- Email: [email protected]
Predictors of malignancy in childhood peripheral lymphadenopathy
Predictors of Malignancy in Childhood Peripheral Lymphadenopathy By Oliver
S. Soldes,
John Ann
Purpose: The aim of this study associated with malignant etiologies lymphadenopathy and to construct to assess the risk of malignancy.
G. Younger, Arbor,
was to identify factors of childhood peripheral a model that may be used
Methods: The medical records of 60 consecutive years old or less who underwent peripheral biopsies were reviewed.
patients 18 lymph node
Results: Increasing node size, number of sites of adenopathy, and age were associated with an increasing risk of malignancy (P< .05 for all variables). Graphs useful for risk determination were constructed based on these variables. Additional factors associated with malignancy included the presence of supraclavicular adenopathy (P < .Ol), an abnormal chest x-ray (PC .Ol), and fixed nodes (P< .Ol). Variables that were not statistically different between patients with benign and malignant adenopathy included the duration of
P
ERIPHERAL LYMPHADENOPATHY is commonly encountered in pediatric surgical practice. An enlarged lymph node in a young child may provoke significant concern in the pediatrician and corresponding anxiety in the parents, even when the probability of malignancy is low. Fear of malignancy is the driving indication for biopsy in the majority of cases. The pediatric surgeon is often called on to allay these fears by performing a biopsy. Although most often benign,1-3 malignant causesof lymphadenopathy are frequent enough to necessitate a significant number of biopsies. In a retrospective review of lymph node biopsy results, Lee et al3 found that 77% of lymph node biopsy findings were benign in children 1 to 10 years of age.3 Furthermore, the therapeutic yield of peripheral lymph node biopsy in children is only 20%.4 Several studies have attempted to identify clinical criteria that predict the etiology of lymphadenopathy and the need for peripheral lymph node biopsy.3-4 If reliable predictors of malignancy could be identified to select patients for biopsy, a large number of unnecessary procedures could be avoided. A survey of the prevalence of adenopathy of the head and neck showed that 44% of children 5 years of age or less seen for well-child visits had palpable lymph nodes. Of children seen for sick visits, 64% had palpable nodes7 Given the prevalence of enlarged palpable nodes in children, identification of predictors of malignancy would provide a significant savings of health care resources and the pediatric surJournalofPediatric
Surgery,
Vol34,
No 10 (October),
1999: pp 1447-1452
and Ronald
B. Hirsch1
Michigan
adenopathy (P = .43), the presence of fever (P = .36), cough (P = .14), splenomegaly (P = .93), skin involvement (P = .39/, tenderness (P = .49), and bilateral adenopathy (P = .39). Fluctuance was associated with benign adenopathy (P < .04). Conclusions: The risk of malignancy increased with increasing size and number of sites of adenopathy and age. Other significant predictors of malignancy included supraclavicular location, an abnormal chest x-ray, and fixed nodes. These data may be used to supplement clinical judgment to predict the risk of malignancy. J Pediatr Srrrg 34:1447-1452. Copyright o 1999 by W3. Saunders Company.
INDEX biopsy.
WORDS:
Lymphadenopathy,
malignancy,
surgery,
geon’s time. In addition, parental anxiety may be allayed, and children could avoid the discomforts of surgery and risks of an anesthetic. The purpose of this study was to identify clinical criteria associated with malignant lymphadenopathy and to develop a model that would predict the risk of malignancy. The experience of the section of pediatric surgery at the University of Michigan with peripheral lymph node biopsy was reviewed. Several preoperative historical, radiological, and physical findings associated with malignant histology in children with peripheral lymphadenopathy were identified. MATERIALS
AND
METHODS
The medical records of 60 consecutive patients age 18 or less who underwent surgical procedures for peripheral lymphadenopathy from January 1991 to December 1993 at the C.S. Mott Children’s Hospital were reviewed. Subsequent procedures on the same patients were
From the Department of Surgery, Sections of Pediatric Surgery and Emergency Medicine, University of Michigan, C.S. Mott Children’s Hospital, Ann Arboc Ml. Presented as a poster at the 29th Annual Meeting of the American Pediatric Surgical Association, Hilton Head, South Carolina, May 10-13, 1998. This work was supported by NIH Training Grant CAO9672-06 [O.S.S]. Address reprint requests to Ronald B. Hirsch& MD, MS, Associate Professor of Surgery, University of Michigan, F3970 C.S. Mott Children S Hospitai, Ann Arbol; MI 48109-0245. Copyrrght 0 1999 by WB. Saunders Company 0022-3468#9/3410-0001$03.00/O 1447
1448
SOLDES,
excluded from the analysis. Data regarding the physical findings, history of illness, laboratory and radiological evaluation, and pathological diagnoses were abstracted from the patient record. Information on physical findings included the location, size (in centimeters), and number of sites of adenopathy. Nodes were characterized by location as cervical, supraclavicular, submandibular, submental, preauricular-parotid, axillary, inguinal, and other sites. The node size and site was based on the examining clinician’s individual estimate of node size and location as described in the patient’s medical record. Bilateral nodes in the same anatomic location (eg, bilateral cervical nodes) were considered to be 2 sites. The presence of fever, cough, tenderness, skin involvement, splenomegaly, and whether the node was fixed were noted. Night sweats and weight loss were often not reported on the clinic record and were omitted from the analysis. Historical data collected included the duration of lymphadenopathy, history of malignancy, immunodeficiency (human innnunodeficiency virus or other cause), transplantation, recent antibiotic therapy, and animal or tuberculosis exposures. Laboratory and radiological data recorded included any preoperative tests obtained during the evaluation before surgical intervention. Some data on preoperative testing (eg, complete blood count, chest x-ray) were not available on all patients, because not all patients had identical laboratory and radiological evaluation before referral and biopsy. Fine-needle aspiration was not used in the diagnosis of adenopathy. Logistic models may be used in the analysis of clinical studies to predict statistical prognosis. Multivariate logistic regression was applied to determine the significant risk factors predicting malignancy in pediatric patients presenting with lymphadenopathy. In this model, the risk factors of age, size, and number of sites of adenopathy were defined as prognostic variables (Xi, X2,. . , XrJ with associated coefficients pi, . , Bk, which were estimated regression coefficients determined from the data. The probability of malignancy, P, was the dependent variable. A multiple regression model was constructed that satisfied the equation: logit(P)=E,+B,X,+B,X,+....+E,X, Adetailed discussion of such methods has been presented elsewhere.* Comparisons between continuous variables were performed using an independent t test, whereas categorical variables were analyzed by x2 or Fisher’s Exact test, as appropriate. All data are presented as mean ? SEM. Statistical analyses were performed using the SAS statistical package (SAS Institute, Guy, NC).
