Signal Detection of Oral Drug-Induced Dementia in Chronic Kidney Disease Patients Using Association Rule Mining and Bayesian Confidence Propagation Neural Network

ORIGINAL ARTICLES

Laboratory of Clinical Pharmacy1, Gifu Pharmaceutical University, Gifu; Department of Pharmacy2, Secomedic Hospital, Funabashi; Community Health Support and Research Center3; Laboratory of Community Healthcare Pharmacy4, Gifu Pharmaceutical University, Gifu, Japan

Signal detection of oral drug-induced dementia in chronic kidney disease patients using association rule mining and Bayesian confidence propagation neural network

Y. N OGUCHI1,*, H. NAGASAWA2, T. TACHI1, T. TSUCHIYA3, H. TERAMACHI1,4,*

Received February 28, 2019, accepted April 22, 2019 *Corresponding authors: Yoshiro Noguchi, Hitomi Teremacji, Laborataory of Clinical Pharmacy, Gifu Pharmaceu- tical University, 1-25-4, Daigakunishi, Gifu 501-1196, Japan [email protected]; [email protected] Pharmazie 74: 570-574 (2019) doi: 10.1691/ph.2019.9426

Among the mechanisms responsible for cognitive dysfunction in chronic kidney disease (CKD) are albuminuria and oxidative stress. However, there may be other causes not yet identified. In fact, the full relevance of CKD patient drug use and its relationship to dementia has hardly been barely investigated. We identified drugs affecting cognitive function in CKD patients by analyzing the spontaneous reporting system in Japan using Association rule mining (ARM) and Bayesian confidence propagation neural network (BCPNN). The signal

detection criterion used were as follows: case ≥ 3, lift > 1, conviction > 1 (ARM) and IC025 >0 (BCPNN). Drugs

with more than 20 cases were valaciclovir (lift: 11.21, conviction: 1.28, IC025: 3.12), amantadine (lift: 19.69, convic-

tion: 1.68, IC025: 3.05), nalfurafine (lift: 8.35, conviction: 1.19, IC025: 2.18), pregabalin (lift: 6.05, conviction: 1.12,

IC025: 1.78), and acyclovir (lift: 5.89, conviction: 1.12, IC025: 1.68). This study is the first report to use a large-scale medical database to identify drugs related to oral drugs-induced dementia in CKD.

1. Introduction risk of dementia, understanding the effect of drugs on cognitive Rates of end stage kidney disease (ESKD) have increased world- function is an important issue for providing safe medical care. wide and the number of dialysis patients is increasing rapidly. For early detection of unknown drug induced adverse events Patients with ESKD have a significantly shorter life expectancy (AEs), signal detection based on disproportionality analysis has than individuals without ESKD. Therefore, the rapid increase generally been used. in dialysis patients not only strains patients themselves but also As an indicator of signal detection, detection models using creates an economic burden to the healthcare system (Etgen et frequency statistics include proportional reporting ratio (PRR) al. 2012). Therefore, the goal of chronic kidney disease (CKD) (Evans et al. 2001) and reporting odds ratio (ROR) (Rothman et treatment is to reduce the incidence of ESKD by controlling CKD. al. 2004). As other detection models have information compo- However, it has been reported that CKD patients may have cogni- nent (IC) given by Bayesian confidence propagation neural tive function disorders (Seliger et al. 2004; Etgen et al. 2012; network (BCPNN) (Bate et al. 1998) and empirical Bayes Deckers et al. 2017). Declining cognitive function, including geometric mean (EBGM) (Szarfman et al. 2002) using Bayesian dementia, decreases patient’s quality of life and increases the estimations. burden on caregivers. Once developed, dementia is often irrevers- While association rule mining (ARM) is a signal detection model ible, so its prevention is important. To provide effective and safe which considers drug interactions and primary disease, has been medical care to CKD patients with various complications, it is reported to have a signal detection power equivalent to that of important to know drugs that affect their cognitive function. disproportionality analysis, and is a simpler model (Noguchi et al. CKD has been shown to be an important risk factor for the onset 2018a,b). Additionally, the signal detection by BCPNN is stable of cardiovascular disease (CVD) and CVD death, but recently, an even if the reported numbers are small. Therefore, to detect signals association between CKD and dementia has also been reported of oral drug-induced dementia in CKD patients in a pharmaco- (Seliger et al. 2004; Etgen et al. 2012; Fujisaki et al. 2014; Deckers epidemiologic model, we used ARM and BCPNN to analyze the et al. 2017). A clinical study on the association between CKD and Japanese Adverse Drug Event Report (JADER) database as spon- dementia in the US showed a 37% increase in dementia risk in taneous reporting system in Japan. people aged 65 years or older with moderate renal dysfunction (Seliger et al 2004). Among the mechanisms of suggested to contribute to cognitive 2. Investigations and results dysfunction in CKD are albuminuria and oxidative stress. These Table 1 shows the characteristics of CKD patients registered in were verified by meta-analysis (Deckers et al. 2017) and animal the JADER. The number of cases of CKD patients registered in experiments in mice (Fujisaki et al. 2014). However, there may JADER was 28,817. Although, since JADER is based on sponta- be other causes that have not been clarified yet. For example, no neous reporting, the number of CKD patients using drugs cannot analysis has been made of a large-scale medical database to iden- be clearly defined, among them, 1.33 % under the age of 60 were tify cognitive dysfunction caused by drugs used by CKD patients, reported to have developed drug-induced dementia, and 2.51 % of the full relevance of CKD patient’s drug use and its relationship to patients over the age of 60 years. The report of males was 2.14 % dementia has been barely clarified. In CKD patients with a high and that of females was 2.14 %. 570 Pharmazie 74 (2019) ORIGINAL ARTICLES

