Patient safety incidents within adult community-based mental health services in England: A mixed-methods examination of reported incidents, contributory factors, and proposed solutions

Abstract

Background

Relatively little is known about mental healthcare-related harm, with patient safety incidents (PSIs) in community-based services particularly poorly understood. We aimed to characterize PSIs, contributory factors, and reporter-identified solutions within community-based mental health services for working-age adults.

Methods

We obtained data on PSIs reported within English services from the National Reporting and Learning System. Of retrieved reports, we sampled all incidents reportedly involving ‘Death’, ‘Severe harm’, or ‘Moderate harm’, and random samples of a proportion of ‘Low harm’ or ‘No harm’ incidents. PSIs and contributory factors were classified through qualitative content analysis using existing frameworks. Frequencies and proportions of incident types were computed, and reporter-identified solutions were inductively categorized.

Results

Of 1825 sampled reports, 1443 were eligible and classified into nine categories. Harmful outcomes, wherein service influence was unclear, were widely observed, with self-harm the modal concern amongst ‘No harm’ (15.0%), ‘Low harm’ (62.8%), and ‘Moderate harm’ (37.6%) categories. Attempted suicides (51.7%) and suicides (52.1%) were the most frequently reported events under ‘Severe harm’ or ‘Death’ outcomes, respectively. Incidents common to most healthcare settings were identified (e.g. medication errors), alongside specialty-specific incidents (e.g. Mental Health Act administration errors). Contributory factors were wide-ranging, with situational failures (e.g. team function failures) and local working conditions (e.g. unmanageable workload) widely reported. Solution categories included service user-directed actions and policy introduction or reinforcement.

Conclusions

Study findings provide novel insights into incidents, contributory factors, and reported solutions within community-based mental healthcare. Targets for safety improvement are outlined, aimed at strengthening system-based prevention of incidents.

Introduction

Internationally, there is growing recognition of patient safety and healthcare-related harms (World Health Organization, 2021). Until recently, patient harm from poor quality or unsafe care in mental health services had received relatively limited research attention. Safety challenges in psychiatric services are increasingly acknowledged within emerging evidence and clinical service improvement agendas (Berzins et al., 2018; D’Lima et al., 2017; Thibaut et al., 2019). Indeed, self-harm, suicide, and violence constitute dominant concerns within the field (Bowers et al., 2011; Bowers et al., 2015; James et al., 2012), with national targets for their reduction within inpatient settings, alongside efforts to mitigate restrictive practices which follow (e.g. physical restraint), established within the English National Health Service (NHS) Mental Health Safety Improvement Programme (NHS England, 2021c).

Harms such as self-harm, suicide, and aggressive behavior can be understood as unsafe service user actions (Dewa et al., 2018). Challenges for the patient safety field, particularly in community-based care where the whole patient journey is often unknown, lie in determining the preventability of such events, which may arise as outcomes of unsafe care (e.g. ineffective risk management, misdiagnosis), but may also occur in the absence of identifiable service shortfalls (Averill et al., 2023). In many countries, mental healthcare is delivered primarily in community settings, yet patient safety outcomes in community-based mental health services remain poorly understood.

Patient safety incident (PSI) reporting by clinical teams constitutes a valuable, yet underutilized source of intelligence about patient safety challenges in mental healthcare (Barnes et al., 2022; Mishina et al., 2023). England and Wales’ National Reporting and Learning System (NRLS) was launched in 2003, encouraging voluntary reporting to generate learning about PSIs and identify solutions to mitigate risks to patients (NHS England, 2021b). These data have provided insights into various mental healthcare safety challenges, their amelioration, and remedial actions by professionals, including medication safety concerns (Alshehri et al., 2021; Dabba et al., 2019; Innes & Curtis, 2015; Young et al., 2020); prevention of and staff responses to inpatient suicides and self-harm (Bowers et al., 2011; James et al., 2012); and PSIs associated with discharge from specialist services (NHS England Patient Safety Domain, 2014).

In 2005, a quantitative overview of the relative frequencies of different PSIs within inpatient and community-based mental health services was published, using structured NRLS data about the incident type (National Patient Safety Agency, 2006). Safety improvement priorities were then examined within subsequent mixed-methods analyses focused primarily on inpatient safety, with themes such as aggressive behavior, sexual safety, and patient absconsion elucidated (National Patient Safety Agency, 2006). However, neither incident contributory factors, nor potential solutions to safety problems in community-based care, were explored. Moreover, it is also acknowledged that free-text incident accounts written by the reporting clinical team, as opposed to structured data, offer the greatest opportunities for learning from NRLS data (Mayer et al., 2016).

