CLINICAL TRIAL

One hundred and eighteen patients were allocated to standard treatment and 118 to NIV. The characteristics of both groups were similar (table 1). Thirty two of the 118 patients (27.1%) in the standard treatment group met the primary end point “need for intubation” compared with 18 of the 118 (15.3%) in the NIV group (p<0.02). In-hospital mortality was also reduced: 24/118 (20.3%) of the standard group died compared with 12/118 in the NIV group (10.2%) (p=0.046).7

FACTORS ASSOCIATED WITH THE FAILURE OF TREATMENT

At enrolment only three variables were associated with subsequent failure of treatment on univariate analysis (table 2): severity of acidosis (p<0.001), degree of hypercapnia (p<0.001), and severity of hypoxia (p=0.035). Allocation to NIV was protective (p=0.038). On multivariate analysis only three variables remained: H⁺concentration (OR 1.22 per nmol/l, 95% CI 1.09 to 1.37, p<0.01), Paco ₂ (OR 1.14 per kPa, 95% CI 1.14 to 1.81, p<0.01), and allocation to NIV (OR 0.39, 95% CI 0.19 to 0.80, p<0.01). However, this model had poor discriminant value, only predicting 82% of the outcomes correctly with a sensitivity of 22% and a specificity of 97%. The predictive value of failure was 67% and of success was 83%. Despite this, these three variables are useful for stratifying patients according to risk (table 3). Compared with a patient with a pH of 7.35 and a Paco ₂ of 6 kPa, a patient with a pH of 7.25 and a Paco ₂of 12 kPa has a 22-fold higher risk of meeting criteria for intubation. It can also be seen that the degree of acidosis is a greater risk factor clinically than the extent of hypercapnia.

After 4 hours of treatment two additional variables were associated with treatment success on univariate analysis (table 4): fall in respiratory rate (p=0.009) and an improvement in H⁺concentration (p=0.035). All six univariate variables were entered into the multivariate analysis of which four remained. At 4 hours the factors associated with a worse outcome were severity of acidosis (OR 1.23 per nmol/l, 95% CI 1.05 to 1.43, p<0.01) and hypercapnia (OR 1.77 per kPa, 95% CI 1.28 to 2.45, p<0.01) at enrolment. Fall in respiratory rate between admission and 4 hours was associated with success (OR 0.92 per breaths/min, 95% CI 0.84 to 0.99, p=0.04) as was fall in H⁺ ion concentration over the same time period (OR 0.89 per nmol/l, 95% CI 0.82 to 0.97, p<0.01). Allocation to NIV was no longer a statistically significant variable. However, this model also lacked discriminant ability, only predicting 85% of the outcomes correctly with a sensitivity of 30% and a specificity of 97%. The predictive value of failure was 69% and of success was 87%.

Despite this, the changes in H⁺ concentration and respiratory rate can be used to assess risk in a clinically meaningful fashion (table 5). An improvement in pH over the first 4 hours from 7.30 to 7.35 is associated with a 47% reduction in the risk of failure. Similarly, a fall in respiratory rate of 8 breaths/min is associated with a 49% reduction in risk.

LONG TERM SURVIVAL

Follow up was achieved for all 236 patients to either death or 1 January 1999. The median survival for all patients enrolled into the study was 15.9 months (95% CI 12.6 to 19.2) with 58.4% surviving at 1 year. There was a trend to improved survival in the group receiving NIV (fig 1). The 1 year survival in the NIV group was 61.6% compared with 53.9% in the standard treatment group. The median survival rates were 16.8 and 13.4 months, respectively. However, these differences were not statistically significant (p=0.12). The curves suggest that the decline in survival was parallel after the first 3 months. Indeed, at 1 year 78.2% of the NIV patients who were alive at 3 months remained alive, which is very similar to the 75.7% of the standard treatment group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Kaplan-Meier plot of survival from enrolment. Median survival 13.4 months in standard treatment group v 16.8 months in NIV group (p=0.12).

LONG TERM SURVIVAL

The patients enrolled in the randomised controlled trial had a 1 year survival of 58%. This is comparable to other long term studies of patients with COPD admitted with respiratory failure where 1 year survival has ranged from 57% to 88% in series where only a proportion of patients required ventilatory support,9-11 and from 34% to 56% in series where all patients receive ventilatory support.12-16 This suggests that our survival rates are typical. However, it is difficult to make accurate comparisons between these studies because of variations in study population (especially co-morbidity), the interventions received during the exacerbation (for example, invasive mechanical ventilation), and subsequent outpatient care.

There was a non-significant trend to improved survival in the NIV group compared with the standard treatment group with 1 year survival rates of 62% and 58%, respectively, despite the fact that the study design was biased against such a finding. The control element of the study was lost at the point of discharge. No control was maintained over outpatient care or the management of future exacerbations. The latter is important as some patients would have received NIV in subsequent exacerbations and others—for example, in centres where NIV was only available in trial—would not have received NIV due to prior enrolment. Hence, the comparison of long term survival was biased against showing a survival advantage.