RESULTS
The clinicopathologic diagnoses of the 60 consecutive patients are shown in Table 1. Malignant etiologies of lymphadenopathy occurred in 27% and benign etiologies in 73% of cases. As may be expected, Hodgkin’s disease and non-Hodgkin’s lymphoma were the most common malignant causes of lymphadenopathy (17% and 5% of all cases, respectively). Other metastatic tumors were uncommon (neuroblastoma, n = 2, 3%; alveolar rhabdomyosarcoma, n = 1,2%). Eight of the 60 patients had a history of malignancy before presentation (alveolar rhabdomyosarcoma, n = 2; Hodgkin’s lymphoma, n = 1; non-Hodgkin’s lymphoma, n = 1; Wilms’ tumor, n = 1; neuroblastoma, n = 1; acute lymphocytic leukemia, n = 1; endodermal sinus tumor, n = 1). Two of the 8 patients (25%) with a history of malignancy had a recurrent malignancy at biopsy. Most benign etiologies of lymphadenopathy were
Table 1. Etiology
YOUNGER,
Malignant Hodgkin’s disease Non-Hodgkin’s lymphoma Leukemia Alveolar
(metastatic)
Percent
16
27
1
(metastatic)
2 44
Noninfectious
7
Capillary hemangioma Sinus histiocytosis
1 2
Langerhans call histiocytosis (X) Necrotic tissue w/atypical infiltrate
1 1
lymph node muscle, adipose,
Pyoganic coccus)
bacteria
nerve
and cervical
(Sfaphy/ococcus, avium pus
virus
Probably infectious Reactive hyperplasia Granulomatous Both reactive inflammation
1 1 16
27
Strepto-
intracellulare
Cat scratch disease Toxoplasmosis Epstein-Barr
rib
73 12
6
Mycobacterium Culture-negative
Total
NO. (n = 60)
10 3 0
rhabdomyosarcoma
Normal Normal Infectious
HIRSCHL
of Lymphadenopathy
Diagnosis
Neuroblastoma Benign
AND
4 2 1 1
(EBV) only
inflammation only hyparplasia and granulomatous
2 21 17
35
1 3 60
infectious or probably infectious (62% of all cases). Reactive hyperplasia, with or without granulomatous inflammation, was by far the most common (n = 20, 33%). Pyogenic bacterial (n = 6, 10%: Stu~hylococc~~, n = 5 and Streptococcus, n = 1) and mycobacterium avium intracellulare (MAI, n = 4,7%) infection were the next most common. Other benign, noninfectious etiologies (n = 7) were responsible for the remaining 12% of cases. A goal of the study was to determine which factors were useful for predicting malignancy in the child who presents without an obvious diagnosis for their adenopathy. Children with a history of cancer who present with adenopathy (n = 8 of 60, 13%) almost invariably will undergo lymph node biopsy to rule out recurrence. These patients have an obvious potential cause of their adenopathy that precludes conservative management without biopsy. The presence of fluctuant nodes (n = 7 of 60, 12%) was associated with a benign etiology of the adenopathy (P < .04). None of the patients with palpably fluctuant nodes had malignancy. Children with grossly fluctuant nodes generally have an obviously infectious cause of adenopathy and will undergo aspiration, drainage, or excision of the affected nodes. These patients also pose no dilemma as to the underlying pathological process and the need for treatment. Therefore, we ex-
RISK OF MALIGNANT
1449
LYMPHADENOPATHY
eluded these 15 patients with a history of malignancy or fluctuant nodes who would require a surgical procedure regardless of the other features of the adenopathy. Of the remaining 45 patients, 14 had malignant lymphadenopathy (Hodgkin’s lymphoma, n = 10,22%; non-Hodgkin’s lymphoma, n = 2,4%; metastatic neuroblastoma, n = 2, 4%). These 45 patients formed the basis for our subsequent analysis. Twenty-three (51%) of the patients were boys and 22 (49%) girls. Five (22%) of the boys and 9 (41%) of the girls had malignant adenopathy. The mean age of patients was 7.8 + 0.9 years (range, 2 months to 18 years; Table 2). Several physical findings were useful for discriminating between malignant and benign causes of adenopathy. Children found to have malignant adenopathy were significantly older (mean, 11.7 + 1.5 years v 6.0 + .9, P < .Ol) than those with benign disease (Table 2). Larger size (P < .05) and number of sites (P < .05) of adenopathy were associated with a greater risk of malignancy (Table 3). The location of adenopathy also influenced the risk of cancer, with supraclavicular adenopathy most likely to be malignant (75% v 22% for all other sites, OR = 10.9, 95% confidence interval [CI] 1.8 to 64.6, P < .Ol). However, bilateral adenopathy was not significantly more often malignant than unilateral adenopathy (42% v 28%, OR = 1.8, 95% CI: .5 to 7.3, P = .39). If the nodes were fixed to adjacent tissues, the risk of malignancy was greatly increased (7 1% v 18%, odds ratio [OR] = 11.5,95% CI: 1.7 to 77.2, P < .Ol; Table 4). Several other common findings were not useful in determining malignancy in this group of patients. The benign and malignant node groups did not differ significantly in the duration of adenopathy (17.3 + 4.7 weeks v 12.4 + 4.0 weeks, P = .43) or preoperative antibiotic use (P = .40; Table 2). The presence of fever (P = .36), cough (P = .14), splenomegaly (P = .93), skin involvement (defined as erythema, discoloration, or induration: P = .39), or nodal tenderness (P = .49) were not associated significantly with the occurrence of malignant lymphadenopathy (Table 4). A chest x-ray (CXR) was ordered in 73% of the 45 patients (Table 4). The presence of an abnormal CXR was
Table 2. Comparison
Vanable
Sex, male, No. (%) Age W Duration of adenopathy Preoperative otics,
(wk) antibi-
No. (%)
of Node
Groups
All Patients In = 45)
Benign Nodes In = 31)
Malignant Nodes In = 14)
P Value
23 (51) 7.8 k 0.9
18 (58) 6.0 f 0.9
5 (36) 11.7 + 1.5
.I7 <.Ol
15.9 t 3.5
17.3 k 4.7
12.4 + 4.0
43
29 (64)
21 (68)
8 (57)
.40
Table 3. Risk of Malignancy NO. In = 45)
Varmble
by Clinical
Variable
Percent Malignant
P Value
Size (cm) 51 >I c2
11 11
18 9
>2<3 >3
8 12
38 58
3
33
No. of Sites 1
Unknown
22
23
2 23 Unknown
15 7 1
27 71 0
Unilateral Bilateral Unknown
32 12 1
28 42 0
Cervical Supraclavicular
28 8
32 75
Axillary
10
30
6 7 0
50 14 -
.05
.05
.39
Location
lnguinal Submandibular Submental Preauricular/parotid
1
0
Other
5
40
*Pvalue
versus
all other
<.01*
sites.