Given the above limitations, the present study comprised a detailed examination of PSIs within community-based mental healthcare settings for working-age adults. First, we aimed to characterize the nature of PSIs and their contributory factors as described within reports. Second, we explored the types of solutions to PSIs presented by incident reporters.

Method

Design

This study adopted a sequential exploratory mixed-methods design (Halcomb & Hickman, 2015). We examined incident types, contributory factors, and reported solutions using qualitative content analysis. Next, within each incident category, we computed descriptive statistics to explore the frequencies and proportions of each PSI type. Study reporting corresponds to the Reporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement (Benchimol et al., 2015).

Data source

We obtained anonymized PSI data from the NRLS database, managed by NHS England (NHSE). The NRLS is one of the largest databases of PSIs, receiving over 2.4 million reports from NHS or independent providers in 2021 (NHS England, 2022b). PSIs are also reportable to the NRLS by public members via an online form; however, data as to the extent to which this mechanism is utilized are not publicly available. NRLS data are used by multiple stakeholders, including the Health Services Safety Investigations Body, for identifying investigation foci (NHS England, 2021a). Through structured fields, reporters provide information about PSIs, those involved, and appraise the degree of harm related to the incident. Three free-text fields allow reporters to describe what happened, identify contributory factors, and report actions taken to prevent incident reoccurrence.

Eligibility criteria

We analyzed reports concerning potential or actual PSIs relating to community-based mental healthcare provision for working-age adults within primary or secondary care. Mental health patient safety was defined broadly: ‘The avoidance of unintended unsafe or iatrogenic harm associated with mental healthcare – either an error in inappropriate treatment, or an omission to detect unsafe behavior’ (Dewa et al., 2018). Therefore, besides PSIs recognized across most healthcare contexts (e.g. medication errors), reports describing potential or actual harmful outcomes for community mental health patients where service influence is less clear (e.g. suicides, interpersonal violence), were included. This patient safety definition was adopted over widely cited alternatives, such as ‘freedom from accidental or preventable injuries produced by medical care’ (Agency for Healthcare Research and Quality, n.d.), which insufficiently represent the range and mechanisms of potential mental healthcare harms (e.g. psychological harm from coercion, ineffective management of unsafe patient actions).

Sample selection

There is no agreed systematic way to sample NRLS reports relating to community-based mental health services. When refining our search criteria iteratively with NHSE support, we optimized sample relevance (specificity) over sensitivity. To retrieve data pertaining to routine care provision, rather than care delivered during the subsequent COVID-19 pandemic, a search executed by the NHSE Patient Safety Data Team (date: 16/12/2021) captured PSIs reported within 2019 (01/01/2019–31/12/2019), involving patients aged 18–65 years where care setting was defined as ‘mental health services’. To remove irrelevant reports from community physical healthcare, the search was limited to 22 English NHS Trusts identified by the authors as providing mental healthcare primarily or exclusively. Incident categories such as ‘failure to return from authorized leave’, and locations such as ‘prison/remand center’, ‘nursing home’ were excluded, owing to associations with institutional mental healthcare models or older adult services. Free-text data were then searched for terms associated with institutional mental healthcare settings (e.g. ‘escorted leave’) and excluded.

Guided by NHSE as to sample size considerations, all retrieved PSIs associated with reported ‘Low harm’, ‘Moderate harm’, ‘Severe harm’, or ‘Death’ outcomes were extracted, alongside a random sample of 1000 reports resulting in ‘No harm’, as judged by incident reporters. Due to large report numbers, the research team randomly sampled half of the ‘No harm’ and ‘Low harm’ reports received in the dataset. Supplementary Material 1 details the full search strategy, and Table 1 presents an overview of the proportion of reports sampled for analysis.

Table 1. Reports retrieved by search within sampling period and proportion sampled for analysis

Data coding and analysis

Free-text data were analyzed using qualitative content analysis (Elo & Kyngäs, 2008; Hsieh & Shannon, 2005). To incorporate learning from existing evidence-based PSI analysis frameworks, whilst retaining the flexibility to introduce new codes, incident type and contributory factors were examined with a hybrid inductive-deductive approach (Aim 1). Conventional inductive qualitative content analysis methods were applied to explore reporter-identified safety solutions (Aim 2), where appropriate coding frameworks do not already exist. Reports were treated as single units of analysis (Elo & Kyngäs, 2008). Data were coded for explicit, manifest content only. Where reports described multiple potential incidents or harms, including chains of contributory incidents, the incident type was classified according to the primary incident most proximal to the outcome for patients or staff.

