Predictors of outcomes in COPD exacerbation cases presenting to the emergency department

Study design

Following a feasibility study in three centres, the present prospective observational study was conducted in 113 centres representative of French hospitals. The centres were asked to include all consecutive patients visiting their EDs between November 2003 and February 2004 with a diagnosis of certain or probable exacerbation of COPD at entry. Included patients were followed until discharge from hospital.

Since this was meant to be a real-life study, there was no constraint or guideline regarding investigations, admission and discharge decisions or treatments. In each centre, an emergency physician and a respiratory physician were in charge of quality control.

Inclusion criteria

Patients were included if they presented with a probable or certain exacerbation of COPD, as defined by the appearance or increase of dyspnoea in a patient with known COPD or risk factors for COPD (age >40 yrs and current or past cumulative tobacco consumption of >10 pack-yrs). Patients meeting these criteria were not included when the investigator judged that the predominant diagnosis was or was likely to be pulmonary embolism, congestive heart failure, pneumothorax or pleurisy, or when the predominant underlying respiratory disease was asthma, bronchiectasis, heart failure, thromboembolic disease or restrictive respiratory insufficiency. All of these exclusion criteria were left to the clinician’s discretion in order to ensure that the real-life nature of the study was respected. Patients requiring immediate ICU monitoring and/or noninvasive or invasive ventilation, as judged by the ED physician, were not included.

Data collected

The data collected were limited to observations and results that are readily available in all EDs, and addressed the following issues.

Statistical analysis

Data are presented as mean±sd or percentage. The in-hospital outcomes of interest were death and requirement for post-hospital support (either convalescence, nursing home or domiciliary support). A two-step analytical plan was applied.

The first step was the exploratory assessment of predictors of death and the need for post-hospital support in the whole population in order to ensure maximal statistical power. Univariate analyses (Chi-squared test for categorical variables and an unpaired t-test or ANOVA for continuous variables) were first performed in order to determine which variables were associated with these outcomes. Then, initial symptoms and signs associated with an increased risk of in-hospital death on univariate analysis were aggregated in a clinical severity index, which was integrated into the subsequent analysis. Finally, multivariate analyses were performed; a backward stepwise procedure using a logistic regression model was applied, including the variables that were significant at a p-value of <0.20 in the univariate analyses. Variables were eliminated one at a time from the model based on likelihood ratio tests. Variables were eligible for inclusion into the final model if they were significantly associated with in-hospital outcomes, with a p-value of <0.05.

The second step aimed to confirm the performance of the proposed group of items as predictors of risk of death. The population sample was randomly split into two groups, a derivation cohort (n = 353) and a validation cohort (n = 334). In the derivation cohort, a backward stepwise logistic regression procedure was performed as described previously, and used to develop a point-based risk scoring system; the number of points assigned to each risk factor was obtained by dividing each β coefficient by the smallest β coefficient significantly different from 0 and rounding to the nearest integer. A risk score was assigned to each participant in both the derivation and the validation cohorts by summing the number of points corresponding to each risk factor. Model discrimination was assessed using the c-statistic, which is analogous to the area under the receiver operating characteristic curve 25. In both the derivation and the validation cohorts, subjects were divided into three groups depending upon the tertile of the score, and mortality rates were compared between groups.

Patient characteristics

Table 1⇓ describes the demographic characteristics of the population and gives details of risk factors and medical history. The proportion of retired patients was 78.6%. In addition, 9.0% did not work any more due to disability. Among the whole population, 27.1% received home support and 8.4% lived in an institution. At least one emergency visit during the previous 6 months was reported by 68.2% and at least one hospitalisation by 37.6%. Initial clinical presentation, severity index, diagnostic procedures and treatments in the ED are detailed in table 2⇓. Wheezing was recorded in 69.3% and focal crackles in 28.9%. The vast majority (72.4%) of patients presented with at least one severity criterion.

Only 84 (10.5%) patients were discharged home from the ED, whereas 533 (67.1%) were hospitalised in a medical ward, 95 (11.9%) in an ICU and 62 (7.8%) in another hospital.

Compared to those hospitalised in a medical ward or an ICU, patients discharged home from the ED were younger and less often referred by a doctor; in addition, they showed fewer signs of severity and lower cardiac and respiratory frequencies (data not shown).

At discharge from hospital, the diagnosis of COPD exacerbation was considered as confirmed in 723 (91.6%) patients. Alternative diagnoses were another respiratory disease in 37 patients, a cardiovascular disease in 19 and another illness in six. When the final diagnosis was a COPD exacerbation, the two main reported aetiologies were bronchial infection in 68.1% and pneumonia in 14.5%. In 4.7% of cases, investigators reported that no aetiology was found.

