Professional Documents
Culture Documents
00
Printedin GreatBritain.All rightsreserved Copyright 8, 1984 Pergamon Press Ltd
Abstract-Cough is the fifth most common reason for physician visits, but data on acute cough
have rarely been collected in a standardized manner and have not been analyzed in a multivariate
fashion. We report data on 1819 patients presenting with cough, all of whom received a
standardized history and physical, and a chest X-ray. Only 48 (2.6%) were found to have an acute
radiographic infiltrate (pneumonia). The prevalence of common signs and symptoms is shown for
the patients with and without pneumonia. Thirty-two of these findings were significant predictors
of pneumonia (p < 0.05, one-tailed). These 32 did not include some of the expected predictors of
pneumonia and did include some predictors not previously described in the literature. A diagnostic
rule is developed which identifies pneumonia patients with 91% sensitivity and 40% specificity, or
74% sensitivity and 70% specificity. The study results suggest that many pneumonias could be
predicted based only on the patients histories. Physician visits to determine physical findings and
chest X-rays might be avoided by telephone triage, with substantial cost savings.
I. INTRODUCTION
COUGH, the fifth most frequent principal reason for patients visiting physicians, resulted
in approximately 29 million visits to office-based physicians in 1977-78. Fifty-four per cent
of these visits were by adults, over 11% of whom had chest radiographs [l].
Although acute coughs have major impact on the health care costs and morbidity of
adults, there are few data from the research literature to help providers identify and choose
the best management strategies in primary care populations. Traditionally acute cough is
evaluated by a limited, individualized patient history and physical, frequently by chest
X-rays (1 in 7 patients [2]), and other tests deemed necessary by the providers.
The physicians concern about the possibility of pneumonia (i.e. the presence of an acute
infiltrate on chest radiographs) appears to be the chief reason for obtaining chest X-rays
[2]. However, there have been no studies of the relationships between clinical findings and
the presence of these radiographic infiltrates in primary care patients presenting with
cough. The study presented here is the only one which includes all of the following features:
defined entry criteria capturing adults with acute cough in a typical primary care
population; standardized extensive initial history and physical evaluation of all patients;
and chest radiographs obtained on all patients.
All correspondence should be addressed to: Dr Paula Diehr, Health Services Research, Pacific Medical Center.
1131 14th Avenue South, Seattle, WA 98114, U.S.A.
This research was supported in part by the U.S. Army Health Services command through Contracts
DADA1 1-77-C-0008 and DADAll-78-C-0009, and by DHHS Bureau of Medical Services.
The views expressed here are not necessarily those of the U.S. Army or of DHHS.
216 PAULA DIEHR et al.
Study goals
In this paper we analyze data on 150 symptoms and signs collected from 1819 cough
patients in an acute care setting to address several questions:
(a) Which of the commonly accepted predictors of pneumonia in this data base are
significantly related to radiographic pneumonia?
(b) Which commonly accepted predictors are not actually significantly related to
pneumonia?
(c) Which non-standard predictors are significantly related to pneumonia?
(3) Can we develop a simple rule to separate cough patients with pneumonia (defined
here as acute radiographic infiltrate) from other cough patients on the basis of the
signs and symptoms? What is the optimal decision rule to accomplish this?
II. METHODS
The study was conducted at the emergency department of Brooke Army Medical Center
(BAMC) at Fort Sam Houston, Texas. We studied all consenting, non-pregnant, adult
walk-in patients seeking medical care for the first time for coughs of less than one months
duration. Patients with a pulse rate of 160/min or more, a temperature of 104F or higher,
or a systolic blood pressure of 90 mmHg or lower, were excluded from the study, as were
patients arriving by ambulance. Approximately one fourth of the eligible patients declined
to participate in this study, which involved randomization and follow-up. Nearly all gave
the amount of time required as the reason for refusal, and none mentioned health as the
reason.
Postero-anterior and lateral chest films were taken of all patients. Radiology residents,
with staff radiologists available for consultation, and knowing only that the patient had
an acute cough, interpreted each chest radiograph. Previous chest films, when available,
were used in the interpretation of the films.
Without knowledge of chest film results, research assistants collected a standard history
on all patients. Physicians reviewed and confirmed the history and performed a standard
physical examination. They also stated whether they would have ordered a CXR or
antibiotics for each patient. The physicians were all board-eligible or board certified in
either internal or adolescent medicine. We attempted to obtain follow-up chest X-rays and
to identify the etiologies of all infiltrates found at the initial visit.
