the hematology profile, the absolute eosinophil count, the clinical situation responsible for the hematologic profile determination, and the probable cause of ...
Eosinophilia Detected by Automated Blood Cell Counting in Ambulatory North American Outpatients Incidence and Clinical Significance Malcolm Brigden, MD, FRCPC, Christine Graydon, RT
• Objectives.-To audit a cohort of ambulatory outpatients with eosinophilia detected on automated blood cell counting. Specific objectives included the determination of whether the eosinophilia had been anticipated, the etiology of the eosinophilia, the clinical follow-up and investigations performed on patients with eosinophilia, and the effect of the detection of eosinophilia on patient management and ultimate clinical outcome. Design.-A year-long retrospective review of all patients with an absolute eosinophil count of greater than 0.7 x 1 09 / L. Setting.-A large outpatient laboratory system. The patient population was managed by family physicians and specialists. lntervention.-Data collection included the results of the hematology profile, the absolute eosinophil count, the clinical situation responsible for the hematologic profile determination, and the probable cause of eosinophilia. Individual physicians were surveyed to determine if discovery of the eosinophilia had changed patient management plan or clinical outcome. Principal Results.-Out of 195 300 patients who had a hematology profile performed, 225 were found to have an ab-
osinophilia has been defined as either an eosinophil accumulation of more than 5% in the circulating white blood cells (WBCs) or as an absolute eosinophil count of greater than 0.7 X 109 I L. 1- 3 Relatively little has been published in modem literature on the incidence, etiology, and clinical significance of eosinophilia in outpatients. Today's sophisticated hematologic instruments provide automated five-part WBC differentials, which are now identifying many new cases of eosinophilia.4•5 Under these circumstances, an eosinophilia is frequently not anticipated by the physician responsible for ordering the hematology profile. The extent to which patients with mild to moderate eosinophilia should be investigated remains controversial. We elected to follow up on an earlier limited study of patients with unsuspected eosinophilia and a normal hematology profile by including all patients with eosino-
E
Accepted for publication March 10, 1997. From the Medical Oncology Clinic, Penticton Regional Hospital, British Columbia, Canada. Reprint requests to Medical Oncology Clinic, Penticton Regional Hospital, 550 Carmi Ave, Penticton, British Columbia, Canada V2A 3G6 (Dr Brigden). Arch Pathol Lab Med-Vol 12 1, September 1997
solute eosinophilia count higher than 0.7 x 109 / l. The overal incidence of eosinophilia in the study population was 0.1 % The eosinophilia was not anticipated in 85% of patients. N< obvious cause was detected for the eosinophilia in 36% o patients. Various allergic diseases were responsible for th1 eosinophilia in the majority of the remaining patients. Fewe1 than 9% of individuals manifested a serious systemic illnes! or parasitemia. Further clinical follow-up had been per formed in 69% of patients. Additional laboratory tests ha< been ordered in 59% of patients. The laboratory tests mos frequently ordered were a repeat hematology profile or stoo examinations for ova and parasites. In only two instances di< the discovery of the eosinophilia appear to result in a signif icant change in patient management or ultimate clinical income. Conclusion.-The vast majority of eosinophilias detectec in ambulatory outpatients are associated with allergic processes. An extensive investigation of eosinophilia in ambulatory North American outpatients does not appear tc be warranted unless specifically indicated by the results 0 1 the history and physical examination. (Arch Pathol Lab Med. 1997;121 :963-967)
philia detected as part of an instrument-generated five· part WBC differential. The overall purpose of this inves· tigation was threefold: first, to document the range of eo· sinophilia noted in ambulatory outpatients; second, to ascertain the etiologic conditions responsible f01 eosinophilia in an ambulatory outpatient population; fi· nally, based on the etiologies identified and the ensuing investigations actually performed, to develop appropriate recommendations for the investigation of eosinophilia in similar patient populations. MATERIALS AND METHODS Metro-McNair Clinical Laboratories (Island Division) is a busy outpatient facility servicing over 500000 people on Vancouver Is· land, British Columbia, Canada. Over a 12-month period, follow· up was completed on all patients w hose automated five-part dif· ferential revealed an absolute eosinophilia. Res ults were obtained on the Sysmex NE-8000 (TOA Medical, Kobe, Japan) used to determine hemoglobin, hematocrit, red blood cell count (RBC), WBC count, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, and automated WBC differential and platelet count. All s pecimens were processed w ithin 4 hours of venipuncture. Patients were classified as having eosinophilia w hen the absolute eosinophil count gen-
Kange Patient age, y Absolute eosinophil count, x 1 0 9/L
1- 103
Mean 44
Male, 122
Patient sex, n
43
1.7
0 .8-11 .9
1.4
Female, 103
!rated by machine exceeded 0.7 x 100 I L. Instrument-generated tbsolute eosinophil counts were confirmed by manual slide reriew. The ordering physician was s ubsequently contacted rega rding !ach patient in whom an eosinophilia had been detected. In each :ase, initia l contact was made at least 4 weeks after the detection >f the eosino philia so that individual physician a pproaches to !OSinophilia would not be influenced by the interview process. 'atient data were reviewed to try to pinpoint the clinical situation ·esponsible for the hematologic profile determination and the >robable cause of the eosinophilia. Specific physician questions md examination of patient records were used to determine the !xistence and extent of drug consumption; historical evidence of tllergic d isease, allergic rhinitis, eczema, dermatitis, or other alergic skin disease; asthma; parasitic disease; cancer; collagen dis!ase; primary blood disease; or pathologic pulmonary conditions. f patients had the results of investigations pending, the individJa! physicians were recontacted at a later d ate to determine the 1ltimate outcome of these studies. In cases where the eosino>hitia had been unsuspected, physicians were also surveyed to ind out if discovery of the eosinophilia had changed patient nanagement or final clinical outcome.