strongly correlated with malignancy (69% v 15%, OR = 12.8, 95% CI: 2.3 to 69.9, P < .Ol). A complete blood count (CBC) was ordered in 80% of all patients. No patient had leukopenia (white blood count [WBC] < 4,000). We could not identify a significant difference between benign and malignant groups (P = .20) in the occurrence of leukocytosis (WBC > 12,000). We excluded patients with fluctuant nodes as well as those with a history of malignancy to develop a model that would better reflect the child who presents a diagnostic challenge and could avoid a surgical procedure if a benign etiology could be predicted. Multivariable logistic regression was applied to identify predictors of malignancy in this group of patients. This logistic model identified age, number of sites of adenopathy, and node size as significant predictors of malignancy. Equations describing the relationship between the probability of malignancy, age, and node size (equation 1); probability, age, and numbers of sites of adenopathy (equation 2); and probability, age, size, and number of sites (equation 3) were derived: logit (P) = - 4.26 + 0.18 (age in years)
+ 0.69 (size in cm)
(1)
logit (P) = - 4.25 + 0.22 (age in years)
NOTE. Plus-minus values are means comparison of the benign to the malignant
2 SEM. group.
P values
are
+ 0.80 (number of sites)
(2)
logit (P) = - 5.51 + 0.20 (age in years)
for
+ 0.57 (size in cm) + 0.70 (number of sites)
(3)
1450
SOLDES,
Table 4. Clinical Patients With Variable
Clinical Variable
No. (%) Sign or symptom Fixed
Skin involvement Tender to palpation Radiological/laboratory tests Abnormal chest X-rayt WBC >12,000$ when
presence
tA chest x-ray SA complete
blood
count
% Malignant
28 (80)
18
10 (22)
33 (73)
27
5(11) 3 (7) 6 (14)
60 33 17
40 (89) 42 (93) 38 (86)
28 31 34
0 (0) 0 (0) 0 (0)
7 (16)
75
37 (84)
30
13 (39)
69 55
20 (61) 25
15 32
11 variable
in 33 of 45 patients
(CBC) was obtained
is compared
with absence
in 36 of the 45 patients
6 5
B
.5 3 %
OX
2 3 2
0.4
E
02
0
’ 6
+
>2<3cm
-
>3cm
3
6
9
12
15
18
11.5 (1.7-77.2) 1.9 (.5-7.6)
-
4.0 (.6-27.0) 1.1 (.l-13.4) 0 0
-
of the clinical
40 -
(2) (2)
co1 .36 .I4 .93 .39 .49
.4 (.04-3.6) 1.8 (.3-9.3) 12.8 (2.3-69.9)
-
co1
2.6 (.6-10.9)
.20
variable.
c
-=-Llcm
rlcm
t>lrZcm
0.8-
% Malignant
represents a diagnostic dilemma, without other factors that mandate a surgical intervention (history of malignancy or grossly fluctuant nodes), and could avoid a procedure. The model and graphs derived from it incorporate patient age, number, and size of the nodes to determine the risk of malignancy. Increasing node size, number of sites of adenopathy, and age all correlated with an increasing risk of malignancy. These three clinical characteristics are easily and universally obtainable in all patients at the time of a clinic visit. Other findings, although highly predictive of malignancy, may or may not be present in a given patient or require supplemental laboratory or radiological testing. Additional clinical factors that increased the likelihood of malignancy included the presence of fixed nodes, supraclavicular location, and an abnormal chest x-ray. The presence of fever, cough, splenomegaly, skin involvement, tenderness, and bilateral nodes did not increase the risk that adenopathy was malignant. To use this model, the surgeon must first determine if the child with peripheral lymphadenopathy has a history of malignancy or fluctuant nodes. If not, then the model
DISCUSSION
-a-
1 1
P Value*
(80%).
We have identified several clinical findings useful for predicting the risk of malignancy in children 18 years of age or less with peripheral lymphadenopathy and used these characteristics to construct a risk model. The logistic models used in this report are a standard method of relating a response variable (risk of malignancy, prognosis) to multiple prognostic variables. They have been applied previously to predict the statistical probability of developing breast cancer,g to attempt to identify high-risk groups for breast cancer,‘O and to estimate relative risks in a study of dietary fat and the risk of prostate cancer. l1 Logistic models have in common the identification of prognostic variables (eg, risk factors), and the estimation of coefficients based on the data to determine the mean response to the prognostic variables.8 Our model was particularly concerned with the child who ‘-
HIRSCHL
(73%).
Graphs that allow the prediction of the risk of malignancy for a lymph node were constructed based on equation 3 (P < .Ol by multivariable logistic regression) and are shown in Fig 1.
A
No. (%)
42
of the clinical
was obtained
No. (%)
Odds Ratio (95% Cl)
Unknown
71
12 (7)
AND
and the Risk of Malignancy
Patients Without Variable
% Malignant
7 (20)
Fever Cough Splenomegaly
‘Pvalue
Variables
YOUNGER,
-=-
'1
0.8
d
Age In Years Fig 1. The probability of a malignant histological diagnosis or 3 or more (C) sites of adenopathy (P < .Ol by multivariate largest node.
