Research Protocol： A pilot study of Japanese version of PEM on losartan

[1]Research Protocol： A pilot study of Japanese version of PEM on losartan
[2]Contents
[3]Title 51
[4]Subtitle 53
[5]Date of approval 55
[6]Investigators and parties supporting the pilot study 57
[7]Sponsors 112
[8]Abstract of the protocol 118
[9]A　　Medical environment associated with PEM 119
[10]B Processes of the study after the identification of patients in J-PEM 134
[11]Time line of the study 138
[12]A　 Marketing losartan in Japan 139
[13]B　 Recruitment of participating pharmacies and hospitals 141
[14]C　 Registration period 150
[15]D　 Sending questionnaires and publication of results 153
[16]Research objectives 156
[17]A Short-term study 158
[18]B Long-term study 169
[19]Review of the literature 174
[20]A　 Historical consideration 175
[21]B Hypothesis-generating function in PEM 206
[22]C Problems associated with T1/T2 comparison 215
[23]D Concurrent control in PEM 229
[24]E Abandonment of concurrent control in the UK PEM 237
[25]F Information on possible confounding variables in PEM 254
[26]G Lessons for Japanese version of PEM (J-PEM) obtained from review of the literature associated with the UK PEM 280
[27]Research Methods 291
[28]A Overall design 292
[29]B Population and sample 297
[30]C　　Data sources for exposure and risk factors 312
[31]C1 Identification of drug, patient and prescribing doctor 313
[32]C2 Co-operation between pharmacists and doctors 317
[33]C3 Questionnaires 319
[34]C4 Answers to questionnaires 338
[35]C5 Information on exposure and risk factors obtained from prescription and other data sources used to identify patients
342
[36]C6 Information on exposure and risk factors obtained from the questionnaire 349
[37]D Criteria for exposed and non-exposed 371
[38]E Events 385
[39]F Sample size 403
[40]G Methods to assemble data 422
[41]H Data management 447
[42]I Methods　of data analysis 465
[43]J　Quality assurance 486
[44]K Limitations of the study 493
[45]Protecting human subjects 497
[46]Publication of results 503
[47]Resources 507
[48]Amendments to the protocol 521
[49]References 524
[50]
[51]Title
[52]Prescription-Event Monitoring (PEM) on losartan, the first angiotensin II receptor blocker marketed in Japan
[53]Subtitle
[54]A pilot study for Japanese version of PEM
[55]Date of approval
[56]Ministry of Health and Welfare (MHW) evaluated the plan of this study regarding various aspects including ethical one after
consulting peer reviewers outside MHW and thereafter approved and decided to support this study in July 1998.
[57]Investigators and parties supporting the pilot study
[58]Principal investigator
[59]Kiyoshi Kubota MD PhD
[60] Associate Professor
[61] Department of Pharmacoepidemiology
[62] Faculty of Medicine, University of Tokyo
[63] 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8655 Japan
[64] (Former Honorary Research Fellow of the Drug Safety Research Unit, Southampton, England)
[65]Major co-investigators
[66]Shiro Hinotsu MD PhD
[67] Department of Epidemiology and Biostatistics
[68] Faculty of Medicine, University of Tokyo
[69] 7-3-1 Hongo, Bunkyo-ku, Tokyo 113 Japan
[70]Miki Yokotsuka BSc
[71] A Research Resident on loan to Department of Pharmacoepidemiology, Faculty of Medicine, University of Tokyo
[72] Society of Japanese Pharmacopoeia
[73] Nagai-Kinen-Kan 8F, 2-12-15, Shibuya, Sibuya-ku Tokyo 150-0002 Japan
[74]Misako Aoyama BSc
[75] Department of Pharmacoepidemiology
[76]Yasuko Morita BSc
[77] Department of Pharmacoepidemiology
[78]Kazuo Samizo BSc
[79] Department of Pharmacoepidemiology
[80]Chikuma Hamada PhD
[81] Department of Pharmacoepidemiology
[82]Akifumi Yafune MD PhD
[83] Department of Pharmacoepidemiology
[84]Daisuke Koide PhD
[85] Department of Pharmacoepidemiology
[86]Tsugumichi Sato PhD
[87] Hokkaido Women's University, School of Human Services
[88]Other staffs
[89]Mrs. Chieko Shigemi
[90]Mrs. Miyako Inada
[91]Miss Chiho Shigemi
[92] Department of Pharmacoepidemiology
[93]Important modifications on this protocol are to be made when the following parties reach an agreement (note: all enquiries
on the study plan for J-PEM must be normally addressed to Department of Pharmacoepidemiology, but in a special case where
an enquiry is directly addressed to one of the three parties below, a person with asterisk (*) is one to respond).
[94]A. Japan Medical Association Research Institute (JMARI)
[95] where major participants are
[96]Dr. Eikichi Itouji (President, JMARI; Vice-president, Japan Medical Association, JMA)
[97]Dr. Takashi Aoyagi (Executive member, Board of trustees, JMA)
[98]Dr. Hokuto Hoshi
[99]*Dr. Yuichiro Kuwama
[100]B. Japan Pharmaceutical Association (JPA）
[101] where major participants are
[102]Dr. Keiichi Saya (President, JPA)
[103]Dr. Noriyuki Takahashi (Vice-president, JPA)
[104]Dr. Minoru Urushibata (Executive Director, JPA)
[105]*Dr. Yukimitsu Kono (Pharmaceutical information center, JPA)
[106]C. Japanese Society of Hospital Pharmacies (JSHP)
[107] where major participants are
[108]Dr. Shikifumi Kitazawa (President, JSHP)
[109]Dr. Masahide Ishii (Chief of Secretary Division, JSHP)
[110]*Dr. Masahiro Hayashi (Chairman, Drug Information Committee, JSHP)
[111]Dr. Masako Fujikami (Chairman, Committee on measure toward small- and medium-sized hospitals、JSHP)
[112]Sponsors
[113]This study is conducted as a Health Science Research supported mainly by Ministry of Health and Welfare (MHW) Japan.
[114]A part of the current PEM study where JMARI recruits participating hospitals is conducted also as a part of one of
JMARI's projects.
[115]JMARI supports the part of the current PEM study conducted by hospitals recruited by JMARI financially.
[116]At least when this protocol is firstly issued, no support is to be obtained from any drug companies including the
manufacturer of losartan.
[117]The latest information will be added to this protocol immediately when any change is made on this aspect.
[118]Abstract of the protocol
[119]A　　Medical environment associated with PEM
[120]Prescription-Event Monitoring (PEM) has been in operation in England since early 1980s to monitor drug safety (and partly
effectiveness) of newly marketed medicines.
[121]To understand Japanese version of PEM, one may realise the difference of circumstances between England and Japan
associated with drug prescription and dispension.
[122]In the British National Health Services (NHS), drugs prescribed by General Practitioners (GPs) are dispensed by chemists
(pharmacists) of pharmacies located elsewhere in the villages, towns or cities.
[123]Patients receive drugs for free and drug fees are reimbursed by the Prescription Pricing Authority (PPA).
[124]The PPA collects all of the original prescriptions issued by GPs and prices each original prescription, and therefore, any
one of drugs described on the prescription is not missed in the pricing processes.
[125]In Japan, more than 70 % of drugs (a figure obtained in 1997) prescribed for outpatients are given to patients directly from
hospitals and doctor's private office where a doctor sees patients.
[126]The rest of drugs prescribed for outpatients are dispensed by pharmacists of pharmacies outside hospital or doctor's
private office.
[127]Every Japanese must be, by the law, an insurant of one of health insurance unions.
[128]Under Japanese comprehensive medical service plan, a patient receives drugs by paying a part of drug fees and the rest of
the drug fee is reimbursed by the insurer to which the patient belongs.
[129]Drug fees are reimbursed according to monthly claims (called as 'Rezept') issued by hospitals and doctor's private office
as well as pharmacies outside hospitals where drugs are dispensed according to the prescription issued by doctors.
[130]Before reimbursing hospitals drug fees (together with other medical fees) the claims are evaluated as to whether they
were made reasonably well and many hot debates between hospitals and insurers follow these evaluation processes.
[131]Mainly because of this opposition between insurers and hospitals associated with 'Rezept', nobody has so far succeeded
in utilising 'Rezept's for pharmacoepidemiology studies except for a local use in a specific small project.
[132]In this protocol on a pilot study of J-PEM, it is not assumed that 'Rezept's can be used in identification of drug, patient
and prescribing doctors.
[133]Drug, patient and prescribing doctor will be identified by pharmacists who dispense the drug according to a prescription or
by computerised prescription data as detailed later.
[134]B Processes of the study after the identification of patients in J-PEM
[135]Being different from the UK PEM where a questionnaire is sent to the prescribing doctor only, in J-PEM, a pair of
questionnaires are sent to both of the pharmacist who has registered the patient and the prescribing doctor.
[136]Data are analysed mainly by comparing event rates between 'test drug' and 'control drug' .
[137]Finally, the results are published both by internet and as a scientific paper in medical journals.
[138]Time line of the study
[139]A　 Marketing losartan in Japan
[140]In Japan, losaratan has been marketed on the 28th August, 1998 as 'Nu-lotan' by Banyu Pharmaceutical Co. LTD.
[141]B　 Recruitment of participating pharmacies and hospitals
[142]Recruitment of participating pharmacies has been started since the 21st August, 1998 when Department of
Pharmacoepidemiology, Faculty of Medicine, University of Tokyo sent a letter to more than 1000 pharmacies which have
participated in a preceding PEM pilot study on troglitazone having been conducted since July 1997 (including those which have
declined later after the study on troglitazone has been started) inviting them to the next study on losartan.
[143]Of those pharmacies, 333 have registered as participating pharmacies until the 18th December 1998.
[144]JPA and JSHP are preparing to recruit more pharmacies outside and inside hospitals.
[145]The list of participating pharmacies outside hospitals will be prepared by JPA until around late December 1998 and that of
participating pharmacies inside hospitals will be prepared by JSHP a little later.
[146]More than 10 hospitals have agreed to participate in the study according to the request made by JMARI.
[147]In addition to the pharmacies and hospitals as above, Department of Pharmacoepidemiology, Faculty of Medicine,
University of Tokyo is going to ask directly some more pharmacies and hospitals to participate in the study.
[148]It is also possible that some pharmacies and hospitals may propose to participate in the study voluntarily because it is
planned to recruit participating pharmacies and hospitals through advertisement using journals and internet.
[149]New participating pharmacies and hospitals are welcomed provided that they start to participate in the study
approximately 4 months before the last day of the registration period.
