Antibiotic Therapy in Exacerbations of Chronic Obstructive Pulmonary

Disease
N. R. ANTHONISEN, M.D.; J. MANFREDA, M.D.; C. P. W. WARREN, M.D.; E. S. HERSHFIELD, M.D.;
G. K. M. HARDING, M.D.; and N. A. NELSON, Ph.D.; Winnipeg, Manitoba, Canada

The effects of broad-spectrum antibiotic and placebo Tager and Speizer (5) reviewed the problem in 1975 and
therapy in patients with chronic obstructive pulmonary concluded that no good evidence existed that antibiotic
disease in exacerbation were compared in a randomized,
therapy was of either short-term or long-term benefit,
double-blinded, crossover trial. Exacerbations were
defined in terms of increased dyspnea, sputum and they suggested that this form of treatment be reas-
production, and sputum purulence. Exacerbations were sessed. Since 1975, the problem has been little investigat-
followed at 3-day intervals by home visits, and those that ed; one recent report has found antibiotic therapy to be of
resolved in 2 1 days were designated treatment no benefit in patients hospitalized with exacerbations ( 6 ) .
successes. Treatment failures included exacerbations in
which symptoms did not resolve but no intervention was Because chronic obstructive pulmonary disease is a
necessary, and those in which the patient's condition common disease, and patients are reported to have an
deteriorated so that intervention was necessary. Over 3.5 average of one to four exacerbations per year, antibiotic
years in 173 patients, 3 6 2 exacerbations were treated, therapy of these exacerbations is a frequently used
180 with placebo and 182 with antibiotic. The success
rate with placebo was 5 5 % and with antibiotic 6 8 % . The
unproven treatment. For this reason, we conducted a tri-
rate of failure with deterioration was 1 9 % with placebo al of antibiotic therapy for such exacerbations.
and 1 0 % with antibiotic. There was a significant benefit
associated with antibiotic. Peak flow recovered more Materials and Methods
rapidly with antibiotic treatment than with placebo. Side PATIENTS
effects were uncommon and did not differ between
antibiotic and placebo.
We recruited patients with stable chronic obstructive pulmo-
nary disease, characterized them in terms of symptoms and
lung function, and followed them. When an exacerbation devel-
EXACERBATIONS of chronic obstructive pulmonary dis- oped, the patient was given, in a double-blinded manner, a 10-
ease are commonly characterized by increases in dysp- day course of either placebo or antibiotic according to a prear-
ranged random schedule, and the exacerbation was followed
nea, cough, and sputum production with increased closely for 3 weeks. Subsequently, the patient continued to be
purulence in sputum. It is recommended that these exac- followed, and further exacerbations were treated in a crossover
erbations be treated with a 7- to 10-day course of broad- manner alternating placebo and antibiotic. Patients were seen in
spectrum antibiotics (1-4), though the role of bacterial the clinic at least every 3 months and fully reevaluated at yearly
intervals.
infection in exacerbations is recognized as unclear. Sur- Patients were eligible for the trial if they were at least 35
prisingly, the efficacy of the therapy is also not clear. years old and had a clinical diagnosis of chronic obstructive
disease, not asthma. In addition, they had to live close enough
• From the Sections of Respiratory and Infectio.us Diseases, Department of Inter-
nal Medicine, and the Department of Social and Preventive Medicine, University to the clinical center for home visits and to be reliable (that is,
of Manitoba; Winnipeg, Manitoba, Canada. to keep two consecutive outpatient appointments). They were
196 Annals of Internal Medicine. 1987;106:196-204. © 1987 American College of Physicians

