Aretrospective study was performed to identify all patients who had
undergone a hip arthrocentesis for the diagnostic workup for an acutely
irritable hip at our tertiary care children's hospital between January 1,
1992, and December 31, 2000. The study was approved by our institutional
review board. A total of 263 patients underwent a hip arthrocentesis during
the study period. All evaluations included a history, physical examination,
and laboratory studies, with a complete blood-cell count with differential,
measurement of the erythrocyte sedimentation rate, and blood cultures.
Additional serum analyses, such as measurements of the C-reactive protein
level and testing for antinuclear antibody titers and rheumatoid factor, were
performed on the basis of the physician's preference and the clinical
presentation. Plain radiographs of the pelvis and the proximal part of the
femur were made for all patients and were evaluated for the presence of
fractures or other osseous lesions. If septic arthritis was a possible
diagnosis, the patient underwent an ultrasound examination of both hips to
look for hip joint effusion. If an effusion was documented, arthrocentesis was
performed under fluoroscopic guidance in the radiology or operating suite,
with arthrographic confirmation of the intra-articular position of the needle;
ultrasound was not utilized for needle-positioning at the time of
arthrocentesis during the study period. Patients were excluded from the study
if no synovial fluid could be obtained with the arthrocentesis. Analysis of
synovial fluid included a white blood-cell count and differential, Gram stain,
and culture. After evaluation, patients with the diagnosis of transient
synovitis were treated with oral analgesics. Patients with the diagnosis of
septic arthritis underwent emergent surgical drainage of the hip joint and
were started on empiric intravenous antibiotics.
Medical records were reviewed for patient age, gender, disease history
(duration of symptoms, previous health-care visit, recent antibiotic therapy
and reason for the therapy, fever, and weight-bearing status), clinical
findings (body temperature), radiographic findings, ultrasound findings,
results of the arthrocentesis (amount and appearance of the aspirate),
laboratory studies (measurement of the erythrocyte sedimentation rate, serum
white blood-cell count with differential, white blood-cell count with
differential in the synovial fluid, and results of cultures of blood and
synovial fluid), treatment, and complications. Weight-bearing status was
determined from the clinical history. Fever was defined as an oral temperature
of =38.5°C during the week prior to the evaluation or at the emergency
room visit. A previous health-care visit was defined as any evaluation of the
irritable hip by a health-care provider during the present illness.
Three separate diagnostic groups were established on the basis of the
criteria of Kocher et
al.2: true septic
arthritis, presumed septic arthritis, and transient synovitis
(Table I). The diagnosis of
true septic arthritis (twenty patients; twenty hips) was assigned when there
was bacterial growth on culture of synovial fluid or both bacterial growth on
culture of blood and a white blood-cell count of =50,000
cells/mm3 (=50.0 × 109/L) in the synovial
fluid. The diagnosis of presumed septic arthritis (twenty-six patients;
twenty-seven hips) was assigned when there was no growth on culture of
synovial fluid or blood but the white blood-cell count in the synovial fluid
was =50,000 cells/mm3 (=50.0 × 109/L). The
diagnosis of transient synovitis (117 patients; 118 hips) was assigned when
there was no growth on culture of synovial fluid or blood, the white
blood-cell count in the synovial fluid was <50,000 cells/mm3
(<50.0 × 109/L), the symptoms resolved without intravenous
antibiotics or surgical intervention, and there was no further development of
disease as reported in the medical records.
On the basis of the study parameters, three groups of patients were
excluded from the original series of 263 patients. One group of forty-six
patients (forty-six hips) was excluded because no synovial fluid could be
obtained at the time of the arthrocentesis despite ultrasound confirmation of
an effusion. Transient synovitis was the diagnosis in twenty-six of these
patients; osteomyelitis, in five patients; cellulitis, in five; septicemia, in
four; myositis, in two; abscess, in two; and sickle cell disease and trauma in
one patient each. A failure to obtain fluid despite intra-articular placement
of the needle indicates that there was a very small amount of fluid within the
hip. Septic arthritis did not develop in any of these excluded patients.