[150]C　 Registration period
[151]Registration form used to register patient codes will be sent to participating pharmacies from Department of
Pharmacoepidemiology, Faculty of Medicine, University of Tokyo from late November 1998 and pharmacies may start
registration immediately they receive registration form.
[152]Registration period will end on the 31st of December 1999 but may be prolonged if, for example, a relatively large number
of new pharmacies start to participate in the study in late 1999 and require longer registration period.
[153]D　 Sending questionnaires and publication of results
[154]A questionnaire asking to report events and other information for a particular individual patient will be mailed to
pharmacists and doctors when more than 6 months have elapsed after the first prescription of the drug monitored has been
issued to that patient.
[155]Mailing questionnaires will be started from around March 1999 until around June 2000 but most questionnaires will be
mailed until early 2000.
[156]Research objectives
[157]Japanese version of PEM may be divided into two observational studies; ie, a short-term study and long-term study.
[158]A Short-term study
[159]In a short-term study, the safety of a newly marketed drug obtained from answers to questionnaires sent to pharmacists
and doctors when more than 6 months have elapsed after the first prescription has been issued to the individual patients is the
main information to be collected.
[160]The information of the safety is obtained mainly from life events reported by pharmacists and doctors.
[161]Event rates are calculated for each of reported events and compared between a test drug and control drug(s) where　the
information on some of major possible confounding factors are taken into account.
[162]From answers to the questionnaires obtained from hospitals where drug, patient and prescribing doctor are identified from
computerized prescription data for those patients prescribed the test drug only, event rates are calculated but the between-
drug comparison is done while some of major possible confounding factors are taken into account when another PEM study on
a similar drug is conducted in the future.
[163]It is therefore important to make the results of one PEM study and those of another PEM study on a similar drug
conducted several years after the previous PEM to be comparable.
[164]To achieve the above objectives, the same information on possible confounding factors must be collected for a series of
PEM studies on similar drugs (eg,　a group of antidiabetics, antihypertensives, etc.)
[165]Another important point to make the results of two PEM studies on two similar drugs be comparable to each other is to
collect the information on events but not that on 'possible ADRs' like 'Drug Use Investigation' defined in 'Good PMS Practice'
issued from MHW Japan.
[166]In 'Drug Use Investigation', the list of ADRs known in the pre-marketing clinical studies is listed in the questionnaire not to
miss any possible ADR.
[167]Though no definite evidence is available, the reporting rate of events which are shown in the list as possible ADRs is likely
to be different from the reporting rate were such a list not shown.
[168]The results of two 'Drug Use Investigations' may not be comparable at least for the events listed in the questionnaire as
possible ADRs used in either study because ADRs to different drugs known in the pre-marketing studies generally differ
between drugs, and indeed, any between-drug comparison is not supposed to conduct when analysing data in the 'Drug Use
Investigation'.
[169]B Long-term study
[170]In a long-term study, the information on the long-term safety (and in part long-term effectiveness) of a newly marketed
drug is obtained mainly from answers to questionnaires sent, once a year over several years, to patients who registered their
own names and addresses directly to Department of Pharmacoepidemiology, Faculty of Medicine, University of Tokyo.
[171]To obtain reliable information from patients, only a limited number of questions on main life events including those
associated with 'true end points' are given in the questionnaires sent to patients.
[172]Event rates are calculated and compared between drugs where the information on possible confounding factors obtained
from doctors in the short-term study is taken into account.
[173]The information on possible confounding factors may be obtained from doctors for most patients as a registration form for
patient is passed to the patient from a doctor only when the doctor judges it is appropriate to do so and such doctor normally
responds to the questionnaire to a doctor as well.
[174]Review of the literature
[175]A　 Historical consideration
[176]It was in 1980 that the Drug Surveillance (later Safety) Research Unit (DSRU) was set up at the University of
Southampton, England by Professor WHW Inman (1).
[177]In order to understand the role of the DSRU and Prescription-Event Monitoring (PEM) which has been the main activity of
the DSRU, it is important to know why the DSRU has been set up in 1980.
[178]It is well known that the thalidomide disaster has led to the 'Yellow Card Scheme', the first national scheme in the United
Kingdom proposed by the subcommittee on adverse reactions of the Committee on Safety of Drugs (later Medicines)
abbreviated as CSD (later as CSM) in 1964 to collect reports on suspected drug reactions (2).
[179]Similar schemes were also established in some other countries in mid to late 1960s (3).
[180]The system is often called as 'spontaneous' or voluntary reporting system (VRS) (4) as it depends on the voluntary
reporting from medical doctors (and other health professionals including pharmacists in some countries such as USA (5)).
[181]On the other hand, PEM was designed to complement VRS after the 'Yellow Card Scheme' was found to have failed to give
early warning signals to 'oculomucocutaneous syndrome' caused by a beta-blocker, practolol (6).
[182]The 'Yellow Card Scheme' failed to detect practolol syndrome during the first four years after practolol was marketed in
1970 because its major manifestations such as dry eye, deafness and psoriasis-like eruption mimicked commonplace illnesses
which doctors often encountered in their clinical practice.
[183]In addition, in many cases, this ADR developed months or even years after the first prescription of the causative agent.
[184]Because of these factors, most doctors did not suspect the causal relationship between those adverse events and the
drug.
[185]It was only after 1974 to 1975 when a couple of astute doctors sent case reports of this new ADR to medical journals
(7-9) that hundreds of 'Yellow Cards' on the cases of suspected practolol syndrome were rushed to VRS (10).
[186]Subacute myelo-optico-neuropathy (SMON) due to chloroquinol, having clustered from 1960s to early 1970s particularly
in Japan (11) had similar features to practolol syndrome and other ADRs difficult to recognise.
[187]In typical SMON, patients had acute abdominal symptoms such as diarrhoea and abdominal pain suddenly at some specific
time point before neurological signs developed even if patients had started taking the causative agent long before the first
manifestation of this ADR (12).
[188]Although thousands of patients suffered from SMON, virtually no Japanese practitioners recognised the causal
relationship from the clinical observation.
[189]The original 4 objectives of the DSRU in the UK
[190]The DSRU had the following 4 objectives when it was established in 1980 (13).
[191]1 To establish a second nationwide drug safety screening programme called Prescription-Event Monitoring (PEM) in
collaboration with Prescription Pricing Authority.
[192]2 To conduct epidemiological investigations into specific drug safety problem on a local or national scale.
[193]3 To develop training and conference facilities.
[194]4 To study methods for improving the public's perception of the balance of drug risks and benefits.
[195]The original 10 objectives of PEM (10)
[196]1 It should enable us to estimate the incidence of adverse events.
[197]2 It should record all events and not merely those which have been thought to have been drug-induced.
[198]3 It should include all the users of a drug for at least as long as required to assemble a population capable of revealing
comparatively uncommon drug-events (eg in the range 1 % to 0.1 %).
[199]4 It should permit long-term follow-up.
[200]5 It should not influence prescribing (ie no inducement).
[201]6 It should not increase medico-legal risk.
[202]7 It should permit fast communication between researchers, prescribers, regulatory authorities and manufacturers.
[203]8 It should be standardized so that groups of patients treated with one drug could readily be compared with other drugs.
[204]9 It should be 'doctor-friendly'.
[205]10 It should be inexpensive.
[206]B Hypothesis-generating function in PEM
[207]Even if not explicitly described in 10 objectives of PEM above, one of the most important functions which PEM was
originally expected to exert, indicated by the historical matter that the paractorol affair led to the establishment of the DSRU,
was that of generating hypotheses on the new ADRs so far unrecognised or the new problems associated with the known ADRs.
[208]A statement 'PEM is a method of hypothesis-generating pharmacovigillance' can be found in a paper published from the
DSRU in as late as 1998 (14) though some textbook evaluates PEM as a system more suitable for hypothesis-testing than
hypothesis-generating (15).
[209]Being different from VRS, PEM should have some mechanism providing a hypothesis-generating function by analysing
'events' without depending on the individual doctor's suspicion on the causal relationship between the event and drug.
[210]As one of methods associated with such mechanisms, a within-drug comparison of T1 and T2, where T1 is the incidence
of an event measured in the first month after the first prescription is issued to individual patients while T2 is the monthly rate
of the same event measured during the succeeding period (eg month 2 to 6), has been tried (16-27) but with no remarkable
success.
[211]In the recent paper from the DSRU, the following statement is found (14); "Signals can be generated by an event having an
unusually high ID (incidence density) or high ranking in the 'Reasons for stopping' medication".
[212]The judgement that an ID is 'unusually' high is based on sort of an intuitive comparison between people who took the drug
and others who did not take the drug but this type of comparison is vague and may be therefore effective only for small number
of exceptional events.
[213] On the other hand, 'reasons for stopping medication' may be considered to be some evidence suggesting that individual
doctors have had suspicion about the causal relationship between the drug and event.
[214]In the recent paper from the DSRU, the following statement is found (14) where a within-drug comparison is raised as an
'alternative method' such as; "Alternatively it may be noticed that the pattern of the number of reports from week to week or
month to month may suggest a relationship to exposure to the drug being monitored".
[215]C Problems associated with T1/T2 comparison
[216]Though no definite proof has been published so far, as stated in some publications from the DSRU (16-27), the method of
T1/T2 (rate ratio) can pick up most of the major events which have been already known to be an ADR to the drug as events
with a high T1/T2 value.
[217]The method is explained as follows (16): "Experience has shown that, where the rate ratio exceeds 3.0, the events are
either the result of a reaction to the drug, or a sign or symptom of the disorder being treated."
[218]The method has two major problems of which the first one is associated with the process of finding events with high rate
ratio.
[219]One big problem associated with the process of finding events with high rate ratio is low efficiency because the method
requires relatively large number of events to generate 'signals'.
[220]To calculate T1 and T2, one must divide reports of one event into two portions; those reported in month 1 and those
reported month 2 or later.
[221]In addition, if month 2-6 is adopted as a period to calculate T2, those reported in month 7 or later are discarded when
calculating T1 and T2.
[222]To generate a 'signal', the number of reports on the event should be big enough to make the comparison of two rates T1
and T2 meaningful, and therefore, except for a few exceptional events where almost all reports are those in month 1, the event
which may be picked up as a 'signal' by the method may have to have the allover incidence measured through the whole
observation period (or whole period on treatment) of 1 % or more unless the size of population studied is much greater than the
current size studied in the UK PEM (aiming to exceed 10,000 subjects).
[223]Another problem is associated with the process of refining 'signals' generated by the mechanical use of T1/T2 method so
as to select 'signals' to be published as possible ADRs.
