Background: Our preliminary analyses of an adverse event database found a “signal” on actaminophen-associated abnormal behavior. Abnormal behavior by Japanese children taking oseltamivir has been repeatedly reported and described. However there is no literature dealing with acetaminophen-associated abnormal behaviors, which is analyzed and further described in this paper.
Methods: 216,945 records registered to JADER between 2004 and 2011 were analyzed. Age, sex, reporting year, and underlying illness were compared between cases of abnormal behavior and controls without abnormal behavior. Multivariate logistic regression modeling was used to investigate confounding and interaction with age group, gender, reporting year, and an underlying disease.
Results: Of 4,544 acetaminophen-treated cases, 260 (5.7% [95%CI; 5.2-6.4]) exhibited abnormal behaviors. Of the 260 cases, factors associated with abnormal behavior were male (196 case; 75.4% [69.7 – 80.5]), under 20 years old (242; 93.1% [89.3 – 95.8]), reported in 2007 or 2008 (169; 65.0% [58.9 – 70.8]), and with influenza (253; 97.3% [94.5 – 98.9]) as an underlying disease. Multivariate logistic regression analysis indicated a relationship between influenza and acetaminophen and an interaction between acetaminophen and oseltamivir. According to subgroup analysis with influenza (n=2,164), patients taking oseltamivir had fewer abnormal behaviors, because teenagers had been reported to be at high risk for abnormal behaviors and doctors were advised to avoid prescribing oseltamivir to them.
Conclusion: Acetaminophen-associated abnormal behavior occurred mostly in persons with influenza. Given the relationship to influenza and the interaction of oseltamivir, the reported acetaminophen-associated abnormal behaviors were most likely induced by the underlying disease, not by the drug itself.
After reports of neuropsychiatric adverse events and abnormal behaviors by Japanese children taking oseltamivir since the 2004-2005 influenza season, this type of adverse event has become the focus of much attention. Exhaustive efforts have been made to clarify a causative relationship between the suspected drug, oseltamivir, and abnormal behavior. These included comparison of post-marketing spontaneous adverse event reports, preclinical and clinical studies, analysis of medical records in the UK General Practice Research Database, and the abnormal behaviors of Japanese and Taiwanese children with influenza. Reviews of the accumulated data did not suggest causality (1, 2). In contrast, acetaminophen has escaped scrutiny. Because physicians commonly use acetaminophen in treatment, they believe that it is generally safe as long as it is used properly.
Because the number of clinical trial participants is much fewer than the number of patients who will be treated with the investigational drug in clinical practice, premarketing clinical trials may fail to detect rare ADRs or ADRs specific to particular population subgroups. Thus, many ADRs are unknown at the time of the launch or even long after the launch of a drug. In order to track ADRs, the World Health Organization (WHO) and health authorities such as the U.S. Food and Drug Administration (FDA) run safety databases. The WHO safety database, termed VigiBase, contained more than 6.0 million individual case safety reports (ICSRs) as of Jan 2013 (http://www.bridgetodata.org/ , last accessed on Mar 2013) and the Adverse Event Reporting System (FAERS) of the U.S. FDA had over 4.0 million ICSRs as of Sep 2012 (http://www.fda.gov/ , last accessed on Mar 2013). In order to assist pharmacovigilance activity through these large databases, several statistical methods have been promulgated that automatically extract a manageable number of “signals” for further evaluation. The commonly used methods are the proportional reporting ratio (PRR) (3), the reporting odds ratio (ROR) (4), the Baysian Confidence Propagation Neural Network (BCPNN) (5), and the (multi) Gamma Poisson Shrinker (mGPS) (6). The BCPNN and the mGPS of the above mentioned methods are used for the WHO and FDA safety databases, respectively (7, 8).
The Japanese Pharmaceutical and Medical Devices Agency (PMDA) has run its own safety database since 2004. The safety database is termed JADER (Japanese Adverse Drug Event Report database) and has been open to the public since May 2012. The publicly available JADER allows researchers to study adverse events and to assess the potential risk of pharmaceutical drugs.In our preliminary screening, we used the BCPNN method with the JADER database to search for ”signals”. The objective of this study is to describe and to investigate one of the detected signals, acetaminophen-associated abnormal behavior.