Table 1: Characteristics of chronic kidney disease patients 3. Discussion

Reporting The number of cases of CKD patients registered in JADER was Case CKD rate (Dementia) patients 28,817, which was 8.2 % of the total. JADER cannot calculate the (%) average age of enrolled patients because cases are registered for Total 616 28,817 2.14 every age of 10 years, but the distribution of reporting proportions Sex in each age was the same for males and females. Although, since the database used is based on spontaneous reporting, the number Male 363 16,993 2.14 of CKD patients using drugs cannot be clearly defined, regardless Female 253 11,824 2.14 of sex, the reported proportion of CKD patients reporting cognitive function-related adverse events was highest in the age over Age 60 years. – 19 11 1,271 0.87 Using the ARM, the apriori algorithm can be used to reduce the 20 – 29 7 684 1.02 number of calculations. The apriori algorithm is based on the principle that “support of a certain item set is always less than or equal 30 – 39 12 1,455 0.82 to support of its partial item set” (Agrawal and Srikant 1994). That 40 – 49 23 1,996 1.15 is, ARM omits calculation except for unreasonable combinations 50 – 59 67 3,644 1.84 among various combinations extracted from large database by apriori algorithm, so there is a possibility that signals of various 60 – 69 157 6,808 2.31 conditions can be detected at an early stage (Noguchi et al. 2018a). 70 – 79 187 7,634 2.45 In our previous studies, ARM showed a sensitivity of 99 %, a 80 – 89 127 4,666 2.72 specificity of 94 %, and a Youden’s index of 0.94. Compared with PRR of the conventional signal detection model, ARM has the 90 – 25 659 3.79 same degree of detectability despite a simple calculation model < 59 120 9,050 1.33 (Noguchi et al. 2018a). 60 ≤ 496 19,767 2.51 On the other hand, when the reported number is small, it is predicted that signal detection by PRR based on the frequency-the- oretic statistical model becomes unstable compared to BCPNN 1524 drugs were registered with adverse events in JADER, and based on the Bayesian statistical model. Therefore, in this study, with a wide range search using ARM, 55 drugs were found to have we first performed prescreening using ARM and used BCPNN as a a signal of dementia in CKD patients (Table 2). Among them, the verification of the obtained signals. drug whose signal was detected by BCPNN was 30 drugs. In CKD patients, there were 55 drugs for which signals of cogni- Furthermore, among these 30 drugs, drugs with more than 20 cases tive function related adverse events included in the dementia wide range search were detected using ARM, and 30 drugs of signals were valaciclovir (case: 174, lift: 11.21, conviction: 1.28, IC025: 3.12), amantadine (case: 33, lift: 19.69, conviction: 1.68, IC : using both ARM and BCPNN. 025 And of them, many were drugs classified as NERVOUS SYSTEM 3.05), nalfurafine (case: 31, lift: 8.35, conviction: 1.19, IC025: 2.18), pregabalin (case: 29, lift: 6.05, conviction: 1.12, IC : 1.78), and (18 drugs) and ANTIINFECTIVES FOR SYSTEMIC USE (6 025 drugs) by the Anatomical Therapeutic Chemical (ATC) Classifica- acyclovir (case: 25, lift: 5.89, conviction: 1.12, IC025: 1.68). In addition, among surveyed patients with dementia signals in the tion System. The drugs with more than 20 cases were valaciclovir, wide range search of SMQ, signals of poor prognosis were detected amantadine, nalfurafine and pregabalin. Furthermore, in the survey with the use of nalfurafine (case: 8, lift: 23.90, conviction: 1.05), on poor prognosis, signals of nalfurafine, valaciclovir, amantadine valaciclovir (case: 8, lift: 5.71, conviction: 1.01), amantadine (case: 5, and pregabalin were detected among the 4 drugs with more than lift: 33.09, conviction: 1.07), sitagliptin (case: 3, lift: 8.96, conviction: 20 cases in the wide range search of dementia. Therefore, in this 1.02), and pregabalin (case: 3, lift: 6.94, conviction: 1.01) (Table 3). paper, we will concentrate on these 4 drugs. Dementia signal detection in a narrow range search identified Valaciclovir is a prodrug of acyclovir with improved oral absorb- nalfurafine (case: 6, lift: 26.79, conviction: 1.03) and pregabalin ability and which causes neuropsychiatric symptoms, namely (case: 4, lift: 13.82, conviction: 1.01). acyclovir encephalopathy.

