A dangerously underrated entity? Non-specific complaints at emergency department presentation are associated with utilisation of less diagnostic resources

This study is a retrospective analysis at the ED of the Bern University Hospital in Bern, Switzerland. Our ED is a tertiary care centre, caring for a patient population of around 2 million and treating over 50,000 adult patients each year with an interdisciplinary team [23]. All patients are triaged by registered nurses using the Swiss triage scale that encompasses four levels (life-threatening, highlyurgent, urgent, non-urgent) [24]. Every Swiss inhabitant is obliged to obtain healthcare insurance that - in its base tariff - covers the costs of healthcare and medication. Private complementary medical insurance is available to pay for services not covered by mandatory health insurance (e.g. complementary medicine, acupuncture), in order to ensure free choice of hospitals or doctors and preferred hospital accommodation. However, this has no influence on emergency examinations, since in Switzerland these are reimbursed as follows: In the ED, all procedures are coded by the person who performed the procedure - using a procedural code of the Tarmed Suisse catalogue (TARMED Suisse. TARMED 01.08.0000). For each procedural code, a number measured in tax points is assigned, that reflects the effort of the procedure, independently of medical insurance status. There is no flat rate per case or DRG system (diagnosis related groups) for emergency treatments. The tax point is a hospital-specific, medical currency (1 tax point is roughly 1 Swiss Franc) and may depend on the individual hospital.

This study was classified as a quality evaluation study by the local institutional review board (KEK-2018-00198) and the need for informed consent was waived.

We randomly chose 1500 patients encounters from all adult patients (> 16 years of age) presenting to the ED between January 1st, 2013 and December 31st, 2017 triaged as urgent to highly-urgent, with a medical presenting complaint (Fig. 1. Flowchart). Life-threatening emergencies (around 8% of our patients in the ED) were excluded, as these are generally cared for within one of the three resuscitation bays and usually consume a lot of resources in a short period of time, as they do not stay long until being referred for further treatment (i.e. ICU, operation theatre). Non-urgent acute patients are assessed as being in a stable condition of health that does not actually require emergency medical therapy, and were therefore excluded (i.e. follow-up, concern related to a non-urgent health problem, medical certificate that is required for a specific administrative purpose, loss of a medical prescription).

Patients were sampled from all patients with complete medical and administrative records, and no previous medical contact for the current admission to the ED (previously seen by general practitioner, internal (i.e. specialist clinic) or external (i.e. external hospital, external specialist) referral, repatriation from a hospital in another country).

Sociodemographic (age, sex, Swiss nationality, private complementary medical insurance and medical (important comorbidities and drug intake) patient characteristics, encounter characteristics (time and day of presentation (night-time admission to ED 7 pm-7 am; effective weekend admission to ED 7 pm Friday to 7 am Monday), transport by emergency medical service (EMS), Swiss triage scale, and patient outcome data (length of stay in the ED and hospital, intensive care unit (ICU) admission, in-hospital mortality, ED return visits for 30 and 365 days, respectively), as well as study outcome data (physician and nursing, material and medication, laboratory, and radiological resources) were collected from the hospital’s electronic patient documentation systems (E-Care, ED 2.1.3.0, Turnhout Belgium and i-pdos_Prod_ODA 7.10.1.2).

Important comorbidities i.e. COPD, diabetes, liver disease, dementia, malignancy, cerebrovascular disease, peripheral artery disease, past myocardial infarction, and chronic kidney disease in the diagnosis list and specific drug intake were determined through a previously validated full-text parser of the ED medical report (see S1 Table). The Anatomical Therapeutic Chemical (ATC) classification system was used to categorise medication intake. The following groups were used: antidiabetics (ATC code A10), antithrombotics (B01), antihypertensives (C02, C04-C09), diuretics (C03), opioids (N02A), antiepileptics (N03), and psycholeptics (N05) [25].

