Quantitative Evaluation of Initial Symptoms as Predictors to Detect Adverse Drug Reactions Using Bayes’ Theory: Expansion and Evaluation of Drug-Adverse Drug Reaction–Initial Symptom Combinations Using Adverse Event Reporting System Database

Daisuke Kobayashi; Shigeru Hosaka; Emiko Inoue; Kimie Ohshima; Nobuaki Kutsuma; Shinji Oshima; Yasushi Okuno

doi:10.1248/bpb.b12-01006

Abstract

In prescription dispensing in Japan, to avoid adverse drug reactions (ADR) pharmacists provide patients with information concerning the initial symptoms (IS) of any ADR that might be caused by the drugs they have been prescribed. However, the usefulness of such information for preventing ADR has not been quantitatively evaluated. We previously performed a trial calculation of the usefulness of rash as a predictor of drug-induced liver disorders by applying Bayes’ theorem and showed that the predictive utility of IS can be quantitatively evaluated using likelihood ratios. However, for other drug-ADR-IS combinations it was difficult to obtain the information required for the calculations from Japanese data alone. In this study, using the Adverse Event Reporting System (AERS) database of the U.S. Food and Drug Administration (FDA), we evaluated 132 drug-ADR-IS combinations that were considered to be potentially clinical significant. Regarding bezafibrate-associated rhabdomyolysis and cibenzoline-associated hypoglycemia, these ADR were not detected in cases involving monotherapy. For 58 combinations, no events that were considered to be IS of the target ADR developed. Fever, nausea, and decreased appetite were the IS of many ADR, making them very useful predictors. In contrast, pruritus and rash were not very useful. Fever might be a predictor of thiamazole-induced agranulocytosis or levofloxacin- or terbinafine-induced liver disorder, tremors might be useful for predicting paroxetine-induced serotonin syndrome, and decreased appetite might be a useful indicator of terbinafine-induced liver dysfunction.

In Japan, when pharmacists dispense drugs to patients they provide information about any symptoms that the prescribed medication might induce that should cause the patient to consult a physician in order to avoid adverse drug reactions (ADR). Specifically, article 25-2 of the Pharmacists Act (1997) obliges pharmacists to describe to patients the initial symptoms (IS) of any ADR that might be induced by the drugs they have been prescribed.

Since the information provided about the IS of ADR has to be easy for patients to understand, less specific symptoms, mainly changes in their physical condition, such as fever, rash, or muscular pain, are described. However, it is doubtful whether such information is useful for detecting ADR.

In our previous report, we performed a trial calculation of the usefulness of rash as an IS of liver disorder; i.e., as an ADR predictor, using Bayes’ theorem.¹⁾ For voriconazole, the prior probability of liver disorder, 0.05, was increased to 0.2 when a rash was present, whereas liver disorders only developed in 2 of 1000 patients who developed a rash after taking loxoprofen sodium, showing that the value of an IS as a predictor varies depending on the drug being taken.

Bayes’ theorem is widely used to evaluate the predictability of a single or a combination of multiple diagnoses.²⁾ However, to use Bayes’ theorem to assess the utility of IS of ADR as diagnostic predictors, it is necessary to identify the probability of a specific symptom accompanying a particular ADR, and we were only able to determine these probabilities for a small number of drugs using Japanese data in our previous study.

The Adverse Event Reporting System (AERS) database, which is run by the U.S. Food and Drug Administration (FDA), contains over four million reports of adverse events (AE). It relies on reports of spontaneous adverse events being submitted to the FDA by health professionals, consumers, and manufacturers. In this study, we used this database in order to markedly increase the number of drug-ADR-IS combinations involving commonly used drugs for which the above mentioned calculations could be performed. In addition, the results of the calculations provided us with some interesting insights.

THEORY

Calculation of Posterior Probability

When an IS (I) that is considered to be a predictor of a particular ADR (A) develops after a certain drug (D) is taken, the probability of A actually developing (P_post) is given by the equation below:

(1)

(2)

where: P_post: Incidence of A accompanied by I, P_pre: Incidence of A, P_other: Incidence of ADR other than A, P_non: Probability of no ADR developing, Q_pre: Probability of D-induced I accompanying A, Q_other: Probability of D-induced I accompanying an ADR other than A, Q_non: Incidence of I-like symptoms that are not drug-related, Q_t: Total incidence of D-induced I in clinical trials

The addition of Q_non to Q_pre and Q_other in Eq. 1 assumes that the incidence of I-like symptoms that are not drug-related is not included in Q_pre or Q_other. ADR are defined as reactions for which an association with a particular drug cannot be ruled out, and IS are mild ADR. Accordingly, it can be assumed that Q_non is not included in Q_pre or Q_other.

Q_pre and Q_other were calculated using the AERS data. Since the AERS data are comprised of AE reports, Q_pre and Q_other were considered to include Q_non. However, since we calculated Q_pre and Q_other from the data of drugs used alone. Therefore, it is strongly assumed that reporters of AE considered the drug to be the primary suspect and reported the event as an ADR. Thus, in the calculation of P_post, we made 2 calculations: Q_non was added to Q_pre and Q_other in one (CalQ+) and not added in the other (CalQ0), broadening the range of interpretation of the results. The P_post values produced using CalQ+ and CalQ0 are presented as P_postQ+ and P_postQ0, respectively.

Equation 2 is only applicable when the total number of individuals who took a particular medication is known, as the latter figure is required Q_t. Since the AERS data did not describe the number of patients who took each drug without developing AE, we did not use Eq. 2 in this study.

Calculation of Likelihood Ratios

The positive likelihood ratio was calculated using Eq. 3.

(3)

where: LR+: Likelihood ratio, Od_pre: Prior odds, Od_post: Posterior odds

LR+ values that were calculated using P_postQ+ and P_postQ0 as P_post were designated as LR+Q+and LR+Q0, respectively.

METHODS

Incidence of IS-Like Symptoms That Were Not Caused by the Target Drug (Q_non)

The IS of ADR, such as fever, headache, and muscular pain, are also experienced as changes in physical condition that are not caused by drugs in daily life. Of these changes in physical condition, those that occur without an obvious cause, such as a cold or excess exercise, need to be differentiated from the IS of ADR, and knowledge about their incidence is required to calculate the frequency of non-specific background IS-like symptoms; i.e., Q_non in Eq. 1. Thus, we first investigated Q_non.

Selection of IS for the Q_non Investigation

The examined IS were selected from “the Manual for Handling Disorders due to Adverse Drug Reactions” (MHD).³⁾ The MHD was published by the Ministry of Health, Labour, and Welfare and was produced in cooperation with academics in order to develop a strategy for dealing with disorders caused by ADR. In particular, the MHD aims to prevent severe ADR being overlooked during their early stages.⁴⁾ So, we selected IS that exhibited high incidences from the MHD for our investigation of Q_non.

Survey of Q_non

Q_non was investigated using an internet-based questionnaire developed by Rakuten Research (Rakuten Research, Inc., Shinagawa, Tokyo, Japan). Nine hundred males and 900 females belonging to 6 decile age groups, ranging from subjects in their 20s to patients in their 70s (150 males and 150 females in each age group), were surveyed about any changes in their physical condition of unknown cause that had occurred during the last 3 months. The survey period was set at 3 months because many ADR develop within 3 months of the initiation of drug treatment.⁵⁾ We have reported the findings of this survey previously.¹⁾ The results are shown together with the incidence of urticaria-like changes in physical condition in Table 1.