Study design: This study is composed of mainly two parts. The first part is an explorative signal detection study in which we describe “signals” or potential drug risks. The second part of the study is a case-control study where we compare demographical and underlying disease regarding the potential risk.
Setting: We used a public version of the JADER database, from the PMDA that was released in May 2012. This version of the database includes adverse drug reactions reported to the PMDA between Apr 2004 and Dec 2011.
Subjects: We prepared four datasets for specific purposes (Table 1). The first dataset was the whole JADER database (Dataset 1; n=216,945), which was used for “signal” detection. The second dataset included only acetaminophen-treated patients (Dataset 2; n=4,544). This dataset was used to describe and compare the demographic data of patients exhibiting abnormal behaviors and other acetaminophen-treated patients who did not. It was extracted by a search for their drug (generic name) in the “Drug201204_teiseiban” table that included “Acetaminophen.” The third dataset contains influenza patients (Dataset3; n=2,164), and was used for a stratified analysis. Dataset 3 was extracted by searching the Hist201204_teiseiban table that included “influenza” as the underlying diseases. The fourth dataset was of influenza patients treated with acetaminophen (Dataset 4; n=671). Dataset 4 includes the patients of Datasets 2 and 3. Dataset 4 was used to compare the ADRs of oseltamivir and acetaminophen-treated influenza patients. The raw data handling and basic tabulations were performed as previously reported (9-12).
Analysis: The potential risk or “signal” for the combination of a specific drug and an adverse event was calculated based on the Information Component (IC) value as reported before (5, 7). The detected 65,952 signals were sorted by “McLachlan’s sensitivity”, one of the Bayesian estimators (13-15). The rates of the abnormal behaviors were shown for the indicated drugs. The 95% confidence interval (95%CI) for the rate was calculated as previously reported (16). The indicated numbers are the number of cases or non-cases of the conditions indexed for calculation of the rate. The number of identified abnormal behavior cases and comparisons was tabulated by reporting year, sex, age group, and underlying illness. Demographic data are shown with the number of cases, the crude odds ratio with 95%CI, and P-value calculated with two-sided Fisher’s exact test or Chi-squared test. Because of the nature of spontaneous reports, the calculated P-values are shown to highlight the differences in characteristics between patients with abnormal behaviors and the comparison group, but were not intended to test for statistical differences. Variables showing an association with an abnormal behavior by univariate analysis were further analyzed by multivariate unconditional logistic regression analysis using the R-statistical package (R Foundation for Statistical Computing, Vienna Austria 2012). Records with missing data were removed from the multiple logistic regression analysis.