One researcher (P.A.) iteratively reviewed a stratified random 10% subset of sampled reports (n = 183), making inductive open coding notes within Microsoft Excel (see Supplementary Material 2 for examples). Informed by this preparatory sense-making analysis phase, several tools for classifying incident type and contributory factors were then applied to the subset of reports to assess their suitability (Aim 1) (Barnes et al., 2022; Berzins et al., 2018; Carson-Stevens et al., 2015). We selected a framework adapted from the Primary Care Patient Safety (PISA) Classification System for characterizing incident type and outcome (Carson-Stevens et al., 2015), with the Yorkshire Contributory Factors Framework – Mental Health adaptation (YCFF-MH) used to code contributory factors (Berzins et al., 2018). Several new codes for the incident type were inductively generated and applied to all relevant reports (see Supplementary Material 3). Frequencies and proportions of incident types were calculated, and organized by incident category. Inductive qualitative content analysis of reporter-identified safety solutions (Aim 2) began with a familiarization phase as described above. The range and nature of solutions to PSIs described within the dataset were then inductively coded.

Once the coding manual had been piloted and finalized (Supplementary Material 4), a second coder independently coded 10% of reports (divided equally between C.H., Consultant Psychiatrist and N.S., senior patient safety expert) and Cohen’s kappa statistic was computed to assess incident type agreement adequacy. Coding disagreements were discussed amongst coders; the remaining researcher provided independent arbitration. The remaining sample was systematically coded by one researcher (P.A.). Illustrative quotations and descriptive statistics are presented to elucidate key findings.

Research ethics considerations

The Health Research Authority confirmed study exemption from research ethics approval, as NRLS data are anonymized before researcher analysis (IRAS ID: 311051). Where reports are quoted, minor editorial adjustments were made to further ensure confidentiality.

Results

Of 1825 reports, 1443 met the analysis criteria. Eligible reports were all appraised to have been written by clinical services staff. We excluded 382 (20.6%) reports for reasons such as not describing a PSI or harmful outcome, providing insufficient detail for coding, or not relating to community-based mental healthcare. Cohen’s kappa statistic indicated suitable coder agreement (κ = 0.77) (McHugh, 2012). Table 2 details reported patient characteristics.

Table 2. Patient characteristics as documented within eligible reports

Table 3 shows the nature and frequency of each PSI type, disaggregated by reporting-staff-rated degree of harm. PSIs were classified into nine categories: investigations; documentation; referral; communication; administration; treatment and procedure; medication; diagnosis and assessment; and potential or actual harmful outcomes where service influence is unclear. The latter group described scenarios involving patient suicides, violence, or risks from others. No reports concerned a tenth incident category, about equipment-related PSIs. Using the YCFF-MH (Berzins et al., 2018), identified contributory factors and active failures (errors by clinical staff at the ‘sharp end’) are summarized for each incident type. See Supplementary Material 4 for code definitions and Table 4 for illustrative examples identified within reports. Finally, Table 5 outlines reported safety solutions by incident category.

Table 3. The nature and frequency of incidents at each reported degree of harm

Notes: Frequencies and proportions presented pertain to the analysis of samples of 16.6% (500/3020) of ‘No harm’ and 50% (459/918) of ‘Low harm’ reports out of the total volume of reports retrieved for these harm categories. During manual coding, further reports were identified as not meeting study criteria and were excluded (n = 382); 1443 eligible reports were tabulated. Proportions are rounded for ease of reporting and therefore do not always sum to 100%.

Table 4. The nature of identified contributory factors by incident category and type

Notes: See Lawton et al. (2012) and Berzins et al. (2018) or Supplementary Material 4 for descriptions of each factor. Several reports detailed multiple active failures and/or contributory factors.

Table 5. The nature of reported safety solutions for each incident category

Investigations

PSIs concerning clinical investigations were observed amongst ‘No harm’ and ‘Low harm’ reports only (Table 3). These included incidents stemming from inadequate support from other departments (e.g. laboratory services) culminating in delayed specimen processing. Key examples involved tests required for clozapine prescription, where delays rendered blood samples unusable. Proposed solutions, corresponding to incorrect test ordering, reinforced the need to clearly stipulate the specific investigation(s) required at the point of order (Table 5).