3
6
9
Age in Years
12
15
18
0
3
6
9
12
15
18
Age in Years
for a peripheral lymph node versus the patient’s age and node size, for 1 (A), 2 (6) logistic regression). The node size refers to the diameter (in centimeters) of the
RISK OF MALIGNANT
LYMPHADENOPATHY
may be used. The surgeon must determine the patient’s age, the largest node size (in centimeters), and the number of sites of adenopathy. These data are plotted on the graphs in Fig 1 to arrive at an estimate of the risk of adenopathy. Extra weight should be assigned to the possibility of malignancy if an abnormal chest x-ray, supraclavicular, or fixed nodes are present. A number of retrospective studies have examined the presenting features, etiologies, and indications for biopsy of childhood peripheral adenopathy.4-6 These studies, while providing useful information on individual risk factors, did not synthesize the data in a way that was useful for determining the risk of malignancy in a particular child. Slap et al6 devised a predictive model using stepwise discriminant analysis to differentiate patients whose biopsy results did lead to treatment from those whose results did not. Their model was applicable for patients 9 to 25 years old and was based on lymph node size; history of ear, nose, and throat (ENT) symptoms; and chest x-rays. They defined patients with normal, hyperplastic, and inflammatory nodes as the group in which biopsy results did not lead to treatment and patients with granulomatous or malignant nodes as those whose results did lead to treatment. Four factors were associated with granuloma or tumor (“treatment group”) at P < .05; abnormal chest x-ray, lymph node size larger than 2.0 cm, a history of night sweats, and a history of weight loss. A history of ENT symptoms was the only variable associated with the absence of granuloma or tumor. However, night sweats and weight loss did not contribute to discrimination between groups and were dropped from their analysis. Their model is derived from many patients at risk for lymphoma not usually seen by the pediatric surgeon (adults age 19 to 25) and omits a group at relatively low risk for lymphoma that comprises a large proportion of children (age 0 to 8 years). Subsequently, its value to the pediatric surgeon may be limited. It should be noted that this is a retrospective analysis with associated limitations. The sample drawn from our study was limited to the referral pattern of the C.S. Mott Children’s Hospital and, therefore, limitations in generalization to other hospitals and practices should be recognized. The C.S. Mott Hospital also is a suburban referral center for pediatric surgical cases for the state of Michigan. As such, it is possible that our sample was biased toward malignant disease. The incidence of malignancy among our patients (14 of 45; 31%) was higher than that of other university centers. The study by Lake and O&i5 study had 13 of 75 (17%) patients with malignant disease. The study of lymph node biopsies by Knight et al4 identified 31 cases of malignant adenopathy out of 239 (11%). The C.S. Mott Hospital may be expected to have a relatively greater concentration.of high-risk (malignant)
1451
adenopathy than the average community setting. Subsequently, it would be intuitive to predict that data based on this experience would show a higher concentration (and risk) of malignant cases. The experience in community practice with patients deemed to have low-risk adenopathy by the model should reflect an even lower true risk of malignancy than at C.S. Mott. Given that a goal of the study was to provide a model that would identify patients who may avoid biopsy, this selection bias should not be expected to diminish the utility of the model. Furthermore, by eliminating patients with a known history of malignancy from the analysis, we also eliminate some of the selection bias toward patients likely to have malignant lymphadenopathy. No cases of tuberculosis were identified in our study. Three of the 14 malignant cases were caused by new metastatic malignancies (neuroblastoma or alveolar rhabdomyosarcoma). The predictors we identified were all statistically associated with malignancy to the P < .05 level. It is possible; however, that the relatively small sample size may have contributed to an (Yerror. Some factors that did not reach statistical significance may have become relevant determinants of risk if the sample size were larger (eg, cough, P = .14). Nonetheless, the factors of age, size, supraclavicular adenopathy, and abnormal chest x-ray that we identified as associated with malignancy also have been identified in other studies.3,4,6The duration of adenopathy, fever, and tenderness have been noted previously to be of little value in the differential diagnosis of peripheral adenopathy.4.5 Our model and the associated graphs are useful for predicting the likelihood of malignant histology and the need for biopsy in children with peripheral lymphadenopathy. Based on this model, the following approach is suggested: older children (~8 years) with larger nodes (> 1 cm), multiple sites of adenopathy, supraclavicular lymph nodes, an abnormal chest x-ray, or fixed nodes should undergo lymph node biopsy without delay. In contrast, young children (<8 years), with a single small node in one site have a low risk of malignancy (15%). They may be managed with further testing (serology, chest x-ray, purified protein derivative skin testing), medical therapy (antibiotics), or with close follow-up as clinical circumstances dictate. To avoid the possibility of delay in the diagnosis of cancer, these data should supplement, but not replace, clinical judgment regarding the risk of malignancy. This report is unique in that it applies rigorous statistical techniques to analyze the risk of malignancy and synthesizes them into a statistical model that is useful to predict the risk of malignancy in a specific patient at the time of an office visit. Previous studies of lymphadenopathy simply identify risk factors or describe the distribution of benign and malignant nodes according to
1452
SOLDES,
individual clinical characteristics without presenting them in a manner that can be used to predict risk in a given patient.3-5 The factors analyzed in this study are clinical data easily determinable in the outpatient setting. This report identifies several factors noted by other investigators, verifying the validity of the data underlying the
YOUNGER,
AND
HIRSCHL
construction of the model. Future studies should focus on collection of data prospectively, on a larger number of patients, in multiple locations to validate and refine the model. Such a strategy would improve the quality of the data, eliminate site-specific bias, and improve the generalizability of the findings.
REFERENCES 1. Zeulzer WW, Kaplan J: The child with lymphadenopathy. Semin Hematol 12:323-334, 1975 2. Ritchie AK: Lymphadenopathy in children: A concise review. WV Med .I 86:402-404,199O 3. Lee YN, Terry R, Lukes RJ: Lymph node biopsy for diagnosis: A statistical study. J Surg Oncol14:53-60,198O 4. Knight PJ, Mulne AF, Vassy LE: When is lymph node biopsy indicated in children with enlarged peripheral nodes? Pediatrics 69:391-3%,1982 5. Lake AM, Oski FA: Peripheral lymphadenopathy in childhood. Am J Dis Child 132:357-359,1978 6. Slap GB, Brooks JSJ, Schwartz JS: When to perform biopsies of enlarged peripheral lymph nodes in young patients. JAMA 252:13211326,1984
7. Herzog LW Prevalence of lymphadenopatby in infants and children. Clin Pediatr 22:485-487,
of the head and neck 1983
8. Armitage P, Gehan EA: Statistical methods for the identification and use of prognostic factors. Jnt J Cancer 13:16-36,1974 9. Farewell VT, Math B, Math M: The combined effect of breast cancer risk factors. Cancer 40:931-936,1977 10. Negri E, Decarli A, La Vecchia C, et al: Identification of high risk groups for breast cancer by means of logistic models. J Clin Epidemiol 43:413-418,199O 11. Giovannucci E, Rimm EB, Colditz GA, et al: A prospective study of dietary fat and risk of prostate cancer. J Nat1 Cancer Inst 85:15711579.1993
S. Soldes,
John Ann
Purpose: The aim of this study associated with malignant etiologies lymphadenopathy and to construct to assess the risk of malignancy.
