Researches On Theory And Application Of Adverse Drug Reaction Signal Detection

China Adverse Drug Reaction Monitoring Centre, established in 1998, mainly rely on spontaneous reporting system (SRS) to collect ADR reports. SRS is currently the most important ADR monitoring system which has earned a reputation of offering a fast and reasonably efficient way of detecting ADR signal. The signal is denoted as possible causal relationship between drug and adverse event, which is unknown or can not be fully confirmed by the previous literature. Manual review of all reports from SRS is time consuming and is difficult to establish more flexible associations among patient characteristics, the reported ADRs and suspected drugs. If we combine computer technology, pharmacoepidemiology and statistical methods to evaluate the enormous amount of information in the database, the speed of signal detection can be no doubt greatly enhanced, which enable the government to make early warning response. Therefore the establishment of a web-based signal detection and early warning system is needed urgently.We standardized Guangdong ADR database including generic drug name and ADR name at beginning. Then nine signal detection methods were compared based on the standardization database and simulation database. Bayesian confidence neural network method was applied to detect signal from Guangdong ADR database, and signal detection and warning modules were constructed. The main works and results of the study are as follows:1 Information standardizationTo promote information standardization of adverse drug reaction database of spontaneous reporting system in Guangdong and to normalize the database including 41 619 records collected from 1 January 2002 to 30 June 2007, over 60 000 generic drug names were standardized based on New Pharmacology (the 15th edition) and MCDEX Clinical Drug Reference (2006). 41 619 records of adverse drug reaction name were normalized with the WHO Adverse Reaction Terminology and International Classification of Diseases-Ten Revision. 5 407 rules of generic drug name and 4 031 standard drug names were created. 8 715 rules of adverse event name and 2 332 standard names of adverse event were established. The rule database was capable of automatic completion 80% of the standardization work of new cases. The data standardization translated the nick name to standardized information, and provided the foundation for the reuse of data such as signal detection and data mining.2 Signal detection and early warning systemWe established a web-based quantitative signal detection system on ADR spontaneous reporting database of Guangdong province in China. The system with three software modules was programmed to perform data preparation, signal detection and report generation respectively. BCPNN, the international recognized measure of disproportionality for quantitative signal detection was introduced and integrated into the system, and its capacity for signal detection was tested with ADR reports collected from 1 January 2002 to 30 June 2007 in Guangdong. 9 046 combinations from'drug-ADR','drug category-ADR','drug-ADR category'and'drug category-ADR category'combinations were collected. The data presented in'drug-ADR'signal detection which is carried out at the level of specific drug and preferred term are the most important, a total of 3 662, accounting for 40.48 percent. 2 496 suspected signals including 56 (0.62%) strong signals were mined from four types signal detection depending on the test database. Known signals such as'Cefradine induced haematuria'and'Captopril induced coughing'were found early by using the BCPNN. In addition, 291 special signals were alerted on the second quarter of 2007. The system can be used in the detection of significant signals from the database of Guangdong drug monitoring and be an extremely useful adjunct to the expert assessment of large numbers of spontaneously reported ADRs.3 Comparison of signal detection methods3.1 Comparison based on Guangdong databaseWe examined the concordance of various measures to BCPNN measure based on the database of Guangdong province in China using sensitivity, specificity and correlation coefficient. 1) ROR, PRR, Yule's Q and Poisson probability methods were comparable especially when three or more cases per combination have been collected. Sensitivity and specificity indexes increased when A (the number of combination between a specific drug and a suspected ADR) increased. 2) The sensitivity of MHRA method and the specificity of Chi-square test decreased with A increased. These two trends in'drug category-ADR category'combination are more obvious than those in'drug-ADR','drug category-ADR'or'drug-ADR category'combinations. 3) SPRT has lower sensitivity and RR method has lower specificity when A is small. 