Problem: Speech recognition, or speech-to-text, is the ability of a computer software system to identify spoken words and then convert them into readable text. It is something most people use daily through cell phone texting and dialing, commands given to smart home devices, virtual meeting transcripts, and closed captioning on television programs. In healthcare, it is used for scheduling, office visit summaries, and other electronic health record (EHR) documentation. A recent study, Speech recognition for medical documentation: an analysis of time, cost efficiency and acceptance in a clinical setting (www.ismp.org/ext/1353), found that medical documentation using speech recognition improved efficiency resulting in significant time savings with lower error rates compared to typing. Like other technologies, we expect the use of speech recognition to continue to expand in the healthcare setting. Be that as it may, medication errors and close calls have been reported by organizations that have been using speech recognition software.
Speech Recognition-Related Medication Errors
In 2022, The Joint Commission (TJC) updated the Quick Safety alert titled, Speech recognition technology translates to patient risk (www.ismp.org/ext/1085). The alert provides examples of speech recognition documentation errors (e.g., allergy to XOPENEX [levalbuterol] documented instead of allergy to sulfa when “sulfa [next]” was spoken). The alert discussed a case in which a patient died after they were discharged from a hospital to a rehabilitation facility. Unbeknownst to the prescriber, the note that was transcribed using speech recognition software included an incorrect insulin dosage of 80 units rather than 8 units (10 times the intended dose). The alert also noted that some EHR documents were found to contain disclaimers to point out the problems with speech recognition that can result in a patient safety event. For example, a disclaimer may state, “Portions of the record may have been created with voice recognition software. Occasional wrong-word or ‘sound-alike’ substitutions may have occurred due to the inherent limitations of voice recognition software. Read the chart carefully and recognize, using context, where substitutions have occurred.”
Recently, an event reported to ISMP involved a pharmacist who received an order for chemotherapy, procarbazine 50 mg orally once daily for a patient. Since the patient did not have a history of cancer or an oncology consult documented, the pharmacist called the prescriber to clarify the indication. The prescriber stated that he intended to order prochlorperazine intravenously (IV) and that there must have been an error when placing the order in the EHR using speech recognition software. When the prescriber said the medication name using the computer dictation software, he was presented with a list of medication options and inadvertently selected the wrong medication and order sentence. After the error was identified, the prescriber discontinued the procarbazine order and entered an order for prochlorperazine 5 mg IV as needed. It is unknown which medication the prescriber had actually dictated.
The pharmacist was unaware that the hospital had recently implemented speech recognition software for prescribers to enter medication orders. The pharmacist who discovered the error notified the medication safety pharmacist who was also not aware of this software capability. In this organization, prescribers were expected to review dictated orders for accuracy before signing, but it seemed, in this case, a gap existed in the process.
Safe Practice Recommendations: The decision for an organization to implement software with speech recognition should be made with input and agreement of senior leadership, department heads, and safety/quality representation (e.g., pharmacy director, medication safety officer, risk management). If your organization is considering its use, follow these safeguards to mitigate the risk of error.
Designate a team. A designated team (e.g., prescribers, pharmacists, nurses, informatics, safety/quality) should invest time in evaluating and planning the use of speech recognition software for medication order entry. The team should proactively test the accuracy of the software for transcribing medication names. Include sound-alike medication name pairs (e.g., clonazePAM and cloBAZam) in testing scenarios and share these examples with staff to demonstrate the type of errors that could happen. Consider completing a failure mode and effects analysis (FMEA) to determine potential failure points and mitigation strategies.
Use simulation. Before implementing speech recognition software for medication orders, use simulation to evaluate the system in a test environment. Work directly with software vendors to understand potential problems that have been reported and recommendations to prevent them. Simulate the workflow to test what does and does not work, gain crucial feedback from frontline staff, and identify any potential safety gaps. Consider holding “a day in the life” to run real-life simulations to see how speech recognition software works in your clinical settings with a diverse group of end users, incorporating various background noises (e.g., emergency department, intensive care unit, cafeteria), and compare to vendors’ testing environments. Evaluate all devices practitioners may use (e.g., laptop, computer on wheels, smartphone) and ask end users to identify vulnerabilities and discuss concerns with the team so they can correct any issues before implementation.
Define the process. Evaluate if there should be any restrictions on using speech recognition software for medication orders (e.g., consider prohibiting use for chemotherapy/other high-alert medications, nonformulary medications). Ensure prescribers understand the risk of transcription errors and the requirements for clear communication of orders considering the impact of background noises.
Use clinical decision support. After they say the medication name, ensure prescribers are prompted to select from a list of medications and corresponding order sentences that go through the same clinical decision support (CDS) as typed orders. Make enhancements to CDS (e.g., indication selection, dose range checking, drug-disease interactions) if needed.
Review the order. As with any medication order, ensure there is a process that requires prescribers to review the medication order prior to submitting it.
Verify the indication. When verifying medication orders, pharmacists should also review the order, evaluate the indication, and ensure the prescribed medication and dose make sense in the context of the patient’s condition. This helps to identify errors with sound-alike drug names. If a pharmacist or nurse has any uncertainty with the medication order, they should contact the prescriber for clarification prior to dispensing or administering the medication.
Educate practitioners. Educate practitioners (e.g., prescribers, pharmacists, nurses) who may use speech recognition software for ordering, or who may receive an order using this technology, about the appropriate steps to ensure safety. Ensure prescribers understand the risk of patient harm if they do not review the order for accuracy prior to signing it.
Seek feedback and report errors. Establish a feedback mechanism for practitioners to report concerns, close calls, and errors with using speech recognition software and share these issues with the medication safety committee. Also report errors to the US Food and Drug Administration (FDA), software vendors, and ISMP. Share internal and external events and lessons learned with end users along with the best workflow to reduce the risk of these errors.
Implement quality checks. Establish and implement policies and procedures for regularly monitoring the quality of medication orders produced using speech recognition software. Identify issues and develop quality action plans to improve processes. Continually assess speech recognition software available on the market to determine if other software solutions offer a superior product.
Provide feedback to vendors. Organizations should collaborate with software vendors and provide feedback, as needed, to improve this technology. We expect this technology will become more advanced and encourage practitioners to become involved in this undertaking.
Suggested citation:
Institute for Safe Medication Practices (ISMP). Can we ensure medication safety with the use of speech recognition software? ISMP Medication Safety Alert! Acute Care. 2024;29(17):1-3.