Documentation

Documentation PSIs, identified amongst ‘No harm’ and ‘Low harm’ reports only, comprised problems arising from unavailable, incomplete, or inaccurate healthcare records (Table 3). Active failures were often implicated in incident causation, including failure to document risks or loss of paper records (Table 4). Reported consequences of documentation PSIs included clinicians conducting home visits unaware of prior patient violence; risk of patient deterioration; and repeated clinical investigations:

‘Service user had three blood tests to be supplied with clozapine, advised by pharmacy that none of the results are on the system.’

Potential solutions were not addressed by reporters.

Referral

Referral-related PSIs included decision-making errors, such as inappropriate referrals, or decisions not to refer, for example onward referral to specialist support not being completed for patients who subsequently died by suicide (Table 3). Patients were encouraged to contact these services directly themselves, but it was not ascertained whether contact was established.

Referrals were also impacted by administration errors, including delays or failures in sending referrals. Active failures included a routine rather than urgent referral for eating disorders care, since the referring general practitioner (GP) did not indicate patient weight or acuity, nor did the receiving team seek to validate this information:

‘Patient waited seven weeks to be seen despite the target being four weeks and when seen was critically ill…’.

Proposed solutions comprised reinforcement or introduction of policies and procedures; improving communication and joint working between services; and establishing new systems for referral management (Table 5).

Communication

PSIs within this category included staff-patient communication problems and communication breakdowns between healthcare professionals (Table 3). For instance, due to insufficient handover between crisis team shifts, an individual at high risk of self-harm was not visited as intended. In other examples, patients received poorly coordinated care, or no care at all, owing to unclear lines of responsibility between teams: ‘There was no working together between different agencies.’ Finally, further reports described communication problems between healthcare professionals and non-healthcare professionals, such as receptionists, security staff, and the police. Reported solutions included proposals to optimize handover practices and configure joined-up electronic patient record access to foster improved communication and information-sharing (Table 5).

Administration

PSIs concerning care administration were observed at the greatest relative frequency among ‘No harm’ reports (13.0%; Table 3). Incidents included staff performance-related problems, for example confidentiality breaches, and professionalism issues. These active failures were typically unintentional errors, although procedural violations were also described:

‘Student nurse observed qualified nurse who did not wash hands between patients, did not use gloves and drew up risperidone incorrectly by using a needle through the rubber seal, then using the same needle to administer the injection to the patient.’

Further PSIs comprised problems in accessing healthcare providers, such as missed or canceled crisis care appointments. Reported contributory factors included staffing shortfalls and unclear lines of responsibility between services. Errors in legal administration were also inductively introduced into the coding framework (Carson-Stevens et al., 2015), including errors in Mental Health Act (MHA) administrative paperwork pertaining to patients treated under a Community Treatment Order (CTO):

‘Service user who I am care coordinator for was on CTO…CTO expired and I was not contacted by Mental Health Act office until two weeks afterwards.’

Multiple solutions were proposed (Table 5), including reinforcement or introduction of various policies and procedures: ‘Clinicians asked to use the caseload system [digital software] so that it’s easier to manage their caseloads in the event of unforeseen absence.’ Moreover, double checks were advised to prevent administrative errors.

Treatment and procedure

PSIs involving treatments other than medications included errors in treatment decision-making, with insufficient treatment, care, or monitoring provided:

‘There were some opportunities missed by the perinatal mental health service to assess her mental state and create comprehensive relapse prevention and crisis plans.’

Further reports detailed errors in treatment implementation or delivery, where active failures and communication breakdowns were cited as contributory factors (Table 4). These reports described care delivery that deviated from service procedures, including the failure to assign patients a care coordinator, or insufficient follow-up after patient non-attendance:

‘Received notification that service user was found dead…there is a need for staff to fully adhere to the ‘Did Not Attend’ procedures…no evidence that the keyworker was contacted and no written records that highlight any action was undertaken to re-engage them.’

Treatment not given in a timely fashion was a salient incident type, comprising 5.1%, 3.3%, and 4.7% of sampled ‘No harm’, ‘Low harm’, and ‘Moderate harm’ incidents respectively. Delayed admission for patients assessed as requiring hospitalization, voluntarily or under MHA legislation, was a key example. High bed occupancy within mental health units, and limited police availability to facilitate CTO recalls, were important contributory factors.