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 required to have a forced expired volume in 1 second (FEVj) tion of exacerbations was done prospectively and for analytical
that was less than 70% of the predicted value and less than purposes only, as management was the same for all exacerba-
70% of the forced vital capacity. The total lung capacity was tions.
required to be greater than 80% of the predicted value. Patients Patients who had an increase in symptoms notified the center
were excluded if the FEVj increased to 80% of the predicted and were seen on the same day by an experienced nurse-practi-
value with use of an inhaled bronchodilator product. Patients tioner, who decided whether the symptoms fulfilled the criteria
were also excluded if they had other disease serious enough to outlined earlier. If the patient was judged to be dangerously ill,
influence their quality of life or clinical course, such as cancer, the patient's physician was notified and participated in the deci-
left ventricular failure, or stroke, or if they had other disease sion as to whether the patient would begin the study protocol. If
likely to require antibiotic therapy, such as recurrent sinusitis the patient was thought to have an exacerbation, the nurse-
or urinary tract infection. practitioner checked symptoms, measured peak flow, and pre-
STUDY PROTOCOL AND TESTING
scribed a ten-day course of either antibiotic or placebo at the
initial visit. All exacerbations were treated in the same way.
When suitable patients were identified, they were placed on a During all exacerbations, repeat visits were made at 3-day inter-
standard regimen (see below) and followed for 2 weeks. At the vals until day 15, and if symptoms had not resolved by then,
end of that time, if they still fulfilled the entry criteria, they had another visit was made at day 21. On each occasion, symptoms
baseline studies. These studies consisted of a symptom history and peak flow were checked, the patient questioned about
in the form of a modified American Thoracic Society question- symptoms associated with the medication, and the remaining
naire ( 7 ) , an examination of the heart and lungs, a chest radio- antibiotic or placebo pills counted. Peak flow was measured
graph and an electrocardiogram if one was not available from three times at each visit, and the largest value recorded.
less than 1 year previously, and lung function tests. Forced vital
capacity and FEVj were measured with a rolling seal spirome- The antibiotics and regimens used were trimethoprim-
ter. Of three efforts, the one with the largest sum of FEVj and sulfamethoxazole, one tablet (160 mg/800 mg) twice daily;
forced vital capacity was recorded and compared with normal amoxicillin, 0.25 g four times daily; or doxycycline, 200 mg
values ( 8 ) . Measurements were repeated after the inhalation of initially followed by 0.1 g daily. The choice of antibiotic was
250 u.g of isoproterenol. Peak flow was measured with a Wright made by the patient's physician, since these agents appear not
peak flow meter (Clement Clark International, London, En- to differ in effectiveness in this situation (12-15). We were
gland), with the best of three efforts being recorded. Functional supplied an appropriate placebo for each drug, so that neither
residual capacity was measured by plethysmography and com- the patient nor the medical staff knew which medication was
pared with normal values ( 9 ) . Inspiratory capacity and expira- active.
tory reserve volume were measured by spirometry, allowing The outcome of the exacerbations was defined in terms of
computation of total lung capacity and residual volume that symptoms. A "treatment success" was defined as resolution
were also compared with normal values ( 9 ) . Arterial blood within 21 days of all symptoms that accompanied the exacerba-
samples were obtained while the patients were semirecumbent tion. The date of resolution was noted. Treatment failures were
at rest and were analyzed for (P02) and (PCO2) with appropri- of two types: "no resolution" included exacerbations in which
ately calibrated electrodes. all symptoms did not resolve in 21 days but no further interven-
tion was deemed necessary; in "failure with deterioration,"
When not having an exacerbation, patients were seen at the
symptoms worsened during the exacerbation to the point that
clinical center at 3-month intervals. At these visits, symptoms
further intervention (hospitalization or treatment with antibiot-
were evaluated, the heart and lungs reexamined, and FEVj,
ic in unblinded fashion) was deemed necessary. These interven-
forced vital capacity, and peak flow measured. At yearly inter-
tions could only be instituted after consultation with a physi-
vals, all baseline studies were repeated. Patients were managed
between exacerbations with a standard regimen. All patients cian, usually the patient's own, and only after more than 72
received inhaled albuterol supplied by metered-dose inhalers or hours of antibiotic or placebo treatment had been given. If in-
compressor nebulizers, or both. All patients received oral theo- tervention was thought necessary less than 72 hours after treat-
phylline, usually in the long-acting form and usually in stan- ment was begun, the treatment was not designated a success or
dard doses. Some patients who had serious symptoms and who failure because we believed that such rapid deterioration was
were thought to be steroid responsive (10) were maintained on unlikely to be related to antibiotic or placebo therapy.
prednisone, 5 to 10 mg/d. Diuretics were used to manage ede- After an exacerbation, patients were required to have a 2-
ma, and patients who fulfilled standard criteria for home oxy- week period with the same symptoms as those present before
gen therapy received such therapy for at least 18 h/d (11). All the exacerbation before another exacerbation was studied. If
patients received polyvalent anti-influenza vaccine each au- patients were hospitalized, new baseline measurements were
tumn. Patients thought to be steroid responsive received in- made before another exacerbation was studied. At their quar-
creased doses of these agents during exacerbations. The regimen terly clinic visits, patients were questioned closely about
employed usually consisted of 40 mg of prednisone for 3 days, episodes of illness that were not studied; if these had the charac-
with the dosage being tapered to the maintenance dose in 9 to teristics of an exacerbation, they were classified as a "missed"
12 days. Although the dose schedule varied somewhat among exacerbation.
patients, it was the same during all exacerbations in any given
ANALYSIS OF DATA
patient.
In 61 exacerbations, the symptom questionnaire administered
P R O T O C O L D U R I N G EXACERBATIONS by the nurse-practitioners was checked for accuracy and repeat-
Exacerbations were defined in terms of symptoms. The oc- ability. This was done by the patient's physician, who tele-
currence of increased dyspnea, sputum volume, and sputum pu- phoned the patient and administered the same questionnaire
rulence was defined as a type-1 exacerbation. Type-2 exacerba- within 24 hours of the nurse-practitioner's visit. Calls were
tions were defined as occurring when two of these three made after the first visit and again at the end of the observation
symptoms were present. Type-3 exacerbations were defined as period—days 15 or 21. The physicians were not informed of the
occurring when one of the three symptoms was present in addi- patient's progress.
tion to at least one of the following findings: upper respiratory All data were recorded on special forms, checked for accura-
infection (sore throat, nasal discharge) within the past 5 days; cy, and entered into a computer file. The SAS statistical pack-
fever without other cause; increased wheezing; increased cough; age (SAS Institute Inc., Cary, North Carolina) was used in
or increase in respiratory rate or heart rate by 20% as com- analysis. For discrete variables such as treatment success or
pared with baseline. If during the course of an exacerbation new failure, the chi-square test statistic was used when only the first
symptoms appeared, reclassification was done; that is, an exac- exacerbation for each patient was analyzed. When exacerbation
erbation that was initially type 3 could be reclassified as type 1 outcomes were analyzed in patients with multiple exacerba-
or 2, but not the reverse. It should be noted that the classifica- tions, and when comparisons were made within patients, the