Nemec et al. defined NSC as an entity of complaints that is not part of the set of specific complaints or signs for which evidence-based management protocols for the ED physician exist (e.g. “generalised weakness”, “decreased general condition”, “feeling exhausted”, “dizziness”) [13]. As they also excluded trauma patients, we focussed our investigations on patients presenting with medical complaints. Other investigators have applied the same definition using this “rule-out” or “absence of specific symptoms” [5, 11]. The presenting complaint collected from the hospital’s electronic patient documentation system was classified as either specific or non-specific according to the proposed scheme by two authors independently (TB, AG). In case of discrepancies between the two reviewers, a third reviewer (TCS) was involved.

The primary outcome was the total utilisation of ED diagnostic resources, defined as the sum of all physician resources (physician patient = consultation time; physician admin = physician administrative work; physician medical record; physician other = other physician services, e.g. arterial puncture; counselling = specific medical counselling, e.g. smoking cessation), as well as nursing, material (e.g. wound dressing, intravenous catheter, but also including medication costs), laboratory, and radiological resources. In the ED, all procedures are coded by the person who performed the procedure - using a procedural code of the Tarmed Suisse catalogue (TARMED Suisse. TARMED 01.08.0000). For each procedural code, a number measured in tax points is assigned, that reflects the effort of the procedure. The tax point is a hospital-specific, medical currency (1 tax point is roughly 1 Swiss Franc) and may depend on the individual hospital.

Secondary outcomes were hospitalisation, length of stay in the ED and hospital, ICU admission, in-hospital mortality, as well as ED return visits within 30 and 365 days.

The geometric mean of a distribution can be obtained by taking the exponentiated mean of the ln-transformed values. On the assumption that the mean of the ln-transformed total ED utilisation of diagnostic resources was about 6.55 (geometric mean of 700 tax points) with a standard deviation (SD) of 0.65, as based on a random sample of 100 consultations and a ratio between NSC- and non-NSC patients of 0.1, a sample size of 1500 patients is sufficient to show a difference of more than 20% in the geometric mean ratio (0.8–1.2) of the ED utilisation of diagnostic resources.

Data was analysed in Stata® 13.1 (StataCorp, The College Station, Texas, USA).

Baseline characteristics are presented as numbers and percentages or medians and interquartile ranges (IQR), using descriptive statistics as appropriate.

To assess the interrater agreement, we used Gwet’s AC - the test parameter of choice for the case of two raters and low prevalence data [26]. Gwet’s AC > 0.8 was considered to be a “very good” agreement [27].

Groups of patients with specific or NSC were compared with respect to presentation, ED and hospital outcome, as well as utilisation of ED resources, using Chi-square and Wilcoxon rank sum tests as applicable. Logistic regressions were performed to reveal associations with a binary outcome - i.e. NSC; the effect sizes are presented as odds ratios (OR) and 95% confidence intervals (CI).

As the primary outcome, total utilisation of ED diagnostic resources, was not normally distributed, the variable was ln-transformed before analysis. To reduce the risk of confounding, variables that showed at least evidence for a weak association (p < 0.2) – both with the primary outcome (utilisation of ED diagnostic resources) and exposure (NSC) – were considered as potential confounders [28, 29]. A stepwise backward linear regression analysis including all potential confounders was used to control the association between NSC and utilisation of ED diagnostic resources. The effect sizes of the linear regression are presented as geometrical mean ratios (GMR) and 95% confidence intervals (CI), as the exponentiated coefficients of a linear regression analysis with ln-transformed outcome corresponds to the GMR of the outcome without ln-transformation. The likelihood-ratio test was used to test if the final model improved adding an interaction term between i) NSC and age, ii) NSC and sex, iii) NSC and high triage category, and iv) triage category and age.