G. Younger, Arbor,
was to identify factors of childhood peripheral a model that may be used
Methods: The medical records of 60 consecutive years old or less who underwent peripheral biopsies were reviewed.
patients 18 lymph node
Results: Increasing node size, number of sites of adenopathy, and age were associated with an increasing risk of malignancy (P< .05 for all variables). Graphs useful for risk determination were constructed based on these variables. Additional factors associated with malignancy included the presence of supraclavicular adenopathy (P < .Ol), an abnormal chest x-ray (PC .Ol), and fixed nodes (P< .Ol). Variables that were not statistically different between patients with benign and malignant adenopathy included the duration of
P
ERIPHERAL LYMPHADENOPATHY is commonly encountered in pediatric surgical practice. An enlarged lymph node in a young child may provoke significant concern in the pediatrician and corresponding anxiety in the parents, even when the probability of malignancy is low. Fear of malignancy is the driving indication for biopsy in the majority of cases. The pediatric surgeon is often called on to allay these fears by performing a biopsy. Although most often benign,1-3 malignant causesof lymphadenopathy are frequent enough to necessitate a significant number of biopsies. In a retrospective review of lymph node biopsy results, Lee et al3 found that 77% of lymph node biopsy findings were benign in children 1 to 10 years of age.3 Furthermore, the therapeutic yield of peripheral lymph node biopsy in children is only 20%.4 Several studies have attempted to identify clinical criteria that predict the etiology of lymphadenopathy and the need for peripheral lymph node biopsy.3-4 If reliable predictors of malignancy could be identified to select patients for biopsy, a large number of unnecessary procedures could be avoided. A survey of the prevalence of adenopathy of the head and neck showed that 44% of children 5 years of age or less seen for well-child visits had palpable lymph nodes. Of children seen for sick visits, 64% had palpable nodes7 Given the prevalence of enlarged palpable nodes in children, identification of predictors of malignancy would provide a significant savings of health care resources and the pediatric surJournalofPediatric
Surgery,
Vol34,
No 10 (October),
1999: pp 1447-1452
and Ronald
B. Hirsch1
Michigan
adenopathy (P = .43), the presence of fever (P = .36), cough (P = .14), splenomegaly (P = .93), skin involvement (P = .39/, tenderness (P = .49), and bilateral adenopathy (P = .39). Fluctuance was associated with benign adenopathy (P < .04). Conclusions: The risk of malignancy increased with increasing size and number of sites of adenopathy and age. Other significant predictors of malignancy included supraclavicular location, an abnormal chest x-ray, and fixed nodes. These data may be used to supplement clinical judgment to predict the risk of malignancy. J Pediatr Srrrg 34:1447-1452. Copyright o 1999 by W3. Saunders Company.
INDEX biopsy.
WORDS:
Lymphadenopathy,
malignancy,
surgery,
geon’s time. In addition, parental anxiety may be allayed, and children could avoid the discomforts of surgery and risks of an anesthetic. The purpose of this study was to identify clinical criteria associated with malignant lymphadenopathy and to develop a model that would predict the risk of malignancy. The experience of the section of pediatric surgery at the University of Michigan with peripheral lymph node biopsy was reviewed. Several preoperative historical, radiological, and physical findings associated with malignant histology in children with peripheral lymphadenopathy were identified. MATERIALS
AND
METHODS
The medical records of 60 consecutive patients age 18 or less who underwent surgical procedures for peripheral lymphadenopathy from January 1991 to December 1993 at the C.S. Mott Children’s Hospital were reviewed. Subsequent procedures on the same patients were
From the Department of Surgery, Sections of Pediatric Surgery and Emergency Medicine, University of Michigan, C.S. Mott Children’s Hospital, Ann Arboc Ml. Presented as a poster at the 29th Annual Meeting of the American Pediatric Surgical Association, Hilton Head, South Carolina, May 10-13, 1998. This work was supported by NIH Training Grant CAO9672-06 [O.S.S]. Address reprint requests to Ronald B. Hirsch& MD, MS, Associate Professor of Surgery, University of Michigan, F3970 C.S. Mott Children S Hospitai, Ann Arbol; MI 48109-0245. Copyrrght 0 1999 by WB. Saunders Company 0022-3468#9/3410-0001$03.00/O 1447
1448
SOLDES,
excluded from the analysis. Data regarding the physical findings, history of illness, laboratory and radiological evaluation, and pathological diagnoses were abstracted from the patient record. Information on physical findings included the location, size (in centimeters), and number of sites of adenopathy. Nodes were characterized by location as cervical, supraclavicular, submandibular, submental, preauricular-parotid, axillary, inguinal, and other sites. The node size and site was based on the examining clinician’s individual estimate of node size and location as described in the patient’s medical record. Bilateral nodes in the same anatomic location (eg, bilateral cervical nodes) were considered to be 2 sites. The presence of fever, cough, tenderness, skin involvement, splenomegaly, and whether the node was fixed were noted. Night sweats and weight loss were often not reported on the clinic record and were omitted from the analysis. Historical data collected included the duration of lymphadenopathy, history of malignancy, immunodeficiency (human innnunodeficiency virus or other cause), transplantation, recent antibiotic therapy, and animal or tuberculosis exposures. Laboratory and radiological data recorded included any preoperative tests obtained during the evaluation before surgical intervention. Some data on preoperative testing (eg, complete blood count, chest x-ray) were not available on all patients, because not all patients had identical laboratory and radiological evaluation before referral and biopsy. Fine-needle aspiration was not used in the diagnosis of adenopathy. Logistic models may be used in the analysis of clinical studies to predict statistical prognosis. Multivariate logistic regression was applied to determine the significant risk factors predicting malignancy in pediatric patients presenting with lymphadenopathy. In this model, the risk factors of age, size, and number of sites of adenopathy were defined as prognostic variables (Xi, X2,. . , XrJ with associated coefficients pi, . , Bk, which were estimated regression coefficients determined from the data. The probability of malignancy, P, was the dependent variable. A multiple regression model was constructed that satisfied the equation: logit(P)=E,+B,X,+B,X,+....+E,X, Adetailed discussion of such methods has been presented elsewhere.* Comparisons between continuous variables were performed using an independent t test, whereas categorical variables were analyzed by x2 or Fisher’s Exact test, as appropriate. All data are presented as mean ? SEM. Statistical analyses were performed using the SAS statistical package (SAS Institute, Guy, NC).