[224]As in the statement shown above (16) most events with high T1/T2 are either ADRs to the drug or a sign or symptom of
the disorder being treated (or events confounded by indication).
[225]Therefore, it is important to have a method to judge whether the finding is confounded by indication.
[226]In order to judge whether confounding by indication is a likely scenario, one must have a good control and the information
on possible confounding variables.
[227]The information on possible confounding variables is important to examine whether the finding associated with a known
drug effect (eg change in asthma mortality in asthmatic patients treated with antiasthmatics) is in fact confounded by
indication.
[228]However, in PEM the information on possible confounding variables is mainly required to examine whether the event is
likely or unlikely to be an ADR.
[229]D Concurrent control in PEM
[230]It is known that choosing 'concurrent control' (ie, patients treated during the same period of time) is, in general, a better
way of avoiding bias when compared with 'historical control' (28).
[231]At least in the early stage of PEM, the between-drug comparison by using concurrent control was thought to be the major
means to analyse the data (29, 30).
[232]For example, in 1981 Inman wrote (30): Each "test" drug will be matched with a "control drug", the test drug being one
that has recently been granted a product license and the control drug will usually be a chemically or pharmacologically similar
drug already marketed for the same indication.
[233]The first task will be to process prescriptions for the control drug in such a way that contemporary treatments may be
selected for comparison with new product (30).
[234]In practice this will be done by monitoring the prescriptions for the control drug for several months until patients who are
"new to the system" appear (30).
[235]These will be put on one side for further study and the remainder discarded (30).
[236]There would be little point, for example, in comparing patients starting treatment with a β-blocking agent marketed for
the first time in 1981 with controls who might have been taking propranolol for ten or more years (30).
[237]E Abandonment of concurrent control in the UK PEM
[238]By some reason which is uncertain from any literature, the idea to obtain concurrent control given in the paper published
in 1981 (30) has been not materialised.
[239]Instead, 'control' which could never be regarded to be 'concurrent' was once tried resulting in a discouraging result as
correctly predicted in the 1981 paper (30).
[240]Probably, the experience with 'Indocid' gave the most important driving force to make the DSRU give up the use of
concurrent control in PEM.
[241]In 1983 the study of two NSAIDs conducted since 1982 (1), benoxaprofen and fenbufen was extended and later in the
same year zomepirac, piroxicam and 'Osmosin' (indomethacin slow-release) were added (31,32).
[242]Though not included in the report of the PEM studies on 5 NSAIDs in 1985 (33), the DSRU has also conducted, at the
same time, a study of 'Indocid' (old preparation of indomethacin).
[243]The original idea was to explore the possibility that 'Osmosin' and 'Indocid' might show different rates of gastrointestinal
disorder.
[244]However, according to the description given by Inman (6): "the results of this experiment were very instructive from the
methodological point of view.
[245]Many patients in the 'Indocid' group had been using the drug for very long periods, sometimes as long as 20 years.
[246]The number of events recorded was much lower than with the other five preparations and the data are considered not to
be comparable.
[247]Many doctors complained that it took too much of their time to record events in patients who had used 'Indocid' often for
many years and asked us not to embark on other studies on well-established drugs."
[248]The problem described here is known as the problems associated with 'survival cohorts' in addition to an extraordinary
burden to doctors who were asked to record events which occurred during the preceding several years in the patients with
'Indocid'.
[249]True cohort studies should be distinguished from studies of 'survival cohorts' or 'available patient cohorts' (28).
[250]Many of the current users of an old drug, particularly that used for a chronic condition, are often 'survivors' who have not
had any serious adverse events and, at the same time, have benefited from the long-term use of the drug.
[251]The DSRU's policy not to employ a concurrent control was already established before 1985 when a PEM study on
ranitidine was conducted as seen in the following description:
[252]"For a number of reasons, we did not initiate a green form exercise for cimetidine.---
[253]The main value of the ranitidine data will be for comparison with any new H2-receptor antagonist which may be marketed
in the future" (6).
[254]F Information on possible confounding variables in PEM
[255]It may be considered to be reasonable that very few questions on possible confounding variables are made in the green
form used in the UK PEM to maintain a fair response rate (58.2%） (14) while doctors are expected to answer for free.
[256]The response rate from doctors in the UK PEM is fairly high and more or less constant though it has been observed that
the response rate tends to decrease slowly when measured over years (14, 33).
[257]Nevertheless, it is sometimes difficult to judge whether the finding suggesting some relationship between the event and
drug is confounded by indication or in fact indicating that the event is an ADR to the drug because of a paucity of the
information on possible confounding variables.
[258]Recently, the DSRU adopted a policy that the event rate is adjusted by age and sex in general when compared between
drugs (14, 34).
[259]However, it is rare that age or sex is a major confounding factor to be adjusted when the event rate is compared between
drugs.
[260]Similar to other pharmacoepidemiology studies, in PEM the most important confounding variables are in general variables
assoicated with the indication of the drug.
[261]The problems associated with a paucity of the information on possible confounding variables may be illuminated by
exploring the role the PEM study on enalapril has played in the processes to examine the relationship between renal failure
and enalapril.
[262]The PEM study on enalapril was conducted between 1985 and 1986, where similar to the ranitidine study, no information
for the patients using captopril was collected in the study.
[263]When the results of the enalapril study were analyzed, the main concern was a large number of fatal cases of which a
considerable proportion was associated with renal failure occurring with a much higher rate than expected.
[264]In 1988 a series of two reports on the PEM study of enalapril were published in BMJ (35, 36) and the subtitle of the second
of the series was "investigation of the potential role of enalapril in deaths with renal failure" signalling a possible relationship
between enalapril and death from renal failure.
[265]In the PEM study of enalapril, 1098 of 15169 green forms recorded patients as having died and after examining patient's
notes for 913 of 1098 fatal cases, in 10 patients enalapril was thought to have contributed to a deterioration in renal function
sufficient to be a factor in the subsequent deaths.
[266]Those 10 patients were characterised by the old age, use of high dose, concomitant use of potassium sparing diuretics or
NSAIDs and pre-existing renal disease (36).
[267]It is now well known that ACE inhibitors have 'reno-protective' effects in the patients with diabetic nephropathy (37-41)
and they may be 'reno-protective' also in the patients with other renal impairment (42, 43).
[268]However, it is also known that in patients with severe bilateral renal artery stenosis, ACE inhibitors reduce or abolish
glomerular filtration resulting in progressive renal failure (44) and in some report, progressive renal failure was recorded in about
5 % of patients with documented renovascular hypertension treated with ACE inhibitors (45).
[269]There seems to be a need to resolve an apparent discrepancy between the report on 'potential role of enalapril in deaths
with renal failure' from the DSRU (36) and those on the 'renoprotective' effect of ACE inhibitors (37-43).
[270]Noteworthily, the rate or renal failure has dramatically decreased in the sequential manner with time when compared
among the PEM studies of four ACE inhibitors including 3 conducted after the PEM study of enalapril ('85-86); ie lisinopril
('88-89), ramipril ('90-91) and perindopril ('90-).
[271]The possible mechanism for the difference of the rates of renal failure was given when the results of the lisinopril study
became available as follows (10):
[272]"The rate of renal failure during treatment with enalapril was calculated to be 6.5 per thousand patient-years of use of the
drug.
[273]The corresponding rate for lisinopril is 3.9 per thousand patient-years.
[274]We believe that this may reflect a difference in the selection of patients, the newer drug being used in patients with less
advanced cardiovascular disease.
[275]This is suggested by the fact that, during the first six months' treatment with enalapril there were 403 deaths (2.9%)
compared with 157 (1.3 %) in the lisinopril group during the same period" (10).
[276]No PEM study was conducted on captopril, and therefore, it is difficult to guess what would have happened and whether
the "selection of patients" (10) would have been easier to recognise if patients with captopril had been studied as a concurrent
control.
[277]However, even when no control was used or unable to use, if a simple question on the concomitant disease or concurrent
abnormal condition was added to the green form, it might be not very difficult to recognise that among the patients prescribed
enalapril, the fraction of those who had already had renal dysfunction was higher than expected.
[278]Even if more direct information on the "selection of patients" (10) had been available, the conclusion of the enalapril study
(ie enalapril might contribute to the 10 deaths) would not have been largely altered.
[279]However, mentioning the possibility of "selection of patients" would, if possible, have made the report more informative.
[280]G Lessons for Japanese version of PEM (J-PEM) obtained from review of literature associated with the UK PEM
[281]Ｅｖｅｎ in the current era where many high-quality medical databases including General Practice Research Database
(GPRD) formally called as Value Added Medical Products (VAMP) in the UK (46-49) are available, the UK PEM still has its own
strength particularly in that it can identify virtually all of the patients in England prescribed a new drug by GPs.
[282]This point is important because such a scenario is possible that, for example, some target number of patients with a
particular new drug is attained in 12 months after marketing the drug in the UK PEM, while the same number may be not
attained until more than 10 years elapse when using the database covering less than 10 % of the population.
[283]When this strength with the UK PEM is taken into account, lack of the concurrent control and a paucity of the information
on confounding variables may be considered to be acceptable.
[284]On the other hand, at least currently, there is no practical means to identify all of the users of some drug in Japan.
[285]As described above, a monthly claim called as 'Rezept' is issued for virtually all of the patients who have received any kind
of medical service in Japan but the resource is currently difficult to use efficiently for the pharmacoepidemiology studies.
[286]In addition to the red tape and conflict of the interest between insurers and hospitals, no systematic record of 'Rezept's
detailed enough to be used as an epidemiological resource is currently available.
[287]Some large insurers make and keep a detailed computerized record made from 'Rezepts' usually issued as hard copies but
the others, particularly small ones, do not.
[288]Therefore, even if the problems associated with the red tape and conflict between insurers and hospitals are resolved,
only a fraction of patients unlikely to represent whole patients are possible to identify from 'Rezepts' though the situation
seems gradually improved (eg 'Rezepts' issued through some electronic media which have been tried in some restricted
districts).
[289]Where no means to identify all of the drug users is available, two problems, no concurrent control and a paucity of the
information on confounding variables illuminated from the above review of the literature associated with the UK PEM may
emerge as fatal problems.
[290]Another factor which makes two issues, concurrent control and more information on confounding variables, be requisite
components of Japanese version of PEM is that there has been and will be (at least for the time being) no good medical
database suitable for 'hypothesis-testing' in Japan so that Japanese version of PEM must have a mechanism for
'hypothesis-testing' within PEM itself.
[291]Research Methods
[292]A Overall design
[293]Similar to the UK PEM, J-PEM is conducted as a cohort study.