Table 1. Incidence Rates of Changes in Body Condition

ID	Symptom	Incidence	ID	Symptom	Incidence
1	Fatigue	0.1644	27	Increased blood pressure	0.0494
2	Pruritus	0.1450	28	Dry cough	0.0494
3	Malaise	0.1339	29	Edema	0.0439
4	Headache	0.1250	30	Nausea	0.0428
5	Diarrhea	0.1161	31	Anemia	0.0417
6	Arthralgia	0.1139	32	Musculoskeletal stiffness	0.0411
7	Tinnitus	0.1000	33	Hyperhidrosis	0.0406
8	Abdominal distensiom	0.0961	34	Erythema	0.0389
9	Constipation	0.0883	35	Wet cough	0.0389
10	Somnolence	0.0872	36	Decreased appetite	0.0378
11	Dizziness	0.0861	37	Pyrexia	0.0350
12	Rash	0.0833	38	Facial edema	0.0294
13	Insomnia	0.0800	39	Tachycardia	0.0289
14	Pollakiuria	0.0783	40	Dysponea	0.0250
15	Oropharyngeal pain	0.0783	41	Bruising	0.0239
16	Abdominal pain	0.0761	42	Black stools	0.0228
17	Abdominal discomfort	0.0700	43	Dyslalia	0.0206
18	Urticaria	0.0614	44	Asthenia	0.0200
19	Palpitations	0.0606	45	Tremors	0.0194
20	Muscle ache	0.0572	46	Oliguria	0.0189
21	Epigastrial pain	0.0561	47	Vomiting	0.0189
22	Hemorrhage(*)	0.0556	48	Wheezing	0.0161
23	Thirst	0.0544	49	Ageusia	0.0150
24	Heart burn	0.0528	50	Oral hypoesthesia	0.0078
25	Chest pain	0.0517	51	Hematuria	0.0078
26	Hypesthesia	0.0517	51	Hematuria	0.0078

(*) Subcutaneous hemorrhaging+gingival bleeding+epistaxis.

Selection of Drug-ADR-IS Combinations

On the condition that one of the target IS examined in the investigation of Q_non was included in each combination, 3 pharmacists selected drug-ADR-IS combinations from the MHD from the viewpoint of clinical importance about patient compliance instruction. That is, the process used to select the drug-ADR-IS combinations was as follows: First, IS that were frequently mentioned in the MHD were extracted. Then, we identified ADR involving these IS. Finally, we extracted drugs associated with these ADR.

The terms used to describe the ADR and IS were standardized to the Preferred Terms outlined in the Japanese edition of the ICH Medical Dictionary for Regulatory Activities Terminology (MedDR/J).

Data Collection. P_pre, P_other, and P_non

The probabilities of the target ADR (P_pre), ADR other than the target (P_other), and no ADR developing (P_non) were adopted from the “ADR frequencies classified by item and a list of abnormal clinical laboratory test results” (a list of ADR) section of the post-marketing surveillance (PMS) data on each drug’s interview form (IF). The IF of brand name drugs were used.

Q_pre and Q_other

Using the AERS data, the total number of cases of ADR, the number of cases of the target ADR, and the number of cases of the target ADR that were accompanied by the target IS were investigated for each drug used alone, and Q_pre and Q_other were determined.

Statistical Analysis

The difference between P_pre and P_post was analyzed arbitrarily by setting the population size to that of the PMS.

RESULTS

Drug-ADR–IS (D-A-I) Combinations and the Results of the AERS Survey

The target D-A-I combinations are shown in Table 2. “★” represents IS that were not found in the AERS database.

Table 2. Selected Drug-ADR-IS Combinations and Results of an AERS Database Search

