Missed presentations, missed opportunities: A cross‐sectional study of mental health presentation undercounting in the emergency department

The burden of mental illness is increasing across developed countries. EDs are often used as access points by people experiencing mental health crises, with such rising demand in Australasia. Accurate data is critical to track and address this need, but research suggests that current data collection methods undercount mental health presentations to the EDs. The present study aimed to quantify and characterise ED mental health presentations that were not identified by usual clinical coding processes.


Introduction
New Zealand (NZ) EDs are uniquely positioned in providing a 24-h, selfreferred service that is free for NZ residents at the point of access. For these reasons, people may attend an ED instead of other providers, like general practitioners, which incur charges. There are around one million attendances each year across 43 publicly funded EDs, with higher rates at weekends and in winter and summer months. 1 Patients are triaged on arrival and prioritised according to urgency of physical and mental concerns. Staff will aim to complete physical examination, laboratory tests and imaging to decide on treatments, referrals and admission or discharge within 6 h, a target that • This study demonstrated that clinicians are recognising and documenting signs of mental illness, but this is not always reflected in diagnosis and coded data. • Deliberate self-harm is commonly recorded as physical injury rather than a mental health problem. • There was an association between contact with mental health services and likelihood of being coded correctly when presenting with a mental health problem.
many EDs are not achieving. 2 In NZ and internationally, the rate of mental health-related presentations is increasing more rapidly than overall ED attendance. 1,3- 6 The prevalence of mental health disorders in ED attendees is high; 48% of patients admitted from one ED for observation (all causes) had anxiety or co-morbid psychiatric conditions. 7 Presentations 'primarily' attributed to mental health complaints are less common, ranging from 5% to 8% of ED attendances. [8][9][10] This highlights the inadequacy of a single primary diagnosis; mental health issues may be a secondary or tertiary diagnosis or evolving and not yet meeting diagnostic criteria, so deprioritised or unrecorded. Information on ED utilisation for mental health may offer insights into unmet need, unreached groups, or overflow from other services, and offer clues on ways to alleviate pressures on EDs.
Most studies seeking to identify and analyse mental health presentations to the EDs concede that undercounting of cases presents a significant threat to validity. 8,11,12 Attempts to quantify this undercount have mainly focused on self-harm, suicide attempts or substance use disorders rather than mental health presentations in general. 11,[13][14][15] This hinders efforts to identify need, direct resources and enable interventions in the area of mental health. The aim of the present study was to quantify and characterise ED mental health presentations that were not identified by usual clinical coding processes.

Setting and sample
The ED of Wellington Regional Hospital serves a population of about 324 000. 16 All presentations occurring between 1 July 2017 and 30 June 2018 of patients aged 10 years and over were included, as marked increases in mental healthrelated ED presentations have been observed from this age. 17

Standard recording of ED data
After assessing every patient in ED, doctors select the most suitable diagnosis from a drop-down box, with optional fields for additional diagnoses. Any patient treated in ED for more than 3 h and/or admitted to a ward is given a diagnostic code from the International Classification of Diseases, 10th revision (ICD-10), a requirement that forms a national dataset. 18 Additionally, free-text electronic notes are entered by triage and ED nurses, doctors and multi-disciplinary mental health clinicians.

Identification of cases
In Phase 1 of the sample selection process (Fig. 1), both the ICD-10 codes (present in only 53.1% of presentations) and primary discharge diagnoses (available for all presentations) were screened for mental health presentations by SK (ED and mental health nurse). Appendices S1 and S2 contain the coding framework and a two-researcher coding procedure. Presentations identified as mental health included mental health and addiction diagnoses and intentional self-harm. Acute intoxication and tobacco use, and dementias, delirium and intellectual disability were excluded. Presentations identified as 'unclear' whether mental health-related (n = 384) included accidental harm (without detailing intent within coding or discharge diagnosis), foreign body ingestions, altered mental status (e.g. hallucinations without attribution to mental illness or substance use, insomnia, non-compliance to medication, unspecified somatoform disorder).
This article describes Phase 2 of the study (Fig. 1). The electronic clinical records relating to the 384 'unclear' presentations were accessed and reviewed by CW (trainee psychiatrist) using the same criteria as Phase 1 to determine whether the underlying problem was mental health. The review was restricted to the ED documentation relating to the index presentation. The majority of cases were easily classified on examination of clinical documentation. A minority (approximately 5%) were ambiguous; group consensus (CW, SK, SEP, AF) was reached by discussion of each (anonymised) case to ensure internal validity. Diagnoses coded as substance-related (e.g. injuries sustained while intoxicated) were noted but excluded from analysis as it was impossible to determine from isolated presentations whether the substance use would reach the threshold of a 'disorder' according to diagnostic criteria (ICD-10).