[294]One of the most important aims of PEM is to find new ADRs unrecognised so far or new problems associated with ADRs
already recognised.
[295]Being different from some other epidemiological study designs employed for pharmacoepidemiology studies including
case-control study, a cohort study has an advantage in that it may identify any unexpected events which may occur in an
observed cohort so that this study design meets the objective of PEM.
[296]For instance, in the cohort study designed with the major objective to study the relationship between smoking and lung
cancer, where those who smoke and those who do not were compared with each other, the results gave researchers some
'bonus' that is a chance to study the relationship between smoking and other malignancies and some of other diseases (50).
[297]B Population and sample
[298]Population in Japanese version of PEM (J-PEM) consists of all of the patients in Japan prescribed either 'test drug' or one
of 'control drugs' on the date of marketing the 'test drug' (defined as 'index date') or later for the first time ever in their lives.
[299]A sample consists of a fraction of patients representing the whole patients with 'test drug' as well as the matched group of
patients prescribed 'control drugs' selected from the population according to the method given below.
[300]As stated in the original design of the UK PEM in 1981 (30), in a study of a newly marketed drug, concurrent controls are
new users of the 'control drug' which is a chemically or pharmacologically similar drug already marketed for the same indication.
[301]If a patient, who has been already prescribed a drug under one particular drug class (eg NSAIDs, sulfonylureas,
dihydropyridine calcium blockers, etc.), is 'newly' prescribed one of 'control drugs' which is not the same in terms of the
chemical structure but is under the same drug class as the drug prescribed before, after the 'index date' for the first time ever
in his/her life, he/she is considered to be a member of the population.
[302]On the other hand, various brand names and formulae (eg injection, tablet, suppository etc.) are considered to be all the
same if the component is chemically identical to one another.
[303]It seems very difficult to know whether the patient has been prescribed any 'old' drug for the first time ever by means of
centrally collected prescriptions, so that concurrent controls may be actually impossible to obtain, for example, in the UK PEM
where drugs, patients and doctors are identified by prescriptions collected by the PPA though some idea to identify concurrent
controls from centrally collected prescriptions has been presented (30).
[304]In J-PEM, patients prescribed 'test drug' as well as controls are found during the day-to-day activity of participating
pharmacies.
[305]In a pilot study of PEM on losartan, 'controls' consist of a group of patients prescribed one of ACE inhibitors or one of
dihydropiridine calcium antagonists for the first time ever for that patients after 1 September 1998.
[306]To find a control, a pharmacist in a participating pharmacy asks the patient whether or not he/she has ever used that drug
so far when the pharmacist finds an antihypertensive which is considered to be potentially new for that patient.
[307]In one of the relatively efficient ways to find a control, a patient is judged to be prescribed a 'control' drug potentially for
the first time in the patient's life if the drug prescribed today has not been prescribed before.
[308]If the patient replies, responding to the question made by a pharmacist, that he/she has an experience to have used that
drug before, or, if the patient is uncertain, the patient is excluded from the list of the concurrent controls and only the patient
who is certainly a new user is included.
[309]Even if a patient has a disease which is the indication of the drug monitored, the patient is not considered to be a member
of the population unless one of drugs monitored is prescribed after the index date.
[310]A patient prescribed the drug monitored (either 'test drug' or one of 'control drugs') already before the 'index date' is not
considered to be a member of the population so that a large fraction of patients prescribed one of 'control drugs' as well as a
small fraction of patients prescribed 'test drug' already before the 'index date' (eg, in the pre-marketing clinical studies) are
excluded.
[311]It is likely that the current sample in J-PEM do not fully represent the population defined above so that every effort must
be made to improve the method to obtain a　sample which fully represents the population mainly by increasing the number of
participating hospitals and pharmacies as well as by trying to use 'Rezepts' to identify patients to be studied particularly at the
stage where the record of 'Rezepts' is improved so that it becomes a good resource suitable for pharmacoepidemiology studies.
[312]C　　Data sources for exposure and risk factors
[313]C1 Identification of drug, patient and prescribing doctor
[314]In a pilot study of Japanese version of PEM on losartan proposed in this protocol, drugs, patients and prescribing doctors
are identified, being independent of doctors, mainly by pharmacists in the pharmacies outside and inside hospitals as described
earlier.
[315]Participation of doctors of hospitals and doctor's private offices, where the method of identification by pharmacists is by
some reason unavailable, are welcomed to the PEM study provided that an alternative method is available where all of the
patients prescribed the test drug as well as prescribing doctors are identified being independent of prescribing doctors.
[316]A typical one of such alternative methods is the use of computerized prescription data.
[317]C2 Co-operation between pharmacists and doctors
[318]Although doctors are not informed of which particular patient may be a subject of monitoring, in the main part of the
current PEM study where drug, patient and prescribing doctor are identified by pharmacists, it is preferable and often essential
to ask doctors' co-operation prior to the moment when the first questionnaire is delivered to the doctor.
[319]C3 Questionnaires (see 'Questionnaires' attached to this protocol or somewhere around the place where this protocol is
placed in homepage)
[320]When more than 6 months elapse after the first prescription is issued for each patient, two questionnaires per patient are
mailed to participating pharmacies.
[321]One questionnaire is to be answered by the pharmacist.
[322]The other one is to be answered by a doctor which is forwarded to the doctor by the pharmacist after the pharmacist
describes, by hand, patient's name and doctor's name on its envelope according to the master file he/she keeps inside his/her
pharmacy.
[323]When the participating pharmacy is located outside hospital and when the pharmacy inside hospital agrees to co-operate,
the questionnaire for doctors are forwarded to the hospital pharmacy and pharmacists inside hospitals pass the questionnaire
to the doctor who is considered to be the most appropriate to answer to that question (and may differ from the prescribing
doctor).
[324]In a pilot study on troglitazone, preceding the study on losartan, an additional questionnaire has been issued to the co-
operative pharmacist inside hospital to obtain the information mainly on prescribed drugs which may not be (completely)
obtained from the pharmacies outside hospital.
[325]However, it has turned out that the information obtained from the pharmacy outside the hospital is in general as good as
that from hospital pharmacy and therefore, the additional questionnaire is not issued to the hospital pharmacy in the PEM study
on losartan.
[326]Prior to starting answering the questionnaire on a patient with one of 'control drugs', it is asked that the person who are
going to answer the questionnaire checks whether the first prescription date of the subject is after the 'index date', and if not,
the person who has noticed that the patient is not eligible is asked to return the questionnaire without entering any other
information.
[327]In a pilot study on troglitazone, approximately 10 % of questionnaires to the doctor issued for patients with a 'control drug'
have been returned mostly by the pharmacist who has registered the patient before being forwarded or passed to the doctor
because the first prescription date is in fact prior to the 'index date', but sometimes by a pharmacist in a hospital pharmacy to
whom the questionnaire to the doctor has been forwarded from the pharmacist outside hospital because of this reason, and
rarely a doctor has indicated that the patient is in fact not eligible.
[328]A couple of questionnaires have returned from doctors who answered all of the questions in the questionnaire issued to
the patient with a 'control drug' prescribed before the 'index date'.
[329]To maintain some level of the response rate from doctors, total number of the questionnaires (either on 'test drug' or
'control drug') sent to one doctor will not normally exceed 10 in one study and the questionnaire is sent at a rate of
1questionnaire per month or less.
[330]If a doctor indicates that he does not want to co-operate any more on one of the questionnaires returned, any more
questionnaire is issued to that doctor (it can be known from doctor code that the prescription has been issued by some specific
doctor).
[331]If 3 questionnaires are sent to one doctor but no questionnaire is returned in two months after the last questionnaire is
sent, a special form is sent to the doctor to ask whether he/she could co-operate if the questionnaire is sent once more and if
he/she answers that he/she cannot co-operate or he/she does not respond, no more questionnaires are issued to that doctor.
[332]In such a case where no co-operation of a doctor is expected, it is considered that further questionnaires for the patients
with a drug prescribed by that doctor are mailed to the pharmacist only because it is not possible nor requisite that a group of
patients monitored by pharmacists is exactly the same as that monitored by doctors (eg, in the pilot study on troglitazone, the
response rate with pharmacists is much higher than that with doctors, and, the data obtained from pharmacists and those from
doctors are analysed separately as below).
[333]A doctor after completing the questionnaire sends back the questionnaire to Department of Pharmacoepidemiology,
Faculty of Medicine, University of Tokyo after giving his/her signature and address but removing the envelope where the
patient's name is given.
[334]A participating pharmacist also sends back the questionnaire without patient's name but with his/her signature and
address.
[335]Inside the envelope where a questionnaire for the doctor is inserted, another small envelope for a patient is also placed
and the doctor is asked, if acceptable, to give the document inside the envelope to the patient.
[336]After receiving the document from his/her doctor and having agreed to participate in the study, a patient registers his/her
name and address directly to Department of Pharmacoepidemiology and a questionnaire is mailed directly to the patient once a
year for several years to monitor major event.
[337]It may be also planned that an additional questionnaire is mailed to a doctor or pharmacist 2 to 3 years later to have the
information on events associated with a long-term effect of drugs.
[338]C4 Answers to questionnaires
[339]As in the pilot study on troglitazone, in the study on losartan, the response rate with participating pharmacists is expected
to be higher than that with doctors mainly because pharmacists are those who want to participate in the study and register
patients' codes while some doctors may not know that they have been involved in the study until they receive a questionnaire.
[340]In the first and small pilot study of PEM on troglitazone being conducted since July 1997, preceding the PEM study on
losartan, the final response rate from the participating pharmacists will be more than 80 % while that from doctors will be more
than 35 %.
[341]In the PEM study on troglitazone, approximately 10 % of the patients for whom a questionnaire was sent to a doctor
registered his/her name to Department.
[342]C5 Information on exposure and risk factors obtained from prescription and other means used to identify patients
[343]As stated above prescription is the means to identify the patient exposed to the drug monitored in pharmacies outside the
hospital most of which are recruited by JPA.
[344]From prescription, the information on some of risk factors, such as age and sex is also obtained.
[345]In the registration form used in the pilot study on losartan (see 'Registration Form' attached to this protocol or somewhere
around the place where this protocol is placed in homepage), in addition to patient's code (already printed on the form),
patient's age and sex, doctoｒ code, drug name and the first prescription date it is asked to give the followings: the name (or
initial, if there is any problem to give the name) of the institute (hospital or doctor's office), size of the hospital or doctor's office
and whether or not hospital pharmacy is available in the place where the prescription has been issued.
[346]Prescription may be also used when a patient is identified by pharmacists of a hospital pharmacy recruited by JSHP, or by
pharmacists or other clerical hospital staff of the hospitals recruited by JMARI.