ID	Drg	AR	IS	Drug-ADR-IS, abbreviation
1	Acarbose	Abnormal hepatic function	Decreased appetite	Ac-Hf-Da
2	Acarbose	Abnormal hepatic function	Malaise★	Ac-Hf-Ml
3	Acarbose	Abnormal hepatic function	Nausea★	Ac-Hf-Na
4	Acarbose	Abnormal hepatic function	Pruritus★	Ac-Hf-Pr
5	Acarbose	Abnormal hepatic function	Pyrexia★	Ac-Hf-Py
6	Acarbose	Abnormal hepatic function	Rash★	Ac-Hf-Ra
7	Acarbose	Abnormal hepatic function	Vomiting★	Ac-Hf-Vo
8	Acarbose	Liver disorder	Decreased appetite★	Ac-Ld-Da
9	Acarbose	Liver disorder	Malaise★	Ac-Ld-Ml
10	Acarbose	Liver disorder	Nausea★	Ac-Ld-Na
11	Acarbose	Liver disorder	Pruritus★	Ac-Ld-Pr
12	Acarbose	Liver disorder	Pyrexia★	Ac-Ld-Py
13	Acarbose	Liver disorder	Rash★	Ac-Ld-Ra
14	Acarbose	Liver disorder	Vomiting★	Ac-Ld-Vo
15	Atorvastatin calcium hydrate	Abnormal hepatic function	Decreased appetite	At-Hf-Da
16	Atorvastatin calcium hydrate	Abnormal hepatic function	Malaise	At-Hf-Ml
17	Atorvastatin calcium hydrate	Abnormal hepatic function	Nausea	At-Hf-Na
18	Atorvastatin calcium hydrate	Abnormal hepatic function	Pruritus	At-Hf-Pr
19	Atorvastatin calcium hydrate	Abnormal hepatic function	Pyrexia	At-Hf-Py
20	Atorvastatin calcium hydrate	Abnormal hepatic function	Rash	At-Hf-Ra
21	Atorvastatin calcium hydrate	Abnormal hepatic function	Vomiting	At-Hf-Vo
22	Atorvastatin calcium hydrate	Liver disorder	Decreased appetite	At-Ld-Da
23	Atorvastatin calcium hydrate	Liver disorder	Malaise	At-Ld-Ml
24	Atorvastatin calcium hydrate	Liver disorder	Nausea	At-Ld-Na
25	Atorvastatin calcium hydrate	Liver disorder	Pruritus★	At-Ld-Pr
26	Atorvastatin calcium hydrate	Liver disorder	Pyrexia	At-Ld-Py
27	Atorvastatin calcium hydrate	Liver disorder	Rash★	At-Ld-Ra
28	Atorvastatin calcium hydrate	Liver disorder	Vomiting	At-Ld-Vo
29	Bezafibrate	Rhabdomyolysis★★	Asthenia	Be-Rh-As
30	Bezafibrate	Rhabdomyolysis★★	Hematuria	Be-Rh-Hu
31	Bezafibrate	Rhabdomyolysis★★	Malaise	Be-Rh-Ml
32	Bezafibrate	Rhabdomyolysis★★	Myalgia	Be-Rh-My
33	Cibenzoline succinate	Hypoglycemia★★	Asthenia	Ci-Hy-As
34	Cibenzoline succinate	Hypoglycemia★★	Dizziness	Ci-Hy-Dz
35	Cibenzoline succinate	Hypoglycemia★★	Dyslalia	Ci-Hy-Dl
36	Cibenzoline succinate	Hypoglycemia★★	Headache	Ci-Hy-Hd
37	Cibenzoline succinate	Hypoglycemia★★	Nausea	Ci-Hy-Na
38	Cibenzoline succinate	Hypoglycemia★★	Palpitations	Ci-Hy-Pa
39	Cibenzoline succinate	Hypoglycemia★★	Somnolence	Ci-Hy-So
40	Cibenzoline succinate	Hypoglycemia★★	Tremors	Ci-Hy-Tr
41	Famotidine	Neutropenia	Oropharyngeal pain★	Fa-Np-Or
42	Famotidine	Neutropenia	Pyrexia★	Fa-Np-Py
43	Glimepiride	Hypoglycemia	Asthenia	Gl-Hy-As
44	Glimepiride	Hypoglycemia	Dizziness	Gl-Hy-Dz
45	Glimepiride	Hypoglycemia	Dyslalia★	Gl-Hy-Dl
46	Glimepiride	Hypoglycemia	Headache★	Gl-Hy-Hd
47	Glimepiride	Hypoglycemia	Nausea★	Gl-Hy-Na
48	Glimepiride	Hypoglycemia	Palpitations★	Gl-Hy-Pa
49	Glimepiride	Hypoglycemia	Somnolence★	Gl-Hy-So
50	Glimepiride	Hypoglycemia	Tremors	Gl-Hy-Tr
51	Levofloxacin hydrate	Anaphylactoid reaction	Dyspnoea	Le-Ar-Dn
52	Levofloxacin hydrate	Anaphylactoid reaction	Palpitations★	Le-Ar-Pa
53	Levofloxacin hydrate	Anaphylactoid reaction	Pruritus★	Le-Ar-Pr
54	Levofloxacin hydrate	Anaphylactoid reaction	Urticaria	Le-Ar-Ur
55	Levofloxacin hydrate	Anaphylactoid shock	Dyspnoea	Le-As-Dn
56	Levofloxacin hydrate	Anaphylactoid shock	Palpitations★	Le-As-Pa
57	Levofloxacin hydrate	Anaphylactoid shock	Pruritus★	Le-As-Pr
58	Levofloxacin hydrate	Anaphylactoid shock	Urticaria	Le-As-Ur
59	Levofloxacin hydrate	Abnormal hepatic function	Decreased appetite	Le-Hf-Da
60	Levofloxacin hydrate	Abnormal hepatic function	Malaise	Le-Hf-Ml
61	Levofloxacin hydrate	Abnormal hepatic function	Nausea★	Le-Hf-Na
62	Levofloxacin hydrate	Abnormal hepatic function	Pruritus★	Le-Hf-Pr
63	Levofloxacin hydrate	Abnormal hepatic function	Pyrexia	Le-Hf-Py
64	Levofloxacin hydrate	Abnormal hepatic function	Rash	Le-Hf-Ra
65	Levofloxacin hydrate	Abnormal hepatic function	Vomiting★	Le-Hf-Vo
66	Levofloxacin hydrate	Liver disorder	Decreased appetite★	Le-Ld-Da
67	Levofloxacin hydrate	Liver disorder	Malaise	Le-Ld-Ml
68	Levofloxacin hydrate	Liver disorder	Nausea	Le-Ld-Na
69	Levofloxacin hydrate	Liver disorder	Pruritus★	Le-Ld-Pr
