`
`Ó Springer-Verlag 1997
`
`P H A R M A C O E P I D E M I O L O G Y A N D P R E S C R I P T I O N
`
`P. E. Gro¨ nroos á K. M. Irjala á R. K. Huupponen á H. Scheinin
`J. Forsstro¨ m á J. J. Forsstro¨ m
`A medication database ± a tool for detecting drug interactions in hospital
`
`Received: 15 April 1996 / Accepted in revised form: 17 June 1996
`
`Abstract Objective: Drug interactions may lead to life-
`threatening injuries. More often, however, they lead to
`slow recovery, induce slight symptoms or result only in
`potential
`injury. Therefore, clinicians are not always
`aware of using potentially interacting drug combina-
`tions. An on-line alarming system of potential drug in-
`teractions was developed in Turku University Central
`Hospital. In the present study, we utilised the system to
`find out the incidence and nature of potential drug in-
`teractions occurring in a representative hospital patient
`population.
`Methods: Computerised anatomical therapeutic chemi-
`cal (ATC)-coded patient medication data of 2547 pa-
`tients, treated in two internal medicine wards, were
`combined with an ATC-coded rule base of drug inter-
`actions. All potential drug interactions in the study
`population were searched for.
`Results: A total of 326 potentially serious drug interac-
`tions were detected in the study population. The number
`of patients in this group was 173, i.e. 6.8% of all patients
`had one or several drug combinations which might have
`led to serious clinical consequences. Concomitant use of
`calcium and fluoroquinolones (decreased absorption)
`was the most common mistake (66 prescriptions).
`Conclusions: Potentially inappropriate drug combina-
`tions seem to occur frequently. Structured and coded
`medication data can be utilised e(cid:129)ciently to detect po-
`tential drug interactions in hospital. Computerised on-
`line monitoring and automatic alarming of potentially
`hazardous drug combinations might help clinicians to
`
`P.E. Gro¨ nroos (&) á K.M. Irjala
`Central Laboratory, Turku University Central Hospital,
`Kiinamyllynkatu 4–8, FIN-20520 Turku, Finland
`Tel +358-2-2612914; Fax +358-2-2613920;
`e-mail paula.gronroos@utu.fi
`R.K. Huupponen á H. Scheinin
`Department of Pharmacology and Clinical Pharmacology,
`University of Turku, Finland
`J. Forsstro¨ m á J.J. Forsstro¨ m
`Department of medicine, University of Turku, Finland
`
`prescribe more safely, but further development of the
`system is needed to avoid unnecessary alarms.
`
`Key words Drug Interactions, Hospital
`
`Introduction
`
`A substantial part of medical treatments lead to injuries
`[1]. The most common reason (19.4%) for these injuries
`is drug complications [2], which are often due to errors
`in the use of drugs [3]. According to previous studies,
`medication errors occur in 2–14% of patients admitted
`to hospitals [4, 5], but fortunately, most do not result in
`injury [6, 7]. However, the goal should be that no errors
`reach the patient [8]. Computerised approaches are ideal
`for this because reliability can approach 100%, while
`methods that rely on human inspection will always miss
`some errors [3].
`According to Bates et al., the leading causes of
`medication errors – drug interactions, negligence of
`known allergies, overdoses, underdoses, wrong choices
`and wrong medication frequencies – were found to be
`potentially preventable by computerised order checking
`[3]. In Turku University Central Hospital (TUCH), an
`integrated computerised system with a structured med-
`ication database was introduced to detect and avoid
`medication errors. Our interest focuses on warning of
`drug interactions and known allergies as well as drug
`eects on laboratory tests.
`Cumulative individual patient medication data are
`stored continuously in an ATC-coded medication data-
`base. The structured form of the medication database
`enables us to process and utilise the medication data in
`several applications [9]. The medication database can be
`combined with structured knowledge and rule bases,
`which makes automatic alarming of errors possible. We
`already have a structured knowledge base for drug in-
`teractions [10] and we are building one for drug eects
`on laboratory tests [11]. Further, the medication data-
`base includes an option to store the reason for discon-
`
`CFAD VI 1022-0001
`
`
`
`14
`
`tinuing a certain medication. If someone intends to start
`the same medication again, the system will automatically
`display the earlier reason for discontinuation, for ex-
`ample an allergic reaction.