[347]Other source used to identify the patients exposed to the drug monitored is the computerized prescription data and such
resource may be used to know some of other risk factors as well.
[348]Other risk factors and some of the information associated with the exposure (eg, dose, date when the drug was stopped,
etc.) are obtained by questionnaires sent to a pharmacist or doctor as given below.
[349]C6 Information on exposure and risk factors obtained from the questionnaire (see 'Questionnaires' attached to this
protocol or somewhere around the place where this protocol is placed in homepage)
[350]In the questionnaire to doctors in J-PEM, the first prescription date, indication of the drug monitored, the date when the
indication developed for the first time, and the date when drug treatment was started for the indication of the drug monitored
are asked.
[351]The question on the date when drug treatment was started for the indication of the drug monitored may be considered to
be the information more relevant to the severity of indication (thus a possible confounding variable) rather than the information
on exposure and this question as well as the question on the date when the indication developed for the first time are unique to
J-PEM as they are not given as specific items to be answered in the questionnaire used in the UK PEM.
[352]In addition, a representative daily dose of the drug monitored, whether or not the patient is still on treatment with the drug
monitored, the date when the drug monitored was stopped or substituted by another drug and its name are asked in the
questionnaire to the doctor.
[353]Of the above questions, that on representative daily dose of the drug monitored is unique to J-PEM but other questions
are included in the questionnaire in the UK PEM as well.
[354]Another question in the questionnaire to doctors associated with exposure to the drug monitored is daily dose when each
event reported occurred.
[355]This question is not unique to J-PEM but has been included in the questionnaire used in the UK PEM since 1994.
[356]In the questionnaire to pharmacists, the information on the duration and daily dose of the drug monitored as well as the
information on concurrent drugs with the same indication as that of the drug monitored and other concurrent drugs with any
indication including OTC drugs are asked to describe.
[357]In addition, a pharmacist is asked to evaluate whether his/her description given in the questionnaire covers the
information on all drugs which the patient has actually obtained including those from other pharmacies or hospitals during the
corresponding period.
[358]Questions on risk factors in the questionnaires to doctors include the date when the indication of the drug monitored
developed for the first time and the date when the drug treatment was started for the indication as described above.
[359]In addition, concurrent diseases and other concurrent conditions which the patient had around the first prescription date
are asked in the questionnaire to doctors where major complications as well as major risk factors are listed so that doctors may
just encircle the terms where appropriate.
[360]The concurrent diseases or conditions which are not shown in the list are asked to itemise.
[361]The list of the concurrent diseases and risk factors shows those which are likely to be confounding variables when events
associated with the indication are analysed.
[362]As detailed in the review of literature, in PEM it is difficult to know what events become those to be analysed in depth so
that it is also difficult to know the best question on possible confounding variables prior to the start of the study.
[363]Another factor to be taken into account when selecting the questions on possible confounding variables is the fact that
the data obtained from some hospitals recruited by JMARI as well as some of those hospitals and doctor's offices who may
want to participate in the PEM studies where concurrent control is not available, may be analysed in the future by comparing
the data obtained from other PEM studies on the drugs with similar indication.
[364]To make such comparison effective, the question on the information on possible confounding variables may be the same
for one class of drugs.
[365]In the questionnaires used in the study of losartan, questions on major risk factors and target organ damage and clinical
cardiovascular disease associated with hypertension listed in the report from 'The Sixth Joint National Committee on
Prevention, Detection, Evaluation, and Treatment of High Blood Pressure' (51) are asked.
[366]The questions include whether the patient had on the first prescription date, complications of hypertension such as left
ventricular hypertrophy, angina, myocardial infarction as well as some of risk factors such as smoking and family history of
cardiovascular disease: women <65 or men < 55 y.
[367]The questions on results of small number of laboratory data obtained around the first prescription date as well as body
weight, body height, are also asked.
[368]In addition, it is required to describe a representative value of blood pressure measured on a particular day before starting
the drug monitored and that during the treatment with the drug monitored as well as to answer whether blood pressure has
been stable in each occasion when measuring blood pressure.
[369]Finally, patient's compliance to doctor's indication on medication etc. are requested to answer.
[370]In the questionnaire to pharmacists, date when the indication of drug has developed for the first time, date when the drug
treatment for that condition has been started for the first time, concurrent diseases as well as patient's compliance to
medication are asked.
[371]D Criteria for exposed and non-exposed
[372]For a 'test drug', the criteria for exposed is very straightforward; just if a 'test drug' is described on the prescription or
computerized prescription data, the patient is judged to be exposed to the 'test drug' and included in the list to be registered
except for a small number of cases where the patient has been exposed to the drug prior to the 'index date' (eg, in a pre-
marketing clinical study).
[373]If it turns out that the patient has not at all used the drug prescribed, the patient is excluded when this fact is found and
such a scenario, if any, will be usually known when the answer to the questionnaire is returned.
[374]On the other hand, for the 'control drug' the criteria for exposed may be considered not so straightforward as that for the
'test drug'.
[375]One of the important points when evaluating the comparability between patients with 'control drugs' and those with 'tesｔ
drug' is whether or not either of the patient and doctor has realised the fact that the patient has ever used that drug in the past
when the drug was 'newly' prescribed.
[376]If such an information is available that the drug has probably caused some untoward reaction before, or that the results
observed after the previous prescription of the drug have been preferable, the information is likely to be considered when
deciding whether or not that drug is prescribed to the patient.
[377]Therefore, the situation encountered when prescribing one of 'old' drugs is different from that when prescribing the newly
marketed 'test drug' where no previous information is available for virtually all of the patients prescribed the 'test drug'.
[378]However, it is likely to be difficult for the pharmacists to know whether or not either of the patient and doctor has realised
the fact that the patient has ever used that drug in the past when the drug has been 'newly' prescribed.
[379]In addition, the chance that the patient remembers that he/she has been prescribed a drug before may be larger when the
drug has caused an untoward effect than that when the drug has not caused any problem so that the previous exposure may
produce differential effect on the patient's memory.
[380]Therefore, it is probably safe that the patient is excluded from the list of controlｓ if it is found that the patient has ever
used the drug prior to the index day (ie, the date when the 'test drug' is marketed) irrespective of whether or not such an
information has been considered when the drug has been prescribed for the first time.
[381]It is ideal that pharmacists in participating pharmacies identify and register all of the patients with the 'test drug' and all of
the eligible patients with 'control drugs' (ie, those who are sure that one of 'control drugs' given by a doctor after the 'index
date' has been prescribed for the first time ever in their lives).
[382]However, it is acceptable that some fraction of eligible patients are registered provided that the patients are not selected
in a way that bias is likely introduced.
[383]For instance, if all of the eligible patients either with the 'test drug' or one of 'control drugs' who visit a participating
pharmacy in the morning of every Monday are registered, this manner may be acceptable as it may not largely affect the
internal validity although patients who visit that pharmacy Monday morning may not represent whole patients who may visit that
pharmacy (eg, patients prescribed a drug by a doctor who does not work in outpatient clinic Monday morning are less likely to
be identified than those by a doctor who does).
[384]On the other hand, if only the patients who have used the drug ('test drug' or 'control drug') more than 30 days are
registered, this may abolish the internal validity because the duration of drug use may be associated with the exposure so that
this way of registration is an arbitrary selection of the patients which is differential (ie the different effect may be produced
between 'test drug' and 'control drugs').
[385]E Events
[386]Health outcomes or any events observed in patients with the drug monitored are obtained from questionnaires sent to
pharmacists and doctors.
[387]The definition of 'event' given in every questionnaire sent to a doctor is almost the same as that in the UK PEM.
[388]Event' is defined in the questionnaire sent to a doctor as: Any new diagnosis, symptom or sign and abnormal laboratory
data, any reason for a referral to a consultant or admission to hospital, any unexpected deterioration (or improvement) in a
concurrent illness, any suspected drug reaction, or any other complaint which was considered of sufficient importance to enter
into patient's notes.
[389]In J-PEM '(any new) symptom or sign and abnormal laboratory data' is added to 'any new diagnosis' to clarify what is to be
reported but otherwise, the definition is exactly the same as that in the UK PEM.
[390]Following the definition, examples are given such as: 'Peptic ulcer, heart burn, anxiety, tachycardia, increase in GOT (AST),
admission to a hospital due to broken leg etc'.
[391]Those examples are further followed by an explanation: "Please report any event even if considered to be irrelevant to the
drug; for example, if more fractures were associated with this drug they could have been due to hypotension, CNS effects or
metabolic bone changes".
[392]The above explanation is exactly the same as that in the green form in the UK PEM used until 1994 when the example was
deleted.
[393]An example of 'event' (broken leg) is shown on the questionnaire used both before and after 1994 in the UK PEM while 6
examples are shown to clarify what is to be reported in J-PEM.
[394]Under the area where events are described in the questionnaire for doctors used in J-PEM,　the following two sentences
are given.
[395]If you judge that the event is likely to be an ADR to any drug used (where the probability that the event is an ADR exceeds
the probability that it is not), add '(probably) an ADR to ---'.
[396]In addition, if that event has been reported to a drug company or MHW etc., please add 'reported to MHW (or the name of
the drug company) as an ADR'.
[397]Of the two lines given above, the former has its counterpart in the questionnaire in the UK PEM which has been used from
1994 where the following is given: "Please indicate if you suspect an EVENT to be an adverse reaction to a drug" (14).
[398]The latter has also its counterpart in the questionnaire in the UK PEM used since 1986 where the following is given
"Please indicate any event reported to CSM or manufacturer" (14).
[399]One more note is given under the area where events are described by doctors in the questionnaire in J-PEM such that:
when reporting abnormal laboratory data as an event, please indicate that the value is outside normal limits like 'GPT42↑' or
'WBC4600↓' as normal range differs between laboratories.
[400]In the questionnaire sent to pharmacists, an event is defined as: Patient's complaint on any symptom or sign (if diagnosis
is known from the patient, describe it), any reason for a referral to a new hospital or admission to hospital, any drug reaction
suspected by doctor or pharmacist, or any other complaint which was considered of sufficient importance to enter into
patient's notes which occurred after the use of the drug monitored.
[401]Definition of an 'event' in the questionnaire to pharmacists is slightly different from that to doctors and the largest
difference is that one component 'any unexpected deterioration (or improvement) in a concurrent illness' in the definition for
doctors is removed from the definition for pharmacists as it may be inappropriate to ask pharmacists whether or not any
deterioration (or improvement) is 'expected' or 'unexpected'.