RESULTS
The clinicopathologic diagnoses of the 60 consecutive patients are shown in Table 1. Malignant etiologies of lymphadenopathy occurred in 27% and benign etiologies in 73% of cases. As may be expected, Hodgkin’s disease and non-Hodgkin’s lymphoma were the most common malignant causes of lymphadenopathy (17% and 5% of all cases, respectively). Other metastatic tumors were uncommon (neuroblastoma, n = 2, 3%; alveolar rhabdomyosarcoma, n = 1,2%). Eight of the 60 patients had a history of malignancy before presentation (alveolar rhabdomyosarcoma, n = 2; Hodgkin’s lymphoma, n = 1; non-Hodgkin’s lymphoma, n = 1; Wilms’ tumor, n = 1; neuroblastoma, n = 1; acute lymphocytic leukemia, n = 1; endodermal sinus tumor, n = 1). Two of the 8 patients (25%) with a history of malignancy had a recurrent malignancy at biopsy. Most benign etiologies of lymphadenopathy were
Table 1. Etiology
YOUNGER,
Malignant Hodgkin’s disease Non-Hodgkin’s lymphoma Leukemia Alveolar
(metastatic)
Percent
16
27
1
(metastatic)
2 44
Noninfectious
7
Capillary hemangioma Sinus histiocytosis
1 2
Langerhans call histiocytosis (X) Necrotic tissue w/atypical infiltrate
1 1
lymph node muscle, adipose,
Pyoganic coccus)
bacteria
nerve
and cervical
(Sfaphy/ococcus, avium pus
virus
Probably infectious Reactive hyperplasia Granulomatous Both reactive inflammation
1 1 16
27
Strepto-
intracellulare
Cat scratch disease Toxoplasmosis Epstein-Barr
rib
73 12
6
Mycobacterium Culture-negative
Total
NO. (n = 60)
10 3 0
rhabdomyosarcoma
Normal Normal Infectious
HIRSCHL
of Lymphadenopathy
Diagnosis
Neuroblastoma Benign
AND
4 2 1 1
(EBV) only
inflammation only hyparplasia and granulomatous
2 21 17
35
1 3 60
infectious or probably infectious (62% of all cases). Reactive hyperplasia, with or without granulomatous inflammation, was by far the most common (n = 20, 33%). Pyogenic bacterial (n = 6, 10%: Stu~hylococc~~, n = 5 and Streptococcus, n = 1) and mycobacterium avium intracellulare (MAI, n = 4,7%) infection were the next most common. Other benign, noninfectious etiologies (n = 7) were responsible for the remaining 12% of cases. A goal of the study was to determine which factors were useful for predicting malignancy in the child who presents without an obvious diagnosis for their adenopathy. Children with a history of cancer who present with adenopathy (n = 8 of 60, 13%) almost invariably will undergo lymph node biopsy to rule out recurrence. These patients have an obvious potential cause of their adenopathy that precludes conservative management without biopsy. The presence of fluctuant nodes (n = 7 of 60, 12%) was associated with a benign etiology of the adenopathy (P < .04). None of the patients with palpably fluctuant nodes had malignancy. Children with grossly fluctuant nodes generally have an obviously infectious cause of adenopathy and will undergo aspiration, drainage, or excision of the affected nodes. These patients also pose no dilemma as to the underlying pathological process and the need for treatment. Therefore, we ex-
RISK OF MALIGNANT
1449
LYMPHADENOPATHY
eluded these 15 patients with a history of malignancy or fluctuant nodes who would require a surgical procedure regardless of the other features of the adenopathy. Of the remaining 45 patients, 14 had malignant lymphadenopathy (Hodgkin’s lymphoma, n = 10,22%; non-Hodgkin’s lymphoma, n = 2,4%; metastatic neuroblastoma, n = 2, 4%). These 45 patients formed the basis for our subsequent analysis. Twenty-three (51%) of the patients were boys and 22 (49%) girls. Five (22%) of the boys and 9 (41%) of the girls had malignant adenopathy. The mean age of patients was 7.8 + 0.9 years (range, 2 months to 18 years; Table 2). Several physical findings were useful for discriminating between malignant and benign causes of adenopathy. Children found to have malignant adenopathy were significantly older (mean, 11.7 + 1.5 years v 6.0 + .9, P < .Ol) than those with benign disease (Table 2). Larger size (P < .05) and number of sites (P < .05) of adenopathy were associated with a greater risk of malignancy (Table 3). The location of adenopathy also influenced the risk of cancer, with supraclavicular adenopathy most likely to be malignant (75% v 22% for all other sites, OR = 10.9, 95% confidence interval [CI] 1.8 to 64.6, P < .Ol). However, bilateral adenopathy was not significantly more often malignant than unilateral adenopathy (42% v 28%, OR = 1.8, 95% CI: .5 to 7.3, P = .39). If the nodes were fixed to adjacent tissues, the risk of malignancy was greatly increased (7 1% v 18%, odds ratio [OR] = 11.5,95% CI: 1.7 to 77.2, P < .Ol; Table 4). Several other common findings were not useful in determining malignancy in this group of patients. The benign and malignant node groups did not differ significantly in the duration of adenopathy (17.3 + 4.7 weeks v 12.4 + 4.0 weeks, P = .43) or preoperative antibiotic use (P = .40; Table 2). The presence of fever (P = .36), cough (P = .14), splenomegaly (P = .93), skin involvement (defined as erythema, discoloration, or induration: P = .39), or nodal tenderness (P = .49) were not associated significantly with the occurrence of malignant lymphadenopathy (Table 4). A chest x-ray (CXR) was ordered in 73% of the 45 patients (Table 4). The presence of an abnormal CXR was
Table 2. Comparison
Vanable
Sex, male, No. (%) Age W Duration of adenopathy Preoperative otics,
(wk) antibi-
No. (%)
of Node
Groups
All Patients In = 45)
Benign Nodes In = 31)
Malignant Nodes In = 14)
P Value
23 (51) 7.8 k 0.9
18 (58) 6.0 f 0.9
5 (36) 11.7 + 1.5
.I7 <.Ol
15.9 t 3.5
17.3 k 4.7
12.4 + 4.0
43
29 (64)
21 (68)
8 (57)
.40
Table 3. Risk of Malignancy NO. In = 45)
Varmble
by Clinical
Variable
Percent Malignant
P Value
Size (cm) 51 >I c2
11 11
18 9
>2<3 >3
8 12
38 58
3
33
No. of Sites 1
Unknown
22
23
2 23 Unknown
15 7 1
27 71 0
Unilateral Bilateral Unknown
32 12 1
28 42 0
Cervical Supraclavicular
28 8
32 75
Axillary
10
30
6 7 0
50 14 -
.05
.05
.39
Location
lnguinal Submandibular Submental Preauricular/parotid
1
0
Other
5
40
*Pvalue
versus
all other
<.01*
sites.