`Drug interactions are typical examples of medication
`errors that may lead to serious injuries [12], but are
`potentially preventable by computer systems [3]. In our
`system, the interaction database is integrated into the
`medication database for continuous monitoring of the
`current medication profiles of individual patients. On-
`line alarms of potential drug interactions can be pro-
`duced directly to clinicians on the wards. To evaluate the
`system and to find out the incidence and nature of po-
`tential drug interactions occurring in a hospital patient
`population, a retrospective study utilising the combined
`medication and interaction databases was performed.
`
`Subjects and methods
`
`Study population
`
`A total of 2547 patients were included in the study. The study
`population consisted of patients treated in two internal medicine
`wards of the TUCH. The two study wards were the nephrological
`(595 patients) and the cardiological units (1952 patients). The in-
`dividual medication data of all patients in these units were man-
`aged by the computerised system. The nephrological unit had used
`the system for 13 months and the cardiological unit for 7 months.
`These periods determined the respective periods of gathering the
`data. In the study wards, automatic alarms of drug interactions
`were not used and the wards were not aware of the study.
`The study wards are representative examples of great drug
`consumers in hospital and, therefore, drug interactions are likely.
`In the nephrological unit, drug treatments are far more complicated
`than in most other departments. The most di(cid:129)cult cases of hy-
`pertension, complications of autoimmune diseases or diabetes, as
`well as severe infections in immunocompromised patients are quite
`common. The cardiological unit covers the most usual diagnoses in
`internal medicine and in earlier studies [13], cardiovascular drugs
`were found to be represented in the majority of potential drug
`interactions.
`
`Medication database and coding of medication data
`
`The medication data of individual patients were stored in the
`medication database,
`in which the trade names of drugs were
`converted into their respective anatomical therapeutic chemical
`(ATC) codes. The ATC code [14] was used for coding drugs also in
`the interaction database. The basic data structure in the medication
`database is called a medication line. The information in each line
`consists of nine fields: (1) social security number (identification) of
`the patient; (2) ward; (3) trade name and strength of the drug; (4)
`pharmaceutical form of the drug; (5) dose of the drug; (6) ATC
`code of the drug; (7) date of onset of the medication; (8) date of
`stopping the medication; and (9) reason for stopping the medica-
`tion, if required. Typically, one patient has several medication lines.
`
`Interaction database
`
`The data on drug interactions were based on FASS (Farm-
`akologiska Specialiteter i Sverige) 1995 [10]. FASS is the Swedish
`physician’s desk reference containing all registered drugs in Swe-
`den. It includes a comprehensive chapter on drug interactions
`gathered by Professor Sjo¨ qvist [10] over 20 years. Overall, the in-
`teraction catalogue includes 671 drug–drug interactions or inter-
`
`them are classified
`actions between drug groups and all of
`according to the clinical importance and level of scientific docu-
`mentation. The clinical importance of the interaction is coded with
`letters from A to D. Letter A corresponds to ‘‘probably no clinical
`importance’’, B to ‘‘clinical importance not yet confirmed’’, C to
`‘‘combinations which may require a modified drug dosage sched-
`ule’’ and D to ‘‘interactions which may result in serious clinical
`consequences’’. The level of scientific documentation is coded with
`numbers from 1 to 4. Number 1 refers to ‘‘incomplete case re-
`ports’’, 2 to ‘‘well-documented case reports’’, 3 to ‘‘studies with
`healthy volunteers’’ and 4 to ‘‘controlled studies of the relevant
`patient material’’. For each interaction, the catalogue includes a
`comment on the nature of interaction and, if possible, short in-
`structions on how to avoid the interaction. Our interaction data-
`base was built on the basis of the FASS data.
`
`Combining the medication database with the interaction database
`
`All drug treatments stored in the medication database at the study
`wards during the study periods were analysed. The interaction
`database, including the ATC codes of the interacting drug combi-
`nations listed in FASS 1995, was used to find out ‘‘forbidden’’
`combinations in the medication database. As a result, we obtained
`a list of all potential drug interactions in the study population. The
`list included trade names, forms, doses and ATC codes of the in-
`teracting drugs, social security numbers and wards of the patients,
`dates of starting and stopping the medication as well as clinical
`importance and level of documentation of the detected interactions.