[402]In addition, the description to be noted when reporting abnormal laboratory data as an event given in the questionnaire to
doctors has been removed from the questionnaire to pharmacists because abnormal laboratory data are rarely reported by
pharmacists and it is considered to be in general irrelevant.
[403]F Sample size
[404]In the UK PEM, the target number of patients to be observed has been set as 10,000 which may be associated with the
description in the original design published in 1980 (30).
[405]In the paper describing the original design, Inman indicated that; Statisticians tell us that to be 95 % certain of detecting a
risk of 0.1 % we would require a minimum of about 3000 observations --- the larger figure of 10,000 is intended to allow for
incomplete reporting.
[406]The similar description is also seen in a document issued from MHW which determines the detailed method of conducting
the 'Drug Use Investigation' (52)
[407]It indicates that 'the number of observations should be that which allows to detect previously unknown ADRs with the rate
of 0.1 % or more with 95 % confidentiality, and therefore, the number should be usually at least 3000.'
[408]However, two descriptions above are in fact different from each other regarding what is to be found by '3000': Inman
indicated 'a risk of 0.1 %' while MHW indicates 'previously unknown ADRs with the rate of 0.1% or more'.
[409]Assuming Poisson distribution when no event is detected in N observations, the upper limit of 95 % confidence of the risk
is approximately 3/N (53).
[410]Nevertheless, the statement 'if no event in 3000 observations then the upper limit of 95 % confidence interval of the risk is
3/3000 (ie, 0.1 %)' does not necessarily imply 'if the risk of an ADR is more than 0.1 %, then the ADR is recognised if at least
3000 are observed'.
[411]Whether or not an even observed is recognised as an ADR depends on the method of how to recognise the relationship.
[412]Under some circumstances, the causal relationship between the event and drug can be recognised by the observation of
a single case (eg, clear results obtained by rechallenge) and if and only if the previously unrecognised ADR is able to be
recognised by a single observation, 3000 observations may allow the detection of that ADR with the risk of 0.1 % or more.
[413]In other words, even if the above statement on 3000 observations given by MHW is derived from a statistical theory, it
assumes that the detection or recognition of a previously unknown ADR is possible by the method which is not statistical.
[414]A 'previously unknown ADR' with the frequency of 0.1% or more expected to detect from the result of a PMS study
observing thousands of patients often has such nature that the causal relationship between the event and drug is difficult to
recognise from the observation of the individual cases.
[415]If the risk of an ADR is as frequent as 0.1 % or more and if that ADR can be recognised from the observation of a single
case, such an ADR is normally expected to detect already in pre-marketing clinical studies or in the first several months after
the marketing the drug before the result of the PMS study observing thousands of patients becomes available.
[416]Supposing that the relationship may be recognised being mediated by the observed significant difference of the rates,
such a signal can be raised only for events with a rate larger than 0.1%.
[417]For instance, in an observational cohort study of 10,000 patients where 5,000 take 'test drug' and the remaining 5,000 take
'control drug', the statistical difference of the risk with the relative risk=5 between two groups can be detected, with α=0.05
and β=0.8, if rate=0.003 or more in the group with 'test drug' (or rate=0.0006 with 'control drug') (54).
[418]If rate=0.001 (0.1 %) with 'test drug' and the relative risk=5 (ie rate= 0.0002 or 0.02 % with 'control drug') are assumed,
approximately 30,000 observations (15,000 for 'test drug' and 15,000 for 'control drug'), 10 times larger than 3,000, are required
in order to detect the relationship 'with 95 % confidentiality' (54).
[419]Probably, total number of 10,000 (5,000 with 'test drug' and 5,000 with 'control drug') may be a realistic number of patients
to be attained in J-PEM, though it takes several years for J-PEM to reach the stage where this goal is usually achieved.
[420]With this number, as shown above, an event with the risk of rate=0.003 (0.3%) is detected as the event where the
difference of the risk is statistically significant between 'test drug' and 'control drug' if the relative risk = 5 or more, while where
the risk is larger than rate=0.003, the difference may be significant even if the relative risk is less than 5.
[421]In a pilot study on losartan, the target number patient codes to be registered is 10,000 (approximately half is for 'losartan'
and the other half for 'control drug') and for 50-60 % of patients registered a pair of questionnaires are issued.
[422]G Methods to assemble data
[423]As described earlier, in PEM, it is normally unpredictable which event is illuminated as an event to be examined in depth
before starting the study.
[424]This means that there is no way to predict what is the 'best' information on possible confounding variables used to
examine some problems associated with a particular event.
[425]To collect the information on possible confounding factors, a practical policy is to ask several factors which may be
considered by doctors when prescribing a patient the test or control drug.
[426]Factors considered when doctors prescribe a patient a drug are varied with different group of drugs (eg, antidepressants,
antihypertensives , antidiabetics or NSAIDs etc.).
[427]In addition, relevant factors may differ among drugs under the same group (eg, antihypertensives).
[428]However, to make a comparison effective, the information to be collected should be the same for drugs to be compared
with one another.
[429]The information collected on possible confounding variables therefore cannot be too specific so that the information on a
limited number of variables likely to be considered by doctors when prescribing most drugs under one group may be obtained.
[430]In the UK PEM, the questionnaires are virtually the same for all of the drugs monitored except for a few exceptional
additional questions but in Japanese version of PEM, the questionnaire consists of two different kinds of questions.
[431]One kind of questions are given in an exactly the same manner for all kind of drugs (eg, questions on events, first day of
prescribing the drug monitored, reason for stopping the drug etc.) but the other kind of questions are specific to the drug group
(eg, information on the base line conditions including available results of a few selected laboratory examinations).
[432]The biggest issue associated with the processes to collect data in PEM is the response rate.
[433]As given above, the expected value of the final response rate from doctors in a pilot study of PEM on troglitazone is a little
more than 35 % which is considerably less than that in the UK PEM (overall response rate is 58 % and 52 % provide clinically
useful data).
[434]To make the response rate higher, several points need to be improved.
[435]One point is to make the questionnaire be simpler; for example, in the pilot study of J-PEM on troglitazone it turned out
that where doctors send back the questionnaire, approximately 97 % of pharmacists who have forwarded a questionnaire to the
doctor also respond and the information on concurrent drugs provided by pharmacists is usually more accurate than that from
doctors.
[436]Based on this finding, virtually all of the questions on concurrent drugs have been removed from the questionnaire for
doctors in a succeeding pilot study on losartan.
[437]As given above, in J-PEM, the amount of questions per patient is larger than that in the UK PEM and approximate time to
fill out one questionnaire in J-PEM (15-20 min) is longer than that in the UK PEM (3 min) which may warrant paying doctor and
pharmacist some amount of money which does not considered to work as an inducement to prescribing any particular drug; cf.
the detailed description under 'Resources'.
[438]Where questionnaires are used to collect information, the response rate cannot reach 100 % and there is a need to
evaluate whether or not patients for whom doctors or pharmacist respond is systematically different from those for whom they
do not.
[439]The comparison between the information provided by pharmacists and that by doctors may be used as a tool to evaluate
this point as the response rate from pharmacists (more than 85 %) is much larger than that from doctors (35-40%).
[440]For example whether or not mortality differs between patients for whom doctors responded and those for whom doctors
did not may be examined though the comparison can be done only among the patients for whom pharmacists have responded.
[441]To make the study noninterventional, doctors are not informed of which particular patient is the study subject until they
receive a questionnaire but pharmacists have some idea about who are possible study subjects as the study subjects are
registered by pharmacists.
[442]In the pilot study of J-PEM on troglitazone, we have found that some of pharmacists monitor patients who are possible
study subjects more closely than the other patients.
[443]The close monitoring by a pharmacist may not usually alter treatment of a patient provided by the doctor so that such a
monitoring may not become a serious problem as long as patients with a 'test drug' and those with a 'control drug' are
monitored with the same level of enthusiasm.
[444]It has been found that pharmacists sometimes report, as an event, a fact that the patient has been examined by some
laboratory test, CT or endoscopy without the information on the result of such a test where the reason why the patient has
been examined would give a better information when reported as an 'event'.
[445]Some idea about how to report events to give the useful information may improve the quality of data if this information is
effectively given to pharmacists.
[446]Currently, the response rate for questionnaires issued directly to patients and problems associated with answers provided
directly by patients who have registered cannot be evaluated as no questionnaires for patients have been issued so far.
[447]H Data management
[448]Currently, all of the data handling as well as the processes to issue registration form and questionnaires are managed by
PCs using a relational database program, Microsoft Access.
[449]When the method of data handling is established after having experienced most problems in pilot studies of J-PEM, the
use of other program suitable for handling large data may be considered, but during the stage of the pilot study, a computer
program for PCs such as Microsoft Access is suitable as the modification of the system is easy whenever needed.
[450]Reports on events are firstly recorded exactly as reported by a doctor or pharmacist and then each event is coded by
Japanese version of 'Adverse Drug Reaction Terminology' (J-ART) developed being based on a Japanese translation of WHO's
'Adverse Reaction Terminology' called as WHO-ART (55) and issued from MHW.
[451]It is planned that MEDDRA, a single global terminology including terms used to describe adverse events agreed between
Europe, the US and Japan, is adopted in J-PEM when its Japanese translation is released and this change would not create any
large problems as all of the terms in J-ART are adopted as formal terms of MEDDRA.
[452]In the UK PEM, an 'event dictionary' has been developed while monitoring various drugs (56) but in J-PEM this is not used
as wording by doctors in the UK is somewhat different from that in Japanese doctors and 'Adverse Drug Reaction Terminology'
which has been developed while coding many reports by Japanese doctors is in general better to code events in J-PEM.
[453]Drugs are coded by using 'Drug Name Database' issued by 'Iyaku-Joho-Kennkyu-Jo' (Drug Information Research
Laboratory), Chiyoda-ku, Tokyo.
[454]Concurrent diseases and concurrent conditions which a patient may have when the drug is prescribed is coded by
International Classification of Diseases (ICD) 10 issued by WHO (57).
[455]Details on how to code events using terminology seems to be a problem which has not been fully addressed so far.
[456]In the UK PEM, events are coded accoding to the following principles　（16-25）.
[457]First, pre-existing diseases are not coded unless an exacerbation occurs.
[458]Second, where the same event occurred more than once, only the first episode is coded.
[459]Third, if an event is a diagnosis or syndrome, related individual signs, symptoms or laboratory test results are not coded.
[460]Fourth, where the events are related, only the more serious is coded, signs taking precedence over symptoms.
[461]Fifth, more than one event in the same class may be coded for one patient.