Table 2: Signal of drugs affecting cognitive function in chronic kidney disease patients

ATC Class Drug name Case Lift Conv. IC level 1 025 A Alimentary tract and metab- Famotidine 15 2.59 1.04 0.47* olism Omeprazole 6 1.57 1.01 -0.59 Cimetidine 3 7.44 1.16 -0.10 Prednisolone 4 2.51 1.03 -0.41 Vildagliptin 11 1.06 1.00 -0.79 Sitagliptin 5 1.35 1.01 -0.87 C Cardiovascular system Digoxin 4 3.37 1.05 -0.18 Aprindine 5 4.53 1.08 0.24* Methyldopa 3 5.66 1.11 -0.19 J Antiinfectives for systemic use Ceftriaxone 5 1.39 1.01 -0.84 Cefepime 3 1.54 1.01 -1.06 Levofloxacin 6 1.43 1.01 -0.70 Voriconazole 9 5.11 1.10 0.87*

Pharmazie 74 (2019) 571 ORIGINAL ARTICLES

ATC Class Drug name Case Lift Conv. IC level 1 025 J Antiinfectives for systemic use Cycloserine 4 31.43 2.94 0.43* Delamanid 4 20.95 1.76 0.46* Ethambutol 3 1.89 1.02 -0.88 Isoniazid 3 1.30 1.01 -1.22 Valaciclovir 174 11.21 1.28 3.12* Aciclovir 25 5.89 1.12 1.68* Oseltamivir 14 9.17 1.22 1.72* Famciclovir 4 11.09 1.28 0.39* Daclatasvir 3 3.01 1.05 -0.52 Asunaprevir 3 3.01 1.05 -0.52 M Musculo-skeletal system Baclofen 6 12.30 1.32 0.97* N Nervous system Fentanyl 5 4.91 1.09 0.30* Tramadol hydrochloride - Ac- 10 6.64 1.14 1.17* etaminophen mixt. Tramadol 5 16.84 1.52 0.79* Oxycodone 5 7.60 1.17 0.55* Morphine 3 15.72 1.47 0.01* Pentazocine 3 7.07 1.15 -0.11 Pregabalin 29 6.05 1.12 1.78* Levetiracetam 7 7.67 1.17 0.93* Carbamazepine 4 1.96 1.02 -0.63 Amantadine 33 19.69 1.68 3.05* Rotigotine 5 16.84 1.52 0.79* Risperidone 5 4.29 1.08 0.20* Sulpiride 3 4.71 1.09 -0.27 Quetiapine 3 3.93 1.07 -0.36 Etizolam 8 10.19 1.25 1.24* Hydroxyzine 3 10.10 1.25 -0.03 Zolpidem 13 9.89 1.24 1.72* Brotizolam 8 7.86 1.17 1.08* Suvorexant 4 13.47 1.37 0.43* Triazolam 3 6.15 1.13 -0.16 Paroxetine 6 6.15 1.13 0.63* Sertraline 5 11.79 1.31 0.71* Donepezil 5 4.45 1.08 0.23* Galantamine 3 6.74 1.14 -0.13 Rivastigmine 3 4.04 1.07 -0.35 Varenicline 3 15.72 1.47 0.01* R Respiratory system Theophylline 3 7.07 1.15 -0.11 Levocetirizine 3 17.68 1.57 0.01* Epinastine 3 7.86 1.17 -0.08 V Various Nalfurafine 31 8.35 1.19 2.18* (NA) (NA) Mosapride 3 4.04 1.07 -0.35

ATC: Anatomical Therapeutic Chemical (classification system), Conv.: conviction, IC025: the lower end of 95% Confidence interval for the information component, NA: not applicable, *: the signal detection of both the association rule mining and the Bayesian confidence propagation neural network.

Acyclovir encephalopathy produces a high rate of consciousness acyclovir is several hours for subjects with normal renal function; disorders (Adair et al. 1994; Asahi et al. 2009), hallucinations (Asahi however, in renal failure patients, half-life extends to about 20 h. et al. 2009), and involuntary movements (Adair et al. 1994), but This is also considered as a cause of side effects. these symptoms are varied and non-specific (Rashiq et al. 1993). It In healthy elderly patients, it was reported that memory function has been reported that an increase in concentration of acyclovir and under amantadine was significantly reduced in comparison to its metabolites (9-carboxymethoxymethylguanine) is involved in the treatment with trihexyphenidyl (McEvoy et al. 1987). Amantadine onset of symptoms (Helldén et al. 2003, 2006). Blood half-life of is excreted in the urine largely unchanged, and accumulates in the

572 Pharmazie 74 (2019) ORIGINAL ARTICLES

Table 3: Signals with poor prognosis of drugs affecting cognitive function in chronic kidney disease patients

ATC Class Drug name case Lift Conv. level 1 A Alimentary tract and metabolism Sitagliptin 3 8.96 1.02

J Antiinfectives for systemic use Valacyclovir 8 5.71 1.01 N Nervous system Pregabalin 3 6.94 1.01 Amantadine 5 33.09 1.07 V Various Nalfurafine 8 23.90 1.05

