- Email: [email protected]
Translation and interpretation: The hidden processes and problems revealed by computerized physician order entry systems
Translation and Interpretation: The Hidden Processes and Problems Revealed by Computerized Physician Order Entry Systems Reid W. Coleman Even the most basic computerized physician order entry systems can reduce medication error rates, improve the quality, and decrease the costs of medical care. Routine tasks such as decryption, triage, transcription, and transmission are eliminated or streamlined, reducing the source and likelihood of human errors. Translation of physician intent into actual orders requires more advanced computer systems with
sophisticated algorithms built-in. Further, adding an interpretative function to understand and transmit orders that could have subtly different meanings will be challenging. Extensive analysis and the cooperative efforts of multidisciplinary teams will be required to add incremental value to computerized physician order entry systems. © 2004 Elsevier Inc. All rights reserved.
T
slips, and cardexes. Transcribed orders are also not entirely exempt from errors due to poor handwriting or errors of omission, in this case, due to the transcriber. Finally, these paper orders must be physically carried, faxed, or “tubed” to the appropriate location before they can be conducted. Their final destinations diverse, orders can be wrongly routed or sometimes lost, leading to delays in the workflow.
HE PHYSICIAN’S order is at a “process crossroad.” Writing an order represents the last step of the physician’s workflow, but is the first step of a very complex cascade of processing events involving nurses, secretaries, and a host of other ancillary personnel and support services. After an order has been written but before it can be conducted, a number of discrete steps are noted to occur. The most easily recognized of these steps are decryption, triage, transcription, and transmission. Poor physician handwriting and decryption errors represent some of the most common and obvious factors driving implementation of computerized physician order entry (CPOE). Decryption of handwritten orders is both an old joke (Fig 1) and a source of significant errors, in addition to potential liability.1 Every hospital has secretaries, nurses, or pharmacists who can decipher even the most obscure hieroglyphics created by many physicians, but all too often the best guess about what was written is incomplete or wrong (Fig 2). Incorrect decryption is a contributing factor to mishaps resulting from wrong drug name, calculation, dosage form, abbreviation, decimal placement, unit, or rate expression, which, as a group comprise one third of all possible medication errors.2 Once the physician’s order has been decrypted, a triage function often follows. After the most urgent or “stat orders” are processed, other orders are handled, typically without any clear-cut rules for prioritization. Ease of processing is often the factor determining which orders get through first, with those requiring complex forms (eg, total parenteral nutrition, blood products, patient-controlled analgesia), telephone calls for scheduling, or clarification left for last. Physician orders are next transcribed onto a variety of medication sheets, laboratory or x-ray Journal of Critical Care, Vol 19, No 4 (December), 2004: pp 279-282
THE CASE FOR COMPUTERIZED PHYSICIAN ORDER ENTRY
Decryption, Triage, Transcription, and Transmission When each of these steps is examined, the case for CPOE becomes compelling and the process appears deceptively simple. The most primitive order entry systems, which have little more than word processing functionality, allow physicians to: 1) type clear orders or 2) pick correctly spelled medication and/or diagnostic test names from a pick list. It is evident that even this most basic of computerized systems has the capacity to eliminate all decryption errors. Medication transcription errors were reduced 100% in one study examining CPOE system implementation at a large academic medical center, and was one of the major documented benefits.3 Slightly more sophisticated systems with unidirectional interfaces can be built to From Lifespan, Providence, RI. Presented at Computers and Information Systems in the ICU: An Expert Roundtable Conference, November 7-9, 2003, Boston, MA. Address reprint requests to Reid W. Coleman, MD, Lifespan, 167 Point St, Providence, RI 02903; e-mail: rcoleman@ lifespan.org. © 2004 Elsevier Inc. All rights reserved. 0883-9441/04/1904-0013$30.00/0 doi:10.1016/j.jcrc.2004.09.001 279
280
REID W. COLEMAN
Fig 1. Decryption errors-–A problem eliminated with CPOE. (Reprinted with permission from http://www.cartoonbank.com/ search_results_category.asp?oldSectionⴝall&keywordⴝhandwriting&artistⴝ&captionⴝ&artIDⴝ&topicⴝall&pubDateFromⴝ &pubDateToⴝ&pubDateMonⴝ&pubDateDayⴝ&pubNYⴝ2&advancedⴝ1&xⴝ22&yⴝ6§ionⴝnotecards. © 2004 Jack Ziegler from cartoonbank.com. All rights reserved.)
transmit orders directly to computer systems in various processing departments, and CPOE takes another forward step toward safety and improved workflow. With orders routed to different printers, triage then becomes a function of the person at the receiving printer and transmission has been accomplished. The secretary/nurse/pharmacist combination may still transcribe orders onto medication administration records and cardexes, but legible transcription is much more likely. Thus, even the most crude technological advances have the capability to promote improvements in order flow, medication-error prevention, and patient safety, and a major step forward can take place.4 Hidden Process: Translation Immediately such systems are seen as limited. Many points of potential failure remain; much of the order processing is manual3 and still performed by error-prone humans. Despite the advancements made with pick-lists, unidirectional interfaces, and improved order routing, it becomes clear that certain other underlying processes are also taking place.