Factors associated with in-hospital death, need for post-hospital support and length of stay

The in-hospital mortality rate was 7.4% (59 deaths). At the end of hospital stay, discharge to home without any particular support was possible in 472 (59.5%) patients, whereas 214 (26.9%) required home support or were transferred to convalescence or nursing homes. Since it was not planned to follow patients transferred to other hospitals, 49 (6.2%) patients were excluded from the analysis. Table 3⇓ shows factors associated with death and need for post-hospital support (at home or in convalescence or nursing homes) on univariate analysis, and table 4⇓ presents the results of multivariate logistic regression analysis when all clinical signs of severity (cyanosis, impaired neurological status, lower limb oedema, asterixis, use of inspiratory accessory muscles and expiratory use of abdominal muscles) were individually integrated into the model. Cardiac and respiratory frequencies did not differ between patients who died and survivors (98.3±23.9 versus 98.5±19.5 beats·min⁻¹ and 28.2±8.2 versus 26.7±7.1 breaths·min⁻¹, respectively). As mentioned in the Statistical analysis section, a three-grade clinical severity index (no severity criteria, one or two severity criteria, and at least thee severity criteria) was constructed on the basis of the number of clinical signs of severity at entry. Using this score, independent prognostic factors on multivariate analysis were age, clinical severity index at entry and baseline dyspnoea grade (table 5⇓). Determinants of the need for post-hospital support on multivariate analysis were sex and risk factors for death (tables 4⇓ and 5⇓).

The need for post-hospital support increased length of stay on univariate analysis (17.3±13.9 versus 10.5±7.1 days; p<0.0001), but this effect interacted with the clinical severity index, which was more powerfully associated with length of stay on multivariate analysis (data not shown).

Derivation and validation of the prediction score

In the derivation cohort, the variables independently associated with risk of death were the same as in the whole population (age, clinical severity index at entry and baseline dyspnoea grade; table 5⇑). The points assigned to each of these three predictors using β coefficients were as follows: 1) 0: age <70 yrs; 1: age ≥70 yrs; 2) 0: no severity signs; 2: 1–2 signs; 3: ≥3 signs; and 3) 0: baseline dyspnoea grades 0–1; 1: grades 2–3; 2: grades 4–5.

Table 6⇓ shows the risk of death in the derivation and validation cohorts dependent upon this score. The prediction characteristics are depicted using the c-statistic and sensitivity and specificity.

Mortality strongly increased with increasing prediction score in both the derivation and the validation cohort. The prediction score showed good discrimination for mortality, with a c-statistic of 0.79 in the derivation cohort and 0.83 in the validation cohort.

Patient characteristics

Despite the recruitment through EDs, a relatively high proportion (41%) of the patients were followed by a respiratory physician, and 70% received treatment for COPD. Although 70% reported undergoing lung function testing or peak expiratory flow rate measurement during past years, only very few of them were able to provide the results of such measurements. In line with these figures, a high proportion (85%) of subjects reported knowing that they had COPD, emphysema or chronic bronchitis. This is probably related to the mode of recruitment used in the present study; physicians were asked to include patients with increased dyspnoea (this criterion was indeed satisfied in almost all of the patients) and a diagnosis of certain or probable COPD exacerbation, with no other dominant respiratory disease or cause of acute respiratory failure (e.g. left heart failure or pulmonary embolism). Thus they probably tended to include mainly patients with known COPD, omitting most of the unknown proportion of patients in whom the disease is discovered when the first exacerbation occurs. This would also explain the relatively high level of baseline dyspnoea in most patients and their frequent recourse to healthcare resources in the months preceding the study.

Despite this, the results of lung function tests were known for only a few of the patients, which suggests the need for improving the education of both doctors and patients. Such information could be of interest for the evaluation of the prognosis of an exacerbation and in the decision as to whether and where admission should be considered, but, at present, lung function test results are too infrequently available to be included in a prediction rule, at least in the setting of French hospitals.

Similarly, peak expiratory flow rate was seldom measured, which makes it difficult to interpret its lack of predictive potential. Finally, the conditions of measurement of arterial blood gas levels (i.e. with or without supplemental oxygen) were too heterogeneous to permit any conclusion as to their prognostic value in the real-life ED setting.

Risk factors for in-hospital death

Although all guidelines list clinical and biological signs of severity that should lead to hospitalisation or admission to the ICU 26–28, only a few studies have addressed this important issue. These studies were reviewed in 2001 in an evidence-based process conducted jointly by the American College of Physicians and the American Society of Internal Medicine 29. At that time, only 11 studies had been conducted to identify predictors of in-patient mortality, and eight had identified such factors in populations ranging 74–3,050 subjects, recruited with markedly variable inclusion criteria. Of these studies, only five performed multivariate analysis and four were performed in ICU patients alone. The internal validity scores were fairly good but external validity and degree of validation were poor. In the five largest studies with multivariate analysis (n = 322–3,050), the independent predictors of in-hospital death were age, acute physiology score, body mass index, functional status before the exacerbation, P_a,O2/inspiratory oxygen fraction ratio, need for mechanical ventilation, serum albumin and sodium levels, and cardiac comorbid conditions. None of these studies provides physicians with a simple prediction rule to aid in the decision as to where a given patient should be managed. In the present study, it was found that three simple clinical criteria were potent predictors of mortality, suggesting that they should be systematically assessed at entry in COPD exacerbations and used to decide where patients should be oriented. These criteria were age, number of clinical signs of severity at entry and baseline dyspnoea grade. Among individual clinical signs of severity, neurological impairment and use of inspiratory accessory muscles were the only factors independently associated with an increased risk of death; however, adding other signs of severity through the aggregated clinical severity index increased the predictive value of the model.