[462]The above 5 principles are basically agreed in J-PEM though some difference exists; eg, in J-PEM, the second or third
episode of the same event are also coded but with a flag indicating that the episode is not the first one.
[463]Another point adopted in J-PEM is that if a doctor or pharmacist indicates that several events as a group (eg, events have
appeared on the same day), they are coded with a flag which indicates this point so that it is possible to count the number of
some specific combination of events in a special analysis done in addition to the standard listing of event codes to find possible
new problems such as a new syndrome.
[464]To manage a variety of problems encountered in coding events, how to code may be summarised as a booklet which may
be given to data entry personnel in Department of Pharmacoepidemiology (or monitoring center in the future).
[465]I Methods　of data analysis
[466]Though the method of data analysis is being developed, some principles may be now possible to outline.
[467]In a pilot study on troglitazone, it has been found that events reported by doctors and those by pharmacistｓ are not
identical to each other where both of the doctor and pharmacist have returned the questionnaires on the same patient.
[468]For instance, in approximately 50 % of cases where both of the pharmacist and doctor have returned the questionnaire for
the same patient, only either doctor or pharmacist has described one or more events but nothing has been described by his/her
counterpart.
[469]In other 30 %, both of the pharmacist and doctor did not describe any event and only in the remaining 20 %, both of the
doctor and pharmacist described one or more events for the same patient.
[470]In more than half of the last cases where both of the doctor and pharmacist described one or more events for the same
patient, the doctor and pharmacist raised events which are totally different from each other or only a part of events were
identical to each other.
[471]Taking into this point into account, it has been judged to be appropriate to analyse answers from pharmacists and those
from doctors separately.
[472]First, answers from pharmacists of which the response rate is much higher than doctors are analysed to find those events
of which crude event rates are significantly different between 'test drug' and 'control drug'.
[473]Event rates are calculated for each of patient groups with 'test drug' and 'control drug' and the odds ratio (OR) is
calculated to estimate the risk ratio (RR).
[474]Fisher's exact test is used to test the difference of event rates between two patient groups.
[475]Second, the problems illuminated by the analysis of answers from pharmacists associated with the events with crude
event rates significantly different between 'test drug' and 'control drug' are further analysed by answers from doctors where
relatively large amount of the information on possible confounding factors is available.
[476]Serious events which are known to be often potentially an ADR (eg, aplastic anaemia) to drugs may be examined in depth
mainly by the follow-up study.
[477]Careful inspection by human eyes may be also important to pick up any other problems to be examined in depth.
[478]Where appropriate, confounding variables may be taken into account and the adjusted odds ratio is calculated by the
logistic regression analysis for categorical variables and the multiple regression analysis is employed for a small number of
continuous variables.
[479]In PEM, it may be noted that adjustment by confounding variables is likely to be incomplete when analysing the results
because the hypothesis tested cannot be specified prior to the study.
[480]Therefore, if it is found that the difference of crude rates of some event is at least in part explained by the difference of
patients with 'test drug' and those with 'control drug' regarding some confounding variables, this fact may be sometimes judged
to be an evidence suggesting that the event is unlikely to be an ADR.
[481]In some of hospitals (particularly those recruited by JMARI), only patients with 'test drug' but no patients with 'control
drug' are identified to register so that questionnaires will be issued and the information will be obtained only for those patients
with 'test drug'.
[482]In data analysis, the event rate from those data obtained for patients with 'test drug' only is calculated as an additional
analysis and the results will be shown separately from the other results obtained from the data of both of the patients with 'test
drug' and those with 'control drug' are available.
[483]Thus, such data will be mainly used for the descriptive purpose for the time being but for the analytical purpose when the
data from PEM study on another 'test drug' similar to the current 'test drug' become available in the future.
[484]In addition, where only patients with 'test drug' are registered, it is required to identify all of the possible subjects by using
computerised prescription data or by any means so that the results may be considered to be in general more representative of
the patients in participating hospitals as compared with the results obtained from patients with 'test drug' and 'control drug'
identified by pharmacists often only for a fraction of whole patients who are eligible as study subjects.
[485]J　Quality assurance
[486]In the pilot studies, all of the event codes and ICD codes for the concurrent diseases coded by any data entry personnel
are reviewed in a meeting held once per one to three weeks in Department of Pharmacoepidemiology in order to standardise
the method of coding.
[487]In the future, the validity of coding may be checked with other professional(s) being independent of data entry personnel.
[488]In the future some program to examine whether any bias is introduced when patients are registered may be designed.
[489]In addition, a validation study may be conducted to check the compatibility between the description on the original
patient's note and that in a questionnaire.
[490]Another problem may be the validation of the data obtained in the follow-up study of which the method of examining the
validation may be examined in the future.
[491]Regarding the analysis of data obtained directly from patients in the long-term study, most problems will be discussed in
the future as even in the first J-PEM pilot study on troglitazone, no questionnaire has been issued to patients registered by
themselves.
[492]K Limitations of the study
[493]As described earlier one of the problems inherent to the study using questionnaires is that the response rate which
cannot reach 100 %.
[494]The amount of the information including that associated with confounding variables is minimised not to make the response
rate too low and what are the best confounding variables which differ between events to be examined cannot be exactly defined
because events emerging from the inspection of the data by human eyes or the simple comparison of crude rates between 'test
drug' and 'control drug' as those to be scrutinised in depth cannot be predicted prior to the study.
[495]Unless the number of participating pharmacies and hospitals is much increased or other method such as the systematic
use of 'Rezept' data which can identify patients which represent whole drug users, there is a room to raise questions on the
generalizability of the study results.
[496]A random error is directly associated with the sample size and it should be much increased from that in the pilot study in
the future.
[497]Protecting human subjects
[498]In PEM where the study is conducted as a strictly noninterventional study, there is little possibility that the patient is
placed at risk as a result of the study.
[499]In addition to the noninterventional nature, the privacy is carefully protected as code number is used both for patients and
doctors.
[500]Thus, the informed consent is not an issue which is requisite but where possible a poster is made and posted inside
participating pharmacies and hospitals noticing patients the following: "a PEM study is being conducted where your data are
potentially used" .
[501]Though the poster is made and distributed by Department of Pharmacoepidemiology, it is posted by pharmacies and
hospitals after entering their names, by hand, as those who have the responsibility for posting it.
[502]Strict confidentiality of the information of the doctor or pharmacist who have returned the questionnaire with his/her
signature as well as that of the patients who have registered by themselves will be maintained.
[503]Publication of results
[504]A table where the number of pharmacies and hospitals which participate in the study and the number of patient codes
registered so far are shown for each prefecture, another table where the number of questionnaires sent and that returned from
pharmacists and doctors are shown, as well as the two kinds of lists (obtained from doctors and from pharmacists) of the
number of all kinds of events coded for 'test drug' and 'control drug' will be presented by internet in homepage of Department of
Pharmacoepdemiology （http://square.umin.ac.jp/~pe//） and those data will be replaced once to twice a month.
[505]With the list of the number of events the following two notes will be added: �@ "The list below indicates events observed
after prescribing each drug but not ADRs and most of reported events are not associated with drugs" and �A "Even if the
crude rates are different between 'test drug' and 'control drug', the difference may be associated with the factors other than
drugs and the final conclusion will be made after carefully taking various variables into account.'
[506]The final results will be published elsewhere in medical journal as a scientific paper.
[507]Resources
[508]Excluding time used to design forms or analyse data, time for routine work needed to send registration forms, enter patient
codes registered from participating pharmacists or hospitals, send pairs of questionnaires for pharmacist and doctor, code the
information obtained from the returned questionnaires and make various inquiries about registration etc. is calculated to be
around 30-45 min per patient though this time may be slightly shortened by improving the system.
[509]Total cost for stamps, papers and envelops, fee for Xerox and filing returned questionnaires in the pilot study on
troglitazone has been calculated to be \1,500-2,000 per each patient code for whom a questionnaire has been issued though
this may be considerably reduced if the most efficient way is found and fixed.
[510]In addition to computers and computer program, one to two full time worker(s) are needed to maintain the system to
manage registration of patients, sending questionnaires and coding the information on the questionnaires answered.
[511]It may be noted that the average time to fill out one questionnaire in J-PEM (15-20 min) is much longer than that in the
UK PEM (3 min) because the number of questions in one questionnaire sent to doctors in J-PEM is larger than those in the UK
PEM.
[512]It is impossible to compare the amount of work done by the pharmacist between Japan and the UK because pharmacists
are not involved in the UK PEM but the work load of a pharmacist may be judged to be not trivial.
[513]Though in the first pilot study of J-PEM having been conducted as a study on troglitazone since July 1997, no money has
been paid to doctor and pharmacist, in the pilot study of J-PEM on losartan, \500 is paid per questionnaire returned with useful
information irrespective of whether the person who has responded is a doctor or pharmacist.
[514]To the pharmacy inside the hospital which has been co-operated in passing the questionnaire forwarded from pharmacists
in pharmacies outside hospital to a doctor, \500 is paid per 3 questionnaires with useful information returned from doctors to
whom the questionnaire has been passed via that pharmacy inside the hospital.
[515]In the questionnaire sent to a pharmacist of the pharmacy outside hospital, the date when the questionnaire to a doctor
on that patient was forwarded as well as whether the questionnaire to a doctor was forwarded directly to the doctor or via
pharmacy inside the hospital are requested to answer.
[516]Therefore, it is possible to know how many questionnaires to doctors have been returned with useful information via a
particular hospital pharmacy.
[517]In the future, a little higher cost (at least \1,000 per questionnaire returned) may be paid as a reward.
[518]Currently, all of the work is managed in Department of Pharmacoepidemiology, but, when more than 2 drugs are monitored
simultaneously, the number of persons and space of work place required will exceed those one Department in a university could
provide.
[519]In such a case, maintenance cost for housing, electricity etc should be counted when resources to be used are calculated.
[520]The pilot study of J-PEM on losartan is conducted mainly as a Health Science Research supported by MHW but how to
maintain J-PEM as a stable scheme particularly in association with the 'Drug Use Investigation' is to be discussed in the near
future.
[521]Amendments to the protocol
[522]Any significant change in the protocol will be documented in writing with the date when such a change is made as well as
its reason.
[523]Note: This protocol has been prepared according to guidelines made by ISPE members (58)　'Guidelines for good
epidemiology practices for drug, device, and vaccine research in the United States'.
[524]References
[525]1 Inman WHW. PEM News No.1, Southampton; Drug Surveillance Research Unit, 1983.
[526]2 Inman WHW, Weber JCP. The United Kingdom. In: Inman WHW, ed. Monitoring for Drug Safety, 2nd ed. Lancaster;
MTP Press, 1985: 13-47.
