Development and evaluation of a computerised clinical decision support system for switching drugs at the interface between primary and tertiary care
© Pruszydlo et al.; licensee BioMed Central Ltd. 2012
Received: 2 August 2012
Accepted: 21 November 2012
Published: 27 November 2012
Upon admission to a hospital patients’ medications are frequently switched to alternative drugs compiled in so called hospital drug formularies. This substitution process is a laborious and error-prone task which should be supported by sophisticated electronic tools. We developed a computerised decision support system and evaluated benefit and potential harm associated with its use.
Based on a multi-step algorithm we identified drug classes suitable for exchange, defined conversion factors for therapeutic interchange, built a web-based decision support system, and implemented it into the computerised physician order entry of a large university hospital. For evaluation we compared medications manually switched by clinical pharmacists with the results of automated switching by the newly developed computer system and optimised the system in an iterative process. Thereafter the final system was tested in an independent set of prescriptions.
After iterative optimisation of the logical framework the tool was able to switch drugs to pharmaceutical equivalents and alternatives; in addition, it contained 21 different drug classes for therapeutic substitution. In this final version it switched 91.6% of 202 documented medication consultations (containing 1,333 drugs) automatically, leaving 8.4% for manual processing by clinical professionals. No incorrect drug switches were found.
A large majority (>90%) of drug switches performed at the interface between primary and tertiary care can be handled automatically using electronic decision support systems, indicating that medication errors and workload of healthcare professionals can be considerably reduced.
KeywordsClinical decision support systems Drug information services Drug switching
Drug switching at the interface between primary and tertiary care is a time-consuming and error-prone task in inpatient care . In hospitals local drug committees and pharmacies usually define hospital drug formularies (HDF). HDF are restricted prescribing lists containing a subset of all available drugs with the intention to assure quality of in-house prescribing, simplify logistics, and save costs in drug therapy [2, 3]. Given the large size of the German drug market the likelihood of drugs being on the HDF is small. Indeed, upon admission to a German hospital about 50% of all previous drugs are switched [1, 4]. Appropriate switching is a laborious task that must carefully consider combination products, routes of administration, and also therapeutic equivalents and their (often differing) doses, if the same active ingredient is not available.
In a pilot investigation assessing the quality of drug switching we evaluated 128 switches in 30 consecutively admitted patients and learnt that one in five drug substitutions (21%) was wrong, mainly due to dose errors or mistakes in switching combination drugs to multiple single agents. The number and severity of these errors, which compromised optimum treatment already on the first day of hospitalisation, was alarming and we, thus, decided to support the process of drug switching electronically.
To implement a CDSS for drug switching upon admission we developed a multi-step algorithm and in a first evaluation it qualified as a logical framework to ease the practice of switching drugs in a standardised and reliable way . We then built a CDSS, based on the algorithm, implemented it into the CPOE of a large university hospital, and evaluated benefit and potential harm.
Development and implementation of the CDSS
Drug classes integrated into the final CDSS version for automatic switching to therapeutic equivalents
A02AA, A02AB, A02AC, A02AD, A02AF, A02AH
Histamine H2-receptor antagonists
Proton pump inhibitors
Serotonin (5-HT3) antagonists
Blood glucose lowering drugs, excl. insulins
A12AA, A12BA, A12CB, A12CC
Sulfonamides, plain (low-ceiling diuretics)
Sulfonamides, plain (high-ceiling diuretics)
C07AA, C07AB, C07AG
Calcium-channel blocking agents
Angiotensin-converting enzyme inhibitors
Angiotensin receptor antagonists
HMG-CoA reductase inhibitors
Selective serotonin (5-HT1) agonists
Benzodiazepine (hypnotics and sedatives)
Benzodiazepine related drugs
To implement the described algorithm, well structured data of all available drugs of the German market were necessary. Based upon this data (MMI Pharmindex, Medizinische Medien Informations GmbH, Germany) the CDSS compares different brands regarding important drug characteristics such as active ingredients, strengths, dosage forms, and ATC classification and also switches combination products. Depending on the step of the algorithm further subroutines of the CDSS provide additional data to be displayed (e.g. newly adjusted dosage regimens, information about tablet splitting, or hints and warnings related to the suggested substitution (Figure 3)).