70	Levofloxacin hydrate	Liver disorder	Pyrexia	Le-Ld-Py
71	Levofloxacin hydrate	Liver disorder	Rash★	Le-Ld-Ra
72	Levofloxacin hydrate	Liver disorder	Vomiting	Le-Ld-Vo
73	Paroxetine hydrochloride hydrate	Serotonin syndrome	Diarrhea	Pa-Ss-Dr
74	Paroxetine hydrochloride hydrate	Serotonin syndrome	Hyperhidrosis	Pa-Ss-Hh
75	Paroxetine hydrochloride hydrate	Serotonin syndrome	Pyrexia	Pa-Ss-Py
76	Paroxetine hydrochloride hydrate	Serotonin syndrome	Tachycardia	Pa-Ss-Ta
77	Paroxetine hydrochloride hydrate	Serotonin syndrome	Tremors	Pa-Ss-Tr
78	Pioglitazone hydrochloride	Congestive heart failure	Dyspnoea	Pi-Cf-Dn
79	Pioglitazone hydrochloride	Congestive heart failure	Fatigue	Pi-Cf-Fg
80	Pioglitazone hydrochloride	Congestive heart failure	Edema	Pi-Cf-Oe
81	Raloxifene hydrochloride	Deep vein thrombosis	Chest pain	Ra-Dt-Cp
82	Raloxifene hydrochloride	Deep vein thrombosis	Dyslalia★	Ra-Dt-Dl
83	Raloxifene hydrochloride	Deep vein thrombosis	Dyspnoea	Ra-Dt-Dn
84	Raloxifene hydrochloride	Venous thrombosis	Chest pain★	Ra-Vt-Cp
85	Raloxifene hydrochloride	Venous thrombosis	Dyslalia★	Ra-Vt-Dl
86	Raloxifene hydrochloride	Venous thrombosis	Dyspnoea★	Ra-Vt-Dn
87	Risperidone	Neuroleptic malignant syndrome	Increased blood pressure	Ri-Nm-Bi
88	Risperidone	Neuroleptic malignant syndrome	Dyslalia	Ri-Nm-Dl
89	Risperidone	Neuroleptic malignant syndrome	Hyperhidrosis	Ri-Nm-Hh
90	Risperidone	Neuroleptic malignant syndrome	Musculoskeletal stiffness	Ri-Nm-Ms
91	Risperidone	Neuroleptic malignant syndrome	Pyrexia	Ri-Nm-Py
92	Risperidone	Neuroleptic malignant syndrome	Tachycardia	Ri-Nm-Ta
93	Risperidone	Neuroleptic malignant syndrome	Tremors	Ri-Nm-Tr
94	Terbinafine hydrochloride	Abnormal hepatic function	Decreased appetite	Te-Hf-Da
95	Terbinafine hydrochloride	Abnormal hepatic function	Malaise	Te-Hf-Ml
96	Terbinafine hydrochloride	Abnormal hepatic function	Nausea	Te-Hf-Na
97	Terbinafine hydrochloride	Abnormal hepatic function	Pruritus	Te-Hf-Pr
98	Terbinafine hydrochloride	Abnormal hepatic function	Pyrexia	Te-Hf-Py
99	Terbinafine hydrochloride	Abnormal hepatic function	Rash★	Te-Hf-Ra
100	Terbinafine hydrochloride	Abnormal hepatic function	Vomiting	Te-Hf-Vo
101	Terbinafine hydrochloride	Liver disorder	Decreased appetite	Te-Ld-Da
102	Terbinafine hydrochloride	Liver disorder	Malaise	Te-Ld-Ml
103	Terbinafine hydrochloride	Liver disorder	Nausea	Te-Ld-Na
104	Terbinafine hydrochloride	Liver disorder	Pruritus	Te-Ld-Pr
105	Terbinafine hydrochloride	Liver disorder	Pyrexia	Te-Ld-Py
106	Terbinafine hydrochloride	Liver disorder	Rash	Te-Ld-Ra
107	Terbinafine hydrochloride	Liver disorder	Vomiting	Te-Ld-Vo
108	Thiamazole	Agranulocytosis	Oropharyngeal pain	Th-Ag-Or
109	Thiamazole	Agranulocytosis	Pyrexia	Th-Ag-Py
110	Ticlopidine hydrochloride	Anemia	Fatigue★	Ti-An-Fg
111	Ticlopidine hydrochloride	Anemia	Headache	Ti-An-Hd
112	Ticlopidine hydrochloride	Anemia	Malaise★	Ti-An-Ml
113	Ticlopidine hydrochloride	Anemia	Palpitations★	Ti-An-Pa
114	Ticlopidine hydrochloride	Cerebral hemorrhage	Hemorrhage(*)★	Ti-Ch-Hr
115	Ticlopidine hydrochloride	Granulocytopenia	Oropharyngeal pain★	Ti-Gp-Or
116	Ticlopidine hydrochloride	Granulocytopenia	Pyrexia★	Ti-Gp-Py
117	Ticlopidine hydrochloride	Liver disorder	Decreased appetite★	Ti-Ld-Da
118	Ticlopidine hydrochloride	Liver disorder	Malaise★	Ti-Ld-Ml
119	Ticlopidine hydrochloride	Liver disorder	Nausea★	Ti-Ld-Na
120	Ticlopidine hydrochloride	Liver disorder	Pruritus★	Ti-Ld-Pr
121	Ticlopidine hydrochloride	Liver disorder	Pyrexia	Ti-Ld-Py
122	Ticlopidine hydrochloride	Liver disorder	Rash★	Ti-Ld-Ra
123	Ticlopidine hydrochloride	Liver disorder	Vomiting★	Ti-Ld-Vo
124	Ticlopidine hydrochloride	Pancytopenia	Dizziness★	Ti-Pp-Dz
125	Ticlopidine hydrochloride	Pancytopenia	Dyspnoea★	Ti-Pp-Dn
126	Ticlopidine hydrochloride	Pancytopenia	Fatigue★	Ti-Pp-Fg
127	Ticlopidine hydrochloride	Pancytopenia	Hematuria★	Ti-Pp-Hu
128	Ticlopidine hydrochloride	Pancytopenia	Hemorrhage(*)★	Ti-Pp-Hr
129	Ticlopidine hydrochloride	Pancytopenia	Oropharyngeal pain★	Ti-Pp-Or
130	Ticlopidine hydrochloride	Pancytopenia	Palpitations★	Ti-Pp-Pa
131	Ticlopidine hydrochloride	Pancytopenia	Pyrexia★	Ti-Pp-Py
132	Ticlopidine hydrochloride	Thrombocytopenia	Hemorrhage(*)★	Ti-Tp-Hr

