test cases. We used medication list as input and International Statistical Classification of Diseases 10 (ICD 10) code as output. After data preprocess, we used word2vector CNN to predict ICD 10 code. This study shows all the three physicians from three departments achieved better performance. The best performance of model was a physician from cardiology department achieved precision 69%, recall 89% and F measure 78%. We need to include more component such as text data, lab report for evaluation.

Digital Medicine: Information and Technological Bases for Using Lexical Analysis in Formalization of Diagnosis Coding

The article brings forth the results of analysis of the existing practice of coding diagnoses, discusses the drawbacks of the existing model leading, in the long run, to impossibility of literate building of a monitoring system for morbidity rate and death rate. It substantiates the approaches, developed and proposed for implementation, to creating a lexical analysis module in terms of automated information systems of supporting the coding. The lexical analysis module under formation in terms of the automated coding system fits both for usage immediately at the moment of coding and post-factum and is topical (with certain technological improvement) for any version of the International Classification of Diseases. The efficiency of implementing the principles of creating and functioning of the system of the automated coding support in terms of the International Classification of Diseases with the lexical analysis module with support of transforming the clinical diagnosis into a formalized code is determined by the fact that usage of such a system increases the quality of forming the statistics of morbidity and death rates in the Russian Federation, which must allow to transfer the planning of medical aid and resource infrastructure to a new qualitative level.

The incomparability of cause of death statistics under \u201cone country, two systems\u201d: Shanghai versus Hong Kong

BackgroundValid and comparable cause of death (COD) statistics are crucial for health policy analyses. Variations in COD assignment across geographical areas are well-documented while socio-institutional factors may affect the process of COD and underlying cause of death (UCD) determination. This study examines the comparability of UCD statistics in Hong Kong and Shanghai, having two political systems within one country, and assesses how socio-institutional factors influence UCD comparability.MethodsA mixed method was used. Quantitative analyses involved anonymized official mortality records. Mortality rates were analyzed by location of death. To analyze the odds ratio of being assigned to a particular UCD, logistic regressions were performed. Qualitative analyses involved literature reviews and semi-structural interviews with key stakeholders in death registration practices. Thematic analysis was used.ResultsAge-standardized death rates from certain immediate conditions (e.g., septicemia, pneumonia, and renal failure) were higher in Hong Kong. Variations in UCD determination may be attributed to preference of location of death, procedures of registering deaths outside hospital, perceptions on the causal chain of COD, implications of the selected UCD for doctors’ professional performance, and governance and processes of data quality review.ConclusionsVariations in socio-institutional factors were related to the process of certifying and registering COD in Hong Kong and Shanghai. To improve regional data comparability, health authorities should develop standard procedures for registering deaths outside hospital, provide guidelines and regular training for doctors, develop a unified automated coding system, consolidate a standard procedure for data review and validity checks, and disseminate information concerning both UCD and multiple causes of death.

Comparing Cancer Mortality Estimates in Mexico By Using Two National Death Registries