Another group of patients, of thirty-six children (thirty-six hips), was
excluded from the analysis on the basis of their final diagnosis. A malignant
tumor was diagnosed in seven of these patients; rheumatological disease, in
five; osteomyelitis, in five; sickle cell crisis, in three;
Legg-Calvé-Perthes disease, in three; immunocompromise, in three; a
gunshot wound, femoral fracture, and postoperative infection in two patients
each; and cellulitis, phlebitis, dermatomyositis, and systemic sepsis in one
patient each. A third group, of eighteen patients (eighteen hips), was
excluded because the white blood-cell count in the synovial fluid was
<50,000 cells/mm3 (<50.0 × 109/L) but the
patients were treated with intravenous antibiotics and/or surgical drainage.
Overall, 100 patients in the above three groups were excluded, leaving 163
patients with 165 hips who satisfied the inclusion criteria.
Univariate analyses were performed with the use of the two-sample Student t
test for continuous variables and the Fisher exact test for categorical
variables. The group with septic arthritis was compared with the group with
transient synovitis. Multiple logistic regression was performed to identify
the most parsimonious model of independent predictors on the basis of two
criteria: first, a p value of 0.05 for the likelihood ratio test of a model
with and a model without a variable and, second, clinically important changes
in parameter estimates of variables included in a multivariate model. Adjusted
odds ratios and 95% confidence intervals were calculated on the basis of the
logistic regression models. The probability of septic arthritis was estimated
for each combination of independent predictors included in the logistic
models. A receiver operating characteristic curve was constructed to assess
the diagnostic performance of each set of independent predictors included in
the models. All statistical analyses were performed with Statistical Analysis
System (SAS) software (version 8.02; SAS Institute, Cary, North Carolina). P
values of <0.05 were considered to be significant.
Of the 163 patients (165 hips), 110 were boys and fifty-three were girls.
The left hip was involved in seventy patients; the right hip, in ninety-one;
and both hips, in two. Of the hips with septic arthritis (twenty with true
septic arthritis and twenty-seven with presumed septic arthritis), only twenty
(43%) had bacterial growth on culture of synovial fluid and/or blood. Of the
twenty hips with true septic arthritis, seventeen (85%) had growth on synovial
fluid culture and no growth on blood culture and three (15%) had no growth on
synovial fluid culture but did have growth on blood culture. Organisms
isolated on culture of synovial fluid and blood included coagulase-negative
Staphylococcus (seven hips, 35%), Staphylococcus aureus (six hips,
30%), Streptococcus viridans (four hips, 20%), Haemophilus
influenza Type b (two hips, 10%), and Pseudomonas aeruginosa
(one hip, 5%). Gram stain analysis identified an organism in nine hips (45%)
with true septic arthritis: seven of the seventeen with positive findings on
synovial fluid culture and two of the three with positive findings on blood
culture. One hip with presumed septic arthritis (4%) had a positive finding on
Gram staining of synovial fluid but no growth on culture of blood or synovial
fluid.
Univariate analysis demonstrated significant differences (p < 0.05)
between the forty-seven hips with septic arthritis and the 118 hips with
transient synovitis with regard to ten measures
(Table II): erythrocyte
sedimentation rate; peripheral percentages of lymphocytes and neutrophils;
total white blood-cell count and percentages of lymphocytes, neutrophils, and
monocytes in the synovial fluid; male gender; a previous health-care visit;
and a history of fever. Univariate analysis demonstrated significant
differences (p < 0.05) between the twenty hips with true septic arthritis
and the twenty-seven hips with presumed septic arthritis with regard to four
factors: age, duration of symptoms, white blood-cell count in the synovial
fluid, and temperature (Table
III). There were no other significant differences between the two
groups.
The four independent multivariate predictors determined by Kocher et
al.2
(Table IV) were tested for
their ability to differentiate between septic arthritis (true and presumed)
and transient synovitis in our population. Two of the variables,
non-weight-bearing status and erythrocyte sedimentation rate, had a p value of
>0.05 and a confidence interval including 1; therefore, they did not
contribute to the model. The algorithm was tested to determine the predicted
probability of a patient having septic arthritis in the presence of all four
presenting variables (Table V),
and it was found to be only 59% in our patient population, which was much
lower than the 99.6% value in the patient population of Kocher et al. The
Hosmer-Lemeshow goodness-of-fit test showed a fit of p = 0.3382.