(*) Subcutaneous hemorrhaging+gingival bleeding+epistaxis. ★ This IS was not reported. ★★ This ADR was not reported in cases involving monotherapy.

Regarding bezafibrate-associatefd rhabdomyolysis and cibenzoline-associated hypoglycemia, the ADR were not induced by these drugs alone. Therefore, they have been marked with “★★.”

Sixty-two of the 132 combinations were not marked with “★” or “★★” and were subjected to the calculation.

CalQ0 and CalQ+

The results of the survey and the calculations for the 62 combinations are shown in Table 3.

Table 3. Calculations and Survey Results for 62 Drug-ADR-IS Combinations

Rank	ID	Drug-ADR-IS abbreviation	PMS			AERS (single drug therapy)				Q_pre	Q_non	P_pre	CalQ0			CalQ+
Rank	ID	Drug-ADR-IS abbreviation	Surveyed cases ①	Total ADR cases ②	Target ADR cases ③	Total ADR/AE cases ④	Target ADR/AE cases ⑤	Total ADR/AE cases with IS ⑥	Target ADR/AE cases with IS ⑦	Q_pre	Q_non	P_pre	P_postQ0	Sig. level vs. P_pre (※1)	LR+Q0	P_postQ+	Sig. level vs. P_pre (※1)	LR+Q+
1	121	Ti-Ld-Py	6813	461	1	1609	1	44	1	1.0000000	0.0350000	0.0001468	0.0042441	****	29.03	—	—	—
2	109	Th-Ag-Py	732	82	5	1318	23	37	11	0.4782609	0.0350000	0.0068306	0.0896047	****	14.31	0.0868249	****	13.82
3	1	Ac-Hf-Da	3634	799	2	379	3	3	1	0.3333333	0.0377778	0.0005504	0.0059521	****	10.87	0.0052199	***	9.53
4	77	Pa-Ss-Tr	5561	1201	2	76523	103	1098	18	0.1747573	0.0194444	0.0003596	0.0034242	***	9.55	0.0030967	***	8.63
5	70	Le-Ld-Py	29880	482	8	28536	13	129	3	0.2307692	0.0350000	0.0002677	0.0017874	****	6.69	0.0020254	****	7.58
6	105	Te-Ld-Py	6929	825	23	11153	56	130	11	0.1964286	0.0350000	0.0033194	0.0199237	****	6.10	0.0208222	****	6.39
7	94	Te-Hf-Da	6929	825	62	11153	46	102	8	0.1739130	0.0377778	0.0089479	0.0435071	****	5.04	0.0470420	****	5.47
8	78	Pi-Cf-Dn	3421	556	1	4263	135	98	16	0.1185185	0.0250000	0.0002923	0.0014319	NS	4.90	0.0014847	NS	5.09
9	98	Te-Hf-Py	6929	825	62	11153	46	130	7	0.1521739	0.0350000	0.0089479	0.0407507	****	4.71	0.0445653	****	5.17
10	76	Pa-Ss-Ta	5561	1201	2	76523	103	60	10	0.0970874	0.0288889	0.0003596	0.0015297	*	4.26	0.0015588	*	4.34
11	72	Le-Ld-Vo	29880	482	8	28536	13	138	1	0.0769231	0.0188889	0.0002677	0.0011025	****	4.12	0.0013511	****	5.05
12	75	Pa-Ss-Py	5561	1201	2	76523	103	167	10	0.0970874	0.0350000	0.0003596	0.0012507	NS	3.48	0.0013390	NS	3.73
13	91	Ri-Nm-Py	3902	1025	7	17201	122	78	11	0.0901639	0.0350000	0.0017940	0.0059927	**	3.35	0.0062052	**	3.47
14	103	Te-Ld-Na	6929	825	23	11153	56	152	7	0.1250000	0.0427778	0.0033194	0.0104747	****	3.18	0.0124576	****	3.79
15	107	Te-Ld-Vo	6929	825	23	11153	56	65	3	0.0535714	0.0188889	0.0033194	0.0101821	****	3.09	0.0122010	****	3.71
16	74	Pa-Ss-Hh	5561	1201	2	76523	103	1025	11	0.1067961	0.0405556	0.0003596	0.0011070	NS	3.08	0.0012195	NS	3.39
17	21	At-Hf-Vo	4805	576	68	32932	38	80	2	0.0526316	0.0188889	0.0141519	0.0422720	****	3.07	0.0509018	****	3.74
18	63	Le-Hf-Py	29880	482	14	28536	20	129	2	0.1000000	0.0350000	0.0004685	0.0013560	***	2.90	0.0018012	****	3.85
19	80	Pi-Cf-Oe	3421	556	1	4263	135	106	15	0.1111111	0.0438889	0.0002923	0.0008046	NS	2.75	0.0009539	NS	3.27
20	100	Te-Hf-Vo	6929	825	62	11153	46	65	2	0.0434783	0.0188889	0.0089479	0.0220375	****	2.50	0.0280394	****	3.20
21	102	Te-Ld-Ml	6929	825	23	11153	56	97	16	0.2857143	0.1338889	0.0033194	0.0079204	***	2.40	0.0102653	****	3.11
22	101	Te-Ld-Da	6929	825	23	11153	56	102	4	0.0714286	0.0377778	0.0033194	0.0068647	**	2.08	0.0092860	****	2.81
23	108	Th-Ag-Or	732	82	5	1318	23	16	3	0.1304348	0.0783333	0.0068306	0.0124599	NS	1.83	0.0177629	NS	2.63
24	68	Le-Ld-Na	29880	482	8	28536	13	289	1	0.0769231	0.0427778	0.0002677	0.0004872	NS	1.82	0.0007460	*	2.79
25	96	Te-Hf-Na	6929	825	62	11153	46	152	3	0.0652174	0.0427778	0.0089479	0.0146825	**	1.65	0.0215513	****	2.44
26	93	Ri-Nm-Tr	3902	1025	7	17201	122	121	3	0.0245902	0.0194444	0.0017940	0.0027259	NS	1.52	0.0037103	NS	2.07
27	95	Te-Hf-Ml	6929	825	62	11153	46	97	8	0.1739130	0.1338889	0.0089479	0.0129264	*	1.45	0.0202027	****	2.28
28	51	Le-Ar-Dn	29880	482	1	28536	57	313	2	0.0350877	0.0250000	0.0000335	0.0000474	NS	1.42	0.0000799	NS	2.39
29	24	At-Ld-Na	4805	576	11	32932	131	202	7	0.0534351	0.0427778	0.0022893	0.0031797	NS	1.39	0.0050519	*	2.21
30	59	Le-Hf-Da	29880	482	14	28536	20	99	1	0.0500000	0.0377778	0.0004685	0.0006290	NS	1.34	0.0010864	**	2.32
31	89	Ri-Nm-Hh	3902	1025	7	17201	122	47	5	0.0409836	0.0405556	0.0017940	0.0024013	NS	1.34	0.0035443	NS	1.98
32	92	Ri-Nm-Ta	3902	1025	7	17201	122	105	3	0.0245902	0.0288889	0.0017940	0.0019262	NS	1.07	0.0031465	NS	1.76
33	83	Ra-Dt-Dn	6967	776	7	6770	137	51	3	0.0218978	0.0250000	0.0010047	0.0009551	NS	0.95	0.0018249	NS	1.82
34	73	Pa-Ss-Dr	5561	1201	2	76523	103	634	8	0.0776699	0.1161111	0.0003596	0.0003009	NS	0.84	0.0005911	NS	1.64
35	19	At-Hf-Py	4805	576	68	32932	38	97	1	0.0263158	0.0350000	0.0141519	0.0118283	NS	0.83	0.0243193	***	1.74
36	15	At-Hf-Da	4805	576	68	32932	38	50	1	0.0263158	0.0377778	0.0141519	0.0110251	NS	0.78	0.0236779	***	1.69
37	50	Gl-Hy-Tr	3409	146	49	1095	68	5	1	0.0147059	0.0194444	0.0143737	0.0111640	NS	0.77	0.0248331	**	1.75
38	43	Gl-Hy-As	3409	146	49	1095	68	11	1	0.0147059	0.0200000	0.0143737	0.0107671	NS	0.75	0.0243480	**	1.71
39	55	Le-As-Dn	29880	482	1	28536	55	313	1	0.0181818	0.0250000	0.0000335	0.0000246	NS	0.73	0.0000574	NS	1.72
40	106	Te-Ld-Ra	6929	825	23	11153	56	144	3	0.0535714	0.0833333	0.0033194	0.0023691	NS	0.71	0.0053475	*	1.61
41	17	At-Hf-Na	4805	576	68	32932	38	202	1	0.0263158	0.0427778	0.0141519	0.0096311	*	0.68	0.0223263	**	1.59
42	111	Ti-An-Hd	6813	461	13	1609	14	11	1	0.0714286	0.1250000	0.0019081	0.0011640	NS	0.61	0.0029854	NS	1.57
43	64	Le-Hf-Ra	29880	482	14	28536	20	214	1	0.0500000	0.0833333	0.0004685	0.0002852	NS	0.61	0.0007484	NS	1.60
44	67	Le-Ld-Ml	29880	482	8	28536	13	132	1	0.0769231	0.1338889	0.0002677	0.0001562	NS	0.58	0.0004213	NS	1.57
45	54	Le-Ar-Ur	29880	482	1	28536	57	176	2	0.0350877	0.0614192	0.0000335	0.0000194	NS	0.58	0.0000525	NS	1.57
46	88	Ri-Nm-Dl	3902	1025	7	17201	122	1	1	0.0081967	0.0205556	0.0017940	0.0009693	NS	0.54	0.0025075	NS	1.40
47	97	Te-Hf-Pr	6929	825	62	11153	46	204	3	0.0652174	0.1450000	0.0089479	0.0044782	**	0.50	0.0127460	*	1.43
48	26	At-Ld-Py	4805	576	11	32932	131	97	2	0.0152672	0.0350000	0.0022893	0.0011209	NS	0.49	0.0032530	NS	1.42
49	28	At-Ld-Vo	4805	576	11	32932	131	80	1	0.0076336	0.0188889	0.0022893	0.0010325	NS	0.45	0.0031641	NS	1.38
50	23	At-Ld-Ml	4805	576	11	32932	131	195	7	0.0534351	0.1338889	0.0022893	0.0010311	NS	0.45	0.0031840	NS	1.39
51	87	Ri-Nm-Bi	3902	1025	7	17201	122	16	2	0.0163934	0.0494444	0.0017940	0.0008014	NS	0.45	0.0023770	NS	1.33
52	16	At-Hf-Ml	4805	576	68	32932	38	195	2	0.0526316	0.1338889	0.0141519	0.0062484	***	0.44	0.0195160	*	1.39
53	18	At-Hf-Pr	4805	576	68	32932	38	104	2	0.0526316	0.1450000	0.0141519	0.0057878	****	0.41	0.0191471	NS	1.36
54	90	Ri-Nm-Ms	3902	1025	7	17201	122	45	2	0.0163934	0.0572222	0.0017940	0.0006859	NS	0.38	0.0022805	NS	1.27
55	60	Le-Hf-Ml	29880	482	14	28536	20	132	1	0.0500000	0.1338889	0.0004685	0.0001777	*	0.38	0.0006431	NS	1.37
56	20	At-Hf-Ra	4805	576	68	32932	38	125	1	0.0263158	0.0833333	0.0141519	0.0050249	****	0.35	0.0184503	NS	1.31
57	44	Gl-Hy-Dz	3409	146	49	1095	68	8	2	0.0294118	0.0861111	0.0143737	0.0050927	****	0.35	0.0191521	NS	1.34
58	81	Ra-Dt-Cp	6967	776	7	6770	137	47	2	0.0145985	0.0516667	0.0010047	0.0003142	NS	0.31	0.0012699	NS	1.26
59	58	Le-As-Ur	29880	482	1	28536	55	176	1	0.0181818	0.0614192	0.0000335	0.0000101	NS	0.30	0.0000433	NS	1.29
60	104	Te-Ld-Pr	6929	825	23	11153	56	204	2	0.0357143	0.1450000	0.0033194	0.0009122	**	0.27	0.0040744	NS	1.23
61	22	At-Ld-Da	4805	576	11	32932	131	50	1	0.0076336	0.0377778	0.0022893	0.0005226	*	0.23	0.0027379	NS	1.20
62	79	Pi-Cf-Fg	3421	556	1	4263	135	55	2	0.0148148	0.1644444	0.0002923	0.0000310	NS	0.11	0.0003147	NS	1.08

※1 **** p<0.0001, *** 0.0001<p<0.001, ** 0.001<p<0.01, * 0.01<p<0.05. P_pre =③/① P_other=(②−③)/① P_non=1−P_pre−P_other Q_pre=⑦/⑤ Q_other=(⑥−⑦)/(④−⑤)

A simplified list of the LR+Q0 and LR+Q+ calculation results, as well as drug names and fully spelt out descriptions of the ADR/AE and IS are shown in Table 4.