Anthonisen eta/. • Antibiotic Therapy in COPD 197

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 Table 1 . Symptoms at Baseline in Patients with Chronic Obstruc- ing the first two to four exacerbations. Again, a pair-type analy-
tive Pulmonary Disease sis was used, the general linear model for repeated observations
(19).
Patients Peak flow results were also analyzed by a paired-within-
patient format. We examined peak flow data from 64 patients
% who had at least one antibiotic-treated and one placebo-treated
Cough exacerbation followed for 15 to 21 days (this excluded failures
None 15.0 with deterioration). If a patient had more than one exacer-
Morning only 60.1 bation with a 15- to 21-day follow-up and the same treatment,
Episodes during day 20.8 peak flow values were averaged. Effects of treatment on peak
Nearly continuous 4.0 flow were examined with a repeated-measures analysis of vari-
Sputum quantity ance using peak flow as the dependent variable and days of
None 16.2 testing (0, 3, 6, 12, or 15) and treatment as the independent
Less than 30 mL 57.2 variables.
30-100 mL 22.5 Agreement between nurse-practitioners and physicians with
More than 100 mL 4.0 regard to symptoms was analyzed by the computation of kappa
Sputum quality scores, which correct for concordance due to chance (19, 20).
Mucoid 62.4
Mucopurulent 18.5 Results
Purulent 2.9 Between 1 N o v e m b e r 1981 and September 1984, 173
Dyspnea patients were enrolled in the study. T h e group was 7 9 . 8 %
None 6.3
Only on unusual exertion 28.3 male, and the average age was 67.3 ± 9.0 ( S D ) years.
Present during normal activity 48.4 Nearly all ( 9 3 . 6 % ) had a history of s m o k i n g , averaging
Present at rest 16.7 39.9 ± 28.9 pack-years, although only 2 1 . 4 % were
Wheezing s m o k i n g at the time of enrollment. Other baseline charac-
None 30.0
teristics are s h o w n in Tables 1 and 2. A t baseline, most
Present only on unusual activity or
exacerbations 52.6 patients had morning c o u g h with scanty m u c o i d s p u t u m ,
Present on many daily activities 15.6 developed dyspnea with mild exertions, but w h e e z e d only
Present at rest 1.7 with severer exertion ( T a b l e 1). Lung function studies
( T a b l e 2 ) s h o w e d that the average patient had moderate-
ly severe airways obstruction with a poor bronchodilator
likelihood-ratio chi-square statistic was used. We used the Mest
and repeated-measures analyses of variance to assess continuous response and hyperinflation but near-normal arterial
variables such as duration of exacerbations. blood gases. T h e standard deviations for these average
In the analyses of success and failure rates, several ap- values (Table 2 ) are relatively large, however, indicating
proaches were used. The simplest and easiest to understand was that a wide range of lung function was represented. Seven
the comparison of results in all antibiotic-treated and placebo-
patients were on h o m e o x y g e n therapy for chronic respi-
treated exacerbations. However, many patients had exacerba-
tions treated in each way, and consequently exacerbation out- ratory failure.
comes may not have been independent of each other because of M o r e than 4 0 0 0 patient-months were observed during
differences among patients and similarities (between exacerba- the study; the average patient was followed for 23.7 ±
tions) within patients.
11.3 m o n t h s . O f the 173 patients, 59 dropped out before
We therefore analyzed only the first exacerbation for each
patient, using independent treatment groups that were, howev- the study w a s ended on 1 M a r c h 1985. O f these, 18 died,
er, relatively small. Also, in patients who had at least two exac- although only 1 death occurred during an exacerbation
erbations, we analyzed exacerbations as matched sets, because m a n a g e d by our protocol and this was due to a ruptured
each of these patients had at least one exacerbation treated with aortic aneurysm. Fifteen patients were dropped because
antibiotic and another treated with placebo. The first two to
four exacerbations in each patient were used for this analysis; they did not cooperate with the study protocol, and an-
the fifth and subsequent ones were excluded to prevent the re- other 17 patients were dropped because they developed
sult from being biased by the few patients who had more than other disease or because their physicians believed that the
four exacerbations. The likelihood chi-square test statistics were study was not in their best interest (usually because they
generated for this group. To describe our experience more com-
pletely, we pooled the results from these patients who had mul- were "too sick" for the s t u d y ) . T h e patients w h o dropped
tiple exacerbation with those results from patients who had out did not differ significantly from the remainder of the
only one exacerbation. This pooling was permissible because the
Table 2. Lung Function at Baseline in Patients with Chronic Ob-
two data sets were independent, and the analysis consisted of structive Pulmonary Disease'*
calculating an average value for the two data sets after weight-
ing each of them according to the variability of the estimates. Forced expiratory volume in 1 second,
Odds ratios and one-tailed confidence limits for these results % predicted ' 33.9 ±13.7
were calculated in the usual way (16-18).
Forced vital capacity, % predicted 59.5 ±16.8
Sets with discordant outcomes—that is, sets containing exac- Functional residual capacity, % predicted 164.6 ± 34.4
erbations with both successful and unsuccessful outcomes— Total lung capacity, % predicted 128.9 ±19.7
were examined by using outcome (success or failure) as the Residual volume, % predicted 205.3 ±51.5
dependent variable. Independent variables were exacerbation Bronchodilator response, % FEVj 111.8 ±17.6
order (first or second), exacerbation type (1, 2, or 3) and treat- Peak flow, L/min 227.5 ± 96.1
ment (antibiotic or placebo). Arterial Po 2 , mm Hg 68.3 ± 9.9
The same matched-set model was used to examine the associ- Arterial Pco 2 , mm Hg 36.6 ± 6.1
ation between presenting symptoms (present or absent) at the Arterial pH 7.42 ± 0.03
onset of exacerbations and the treatment. In patients who had
more than one exacerbation, duration was analyzed by examin- * Data given as mean ± SD. F E V | = forced expiratory volume in 1 second.