Variables extracted
For each mental health presentation, the data extracted included demographic information (age, sex, ethnicity), deprivation status (NZ Index of Deprivation, NZDep, a small-area based measure of socioeconomic deprivation, based on each patient's residence at time of presentation), 19 time of presentation, whether a patient was a current or past (within 5 years) client of specialist mental health services, and whether they were admitted or discharged from ED. Presentations were further categorised by reason for presentation identified from clinical notes (Fig. 2). More specific diagnoses were not given because of clinician notes lacking the level of detail required.

Statistical analysis
Statistical analysis was conducted in R 4.0 (R Institute, Vienna, Austria). Differences in case numbers are presented as the percentage of additional cases picked up through the clinical note review (Phase 2) relative to the case count for the clinical code/ diagnosis screening (Phase 1). These differences are presented as percentages with 95% confidence intervals (CIs; percentage estimated as the ratio of the two counts, assumed to follow a Poisson distribution, using the poisson.test function in R). These are presented for the overall count and by sociodemographic and clinical subgroups, with Pearson's chisquared test applied to test whether these undercount rates differed by subgroup.

Ethics and M aori consultation
The

Results
Phase 1 (clinical code/diagnosis screening) identified 1988 mental health presentations. Phase 2 (clinical note review) revealed 91 additional mental health presentations that had not been evident through Phase 1 (Fig. 1). This equated to an additional 4.6% identified mental health presentations (95% CI 3.7-5.6), corresponding to a 4.4% share of presentations identified through Phase 1 and 2 (Fig. 2). Table 1 shows the characteristics of these patients. The 2079 mental health presentations comprised 4.2% of the 49 170 presentations to the ED in the 12-month period. There were 58 presentations where it was still not possible to determine mental health presentation status because of insufficient information. Deliberate selfharm was the most common mental health presentation recorded without indicative discharge diagnoses or ICD-10 codes, with 45 of the 91 additional mental health presentations identified being as a result of poisoning using pharmaceuticals or household agents, self-injury or suicide attempts (Fig. 2).
There was insufficient evidence to indicate whether age, sex, ethnicity, deprivation status or time of presentation were associated with differential levels of presentations being correctly recorded by existing coding processes, or only discovered by in-depth clinical record review (Table 1).
Patients who had never been under local specialist mental health services were less likely to have their mental health presentation coded in the records (e.g. the non-client group contributed an additional 7.5% of cases from the notes review; χ 2 (2) = 16.48, P = <0.001). Patients with mental health presentations discharged directly from ED were also less likely to have their mental health problems coded (7.2% additional cases from notes review, compared to only 1.8% additional cases for those admitted; χ 2 (1) = 31.23, P = <0.001).

