Collecting Infectious Disease Data from LARS and Improving Data Quality in Taiwan

Chih-Ting Yeh, Chai-Lin Li, Chih-Jung Ke, Chi-Ming Chang

Abstract


ObjectiveTo improve data quality and sustain a good quality data collectedby Laboratory Automated Reporting System (LARS), we use a Three-stage Data Quality Correction (3DQC) strategy to ensure data accuracy.IntroductionTo immediately monitor disease outbreaks, the application oflaboratory-based surveillance is more popular in recent years.Taiwan Centers for Disease Control (TCDC) has developed LARS tocollect the laboratory-confirmed cases caused by any of 20 pathogensdaily via automated submitting of reports from hospital laboratoryinformation system (LIS) to LARS since 2014 [1]. LOINC is usedas standardized format for messaging inspection data [1, 2]. Thereare 37 hospitals have joined LARS, coverage rate about 59% of allhospitals in Taiwan. Recently, more than 10,000 of data are collectedweekly and used in monitoring pathogen activity [3]. Therefore, itis important to ensure data quality that the data will lead to valuableinformation for public health surveillance.MethodsA 3DQC strategy was designed to improve data accuracy andcarried out by teamwork among TCDC, Taiwan Association forMedical Informatics (TAMI) and IT Company (Figure 1). In the firststage of 3DQC, IT Company checked data format. In the second stage,TCDC verified information between hospital inspection reports anddata receiving in LARS. In the third stage, TAMI evaluated LOINCmapping and TCDC monitored stability of data transmission. Aftercorrecting the data, hospitals were approved to join LARS.ResultsDuring the first stage of 3DQC, we observed that some problemswith syntax error in data (e.g. incorrect patient identification number,or lack of residence codes). Because some data were stored in Hospitalinformation system (HIS) but not in LIS, an error may occur whilehospital accessed records from HIS. In the second stage, 50-70%of inspection reports provided by each hospital had problems withsemantic information error. For example, a positive result of influenzaA on a screening flu test recorded in LIS but hospital transferred thewrong result with influenza B positive into LARS. In the third stage,we found that 20-30% of terms mismatched to LOINC code. Thisstudy categorized these terms into two groups (1) the Exceptioncodes, which were considered reasonable and (2) the Error codes, andalso reviewed Error codes and made a modified advice for hospitalsto improve LOINC mapping. Through 3DQC strategy, the LOINCmapping rate raised from 40 to 80%, Exception codes mapping was20%, and the total mapping rate was near 95-99% (Figure 2). Sofar, most hospitals have maintained a good quality data even theyformally participate in LARS.ConclusionsThis study suggested that 3DQC can effectively detect problemsand reduce errors of data collected from LARS, and indicated thateffect of 3DQC can be maintained even hospital formally participatesin LARS. Future research will focus on development of automaticprogramming of 3DQC to ensure high-quality data.Figure 1. A Three-stage Data Quality Correction strategy

Full Text:

PDF


DOI: https://doi.org/10.5210/ojphi.v9i1.7610



Online Journal of Public Health Informatics * ISSN 1947-2579 * http://ojphi.org