Proposed solutions were wide-ranging and included a focus on policies and procedures, where established policies for patient risk zoning were reinforced (Table 5). To address hospital bed availability constraints, further reports discussed improving working across community team boundaries, and exploring how crisis teams could better facilitate earlier hospital discharges:

‘Crisis team staff becoming more actively involved in patients admitted onto the wards to try to reduce length of stay.’

Medication

Medication PSIs comprised 17.1% of ‘No harm’ reports, among which administering errors (3.8%), dispensing errors (3.4%), and patient unintentional overdoses (3.1%) were modal incident types (Table 3). First-generation depot antipsychotics were implicated in multiple medication administering errors. Active failures, including staff attentional lapses and mistakes, were prominent contributory factors to medication-related incidents (Table 4). Service user factors were also reported, such as taking medications twice by accident, or concealing illegal medication supplies. Equipment and supply design issues involving similar packaging for different medications, or equipment malfunction, were also described:

‘Most of the depot was given before plunger failed and became stuck, so a small amount was unable to be administered.’

Proposed solutions included calls for improved information sharing between primary and secondary care services and optimizing within-team communication using a structured clinical handover format. Service user-directed measures (e.g. medication counseling), and staff training and supervision were among the further solutions clinicians recommended (Table 5).

Diagnosis and assessment

Table 3 indicates that diagnosis and assessment PSIs comprised over one-third of ‘Moderate harm’ incidents (35.9%) and around one-in-eight ‘No harm’ reports (12.3%). Examples included diagnostic errors and failures to act on patient symptoms:

‘Call from GP. He received an ECG reading [from five days ago]. ECG showed it was abnormal, patient was reporting chest pain and breathlessness…GP advised me that this was a significant cardiac event and [emergency services] should have been called.’

Delayed assessments – primarily delayed MHA assessments – constituted a sizeable proportion of ‘Moderate harm’ PSIs (33.2%). Outcomes included self-harm; deterioration; and risks to others:

‘Patient has relapsed into paranoid psychosis and is at high risk of harm to others, property, public, and above all himself – he has lost a great deal of weight, is looking gaunt, failing to eat and refusing all medication.’

Contributory factors to delayed MHA assessments were multiple (Table 4). Cancelled assessments were widely attributed to service user factors, for example patients not being home. Scheduling and bed management challenges were also commonplace, along with insufficient support from other agencies, including police or interpreter availability for assessments.

Amongst reported solutions (Table 5), proposed actions centered on policies and procedures, including the introduction of procedural changes around risk recording in triage letters. Measures to optimize within- and between-team communication and information sharing to improve safety were also cited.

Potential or actual harmful outcomes where service influence is unclear

Within each harm level, the majority of reports described harmful outcomes where the influence of mental health services was unclear. Self-harm was the modal event amongst ‘No harm’ (15.0%), ‘Low harm’ (62.8%), and ‘Moderate harm’ (37.6%) reports. Reports involving ‘Severe harm’ and ‘Death’ were characterized primarily by attempted suicides (51.7%) and suicides (52.1%) respectively (Table 3). Often, the influence of clinical teams could not be ascertained definitively for reports within this category. For example, from this excerpt following a service user’s death, it is unclear whether assessments were insufficient, or whether risks escalated after the assessment:

‘Staff at the residential home and the patient’s social worker had expressed concern that the patient had deteriorating mental health and may be a risk to himself. However, the mental health assessment informed the clinical decision that the patient did not require secondary mental health services follow up.’

Reported contributory factors are summarized in Table 4. Service user factors such as declining treatment, or impulsive acts exacerbated by intoxication, were widely cited by reporting staff. Proposed solutions in efforts to prevent the reoccurrence of such events included service user-directed actions, such as naloxone supply for reversing the harmful effects of opioid overdoses. Further solutions centered around dispensing pharmacist- or carer-led measures to restrict access to means for overdose or self-harm. Changes to service users’ care packages for preventing incident reoccurrence were also cited, for example referrals for supported accommodation, psychological therapies, or increasing consultation frequency.