Evaluation of the CDSS
In some surgical wards of the University Hospital of Heidelberg switching drugs on admission is routinely performed by a team of clinical pharmacists. Therefore the patient’s drug history is documented by nurses or physicians of the surgical ward and faxed to the hospital pharmacy where the drugs are switched manually to drugs of the HDF. The resulting suggestion for in-house medication is faxed back to the requesting ward and documented on paper. These previously and independently documented medication switches were used to test the functionality of the newly developed CDSS (version 0.9). The medications of consecutively documented drug switch consultations of a three-month period were entered into the CPOE and the manual switches by the clinical pharmacists were compared with the suggestions of the electronic CDSS.
Because some drugs may be switched to more than one compound of the HDF each switch was evaluated by an independent senior clinical pharmacist who was blinded for the origin of the switch. The main goal of this comparison was to decide whether the switching results were identical, equivalent or whether either one suggestion was better. All switching results of the CDSS that were considered worse were reviewed once again by the same clinical pharmacist who judged whether the CDSS suggestion was inadequate/wrong or correct but suboptimal.
On the basis of this assessment the CDSS was slightly modified to further improve the software (version 1.0). Then the evaluation was repeated in an independent set of consecutive switch consultations documented in the three months following the first evaluation period.
Data collection and analysis
This study was approved by the responsible Ethics Committee of the Medical Faculty of Heidelberg University, Germany (protocol # 136/2005) and conducted according the principles of the current version of the Declaration of Helsinki. Only anonymised prescription data of switch consultations were used for this study. Data were described with descriptive statistics and reported as absolute and relative frequencies or arithmetic means with standard deviation (SD). Data entry and analysis were performed on a Microsoft SQL Server 2005 database by using structured query language (SQL) and by using Microsoft Excel 2003.
Pilot evaluation of the CDSS (version 0.9)
In the first evaluation 174 documented drug switch consultations were included containing 1,296 drugs manually switched by the team of clinical pharmacists (mean ± SD: 7.5 ± 4.3 drugs per consultation). 1,176 of these (90.7%) could be entered into the evaluation-database; the remaining 120 (9.3%) were excluded because essential information was missing on the handwritten consultations (e.g. drug strength or dosage regimen).
“Insulins” (ATC code: A10A): Substitution of insulins should be personalised with a tailored monitoring of blood-glucose levels.
“Alpha-adrenoceptor blockers” (ATC code: C02CA): To switch alpha-adrenoceptor blockers a change of the dosage form and thus release characteristics (e.g. slow → instant release) can be necessary which is not yet supported by the given algorithm.
“Drugs for treatment of hyperkalemia and hyperphosphatemia” (ATC code: V03AE): In this group various agents with differing mechanisms of action are clustered that require manual processing (e.g. calcium carbonate → calcium diacetate).
“Other antianemic preparations” (ATC code: B03XA): In this group diverse chemical classes (e.g. biosimilars) are clustered that require processing by experts.
In the revised tool only oral dosage forms were considered thus omitting parenteral and inhaled drugs.
Evaluation of the final CDSS (version 1.0)
The evaluation of the refined CDSS version comprised 202 documented drug switch consultations containing 1,518 drugs (7.5 ± 3.9 switches per consultation). Of these 185 prescriptions (12.2%) were incomplete or inaccurate and thus excluded leaving 1,333 drug switches (87.8%) for evaluation.
947 of 1,333 drugs (71.0%) were substituted similarly by the team of pharmacists and the CDSS. Differences in automatic and manual switching occurred in 386 cases (29.0%). For 58.6% of these cases suggestions of pharmacists and CDSS were equivalent, in 15.5% the CDSS suggestions and in 25.9% the pharmacists’ suggestions were considered superior. The review of the latter discrepant suggestions revealed that all suggestions by the CDSS were correct, but not always the optimum choice, and in no instance (0%) inadequate equivalents were suggested. The results of both evaluations (version 0.9 and 1.0) are summarised in Figure 4.
Switching performance of the CDSS
Performance of the CDSS
CDSS version 0.9 (n=1,176)
CDSS version 1.0 (n=1,333)
Group 1: HDF drug
Group 2: Pharmaceutical equivalent
Group 3: Pharmaceutical alternative
Group 4: Therapeutic equivalent
Group 5: No CDSS-switch
In Germany close to 18 million people are hospitalised every year  and according to their drug history they are prescribed an average of six drugs [1, 13]. Hence, every day roughly 300,000 drug switches are performed in German hospitals and – if performed with similarly poor accuracy as in our pilot study – they will be a major cause for avoidable risks for patients and also a waste of working force.