`
`Results
`
`A total of 22 508 prescriptions were stored in the medi-
`cation database in the study wards during the follow-up
`periods (7 and 13 months). The drugs most commonly
`prescribed in the study wards are listed in Table 1. The
`number of patients receiving two or more drugs con-
`currently was 2347.
`Potentially serious interactions, i.e. interactions be-
`longing to FASS group D, occurred in 326 prescriptions
`out of 22 508 (1.4%). The number of patients in this
`
`Table 1 Drugs most commonly prescribed in study wards
`
`Drug or drug group
`
`Number of prescriptions
`
`Diuretics
`Antibiotics
`b-Adrenoceptor blockers
`Long-acting nitrates
`Calcium channel blockers
`Acetylsalicylic acid
`Hypnotics and sedatives
`Antidiabetic therapy
`Antiepileptics
`Corticosteroids for systemic use
`ACE inhibitors
`Peptic ulcer therapy
`Digoxin
`Dipyridamole
`Nonsteroidal anti-inflammatory drugs
`Calcium salts
`Potassium salts
`Oral anticoagulants
`Laxatives
`Iron
`
`1852
`1818
`1403
`1370
`1145
`1121
`1077
`814
`812
`777
`761
`679
`567
`523
`497
`469
`427
`373
`308
`188
`
`CFAD VI 1022-0002
`
`
`
`group was 173. This means that 6.8% of all patients and
`7.4% of patients taking two or more concurrent drugs
`had one or several drug combinations which might have
`led to serious clinical consequences and, therefore, ought
`to have been avoided. Most of these combinations were
`potentially hazardous due to increased toxicity and the
`rest of them were potentially ineective due to decreased
`absorption of either drug. Potentially toxic drug com-
`binations occurred in 222 prescriptions (1.0%) covering
`121 patients (4.8%). Potentially ineective drug combi-
`nations due to decreased absorption occurred in 104
`prescriptions (0.5%), covering 56 patients (2.2%). Four
`patients (0.2%) had both potentially toxic and ineec-
`tive drug combinations. If we classify these 326 ‘‘group
`D interactions’’ according to the level of documentation
`by FASS, 54 interactions belonged to class 4, 239 to
`class 3, 15 to class 2 and 18 to class 1. Consequently,
`89.9% of the potentially serious interactions detected by
`the system belonged to FASS documentation class 3
`(‘‘healthy volunteers’’) or 4 (‘‘relevant patient material’’)
`and can be considered to be well documented. Fur-
`thermore, 1460 (57.3%) patients were exposed to inter-
`actions classified under groups C, B and A, but these
`prescriptions were not scrutinised further because of
`their minor clinical importance.
`The ten most frequent potentially serious interactions
`between drugs or drug groups and the nature of inter-
`actions are listed in Table 2. These top ten interactions,
`286 in total, correspond to 87.7% of all potentially se-
`rious interactions detected.
`
`Discussion
`
`Prescribing potentially inappropriate drug combinations
`was relatively common in our hospital. Potentially seri-
`ous drug interactions occurred in 1.4% of the prescrip-
`tions and in 7.4% of the patients taking two or more
`
`15
`
`drugs concurrently. Seventy per cent of these patients
`had potentially toxic drug combinations and 32% had
`potentially ineective combinations due to decreased
`absorption of either drug. Furthermore, in our study
`population, 57.3% of the patients were exposed to po-
`tential interactions of minor clinical importance.
`According to Linnarsson’s study [13]
`in primary
`health care, potential drug interactions occurred in 12%
`of patients receiving two or more concurrent drugs and
`1.9% of all prescriptions resulted in a potential drug
`interaction. In a study by Dambro and Kallgren [15] in a
`family practice, 9.2% of the study patients were pre-
`scribed drugs with known interaction potential. Seven-
`teen per cent of
`these potential
`interactions were
`considered to be of ‘‘major’’ clinical significance. In both
`studies, the concurrence of potentially interacting drugs
`was estimated from overlapping prescriptions, while in
`our study, the concurrence was determined by the exact
`dates of starting and stopping the medications. The
`dierences between study populations in hospital and
`primary health care may also have had an impact on the
`incidence of potential drug interactions. In an overview
`of drug interaction screening, Jankel and Speedie [16]
`have evaluated 19 studies aiming at measuring the fre-
`quency of drug interactions. The incidence of all po-
`tential drug interactions varied from 2.2% to 70.3%.