strongly correlated with malignancy (69% v 15%, OR = 12.8, 95% CI: 2.3 to 69.9, P < .Ol). A complete blood count (CBC) was ordered in 80% of all patients. No patient had leukopenia (white blood count [WBC] < 4,000). We could not identify a significant difference between benign and malignant groups (P = .20) in the occurrence of leukocytosis (WBC > 12,000). We excluded patients with fluctuant nodes as well as those with a history of malignancy to develop a model that would better reflect the child who presents a diagnostic challenge and could avoid a surgical procedure if a benign etiology could be predicted. Multivariable logistic regression was applied to identify predictors of malignancy in this group of patients. This logistic model identified age, number of sites of adenopathy, and node size as significant predictors of malignancy. Equations describing the relationship between the probability of malignancy, age, and node size (equation 1); probability, age, and numbers of sites of adenopathy (equation 2); and probability, age, size, and number of sites (equation 3) were derived: logit (P) = - 4.26 + 0.18 (age in years)
+ 0.69 (size in cm)
(1)
logit (P) = - 4.25 + 0.22 (age in years)
NOTE. Plus-minus values are means comparison of the benign to the malignant
2 SEM. group.
P values
are
+ 0.80 (number of sites)
(2)
logit (P) = - 5.51 + 0.20 (age in years)
for
+ 0.57 (size in cm) + 0.70 (number of sites)
(3)
1450
SOLDES,
Table 4. Clinical Patients With Variable
Clinical Variable
No. (%) Sign or symptom Fixed
Skin involvement Tender to palpation Radiological/laboratory tests Abnormal chest X-rayt WBC >12,000$ when
presence
tA chest x-ray SA complete
blood
count
% Malignant
28 (80)
18
10 (22)
33 (73)
27
5(11) 3 (7) 6 (14)
60 33 17
40 (89) 42 (93) 38 (86)
28 31 34
0 (0) 0 (0) 0 (0)
7 (16)
75
37 (84)
30
13 (39)
69 55
20 (61) 25
15 32
11 variable
in 33 of 45 patients
(CBC) was obtained
is compared
with absence
in 36 of the 45 patients
6 5
B
.5 3 %
OX
2 3 2
0.4
E
02
0
’ 6
+
>2<3cm
-
>3cm
3
6
9
12
15
18
11.5 (1.7-77.2) 1.9 (.5-7.6)
-
4.0 (.6-27.0) 1.1 (.l-13.4) 0 0
-
of the clinical
40 -
(2) (2)
co1 .36 .I4 .93 .39 .49
.4 (.04-3.6) 1.8 (.3-9.3) 12.8 (2.3-69.9)
-
co1
2.6 (.6-10.9)
.20
variable.
c
-=-Llcm
rlcm
t>lrZcm
0.8-
% Malignant
represents a diagnostic dilemma, without other factors that mandate a surgical intervention (history of malignancy or grossly fluctuant nodes), and could avoid a procedure. The model and graphs derived from it incorporate patient age, number, and size of the nodes to determine the risk of malignancy. Increasing node size, number of sites of adenopathy, and age all correlated with an increasing risk of malignancy. These three clinical characteristics are easily and universally obtainable in all patients at the time of a clinic visit. Other findings, although highly predictive of malignancy, may or may not be present in a given patient or require supplemental laboratory or radiological testing. Additional clinical factors that increased the likelihood of malignancy included the presence of fixed nodes, supraclavicular location, and an abnormal chest x-ray. The presence of fever, cough, splenomegaly, skin involvement, tenderness, and bilateral nodes did not increase the risk that adenopathy was malignant. To use this model, the surgeon must first determine if the child with peripheral lymphadenopathy has a history of malignancy or fluctuant nodes. If not, then the model
DISCUSSION
-a-
1 1
P Value*
(80%).
We have identified several clinical findings useful for predicting the risk of malignancy in children 18 years of age or less with peripheral lymphadenopathy and used these characteristics to construct a risk model. The logistic models used in this report are a standard method of relating a response variable (risk of malignancy, prognosis) to multiple prognostic variables. They have been applied previously to predict the statistical probability of developing breast cancer,g to attempt to identify high-risk groups for breast cancer,‘O and to estimate relative risks in a study of dietary fat and the risk of prostate cancer. l1 Logistic models have in common the identification of prognostic variables (eg, risk factors), and the estimation of coefficients based on the data to determine the mean response to the prognostic variables.8 Our model was particularly concerned with the child who ‘-
HIRSCHL
(73%).
Graphs that allow the prediction of the risk of malignancy for a lymph node were constructed based on equation 3 (P < .Ol by multivariable logistic regression) and are shown in Fig 1.
A
No. (%)
42
of the clinical
was obtained
No. (%)
Odds Ratio (95% Cl)
Unknown
71
12 (7)
AND
and the Risk of Malignancy
Patients Without Variable
% Malignant
7 (20)
Fever Cough Splenomegaly
‘Pvalue
Variables
YOUNGER,
-=-
'1
0.8
d
Age In Years Fig 1. The probability of a malignant histological diagnosis or 3 or more (C) sites of adenopathy (P < .Ol by multivariate largest node.