The potential for medication errors concerning dose adjustments after switching to therapeutic equivalents is well known. In an American study a significant proportion of patients whose cholesterol lowering medication was switched from atorvastatin to simvastatin thereafter received lower therapeutic doses, potentially impairing the quality of care and effectiveness . But in some cases also the appropriateness of generic substitution is still controversially discussed [15, 16]. Even when therapeutic doses and conversion factors are carefully considered the substitution may lead to critical changes in the exposure with additives [17, 18] and – given the generally accepted range of bioequivalence – switching may cause considerably differing exposures to the active compound, which may be relevant for drugs with a narrow therapeutic window . Therefore tight regulations and recommendations defining suitable drugs and drug classes for substitution might improve physicians’ confidence and compliance in the switching procedure .
Accordingly, we formed an interdisciplinary team of specialists (physicians, pharmacists, and computer scientists) to design and develop an electronic tool, which is a standard procedure to create well-fitted and user-friendly systems . After implementation we evaluated a large independent sample of prospectively documented drug switches performed by experienced clinical pharmacists and thus used real clinical data as the most realistic test-cases.
The first test of our newly developed CDSS already showed good performance of our algorithm (93.6% could be switched electronically) but also revealed weaknesses that might have led to medication errors if the CDSS would have been released into clinical routine before meticulous validation. These weaknesses mainly albeit not exclusively concerned the switch to therapeutic equivalents, which in some cases required additional patient information or a switch to formulations with differing release characteristics. After modification, the second version (version 1.0) of the CDSS enabled automatic switching of 91.6% of the cases without any inadequate suggestions. Considering the huge size of the German drug market the performance of the tool is rather remarkable. Indeed with more than 70,000 pharmaceuticals, Germany has one of the largest drug markets worldwide suggesting that the CDSS will likely also efficiently switch drugs prescribed in other countries. Such a CDSS may even be useful in countries whose reimbursement system currently allows continuation of the patient’s own drugs during hospitalised care (such as the UK) because even then formulary substitutions are still required for example when patients are admitted as emergencies or if there is insufficient quantity of medicine to cover the whole inpatient stay.
Even in the optimised version, some drug switches suggested by the CDSS (100/1,333 observed switches) were judged inferior to the switching result of the clinical pharmacists. Analysis of these situations revealed that in most cases the clinical pharmacists derogated from the basic algorithm to improvise in a non-standard situation. For example our CDSS failed to compute an adequate dosage regimen after switching “Metoprolol 100 retard 1A Pharma” (containing 78.09 mg metoprolol) to “Beloc-Zok Retardtabl” (containing 77.82 mg metoprolol) because of slightly differing drug strengths, whose clinical irrelevance is easily recognised by an expert but requires proper specification for consideration by a computer system. Furthermore the human specialist is able to consult information sources beyond the CDSS database (e.g. by contacting the pharmaceutical manufacturer when additional drug information is needed and not available electronically). At last and in contrast to a CDSS, clinical pharmacists were able to consider special patient characteristics (e.g. age/mental state) and therefore to adjust a dosage regimen seeking to simplify the prescription (e.g. by avoiding tablet splitting). Indeed complex and complicated drug regimens (e.g. regimens with multiple administration times or the need for tablet-splitting) are an important prescription characteristic promoting non-adherence of the patients [20, 21] that could be prevented in a large fraction of all prescriptions .
This emphasises areas of unmet need among professionals for support of drug switching in complicated cases that are not yet covered by the CDSS. Nevertheless, already today the CDSS can reduce the workload of these professionals by reliable handling of the large majority of routine substitutions. This is a substantial reduction of time when considering that manual drug switching by American clinicians was estimated to take 11 minutes per medication .
The thorough analysis of drug pairs not yet automatically switched revealed that a meaningful next step would be to support dose adjustment of different application forms and to consider combination products for therapeutic substitution (step 4). In addition, a future switching tool could also enable adoption of new scenarios like patient admission to an intensive care unit where oral forms often have to be switched to parenteral or intravenous forms and the switch back to ambulatory medication at discharge from hospital.