ATC: Anatomical Therapeutic Chemical (classification system), Conv.: conviction. body when taken by a patient with reduced renal function. More- patients. In the search using ARM, signals (or signal candidates) over, high blood concentrations of amantadine are maintained even were easily obtained from large databases, from which BCPNN after discontinuation. As a result, amantadine use in CKD patients properly detected the signals. may possibly impair cognitive function. Although sufficient attention is required for interpretation of the results Nalfurafine and pregabalin were detected as signals even when obtained from spontaneous reporting, we considered that by avoiding limited to adverse events included in the dementia narrow range the use of drugs related to drug-induced dementia revealed in this report, search, and therefore may be drugs requiring special attention. it is possible to help prevent the onset of dementia in CKD patients. Nalfurafine is used for the treatment of pruritus in hemodialysis Until more detailed clinical research results are reported, the use of these patients. This compound has a particularly high affinity for the drugs requires observation of patient cognitive function over time. opioid k receptor which is involved in hallucinations and delirium, adverse effects that have also been reported in post-marketing surveillance (Kozono et al. 2018). 4. Experimental Pregabalin, unlike benzodiazepines, is said to have no effect on g-am- 4.1. Data source inobutyric acid (GABA) receptors, but is an analogue of GABA, an Authors do not own the data because the Japanese authority, Pharmaceuticals and inhibitory neurotransmitter (Bryans et al. 1999). Therefore, there is Medical Devices Agency (PMDA), does not permit sharing the Japanese Adverse a possibility that central inhibitory action may develop with its use. Drug Event Report database (JADER) directly. Data owned by PMDA can be The use of these drugs requires careful attention, such as observa- accessed directly here: tion of patient cognitive function over time. http://www.info.pmda.go.jp/fukusayoudb/CsvDownload.jsp (only in Japanese). The JADER consists of four comma-separated values (csv) files as data tables: A limitation to this study is that the medical database used is based DEMO.csv (table containing patient information), DRUG.csv (table containing on spontaneous reports and, therefore, is influenced by reporting medicine information), HIST.csv (table including patient past history), and REAC.csv bias (Pariente et al. 2007; Burkey et al. 2008). Such a drawback (adverse event information table). The data sets were created by combining the iden- cannot be solved by changing the analysis method from dispropor- tification numbers assigned to each of the four tables. In this study, data from patients registered in JADER from April 2004 through June 2018 were used. However, reports tionality analysis to ARM. In addition, since data on the grade of with missing information regarding sex, age, or primary disease, and where subjective CKD in patients do not exist on JADER, analysis by grade could not terms such as “youth” and “elderly” were used, were excluded from the analysis. For be performed. Therefore, sufficient attention is required to interpret this reason, 352,372 cases were used for analysis. the result on signals derived from using association analysis. In signal detection using a spontaneous reporting system, the disproportionality measure between the drug registered as a suspected 4.2. Defi nition of primary disease and adverse events drug and the adverse event report is used as a signal. Therefore, We investigated patients with CKD registered as a primary disease. The original diseases registered in JADER are described in the preferred terms (PTs) of the this study cannot clarify the possibility that primary disease is the Medical Dictionary for Regulatory Activities/Japanese version (MedDRA/J). There- cause of adverse events. Additionally, although confounding factors fore, the original diseases extracted were defined as PTs contained in chronic kidney related to dementia vary, the statistical models used in this study disease among standardized MedDRA queries (SMQ). The PTs included in the SMQ cannot take into account anything other than CKD in either model. included “narrow range” when searching for a case highly likely to indicate the state, This problem will need to be clarified in future research. and “wide range” when it was necessary to detect all possible cases. In this study, all 172 PTs of wide range were used in the search. In conclusion, this study is the first report to use a large-scale medical The AE in this study was dementia. Therefore, among SMQ of MedDRA/J, PTs database to identify drugs related to drug-induced dementia in CKD included in dementia (SMQ 20000073) were defined as AEs to be investigated. In the

Fig.: Venn diagram and four-by-two contingency table for signal detection. Pharmazie 74 (2019) 573 ORIGINAL ARTICLES