The resulting data base consisted of the initial interview/physical examination informa-
tion, follow-up and X-ray findings. The data were analyzed in several steps. First, a subset
of patients was chosen, which included all of the pneumonia patients (PN), and a 25%
random subset of the non-pneumonia patients (NPN). The remainder of the NPN patients
were reserved for testing of the models resulting from this analysis. An additional group
of patients with equivocal infiltrates was reserved for later analysis, since we did not know
whether it was more appropriate to group them with the PN or the NPN patients.
Next, the proportion of patients reporting each symptom was tabulated. Each symptom
was then tested with a t or chi-square test or Fishers exact test [3] for significant
association with radiographic pneumonia. Symptoms which were at least marginally
associated with pneumonia (p < 0.10, 2-tailed), and which occurred in at least 10 (20%) of
the pneumonia patients were retained for further analysis.
Stepwise discriminant analysis [4] was used to select a small subset of variables which
could be used to predict the chest X-ray result. The sensitivity, specificity, and reliability
PN in Outpatients with Acute Cough 217
No pneumonia
Sex
Male 18 (37.5) 214 (49.2)
Female 30 (62.5) 221 (50.8)
Age
13-19 I (14.6) 92 (20.2)
2fk29 1 I (22.9) 105 (24.1)
3lS39 l(14.6) 45 (10.4)
4&49 6 (12.5) 58(13.3)
50-59 8 (16.7) 71 (16.3)
6&69 4 (8.3) 48(11.0)
70-19 4 (8.3) 13 (3.0)
80+ I (2.1) 3 (0.7)
48 435
*This table omits 75% of the NPN patients and 49
patients with equivocal pneumonia.
of the resulting rule were examined. The medical records for the misclassified pneumonia
cases were reread, to examine the reasons for and consequences of misdiagnoses. Finally,
the performance of several different diagnostic strategies was examined.
III. FINDINGS
Y0 with finding
the etiologies for cough in these patients. History of previous pneumonia, chronic cough,
and pleuritic chest pain, although associated in the right direction were not significantly
related to pneumonia. Alcoholism and bloody sputum were found more frequently in the
non-pneumonia patients, All of the traditional physical examination items were
significantly related to infiltrate except for pulse above 100 and rhonchi. The mean pulse
rates were 80 for the NPN patients and 85 for the PN patients, which was a significant
difference 0, = 0.012) by a t-test. Rhonchi might have been a significant predictor if more
patients with pneumonia had been available for analysis.
The sensitivity column shows how frequently the findings were present in those patients
with pneumonia. For example, 44% of those cough patients with infiltrates had a history of
pneumonia. Most were quite infrequent. The only two findings which were present in more
than half of the PN patients (sputum production and smoking) were also present in half or
more of the NPN patients. In general, the history items are present more frequently than
the physical findings.
The spec$city of all of the findings for pneumonia is generally very high, especially for
the infrequent findings. However, the specificity is lower for the findings with relatively
high sensitivity.
In general, most findings had relative risks between 1 and 4, meaning that at most any
given finding increased the probability of pneumonia by a factor of about four. Clearly,
no single finding is a good predictor of pneumonia. Two of the symptoms (alcoholism and
bloody sputum) had risks less than 1, indicating that their absence was actually a better
predictor of pneumonia than their presence. However, there were very few patients with
these findings, suggesting that this may be a chance result.