Abstract 49 Background: Mexico has low cancer mortality compared with other Latin American countries and high-income economies. The National Institute of Geography and Statistics (INEGI) uses an automated coding system that has been adapted from the Centers for Disease Control system to process death certificates. In contrast, the System for Epidemiologic Death Statistics (SEED) relied on a standardized manual process. We hypothesized that the low cancer mortality could be explained by processing procedures. We undertook this work to compare cancer mortality rates from two national death registries that independently code and attribute cause of death from death certificates. Methods: We calculated age-standardized cancer mortality rates for all cancers and specific sites by sex for 2010 to 2014 by using data from INEGI and SEED. We calculated stroke mortality rates for comparison. We obtained age-adjusted mortality rates and 95% CIs by using direct method and World Population Prospects 2010 as a standard. Results: Age-standardized mortality rate per 100,000 women for all cancers was 74.8 (95% CI, 74.0 to 75.6) in INEGI and 74.1 (95% CI, 73.3 to 74.9) in SEED for 2010. In contrast, for men, cancer mortality rate was 69.0 (95% CI, 68.2 to 69.7) in INEGI and 68.4 (95% CI, 67.7 to 69.7) in SEED. We did not observe substantial differences across cancer sites and years; however, when analyzing cases with cancer as a contributing condition, INEGI reported approximately 50% more compared with SEED. In contrast, stroke mortality significantly differed between registries for all years. For women, in 2010, INEGI reported 29.4 (95% CI, 29.0 to 29.9) compared with 26.5 (95% CI, 26.0 to 26.9) in SEED. Corresponding estimates for men were 36.6 (95% CI, 36.0 to 37.1) and 33.0 (95% CI, 32.4 to 33.5), respectively. Conclusion: Cancer mortality estimates were minimally affected by use of distinct processing procedures. Death certification in Mexico should be further evaluated. Funding: AstraZeneca, GlaxoSmithKline, and Merck. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Susana Lozano-Esparza No relationship to disclose Dalia Stern No relationship to disclose Juan Eugenio Hernandez-Avila No relationship to disclose Evangelina Morales-Carmona No relationship to disclose Rafael Lozano No relationship to disclose Eduardo Lazcano-Ponce Consulting or Advisory Role: Merck Sharp & Dohme Research Funding: Roche, Roche (Inst), BD, Qiagen (Inst) Travel, Accommodations, Expenses: GlaxoSmithKline, Merck Sharp & Dohme, Qiagen Ruy Lopez-Ridaura Research Funding: AstraZeneca Martin Lajous Research Funding: AstraZeneca

Spontaneous facial expression in unscripted social interactions can be measured automatically.

Methods to assess individual facial actions have potential to shed light on important behavioral phenomena ranging from emotion and social interaction to psychological disorders and health. However, manual coding of such actions is labor intensive and requires extensive training. To date, establishing reliable automated coding of unscripted facial actions has been a daunting challenge impeding development of psychological theories and applications requiring facial expression assessment. It is therefore essential that automated coding systems be developed with enough precision and robustness to ease the burden of manual coding in challenging data involving variation in participant gender, ethnicity, head pose, speech, and occlusion. We report a major advance in automated coding of spontaneous facial actions during an unscripted social interaction involving three strangers. For each participant (n = 80, 47 % women, 15 % Nonwhite), 25 facial action units (AUs) were manually coded from video using the Facial Action Coding System. Twelve AUs occurred more than 3 % of the time and were processed using automated FACS coding. Automated coding showed very strong reliability for the proportion of time that each AU occurred (mean intraclass correlation = 0.89), and the more stringent criterion of frame-by-frame reliability was moderate to strong (mean Matthew's correlation = 0.61). With few exceptions, differences in AU detection related to gender, ethnicity, pose, and average pixel intensity were small. Fewer than 6 % of frames could be coded manually but not automatically. These findings suggest automated FACS coding has progressed sufficiently to be applied to observational research in emotion and related areas of study.

Cause-of-death reporting: how to measure inaccurate completion

Accuracy of cause-of-death statistics primarily depends on quality of causes reporting on the death certificate. This study aims to define a set of indicators for monitoring cause-of-death reporting at the population level using information available during the routine data processing. Reports generated by the automated coding system (ACS) were used to define and compute indicators relating to specific aspects of proper completion of death certificates. The analysis was performed on 577,989 certificates, accounting for 98% of those issued inItalyin 2009 (excluding infant deaths). The mean number of causes reported was 4.3, while 46% of certificates reported one single complete sequence. Duration was reported only in 49% of cases, 8% of them with chronological inconsistency between items. The underlying cause of death (UC) was correctly identified and reported by the certifier in 69% of cases. Specific patterns of sequence reporting by UC and the age of the decedent were also observed. The proposed indicators significantly contribute to the evaluation of quality of cause-of-death statistics, for the aspects concerning certificate completion. This approach enables analysis of a wide range of quality dimensions from large scale data and is a useful tool for those countries using automated coding systems. Francesco GRIPPO, Researcher, Socio-demographic and Environmental Statistics Directorate, Italian National Institute of Statistics (Istat), via Cesare Balbo, 16, 00184-Roma, Italy. E-mail: frgrippo@istat.it . Enrico GRANDE, Researcher, Socio-demographic and Environmental Statistics Directorate, Italian National Institute of Statistics (Istat), via Cesare Balbo, 16, 00184-Roma, Italy. E-mail: grande@istat.it . Monica PACE, Researcher, Socio-demographic and Environmental Statistics Directorate, Italian National Institute of Statistics (Istat), via Cesare Balbo, 16, 00184-Roma, Italy, currently at Eurostat, Directorate F, Social statistics, Italy. E-mail: monica.pace@ec.europa.eu .