With use of our patient database, several logistic regression models were
constructed and tested for their ability to predict septic arthritis. The best
model for our patient population consisted of three variables: a history of
fever (=38.5°C), a peripheral white blood-cell count of
>12,000/mm3 (>12 × 109/L), and a previous
health-care visit (Tables IV
and V). The Hosmer-Lemeshow
goodness-of-fit test indicated a good fit (p = 0.9802). This model improved
the predicted probability of septic arthritis from 59% based on the Kocher et
al.2 criteria to 71%
based on our three-variable model.
The physiologic response to an early bacterial infection can be quite
variable and can even result in serum markers of inflammation within the
normal range of
values15,16.
Diagnosis of an infected hip is especially difficult in the early phase, and
there is no single serum analysis that can serve as a definitive
test15. Only
bacterial growth on culture of synovial fluid can definitively establish the
diagnosis of septic arthritis, and culture results usually are not available
when the patient presents for evaluation and
treatment17.
Furthermore, previous studies have demonstrated that, even in a definite case
of septic arthritis, the laboratory values can be in the normal range: up to
21% (eight of thirty-eight
hips)3 can have a
normal erythrocyte sedimentation rate, 12.5% (one of eight
hips)13 to 58%
(fifteen of twenty-six
hips)14 can have a
normal body temperature, and 25% (two of eight
hips)13 to 74%
(twenty-eight of thirty-eight
hips)3 can have a
normal peripheral white blood-cell
count5,8,13-15.
Because there is no single definitive test, several multifactorial
algorithms have been proposed to minimize the need for ultrasonography and the
painful interventions of arthrocentesis and surgical drainage. Del Becarro et
al.3 recommended
that all patients with an irritable hip be considered for diagnostic hip
arthrocentesis when there is no identifiable source and the erythrocyte
sedimentation rate is =20 mm/hr or the oral temperature is =37.5°C.
The combination of an elevated erythrocyte sedimentation rate and a high body
temperature identified 97% (thirty-seven) of thirty-eight cases of septic
arthritis of the hip in their population. Eich et
al.13 concluded
that a rectal temperature of =38°C, an erythrocyte sedimentation rate
of =20 mm/hr, and a C-reactive protein level of =20 mg/dL (=200 mg/L)
are the most important parameters for differentiation. If two of these factors
were present and ultrasound showed a hip effusion, there was a sensitivity of
100% and a specificity of 89% for septic arthritis.
Beach18 advocated a
scoring system utilizing the factors of hip pain on physical examination,
tenderness, a fever of =38°C, and an erythrocyte sedimentation rate of
=20 mm/hr. Beach found a low chance of infection in hips with no or one
finding, whereas a hip with two or more findings had a high risk of having an
infection and should undergo ultrasonographic evaluation and arthrocentesis if
an effusion was detected.
Kocher et al.2
identified four variables as having a predictive value for septic arthritis: a
history of a fever of =38.5°C, an erythrocyte sedimentation rate of
=40 mm/hr, non-weight-bearing status, and a peripheral white blood-cell
count of >12,000
cells/mm3 (>12.0
× 109/L). The extremely significant adjusted odds ratios for
these four factors, 14.4 to 38.6, was strong evidence that these factors were
highly effective in differentiating septic arthritis from transient
synovitis.
Clinical prediction algorithms must be generalizable, and our hope was to
confirm these findings in our population. However, when the model of Kocher et
al.2 was applied to
our patient population, it did not perform as well (predictive value, 59%) as
it did in their patient population (predictive value, 99.6%). Analysis of our
data determined that a three-variable model performed optimally but achieved
only a 71% predicted probability when all three variables were present. We
were unable to detect a difference in weight-bearing status, which was an
important predictor in the study by Kocher et al. This is admittedly difficult
to assess in a retrospective study, but it seems that children with an acutely
painful hip cannot or will not bear weight regardless of the etiology of their
condition.