There were 216,945 adverse event cases recorded in the JADER database between 2004 and 2011 (Dataset 1). Of these 4,544 were treated with acetaminophen (Dataset 2). 2,164 cases (Dataset 3) were extracted from Dataset 1 whose underlying diseases included influenza. Of these 2,164, 671 were treated with acetaminophen (Dataset 4). (Table 1)
The BCPNN yielded more than 50 thousands combinations with a potential risk of a drug-adverse event (Table 2). We listed drugs frequently reported as being associated with abnormal behaviors in Table 3. Of 216,945 records, 838 (0.4% [95%CI; 0.4-0.4]) were cases of abnormal behaviors. Of these 260 (5.7% [95%CI; 5.1-6.4]) cases of abnormal behavior were associated with 4,544 actaminophen-treated cases (Table 3). The characteristics associated with a high number of reports were male (196 cases), ages under10 (97 cases), age 10-19 (145 cases), and report year of 2007 (108 cases) or 2008 (61 cases) (Table 4). Because there was an increase in the number of abnormal behavior reports in 2007 and 2008, we handled these two years as a single categorical variable, in the logistic regression analysis. Similarly, we handled age under 20 as a single categorized variable. Besides the above, the most characteristic factor was influenza as an underlying disease (253 cases). According to logistic regression analysis of dataset1, concordance was found between acetaminophen and influenza (supplemental table, Full Model; Model A vs. Model B, Minimum Model; Model D vs Model E), but there was no interaction between acetaminophen and influenza (Full Model; Model B vs. Model C, Minimum Model; Model E vs. Model F). Based on these results, we did a more detailed analysis with the 2,164 cases (Dataset3) who had influenza as an underlying disease. The use of oseltamivir was associated with reduced acetaminophen-associated abnormal behavior (Model H vs. Model I). The interaction of oseltamivir and acetaminophen is shown in Fig 1. The rates of abnormal behaviors among the cases without oseltamivir treatment without and with acetaminophen were 29.2% [95%CI; 25.4-33.7] and 52.3% [45.8-58.8] respectively. Thecorresponding rates for cases treated with oseltamivir were 24.4% [21.7-7.1] and 29.6% [25.4-34.2] respectively. Thus, the use of oseltamivir appeares to be related to fewer acetaminophen associated abnormal behaviors. Comparison was done of the age distribution of abnormal behavior between patients with and without oseltamivir for the influenza cases (Fig 2). Most of the cases without oseltamivir were in the 10-19 years age group. Thus, use of oseltamivir might avoid high risk age group. The prescription of oseltamivir appeared to be a confounding factor related to the abnormal behaviors associated with acetaminophen. This appeares to be a well-known epidemiological effect, so-called “confounding by indication”.
The BCPNN that used all of the cases in JADER (Dataset1) yielded more than 50,000 combinations of potential adverse events and drugs (Table 2) Most of the combinations are well known, such as myelosuppression associated anti-neoplastic cytotoxic agents, hemorrhage associated with anti-coagulation agents, hypersensitivity associated with iodinated contrast agents, hypoglycemia associated anti-diabetic agents, or infection associated with immunosuppressants. Among these, abnormal behaviors associated with acetaminophen drew our attention. Although acetaminophen is widely used as an analgesic antipyretic drug and is easily purchased over the counter, abnormal behaviors associated with it have rarely been reported. We have never experienced abnormal behaviors associated with acetaminophen in our own clinical practice. Thus, the result was unexpected and we decided to do further examination using data mining methodology. First, we listed the top 10 most frequently reported drugs associated with abnormal behaviors (Table 3). The rate of abnormal behaviors appeared high for acetaminophen (5.7% [95%CI; 5.1-6.4]) compared with that for all drugs (0.4% [95%CI; 0.4-0.4]). Oseltamivir and zanamivir also have high report rates, which should be interpreted with caution because both drugs have undergone active surveillance for abnormal behaviors and have been widely covered in the media. The images of both have been affected by non-scientific reporting (7).
We compared the demographic data of 4,544 cases with and without abnormal behaviors treated with acetaminophen (Dataset 2) with and without abnormal behavior. The characteristics of ICSR with abnormal behaviors included male, age under 20 years, and reported in 2007 or 2008.
Newly marketed pharmaceuticals, by virtue of their novelty alone, sometimes elicit adverse-event reports at high rates, and the report rate tends to decrease over time. This tendency is known as the Weber effect (17). It is also well known that media coverage accelerates reports of specific adverse events. However, acetaminophen is a medicine that was first synthesized in 1878 and that was introduced for clinical use in 1887 (18). As far as the authors know, acetaminophen was not a topic of wider public attention until recently. Thus, the increased reports from 2007 and 2008 suggest that the increased number of reports was affected by other factor(s) that may have nothing to do with the medicine itself. Moreover, most of the cases with abnormal behavior had influenza, thus the reports of abnormal behavior appears to be related to the underlying disease. We did multivariate logistic regression analysis to clarify how influenza affected the reporting of abnormal behavior from the point of view of an epidemiological effect. The multivariate logistic regression analysis using dataset 1 showed a confounding effect for influenza and acetaminophen. (supplemental table, Full Model; Model A vs. Model B, Minimum Model; Model D vs. Model E, the difference between the full model and the minimum model was the number of concomitantly used medicines. ) Thus, the effect appeared to be a result of the underlying disease. A limitation of the study is the possibility of multiple co-linearity problems when we put the related variables together in one model, such as for influenza as an underlying disease and anti-influenza drugs that are prescribed exclusively for influenza. However, both models indicated the same results regarding the influence of the influenza, the minimum model without anti-influenza drugs and the full model with all potential and available variables, including anti-influenza drugs. Thus, regardless of the model, the confounding of influenza was robust.