The final column shows the predictive value of a positive finding, or the per cent of
patients with that finding who have pneumonia. These are generally very low, indicating
that no single finding will have a high yield of pneumonia cases. Even the five findings with
high predictive values had such a low prevalence that they would not be of much use in
PN in Outpatients with Acute Cough 219
A with finding
Positive
Clinical PN Relative predictive
findings Sig. NPN (sensitivity) Specificity risk WIW
Sputum production
all day *** 40.0 66.7 60.0 2.9 4.4
Smoker * 49.2 62.5 50.8 1. 3.4
Sore throat ** 74.0 58.3 26.0 0.5 4.2
Rhinorrhea *** 86.0 66.7 14.0 0.3 62
Night sweats ** 19.1 33.3 80.9 2.1 46
Chills ** 18.6 31.3 81.4 1.9 4.4
History of fever *** 20.9 43.8 79.1 2.8 54
Myalgia ** 57.5 75.0 42.5 2.2 3.5
Some sputum ** 62.4 79.2 37.6 2.2 34
Diastolic BP < 76 * 65.1 78.3 34.9 2.0 3.2
Hx purulent sputum ** 47.6 64.6 52.4 2.0 3.6
Appears severely ill *** 4.9 19.1 95. I 4.3 Y.7
Severe chest pain ** 9.0 16.7 91.0 2.0 4.x
R&F ** 7.2 19.1 92.8 2.9 6.8
Resp > 25 it* 8.4 28.3 91.6 4.0 x.3
Temp > 99.9 *** 6.2 27.1 93.8 5.1 Ill.7
Positional chest pain ** 5.7 16.7 94.3 3.1 1.4
Low .frequency items (Tested with the Fisher Exact Test)
Hist. of chest pain ** 19.8 6.3 80.2 0.3 0.9
Sputum all day for
those with any sputum ** 64.2 84.2 35.8 2.9 4.4
Tender sinuses ** 16.5 4.3 83.5 0.2 3.1
Cachexla ** 2.6 10.6 97.4 4.2 10.2
Increased Fremltus *** 0.9 8.3 99.1 8.0 20.0
Skin dehydration ** 0.2 4.3 99.8 13.1 33.1
Asymmetric resp ** 0.0 4.3 100.0 39.0 100.0
Pleural rubs * 0.5 4.3 99.5 7.8 20.0
Local dullness * 0.5 4.3 99.5 7.8 20.0
Egophony * 0.5 4.3 99.5 7.8 20.0
PN = pneumonia; NPN = no pneumonia.
tSee appendix for definition of statistics.
*p < 0. IO. two-tailed test: **p < 0.05. two-tailed test; ***p -C0.01. two-tailed test.
screening a similar population of patients with cough. (Only four PN patients had
increased fremitus, and only two PN patients had the other findings).
The Rare variable, a composite of these infrequent findings, is a highly significant
predictor of pneumonia, and is discussed later. Note, however, that even when these rare
findings are combined they still identify only about a quarter of the pneumonia patients.
Table 2 presented the clinical findings usually thought to be related to pneumonia. Table
3 shows all the items which were actually significantly (p < 0.10, two-tailed test) more
frequent in the PN than the NPN patients (or vice-versa), based on a chi-square of Fishers
exact test. The mean age, temperature, pulse, and respiration rates were compared for the
two groups using r-tests; they were also dichotomized and tested using chi-square tests.
Since the results were very similar, only the dichotomized versions are shown in Table 3.
Note that the prevalences of sore throat, rhinorrhea, high diastolic blood pressure, history
of chest pain, and tender sinuses were significantly lower in the PN group. The potential
usefulness of this negative information is discussed elsewhere [5].
Most findings were in the expected direction; however, we cannot explain the apparent
relationships between pneumonia and positional chest pain, low diastolic blood pressure.
and myalgias. Fortunately, we did not need to use these variables further.
TABLE4. STEPWISE
D,SCRM,NANT
ANALYSIS TO PREDK~T f'NE"MONl.4 AMONG
ADULTCOUGH
PATENTS
Disc score y0 with symptom
Variable (Points) No pneumonia Pneumonia
Rhinorrhea -2 86.0 66.7
Sore throat -1 74.0 58.3
Night sweats 1 19.1 33.3
Myalgia I 51.5 75.0
Sputum all day I 40.0 66.7
Respiratory rate > 25 2 8.4 28.3
Temp. 100 or more 2 6.2 21.1
analysis. The other symptoms were excluded, even though they were statistically significant
predictors, because fewer than ten of the PN patients had any single finding.
Table 4 shows the seven variables chosen by a stepwise discriminant analysis of the
46 PN and 428 NPN patients for whom complete data were available. Variables were
chosen in order of the strength of their association with PN, and only variables with a
significant F-ratio (p < 0.05) at that step were allowed to enter. Some of the symptoms
in Table 3 which were significant when considered alone did not add more information
once other items were entered, and so were not selected by the stepwise procedure. Two
of the negative symptoms, rhinorrhea and sore throat, were chosen. It is noteworthy
that none of these findings would require a physical examination by a physician, and that
all might be determined over the telephone.