A number of potential limitations should be considered before generalisation of the results to other settings. (1) In our study we only switched drug combinations of surgical patients thus restricting evidence to patients receiving comparable medication regimens. However, as shown in our previous evaluation performed in the same wards , the Charlson score of these patients is high, reflecting the numerous co-morbidities of these patients and suggesting that their drug regimens likely represent also a population of internal medicine patients. (2) In our evaluation we had to exclude about 10% of the switch requests because essential details of the patients’ prescription were missing (e.g. missing dosage regimen, strengths, or information needed to identify the specific brand). This stresses the advantages of an electronic documentation of patient medications in a CPOE linked to drug databases as it enables exact identification of brands, the key information to relevant drug-related data like drug composition, strength, galenic formulation, and the corresponding SPC . Unfortunately SPC information is currently not available in a well structured format , which would facilitate the development of tools like the one described herein. (3) Our evaluation was conducted by project members and not by clinical staff for whom the CDSS was developed. Thus, possible socio-technical incidents, a group of medication errors originating from interactions between clinical staff and the system , have yet to be investigated. (4) Finally the revision of differences between the switching results of pharmacists and the CDSS was performed by only one expert. This senior clinical pharmacist was considered best choice due to her extensive practical experience in switching drugs for years. However, by blinding this expert for the origin of the switching suggestions (pharmacist or CDSS) bias is minimised.
The results of our study demonstrate that in the overwhelming majority of cases (>90%) a sophisticated electronic CDSS can safely and reliably switch drugs of admitted patients to the locally available drugs as compiled in a HDF. Given the substantial error-rates in this process such support is indeed needed.
Hospital drug formularies
Computerised physician order entry
Clinical decision support systems
Summary of product characteristics
Anatomic therapeutic chemical (classification system)
- Walk SU, Bertsche T, Kaltschmidt J, Pruszydlo MG, Hoppe-Tichy T, Walter-Sack I, Haefeli WE: Rule-based standardised switching of drugs at the interface between primary and tertiary care. Eur J Clin Pharmacol. 2008, 64: 319-327. 10.1007/s00228-007-0402-5.View ArticlePubMedGoogle Scholar
- De Smet M: Drug formularies – good or evil? A view from the EEC. Cardiology. 1994, 85 (Suppl 1): 41-45.View ArticleGoogle Scholar
- Thürmann PA, Harder S, Steiof A: Structure and activities of hospital drug committees in Germany. Eur J Clin Pharmacol. 1997, 52: 429-435. 10.1007/s002280050315.View ArticlePubMedGoogle Scholar
- Himmel W, Tabache M, Kochen MM: What happens to long-term medication when general practice patients are referred to hospital?. Eur J Clin Pharmacol. 1996, 50: 253-257. 10.1007/s002280050103.View ArticlePubMedGoogle Scholar
- Agrawal A: Medication errors: prevention using information technology systems. Br J Clin Pharmacol. 2009, 67: 681-686. 10.1111/j.1365-2125.2009.03427.x.View ArticlePubMedPubMed CentralGoogle Scholar
- Bates DW, Gawande AA: Improving safety with information technology. N Engl J Med. 2003, 348: 2526-2534. 10.1056/NEJMsa020847.View ArticlePubMedGoogle Scholar
- Kuppermann GJ, Bobb A, Payne TH, Avery AJ, Gandhi TK, Burns G, Classen DC, Bates DW: Medication-related clinical decision support in computerized provider order entry systems: a review. J Am Med Inform Assoc. 2007, 14: 29-40.View ArticleGoogle Scholar
- Bates DW, Leape LL, Cullen DJ, Laird N, Petersen LA, Teich JM, Burdick E, Hickey M, Kleefield S, Shea B, Vander Vliet M, Seger DL: Effect of computerized physician order entry and a team intervention on prevention of serious medication errors. JAMA. 1998, 280: 1311-1316. 10.1001/jama.280.15.1311.View ArticlePubMedGoogle Scholar
- Shulman R, Singer M, Goldstone J, Bellingan G: Medication errors: a prospective cohort study of hand-written and computerised physician order entry in the intensive care unit. Crit Care. 2005, 9: R516-R521. 10.1186/cc3793.View ArticlePubMedPubMed CentralGoogle Scholar
- Furberg CD, Herrington DM, Psaty BM: Are drugs within a class interchangeable?. Lancet. 1999, 354: 1202-1204. 10.1016/S0140-6736(99)03190-6.View ArticlePubMedGoogle Scholar
- Kereiakes DJ, Willerson JT: Therapeutic substitution: guilty until proven innocent. Circulation. 2003, 108: 2611-2612. 10.1161/01.CIR.0000103637.50536.CC.View ArticlePubMedGoogle Scholar
- Destatis - Federal Statistical Office of Germany: Basic data of hospitals and prevention/rehabilitation facilities. http://www.destatis.de/jetspeed/portal/cms/Sites/destatis/Internet/DE/Content/Publikationen/Fachveroeffentlichungen/Gesundheit/Krankenhaeuser/GrunddatenKrankenhaeuser2120611097004,property=file.pdf.