3
6
9
Age in Years
12
15
18
0
3
6
9
12
15
18
Age in Years
for a peripheral lymph node versus the patient’s age and node size, for 1 (A), 2 (6) logistic regression). The node size refers to the diameter (in centimeters) of the
RISK OF MALIGNANT
LYMPHADENOPATHY
may be used. The surgeon must determine the patient’s age, the largest node size (in centimeters), and the number of sites of adenopathy. These data are plotted on the graphs in Fig 1 to arrive at an estimate of the risk of adenopathy. Extra weight should be assigned to the possibility of malignancy if an abnormal chest x-ray, supraclavicular, or fixed nodes are present. A number of retrospective studies have examined the presenting features, etiologies, and indications for biopsy of childhood peripheral adenopathy.4-6 These studies, while providing useful information on individual risk factors, did not synthesize the data in a way that was useful for determining the risk of malignancy in a particular child. Slap et al6 devised a predictive model using stepwise discriminant analysis to differentiate patients whose biopsy results did lead to treatment from those whose results did not. Their model was applicable for patients 9 to 25 years old and was based on lymph node size; history of ear, nose, and throat (ENT) symptoms; and chest x-rays. They defined patients with normal, hyperplastic, and inflammatory nodes as the group in which biopsy results did not lead to treatment and patients with granulomatous or malignant nodes as those whose results did lead to treatment. Four factors were associated with granuloma or tumor (“treatment group”) at P < .05; abnormal chest x-ray, lymph node size larger than 2.0 cm, a history of night sweats, and a history of weight loss. A history of ENT symptoms was the only variable associated with the absence of granuloma or tumor. However, night sweats and weight loss did not contribute to discrimination between groups and were dropped from their analysis. Their model is derived from many patients at risk for lymphoma not usually seen by the pediatric surgeon (adults age 19 to 25) and omits a group at relatively low risk for lymphoma that comprises a large proportion of children (age 0 to 8 years). Subsequently, its value to the pediatric surgeon may be limited. It should be noted that this is a retrospective analysis with associated limitations. The sample drawn from our study was limited to the referral pattern of the C.S. Mott Children’s Hospital and, therefore, limitations in generalization to other hospitals and practices should be recognized. The C.S. Mott Hospital also is a suburban referral center for pediatric surgical cases for the state of Michigan. As such, it is possible that our sample was biased toward malignant disease. The incidence of malignancy among our patients (14 of 45; 31%) was higher than that of other university centers. The study by Lake and O&i5 study had 13 of 75 (17%) patients with malignant disease. The study of lymph node biopsies by Knight et al4 identified 31 cases of malignant adenopathy out of 239 (11%). The C.S. Mott Hospital may be expected to have a relatively greater concentration.of high-risk (malignant)
1451
adenopathy than the average community setting. Subsequently, it would be intuitive to predict that data based on this experience would show a higher concentration (and risk) of malignant cases. The experience in community practice with patients deemed to have low-risk adenopathy by the model should reflect an even lower true risk of malignancy than at C.S. Mott. Given that a goal of the study was to provide a model that would identify patients who may avoid biopsy, this selection bias should not be expected to diminish the utility of the model. Furthermore, by eliminating patients with a known history of malignancy from the analysis, we also eliminate some of the selection bias toward patients likely to have malignant lymphadenopathy. No cases of tuberculosis were identified in our study. Three of the 14 malignant cases were caused by new metastatic malignancies (neuroblastoma or alveolar rhabdomyosarcoma). The predictors we identified were all statistically associated with malignancy to the P < .05 level. It is possible; however, that the relatively small sample size may have contributed to an (Yerror. Some factors that did not reach statistical significance may have become relevant determinants of risk if the sample size were larger (eg, cough, P = .14). Nonetheless, the factors of age, size, supraclavicular adenopathy, and abnormal chest x-ray that we identified as associated with malignancy also have been identified in other studies.3,4,6The duration of adenopathy, fever, and tenderness have been noted previously to be of little value in the differential diagnosis of peripheral adenopathy.4.5 Our model and the associated graphs are useful for predicting the likelihood of malignant histology and the need for biopsy in children with peripheral lymphadenopathy. Based on this model, the following approach is suggested: older children (~8 years) with larger nodes (> 1 cm), multiple sites of adenopathy, supraclavicular lymph nodes, an abnormal chest x-ray, or fixed nodes should undergo lymph node biopsy without delay. In contrast, young children (<8 years), with a single small node in one site have a low risk of malignancy (15%). They may be managed with further testing (serology, chest x-ray, purified protein derivative skin testing), medical therapy (antibiotics), or with close follow-up as clinical circumstances dictate. To avoid the possibility of delay in the diagnosis of cancer, these data should supplement, but not replace, clinical judgment regarding the risk of malignancy. This report is unique in that it applies rigorous statistical techniques to analyze the risk of malignancy and synthesizes them into a statistical model that is useful to predict the risk of malignancy in a specific patient at the time of an office visit. Previous studies of lymphadenopathy simply identify risk factors or describe the distribution of benign and malignant nodes according to
1452
SOLDES,
individual clinical characteristics without presenting them in a manner that can be used to predict risk in a given patient.3-5 The factors analyzed in this study are clinical data easily determinable in the outpatient setting. This report identifies several factors noted by other investigators, verifying the validity of the data underlying the
YOUNGER,
AND
HIRSCHL
construction of the model. Future studies should focus on collection of data prospectively, on a larger number of patients, in multiple locations to validate and refine the model. Such a strategy would improve the quality of the data, eliminate site-specific bias, and improve the generalizability of the findings.
REFERENCES 1. Zeulzer WW, Kaplan J: The child with lymphadenopathy. Semin Hematol 12:323-334, 1975 2. Ritchie AK: Lymphadenopathy in children: A concise review. WV Med .I 86:402-404,199O 3. Lee YN, Terry R, Lukes RJ: Lymph node biopsy for diagnosis: A statistical study. J Surg Oncol14:53-60,198O 4. Knight PJ, Mulne AF, Vassy LE: When is lymph node biopsy indicated in children with enlarged peripheral nodes? Pediatrics 69:391-3%,1982 5. Lake AM, Oski FA: Peripheral lymphadenopathy in childhood. Am J Dis Child 132:357-359,1978 6. Slap GB, Brooks JSJ, Schwartz JS: When to perform biopsies of enlarged peripheral lymph nodes in young patients. JAMA 252:13211326,1984
7. Herzog LW Prevalence of lymphadenopatby in infants and children. Clin Pediatr 22:485-487,
of the head and neck 1983
8. Armitage P, Gehan EA: Statistical methods for the identification and use of prognostic factors. Jnt J Cancer 13:16-36,1974 9. Farewell VT, Math B, Math M: The combined effect of breast cancer risk factors. Cancer 40:931-936,1977 10. Negri E, Decarli A, La Vecchia C, et al: Identification of high risk groups for breast cancer by means of logistic models. J Clin Epidemiol 43:413-418,199O 11. Giovannucci E, Rimm EB, Colditz GA, et al: A prospective study of dietary fat and risk of prostate cancer. J Nat1 Cancer Inst 85:15711579.1993