We found several significant differences between our patients with septic
arthritis and those with transient synovitis. In general, the children with
septic arthritis appeared to be more acutely ill with leukocytosis as
reflected in both the serum and the synovial fluid analyses, they had a higher
average erythrocyte sedimentation rate, and higher percentages had a prior
health-care visit and a history of fever. Thus, it seems that septic arthritis
produces a more toxic-appearing clinical picture more quickly than does
transient synovitis.
One interesting finding in our review was the significant effect of a
previous health-care visit on the prediction algorithm. We previously reported
that 33% (twenty-one) of sixty-four children with septic arthritis had been
seen by health-care professionals prior to the definitive diagnosis and 30%
(nineteen) had been treated previously with
antibiotics19. In
the current study, 63% (twenty-nine) of the forty-seven hips in patients with
septic arthritis had been evaluated previously by a health-care provider
compared with 45% (forty-seven) of the 118 hips with transient synovitis.
Considering the other findings in this analysis (elevated erythrocyte
sedimentation rate, fever, and leukocytosis), the effect of a previous
health-care visit may be a measure of the caregiver's opinion of the child's
condition. The more toxic appearing the condition, the more likely the
caregiver is to bring the child in for evaluation of the irritable hip.
The frequency of transient synovitis in our study was much greater than
that in the study by Kocher et
al.2 (72% and 51%,
respectively). Admittedly, our threshold for hip ultrasonography and
arthrocentesis is low, as a result of the known difficulty of diagnosing
septic arthritis clinically and the availability and accuracy of hip
ultrasonography20-22.
Ideally, the 118 patients with transient synovitis in this study would not
have undergone ultrasonography and secondary arthrocentesis if we had been
able to better differentiate between these two disease processes with use of
peripheral serum and/or blood analyses and plain radiographs. Ultimately,
minimizing the pain and distress associated with unnecessary arthrocentesis
and surgical drainage is an important goal of the diagnostic algorithm.
The workup for an irritable hip at our institution consists of a thorough
history and physical examination along with plain radiographs of the hip and
laboratory studies, including a complete blood-cell count with differential,
measurement of the erythrocyte sedimentation rate and C-reactive protein
level, and blood cultures. If there is a clinical suspicion of septic
arthritis of the hip, patients undergo an ultrasound examination of both hips
to identify an effusion. Hip joint arthrocentesis is routinely performed,
typically in the radiology department with arthrographic or ultrasound
confirmation of the intra-articular position of the needle, for all patients
with a documented effusion. Synovial fluid obtained at arthrocentesis is sent
to the laboratory for a cell count and differential, Gram stain, and culture.
The decision regarding treatment is based on all of the collected information.
In general, a white blood-cell count of =50,000/mm3 (=50.0
× 109/L) in the synovial fluid and a positive Gram stain are
the strongest indications to proceed with surgical irrigation and
débridement. However, as evidenced by the eighteen patients who were
excluded from this analysis because of a synovial fluid white blood-cell count
of <50,000/mm3 (<50.0 × 109/L) but still
treated with surgical irrigation and débridement, the decision to
manage an irritable hip surgically is based on the evaluation of all of the
information collected during the diagnostic process.
The use of a validated clinical prediction algorithm for septic arthritis
of the hip should be able to decrease utilization of health-care resources,
minimize the use of painful procedures and unnecessary treatments, and
optimize patient outcome. However, in our patient population, we were unable
to confirm the utility of the clinical prediction algorithm developed by
Kocher et al.2. On
the basis of the low predictive probability of the best model that we could
develop, we will continue to use hip ultrasonography and arthrocentesis as
adjunctive diagnostic modalities in the evaluation of the irritable hip.
Because of the low annual rates of septic arthritis of the hip, the
development of a valid clinical prediction algorithm may best be accomplished
in a prospective, multicenter study. The generalizability of the algorithm
developed by Kocher et al. or of our algorithm is unknown and it is likely
that neither is valid; thus, both should be applied with caution at other
institutions.