198 February 1987 • Annals of Internal Medicine • Volume 106 • Number 2

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 used in 42.8% of antibiotic-treated exacerbations and in
43.5% of placebo-treated ones.
The success and failure rates in 180 placebo-treated
and 182 antibiotic-treated exacerbations are shown in the
top of Table 4. A few episodes were not classified as treat-
ment success or failure, because the patients' condition
deteriorated less than 72 hours after treatment began or
they developed another illness unrelated to the exacerba-
tion. The success rate in antibiotic-treated exacerbations
was about 1.24 times higher than the success rate in pla-
cebo-treated episodes. Failures with deterioration (those
in which the patient's symptoms worsened so that further
intervention was thought necessary) were nearly twice as
common with placebo treatment.
When matched sets of exacerbations and single exacer-
bations were pooled, the relative odds favoring success
with antibiotic treatment were 2.32 times the odds for
placebo, with a lower 95% confidence limit of 1.38. In
these data, the success rate with antibiotics was 18.9%
greater than that with placebo, with a lower 95% confi-
Figure 1 . Cumulative distribution of exacerbations frequency in pa- dence limit of 7.8%. Similarly, when odds ratios obtained
tients with chronic obstructive pulmonary disease followed at least
6 months. The ordinate shows the exacerbation frequency, and the from matched sets and single exacerbations were pooled,
abscissa, the percentage of patients (PTS). deterioration was 0.467 times as common in antibiotic-
treated exacerbations as in placebo-treated ones, with an
patients in symptoms (Table 1) or lung function (Table upper 95% confidence limit of 0.88. In these data there
2). was a 9.1 % difference in the frequency of deterioration,
The 173 patients had 448 exacerbations while under with a lower 95% confidence limit of 1.9%. By matched-
observation. Of these, 86 were "missed," that is, not set analysis, exacerbation outcome was not significantly
treated according to protocol. In 35 cases the patient was related to order of treatment or exacerbation type.
judged by the study team to be "too sick" for the proto- We also analyzed the first exacerbation recorded in
col; in 21 the patient or an outside physician began anti- each patient, 59 of which were treated with placebo and
biotic therapy before notifying the study team; and in 16 57 with antibiotic. The results (Table 4 Bottom) were
the exacerbation occurred when the patient was out of similar to these when all exacerbations were analyzed to-
town. The other 14 exacerbations were missed for various gether. The difference in success rates just missed being
reasons. When all patients and all exacerbations were statistically significant (p = 0.06, chi square) in this rel-
considered, the average patient had 1 exacerbation each atively small series.
9.2 months. This statistic is somewhat misleading, how- Because both wheeze (Table 3) and success (Table 4)
ever, because 10 patients were followed for less than 6 were associated with antibiotic therapy, their relationship
months. When these patients were excluded, the average with each other was examined. Wheeze was not associat-
was 1 exacerbation each 8.5 months, or an average fre-
Table 3. Findings at Onset of Exacerbations of Chronic Obstruc-
quency of 1.3 ± 1.5/yr. A cumulative distribution of tive Pulmonary Disease*
exacerbation frequency in patients followed at least 6
months is shown in Figure 1. About 25% of the patients Placebo Antibiotic
had no exacerbations; the median for the whole group Group Group
was 0.8 exacerbations per year; and a small number of (77=180) (/7=182)
patients had many exacerbations (10% had more than Exacerbation type, %
3.4/yr, and 1 averaged 9.6/yr). Type 1 39.7 38.5
Type 2 41.9 42.3
Symptoms present at the onset of the 362 exacerba-
Type 3 18.4 19.2
tions treated according to protocol are shown in Table 3. Individual findings, %
Increased dyspnea, cough, sputum, sputum purulence, Recent upper respiratory tract infection 54.7 48.9
and wheeze were common, whereas fever and tachycar- Increased dyspnea 87.2 92.3
dia were relatively uncommon. Approximately 40% of Increased sputum production 71.7 67.6
Change in sputum color 60.6 58.8
the exacerbations were type 1 at onset, and another 40% Increased wheeze 67.2 79.1
were type 2. The distribution of exacerbation types was Increased cough 82.2 82.4
essentially the same in antibiotic- and placebo-treated ex- Fever 29.4 29.1
acerbations. The frequency of individual symptoms was Increased heart rate 21.6 21.7
not significantly different between antibiotic- and Increased respiratory rate 44.1 40.7
placebo-treated exacerbations except for wheeze, which * Increased wheeze was commoner (p = 0.01) at the onset of antibiotic-treated
occurred more often (p < 0.01) at the onset of antibiot- exacerbations when results from matched-set and single exacerbations were
pooled. There was no significant difference (p > 0.10) for any other finding
ic-treated exacerbations. Systemic steroid thereapy was shown.