Discussion

Key findings

To our knowledge, this is the first in-depth mixed-methods study of PSIs in community-based mental health services, advancing upon early descriptive statistical findings published shortly after the NRLS’ inception (National Patient Safety Agency, 2006). Informed by empirically developed incident analysis frameworks, nine incident categories (Carson-Stevens et al., 2015), and multiple contributory factors (Berzins et al., 2018), were identified amongst eligible reports. We also present novel findings as to the solutions clinical staff proposed for preventing incident reoccurrence.

Although many reports described issues observed within most care specialties or settings, including documentation, communication, and medication-related PSIs (Barnes et al., 2022; Carson-Stevens et al., 2015; Mishina et al., 2023), certain incidents and contributory factors appear unique to mental healthcare. We coded specialty-specific problems, such as errors in MHA legal administration (Table 3). Delayed MHA assessments and hospital admissions further threatened the safety of community patients, as did unintentional overdoses by patients. Such risks are likely heightened in community-based care, where patients and carers play greater roles in care delivery (Vincent & Amalberti, 2016).

Within each harm level, harmful patient outcomes (e.g. self-harm, attempted suicides, or suicides) were modal reporting foci; the influence of mental health services over the causation or prevention of these events could not be conclusively established. Such safety challenges align poorly with traditional patient safety definitions derived from physical healthcare contexts (Agency for Healthcare Research and Quality, n.d.). However, their salience within the sample suggests that mental healthcare professionals consider such harms as within the remit of patient safety. Nevertheless, a more decisive stance on report eligibility could be taken in future analyses, potentially excluding reports about patient deaths with no associated PSI documented.

Our analyses complement existing research in highlighting myriad contributory factors to mental healthcare-related PSIs, including latent organizational factors, local working conditions, and situational failures (Berzins et al., 2018). That active failures were observed widely is perhaps a surprising finding (Berzins et al., 2018), although reporters may more readily observe causal relationships between staff practice and PSIs in hindsight (Vincent, 2010). Such biases may be reproduced in secondary analyses of these data conducted by those who are external to events surrounding these PSIs. Consequently, the contributory influence of other factors may not have been adequately reflected.

Although neither policies and procedures nor training and education were recorded contributory factors in our sample (Berzins et al., 2018), solutions of this nature were widely cited across different incident categories. This suggests a mismatch between the described incidents and suggested routes to prevention. Overall, proposed preventative actions appeared weak (Veterans Health Administration, 2021), with poor explanations of how measures may prevent future incidents, and overreliance on reinforcing existing policies or service user-directed solutions. Nevertheless, what constitutes an effective solution may differ in psychiatric care contexts compared to other specialties, for which measures directed at the wider care system rather than at individuals may be considered stronger (Veterans Health Administration, 2021). For certain highly-reported challenges within community-based mental healthcare (e.g. self-harm), it is plausible that service user-directed solutions (e.g. medication review, restricting access to means) as part of wider therapeutic engagement, are considered best-placed to prevent harm.

Strengths and limitations

Our research offers novel findings about PSIs, their causation, and proposed safety solutions in adult community-based mental health services. Elucidation of mental healthcare-specific PSIs and harms, including those that are not routinely investigated, such as PSIs resulting in low-moderate levels of harm, constitute additional strengths of this research. Previous research suggests that the use of ‘other’ codes, or potential misclassifications of incident type are common features of structured NRLS data (Dabba et al., 2019; Härkänen et al., 2020; Mainey, 2020). Therefore, a further strength is the systematic, manual approach to coding, applying empirically-informed incident analysis frameworks (Berzins et al., 2018; Carson-Stevens et al., 2015). The sequential exploratory research design has allowed for rich, mixed-methods insights into these routine data (Halcomb & Hickman, 2015).

Study limitations stem primarily from the sampling and features of these routine data. There is no optimal means to systematically sample NRLS data from community-based mental health services for working-age adults; our sampling approach prioritized specificity over sensitivity. Likewise, although reporting cultures and systems are best-established within NHS Trusts, our sample will not have captured reports submitted by independent and voluntary, community, and social enterprise (VCSE) providers within England. Since such organizations are omitted from NRLS Organisation Patient Safety Incident Reports (OPSIR) statistics (NHS England, n.d.), reporting rates within these services remain unclear. Underrepresentation of ‘No harm’ and ‘Low harm’ reports within our search strategy represents a further sampling limitation. Our sample may have omitted key types of PSIs observed in clinical practice and therefore may not effectively establish the prevalence at which staff report different types of PSIs to the NRLS database. Furthermore, the quality of reporting about harmful patient outcomes (e.g. self-harm) often precluded examination of whether such events seemingly occurred following shortfalls in care and the extent to which they might have been prevented. Moreover, owing to reporting biases and practicalities, PSIs resulting in lower harm levels and those occurring in community settings are also particularly prone to underreporting in voluntary reporting systems (National Patient Safety Agency, 2006). However, our quantitative descriptive analyses sought to supplement and enhance our qualitative findings, rather than to support generalizable conclusions.