Table 4 shows the approximate weighting or points for each item in the discriminant
score, which can be used to assign the diagnosis. The score for a patient is calculated by
summing the points for each symptom present; a patient with rhinorrea, fever, and myalgia
would thus have a point score of
(-2)+(+2)+(+1)= 1.
The discriminant rule may be better understood by considering the distribution of points
for the PN and NPN groups, shown in Table 5. For example, 12 (8.8%) of the 136 patients
with a score of 1 had infiltrates. Note that the NPN patients tend to have low scores, while
the PN patients tend to have higher scores. For example, in Table 5, 59% of the PN cases
had a score of 1 or more, compared to only 12% of the NPN cases. A reasonable clinical
decision might be to assume the patient has an infiltrate if his score is 1 or higher. Then,
the hypothetical patient with a score of + 1 would be diagnosed as having pneumonia, and
treated appropriately, without requiring a CXR. The rule which cuts off at + 1 has 59%
sensitivity and 88% specificity. A rule which cut off at zero would have 74% sensitivity and
70% specificity; a rule which cut off at - 1 would have 91% sensitivity and 40% specificity.
The choice of an appropriate cut-off would, of course, be determined by the purposes of
the classification.
TABLE5. DISTRIBUTION
OF POINTSCORES
Point No. with No. w/out A with Cumulative
Score PN PN* PN sensitivity specificity
-3 0 140 0 100 8
-2 4 552 0.7 91 40
-1 8 504 1.6 74 70
0 7 316 2.2 59 88
I 12 124 8.8 33 96
2 6 52 10.3 20 99
3 4 12 25.0 I1 99
4 3 8 27.3 4 100
5 I 4 20.0 2 100
6 I 0 100.0 0 100
46 1712
*Adjusted for 25% sampling of NPN patients. Patients with missing values for any of
the discriminant variables are omitted.
PN in Outpatients with Acute Cough 221
*The discriminant analysis chose different variables for these two rules
222 PAULA DIEHR et al.
antibiotics, 59% of all PN patients would have received them, and only 227 patients total
would have received antibiotics. Again, the rule allows a substantial decrease in patient
exposure to costly, and potentially unnecessary therapy.
The physicians stated [2] that they were ordering CXRs to identify pneumonia, but they
may have been ordering antibiotics for other reasons, which suggests that the comparison
of the two antibiotic-ordering rules may not be completely fair. Both comparisons suggest,
however, that the performance of this rule is better than or comparable to the implicit
decision rules of the physicians who were actually seeing and managing the patients.
Rare jindings
One reason that none of the traditionally taught physical findings entered the discrim-
inant rule is that they were so infrequent in the PN group. To examine whether there was
useful information in these rare findings we created a single variable which was positive
if the patient had any of: asymmetric respiration, rub, egophony, increased fremitus, or
local dullness. As shown in Table 2, Rare is highly associated with the diagnosis of
pneumonia, being present in 21.3% of the PN cases but only 3.5% of the NPN patients
(p < 0.00005). Rare had a relative risk of 6.5 and a positive predictive value of about
14%. However, although Rare is a highly significant predictor of pneumonia, it still
detects only about a fourth of the PN cases.
The discriminant analysis was repeated, allowing Rare as a potential variable. In fact,
Rare was the first variable chosen in the analysis. The other variables which entered were
the same as in Table 4. The rule based on Rare had nearly slightly better sensitivity and
specificity than the rule in Table 4. It was, however, less reliable based on the split sample
analysis. (Further, Rare did not enter one of the half-sample equations.) We thus chose
to evaluate the rule in Table 4 rather than the Rare rule because it did not require
examining the patient for a list of rare findings, and performed just as well.
Equivocal pneumonias
The chest radiographs of 49 patients were interpreted by the radiologists as showing
equivocal pneumonia. These patients were not used in determining the discriminant rule.
The point scores were calculated for the patients with equivocal findings and compared
to the results shown in Table 5. The distribution of points for the PN and equivocal
patients is significantly different, but the distributions for the NPN and the equivocals are
not significantly different. Thus, by these criteria, the equivocal cases may not have had
pneumonia. On the other hand, this may indicate that the rule is not sensitive to
borderline infiltrates. This finding supports our early decision not to include the
equivocal patients with the unequivocal pneumonias, since they are different, at least for
the items used in the discriminant function.