Anthonisen eta/. • Antibiotic Therapy in COPD 199

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 Table 4. Outcome of Exacerbations of Chronic Obstructive Pul- exacerbations deterioration was over twice as common
monary Disease
with placebo as with antibiotic. In type-2 exacerbations,
Placebo Antibiotic antibiotic success rates were somewhat greater and dete-
Group Group rioration somewhat less common than with placebo, but
differences between treatments were less striking than in
%(n) %(n) type-1 exacerbations and deterioration was much less
All exacerbations common whatever the treatment. By both classification
Success 55.0(99) 68.1(124)* systems, type-3 exacerbations showed essentially the
N o resolution 23.3(42) 18.7 (34)
Deterioration 18.9(34) 9.9 ( 1 8 ) f same rates of success and deterioration for antibiotic and
Othert 2.9 (5) 3.2 (6) placebo treatment. Of exacerbations classified as type 3 at
First exacerbations onset, 20 of 33 treated with placebo were reclassified due
Success 52.3(31) 66.8 (38)§ to an increase in symptoms, as compared with 14 of 35
No resolution 27.1(16) 17.5 (10)
treated with antibiotic. However, the outcomes of exacer-
Deterioration 17.0(10) 12.3 (7)
Other! 3.4 (2) 3.4 (2) bations that worsened were not different from those of
exacerbations that did not worsen. Of the 20 placebo-
* p < 0.01 when results from matched-set and single exacerbations were
pooled.
treated exacerbations that were associated with an in-
+ p < 0.05 when matched-set and single exacerbation data were pooled. crease in symptoms, 8 were treatment failures, 3 with
J Includes exacerbations in which patients had deterioration in less than 72
hours or developed unrelated illnesses. deterioration; of 14 similar antibiotic-treated exacerba-
§/> = 0.06. chi-square test. tions, 6 were failures, 4 with deterioration.
When all exacerbations without deterioration were
ed with a successful outcome. In placebo-treated exacer- considered, with treatment failures being assigned a dura-
bations, the success rate was 57.6% in those episodes tion of 22 days, antibiotic-treated exacerbations were
without increased wheezing at onset, and it was 53.7% in shorter than those treated with placebo (14.1 ± 6.3 days
those episodes with increased wheezing. In antibiotic- compared with 15.5 ± 6.1 days). Duration was also ana-
treated exacerbations, the success rate was 76.3% in lyzed by the general linear model technique, which exam-
those episodes without increased wheezing at onset and ined data from the first two to four exacerbations in pa-
65.8% in those with wheezing. The presence of wheeze, tients who had multiple exacerbations. When treatment
though associated with antibiotic therapy (Table 3), was failures were assigned a duration of 22 days, the duration
not associated with treatment success, as was antibiotic of antibiotic-treated exacerbations averaged 2.2 days less
therapy. The same result was obtained when the effect of than those treated with placebo (p = 0.02). When all
wheezing on outcome was examined with the matched- treatment failures were eliminated, antibiotic-treated ex-
set model. The inclusion of wheeze in the model did not acerbations averaged 11.9 zt 5.3 days in length and those
alter the relative odds favoring antibiotic treatment and treated with placebo were 12.8 ± 5.2 days in length, val-
did not significantly improve the fit of the model. ues that were not significantly different. Treatment order
Table 5 presents success and failure rates for exacerba- did not affect these differences.
tions classified as to type, both at onset and when the Peak flow data were analyzed in patients with paired
maximum number of symptoms was present (if addition- exacerbations that were followed for 15 to 21 days. Peak
al symptoms occurred during an exacerbation, it was re- flow declined from a mean of 215 L/min during stable
classified accordingly). By either method of classifica- baseline conditions to 190 L/min at the onset of exacer-
tion, placebo treatment showed a graduation of success bations (p < 0.05, paired Mest). Furthermore, peak flow
rates from type-1 to type-3 exacerbations, with the rate increased during the exacerbation (Figure 2 ) , and the
being highest in the type-3 exacerbations. Differences in rate of increase was faster in antibiotic-treated exacerba-
success rate between antibiotic and placebo were greatest tions than in placebo-treated ones (p < 0.02, repeated-
in exacerbations classified as type 1 at onset, and in these measures analysis of variance).

Table 5. Exacerbation Result by Type*

Placebo Group Antibiotic Group

Type 1 Type 2 Type 3 Type 1 Type 2 Type 3

££/« >

Exacerbation onset
Success 43.0(31) 60.0(45) 69.7(23) 62.9(44) 70.1(54) 74.2(26)
N o resolution 22.2(16) 26.7(20) 18.2 (6) 20.0(14) 20.8(16) 11.4 (4)
Deterioration 30.5(22) 10.7 (8) 12.1 (4) 14.3(10) 5.2 (4) 11.4 (4)
Maximum symptoms
Success 46.0(52) 66.7(36) 84.6(11) 57.7(60) 80.7(46) 85.7(18)
N o resolution 24.8(28) 24.1(13) 7.7 (1) 22.1(23) 15.8 (9) 9.5 (2)
Deterioration 26.6(30) 5.6 (3) 7.7 (1) 15.4(16) 3.5 (2) 0
* Sum of percents is less than 100% because of patients who had deterioration in less than 72 hours or who developed unrelated disease. For definition of types 1, 2, and
3, see Methods.

200 February 1987 • Annals of Internal Medicine • Volume 106 • Number 2

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 particular, agreement as to whether an exacerbation had
ended or not was 78.4% (kappa = 0.57, Table 7).