Translation of common phrases used by physicians into actionable items is one such process that has a substantial impact but may not be appreciated during the initial design phase of a CPOE system. It becomes clear later in the process that the language that physicians use when writing orders is not the language used by ancillary systems. For example, secretaries, nurses, and pharmacists have been translating terms like “in am,” “after meals,” “now,” and “QID” into actual times, reflecting when the dose is expected to be administered. Additionally, unit secretaries have “crib sheets” that list the ordering name of terms such as “hip film,” “persantine-thall,” and “speech eval”; nurses/pharmacists are known to apply certain local conventions, ie, warfarin is given at 4 pm, statins after supper, and some Q6 hour medications, although scheduled for midnight, can be given at 10 pm to avoid waking the patient. Physicians may be only vaguely aware that this translation process has been taking place, but if a CPOE system is to reflect real-life practice, this translation must be integrated into the ordering process. Much of it can be done by more sophisticated interfaces that use
COMPUTERIZED PHYSICIAN ORDER ENTRY SYSTEMS
281
Fig 2. Complete handwritten orders from hospital charts. Chart review indicates that these orders were carried out without any documented clarification.
synonym tables and increasingly complex logic, but there is also an increase in the amount of information that the physician must provide when initiating an order at the computer terminal. Incremental improvement in order systems at this level of development are best appreciated by nurses, secretaries, and pharmacists. Orders are now more complete as well as more readable, and the time spent on clarification is reduced. In contrast, from the physician’s perspective, there is additional work. However, to successfully implement a CPOE system physician acceptance is key and the reluctance-factor might be reduced if value can be attached to the increased workload. The necessary next step is therefore to turn the CPOE system into an order management system. The distinction is somewhat arbitrary, but an order system becomes a management system with the addition of the two-way interface. When orders are checked against existing data so that dosages may be checked against patients’ current weight and renal/liver function, drug-drug interactions and allergies are analyzed, duplicate orders are detected in real time, and all relevant information is fed back to the physician as orders are being written, real benefit is gained by the physician and another
level of improvement in safety is achieved. To this, add disease-specific prompting for tests and treatments and the order entry system has now become a physicians’ order management system. It is at this point, however, that 2 new problems with translation become apparent. The first is the “Babelfish phenomenon.” When a medication order received by a pharmacy system is sent across another interface to a medication administration recording system, translation occurs twice. “Babelfish” is an Internet-based program that performs “translation” in a predominantly word-for-word fashion.5 The classic Babelfish story (which is apocryphal) tells that if the phrase “The spirit is willing but the flesh is weak” is translated into Russian by this system, and the Russian is then translated back into English, the result is “The ghost is willing but the meat is raw.” The story is amusing, but in a medication system this phenomenon can lead to disaster. Sending an order for a medication duration expressed as days to a pharmacy system that calculates duration in hours from the first dose, and then to an administration system that calculates duration in number of doses has the potential to create lost or extra doses. Add to this problem a “one time” medication order
282
REID W. COLEMAN
that, in a CPOE system, has a literal “duration” of zero minutes and thus disappears from the medication administration record as soon as it appears. Given these simple examples, it is easy to see how a CPOE system designed to improve medication safety can evolve into one that initiates and perpetuates significant errors. Hidden Process: Interpretation The second translation problem is more insidious but potentially as dangerous. This particular issue is not caused by computerized order systems, per se, but is certainly revealed by them. Nurses, secretaries, and pharmacists do not only translate physician orders; they interpret them. Whereas translation is the process of communicating the meaning of an order, interpretation is part of communicating its perceived intent. This function is inherent in any paper order system, but is seldom recognized even by those who perform it and is best illustrated with an example. A physician admitted a patient at 9 pm on Saturday with a diagnosis of community acquired pneumonia. Among other orders, azithromycin was prescribed. The frequency choices for this medication displayed by the order system were “one time a day” and “every 24 hours.” The physician chose “one time a day.” The system also requires that a start time for medications be specified, based on a pick-list of descriptive terms. The physician chose “start now, then routine scheduled times.” The system time for “one time a day” is 8 am, and the system scheduled administration times of 9 pm on Saturday and then 8 am every day starting Sunday morning. The nurse reviewing this order used judgment to override the default times and changed the scheduled azithromycin dose to be administered each day at 9 pm. On Tuesday morning the patient was recovering well and at 8:30 am the physician discontinued the intravenous azithromycin and entered a new order for oral azithromycin, choosing “one time a day” for the frequency and “next scheduled time” for the start time. The pharmacy system generated a new admin-
istration schedule starting Wednesday morning at 8 am. If an alert nurse had not noticed this problem, the patient would have gone 35 hours without an antibiotic dose. The example described actually occurred in our hospital system and led to an investigation into the process. The nurse who first interpreted the order believed that the physician intended for the second dose of the antibiotic to be given 24 hours after the first dose instead of 11 hours later, which was the literal meaning of the order as entered. A second nurse was asked her interpretation and stated she would have followed the literal meaning of the order. A third nurse would have adjusted the dosing schedule to every 20 hours until the every 8 am schedule was established. Each nurse was clear that his/her answer was correct according to nursing policy and procedure. Finally, the physician was asked about her intent. The physician, a board certified internist and pulmonologist who is well respected by her colleagues and recognized by the house staff as a superb teacher, replied, “I don’t know. The nurses always handle that.” The lesson to be learned is that communicating the appropriate action between personnel of varying disciplines and motivations is a challenge that may be more complex than originally recognized, especially when there are many choices. CONCLUSIONS
Designing and implementing a computerized order system solves many well-recognized problems, but can also shed light and reveal systematic problems that were not previously understood or appreciated. The need for translation of physician orders can be handled by computer systems. The challenge of effective communication and interpretation cannot. Those who undertake the development and implementation of computer order system must understand that technology is only part of the solution. Examining and improving the processes that are employed in providing patient care is the true path to improving quality and safety.
REFERENCES 1. Hirshborn C: Poor penmanship costs MD $225,000. CMAJ 162:91, 2000 2. Kohn LT, Corrigan JM, Donaldson MS: To err is human: Building a safer health system. Washington, DC, National Academy Press, 1999, pp 1-223 3. Mekhjian HS, Kumar RR, Kuehn L, et al: Immediate
benefits realized following implementation of physician order entry at an academic medical center. JAMIA 9:529-539, 2002 4. Bates DW, Teich JM, Lee J, et al: The impact of computerized physician order entry on medication error prevention. JAMIA 6:313-321, 1999 5. http://en.wikipedia.org/wiki/Babel_fish
sophisticated algorithms built-in. Further, adding an interpretative function to understand and transmit orders that could have subtly different meanings will be challenging. Extensive analysis and the cooperative efforts of multidisciplinary teams will be required to add incremental value to computerized physician order entry systems. © 2004 Elsevier Inc. All rights reserved.
T
slips, and cardexes. Transcribed orders are also not entirely exempt from errors due to poor handwriting or errors of omission, in this case, due to the transcriber. Finally, these paper orders must be physically carried, faxed, or “tubed” to the appropriate location before they can be conducted. Their final destinations diverse, orders can be wrongly routed or sometimes lost, leading to delays in the workflow.
HE PHYSICIAN’S order is at a “process crossroad.” Writing an order represents the last step of the physician’s workflow, but is the first step of a very complex cascade of processing events involving nurses, secretaries, and a host of other ancillary personnel and support services. After an order has been written but before it can be conducted, a number of discrete steps are noted to occur. The most easily recognized of these steps are decryption, triage, transcription, and transmission. Poor physician handwriting and decryption errors represent some of the most common and obvious factors driving implementation of computerized physician order entry (CPOE). Decryption of handwritten orders is both an old joke (Fig 1) and a source of significant errors, in addition to potential liability.1 Every hospital has secretaries, nurses, or pharmacists who can decipher