In court, the prescribing doctor confirmed they had chosen the wrong drug from the drop-down menu.<\/p>\n
This is one tragic example of the risk of \u2018look-alike sound-alike\u2019 (LASA) medication errors in electronic prescribing systems.<\/p>\n
In November 2025, a\u00a0national patient safety alert\u00a0was issued on the risk of harm from healthcare staff incorrectly recording patients\u2019 penicillin allergies as penicillamine allergies, providing a further example\u200b2\u200b.<\/p>\n
\u201cThis error can result in patients with known penicillin allergies being prescribed penicillin-based antibiotics, increasing the risk of a potentially fatal anaphylactic reaction,\u201d it said.<\/p>\n
The alert explained that the error could occur in different ways, including that penicillamine appears above penicillin in an alphabetical drop-down list, or an allergy page displaying drugs by drug name instead of group, meaning that penicillamine will be the only option that comes up when typing \u2018penicill\u2019.<\/p>\n
Beginning in the 1990s, electronic prescribing and medicines administration (ePMA) systems have been rolled out across the NHS, particularly since NHS England\u2019s push to go paperless \u2014 a target first set for 2018 and later changed to 2024 following delays\u200b3\u200b. This deadline was pushed back again to 2025, which the government said was owing to\u00a0delays caused by the COVID-19 pandemic<\/a>.\u00a0<\/p>\n According to the UK government, the use of electronic prescribing can reduce medication errors by 30%\u200b4\u200b. However, the above examples show that there is potential for errors to occur, particularly in relation to LASA medicines, begging the question: have ePMA systems increased the risk of LASA errors?<\/p>\n Incident data<\/p>\n To answer this question,\u00a0The Pharmaceutical Journal\u00a0sent a Freedom of Information request to NHS England for data on patient safety incidents associated with LASA medicines between 2015 and 2025, to determine if there has been an increase in the number of incidents as ePMA systems have been rolled out to increasing numbers of NHS trusts in England (see Figure 1).<\/p>\n However, it was difficult to obtain data specifically on LASA incidents for multiple reasons. Two incident reporting systems have been running concurrently over the past few years, owing to a transition from the original \u2018National reporting and learning system\u2019 (NRLS), which ran from 2003 to 2024, and the \u2018Learn from patient safety events\u2019 (LFPSE) service, which was introduced in 2021. This created the potential issue of dual reporting during the crossover period (2021\u20132024), where both systems were running.<\/p>\n In a report on the transition, NHS England said: \u201cPreliminary analysis suggests that most organisations transitioning to LFPSE continued to submit to NRLS after their transition date. Further analysis will be needed to examine the extent to which this was dual reporting of the same incidents, versus overlapping reporting of different incidents\u200b5\u200b.\u201d<\/p>\n The way in which LASA incidents are reported further confounds the issues. There is not a specific category for LASA incidents; categories can include \u2018wrong\/transposed\/omitted medicine label\u2019 or \u2018poor packaging or design of medicines that might lead to errors\u2019. In addition, reports are often submitted in free text, making it hard to extract data, but also making it easier for staff to report non-LASA incidents.<\/p>\n Figure 1: Potential look-alike sound-like incidents between 2015 and 2025<\/p>\n From the available data, NHS England was able to identify drug pairs that make up the most reported errors (see Table).<\/p>\n Table: Most reported drug pairs involved in LASA\u00a0errorsReplacing one type of error with another<\/p>\n While the available data may not show an increase in errors, it is possible that LASA errors in traditional systems have just been replaced with new LASA errors in electronic systems, suggests Bryony Dean Franklin, professor of medication safety at University College London, and director and executive lead pharmacist for research at the Centre for Medication Safety and Service Quality, Imperial College Healthcare NHS Trust.<\/p>\n \u201cWith paper prescribing, it was more likely that you would write in your head the right drug, but the handwriting would be so bad it would be read differently by the nurse or pharmacist,\u201d she says.