Table 4. Summary of Table 3

Rank	ID	Drug	ADR/AE	IS	LR+Q0	LR+Q+	Rank	ID	Drug	ADR/AE	IS	LR+Q0	LR+Q+
1	121	Ticlopidine	Liver disorder	Pyrexia	29.03	—	32	92	Risperidone	Neuroleptic malignant syndrome	Tachycardia	1.07	1.76
2	109	Thiamazole	Agranulocytosis	Pyrexia	14.31	13.82	33	83	Raloxifene	Deep vein thrombosis	Dyspnoea	0.95	1.82
3	1	Acarbose	Abnormal hepatic function	Decreased appetite	10.87	9.53	34	73	Paroxetine	Serotonin syndrome	Diarrhea	0.84	1.64
4	77	Paroxetine	Serotonin syndrome	Tremors	9.55	8.63	35	19	Atorvastatin	Abnormal hepatic function	Pyrexia	0.83	1.74
5	70	Levofloxacin	Liver disorder	Pyrexia	6.69	7.58	36	15	Atorvastatin	Abnormal hepatic function	Decreased appetite	0.78	1.69
6	105	Terbinafine	Liver disorder	Pyrexia	6.10	6.39	37	50	Glimepiride	Hypoglycemia	Tremors	0.77	1.75
7	94	Terbinafine	Abnormal hepatic function	Decreased appetite	5.04	5.47	38	43	Glimepiride	Hypoglycemia	Asthenia	0.75	1.71
8	78	Pioglitazone	Congestive heart failure	Dyspnoea	4.90	5.09	39	55	Levofloxacin	Anaphylactoid shock	Dyspnoea	0.73	1.72
9	98	Terbinafine	Abnormal hepatic function	Pyrexia	4.71	5.17	40	106	Terbinafine	Liver disorder	Rash	0.71	1.61
10	76	Paroxetine	Serotonin syndrome	Tachycardia	4.26	4.34	41	17	Atorvastatin	Abnormal hepatic function	Nausea	0.68	1.59
11	72	Levofloxacin	Liver disorder	Vomiting	4.12	5.05	42	111	Ticlopidine	Anemia	Headache	0.61	1.57
12	75	Paroxetine	Serotonin syndrome	Pyrexia	3.48	3.73	43	64	Levofloxacin	Abnormal hepatic function	Rash	0.61	1.60
13	91	Risperidone	Neuroleptic malignant s yndrome	Pyrexia	3.35	3.47	44	67	Levofloxacin	Liver disorder	Malaise	0.58	1.57
14	103	Terbinafine	Liver disorder	Nausea	3.18	3.79	45	54	Levofloxacin	Anaphylactoid reaction	Urticaria	0.58	1.57
15	107	Terbinafine	Liver disorder	Vomiting	3.09	3.71	46	88	Risperidone	Neuroleptic malignant syndrome	Dyslalia	0.54	1.40
16	74	Paroxetine	Serotonin syndrome	Hyperhidrosis	3.08	3.39	47	97	Terbinafine	Abnormal hepatic function	Pruritus	0.50	1.43
17	21	Atorvastatin	Abnormal hepatic function	Vomiting	3.07	3.74	48	26	Atorvastatin	Liver disorder	Pyrexia	0.49	1.42
18	63	Levofloxacin	Abnormal hepatic function	Pyrexia	2.90	3.85	49	28	Atorvastatin	Liver disorder	Vomiting	0.45	1.38
19	80	Pioglitazone	Congestive heart failure	Edema	2.75	3.27	50	23	Atorvastatin	Liver disorder	Malaise	0.45	1.39
20	100	Terbinafine	Abnormal hepatic function	Vomiting	2.50	3.20	51	87	Risperidone	Neuroleptic malignant syndrome	Increased blood pressure	0.45	1.33
21	102	Terbinafine	Liver disorder	Malaise	2.40	3.11	52	16	Atorvastatin	Abnormal hepatic function	Malaise	0.44	1.39
22	101	Terbinafine	Liver disorder	Decreased appetite	2.08	2.81	53	18	Atorvastatin	Abnormal hepatic function	Pruritus	0.41	1.36
23	108	Thiamazole	Agranulocytosis	Oropharyngeal pain	1.83	2.63	54	90	Risperidone	Neuroleptic malignant syndrome	Musculos keletal stiffness	0.38	1.27
24	68	Levofloxacin	Liver disorder	Nausea	1.82	2.79	55	60	Levofloxacin	Abnormal hepatic function	Malaise	0.38	1.37
25	96	Terbinafine	Abnormal hepatic function	Nausea	1.65	2.44	56	20	Atorvastatin	Abnormal hepatic function	Rash	0.35	1.31
26	93	Risperidone	Neuroleptic malignant syndrome	Tremors	1.52	2.07	57	44	Glimepiride	Hypoglycemia	Dizziness	0.35	1.34
27	95	Terbinafine	Abnormal hepatic function	Malaise	1.45	2.28	58	81	Raloxifene	Deep vein thrombosis	Chest pain	0.31	1.26
28	51	Levofloxacin	Anaphylactoid reaction	Dyspnoea	1.42	2.39	59	58	Levofloxacin	Anaphylactoid shock	Urticaria	0.30	1.29
29	24	Atorvastatin	Liver disorder	Nausea	1.39	2.21	60	104	Terbinafine	Liver disorder	Pruritus	0.27	1.23
30	59	Levofloxacin	Abnormal hepatic function	Decreased appetite	1.34	2.32	61	22	Atorvastatin	Liver disorder	Decreased appetite	0.23	1.20
31	89	Risperidone	Neuroleptic malignant s yndrome	Hyperhidrosis	1.34	1.98	62	79	Pioglitazone	Congestive heart failure	Fatigue	0.11	1.08

Likelihood ratios in bold: p<0.05 (P_pre vs. P_post), Likelihood ratios in italics: Not significant (P_pre vs. P_post).

For Ti-Ld-Py, which was ranked in first place, P_postQ+ was not calculated because Q_pre+Q_non exceeded 1.

The combinations are listed in decreasing order of LR+Q0 in Tables 3 and 4. Their ranks were similar to those based on LR+Q+ (rank correlation coefficient: 0.993, calculated excluding Ti-Ld-Py). Thirty combinations had LR+Q0 values of less than 1; i.e., their P_post values were smaller than their P_pre values. This was due to the fact that their Q_pre values were lower than their associated Q_non values, indicating that the incidence of IS-like changes in physical condition that occur during daily life and are unrelated to the medication was higher than the incidence of drug-induced IS; therefore, it would be inappropriate to regard the target symptoms as IS of ADR. Although the LR+Q+ values of some combinations exceeded 1, most of these combinations exhibited insignificant differences between their P_pre and P_postQ+ values, and the associated symptoms were demonstrated to be inappropriate for detecting the target ADR, as was found in the LR+Q0-based evaluation.

Usefulness of IS

Development of the IS marked with “★” in Table 2 was not described in ADR cases of the drugs, suggesting low-usefulness as predictors. Provision of the information of these IS are not only less beneficial but also may induce only a nocebo effect, to which attention should be paid.⁶⁾

The combinations for which P_post was calculable (Table 3) are summarized according to IS in Table 5. Combinations for which both P_postQ0 and P_postQ+ were significantly greater than P_pre and LR+Q0 was greater than 1 were regarded as useful.

Table 5. Usefulness of IS

ID	IS	Listed (X)	Calculated(Y)	Useful (Z)
37	Pyrexia	15	10	7
47	Vomiting	9	5	4
36	Decreased appetite	9	6	3
3	Malaise	10	6	2
30	Nausea	10	5	2
45	Tremors	3	3	1
39	Tachycardia	2	2	1
40	Dysponea	6	4	0
2	Pruritus	11	3	0
12	Rash	9	3	0
33	Hyperhidrosis	2	2	0
18	Urticaria	2	2	0
43	Dyslalia	4	1	0
15	Oropharyngeal pain	4	1	0
1	Fatigue	3	1	0
25	Chest pain	2	1	0
11	Dizziness	2	1	0
4	Headache	2	1	0
44	Asthenia	1	1	0
27	Increased blood pressure	1	1	0
5	Diarrhea	1	1	0
32	Musculoskeletal stiffness	1	1	0
29	Edema	1	1	0
19	Palpitations	5	0	0
22	Hemorrhage(*)	3	0	0
49	Hematuria	1	0	0
33	Somnolence	1	0	0

X: number of entries in Table 2, Y: number of drug-ADR-IS for which LR+ could be calculated, Z: number of drug-ADR-IS for which P_postQ0 and P_postQ+ were significantly different from P_pre and LR+Q0 > 1. (*) Subcutaneous hemorrhaging+gingival bleeding+epistaxis.