Automating the Assignment of Diagnosis Codes to Patient Encounters Using Example-based and Machine Learning Techniques

ObjectiveHuman classification of diagnoses is a labor intensive process that consumes significant resources. Most medical practices use specially trained medical coders to categorize diagnoses for billing and research purposes. MethodsWe have developed an automated coding system designed to assign codes to clinical diagnoses. The system uses the notion of certainty to recommend subsequent processing. Codes with the highest certainty are generated by matching the diagnostic text to frequent examples in a database of 22 million manually coded entries. These code assignments are not subject to subsequent manual review. Codes at a lower certainty level are assigned by matching to previously infrequently coded examples. The least certain codes are generated by a naïve Bayes classifier. The latter two types of codes are subsequently manually reviewed. MeasurementsStandard information retrieval accuracy measurements of precision, recall and f-measure were used. Micro- and macro-averaged results were computed. ResultsAt least 48% of all EMR problem list entries at the Mayo Clinic can be automatically classified with macro-averaged 98.0% precision, 98.3% recall and an f-score of 98.2%. An additional 34% of the entries are classified with macro-averaged 90.1% precision, 95.6% recall and 93.1% f-score. The remaining 18% of the entries are classified with macro-averaged 58.5%. ConclusionOver two thirds of all diagnoses are coded automatically with high accuracy. The system has been successfully implemented at the Mayo Clinic, which resulted in a reduction of staff engaged in manual coding from thirty-four coders to seven verifiers.

Prostate cancer (PC) mortality rates: Manual versus ACME (Automated Classification of Medical Entities) coding

4699 Background: Norway has one of the highest PC mortality rates world-wide. Two processes affect the quality and comparability of cause of death statistics: the information basis and handling of the information (codification process). In Statistics Norway (SN) the codification process is manual and information both from the death certificate, autopsy reports, queries to physicians and routinely collected information from the national Cancer Registry of Norway is used. To investigate the impact of SN registration routines on the Norwegian PC mortality rates we compared the underlying cause of death (UCOD) selected by SN to the UCOD selected by ACME. Methods: From the files of SN and Cancer Registry of Norway a total of 2012 Norwegian men with a diagnosis of PC and who died in 1996 were identified. Death certificates were collected from SN and recoded according to ACME specifications with use of death certificate information only. The UCOD selected by ACME was compared to the UCOD selected by SN. Age standardized PC mortality rates (World Population) were calculated. Results: In only 29 cases ACME could not select an UCOD. These cases were coded manually. Using ACME as described the age standardized (World Population) PC mortality rate in Norway for 1996 would have been 24.0 as compared to the figure of 24.8 based on UCODs identified by SN. Overall, ACME selected PC as UCOD in 1124 cases as compared to 1161 according to SN. In 1081 cases both ACME and SN selected PC as UCOD. Conclusions: On a case level there are noticeable differences between the automated coding system ACME and Statistics Norway regarding selection of PC as underlying cause of death. Overall, however, the registration routines do not fully explain the high PC mortality rate in Norway. Author Disclosure Employment or Leadership Consultant or Advisory Stock Ownership Honoraria Research Funding Expert Testimony Other Remuneration Aventis Norway

Managers Reports of Automated Coding System Adoption and Effects on Data Quality

Summary Objective: Assessment of the adoption of automated classification (encoder) systems in healthcare settings and related effects on perceived data quality. Methods: Survey of all U.S. accredited medical records managers, summarizing their reports of automated encoding systems and data quality change following adoption of systems. Results: Significant improvement in data was seen from adoption of automated encoding systems, though variation existed across regions and key demographic variables. Conclusion: At a national level, there is a need to minimize data quality variation and ensure some degree of nationwide uniformity in the performance of coding systems. If healthcare providers are expected to trust coded data for comparative purposes, there will be a like need for more uniform and standardized system-based performance benchmarks.

Managers reports of automated coding system adoption and effects on data quality.