<\/p>\n \u201cSo, the \u2018look-alike\u2019 error would be perhaps at the dispensing or administration stage, rather than the prescribing stage, whereas now I think the error is perhaps more likely at the prescribing stage because it\u2019s picking from a menu.\u201d<\/p>\n It\u2019s almost that we\u2019ve taken away illegibility errors with one hand, and given drop-down menu errors with the other<\/p>\n Julia Scott, pharmacist and chief information officer at Dartford and Gravesham NHS Trust<\/p><\/blockquote>\n Franklin says these errors may have balanced each other out, explaining why there may not be a difference in the data.<\/p>\n Julia Scott, a pharmacist and chief information officer at Dartford and Gravesham NHS Trust,\u00a0<\/strong>echoes this. \u201cIt\u2019s almost that we\u2019ve taken away illegibility errors with one hand, and given drop-down menu errors with the other. So, I wouldn\u2019t expect LASA errors to increase or decrease. I would expect them to change.\u201d<\/p>\n A study comparing the prevalence and types of prescribing errors following the implementation of an ePMA system at a London teaching hospital with paper-based prescribing found that while errors involving incorrect doses and illegible or incomplete orders were less common with electronic prescribing; those involving duplication, omission, incorrect drug and incorrect formulation were more common\u200b6\u200b.<\/p>\n Mitigating errors<\/p>\n One of the tactics used to try and prevent LASA errors in paper-based prescribing is \u2018tall-man lettering\u2019, in which certain letters in drug names are capitalised to distinguish it from others (see Figure 2).\u00a0<\/p>\n Figure 2: Example of tall-man\u00a0lettering Adapted from: Quentin Lohmeyer et al. BMJ Qual Saf 2023;32:26-33<\/p>\n \u201cThere is some evidence it helps, but it doesn\u2019t mitigate risk completely. But can you implement tall man in your e-prescribing system to add an additional layer of prompting?\u201d asks Scott.<\/p>\n She also suggests changing how drugs are grouped. \u201cIf we\u2019re aware of commonly known LASA groups or pairs, can you add anything extra? For example \u2014 in terms of the sorting \u2014 such as forcing things out of alphabetical order if necessary, to take penicillamine and penicillin away from each other in a list.<\/p>\n \u201cHow do humans behave? If your drop-down menu list is too long, is there a tendency to go for the first thing that looks around about right? So how do we think about truncation and things like minimum character sets \u2014 how many letters do I have to type before it will offer me a drop-down list?\u201d she explains.<\/p>\n \u201cIn something like the gabapentin\/pregabalin combo \u2014 which is a classic LASA pair \u2014 if I only have to type \u2018gaba\u2019, I\u2019m going to get offered both. If I have to type five letters and put \u2018gabap\u2019, I\u2019ve immediately lost the LASA error. So, there are things you can build in that will narrow that drop-down menu list and just make it safer, but this needs to be balanced against usability.\u201d<\/p>\n Integrating AI<\/p>\n Scott also hopes that the integration of clinical decision support AI could help prevent LASA errors. \u201cIf you\u2019ve got ePMA ideally integrated with your electronic patient record, can you start having things like natural language processing coming in and applying logic, like \u2018you were writing about a diagnosis of a chest infection, but you\u2019ve selected penicillamine. I know those two things don\u2019t go together\u2019? You can start to bring some really sophisticated, clever prompts in.\u201d<\/p>\n However, she also warns of a \u2018flip side\u2019 for AI in relation to errors from\u00a0ambient voice technology<\/a>\u00a0(AVT), also known as \u2018AI scribes\u2019.\u00a0<\/p>\n AVTs listen to conversations between patients and healthcare professionals, after which they generate a transcript and clinical summary of the consultation. These outputs can be transformed into clinical notes and used to populate health records, or generate letters for patients or referral letters to other clinicians.<\/p>\n In January 2026,\u00a0NHS England urged clinicians<\/a>\u00a0to take advantage of the technologies.<\/p>\n \u201cThat\u2019s going to bring a whole new category of significant sound-alike error risk, because it\u2019s almost back to the days of verbal orders, when verbal orders can be really easily misheard,\u201d Scott explains.