extraction of AEs, 90 PTs included in the wide range search and only 21 PTs included Abbreviations: Adverse event: AE, Association rule mining: ARM, Bayesian confi- only in narrow range search were used. dence propagation neural network: BCPNN, chronic kidney disease: CKD, cardio- In addition, JADER includes the following outcome information on adverse events: vascular disease: CVD, empirical Bayes geometric mean: EBGM, end stage kidney recovery, remission, no recovery, death, after-effects, and unknown. For PTs included disease: ESKD, γ-aminobutyric acid: GABA, information component: IC, Japanese in the wide range search, signals with descriptions of either no recovery, death and Adverse Drug Event Report: JADER, Medical Dictionary for Regulatory Activities/ after-effects were analyzed as “poor prognosis adverse events”. Japanese version: MedDRA/J, Pharmaceuticals and Medical Devices Agency: PMDA, preferred term: PT, proportional reporting ratio: PRR, reporting odds ratio: ROR, standardized MedDRA queries: SMQ. 4.3. Data analysis Conflicts of interest: Although Laboratory of Community Healthcare Pharmacy, Gifu ARM is a simple analysis method for large databases, and appropriate signals can be Pharmaceutical University, is financially supported by donations from WELCIA detected from many combinations by apriori algorithm. Therefore, in this study, we YAKKYOKU CO., LTD., the authors report no conflicts of interest regarding the first performed prescreening using ARM, and used BCPNN as a verification of the content of this article. obtained signals. However, since the BCPNN cannot create a model for detecting signals of events with Funding: This study was supported by JSPS KAKENHI Grant Number JP16K19175. multiple conditions attached to conclusion: Y like signals related to “poor prognosis adverse events”, We performed on signals related to “poor prognostic dementia” using only ARM. References 4.3.1. Signal detection using association rule mining Adair JC, Gold M, Bond RE (1994) Acyclovir neurotoxicity: clinical experience and review of the literature. South Med J 87: 1227-1231. The calculation of lift and conviction are shown in Eqs. (1) and (2). The number of Agrawal R, Srikant R (1994) Fast algorithms for mining association rules. Proc 20th reports required to calculate the lift (Fig.) is as, n : Of CKD patients, the number of 111 int conf very large databases. 1215: 487–499. cases that developed dementia using suspected drugs, n : Of CKD patients, total 11+ Asahi T, Tsutsui M, Wakasugi M, Tange D, Takahashi C, Tokui K, Okazawa S, number of cases of adverse events using suspected drug, n : Of CKD patients, Total ++1 Okudera H (2009) Valaciclovir neurotoxicity: clinical experience and review of the number of cases that developed dementia. literature. Eur J Neurol 16: 457-460. The “lift” of ARM used for signal detection in this research is an index showing the Bate A, Lindquist M, Edwards IR, Olsson S, Orre R, Lansner A, De Freitas RM relative magnitude of the probability of observing conclusion: Y (= dementia) under (1998) A Bayesian neural network method for adverse drug reaction signal gener- the condition of premise: X (= suspect drug with CKD patients). ation. Eur J Clin Pharmacol 54: 315–321. When the lift is 1.0, the 2 events (premise: X and conclusion: Y) are independent of Bryans JS, Wustrow DJ (1999) 3-Substituted GABA analogs with central nervous each