Discussion
POTENTIAL BIASES
This study shows that, compared to placebo, antibiotic
treatment of chronic obstructive pulmonary disease in ex-
acerbation produced significantly earlier resolution of
symptoms. Before this conclusion can be accepted, how-
ever, biases in the study must be examined. The most
notable was that increased wheezing was commoner in
antibiotic-treated exacerbations than in those treated
with placebo ( T a b l e 3 ) ; this observation raises the ques-
tion of whether treatment success was associated with the
increased wheezing and not antibiotics. This was not the
case. In both antibiotic- and placebo-treated episodes, ex-
acerbations having increased wheezing at onset were as-
sociated with lower success rates than those without
wheezing. The reason for the higher assignment of exac-
erbations with increased wheezing to antibiotic treatment
is not clear, but because wheezing was not associated
with successful resolution of symptoms, the increased
success rate observed with antibiotics cannot be ascribed
to wheeze.
Figure 2. Recovery of peak flow (VP) in 6 4 patients with paired A few other systemic biases existed in our study. The
exacerbations that were observed for 15 to 2 1 days. Open symbols
show antibiotic-treated exacerbations; closed circles, those treated study was conducted in double-blinded fashion; placebo
with placebo. medications were not distinguishable from antibiotics,
and neither the patients nor the study team knew which
No significant differences were seen in success rates tablets were active medication. Aside from the use of an-
that depended on the individual antibiotic used. About tibiotic or placebo, treatment regimens during exacerba-
20% of the exacerbations were treated with doxycycline tions did not differ, and our crossover design ensured that
or its placebo, whereas the remainder were evenly divided therapy between exacerbations was not different. System-
between trimethoprim-sulfamethoxazole and amoxicillin ic steroid therapy was used in approximately 42% of
or their placebos. The choice of antibiotic was not sys- both antibiotic- and placebo-treated exacerbations, and
tematically related to baseline lung function or number of the success rate with antibiotics was greater in both ster-
exacerbations. At the onset of exacerbations, the frequen- oid-treated and non-steroid-treated exacerbations. Some
cy of individual findings was similar for all three antibiot- patients dropped out of the study because their physi-
ics; and with all of them the success rate was greater than cians judged them "too sick" to participate (that is, too
that achieved with placebo, although differences were not sick to have exacerbations treated with placebo half the
statistically significant between placebo and individual time), and some exacerbations in study patients were not
agents considered separately. The use of systemic steroid treated on the study protocol for similar reasons. Thus,
treatment during exacerbations did not influence the re- our results may not apply to such patients or exacerba-
sults. In exacerbations treated with steroids, the success tions. However, we believe it unlikely that antibiotics
rate with antibiotic was higher (though not significantly would be less successful than placebo in severely ill pa-
so) than that with placebo, and these rates were similar tients. The judgment of whether a patient was "too sick"
and not significantly different from those in exacerbations
not treated with steroids. Table 6. Percentage of Exacerbations with Side Effects During
The incidence of side effects was low and not signifi- the First 9 Days of Therapy*
cantly different between antibiotic and placebo therapy
Side Effect Placebo Antibiotic
(Table 6). (Table 6 lists individual symptoms during ex- Group Group
acerbations, not patients, and many patients had more
than one side effect during a given exacerbation.) Al- % %
though gastrointestinal symptoms were slightly common- Nausea 8.0 10.1
er in antibiotic-treated exacerbations, all other symptoms Vomiting 1.9 2.8
were commoner during placebo treatment. Symptoms re- Abdominal cramps 4.2 3.2
Diarrhea 2.9 5.0
garded as serious enough to withdraw therapy occurred Rash 1.0 1.2
slightly more often during placebo-treated exacerbations. Other 6.1 1.5
Kappa scores for the comparison of nurse-practitioner Treatment discontinued 2.5 1.5
and physician evaluations of symptoms are shown in Ta- * Incidence of each side effect recorded separately. Many patients had more
ble 7. In general, agreement was fair to good (20, 21). In than one.

Anthonisen et al. • Antibiotic Therapy in COPD 2 0 1

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 Table 7. Kappa Scores for Agreement on Symptoms Between been higher for both placebo- and antibiotic-treated exac-
Physician and Nurse-Practitioner
erbations, and the difference between the two would have
Onset of End of been greater because the duration of successfully treated
EExacerbation Exacerbation exacerbations was shorter for antibiotic than for placebo.
Had we used a longer time period, the final day of evalua-
Upper respiratory infection 0.39 tion would have been at least 2 weeks after the end of the
Increased dyspnea 0.70 0.53
Increased sputum production 0.51 0.25 10-day course of antibiotic therapy, and it would have
Change in color 0.73 0.48 become unclear whether we were dealing with single ex-
Increased wheezing 0.50 0.49 acerbations. In any event, treatment failures with deterio-
Increased cough 0.56 0.57 ration largely occurred within the first 15 days of exacer-
Number of symptoms (out of five) 0.49 0.47
bations, and their definition was not dependent on an
None at end 0.57
arbitrary decision about the acceptable duration of exac-
erbations. These failures were nearly twice as common
was totally subjective, and there likely was considerable with placebo.
overlap between these patients and exacerbations and In summary, we found that antibiotic-treated exacer-
those treated on our protocol. Patients who dropped out bations were significantly more likely to resolve in 21
did not differ significantly from the remainder in symp- days than were placebo-treated ones, and they were sig-
toms or lung function. Furthermore, failures with deteri- nificantly less likely to be associated with clinically
oration occurred nearly twice as often in placebo-treated alarming deterioration after the first 72 hours. This con-
exacerbations, which argues that antibiotic treatment was clusion was supported by the more rapid recovery of
advantageous in severe exacerbations. peak flow in patients with antibiotic-treated exacerba-
Some might argue that our criteria for the onset and tions. We do not believe that these results were due to
resolution of exacerbations were "soft" and subjective, biases in the study or methods of analysis. Finally, the
because they depended on patient reports of symptoms. close similarity of rates in the two panels of Table 4 indi-
We deliberately used symptoms to define the onset of cates that our result was not due to data from a few
exacerbations, because these are the criteria applied in patients with many exacerbations; success and failure
medical practice, and once onset was defined in these rates for all exacerbations were similar to those when
terms, resolution had to be defined in similar ones. only each patient's first exacerbation was considered.
Agreement was fair to good between our nurse-practi- This conclusion was supported by the results of the
tioners and the patients' physicians as to whether in- matched-set analyses, which considered only the first
creased symptoms were present or absent. The kappa four exacerbations in each patient.
scores in Table 7 are substantially higher—indicating
better agreement—than those between radiologists read- COMPARISON W I T H OTHER S T U D I E S
ing xeromammograms (22). This agreement was How did our findings differ from those of others who
achieved despite some differences in methods. The nurse- have studied this question? In 1975, Tager and Speizer
practitioners interviewed patients directly whereas the (5) reviewed six adequately controlled studies of antibi-
physicians did so by telephone, and the timing of the otic therapy in chronic obstructive pulmonary disease in
interviews could vary by as much as 24 hours. Thus, which the patient populations were reasonably well de-
though our criteria may have been "soft," they were re- fined. Two studies were "positive," purporting to show
producible. an advantage for antibiotic, but one (13) used question-
Symptomatic evidence favoring the efficacy of antibiot- able statistical methods. The other (12) compared a
ic therapy was supported by the peak flow results shown group of 86 placebo-treated patients with two groups of
in Figure 2. In antibiotic-treated exacerbations, peak approximately equal size, one of patients treated with tet-
flow, an objective measurement, increased more rapidly racycline and the other of patients treated with chloram-
than it did in exacerbations treated with placebo. It phenicol. The antibiotic-treated groups improved more
should be noted that our analysis of the peak flow results rapidly in terms of "general assessment" and "sputum
was, if anything, biased against an antibiotic advantage. purulence," but early relapse was seen, particularly for
We studied pairs of exacerbations from single patients sputum purulence, in all groups. Of four "negative" stud-
that were observed for 15 to 21 days; exacerbations with ies, three (23-25) showed a trend favoring antibiotic
deterioration were excluded from the analysis. Because therapy; the largest examined results from 71 antibiotic-
such failures occurred more often in placebo-treated epi- treated and 75 placebo-treated exacerbations (23). The
sodes and were probably associated with prolonged de- last of the six studies reviewed had only 9 patients in the
pressions of peak flow, their exclusion tended to increase antibiotic treatment group (26).
apparent recovery of peak flow more in the placebo than More recently, Nicotra and coworkers (6) completed
in the antibiotic group. a careful study of 40 patients with chronic obstructive
Others might argue that our classifying treatment out- pulmonary disease hospitalized with acute exacerbations,
come as success or failure at 21 days was arbitrary and comparing tetracycline treatment with placebo. Their as-
that a different result might have been observed had a sessment was based on lung function tests and blood gas-
different time period been used. Had we used a shorter es, and at 7 days there was no difference between treat-
time period for our criterion, failure rates would have ment groups. However, both arterial P02 and peak flow