Assessment of the adoption of automated classification (encoder) systems in healthcare settings and related effects on perceived data quality. Survey of all U.S. accredited medical records managers, summarizing their reports of automated encoding systems and data quality change following adoption of systems. Significant improvement in data was seen from adoption of automated encoding systems, though variation existed across regions and key demographic variables. At a national level, there is a need to minimize data quality variation and ensure some degree of nationwide uniformity in the performance of coding systems. If healthcare providers are expected to trust coded data for comparative purposes, there will be a like need for more uniform and standardized system-based performance benchmarks.

The Conceptual Complexity of Presidents Carter and Clinton: An Automated Content Analysis of Temporal Stability and Source Bias

Previous work on the measurement of leaders' foreign policy orientations by Margaret G. Hermann argues that spontaneous utterances should provide more useful indications of stable personality characteristics than prepared statements. This paper reconsiders this question for Presidents Jimmy Carter and Bill Clinton through an automated coding system. A large number of spontaneous utterances and prepared statements by these presidents were compared in an effort to determine the stability of their conceptual complexity across source types and over time. The findings support the ability of Hermann's conceptual complexity measurement to capture robust differences between individuals; however, scores derived from spontaneous comments varied significantly from those derived from prepared remarks. In addition, the trait appears to be stable for some subjects (Carter) but not for others (Clinton).

Automated Coding of Patient Discharge Summaries Using Conceptual Graphs

In medicine, as in other domains, indexing and classification is a natural human task which is used for information retrieval and representation. In the medical field, encoding of patient discharge summaries is still a manual time-consuming task. This paper describes an automated coding system of patient discharge summaries from the field of coronary diseases into the ICD-9-CM classification. The system is developed in the context of the European AIM MENELAS project, a natural-language understanding system which uses the conceptual-graph formalism. Indexing is performed by using a two-step processing scheme; a first recognition stage is implemented by a matching procedure and a secondary selection stage is made according to the coding priorities. We show the general features of the necessary translation of the classification terms in the conceptual-graph model, and for the coding rules compliance. An advantage of the system is to provide an objective evaluation and assessment procedure for natural-language understanding.

Research needs for dietary assessment and monitoring in the United States

The 10-y Plan for the National Nutrition Monitoring and Related Research Program calls for research related to dietary data collection and interpretation and for improved data dissemination. The United States Department of Agriculture Human Nutrition Information Service has released data from the 1989 Continuing Survey of Food Intakes by Individuals and is planning for the next one. Respondent burden will be reduced. The sample is designed for improved estimates of dietary intakes for children and elderly individuals. An automated coding system will contribute to improved data management. Current research projects include developing data collection strategies in a cognitive research laboratory, exploring alternative interview structures for children, and developing methods to estimate the distributors of usual intakes of nutrients and foods. Priorities for further research and development are applying cognitive psychology research techniques to improve accuracy of responses, ensuring the temporal validity of food-composition databases, and developing the probability approach for assessing dietary adequacy.

PROCODE—automated coding system in group technology for rotational parts

Group technology (GT) permits to regroup mechanical parts by morpho-dimensional analogy. This constitutes an indispensable stage in automating the methods and integrating the tools into CAD/CAM systems. To create an efficient management of technical records and obtain process planning by the variant's method and interfacing it with CAPM (computer-aided production management), it is necessary to automate the process downstream of a solid modeler. The most important stage in the conception of an automatic codification system involves extracting the topological and geometrical attributes of the workpiece from the geometric modeler. The developed system is tailored for the codification of rotational parts available in the Pro/Engineer® solid modeler based on B-Rep representation. The choice of this kind of modeler was motivated by the versatility of its geometrical and technological data base. The generation of GT code is based on the PMG® system developed by CETIM, which utilizes rational classification principles. The principal results of the system prototype are presented. The system is developed using C language in a Unix environment.

Development of a coding system for recording clinical findings in farm animal practice.

A coding system was developed for recording veterinary information in the University of Liverpool farm animal practice. Existing coding systems were found to be unsuitable for use on a microcomputer in a veterinary practice. A number of new coding formats were developed and tried. The need to record varying detail, to analyse data and to enter data quickly resulted in biaxial hierarchical codes. These are used in an automated coding system to record all veterinary and reproduction information in the practice.