other; when the lift is greater than 1.0, the events are considered dependent, and system activity: A review. Med Res Rev 19: 149–177. the greater the value, the higher the dependence. Burkey BF, Hoffmann PK, Hassiepen U, Trappe J, Juedes M, Foley JE (2008) Adverse effects of dipeptidyl peptidases 8 and 9 inhibition in rodents revisited. Diabetes (1) Obes Metab 10: 1057–1061. Deckers K, Camerino I, van Boxtel MP, Verhey FR, Irving K, Brayne C, Kivipelto M, Starr JM, Yaffe K, de Leeuw PW, Köhler S (2017) Dementia risk in renal dysfunction: A systematic review and meta-analysis of prospective studies. Neurology 88: 198–208. (2) Etgen T, Chonchol M, Förstl H, Sander D (2012) Chronic kidney disease and cognitive impairment: a systematic review and meta-analysis. Am J Nephrol 35: 474–482. Evans SJ, Waller PC, Davis S (2001) Use of proportional reporting ratios (PRRs) for signal generation from spontaneous adverse drug reaction reports. Pharmacoepi- On the other hand, “conviction” is an evaluation indicator, which indicates whether or demiol Drug Saf. 10: 483–486. not the rule does not make a wrong prediction, paying attention to the exclusion event Fujisaki K, Tsuruya K, Yamato M, Toyonaga J, Noguchi H, Nakano T, Taniguchi M, of the conclusion: Y of the obtained rule. If the conviction is large, it is less likely that Tokumoto M, Hirakata H, Kitazono T (2014) Cerebral oxidative stress induces the conclusion: Y is not the case for the premise: X. spatial working memory dysfunction in uremic mice: neuroprotective effect of The apriori algorithm using ARM is based on the principle that “support of a certain tempol. Nephrol Dial Transplant 29: 529–538. item set is always less than or equal to support of its partial item set” (Agrawal and Helldén A, Odar-Cederlöf I, Diener P, Barkholt L, Medin C, Svensson JO, Säwe Srikant 1994). Therefore, unlike BCPNN, in ARM it is not necessary to calculate lift J, Ståhle L (2003) High serum concentrations of the acyclovir main metabolite and conviction of all combinations for signal detection. 9-carboxymethoxymethylguanine in renal failure patients with acyclovir-related In this study, we used visual mining studio version 8.2 (NTT DATA Mathematical neuropsychiatric side effects: an observational study. Nephrol Dial Transplant. 18: Systems Inc, Japan) for ARM. 1135–1141. The detection criterion (case; n ≥ 3, lift > 1, conviction > 1) similar to those used in 111 Helldén A, Lycke J, Vander T, Svensson JO, Odar-Cederlöf I, Ståhle L (2006) The a previous study were used (Noguchi et al. 2018a). aciclovir metabolite CMMG is detectable in the CSF of subjects with neuropsychi- As well, in order to detect the signals related to poor prognosis adverse events, it is atric symptoms during aciclovir and valaciclovir treatment. J Antimicrob Chemo- necessary to set premise: X = suspect drug with CKD patients and conclusion: Y = ther 57: 945–949. poor prognosis dementia. In this paper, the calculation formula based on the number Kozono H, Yoshitani H, Nakano R (2018) Post-marketing surveillance study of the (n) of reports was omitted. safety and efficacy of nalfurafine hydrochloride (Remitch® capsules 2.5 μg) in 3,762 hemodialysis patients with intractable pruritus. Int J Nephrol Renovasc Dis 11: 9–24. 4.3.2. Signal detection using Bayesian confi dence propagation neuralnetwork McEvoy JP, McCue M, Spring B, Mohs RC, Lavori PW, Farr RM (1987) Effects of