As a result, pyrexia was found to be useful in 7 of 10 drug-ADR-IS combinations for which the parameters could be calculated, suggesting its usefulness as an IS. Vomiting (4/5) and decreased appetite (3/6) were also suggested to be useful ADR predictors. Pruritus (0/11) and rash (0/9) are widely informed as IS, but were not found to be useful in any of the drug-ADR-IS combinations examined in this study. However, although pyrexia was considered to be useful, it did not develop in 5 of 15 candidate drug-ADR-IS combinations, which were marked with “★,” demonstrating that the type of IS induced and the likelihood of IS developing differ depending on the type of drug and ADR.

To prevent thiamazole-induced agranulocytosis, which was ranked highly in Tables 3 and 4, we suggest that fever should be selected as the IS, whereas paroxetine-induced serotonin syndrome can be detected by selecting tremors, levofloxacin- or terbinafine-induced liver disorder can be predicted by selecting fever, and decreased appetite might be a useful indicator of terbinafine-induced liver dysfunction. All of these IS were selected because they exhibited LR+ values of greater than 5.0. However, the calculations for fever for ticlopidine-induced liver disorder and decreased appetite for acarbose-induced liver dysfunction may have been less reliable because the numbers of ADR cases and ADR accompanied by IS were small.

DISCUSSION

IS That Were Not Reported to the AERS (★)

Some of the ISs listed in the MHD were not reported in the AERS database. In the MHD, the drug-ADR-IS combinations were selected based on their clinical significance. On the other hand, about 40% of the events in the AERS database were reported by customers.⁷⁾ In addition, most of them were reported after ADR had developed. Therefore, minor changes might have been overlooked in the AERS database, which might have contributed to some of the drug-ADR-IS combinations not being included in the AERS database.

Thus, the IS included in such drug-ADR-IS combinations might be hard to identify by retrospective studies.

ADR/AE That Were Not Found in the AERS Database (★★)

No cases of bezafibrate-associated rhabdomyolysis or cibenzoline succinate-associated hypoglycemia were found in the AERS database. It is known that bezafibrate-associated rhabdomyolysis is mainly caused by co-treatment with HMG-CoA reductase inhibitors.⁸⁾ In this study, the AERS search was limited to cases involving monotherapy might explain why no cases were identified. In addition, cibenzoline succinate-associated hypoglycemia frequently develops in patients with renal dysfunction.⁹⁾ Since most patients with renal dysfunction are prescribed multiple drugs, it is reasonable that no cases were identified when the AERS search was limited to monotherapy-treated patients.

In Japan, package inserts are included with all drugs. These inserts are official documents and provide comprehensive information about the drug in question. Since safety is the primary concern during the production of these package inserts, the reporting of not only ADR that occur in the setting of monotherapy but also those that occur during combination treatment anywhere in the world is requested by the Japanese regulatory authority.¹⁰⁾ Thus the searching for ADR that occurs in monotherapy setting, using official document focusing on safety, might result in some ADR being missed.

LR+Q0 and LR+Q+

LR+Q+ was calculated as well as LR+Q0 because it was assumed that the individuals with monotherapy that had reported the AE strongly suspected that the drug in question was the cause of the reported AE, and AE that were not caused by the drug were not included in the database. Accordingly, Q_non was not included in Q_pre or Q_other. Thus, we added Q_non in another calculation to set 2 ranges of P_post; P_postQ0 and P_postQ+, to avoid biased conclusion. This procedure also aimed at avoidance of LR+ below 1. That is, the higher the Q_non to Q_pre ratio, the larger the difference between LR+Q0 and LR+Q+; i.e., LR+Q0 is less than 1 and shifts closer to zero, whereas LR+Q+ shifts towards 1.

However, the results of the 2 calculations were mostly the same with regard to the usefulness of the target IS as a predictor of ADR in all calculations, showing that consistent evaluation is possible. Even though a mixture of ADR and AE are included the AERS reports, they might be sufficiently useful to allow the feasibility of IS as predictors of ADR to be evaluated.

Problems Associated with the Use of Useful IS as Predictive Markers

In this study, we found that pyrexia and vomiting are useful for predicting severe ADR. However, using such IS could be problematic since telling patients about these potential problems might cause them to become anxious, leading to poor adherence. However, providing information about IS that do not make patients as anxious, such as skin symptoms, including rash and pruritus, and fatigue, might be provided frequently rather than useful IS. Thus, in the clinical setting the information provided to patients about IS should be selected carefully.

Re-calculation of Values Obtained in Our Previous Study¹⁾ Using Eq. 2

Equation 2 was derived from Eq. 1, which was newly developed in the present study. Since Qt is included in Eq. 2, it is only applicable when the total number of users of the target drug is known. In our previous study, we utilized a simplified method to obtain P_postQ+, and LR+Q+ was derived from P_postQ+.¹⁾ In this study, it was possible to calculate P_postQ+ and LR+Q+ using Eq. 2. When we re-calculated the values obtained in our previous study using Eq. 2, we found that the LR+Q0 and LR+Q+ were similar (Table 6).

Table 6. Comparison of P_postQ+ and LR+Q+Values Derived from Eq. 2 with Those Obtained Using a Simplified Method

	P_pre	P_post Q+		LR+Q0	LR+Q+
	P_pre	Previous	Eq. 2	Previous	Previous	Eq. 2
Voriconazole	0.05000	0.04045	0.09814	5.00	0.80	2.07
Fenofibrate	0.00341	0.00313	0.04512	10.82	0.92	13.79
Sulfametoxazole/trimetoprim	0.00091	0.00329	0.01616	21.48	3.64	18.09
Ticlopidine hydrochloride	0.00176	0.00055	0.00863	2.54	0.31	4.93
Itraconazole	0.00081	0.00026	0.00545	4.44	0.32	6.80
Loxoprofen sodium hydrate	0.00007	0.00017	0.00141	27.92	2.35	19.00

Data Sources and Study Limitations

We used the following variables to perform the calculations in this study: (1) ADR probability values (P_pre and P_other), (2) the incidences of cases in which the target IS was accompanied by ADR/AE (Q_pre and Q_other), and (3) the incidence of I-like symptoms that are not drug-related (Q_non). The ADR probability values were obtained from the PMS data reported in drug IF, the IS incidence data were derived from the AERS database, and the data regarding I-like symptoms that are not drug-related were obtained from an internet-based survey of the general public. Each data source has its own problems.

