Association of dyspnoea, mortality and resource use in hospitalised patients

Discussion

Our study is the first large-scale quantification of risk of adverse outcome associated with dyspnoea among all non-ICU patients at the time of hospital admission. There are two reasons to assess dyspnoea: 1) to identify a common and uncomfortable symptom to improve patient comfort [9], and 2) based on our present findings, dyspnoea at any level can identify patients at increased risk of hospital resource use and death.

Prior large-scale investigations have examined the relationship of dyspnoea to risk of harm in particular categories of patients, e.g. those at risk for cardiac disease [16–18], pulmonary disease [19], gastrointestinal disease [20, 21] and others. Other investigations have looked at the relationship of dyspnoea to risk of harm in the general (non-hospitalised) population [8].

In contrast to prior studies of inpatients, our study does not limit the population of interest to a specific diagnosis or category of patients. This universal dyspnoea assessment has important practical consequences: institution of dyspnoea assessment for all patients is more effective, and in some ways easier, than if a diagnosis were required before collecting dyspnoea assessments. Furthermore, our data show that the most important predictions arose in patients who did not have a primary discharge diagnosis of cardiopulmonary disease; these were missed entirely by strategies used in prior studies.

To properly evaluate the benefit of any screening test, one must consider the burden, inconvenience and test characteristics, as well as the effectiveness, risk and cost of possible interventions based on the test results [22]. The bedside measurement of dyspnoea is promising in several ways. First, as we have previously described, the routine evaluation of dyspnoea performed at the time of admission and for each subsequent shift by nurses throughout our hospital is fast, feasible and inexpensive; each evaluation takes <1 min to complete and results are immediately available to care staff [10, 23]. Second, our results suggest that measuring dyspnoea at the bedside is a useful, straightforward way of identifying patients at risk for death during the remainder of the hospitalisation and after discharge. However, we do not yet know whether early intervention in dyspnoea, either in the hospital or on discharge, will improve outcomes for these patients. There are reasons to think this simple assessment will be useful: abnormal vital signs have previously been used in medical emergency team activation to align hospital resources with patients at high risk of clinical decline [24, 25]. The presence of any dyspnoea (i.e. a dyspnoea rating of ≥1) had a specificity of 89% and sensitivity of 30% to identify patients who were at risk of in-hospital death. Patient report of a dyspnoea rating of ≥4 had a specificity of 96% (at the expense of a sensitivity of 13%); the false-positive rate in our study (96%) was identical to the false-positive rate in the National Lung Cancer Screening Trial, which used far greater resources [26]. The absolute risk of death in our study for patients with dyspnoea on admission is comparable to the risk of death from lung cancer for the control group in the Lung Cancer Screening Trial [26]. Finally, apart from its utility as a signal of future outcomes, the symptom burden of dyspnoea is enough to warrant more aggressive attention and treatment [27]. Given the low cost of the screening, the burden of the symptom of dyspnoea for patients and the opportunity to obtain a powerful signal of harm to potentially improve outcomes, we see these results as a call to assess and document dyspnoea in all patients, and to investigate interventions to reduce adverse outcomes.

As we and other authors have noted previously, dyspnoea is most commonly associated with increased respiratory demand combined with cardiopulmonary limitations; ambulatory patients frequently moderate their activity to minimise respiratory demand and avoid discomfort [28]. We found that patients who report dyspnoea at rest prior to arrival at the hospital, and consequently are unable to moderate activity to mitigate symptoms, are particularly vulnerable to harm.

Pain is routinely measured across hospitals. We performed the same statistical analyses to test the relationship between pain and adverse outcomes. In contrast to dyspnoea, pain was not associated with adverse outcomes. In fact, patients with pain fared slightly better than those without pain, indicating that the observation was not due to lack of statistical power. We can imagine three possible reasons that pain did not predict adverse outcomes. 1) Many sources of pain (e.g. from broken bones) are not associated with critical homeostatic systems; an analysis restricted to visceral pain might yield different results, but this information was not recorded in the Initial Patient Assessment. In fact, the clinical service for >30% of the patients who reported pain ≥4 was either the general surgery or orthopaedic surgery service (supplementary table E8). 2) Pain is aggressively managed, which may remove much of the signal; this seems less likely as the range of pain ratings was similar to the range of dyspnoea ratings. 3) Pain is so routinely measured that when we measure pain in everyone, we may be enriching our denominator for the less sick patients, simply because of the routine prevalence of pain.

Our study has several limitations. First, it was conducted at a single, tertiary academic health centre. Second, diagnoses were identified based on billing data, which can only be elicited on discharge. These diagnoses were based on the Clinical Classifications Software proposed by the Agency for Healthcare Research and Quality [14] so as to standardise and replicate the designation of diagnoses into groups such as “respiratory diseases” or “cardiovascular diseases”; however, clustering of diagnoses may be overly broad. Third, real-world measurement of dyspnoea may vary depending on how a nurse or physician asks about patient symptoms; we know that in many cases nurses ask a yes/no question and record a 0 for a no answer, and in some cases nurses modify or replace the patient report with their own judgement based on signs [10]. Fourth, the exact timing of dyspnoea assessment in the Initial Patient Assessment is unknown; it is done sometime during the first 12 h in the unit. Furthermore, dyspnoea rating is not documented on arrival in our emergency department. For these reasons, we do not know the patient's dyspnoea intensity on arrival at the hospital and how dyspnoea has been modified by the first few hours of treatment. Fifth, we hypothesised that a dyspnoea rating of 4 would be a useful cut-off, having based this choice on our research group's finding that two-thirds of patients deemed dyspnoea <4 “acceptable” [29]. However, in the post hoc analyses to evaluate the utility of this cut-off, we found that a cut-off of 3 was associated with similar risks of death. We note that these findings challenge the utility of a cut-off of 4, suggesting that any elevation of dyspnoea is the only major distinction, which may lead to future investigation in both rating use and dyspnoea.

We have shown that a one-time measurement of dyspnoea during the first shift on the hospital unit has a strong predictive value for adverse outcomes; it is likely that repeated dyspnoea assessment at the time of arrival and throughout hospitalisation would further improve risk prediction [30]. This simple assessment can be utilised even in hospitals without the resources to provide more data-intense modelling of illness severity in real-time.