BCPNN is an artificial neural network inspired by Bayesian estimations. IC025 is the amantadine and trihexyphenidyl on memory in elderly normal volunteers. Am J lower end of 95% confidence interval for IC. The detection criterion of this statistical Psychiatry 144: 573-577. model is IC025 > 0 similar to those used in a previous study were used (Bate et al. 1998). Noguchi Y, Ueno A, Otsubo M, Katsuno H, Sugita I, Kanematsu Y, Yoshida A, Esaki A table for calculating IC025 which is an indicator of BCPNN considering the CKD H, Tachi T, Teramachi H (2018a) A simple method for exploring adverse drug patient is shown in the Fig. events in patients with different primary diseases using spontaneous reporting system. BMC Bioinform 19: 124. doi: org/10.1186/s12859-018-2137-y. Noguchi Y, Ueno A, Otsubo M, Katsuno H, Sugita I, Kanematsu Y, Yoshida A, Esaki (3) H, Tachi T, Teramachi H (2018b) A new search method using association rule mining for drug-drug interaction based on spontaneous report system. Front. Phar- macol 9: 197. Pariente A, Gregoire F, Fourrier-Reglat A, Haramburu F, Moore N (2007) Impact of safety alerts on measures of disproportionality in spontaneous reporting databases: the notoriety bias. Drug Saf 30: 891–898. Rashiq S, Briew AM, Mooney T, Giancarlo T, Khatib R, Wilson FM (1993) Distin- (4) guishing acyclovir neurotoxicity from encephalomyelitis. J Intern Med 234: 507–511. Rothman KJ, Lanes S, Sacks ST (2004) The reporting odds ratio and its advantages over the proportional reporting ratio. Pharmacoepidemiol Drug Saf 13: 519–523. Seliger SL, Siscovick DS, Stehman-Breen CO, Gillen DL, Fitzpatrick A, Bleyer A, Kuller LH (2004) Moderate renal impairment and risk of dementia among older (5) adults: the Cardiovascular Health Cognition Study. J Am Soc Nephrol 15: 1904– 1911. Szarfman A, Machado SG, O’Neill RT (2002) Use of screening algorithms and (6) 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.

To calculate the IC025 of the CKD patient, replace it as follows: N11 = n111, N1+ = n11+, N+1 = n111 + n011, N++ = n111 + n101+ n011+ n001. In addition, the calculation method of general IC025 is shown in Eqs. (3) to (6) as follows: In this study, the validity of drugs detected in ARM was examined using BCPNN.

574 Pharmazie 74 (2019)