`Dierences in study designs, methodologies, popula-
`tions and definitions have probably again contributed to
`the considerable variation in the reported incidence
`rates.
`Our study was carried out in a hospital setting in two
`internal medicine wards. Here, the clinicians are experts
`in drug treatments and, therefore, complications of drug
`treatments should be quite rare. In primary health care,
`the incidence of potential drug interactions can be an-
`ticipated to be higher because a wider range of drugs is
`used by general practitioners than hospital specialists,
`who are often experts in certain drug treatments. In
`
`Table 2 The ten most frequent potentially serious drug interactions, number of prescriptions and nature of interaction
`
`Drug(s)
`
`Drug(s)
`
`Number
`
`Nature of interaction
`
`Calcium salts
`
`Fluoroquinolones
`
`Potassium salts
`Verapamil
`Warfarin
`
`Warfarin
`
`Warfarin
`
`Iron
`
`Sucralfate
`
`Morphine
`Diltiazem
`
`Spironolactone
`b-Adrenoceptor blockers
`Acetylsalicylic acid
`
`Amiodarone
`
`Nonsteroidal
`anti-inflammatory drugs
`
`Fluoroquinolones
`
`Fluoroquinolones
`
`Barbiturates
`Nifedipine
`
`66
`
`54
`38
`27
`
`23
`
`23
`
`20
`
`16
`
`10
`9
`
`Calcium inhibits the absorption
`of fluoroquinolones
`Risk for hyperkalaemia
`Risk for bradycardia
`Potentiation of anticoagulation, inhibition
`of thrombocyte function
`Amiodarone inhibits the metabolism
`of warfarin
`Risk for gastrointestinal bleeding due
`to inhibition of platelet aggregation
`and damage of the gastrointestinal
`epithelium
`Iron inhibits the absorption
`of fluoroquinolones
`Sucralfate reduces the absorption
`of fluoroquinolones
`Enhanced depressive eect on respiration
`Diltiazem decreases the clearance
`of nifedipine
`
`CFAD VI 1022-0003
`
`
`
`16
`
`hospital, colleagues may also check each other’s pre-
`scriptions. Obviously, the use of potentially inappro-
`priate drug combinations is sometimes unavoidable.
`When we scrutinised the 286 potential drug interactions
`listed in Table 2, it turned out that many of the detected
`‘‘forbidden’’ drug combinations, for example spirono-
`lactone and potassium chloride as well as verapamil and
`beta-adrenoceptor blockers, were used deliberately. The
`risk of interaction was considered and the doses of drugs
`in question were adjusted to be safe and eective. In
`these cases, specialists may find automatic interaction
`alarms unnecessary and frustrating, but the same alarms
`are probably of great educational importance for junior
`doctors.
`Although potentially serious drug interactions are
`relatively frequent, they seldom lead to serious injuries
`[15, 17]. However, the resulting injuries may be devas-
`tating [12] while often preventable by computer systems
`[3]. We believe that hints of potentially inappropriate
`drug combinations should help clinicians to prescribe
`more safely. In our system, on-line alarms of potentially
`serious drug interactions can be produced directly to
`clinicians in the wards. The system automatically checks
`for all potential drug interactions in the current patient
`medication profiles, not only those suspected by the
`clinicians. This automation is an important aspect, be-
`cause computer-based tools that require additional ef-
`forts, beyond the usual routine, do not easily gain wide
`acceptance among clinicians [18, 19].
`Computers are supposed to provide information
`needed to prescribe safely [20]. As for drug interactions,
`the type of interaction is of particular importance.
