This study used a qualitative design and the setting was community care in England and Wales. This included settings such as general practice, community pharmacy, community mental health services, primary care networks (collaborative/integrated healthcare) (26), district nursing services, nursing/care homes and substance misuse services (27).

In this study medication errors were defined as “any preventable event that may cause or lead to inappropriate medication use or patient harm while the medication is in the control of the health care professional, patient, or consumer” (28). Events were categorised relating to the domains of medication prescribing, monitoring, dispensing and administration. ADEs were defined as an injury due to the use of a medication (29) and considered preventable if best medical practice was not adhered to (30, 31). Serious mental illness was defined as an illness with substantial functional impairment and included non-organic psychotic disorders, major affective disorder and personality disorders (32). Less serious mental illness was any other mental illness falling outside of the serious mental illness definition.

Four individuals with experience of using mental health services or caring for someone in mental health services were consulted on the design of the study and helped refine the interview schedule. Another patient advisory group of three expert patients with lived experience of mental illness were consulted three times throughout the duration of the study in order to provide advice regarding recruitment, validation of identified themes and their interpretation.

The sampling frame comprised of pharmacists, GPs, nurses, psychiatrists and substance misuse specialists with experience of working in a UK-based community service within the last 5 years. Within this frame, participants were recruited via purposive sampling to represent a range of professional roles and community settings. We sought participants who had first-hand experience of dealing with medication errors in relation to patients with mental illness (33). Participants were approached through the research team’s regional and national professional networks or through social media advertisements between June–November 2022. Snowballing was encouraged to reach potential participants that the research team did not have direct access to. Participants who expressed interest to take part were then asked to complete a signed consent form before being eligible to attend the interview. Interviews were conducted until data saturation was met which was predicted to be ~20 interviews, as little new information is generated beyond this number (34). Interviews continued beyond this target to further validate and strengthen the data given the heterogeneity of the professional groups.

Interviews were conducted remotely via Zoom^™. The interview guide (Supplementary file 1) was adapted from a study of medication administration errors using the critical incident technique (CIT) in secondary care mental health services (20) before being refined through consultation with the patient advisory group and piloting. The CIT enables exploration of the processes leading up to an incident and the consequences (35) and has already been applied to medication error and health research in other settings (20, 36–38). The CIT also reduces the propensity of interviewees to provide generic answers by encouraging them to focus on specific details of an incident (35). The interview consisted of first gathering information about the participant such as professional/training background, then a detailed exploration of one or more medication errors and/or preventable ADEs that the participant had made or that had direct first-hand knowledge of including the origin of the error(s) and the reasons for its occurrence. Finally, the participant discussed how they believed these incidents could have been avoided. Participants were encouraged to discuss errors that had happened recently or that could be recalled clearly. MJA conducted the interviews using a semi-structured interview guide. All interviews were video/audio recorded and sent to a professional university approved transcription company to undergo intelligent verbatim transcription. Returned transcripts were then checked against the original recording for accuracy by the main researcher (MJA). The interview length ranged between 23 and 91 minutes with the average length being 50.3 minutes.

Interview transcripts were uploaded into NVivo 12 Plus software for analysis using the framework method (39). This method allows systematic reduction of the data and in-depth analysis of key themes across the entire dataset without losing context (39) and has been used in other health research examining the causes of medication errors (20, 38). The framework method included seven steps: transcribing; reading and becoming familiar with the interviews; specific descriptive codes; developing a framework by grouping into established themes; applying the framework to the remainder of the interviews; charting the data into matrices; and finally interpreting the data. Participants were instructed in the participant information sheet to provide only first-hand accounts. Participant accounts that were based on speculation (i.e. no first-hand experience of the error) about the causes of errors/ADEs were excluded from any analysis. Independent analysis of the transcripts was carried out by the research team (MJA, RNK, PJL and DLP) who then all met to reach consensus, in order to help validate the themes generated. Consultation was carried out with the research team and the patient advisory group regarding the themes that were generated with an iterative process following to refine the data. The data were organised into themes based on the London Protocol (40) which builds upon Reason’s error framework (41), and has been used in other medication error research in community settings (42). The London Protocol proposes an analysis of clinical incidents by placing contributory factors into seven domains: patient; task and technology; individual (staff); team; work environment; organisation and management; and institutional (40).

Participants provided informed written consent to take part in this study. Participants were also given an interview ID to anonymise their contributions. Participants were being asked to recall preventable medication safety incidents which some may have found distressing, therefore, a distress protocol was developed as a precaution. This study was approved by the University of Manchester Research Ethics Committee 1 (reference 2022-13735-23555).

The most common contributory factor discussed was the individual (staff) which was involved in a total of 37 errors/ADEs, followed by the work environment (n = 31) and teams and interfaces (n = 28). A list of the contributory factors and how many were implicated in each error type are displayed in Table 2. A full breakdown of the contributory factors by sub-themes are displayed in Supplementary file 3. Figure 1 demonstrates how multiple factors contributed to one specific error from one of the participant accounts. The main difference between error accounts across community care services were the types of error(s) that professionals focused on for discussion, for example, GPs focused solely on prescribing and monitoring events.