202 February 1987 • Annals of Internal Medicine • Volume 106 • Number 2

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 increased more in the tetracycline group. Indeed, the dif- Type-2 exacerbations, in which two of the three symp-
ferences they observed in peak flow at 7 days were very toms were present at onset, had a 70% success rate with
similar to those we found at 6 days (Figure 2 ) . Presum- antibiotic and a 60% success rate without it; and al-
ably, our differences were statistically significant because though deterioration was relatively uncommon, it oc-
we had a larger patient group and because our paired, curred more often in placebo-treated exacerbations (Ta-
within-patient analysis reduced the variability of the re- ble 5). Antibiotic therapy could probably be justified for
sult. Thus, our results were not qualitatively different such exacerbations in patients known to tolerate the pro-
from those in the literature, and the reason we found posed antibiotic.
statistically significant advantages for antibiotic therapy We also analyzed individual symptoms (Table 3) to
is probably that we observed many more exacerbations see if any were associated with a greater difference be-
than others did. tween antibiotic and placebo therapy than the group as a
whole. We found that although some (notably increased
CLINICAL SIGNIFICANCE dyspnea, cough, sputum, and sputum purulence) were
If the statistical validity of our results is accepted, what associated with a difference between treatments similar to
is their clinical significance? The difference in success rate that of the study as a whole, none clearly exceeded it or
between antibiotic and placebo treatment approximated approached the treatment-related differences noted for
13%, with more than half of the placebo treatments be- type-1 exacerbations.
ing successful. This percentage does not appear to be a ACKNOWLEDGMENTS: The authors thank the nurse-practitioners who
great clinical advantage. The difference in failures with did the study, L. A. Mendella, B. N. Nord, I. Warner, D. Gaborieau, A.
Szabo, and M. A. Robinson; and thank Dr. V. A. Taraska, who contributed
deterioration was probably of greater clinical signifi- patients to the study.
cance; patients having such failures were thought to need Grant support: by a grant from Health and Welfare Canada. Drugs and
additional treatment, which certainly resulted in in- placebos were supplied free of charge by the following manufacturers: Bur-
roughs Wellcome Co. (trimethoprim-sulfamethoxazole, Septra DS, Ayerst
creased cost and likely was associated with increased pa- Laboratories (amoxicillin, Amoxil), and Pfizer Canada, Inc. (doxycycline,
tient morbidity. Although these failures were relatively Vibramycin).
uncommon, they occurred nearly twice as often with pla- • Requests for reprints should be addressed to N. R. Anthonisen, M.D.; F-2
cebo treatment and probably afford the best argument for General Centre, Health Sciences Centre, University of Manitoba, 700 Wil-
antibiotic therapy. liam Avenue; Winnipeg, Manitoba, Canada R3E 0Z3.