This result lead us address to test the possibility that a concomitantly administered acetaminophen happened to be reported when the oseltamivir-associated adverse event was reported. If the acetaminophen were just a concomitant medication, acetaminophen associated abnormal behavior cases would be associated with oseltamivir. On the contrary, oseltamivir was associated with decreased acetaminophen associated abnormal behavior reporting (n = 2,164; Model H vs. Model I in Dataset 3). In dataset 3, because all cases had influenza as an underlying illness, there would not be multiple co-linearity between the use of oseltamivir and influenza, thus an interaction based on abnormal behavior was shown between influenza and acetaminophen (Figure 1). Among the acetaminophen-treated patients, the rates of abnormal behavior with and without oseltamivir were 29.6 % [95%CI; 25.4 - 33.7] and 52.3 % [95%CI; 45.8 - 58.8], respectively. The rate of acetaminophen-associated abnormal behavior appeared to be decreased in patients with oseltamivir. On the other hand, among the patients without acetaminophen with and without oseltamivir the rates were 24.4 % [95%CI; 21.7 – 27.1] and 29.4 % [95%CI; 25.4 – 33.7], respectively. The rate of abnormal behavior was not affected by oseltamivir use. Thus, the interaction indicated by the logistic regression models was confirmed. In order to further explore these effects, we calculated the age distribution of patients treated with acetaminophen (Dataset 4, Figure 2). Few teenagers were treated with oseltamivir. This is because of the health ministry recommendation that doctors not prescribe oseltamivir to teenagers, who had been reported in the media to be at high risk for abnormal behaviors associated with oseltamivir treatment. Physicians would have thus avoided the prescription of oseltamivir for teenagers because of the high risk for abnormal behavior. Patients were often treated with symptomatic therapy based on antipyretics such as acetaminophen. Based on these conditions, the reported acetaminophen-associated abnormal behavior events were most likely influenza-induced events.
Strength and limitations of this study
Screening of existing databases, such as JADER can be done quickly and at an affordable cost. This exploratory approach will provide useful data for interested scientists. Regarding the potential confounding by indication, for example fever, since we have no access to the raw data, we cannot do further evaluation. Moreover, due to the incompleteness and inconsistencies of witnessed adverse event reporting, analyses of databases can supplement, but not replace controlled epidemiologic studies.
Association with abnormal behaviors as an adverse event by patients treated with acetaminophen was found by a data mining method, BCPNN. Given that, the large majority of the cases reported had influenza as the underlying disease, the reported acetaminophen-associated abnormal behaviors were most likely associated with adverse events induced by the influenza itself.
1. Toovey S, Prinssen EP, Rayner CR, Thakrar BT, Dutkowski R, Koerner A, Chu T, Sirzen-Zelenskaya A, Britschgi M, Bansod S, Donner B. Post-marketing assessment of neuropsychiatric adverse events in influenza patients treated with oseltamivir: an updated review. Adv Ther 29:826-848, 2012.
2. Toovey S, Rayner C, Prinssen E, Chu T, Donner B, Thakrar B, Dutkowski R, Hoffmann G, Breidenbach A, Lindemann L, Carey E, Boak L, Gieschke R, Sacks S, Solsky J, Small I, Reddy D. Assessment of neuropsychiatric adverse events in influenza patients treated with oseltamivir: a comprehensive review. Drug Saf 31:1097-1114, 2008.