Three other controlled but not randomised trials17-19assessing the long term survival of patients with COPD treated with NIV have reported similar findings with 1 year survival rates of 64% (95% CI 49 to 79, n=39),18 71% (95% CI 53 to 89, n=24),17 and 87% (95% CI 69 to 100, n=15).19 The equivalent survival rates in the standard treatment groups were less at 30% (95% CI 12 to 47, n=27),18 50% (95% CI 30 to 70, n=24),17 and 53% (95% CI 28 to 79, n=15).19 However, the survival advantage in our study and that of Confalonieri et al 17 appeared to be related to the reduction in in-hospital mortality during the index admission rather than a long term protective effect as the decline in survival after 3 months was identical in the two groups.

The data therefore suggest that the benefit of NIV is in improving survival during the exacerbation itself, and that those in whom in-hospital death was avoided do not represent a high risk group. If they had a very poor prognosis, the survival curves would have converged as these patients died early in the follow up period. Moreover, the median survival of 15.9 months is not sufficiently poor to render NIV inappropriate, at least in terms of survival.

FACTORS ASSOCIATED WITH THE FAILURE OF TREATMENT

On admission patients with the worst degree of acidosis and the highest level of Paco ₂ were the most likely to meet criteria for intubation (table 3). Of these two variables, acidosis was the most important clinically. The risk of needing intubation was reduced by NIV. After 4 hours improvements in respiratory rate and in acidosis were associated with success (table5), but had to be interpreted in the light of the initial pH and Paco _2. However, at neither enrolment nor in the first 4 hours was it possible accurately to predict on an individual basis who will or will not meet the criteria for intubation. However, these variables can be used as a clinical guide.

The findings of this study are consistent with previously published studies. A low pH has been shown to be associated with increased mortality20 ,21 and the need for intubation22in cohorts of patients admitted to hospital with COPD. Moreover, in controlled and uncontrolled trials of NIV a low initial pH3 ,23 and a high Paco ₂ 3 ,23 ,24 have been associated with a poor outcome. A fall in Paco ₂ 23 ,24 and improvement in pH4 ,23-25 have also been shown to be protective.

A change in respiratory rate has not previously been reported as a prognostic factor. The fall in respiratory rate is probably explained by an increase in tidal volume which allows the respiratory rate to fall, thus protecting against fatigue but allowing minute ventilation to be maintained. The improvement in tidal volume may occur due to medical treatment and/or NIV. The absence of “allocation to NIV” as a prognostic variable at 4 hours is initially a little surprising. However, NIV has been shown in randomised controlled trials to improve acidosis and reduce respiratory rate.1 ,4 ,7 These variables therefore reflect not only allocation to NIV, but also response to NIV and, as such, the NIV variable is lost from the logistic regression analysis.

It is also valuable to consider variables that were not significant in the analysis, notably hypoxia and radiographic consolidation. Hypoxia on admission in acute exacerbations has not been shown to be associated with mortality or the need for intubation and, as such, the findings in this study are not surprising.3 ,4 ,20-25Radiographic consolidation has been shown to be a poor prognostic factor.18 ,23 ,26 However, Rieves et al 26 also studied 19 patients with less severe disease and could not find an association with consolidation. They suggested an interplay between disease severity and consolidation. All three studies included patients who were more seriously ill than those in our study. The explanation suggested by Rieveset al 26 may explain the lack of association in our study finding. Alternatively, these studies were small (n=59, 33, and 57, respectively) and the finding may be a statistical anomaly which does not appear in our larger study (n=236).

The lack of discrimination in the prognostic models reflects the methodology used. This was a post hoc analysis and potentially important variables were therefore not considered. For example, Ambrosino et al 23 found that a low body mass index, a high Apache II score, tachycardia, decreased consciousness, and poor compliance with NIV were associated with failure. These factors were not included in the above analyses because they were not available or, in the case of consciousness, those with a low consciousness were excluded from the trial. The lack of these factors may explain the failure of the models to discriminate in a clinically useful fashion. Notably, health status and lung function before admission were not included. However, in the clinical situation this information is often unavailable within the first 4 hours of an admission.

In conclusion, the initial pH and Paco ₂ can be used clinically to stratify patients according to their risk of meeting criteria for intubation. NIV reduces this risk and, once started, the effectiveness of NIV should be monitored using respiratory rate and improvement in acidosis. Changes in Paco ₂ and Pao ₂ are not associated with either a positive or an adverse outcome and should not be the focus of monitoring. Unfortunately, these variables are not sufficiently predictive to be able accurately to discriminate individuals who will fail and require invasive mechanical ventilation despite NIV. Further research is required to develop a predictive model as well as to clarify the importance of radiographic consolidation and of a fall in respiratory rate. With regard to long term survival, NIV appears to be associated with a long term survival advantage but this is achieved by avoiding death during the index admission. Those in whom death is avoided seem to have a similar long term prognosis to the whole group, and the median survival is sufficiently good to support the use of NIV in acute exacerbations of COPD.