- Laroche ML, Charmes JP, Nouaille Y, Fourrier A, Merle L: Impact of hospitalisation in an acute medical geriatric unit on potentially inappropriate medication use. Drugs Aging. 2006, 23: 49-59. 10.2165/00002512-200623010-00005.View ArticlePubMedGoogle Scholar
- Hess G, Sanders KN, Hill J, Liu LZ: Therapeutic dose assessment of patient switching from atorvastatin to simvastatin. Am J Manag Care. 2007, 13 (Suppl 3): S80-S85.PubMedGoogle Scholar
- Ferner RE, Lenney W, Marriott JF: Controversy over generic substitution. BMJ. 2010, 340: c2548-10.1136/bmj.c2548.View ArticlePubMedGoogle Scholar
- Johnston A, Asmar R, Dahlöf B, Hill K, Jones DA, Jordan J, Livingston M, Macgregor G, Sobanja M, Stafylas P, Rosei EA, Zamorano J: Generic and therapeutic substitution: a viewpoint on achieving best practice in Europe. Br J Clin Pharmacol. 2011, 72: 727-730. 10.1111/j.1365-2125.2011.03987.x.View ArticlePubMedPubMed CentralGoogle Scholar
- Dueñas-Laita A, Pineda F, Armentia A: Hypersensitivity to generic drugs with soybean oil. N Engl J Med. 2009, 361: 1317-1318. 10.1056/NEJMc0904562.View ArticlePubMedGoogle Scholar
- Rogkakou A, Guerra L, Scordamaglia A, Canonica GW, Passalacqua G: Severe skin reaction due to excipients of an oral iron treatment. Allergy. 2007, 62: 334-335. 10.1111/j.1398-9995.2006.01287.x.View ArticlePubMedGoogle Scholar
- Went K, Antoniewicz P, Corner DA, Dailly S, Gregor P, Joss J, McIntyre FB, McLeod S, Ricketts IW, Shearer AJ: Reducing prescribing errors: can a well-designed electronic system help?. J Eval Clin Pract. 2010, 16: 556-559.PubMedGoogle Scholar
- Lam PW, Lum CM, Leung MF: Drug non-adherence and associated risk factors among Chinese geriatric patients in Hong Kong. Hong Kong Med J. 2007, 13: 284-292.PubMedGoogle Scholar
- Saini SD, Schoenfeld P, Kaulback K, Dubinsky MC: Effect of medication dosing frequency on adherence in chronic diseases. Am J Manag Care. 2009, 15: e22-e33.PubMedGoogle Scholar
- Witticke D, Seidling HM, Lohmann K, Send AFJ, Haefeli WE: Opportunities to reduce medication regimen complexity: A retrospective analysis of 500 patients at discharge from internal medicine of a university hospital. Drug Saf. 2012, in press.Google Scholar
- Abourjaily P, Gouveia WA, Selker HP, Zucker DR: Evaluating the nondrug costs of formulary coverage restrictions. Manag Care. 2005, 14: 50-62.PubMedGoogle Scholar
- Quinzler R, Schmitt SPW, Szecsenyi J, Haefeli WE: Optimizing information on drug exposure by collection of package code information in questionnaire surveys. Pharmacoepidemiol Drug Saf. 2007, 16: 1024-1030. 10.1002/pds.1406.View ArticlePubMedGoogle Scholar
- Maxwell S, Eichler HG, Bucsics A, Haefeli WE, Gustafsson LL, on behalf of the e-SPC consortium*: e-SPC - delivering drug information in the 21st century: Developing new approaches to deliver drug information to prescribers. Br J Clin Pharmacol. 2012, 73: 12-15. 10.1111/j.1365-2125.2011.03981.x.View ArticlePubMedPubMed CentralGoogle Scholar
- Redwood S, Rajakumar A, Hodson J, Coleman JJ: Does the implementation of an electronic prescribing system create unintended medication errors? A study of the sociotechnical context through the analysis of reported medication incidents. BMC Med Inform Decis Mak. 2011, 11: 29-10.1186/1472-6947-11-29.View ArticlePubMedPubMed CentralGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1472-6947/12/137/prepub
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.