3. Evans SJ, Waller PC, Davis S. Use of proportional reporting ratios (PRRs) for signal generation from spontaneous adverse drug reaction reports. Pharmacoepidemiol Drug Saf 10:483-486, 2001.
4. van Puijenbroek EP, Bate A, Leufkens HG, Lindquist M, Orre R, Egberts AC. A comparison of measures of disproportionality for signal detection in spontaneous reporting systems for adverse drug reactions. Pharmacoepidemiol Drug Saf 11:3-10, 2002.
5. Bate A, Lindquist M, Edwards IR, Olsson S, Orre R, Lansner A, De Freitas RM. A Bayesian neural network method for adverse drug reaction signal generation. Eur J Clin Pharmacol 54:315-321, 1998.
6. Szarfman A, Machado SG, O'Neill RT. Use of screening algorithms and computer systems to efficiently signal higher-than-expected combinations of drugs and events in the US FDA's spontaneous reports database. Drug Saf 25:381-392, 2002.
7. Oshima Y. [Series: For attending physicians; professionalism; causality assessment between reported adverse events and suspected drugs]. Nihon Naika Gakkai Zasshi 100:2323-2333, 2011.
8. Stahl M, Lindquist M, Edwards IR, Brown EG. Introducing triage logic as a new strategy for the detection of signals in the WHO Drug Monitoring Database. Pharmacoepidemiol Drug Saf 13:355-363, 2004.
9. Oshima Y. Characteristics of drug-associated rhabdomyolysis: analysis of 8,610 cases reported to the US Food and Drug Administration. Internal Medicine 50:845-853, 2011.
10. Oshima Y. Characteristics Of 1,544 Drug-induced Hypersensitivity Syndrome Cases Reported To Food And Drug Administration. 2011.
11. Oshima Y, Tsukamoto H, Tojo A. Association of hepatitis B with antirheumatic drugs: a case-control study. Mod Rheumatol, 2012.
12. Oshima Y, Yuji K, Tojo A. Eltrombopag in refractory aplastic anemia. N Engl J Med 367:1162; author reply 1163, 2012.
13. Ahmed I, Dalmasso C, Haramburu F, Thiessard F, Broet P, Tubert-Bitter P. False discovery rate estimation for frequentist pharmacovigilance signal detection methods. Biometrics 66:301-309, 2010.
14. Ahmed I, Haramburu F, Fourrier-Reglat A, Thiessard F, Kreft-Jais C, Miremont-Salame G, Begaud B, Tubert-Bitter P. Bayesian pharmacovigilance signal detection methods revisited in a multiple comparison setting. Stat Med 28:1774-1792, 2009.
15. Ahmed I, Thiessard F, Miremont-Salame G, Begaud B, Tubert-Bitter P. Pharmacovigilance data mining with methods based on false discovery rates: a comparative simulation study. Clin Pharmacol Ther 88:492-498, 2010.
16. Clopper CJ, Pearson ES. The Use of Confidence or Fiducial Limits Illustrated in the Case of the Binomial. Biometrika 26:404-413, 1934.
17. Weber JCP. Epidemiology of adverse reactions to nonsteroidal anti-inflammatory drugs. In: Rainsford KD VG, editor. Advances in Inflammation Research New York, Raven Press; 1984. pp 1-7.
18. Bertolini A, Ferrari A, Ottani A, Guerzoni S, Tacchi R, Leone S. Paracetamol: new vistas of an old drug. CNS Drug Rev 12:250-275, 2006.
Source(s) of Funding
This study has been done without funding.
The authors report the following funding sources which could have possible relevance to the work described here: Yasuo Oshima, MD, PhD, FACP; received a stipend from Sanofi K.K. Hideyuki Ikematsu, MD; received honoraria for advice from Chugai Pharmaceutical Co., Ltd., Taisho Toyama Pharmaceutical co., Ltd., Glaxo Smith Kline K.K., Shionogi & Co., Ltd., and Daiichi Sankyo Co., Ltd.