Although PSIs can be reported to the NRLS by members of the public using an online form, this mechanism appears not to be used widely. In our study, all reports were seemingly authored by staff within clinical services. Service users and carers are important sources of safety intelligence (Armitage et al., 2018); as such, their perspectives are vital. Existing research suggests that service user- and carer-centered paradigms of ‘patient safety’ within community-based mental health services, or at mental healthcare transitions, are more wide-ranging than those espoused by staff. For example, in two studies, service users and carers emphasized the importance of psychological aspects of feeling safe (Averill et al., 2024; Tyler et al., 2021). Whilst the prevention and management of risks (e.g. self-harm, violence) was considered a central tenet of safe community care provision amongst service users, carers, and staff alike, in practice, carers felt that they were left to hold risks with little support from services (Averill et al., 2024). Such nuances remain undetected by traditional incident reporting mechanisms.

Implications

Our findings have implications for a systems-informed understanding of PSIs in mental healthcare. We found evidence of the adverse conditions clinicians inherit within mental healthcare systems (Berzins et al., 2018), rendering unusable the resources or mechanisms designed to keep patients safe. Such problems included the unavailability of police support for MHA assessments, inadequate inpatient bed capacity, and insufficient staffing. Where shortages in UK mental health inpatient bed capacity and staffing are rising (NHS Digital, 2022; OECD, 2021), our findings reinforce imperatives to identify effective, scalable solutions to improve patient access and flow.

Our findings contribute to Safety-I perspectives (Hollnagel et al., 2015), namely, learning from what has gone wrong within mental healthcare. The ongoing national rollout of a new PSI data capture mechanism to succeed the NRLS – the Learn from Patient Safety Events (LFPSE) service – warrants attention in future analyses of routine patient safety data (NHS England, 2022a). Whilst we do not anticipate that the LFPSE launch will materially change reporting behavior and corresponding PSI data, this system also allows for instances of good care to be reported, potentially shedding light on the qualities of resilient systems that support safe care delivery. As such, this new functionality may have implications for Safety-II learning, concerned with ensuring that ‘as many things as possible go right.’ To strengthen the potential to learn from reported harmful patient outcomes, reporters should be prompted to reflect on their antecedents (where known) and to contextualize such events within safety and quality considerations across the wider patient care journey, including the extent to which effective risk management approaches were in place.

A further implication concerns the nature of PSIs selected for in-depth investigation. NHSE’s recent replacement of the Serious Incident Framework with the Patient Safety Incident Response Framework (PSIRF) will grant organizations greater agency over this matter (NHS England, 2022c). Historical requirements to investigate all PSIs classified as ‘serious incidents’ have resulted in repeated investigations of similar incidents. Relatively commonplace PSIs not meeting the ‘serious’ threshold have been insufficiently explored. Our findings suggest several safety challenges that may benefit from a detailed investigation. Indeed, further to harms associated with delayed MHA assessments, another such issue relates to harms stemming from disruption to patients’ medication schedules, with clozapine a widely reported example. Where close monitoring and supervision are required (Khawagi et al., 2020), PSIs such as errors in laboratory phlebotomy processing, documentation of results, medication administration, or supply issues, may culminate in patients missing doses and thus requiring re-titration after a treatment break (Dabba et al., 2019; Khawagi et al., 2020). Repeated clinical investigations, deterioration, and hospitalization were among the reported consequences. In-depth investigation of these routinely occurring problems may help to mitigate factors underpinning these pervasive system challenges.

More broadly, our synthesis of solutions to prevent incident reoccurrence, as proposed by clinical staff, provides an important overview for psychiatrists and other healthcare professionals; service managers; and commissioners seeking to improve safety. Based upon our systems-informed incident analysis, these stakeholders will be well-positioned to ensure that safety improvement strategies suitably target those contributory factors implicated in incident causation. Future research to drive improvement efforts should explore the feasibility and scalability of measures proposed by reporting clinicians. This will form a vital preparatory step towards generating broadly effective and implementable solutions to identified PSIs.