Discussion
Although the true number of mental health presentations to an ED may remain hard to count, the present study provides estimates of the scope of undercounts when relying on routine coding data. To our knowledge, this is the first study seeking to uncover hidden mental health presentations to the ED using clinical notes in addition to routine coding data. Our findings demonstrate that relying on discharge diagnoses and ICD-10 codes alone missed a substantive number of mental health presentations, with almost an additional 5% identified where clinicians had documented clear signs of a mental health presentation in the full notes. The present study indicates that currently there is no straightforward, efficient way to accurately quantify this volume based on common routinely collected data. Only in July 2021 did NZ adopt a unified approach using the Systemized Nomenclature of Medicine Clinical Terms (SNOMED CT) to record ED presenting complaints and diagnoses. Whether this is fully capturing mental health presentations is unknown.
Patients who had previous contact with mental health services had a smaller undercount in the coded data. Learning from a patient's health record that they have already been diagnosed with a mental health problem may influence a doctor's line of enquiry and sidestep the need to forge a new diagnosis when signs and symptoms are consistent with an existing diagnosis. However, this may also be the mechanism by which diagnostic overshadowing occurs, whereby physical symptoms are misattributed to mental illness. 20 Conversely, patients without a history of mental health service contact were less likely to be coded for their mental health presentation. ED clinicians may feel they lack time, skills or resources to first diagnose and communicate with patients presenting with mental health problems. 21 Many patients recorded as being 'admitted' are transferred to a shortstay observation unit (distinct from ED to allow for further observation and treatment, while avoiding blocking ED beds), others go to medical, surgical or psychiatric wards. Admitted patients were more likely to have had their mental health presentation recorded in the ED data, although it is not possible to establish the direction of this relationship. As beds are a limited resource, a threshold necessitating admission may only be met by serious illness and higher risk. Clear reasons to justify an admission are more likely to be coded as a diagnosis. At present, a risk-based approach predominates. 22 Patients judged to be high risk of harm to themselves or others will be prioritised and sometimes even compelled (by use of mental health legislation) to undergo assessment. Those considered at low risk may be discharged and advised to see their general practitioner. Hence mild to moderate presentations may be overlooked and under-served.
Negative attitudes towards selfharm among ED staff are wellrecognised. 21 Perceived stigma may also be reflected in this data, with deliberate self-harm frequently being recorded as physical injury rather than a mental health problem; these cases comprised almost half of the 91 uncoded mental health presentations. This highlights the need for ED information systems that allow vital information regarding intent or mechanism of injury to be recorded as part of the diagnosis, alongside physical details. 23 Patients with mental health diagnoses often face stigma, both interpersonal and structural, 24 which may deter ED doctors from making diagnoses without a high level of certainty, arguably reflected by high specificity in their mental health diagnoses. 25 Many mental health diagnoses need longitudinal assessment and cannot be made following a one-off ED presentation, but recording clinical impressions may assist with future diagnostic assessment, while still allowing diagnostic uncertainty and not increasing workload. Recording this in a way that data can be easily collated (rather than embedded in freetext notes) would improve quantification of mental health presentations, and support calls for increased resourcing for mental health support within EDs. 6 A significant proportion of presentations remained 'unclear' even after indepth review of the clinical notes. Although this partly reflected the high threshold set by authors for defining a mental health presentation, it also demonstrates that essential information is being omitted from the discharge and   6-9.2) †Phase 1: presentations found on clinical code screening (discharge diagnoses and ICD-10 codes). ‡Phase 2: presentations found on clinical note review. §n = 1 with sex recorded as 'Other', excluded from further analysis. ¶NZDep: a residential areabased measure of socioeconomic deprivation in New Zealand, with 1 being the least deprived quintile and 5 being the most deprived quintile. coding data, and may never have been established. Training for ED clinicians in mental health screening and assessment may lead to greater recognition and documentation, 26 and more patients accessing treatment. It has been established that longer duration of untreated symptoms is associated with poorer outcomes, highlighting the importance of timely diagnoses and appropriate care. 27 Research into whether care differs for patients whose mental health presentations are not recorded correctly, and gaining understanding of their experience in the ED, may help to identify mechanisms to improve care.

Limitations
This was a single-centre study, and so care must be taken in applying these findings to other centres because of variations between EDs in coding and clinical practices, departmental and wider culture and population. We suspect the same coding and data issues would apply elsewhere in NZ and internationally. 11,13,14 The finding that mental health constituted 4% of total presentations to the ED is lower than other studies, reinforcing that the populations/systems under study and methods used to analyse data may not be comparable. [8][9][10] Related to fewer-than-expected presentations being found, small numbers of some clinical characteristics may limit the accuracy of comparative analysis. This could be overcome by a longer study period, or a multi-centre study.
Manually reviewing clinical notes is likely inadequate for identifying mental health presentations given that routinely collected hospital data may identify only a third of intentional selfharm. 14,28,29 Since notes were written by the same clinicians making diagnoses, if doctors do not ask the pertinent questions and therefore miss mental health symptoms, 7 this retrospective review will be subject to the same bias. Conversely, 'diagnostic overshadowing' may be present in some cases, leading to false positives, 20 which cannot be detected by the present study design. Coding of diagnoses is imperfect and often dependent on clinicians, whose abilities can be affected by factors inherent to the ED environment, such as time pressures or tiredness. 25 Substance use was not possible to evaluate with the present study design, and may mask commonly comorbid mental illness. A prospective study to establish numbers of mental health presentations may be able to minimise false positives or false negatives, and also consider mild or evolving mental health presentations by using assessment conducted independent from ED clinicians.
Phase 1 screened only the first diagnoses and ICD-10 codes (additional ones were noncompulsory and variably present). Incorporating all available diagnoses and codes may have increased sensitivity, but may have compromised specificity since we sought to count not co-existing mental health issues, but cases where mental health was instrumental in the presentation. Alternatively, manually reviewing every ED presentation in the 12-month period (rather than only presentations considered 'unclear') would overcome these limitations, but would have been unfeasible with over 49 000 presentations to review. The chosen methodology would have been improved by using blinded dual assessment with adjudication to minimise assessor bias.
Description of variables such as triage category, referral and length of stay was beyond the scope of the present study; authors intend to present this in a subsequent paper.

Conclusion
In this ED, some mental health presentations were unable to be extracted, counted or reported. As ED data on presentation numbers, patient characteristics and clinical practice are captured to inform on trends, to raise 'red flags' and to evaluate where to do better, our findings are concerning. Further research is needed as to why this occurs, and whether the current information systems hinder the collection of mental health data in the ED.