Arguments against antibiotic therapy would be cost

and side effects. The cost of broad-spectrum antibiotics References
like the ones we used is relatively small and should not 1. INGRAM RH J R . Chronic bronchitis, emphysema, and airways obstruc-
deter their use. Side effects were not severe and occurred tion. In: P E T E R S D O R F RG, A D A M S R D , B R A U N W A L D E, ISSELBACH-
ER KJ, M A R T I N JB, WILSON JD, eds. Harrison's Principles of Internal
no more often than with placebo (Table 7 ) . It should be Medicine. 10th ed. New York: McGraw-Hill Book Co.; 1983:1550.
noted that the incidence of side effects in our study was 2. W E L C H MH. Obstructive diseases. In: G U E N T E R CA, W E L C H M H , eds.
almost certainly less than that which would be noted if Pulmonary Medicine. 2nd ed. Philadelphia: J. B. Lippincott Company;
1982:697.
patients were randomly exposed to the antibiotics we 3. H O W E L L JBL. Chronic lung disease with airflow obstruction—chronic
used. Most, if not all, of our patients had had courses of bronchitis and emphysema. In: EMERSON P, ed. Thoracic Medicine.
broad-spectrum antibiotics before they entered this trial, London: Butterworth; 1981:456.
4. STUART-HARRIS C. Acute respiratory infection in the patient with
and presumably their physicians specified a particular an- chronic bronchitis and emphysema. In: FISHMAN AP, ed. Pulmonary
tibiotic in light of this past experience. A drug was avoid- Diseases and Disorders. New York: McGraw-Hill Book Co.; 1980:455.
ed if a patient was thought to be sensitive to it, thereby 5. TAGER I, SPEIZER FE. Role of infection in chronic bronchitis. N Engl J
Med. 1975;292:563-70.
decreasing the side effects seen in this study. It would 6. NICOTRA MB, RIVERA M, A W E RJ. Antibiotic therapy in acute exacer-
appear, however, that side effects are likely to be very bations of chronic bronchitis: a controlled study using tetracycline. Ann
Intern Med. 1982;97:18-21.
uncommon and of little significance in patients with es- 7. N A T I O N A L H E A R T , L U N G A N D B L O O D INSTITUTE, DIVISION O F L U N G
tablished chronic obstructive pulmonary disease. DISEASES. Protocol Intermittent Positive Pressure Breathing Collabo-
Antibiotic therapy would be more attractive if we were rative Program. Bethesda, Maryland: National Institutes of Health;
1978.
able to identify kinds of exacerbations that are associated 8. MORRIS JF, KOSKI A, JOHNSON LC. Spirometric standards for healthy
with the greatest difference between antibiotic and place- nonsmoking adults. Am Rev Respir Dis. 1971;103:57-67.
9. G O L D M A N HI, BECKLACKE MR. Respiratory function tests: normal
bo therapy. Furthermore, if these exacerbations could be values at medium altitudes and the prediction of normal results. Am
identified at their onset, the expense of close follow-up in Rev Tuberc. 1959;79:457-67.
all patients could be avoided. Table 5 indicates that exac- 10. M E N D E L L A LA, M A N F R E D A J, W A R R E N CPW, A N T H O N I S E N NR.
Steroid responses in stable chronic obstructive pulmonary disease. Ann
erbations classified as type 1 at onset showed a relatively Intern Med. 1982;96:17-21.
large advantage for antibiotic therapy. There was nearly 11. N O C T U R N A L O X Y G E N T H E R A P Y T R I A L G R O U P . Continuous or noctur-
a 20% difference in success rates, and deterioration oc- nal oxygen therapy in hypoxemic chronic obstructive lung disease: a
clinical trial. Ann Intern Med. 1980;93:391-8.
curred in 30% of placebo-treated episodes, more than 12. P I N E S A, R A A F A I H, G R E E N F I E L D JSB, LINSELL W D , SOLARI ME.
twice as often as with antibiotic. Clearly, exacerbations Antibiotic regimens in moderately ill patients with purulent exacerba-
tions of chronic bronchitis. Br J Dis Chest. 1972;66:107-15.
presenting with increased dyspnea, sputum, and sputum
13. P I N E S A, R A A F A T H, PLUCINSKI K, G R E E N F I E L D JSB, SOLARI M.
purulence should be treated with antibiotic. Antibiotics Antibiotic regimens in severe and acute purulent exacerbations of chron-
appeared to confer no benefit in type-3 exacerbations and ic bronchitis. Br Med J. 1968;2:735-8.
14. H U G H E S D T D , D R E W CD, JOHNSON TBW, JARVIS J D . Trimethoprim
apparently should not be used when only one symptom and sulphamethoxazole in the treatment of chronic chest infections.
(dyspnea, sputum, or sputum purulence) is present. Chemotherapy. 1969;14:151-7.

Anthonisen eta/. • Antibiotic Therapy in COPD 2 0 3

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016

 15. P I N E S A, R A A F A T H, G R E E N F I E L D JSB, SIDDIQUI G, L E N O X - S M I T H I, tics of concordance. Clin Pharmacol Ther. 1981;29:111-23.
LlNSELL WD. The management of purulent exacerbations of chronic 21. LANDIS RH, KOCH GG. The measurement of observer agreement for
bronchitis: a comparison of co-trimoxazole and tetracycline. Practition- categorical data. Biometrics. 1977;33:159-74.
er. 1972;208:265-7. 22 BOYD N F , W O L F S O N C, M O S K O W I T Z M, et al. Observer variation in the
16. BRESLOW NE, DAY NE. Statistical Methods in Cancer Research. Lyon: interpretation of xeromammograms. JNCI. 1982;68:357-63.
International Agency for Research on Cancer; 1980:248-79. ( I A R C Sci- 23. E L M E S PC, F L E T C H E R CM, D U T T O N AAC. Prophylactic use of oxytet-
entific Publications, no. 32.) racycline for exacerbations of chronic bronchitis. Br Med J.
17. KALBFLEISCH J D , PRENTICE RL. The Statistical Analysis of Failure 1957;2:1272-5.
Time Data. New York: John Wiley & Sons; 1980:200. 24. W A D E OL, WILSON TS. Value of ampicillin in the hospital treatment of
18. N A E S S E N S JM, O F F O R D K P , SCOTT WF, D A O O D SL. The MCSTRAT exacerbations of chronic bronchitis. Br Med J. 1965;2:904-8.
procedure. In: Supplemental Library, User's Guide. 5th ed. Cary, North 25. BERRY DG, FRY J, HINDLEY CP, et al. Exacerbations of chronic bron-
Carolina: SAS Institutes Inc.; 1986:307-28. chitis—treatment with oxytetracycline. Lancet. 1960;1:137-9.
19. SAS Users Guide: Statistics. Cary, North Carolina: SAS Institute, Inc.; 26. P E T E R S E N ES, ESMANN V, H O N C K E P, M U N K N E R C. A controlled
1982. study of the effect of treatment on chronic bronchitis: an evaluation
20. K R A M E R MS, FEINSTEIN AR. Clinical biostatistics: LIV. The biostatis- using pulmonary function tests. Acta Med Scand. 1967;182:293-305.

204 Annals of Internal Medicine. 1987;106:204-208. ©1987 American College of Physicians

Downloaded From: http://annals.org/ by a University of Otago User on 12/08/2016