One strength of the present study is that, following identification of predictors of death in the whole study population, two subgroups were randomly constituted to construct a prediction score and assess its validity. As mentioned in the Results section, this score was very easy to calculate, and, when the derivation and validation populations were classified according to score tertile, marked differences in mortality were observed between the two subgroups. In addition, prediction characteristics depicted by the c-statistic, sensitivity and specificity were satisfactory, especially considering the simplicity of the score and particularly the lack of requirement for nonclinical data. Indeed, the c-statistic was higher than that reported by Celli et al. 30 for the BODE (body mass index, airflow obstruction, dyspnoea and exercise capacity index).

Whether or not such a prediction score might improve patient outcome remains to be determined in a prospective validation study.

Predictors of need for post-hospital support

Although the need for post-hospital support is clearly an increasingly important clinical and economic issue, the present authors are not aware of any previous studies aimed at identifying the corresponding clinical predictors.

In the present study, factors associated with an increased risk of death were also predictors of the need for post-hospital support, at home or in intermediate care facilities or institutions. This finding is not surprising; if they survive, the most severe patients show an increased level of dependence when they are discharged. Identification of these factors may be useful in helping to reduce length of stay, since patients who need post-hospital support as a consequence of increased severity stay longer, which probably relates to two different causes; first, the time taken to recover a health status compatible with discharge may be longer; and, secondly, hospitalisation may be prolonged by difficulties in organising adequate post-hospital support. Such difficulties are likely to become increasingly frequent, considering the increasing age of the population 16. This will have not only clinical consequences through the increased hospital-related morbidity, quality-of-life impairment and mortality, but also economic consequences since hospital stays account for approximately half of all COPD-related healthcare expenses, with COPD representing the second leading respiratory cause of direct and indirect healthcare costs in the USA and 3.5% of all healthcare expenses in France 8, 11.

Interestingly, the need for post-hospital support was not independently related to any kind of pre-hospital support (home support or residence in a nursing home); the most likely explanation of this is that, in most cases, pre-hospital support was also determined by dyspnoea severity (data not shown).

Limitations of the present study

As mentioned previously, one main limitation was the lack of lung function and relevant arterial blood gas data, making it impossible to include them in the prediction rule. However, this has to be considered as corresponding to what occurs in the real-life setting; thus it does not decrease the value of the identified prognostic factors.

Another limitation is the lack of post-hospital mortality data, which would be mandatory for: 1) validation of the prognostic factors found in the present study and their transformation into admission/discharge criteria; and 2) assessment of the validity of decisions of immediate discharge from the ED. In order to achieve these goals, a large ongoing prospective study includes detailed follow-up at 3 months and assessment of vital status at 3 yrs 31.

Recruitment bias also has to be considered. Indeed, the mean number of patients recruited per centre per month was quite small (approximately three), far lower than that expected during the winter period. As mentioned previously, recruitment was obviously limited to patients with a fairly certain diagnosis of COPD exacerbation at entry. In addition, most centres were active only during short periods of time (1–4 weeks), and local coordinators checked that cases obviously corresponding to inclusion criteria were not omitted, suggesting that the recruitment was not far from exhaustive during active inclusion periods. Limiting the recruitment period to one winter season ensured that, if some kind of seasonal bias might influence predictors of outcomes, it was homogenously distributed throughout the study period and centres. In addition, since such a bias exists, it corresponds to what happens in real life, and does not compromise the applicability of the results to the period when COPD exacerbations are the greatest cause of concern for clinicians and hospitals.

Finally, the present findings do not apply to COPD cases discovered at the occasion of their first exacerbation since the disease had already been diagnosed in most patients. They also do not apply to patients who require immediate ventilatory support and/or ICU admission when presenting to the ED; inclusion was indeed limited to patients who did not require immediate ventilatory support, as decided by the attending emergency physician. The purpose of this restriction was to limit the heterogeneity of the population and to avoid biasing the results by including a subgroup of patients with very severe exacerbation, which could have induced the identification of factors that might not be applicable to exacerbations of lesser severity.

Conclusions

The present study provides simple prognostic factors for use in patients with chronic obstructive pulmonary disease exacerbations presenting to the emergency department. The corresponding variables are exclusively clinical and very easy to measure at entry in the real-life context, and represent powerful predictors of the risk of both death and need for post-hospital support. Thus, following prospective validation, they may be useful for clinicians in helping them with: 1) admission and orientation decisions; 2) anticipation of the need for intermediate care facilities or home support; and 3) informing patients and relatives on exacerbation-related short-term risks.