even the most obscure hieroglyphics created by many physicians, but all too often the best guess about what was written is incomplete or wrong (Fig 2). Incorrect decryption is a contributing factor to mishaps resulting from wrong drug name, calculation, dosage form, abbreviation, decimal placement, unit, or rate expression, which, as a group comprise one third of all possible medication errors.2 Once the physician’s order has been decrypted, a triage function often follows. After the most urgent or “stat orders” are processed, other orders are handled, typically without any clear-cut rules for prioritization. Ease of processing is often the factor determining which orders get through first, with those requiring complex forms (eg, total parenteral nutrition, blood products, patient-controlled analgesia), telephone calls for scheduling, or clarification left for last. Physician orders are next transcribed onto a variety of medication sheets, laboratory or x-ray Journal of Critical Care, Vol 19, No 4 (December), 2004: pp 279-282
THE CASE FOR COMPUTERIZED PHYSICIAN ORDER ENTRY
Decryption, Triage, Transcription, and Transmission When each of these steps is examined, the case for CPOE becomes compelling and the process appears deceptively simple. The most primitive order entry systems, which have little more than word processing functionality, allow physicians to: 1) type clear orders or 2) pick correctly spelled medication and/or diagnostic test names from a pick list. It is evident that even this most basic of computerized systems has the capacity to eliminate all decryption errors. Medication transcription errors were reduced 100% in one study examining CPOE system implementation at a large academic medical center, and was one of the major documented benefits.3 Slightly more sophisticated systems with unidirectional interfaces can be built to From Lifespan, Providence, RI. Presented at Computers and Information Systems in the ICU: An Expert Roundtable Conference, November 7-9, 2003, Boston, MA. Address reprint requests to Reid W. Coleman, MD, Lifespan, 167 Point St, Providence, RI 02903; e-mail: rcoleman@ lifespan.org. © 2004 Elsevier Inc. All rights reserved. 0883-9441/04/1904-0013$30.00/0 doi:10.1016/j.jcrc.2004.09.001 279
280
REID W. COLEMAN
Fig 1. Decryption errors-–A problem eliminated with CPOE. (Reprinted with permission from http://www.cartoonbank.com/ search_results_category.asp?oldSectionⴝall&keywordⴝhandwriting&artistⴝ&captionⴝ&artIDⴝ&topicⴝall&pubDateFromⴝ &pubDateToⴝ&pubDateMonⴝ&pubDateDayⴝ&pubNYⴝ2&advancedⴝ1&xⴝ22&yⴝ6§ionⴝnotecards. © 2004 Jack Ziegler from cartoonbank.com. All rights reserved.)
transmit orders directly to computer systems in various processing departments, and CPOE takes another forward step toward safety and improved workflow. With orders routed to different printers, triage then becomes a function of the person at the receiving printer and transmission has been accomplished. The secretary/nurse/pharmacist combination may still transcribe orders onto medication administration records and cardexes, but legible transcription is much more likely. Thus, even the most crude technological advances have the capability to promote improvements in order flow, medication-error prevention, and patient safety, and a major step forward can take place.4 Hidden Process: Translation Immediately such systems are seen as limited. Many points of potential failure remain; much of the order processing is manual3 and still performed by error-prone humans. Despite the advancements made with pick-lists, unidirectional interfaces, and improved order routing, it becomes clear that certain other underlying processes are also taking place.
Translation of common phrases used by physicians into actionable items is one such process that has a substantial impact but may not be appreciated during the initial design phase of a CPOE system. It becomes clear later in the process that the language that physicians use when writing orders is not the language used by ancillary systems. For example, secretaries, nurses, and pharmacists have been translating terms like “in am,” “after meals,” “now,” and “QID” into actual times, reflecting when the dose is expected to be administered. Additionally, unit secretaries have “crib sheets” that list the ordering name of terms such as “hip film,” “persantine-thall,” and “speech eval”; nurses/pharmacists are known to apply certain local conventions, ie, warfarin is given at 4 pm, statins after supper, and some Q6 hour medications, although scheduled for midnight, can be given at 10 pm to avoid waking the patient. Physicians may be only vaguely aware that this translation process has been taking place, but if a CPOE system is to reflect real-life practice, this translation must be integrated into the ordering process. Much of it can be done by more sophisticated interfaces that use
COMPUTERIZED PHYSICIAN ORDER ENTRY SYSTEMS
281
Fig 2. Complete handwritten orders from hospital charts. Chart review indicates that these orders were carried out without any documented clarification.