[527]3 Davies DM. History. In: Inman WHW, ed. Monitoring for Drug Safety, 2nd ed. Lancaster; MTP Press, 1985: 1-4.
[528]4 Inman WHW. Post-marketing surveillance in the general population. In: Inman WHW, ed. Monitoring for Drug Safety,
2nd ed. Lancaster; MTP Press, 1985: 3-11.
[529]5 Ahmad SR, Freiman JP, Graham DJ, Nelson RC. Quality of adverse drug experience reports submitted by pharmacists
and physicians to the FDA. Pharmacoepidemiiolo Drug Safety 1996; 5: 1-7.
[530]6 Inman WHW, Rawson NSB, Wilton LV. Prescription-Event Monitoring. In: Inman WHW, ed. Monitoring for Drug
Safety, 2nd ed. Lancaster; MTP Press, 1985: 213-35.
[531]7 Felix RH, Ive FA. Cutaneous and ocular reactions to practolol. BMJ 1974; 4: 321-324.
[532]8 Wright P. Untoward effects associated with practolol administration: BMJ 1975; 1: 595-598.
[533]9 Brown P, Baddely H, Read AE, Davies DJ, McGarry J. Sclerosing peritonitis, an unusual reaction to a β-
adrenergic-working drug (practolol). Lancet 1974; 2: 1477-1481.
[534]10 Inman WHW. PEM News No. 7, Southampton: Drug Safety Research Unit; 1991.
[535]11 Tsubaki T, Homma Y, Hoshi M. Neurological syndrome associated with clioquinol. Lancet 1971; 1: 696-697.
[536]12 Tsubaki T. Etiology of SMON: an early study and its development. In: Soda T, ed. Drug-Induced Sufferings.
Amsterdam, Excerpta Medica, 1980: 423-428.
[537]13 Inman B. 30 years in postmarketing surveillance: a personal perspective. Pharmacoepidemiol Drug Safety 1993; 2:
239-258.
[538]14 Mann RD. Prescription-event monitoring --- recent progress and future horizons. Br J Clin Pharmacol 1998; 46:
195-201.
[539]15 Strom BL. How should one perform pharmacoepidemiology studies ?: choosing among the available alternatives. In:
Strom BL ed. Pharmacoepdemiology, 2nd ed. Chichester: John Wiley & Sons Ltd. 1994: 337-350.
[540]16 Kubota K, Kubota N, Pearce GL, Prescott P, Mann RD. Signalling drug-induced rash with 36 drugs recently marketed
in the United Kingdom and studied by Prescription-Event Monitoring. Int J Clin Pharmacol Ther 1995; 33: 219-225.
[541]17 Inman W, Kubota K, Wilton L, Pearce G. Nicardipine: PEM report number 1. Pharmacoepidemiol Drug Safety 1993; 2:
259-285.
[542]18 Inman W, Kubota K, Pearce G, Wilton L. Fluconazole: PEM report number 3. Pharmacoepidemiol Drug Safety 1993; 2:
321-340.
[543]19 Inman W, Kubota K, Pearce G, Wilton L. Ciprofloxacin: PEM report number 4. Pharmacoepidemiol Drug Safety 1993; 2:
341-364.
[544]20 Inman W, Kubota K, Pearce G, Wilton L. Fluoxetine: PEM report number 5. Pharmacoepidemiol Drug Safety 1993; 2:
365-391.
[545]21 Inman W, Kubota K, Pearce G, Wilton L. Paroxetine: PEM report number 6. Pharmacoepidemiol Drug Safety 1993; 2:
393-422.
[546]22 Inman W, Kubota K, Pearce G, Wilton L. Itraconazole: PEM report number 7. Pharmacoepidemiol Drug Safety 1993; 2:
423-443.
[547]23 Inman W, Kubota K, Pearce G, Wilton L. Cisapride: PEM report number 8. Pharmacoepidemiol Drug Safety 1993; 2:
445-471.
[548]24 Wilton L, Inman W, Kubota K, Pearce G, Skipp M. Omeprazole: PEM report number 9. Pharmacoepidemiol Drug Safety
1993; 2: 473-498.
[549]25 Inman W, Kubota K, Pearce G, Wilton L. Zopiclone: PEM report number 10. Pharmacoepidemiol Drug Safety 1993; 2:
499-521.
[550]26 Edwards JG, Inman WHW, Wilton L, Pearce G. Prescription-Event Monitoring of 10401 patients treated with
fluvoxamine. Br J Pscyichiatr 1994; 164: 387-395.
[551]27 Edwards JG, Inman WHW, Wilton L, Pearce G, Kubota, K: Drug Safety Monitoring of 12692 Patients Treated with
Fluoxetine: Hum Psychopharmacol. 12: 127-137, 1997.
[552]28 Fletcher RH, Fletcher SW, Wagner EH. Risk. In: Clinical Epidemiology. 3rd ed. Baltimore: Williams & Wilkins,
1996: 94-110.
[553]29 Inman WHW. Postmarketing surveillance of adverse drug reactions in general practice. I: Search for new methods.
Br Med J 1981; 282: 1131-1132.
[554]30 Inman WHW. Postmarketing surveillance of adverse drug reactions in general practice. II: Prescription-event
monitoring at the University of Southampton. Br Med J 1981; 282: 1216-1217.
[555]31 Inman WHW. PEM News No.2, Southampton: Drug Surveillance Research Unit, 1984.
[556]32 Inman WHW. PEM News No3, Southampton: Drug Surveillance Research Unit, 1985.
[557]33 Inman W, Pearce G. Prescriber profile and post-marketing surveillance. Lancet 1993; 342: 658-661.
[558]34 Dunn NR, Freemantle SN, Pearce GL, Mann RD. Galactorrhoea with moclobemide. Lancet 1998; 351: 802.
[559]35 Inman WHW, Rawson NSB, Wilton LV, Pearce GL, Speirs CJ. Postmarketing surveillance of enalapril. I: Results of
prescription-event monitoring. Br Med J 1988; 297: 826-829.
[560]36 Speirs CJ, Dollery CT, Inman WHW, Rawson NSB, Wilton LV. Postmarketing surveillance of enalapril. II: Investigation
of the potential role of enalapril in deaths with renal failure. Br Med J 1988; 297: 830-832.
[561]37 Bjorck S, Mulec H, Johnsen SA, Norden G, Aurell M. Renal protective effect of enalapril in diabetic nephropathy. Br
Med J 1992; 304: 339-343
[562]38 Chan JCN, Cockram CS, Nicholls MG, Cheung CK, Swaminathan R. Comparison of enalapril and nifedipine in treating
non-insulin dependent diabetes associated with hypertension: one year analysis. Br Med J 1992; 305: 981-985.
[563]39 Hallab M, Gallois Y, Chatellier G, Rohmer V, Fressinaud P, Marre M. Comparison of reduction in microalbuminuria by
enalapril and hydrochlorothiazide in normotensive patients with insulin dependent diabetes. Br Med J 1993; 306: 175-182.
[564]40 Lewis E, Hunsicker LG, Bain RP, Rohde RD. The effect of angiotensin-converting-enzyme inhibition on diabetic
nephropathy. N Engl J Med 1993; 329: 1456-1462.
[565]41 Ravid M, Lang R, Rachmani R, Lishner M. Long-term renoprotective effect of angiotensin-converting enzyme inhibition
in non-insulin-dependent diabetes mellitus: a 7-year follow-up study. Arch Intern Med 1996; 156: 286-289.
[566]42 Maschio G. Protecting the residual renal function: How do ACE inhibitors and calcium antagonist compare ?
Nephron 1994; 67: 257-262.
[567]43 Fluck RJ, Raine AEG. ACE inhibitors in non diabetic renal disease. Br Heart J 1994; 72 (Suppl): 46-51.
[568]44 Hricik DE, Browning PJ, Kopelman R, et al. Captopril-induced functional renal insufficiency in patients with bilateral
renal artery stenosis or renal artery stenosis in a solitary kidney. N Engl J Med 1983; 308:373-376.
[569]45 Ravid M, Ravid D. ACE inhibitors in elderly patients with hypertension. Drugs & Aging 1996; 8: 29-37.
[570]46 Jick SS, Dean AD, Jick H. Antidepressants and suicide. Br Med J 1995; 310: 215-218.
[571]47 Jick H, Jick SS, Gurewich V, Myers MW, Vasilakis C. Risk of idiopathic cardiovascular death and nonfatal venous
thromboembolism in women using oral contraceptives with differing progetagen components. Lancet 1995; 346: 1589-1593.
[572]48 Jick H, Jick S, Derby LE, Vasilakis C, Myers MW, Meier CR. Calcium-channel blockers and risk of cancer. Lancet
1997; 349: 525-528.
[573]49 Dodriguez LAG, Ruigomez A, Jick H. A review of epidemiologic research on drug-induced acute liver injury using the
general practice research data base in the United Kingdom. Pharmacotherapy 1997; 17: 721-728.
[574]50 Breslow NE, Day NE. The role of cohort studies in cancer epidemiology. In Statistical Methods in Cancer
Research. Vol II The design and analysis of cohort studies. Lyon, IARC. 1987: 2-46.
[575]51 Joint National Committee. The Sixth Report of the Joint National Committee on Prevention, Detection, Evaluation,
and Treatment of High Blood Pressure. Arch Intern Med 1997; 157: 2413-2446.
[576]52 MHW. Guidelines for the method of conducting 'Drug Use Investigation' etc of medicines. 1997.03.27 (Yaku-ann No
34)
[577]53 Arme-P. 'If nothing goes wrong, it everything all right ?' In: Methodological approaches in pharmacoepidemiology:
Application to spontaneous reporting. Amsterdam, Elsevier. 1993: 89-94.
[578]54 Strom BL. Sample size consideration for pharmacoepidemiology study. In: Strom BL ed. Pharmacoepdemiology,
2nd ed. Chichester: John Wiley & Sons Ltd. 1994: 29-38.
[579]55　WHO. International monitoring of adverse drug reactions to drugs: adverse reaction terminology, Uppsala: WHO
collaborating Centre for International Drug monitoring, 1991.
[580]56 Kubota K, Inman WHW: Terminology in Prescription-Event Monitoring. Eur J Clin Pharmacol. 46:497-500, 1994.
[581]57 WHO. International Statistical Classification of Diseases and Related Health Problems. Tenth Revision. (ICD-10)
Geneva 1993.
[582]58 Guidelines for good epidemiology practices for drug, device, and vaccine research in the United States.
Pharmacoepidemiology and Drug Safety 5: 333-338, 1996.