The consultation characteristics of patients with specific vs. non-specific complaints are detailed in Tables 1 and 2. Among the sociodemographic characteristics, age (per year increase) (OR 1.01, 95% CI: 1.00–1.02, p = 0.040), male sex (OR 1.53, 95% CI: 1.12–2.09, p = 0.007), and private insurance (OR 1.78, 95% CI: 1.21–2.61, p = 0.003) demonstrated higher odds for presenting with a NSC. In the context of multimorbidity, only the presence of diabetes and antidiabetic medication had higher odds for presenting with a NSC (OR 2.07, 95% CI: 1.36–3.15, p = 0.001 and OR 1.94, 95% CI: 1.20–3.13, p = 0.007, respectively). Weekend admissions were associated with lower odds for NSC (OR 0.65, 95% CI: 0.47–0.91, p = 0.011). The highly urgent triage category was significantly associated with lower odds for presenting with a NSC (OR 0.50, 95% CI: 0.36–0.70, p < 0.001).

For parameters of patient outcome, no significant difference was detected. If the analysis was restricted to hospitalised patients (n = 588), the patients presenting with an NSC (n = 72) had a significantly (p = 0.039) longer hospital stay (median 5.8 days, IQR: 2.9–9.6) than those presenting with a specific complaint (n = 516) (median 4.3 days, IQR: 2.1–8.0).

In univariate analysis, no significant difference in the total utilisation of diagnostic resources in the ED was found [specific: 844 (577–1313) vs. NSC: 778 (551–1183) tax points, p = 0.092, median (IQR)]. For the different resource categories, patients presenting with NSC utilised significantly more physician resources than patients presenting with a specific complaint [specific: 433 (323–581) vs. NSC: 485 (329–639) tax points, p = 0.048, median (IQR)]. Furthermore, utilisation of material, laboratory and radiology resources was found to be significantly less in patients presenting with a NSC (Fig. 2).

The associations of the various confounders with the total consumption of diagnostic resources are detailed in Table 3.

A backward selection logistic regression model including relevant confounders (variables that showed at least a weak association (p < 0.2) – both with the primary outcome (utilisation of ED diagnostic resources) and exposure (NSC), i.e. age, private insurance, triage category, night or weekend admission and on any antidiabetic) showed that NSC was significantly associated with lower utilisation of ED diagnostic resources [geometric mean ratio (GMR) 0.91, 95% CI: 0.83–1.00, p = 0.040], corresponding to 9% less utilisation of total ED diagnostic resources (Table 4).

Our results must be discussed in the light of some limitations. Firstly, this study was conducted at a single centre, albeit one of the largest of its kind in Switzerland, and thus generalisations should only be drawn carefully. However, 69% of the patients were walk-in patients, thus, the selection bias caused by the university character of our hospital might be small. The data used to describe our study population are definitely helpful in comparing another centre to our patient collective and to decide on the transferability of our results. Furthermore, we have developed a resource consumption score of our patient data elsewhere [22]. The identified predictors are widely accepted and even the magnitude of the confounders could be transferred to other hospitals, e.g. Lebanon, so that we are confident that our results are generalisable if handled carefully. Secondly, there is no gold standard definition of NSC. We used the definition introduced by Nemec et al. [13], which we recently demonstrated to be reliable and reproducible, even in a retrospective analysis [11, 30], and that excludes trauma patients. Furthermore, we focussed on patients with highlyurgent or urgent acuteness. Finally, we concentrated on the total diagnostic resources. Future analysis might consider specific steps of the diagnostic work-up of NSC patients as well as the clinical reasoning process [31, 32]. As with all retrospective data analyses, there are some potential layers of bias to be considered [33]. Most importantly, the quality of source data - the medical report and the correct procedural coding - might not be ensured. However, medical reports are usually checked by senior consultants and the staff is regularly trained by controllers to ensure correct procedural coding. Thus, this bias is thought to be small and in addition, can be assumed to be independent of the type of complaint. As regards data abstraction, the abstractors were experienced physicians and the interrater agreement with regard to complaint specificity was very good. In case of discrepancies between the two reviewers, a third reviewer (TCS) was involved.