<\/p>\n \u201c[For example], the AI scribe has heard penicillin as penicillamine, or you\u2019ve truncated your \u2018cipro\u2019, so it needs to know that you mean ciprofloxacin. So AVT is another layer where we\u2019re gonna have to think very heavily about a brand-new LASA error mechanism that we don\u2019t have at the moment.\u201d<\/p>\n Scott goes on to question how this error could then be mitigated. \u201cAre we going to have to have AI-enabled clinical decision support layered over AVT to retrospectively apply tall man highlighting and things like that to the transcript that you\u2019re checking to help draw your lazy human eye to the verbal errors added to the transcript?<\/p>\n \u201cEverything you introduce, there\u2019ll be a new error mechanism,\u201d she concludes. \u201cIt\u2019s not about humans being fallible and making mistakes. Some of it is about system design, but a huge amount, for me, is about understanding those fundamental cognitive mechanisms that lead to errors and trying to plan ahead for those and think about how we can use design and engineering principles to mitigate the risk.\u201d<\/p>\n Other methods<\/p>\n Franklin also points to other ways of reducing LASA errors. \u201cThere\u2019s a system that we\u2019ve been working on called \u2018Touchdose\u2019, which allows prescribing by indication, so it matches doses to indications, which then does a little bit more sensitive checking. It\u2019s like making the prescriber link in with \u2018what\u2019s the clinical indication that we\u2019re using this drug for?\u2019\u201d<\/p>\n Touchdose<\/a>\u00a0\u2014 developed by Imperial College London spinout Dosium and part funded by the then Royal Pharmaceutical Society, with the investment remaining with the Royal College of\u00a0Pharmacy\u00a0\u00a0\u2014 is a clinical decision support system. It calculates the correct dosage of the drug for the individual patient and indication \u2014 including frequency, duration and route of administration. A patient\u2019s details are pulled from their medical record and the system filters by characteristics, such as age and weight, to suggest only relevant treatment options. It then calculates the correct doses.\u00a0The tool has primarily been designed for use in paediatrics and pulls data from the\u00a0British National Formulary\u00a0(BNF)\u00a0and\u00a0BNF for Children.<\/p>\n \u201cBecause it\u2019s requiring the prescriber to go through a workflow that involves picking the indication, I think this would be less likely to have look-alike type errors, because it\u2019s making you match things up,\u201d Franklin says.<\/p>\n A study of use of the tool in a London teaching hospital showed that, when Touchdose was used, the overall prescribing error rate was 1.2%, compared with 7.1% in standard medication orders\u200b7\u200b.<\/p>\n Touchdose currently sits on top of the Cerner ePMA system, but Franklin says it is being developed to interface with others.\u00a0<\/p>\n Under-reporting<\/p>\n Of course, with limited data currently available, it is difficult to know the true scale of LASA errors, particularly in relation to ePMA systems.<\/p>\n \u201cWe know that only about 1 in 100 prescribing errors and about 1 in 1,000 administration errors ends up reported as an incident report, so we\u2019ve got such massive underreporting that the granularity of picking anything up is quite tricky,\u201d says Franklin.<\/p>\n Reporting requires you to be aware that an error has occurred and be willing to report it<\/p>\n Bryony Dean Franklin, professor of medication safety at University College London, and director and executive lead pharmacist for research at the Centre for Medication Safety and Service Quality, Imperial College Healthcare NHS Trust<\/p><\/blockquote>\n \u201cReporting requires you to be aware that an error has occurred and be willing to report it, have the time and energy to do so and be able to get on a computer on the ward \u2026 so we\u2019ve got massive under reporting for many good reasons, not because people are trying to hide things under the carpet, just because of practicalities.\u201d<\/p>\n Franklin is hopeful that AI will make a difference in analysing LASA error reporting. \u201cYou get thousands of these reports and they\u2019re typically very varied in terms of quality \u2013 some of them are incredibly minimal in terms of the detail.\u00a0<\/p>\n \u201cI\u2019ve done analysis of these where I\u2019ve obtained the whole dataset from NHS England, and [spent] hours and hours of just reading these free-text fields and trying to work out what goes on.