synonym tables and increasingly complex logic, but there is also an increase in the amount of information that the physician must provide when initiating an order at the computer terminal. Incremental improvement in order systems at this level of development are best appreciated by nurses, secretaries, and pharmacists. Orders are now more complete as well as more readable, and the time spent on clarification is reduced. In contrast, from the physician’s perspective, there is additional work. However, to successfully implement a CPOE system physician acceptance is key and the reluctance-factor might be reduced if value can be attached to the increased workload. The necessary next step is therefore to turn the CPOE system into an order management system. The distinction is somewhat arbitrary, but an order system becomes a management system with the addition of the two-way interface. When orders are checked against existing data so that dosages may be checked against patients’ current weight and renal/liver function, drug-drug interactions and allergies are analyzed, duplicate orders are detected in real time, and all relevant information is fed back to the physician as orders are being written, real benefit is gained by the physician and another
level of improvement in safety is achieved. To this, add disease-specific prompting for tests and treatments and the order entry system has now become a physicians’ order management system. It is at this point, however, that 2 new problems with translation become apparent. The first is the “Babelfish phenomenon.” When a medication order received by a pharmacy system is sent across another interface to a medication administration recording system, translation occurs twice. “Babelfish” is an Internet-based program that performs “translation” in a predominantly word-for-word fashion.5 The classic Babelfish story (which is apocryphal) tells that if the phrase “The spirit is willing but the flesh is weak” is translated into Russian by this system, and the Russian is then translated back into English, the result is “The ghost is willing but the meat is raw.” The story is amusing, but in a medication system this phenomenon can lead to disaster. Sending an order for a medication duration expressed as days to a pharmacy system that calculates duration in hours from the first dose, and then to an administration system that calculates duration in number of doses has the potential to create lost or extra doses. Add to this problem a “one time” medication order
282
REID W. COLEMAN
that, in a CPOE system, has a literal “duration” of zero minutes and thus disappears from the medication administration record as soon as it appears. Given these simple examples, it is easy to see how a CPOE system designed to improve medication safety can evolve into one that initiates and perpetuates significant errors. Hidden Process: Interpretation The second translation problem is more insidious but potentially as dangerous. This particular issue is not caused by computerized order systems, per se, but is certainly revealed by them. Nurses, secretaries, and pharmacists do not only translate physician orders; they interpret them. Whereas translation is the process of communicating the meaning of an order, interpretation is part of communicating its perceived intent. This function is inherent in any paper order system, but is seldom recognized even by those who perform it and is best illustrated with an example. A physician admitted a patient at 9 pm on Saturday with a diagnosis of community acquired pneumonia. Among other orders, azithromycin was prescribed. The frequency choices for this medication displayed by the order system were “one time a day” and “every 24 hours.” The physician chose “one time a day.” The system also requires that a start time for medications be specified, based on a pick-list of descriptive terms. The physician chose “start now, then routine scheduled times.” The system time for “one time a day” is 8 am, and the system scheduled administration times of 9 pm on Saturday and then 8 am every day starting Sunday morning. The nurse reviewing this order used judgment to override the default times and changed the scheduled azithromycin dose to be administered each day at 9 pm. On Tuesday morning the patient was recovering well and at 8:30 am the physician discontinued the intravenous azithromycin and entered a new order for oral azithromycin, choosing “one time a day” for the frequency and “next scheduled time” for the start time. The pharmacy system generated a new admin-
istration schedule starting Wednesday morning at 8 am. If an alert nurse had not noticed this problem, the patient would have gone 35 hours without an antibiotic dose. The example described actually occurred in our hospital system and led to an investigation into the process. The nurse who first interpreted the order believed that the physician intended for the second dose of the antibiotic to be given 24 hours after the first dose instead of 11 hours later, which was the literal meaning of the order as entered. A second nurse was asked her interpretation and stated she would have followed the literal meaning of the order. A third nurse would have adjusted the dosing schedule to every 20 hours until the every 8 am schedule was established. Each nurse was clear that his/her answer was correct according to nursing policy and procedure. Finally, the physician was asked about her intent. The physician, a board certified internist and pulmonologist who is well respected by her colleagues and recognized by the house staff as a superb teacher, replied, “I don’t know. The nurses always handle that.” The lesson to be learned is that communicating the appropriate action between personnel of varying disciplines and motivations is a challenge that may be more complex than originally recognized, especially when there are many choices. CONCLUSIONS
Designing and implementing a computerized order system solves many well-recognized problems, but can also shed light and reveal systematic problems that were not previously understood or appreciated. The need for translation of physician orders can be handled by computer systems. The challenge of effective communication and interpretation cannot. Those who undertake the development and implementation of computer order system must understand that technology is only part of the solution. Examining and improving the processes that are employed in providing patient care is the true path to improving quality and safety.
REFERENCES 1. Hirshborn C: Poor penmanship costs MD $225,000. CMAJ 162:91, 2000 2. Kohn LT, Corrigan JM, Donaldson MS: To err is human: Building a safer health system. Washington, DC, National Academy Press, 1999, pp 1-223 3. Mekhjian HS, Kumar RR, Kuehn L, et al: Immediate
benefits realized following implementation of physician order entry at an academic medical center. JAMIA 9:529-539, 2002 4. Bates DW, Teich JM, Lee J, et al: The impact of computerized physician order entry on medication error prevention. JAMIA 6:313-321, 1999 5. http://en.wikipedia.org/wiki/Babel_fish