`Therefore, the type of interaction and short instructions
`how to avoid the interaction are included in the alarm
`given to clinicians, as soon as the medication is stored in
`the medication database, i.e. before the medication is
`given to the patient. Also, the pharmaceutical form of
`the drug plays a major role in drug interactions. If, for
`example, either of the ‘‘interacting’’ drugs is given par-
`enterally,
`it is unnecessary to warn about decreased
`absorption of either drug. In these cases, the nature of
`interaction can be seen in the instruction mentioned
`above, but we should devise a better strategy to avoid
`this problem of ‘‘false alarms’’. Unfortunately, the ATC
`code does not dierentiate between parenteral and per-
`oral dosage of drugs.
`Overall, the ATC coding for drugs appeared to
`function very well. The hierarchical structure of the
`ATC code is practical in our application, because often a
`whole group of drugs interacts with certain drugs. The
`group, for example beta adrenoceptor blockers, can be
`defined with one single ATC code in the interaction
`database but can still be found in the medication data-
`base as individual beta blockers. The only major error in
`the detected potential drug interactions was the combi-
`nation of sodium bicarbonate and ciprofloxacin. The
`reason for this ‘‘false alarm’’ was the fact that, according
`to our interaction database, antacids interact with fluo-
`roquinolones. However, only magnesium, aluminum
`
`and calcium interact with fluoroquinolones [21–23].
`Therefore, in this case, the individual ATC codes for
`antacids containing Mg, Al and Ca should be fed into
`the rule base, instead of the whole group of antacids.
`Neither type of ‘‘false alarms’’ mentioned above was
`included in the reported incidence rates of potential drug
`interactions.
`the main reasons for
`ATC coding was one of
`choosing the FASS as the source of interaction data.
`Another important aspect was the classification used in
`the FASS. It oers an ability to limit the drug interac-
`tion screen to a certain level of clinical significance and
`documentation. However, the final classification of se-
`riousness of the potential
`interaction remains,
`in all
`cases, a clinician’s personal opinion and will, above all,
`vary with the clinical situation.
`The structuring and coding of data in our system is
`the basis for the functioning of the system. Once the
`ATC-coded medication database is built, it can be used
`to avoid medication errors in several connections. Apart
`from drug interactions, overdoses and underdoses could
`be detected. As mentioned earlier, our system already
`monitors known allergies. We are also planning to build
`an alarming system for drug eects on laboratory tests
`[11]. Furthermore, clinical practice can be monitored
`eectively. For instance,
`it is possible to check how
`guidelines
`for
`treating patients with hypercholes-
`terolaemia or hypertension are followed. This kind of
`monitoring helps us to identify the actual problems in
`medication, facilitates quality assurance and enables
`drug utilisation research, which is becoming an impor-
`tant tool in health economics.
`Even if computers can never replace clinical judge-
`ment, computer-based tools assist prescribing in various
`ways [19]. However, there is a lack of studies establishing
`the real benefits brought about by these systems. In this
`study, we have shown how detailed information about
`clinical practice can be easily obtained by means of a
`computerised hospital information system and discussed
`how this system can be utilised to avoid errors in drug
`treatments. In the near future, our aim is to perform a
`prospective study to evaluate, objectively, the impact of
`our alarming system on the quality and costs of patient
`care.