4) The correlation coefficients among the ROR, PRR, BCPNN, RR and Yule's Q close to 1. Low correlation coefficients between Chi-square and other methods were found especially for'drug category-ADR category'signal detection.3.2 Comparison based on simulation databaseWe constructed spontaneous reporting system model based on some SRS references considering several factors, such as frequency of exposure, time to market, the background incidence of adverse events, the severity of ADR as well as of reporting probability (or underreporting coefficient). The model involving 60 kinds of drugs and 40 kinds of adverse events was constructed to compare various signal detection methods based on simulation database by SAS software. The results showed that SPRT and MHRA have low sensitivity, RR and Chi-square with low specificity, and other methods are comparable. The conclusion is comparable with that from comparison based on Guangdong database with BCPNN as reference measure.4 Other relevant researches4.1 Characteristics of Guangdong ADR reportsWe applied statistical methods to analyze characteristics of Guangdong ADR reports including sex, age, nation, ADR status, outcome of ADR, relation between drug and ADR, dosage, usage and etc. 20 962 reports in male and 20 652 reports in female were collected by doctors, nurses and pharmacists et al. 19-29 and 30-39 age groups, probable or possible relation between drug and ADR were usually found. 36 283 reports (87.18%), 3 818 (9.17%), 1 230 (2.96%) and 288 (0.69%) are general ADRs, new ADRs, serious ADRs and new-serious ADRs respectively. The frequency of ADR suffered after taking intravenous injection reaches to 1.47 times, which is higher than 1.36 times on average. Of the top 10 drugs which easiest induced ADRs, half were cephalosporins. 2 835 ADRs were related with ceftriaxone sodium. Rash, anaphylactoid reaction and rigors are the most frequently reported adverse reactions in the database.4.2 Characteristics of ADR induced by cephalosporins antibioticsA retrospective investigation into 8 804 cases of ADR of cephalosporins antibiotics collected by ADR Monitoring Centre of Guangdong was carried out to pay attention to ADRs induced by cephalosporins antibiotics and correlation factor. 41 kinds of cephalosporins antibiotics were found. 2 554 reports (29.01%) were observed after Ceftriaxone Sodium administration, 1157 (13.14%) following Ceftazidime administration, 912 (10.36%) after Cefuroxime Sodium administration and 4 181 reports (47.49%) were further reported during treatments with other cephalosporins antibiotics. ADR involves four body systems, and Skin and its accessories damage accountes for the highest percentage of ADR of cephalosporins antibiotics by signal detection.4.3 Drug-drug interactions and drug-related syndromesWe applied logistic regression model to analyze the interaction of isoniazid and rifampin, and nausea-vomiting syndrome after cefradine administration based on Guangdong ADR database. Examples were provided to explore signal detection for drug-drug interactions and drug-related syndromes and give reference to ADR Monitoring.4.4 Signal testThe signals detected by signal detection system must be reevaluated by the committee of experts. The signal produced by computer is defined as first signal, and the signal produced by experts is denoted as second signal. Shenmai, Danshen and Qinkailing injection signal test were carried out by Guangdong ADR Monitoring Centre after'Yuxingcao'incident.On one hand, for theory, 1) Signal detection methods were compared based on Guangdong ADR database and simulation database, the correlations and differences among them were concluded. 2) Signal detection theory was improved and logistic regression and other statistical methods were applied for drug safety evaluation.On the other hand, for application, 1) A web-based Guangdong ADR signal detection and early warning system, three modules including data preparation, signal detection and report generation were constructed to perform automatic and manual signal detection for'drug/drug category-ADR/ADR category'. 2) Characteristics of Guangdong ADR reports and ADRs induced by cephalosporins antibiotics were analyzed.The breakthrough of this study consists in four points. 1) Guangdong web-based signal detection and early warning system was constructed and several international signal detection methods were applied in Chinese database. ADR and drug standardized database, standardization for signal detection and warning based on Chinese database were established and the detection differences among different national databases were introduced. 2) Nine methods were compared objectively based on Guangdong database considering the frequency distribution of A (the number of combination between a specific drug and a suspected ADR). Not all methods detection capabilities increase with the A value increases. 3) We considered'drug-ADR','drug category-ADR','drug-ADR category'and'drug category-ADR category'four types of signal detection. Comprehensive evaluation was performed. 4) Characteristics of Guangdong ADR reports, characteristics of ADRs induced by cephalosporins antibiotics and other relevant issues were analyzed with general description and statistical inference methods, and signal detection methods.ADR signal detection and early warning system is an automatic and systematic method to mine various types of drug safety database. Potential safety problems can be found early and accurately by the system. Operational processes from data preparation, signal detection to signal test were constructed in this study, which can provide some references for establishment, development and utilization of domestic spontaneous adverse drug reaction databases.