PN in Outpatients with Acute Cough 223
This study has examined the relationships between clinical findings (symptoms and
signs), and the presence of radiographic infiltrates among 1819 acute cough patients. The
basic findings were:
(1) Only 2.4/, of these patients presenting with acute cough had acute radiographic
infiltrates.
(2) Some of the expected predictors of pneumonia were in fact not significantly related
to the presence of an infiltrate among cough patients, as shown in Table 2.
(3) Some unexpected predictors were found, as shown in Table 3.
(4) A rule we developed, based only on presenting symptoms, could predict radiographic
infiltrates better than physicians could.
*A complete list of the data collected is available from the authors on request
224 PAULADIEHR et al.
the findings might be spurious, but the results which disagree with the literature clearly
need to be validated on a new sample of patients.
As we have mentioned above, the failure of some of the expected predictors of
pneumonia to be significantly related to infiltrates may have been due to the small number
of pneumonias studied. On the other hand, it may have been caused by the incomplete
nature of previous studies, where not all patients were X-rayed, thus allowing the bias of
the selection procedure to be mistaken for correlates of pneumonia [8].
These results suggest possible changes in the management of patients presenting with
cough. At present most such patients are seen by physicians who gather clinical data,
frequently perform tests (such as X-rays) and determine management. Pneumonia can be
distinguished with fairly high accuracy simply on the basis of a set of 7 questions which
could be administered by non-physicians, and possibly by the patients themselves. This
application of telephone management could hypothetically save patients many low-yield
physician visits and chest radiographs and could thus reduce overall health care costs,
without compromising patient safety or treatment efficacy. A prospective experimental
evaluation of such a strategy would clearly be valuable at this time.
Acknowledgements-The authors gratefully acknowledge the assistance of Robin Denny in manuscript prepara-
tion, and of Valerie James and Lynn Burgin in data collection. In addition we are grateful for the help of Dr
Theodore McNitt, the emergency room physicians, and the Department of Radiology at Brooke Army Medical
Center for their valuable participation in this study.
REFERENCES
1. Cypress, Beulah K: Patients reasons for visiting physicians. Vital and health statistics. Series 13, Data from
the National Health Survey, No. 56. DHHS publication No. (PHS) 82-1717, 1981
2. Bushyhead JB, Wood RW, Tompkins RK, Wolcott BW, Diehr PH: The effect of chest radiographs on the
management and clinical course of patients with acute cough. Med Care 21: 497-507, 1983
3. Ipsen J, Feigl P: Bancrofts Introduction to Biostatistics. New York: Harper and Row, 1970
4. Kleinbaum DG, Kupper LL: Applied Regression Analysis and other Multivariable Methods. North Scituate,
Massachusetts: Duxbury Press, 1978.
5. Christensen-Szalanski J, Bushyhead J: Physicians misunderstanding of normal findings. Med Decision
Making 3: 169-175, 1983
6. Dixon WJ, Brown MB (Eds): BMDP-77: Biomedical Computer Programs-P Series. Berkeley: University of
California Press, 1977
7. National Center of Health Statistics. Advance Data from Vital and Health Statistics, No. 65. Washington
DC: US Government Printing Office, Nov. 5, 1980
8. Bushyhead JB, Christensen-Szalanski JJJ: Feedback and the illusion of validity in a medical clinic. Med
Decision Making 1: 115-123, 198I
APPENDIX
Calculation of statistics in Tables 2 and 3
(1) Sensitivity is the per cent of PN patients with the finding.
(2) SpeciJicity is the per cent of NPN patients without the finding.
(3) Relative risk is
(No. with finding, with PN)/(No. with finding)
RR=(N o. without finding, with PN)/(No. without finding)
The formula used in Tables 2 and 3, which accounts for 25% sampling of the NPN cases, is as follows:
Let X = sensitivity of a finding (%);
Y = specificity of a finding (%).
X*[48*(100 - X) + 435*4* yl
RR = (100 - X)*[48*X + (100 - Y)*435*4]
For Previous PN--ever, X = 43.8, Y = 64.1, and
RR = 43.8*[48*56.2 +435*4&l. I] = 1,38,
56.2*[48*43.8 + 35.9*435*4]
(4) Positive predictive value is
ppv = (No. Wit;onc$;n;r) x loo%,
For Tables 2 and 3, taking the 25% sampling of the NPN patients into account, the formula is as follows:
PN in Outpatients with Acute Cough 225