Over half of the participants described a lack of knowledge, skills and training amongst community clinicians with regards to mental illnesses, psychotropic medication and of local shared care procedures (transfer of clinical responsibility from hospital to community). This was reported to be caused by unmet training needs but was also linked to a perceived lack of ownership and responsibility clinicians felt they had for what they considered to be a specialist patient group. One GP, when describing the origins of a prescribing error, described this relationship between familiarisation with psychotropics and responsibility:

This GP then went on to add that this lack of detailed knowledge of psychotropics created a barrier to community clinicians conducting independent scrutiny of medication, a view also highlighted by a minority of other participants. This was exemplified by their colleague reporting a practice of leaving other clinicians’ work even when there may be concerns, which contributed to inappropriate prescribing:

Healthcare professionals working in non-specialist community settings reported managing the care of patients with mental illness, yet they also viewed psychiatry as being outside of their remit of responsibility; this belief influenced the subtle decision to not develop more detailed knowledge of the medications involved. This was reported amongst cases involving patients with both serious and less serious mental illness.

A minority described psychiatric practice as “paternalistic” (NURSE01) in nature which was reported to generate a power imbalance between patients and clinicians and may also extend to clinician–clinician relationships. Some participants explained how this power dynamic could result in exaggeration of symptoms in order to validate their prescribing decisions, which contributed to inappropriate prescribing:

The majority of participants discussed workload, time and shift pattern issues. Participants all described how service demand was exceeding the capacity it was designed to handle and how an increased workload was not matched by an increase in staffing. High workloads meant staff reported not having the time to work carefully, communicate with each other effectively or undergo necessary training, which contributed to them making errors because they were feeling stressed and distracted. One specialist pharmacist described how these factors resulted in inappropriate prescribing from clinicians working in a specialist service:

A minority of participants described how workplaces were struggling to recruit as well as retain staff. Absent staff due to sickness, holidays or training left the remaining staff with a higher workload. As a result, participants described relying on agency/locum staff and staggered shift patterns, affecting the skill mix within the team and negatively impacting continuity of care. One locum community pharmacist described how a lack of continuity and variable shift patterns impacted their ability to gather all of the necessary information, resulting in an omission dispensing error:

Nearly a quarter of participants discussed the design of clinical resources, protocols and drug names specific to psychiatry as contributors to error. Clinical resources and protocols regarding shared care (defined earlier) were reported to be unclear and confusing, causing staff to misinterpret them or have difficulty locating the required information. For example, one general practice pharmacist highlighted that it can be difficult locating what information should be included in shared care agreements due to poor organisation of the website resources. The perceived similar sounds of drugs such as “tramadol/trazodone” or “zuclopenthixol acetate/decanoate” caused confusion as illustrated by one GP’s account of an error, when promazine was prescribed instead of promethazine:

Verbal and written communication between specialist mental health teams, general practices and community pharmacies was discussed by over half of participants as an important contributor to medication errors. Specifically, they described how it was difficult to communicate with other clinicians due to a lack of contact details and delays in receipt of clinical letters/information between services - this led to mainly prescribing and monitoring errors. Participants also described the lack of clinical updates and the absence or ambiguity/brevity of the clinical information that was communicated. This led to uncertainty in how to act due to being unaware of critical details that may have otherwise influenced their decisions. Participants highlighted that there could also be brief, as well as absent, verbal and written clinical information from the clinician to the patient. A general practice pharmacist described how a letter from specialist mental health services did not highlight that baseline physical health monitoring had not taken place resulting in a monitoring error:

Over a quarter of participants discussed the ideas of team structure, inclusion and appropriate feedback loops. Participants described the increasing range of healthcare professionals working in community care services and the complexity this brings to the healthcare system. One GP described the team structure within their practice and how this had changed over the years which created a skill mix within the team (as described earlier), with each professional discipline having variable knowledge regarding practice procedures. These contributors impacted continuity of care and team skill mix which ultimately led to a prescription for an inappropriate indication:

Some pharmacists reported that they felt excluded from the healthcare working environment and this was exacerbated by the healthcare system placing emphasis on particular roles such as GPs. Ultimately, this was reported to result in clinicians working independently without knowing the defined roles of colleagues and was linked to diffusion of responsibility which left the patient in “no man’s land.” (GP04) One general practice pharmacist outlined how they felt excluded from carrying out medication related tasks. The exclusion of their expertise contributed to a colleague’s omission and duplication prescribing error as the pharmacist was not able to provide input into the clinical scenario:

A minority of participants highlighted that there could be a lack of continuing supervision and support within their workplaces. Participants reported initial training but no ongoing support which resulted in clinicians not being aware they were doing tasks incorrectly which led to the risk of prescribing, dispensing and administration errors.