<\/p>\n \u00a0\u201cOne of the things with the new LFPSE system is that I think there\u2019s a hope that it will be more amenable to analysis, whether that\u2019s with AI or something else.\u201d<\/p>\n Scott agrees that AI could be the change and says that, in an ideal world, reporting would capture near-misses as well.<\/p>\n \u201cThey\u2019re more powerful data than the actual incidents, and certainly the incidents that make it to harm. In my ideal world, there\u2019s an ePMA system where you can make a mistake, and there\u2019s a way of seeing \u2018I nearly did that, silly me.\u2019\u00a0<\/p>\n \u201cCan I press a button that captures the almost mistake? Because it knows what I\u2019ve done. It knows where I\u2019ve clicked. It knows the workflow I followed. Can I press a button that says capture a near-miss further down the line?\u201d<\/p>\n \u201cYou should very easily be able to have a partially completed incident form, assuming an incident management is integrated with your ePMA system, that pulls across all that demographic stuff, so that you can then just fill out a narrative description of what happened. You\u2019d use AI support so you can narrate what happened, and it\u2019s going to parse it for you into a sensible natural language structure and extract all the coded concepts, pull the drug data from what you\u2019ve said, or from the prescription that it can see, because this clever incident reporting system now has access to the ePMA system.<\/p>\n \u201cIn my ideal world, it just needs you to describe what\u2019s gone wrong, and then the AI is clever enough to start doing a thematic analysis for you very quickly.\u201d<\/p>\n She says that a barrier to reporting in the current system is that staff have to stop what they are doing and open the incident reporting system.<\/p>\n \u201cI think [staff think] \u2018I\u2019ll report that later\u2019, carry on with the day job, because they\u2019re in the middle of patient care, and then that thought is gone. [They can have] the best intentions of reporting it, but it\u2019s never going to happen if they are not able to report it in the moment.\u201d<\/p>\n \u201cHealthcare professionals all understand the benefits of incident reporting and analysis and so on. They do want to do it. They want feedback. They want evidence that it\u2019s making a difference but it\u2019s hard to make a difference if you don\u2019t get reporting in [the right] volumes.\u201d<\/p>\n Scott is hopeful that AI can vastly improve the current system.<\/p>\n \u201c<\/strong>We obviously need to tackle known issues with AI, including environmental impact and ethical concerns, but if we can do that, there is such potential for it to enhance medication safety,\u201d she explains.\u00a0\u00a0<\/p>\n \u201cIn time, there may be all sorts of new risks that we just cannot imagine at the moment; whole new error types that AI may create. But this should not prevent us from taking advantage of all the potential benefits, we just need investment in the skills and knowledge to do this safely.\u201d<\/p>\n It seems clear that LASA errors are unlikely to ever be fully eliminated. However, NHS England is hopeful the LFPSE system will improve things.<\/p>\n Its website says: \u201c[The LFPSE system] introduces improved capabilities for the analysis of patient safety events occurring across healthcare, and enables better use of the latest technology, such as machine learning, to create outputs that offer a greater depth of insight and learning that are more relevant to the current NHS environment\u201d\u200b8\u200b.<\/p>\n Perhaps this system, coupled with the potential of AI, means ePMA systems may eventually help to reduce the rate of LASA errors.\u00a0<\/p>\n 1.<\/p>\n Prescribing error led to death of baby, says coroner. Pharmaceutical Journal. Published online 2026. doi:10.1211\/pj.2026.1.396976<\/a><\/li>\n 6.<\/p>\n Feather C, Appelbaum N, Le Geyt J, Jheeta S, Maconochie I, Franklin BD. Evaluating the impact of an indication-based, patient-specific prescribing tool on prescribing errors in paediatrics: a non-randomised, before-and-after study. bmjpo. 2025;9(1):e003662. doi:10.1136\/bmjpo-2025-003662<\/a><\/li>\n 7.<\/p>\n Franklin BD, Puaar S. What is the impact of introducing inpatient electronic prescribing on prescribing errors? A naturalistic stepped wedge study in an English teaching hospital. Health Informatics J. 2019;26(4):3152-3162. doi:10.1177\/1460458219833112<\/a><\/li>\n\n
![\"\"](\"https:\/\/www.europesays.com\/uk\/wp-content\/uploads\/2026\/05\/Figure-2-1-1024x555.png\")
<\/p>\n\n
\n