`
`References
`
`1. Brennan TA, Leape LL, Laird NM, Hebert L, Localio AR,
`Lawthers AG, Newhouse JP, Weiler PC, Hiatt HH (1991) In-
`cidence of adverse events and negligence in hospitalized pa-
`tients – results of the Harvard Medical Practice Study I. N Engl
`J Med 324: 370–376
`2. Leape LL, Brennan TA, Laird N, Lawthers AG, Localio AR,
`Barnes BA, Hebert L, Newhouse JP, Weiler PC, Hiatt H (1991)
`The nature of adverse events in hospitalized patients – results of
`the Harvard Medical Practice Study II. N Engl J Med 324: 377–
`384
`3. Bates DW, Cullen DJ, Laird N, Petersen LA, Small SD, Serri
`D, Lael G, Sweitzer BJ, Shea BF, Hallisey R, Vander Vliet M,
`Nemeskal R, Leape LL (1995) Incidence of adverse drug
`
`CFAD VI 1022-0004
`
`
`
`events and potential adverse drug events – implications for
`prevention. JAMA 274: 29–34
`4. Lesar TS, Briceland LL, Delcoure K, Parmalee JC, Masta-
`Gornic V, Pohl H (1990) Medication prescribing errors in a
`teaching hospital. JAMA 263: 2329–2334
`5. Classen DC, Pestotnik SL, Evans RS, Burke JP (1991)
`Computerized surveillance of adverse drug events in hospital
`patients. JAMA 266: 2847–2851
`6. Bates DW, Boyle DL, Vander-Vliet MB, Schneider J, Leape L
`(1995) Relationship between medication errors and adverse
`drug events. J Gen Intern Med 10: 199–205
`7. Leape LL (1994) Error in medicine. JAMA 272: 1851–1857
`8. Berwick DM (1989) Continuous improvement as an ideal in
`health care. N Engl J Med 320: 53–56
`9. Gro¨ nroos P, Irjala K, Heiskanen J, Torniainen K, Forsstro¨ m
`JJ (1995) Using computerized individual medication data to
`detect drug eects on clinical laboratory tests. Scand J Clin Lab
`Invest 55 [Suppl 222]: S31–S36
`10. Sjo¨ qvist F (1995) FASS ‘95 Interaction between drugs – sys-
`tematic review. LINFO Drug Information LTD, Stockholm,
`Sweden
`11. Gro¨ nroos P, Irjala K, Forsstro¨ m JJ (1995) Coding drug eects
`on laboratory tests for health care information systems. Proc
`Annu Symp Comput Appl Med Care 1995, Oct 28–Nov 1, New
`Orleans, 449–453
`12. Kuperman GJ, Bates DW, Teich JM, Schneider JR, Cheiman
`D (1994) A new knowledge structure for drug–drug interac-
`tions. Proceedings of the Annual Symposium on Computer
`Applications in Medical Care 1994, Nov 5–9, Washington,
`DC, 836–840
`13. Linnarsson R (1993) Drug interactions in primary health care.
`Scand J Prim Health Care 11: 181–186
`
`17
`
`therapeutic
`14. World Health Organisation (1995) anatomical
`chemical
`(ATC) classification index. WHO Collaborating
`Centre for Drug Statistics Methodology, Oslo, Norway
`15. Dambro MR, Kallgren MA (1988) Drug interactions in a clinic
`using COSTAR. Comput Biol Med 18: 31–38
`16. Jankel CA, Speedie SM (1990) Detecting drug interactions: a
`review of the literature. DICP, Ann Pharmacother 24: 982–989
`17. Gardner RM, Hulse RK, Larsen KG (1990) Assessing the ef-
`fectiveness of a computersed pharmacy system. Proc Annu
`Symp Comput Appl Med Care 1990: 668–672
`18. Wyatt JC (1994) Clinical data systems. Part II. Components
`and techniques. Lancet 344: 1609–1614
`19. Wyatt JC, Walton R (1995) Computer based prescribing. BMJ
`311: 1181–1182
`20. Leape LL, Bates DW, Cullen DJ, Cooper J, Demonaco HJ,
`Gallivan T, Hallisey R, Ives J, Laird N, Lael G, Nemeskal R,
`Petersen LA, Porter K, Servi D, Shea BF, Small SD, Sweitzer
`BJ, Thompson BT, Vander Vliet M (1995) Systems analysis of
`adverse drug events. JAMA 274: 35–43
`21. Frost RW, Lasseter KC, Noe AJ, Shamblen EC, Lettieri JT
`(1992) Eects of aluminum hydroxide and calcium carbonate
`antacids on the bioavailability of ciprofloxacin. Antimicrob
`Agents Chemother 36: 830–832
`22. Radandt JM, Marchbanks CR, Dudley MN (1992) Interac-
`tions of fluoroquinolones with other drugs: mechanisms, vari-
`ability, clinical significance and management. Clin Infect Dis
`14: 272–284
`23. Sahai J, Healy DP, Stotka J, Polk RE (1993) The influence of
`chronic administration of calcium carbonate on the bioavail-
`ability of oral ciprofloxacin. Br J Clin Pharmacol 35: 302–304
`
`CFAD VI 1022-0005
`
`