In Japan, PMS data is collected during clinical practice. Although the resultant PMS data might be affected by selection bias because of the limited number of participating centers and the enrollment of agreeable patient,¹¹⁾ a certain level of quality is assured by the regulatory body since data are collected by pharmaceutical companies under the supervision of the regulatory body.¹²⁾ Furthermore, as the total number of patients that take a particular drug can be obtained from the PMS data, the incidence of ADR is calculable.¹²⁾ Therefore we used PMS data for calculating P_pre and P_other. However, it was difficult to evaluate the rare ADR, since the sample size was small. Similarly, it was difficult to determine the IS accompanied by ADR from the PMS data. A recent report suggested that small changes in the body condition caused by medication cannot be detected from the PMS data.¹³⁾ In the present study, we calculated Q_pre and Q_other using AERS data. Since the AERS database is based on a spontaneous reporting system, it has a tendency of underreporting to a greater extent than the abovementioned PMS data.^11,14,15) However, the cases in the AERS database involve nearly 5 million subjects, which is a sufficient number to compensate for any such underreporting. On the other hand, since the total number of users of a particular drug cannot be obtained from the AERS database, ADR incidences cannot be derived from it. However, this is not a problem for calculating Q_pre and Q_other since these parameters represent incidence rates among AE cases associated with the target drug. However, causal relationships have not been demonstrated to exist between the AE reported in the AERS database and the suspected drugs, and some of the reported IS might have been caused by the disease being treated.¹⁴⁾ Our results might have been influenced by these factors. Although notoriety bias (due to increased vigilance following safety alerts) has been detected in the AERS databses,¹⁶⁾ this might not have influenced the IS incidence data examined in our study, because IS is the mild ADR/AE. Finally, Q_non was obtained from an internet-based survey. Internet-based research (IR) has attracted attention due to the fact that it allows a large amount of data to be obtained cheaply and quickly. Such data are derived from individuals that are registered with research firms. The quality difference (survey results) between research firms is considered to be small¹⁷⁾; however, the response distribution patterns of IR are known to differ from those of mail surveys and detention surveys. Some authors have stated that it is wrong to assume that only data obtained by IR deviates from the “true” rate.¹⁷⁾ In the present study, Q_non was obtained from a pre-coded questionnaire. Concerning the causes of any changes in body condition that the IR subjects had experienced during the previous 3 months, the subjects were asked to select: (1) due to an ADR, (2) due to a cause not related to drugs, (3) an unexplained change in body condition that was not related to ADR, or (4) no change in body condition. The probability of selecting (3) was equivalent to Q_non. By utilizing a multiple choice answer format, the overreporting of changes in body condition due to an increased awareness of the condition, and hence, more frequent self-diagnosis cannot be ruled out. This is equivalent to the underreporting of ADR incidence data that occurs in spontaneous reports compared with that obtained in observational studies.¹⁵⁾ Thus, Q_non might overestimate the “true” probability. Accordingly, this could result in LR+Q0 values of less than 1 for ADR with small Q_pre values, as well as the underestimation of both P_postQ+ and LR+Q+.

Future Subjects

This study focused on drug-ADR-IS combinations that the pharmacists considered clinically important. Thus, other drug-ADR-IS combinations that are encountered in clinical practice remain to be examined. It is necessary to exhaustively investigate useful IS in order to be able to predict ADR.

Ghajar et al. investigated the usefulness of flare as an IS for predicting sulfonamide-induced drug eruption based on the type and onset time of the flare using Bayes’ theorem.¹⁸⁾ It is necessary to modify calculations such as ours in order to increase the probability of prevention, e.g., by adding information concerning the onset time of IS. It is also necessary to investigate the usefulness of evaluating IS in consideration of the concomitant use of multiple drugs because ADR overlap and the probability of an ADR developing alters when multiple drugs are used concomitantly.¹⁹⁾

CONCLUSION

By using the AERS database, we were able to quantitatively evaluate a markedly increased number of drug-ADR-IS combinations using Bayes’ theorem. As a result, it was suggested that fever, nausea, and decreased appetite are useful predictors of ADR, but pruritus and rash are less useful.

For thiamazole-induced agranulocytosis and levofloxacin- or terbinafine-induced liver disorder, fever might be a useful predictor of ADR; tremors might be useful for detecting paroxetine-induced serotonin syndrome; and decreased appetite might be a useful indicator of terbinafine-induced liver dysfunction.

REFERENCES

1) Ohshima S, Ieda M, Yamamoto M, Kobayashi D. Quantitative evaluation method of information value of initial symptoms based on Bayesian theory. Yakugaku Zasshi, 132, 763–768 (2012).
2) Scandellari C. The Bayesian approach to evaluation of diagnostic data. Ann. Ist. Super. Sanita, 27, 385–393 (1991).
3) Ministry of Health, Labour and Welfare. “the Manual for Handling Disorders due to Adverse Drug Reactions.”: ‹http://www.mhlw.go.jp/topics/2006/11/tp1122-1.html›, cited 6 May, 2013.
4) Saito M. Introduction of “the manual for handling disorders due to adverse drug reactions”: focus on the antibiotics related severe adverse drug reactions. Jpn. J. Antibiot., 61, 269–288 (2008).
5) Jose J, Rao PG. Pattern of adverse drug reactions notified by spontaneous reporting in an Indian tertiary care teaching hospital. Pharmacol. Res., 54, 226–233 (2006).
6) Colloca L, Finniss D. Noceboeffects, patient-clinician communication, and therapeutic outcomes. JAMA, 307, 567–568 (2012).
7) U.S. Food and Drug Administration. “AERS Reporting by Healthcare Providers and Consumers by Year (As of December 31, 2010).”: ‹http://www.fda.gov/Drugs/GuidanceComplianceRegulatoryInformation/Surveillance/AdverseDrugEffects/ucm070456.htm›, cited 10 May, 2013.
8) Wu J, Song Y, Li H, Chen J. Rhabdomyolysis associated with fibrate therapy: review of 76 published cases and a new case report. Eur. J. Clin. Pharmacol., 65, 1169–1174 (2009).
9) Interview form of cibenzoline succinate tablets, cited 11 May, 2013.
10) Ministry of Health, Labour and Welfare. “Instructions for Description on Drug Usages” (the 607th Notification from the Pharmaceutical Affairs Bureau issued on April 25th, 2007).
11) Ohashi Y. Progress report of the working group on “future postmarketing surveillance (PMS) in Japan.” Jpn. J. Pharmacoepidemiol., 8, 1–35 (2003).
12) Watanabe S, Nakano Y, Nomura K. Steps in developing a database of drug use-result surveillance: as an example of anti-hyperlipidemia drugs. Jpn. J. Pharmacoepidemiol., 17, 87–97 (2012).
13) Gnädinger M, Mellinghoff HU. The out-of-focus bias in drug surveillance. Eur. J. Clin. Pharmacol., 69, 357–359 (2013).
14) Fujita T. Signal detection of adverse drug reactions. Jpn. J. Pharmacoepidemiol., 14, 27–36 (2009).
15) Fletcher AP. Spontaneous adverse drug reaction reporting vs. event monitoring: a comparison. J. R. Soc. Med., 84, 341–344 (1991).
16) Raschi E, Piccinni C, Poluzzi E, Marchesini G, Ponti F D, The association of pancreatitis with antidiabetic drug use: gaining insight through the FDA pharmacovigilance database, Acta Diabetol: Published on line 19 October 2011.
17) Developing reliable internet-based surveys, SSJDA-42.: ‹https://ssjda.iss.u-tokyo.ac.jp/rps/RPS042.pdf›, cited 11 May, 2013.
18) Ghajar BM, Lanctôt KL, Shear NH, Naranjo CA. Bayesian differential diagnosis of a cutaneous reaction associated with the administration of sulfonamides. Semin. Dermatol., 8, 213–218 (1989).
19) Endo T, Watanabe N, Goto K, Kutsuma N, Komada F, Saitoh Y, Kobayashi D, Morimoto Y. Overlapping of the clinically significant adverse reaction of drugs in the polypharmacy. Jpn. J. Pharm. Health Care Sci., 28, 157–163 (2002).

Corresponding author

Register with J-STAGE for free!