Participants discussed that patients with mental illness could be associated with some unique challenges. During episodes of illness, patients were described as presenting to services in an agitated state. A specialist nurse described how a patients’ challenging behaviour led clinicians to prescribe differently to how they normally would have, which itself contributed to overprescribing:

As highlighted in the previous quote, continuity is important as this pattern of response by clinicians was linked to inexperience of individual staff and was more likely to occur during periods of high demand on services. The specialist nurse quoted above goes on to describe how these illnesses may cause difficulties with patients’ understanding of their medication regimens which contributed to a prescribing error:

Less than a quarter of participants discussed social factors for this patient group. Clinicians reported that some patients may have lived alone and had no family, friends or social support to assist them with regular challenges such as remembering to take medication or attend appointments which contributed to administration and monitoring errors, respectively. This patient group were also reported to lead “chaotic lifestyles” (GP02), i.e. complex psychosocial factors, which participants linked to patients’ difficulties in arranging and attending appointments. One GP described the various modes of communication as well as multiple attempts required to contact patients which contributed to a monitoring error (absence of tests) for an extended period of time:

Participants recognised that patients’ poor engagement with healthcare services was also on occasion a result of their symptoms of mental illness; for example, anxiety or paranoia preventing patients from travelling to monitoring appointments.

A minority of participants discussed a perceived lack of resources (e.g. staff, equipment, funding) necessary to provide effective community care services for this patient group. It was felt that resource constraints within secondary care led to patients being inappropriately discharged to the community to create space, as described by one GP which contributed to absence of monitoring:

A minority of participants reported that there were a high number of patients with mental illness in community care, due to discharges that were deemed to be inappropriate. This problem was exacerbated by the perceived high threshold criteria placed for patients to access secondary care mental health services, resulting in clinicians in the community being left to support patients due to referrals being rejected. This was exemplified by one GP who described a prescribing error for a patient whose complexities, they felt, warranted input from secondary services and required local political involvement to have a referral accepted into secondary services:

Structure and work cultures were mentioned by a minority of participants. Clinical teams were reported to have allowed a culture to develop in which care management shortcuts became routine practice in order to meet organisational goals. These working practices were reported across multiple community services and were used because teams felt that they improved working efficiency with less consideration of whether it was beneficial to the patients and the organisation. This was outlined by one nurse who described how one clinical team had created an environment where they adopted shortcuts in order to meet patient demands. Despite education/training interventions to change their practice, staff still continued to inappropriately prescribe sedatives:

Policies that were not adapted for the needs of this patient group were reported as a contributory factor in medication errors. For example, participants described a policy to discharge patients without follow-up after multiple occasions of appointment non-attendance, however, this approach was deemed inappropriate for those with mental illness whose ability to attend can be a direct result of their mental illness and an indicator of condition deterioration. One general practice pharmacist described how this policy led to a complex patient being managed solely by community care when the service were ill-equipped to support the patient, which ultimately led to a colleagues’ duplication/overprescribing error of multiple antidepressants and mood stabilisers:

Respondents described misplaced and inappropriate targets, which exacerbated the lack of follow-up. Policies allowed clinicians to adopt more of a checklist approach in order to meet criteria to receive payment/funding, after which there was perceived to be less impetus to follow-up patients who do not attend appointments, which resulted in a monitoring error as highlighted by one GP:

The analytical and error frameworks used for data analysis and interpretation are well established and have been used in other health and medication error research (20, 42). A selection of transcripts were independently read by members of the research team to verify any themes emerging from the data and the final set of themes were presented to a patient advisory group and verified. Participants were asked to talk about errors that they made themselves or had direct first-hand knowledge of why an error occurred to ensure contributory factors were not based on generalised views. However, the majority of errors discussed by participants were made by another clinician which raises the possibility that they would not have had full insight into what other people were thinking at the time of the error, even if that person had discussed the incident with them. Future research should explore accounts from professionals who made the error. There may have been a degree of social desirability bias whilst participants recalled incidents due to the sensitive nature of the topic; participants may have modified accounts to make them sound less serious, which may have limited data availability. As participants discussed accounts that happened in the past, there is a potential for recall bias. However, this risk was minimised by encouraging participants to discuss errors that were significant in their practice, as well as allowing participants to use resources such as emails, medical records etc. to aid in the recall of events (these were not shared with the research team). Self-serving attribution bias may also have been a possibility as respondents may have deflected blame onto others to place themselves in a more positive light (73, 74). Previous research has highlighted limitations of remote interviews such as lower quality data and difficulty building rapport (75). This was mitigated by following an interview schedule which helped ensure consistency of data being obtained and the use of a webcam to allow the interviewer to respond to visual cues. There was underrepresentation of some professional groups, for example, psychiatrists. However, there was representation from all eligible professional groups and data collection continued beyond the point of saturation. The research teams professional networks mostly consisted of contacts within the Northwest of England which may impose regional bias. However, there were no differences in accounts between regions and there was representation of professionals from other regions in the UK, as snowballing was encouraged to reach participants that the research team did not have contact with. Finally, error accounts were from the viewpoint of one professional within a healthcare chain and there is the possibility that some details were missed in the series of events, which may have affected their awareness about local versus wider system factors.