RESULTS Over the 12-month period, 195 300 patients had a henatology profile performed. Absolute eosinophilia was di1gnosed in 225 of these patients (Table 1), resulting in an werall incidence of eosinophilia of 0.1%. The male-female :atio was approximately equal, with an overall mean pa:ient age of 44 years. The distribution of the absolute eo;inophil counts for the entire patient population is illus:rated in the Figure. Expressed as an absolute count, the ~osinophil count ranged from 0.8 to 11.9 X 109 /L with a nean count of 1.7 X 109 I L. Seven percent of patients had m absolute eosinophil count greater than 2.5 x 109 / L. In n% of the 225 patients, the proportion of eosinophils in :he differential was 20% or greater.
140 120 "'100
c 2 .5
10 9/L
The distribution of eosinophil count in the study population.
)64 Arch Pathoi Lab Med-Vol 121 , September 1997
Etiologic Factor Unknown Seasonal allergy/allergic rhi nitis Asthma Eczema/dermatitis Cancer Drug allergy Collagen disease Parasitic disease
No. of Patients (n = 225)
Percent of Total
80 66 31 21 7
36 29 14 9 4
7 8 5
3 3 2
Table 3.-Physician Follo w-up in Patients Wit h Eosinophilia No. of Patients (n = 225)
Percent of Total
Clinical follow-up Repeat office visit None
156 69
69 31
laboratory follow-up Further laboratory testing None
132 93
59 41
118
52
46 14 11 7 5
20 6 5 3 3
Follow-up/ Action
Individua l tests ordered Repeat hematologic profile Stool examinations fo r ova and parasites Allergic skin testing Chest radiograph Throat culture Urinalysis
The eosinophilia was not anticipated in 80% of study patients. The 20% of patients in whom the eosinophilia was anticipated virtually all had either allergic rhinitis/ seasonal allergy, asthma, or eczema. A list of identified etiologic causes for the entire patient population is provided in Table 2. In 36% of patients the responsible physician was not able to define a specific medical condition as a causal factor for the eosinophilia. Allergic processes accounted for the majority of patients in whom a cause for eosinophilia was apparent. In only 20 of the 225 patients was there an underlying malignancy, collagen disease, or parasitic infection identified. The eight instances of underlying collagen vascular disease were all known cases and consisted of three cases of rheumatoid arthritis and one case each of systemic lupus, mixed connective tissue disease, polyarteritis nodosa, Hashimotds thyroiditis, and ankylosing spondylitis. The seven cases with underlying cancer were known solid tumors with metastatic disease. The seven presumed drug reactions included two cases of gold therapy, two cases involving nitrofurantoin, as well as one case each involving cephalothin, tegretol, and valproic acid. Each of the presumed drug reaction patients manifested a variety of systemic symptoms (rash, fever, arthralgias), except in the instance of the two individuals receiving gold therapy. The most frequent physician-ordered follow-up investigations involved in the 225 cases are outlined in Table 3. Further clinical follow-up consisting of a repeat office visit was performed in 69% of patients. Subsequent laboratory testing was performed in 59% of patients. A repeat heEosinophilia Detected by Automated Cell Counting--Brigden & Graydon
ur oom compnsea the most trequently performed followup laboratory procedures. Only a few patients had other testing, such as immunoglobulin E or immunoglobulin assays, antinuclear antibody determinations, or serum protein electrophoresis. In no instance was a bone marrow test performed. In three of the cases of parasitemia and three of the presumed drug reactions, detection of the eosinophilia helped to focus on the diagnosis. In only two cases, both involving gold therapy, did knowledge of the eosinophilia result in different patient management. In both instances, the gold therapy was discontinued. Only in these two cases did provision of information regarding the presence of unsuspected eosinophilia appear to have changed the ultimate clinical outcome.
COMMENT The widespread use of automated hematology cell counts has m ade the detection of absolute eosinophilia relatively commonplace.'~-7 Under these circumstances, the discovery of an absolute eosinophilia in an outpatient frequently represents an unanticipated finding.+-6 Although many basic science investigations regarding eosinophil structure and function have been published in the last few years, there have been relatively few studies regarding the etiology and clinical significance of eosinophilia.s-10 The appropriate degree of investigation in cases of eosinophilia, especially in outpatients, remains controversial.4·7·11 •12 The precise absolute eosinophil count at which clinical concern is merited has also been debated.3·7 Lukens 13 and others have attempted to approach this problem by assigning clinical significance to arbitrary levels of eosinophilia. Mild eosinophilia has been considered as an absolute eosinophil count of greater than 0.7 but less than 1.0 X 109I L, moderate eosinophilia as a count of greater than 1.0 but less than 5.0 X 109I L. Exaggerated eosinophilia has been defined as an absolute eosinophil count of greater than 5.0 X 109 IL. Relative to these ranges, most of the patients in the present investigation fell in the moderate eosinophilia category, as the mean eosinophil count in the study was 1.7 X 109I L. There have been relatively few clinical studies involving large numbers of patients that have investigated underlying etiologies responsible for the eosinophilia. In an investigation performed at the Mayo Clinic, up to 40% of patients with eosinophilia had parasitemia, cancer, collagen vascular disease, or other serious systemic illness.3 However, the eosinophilia observed in the patients reported in the Mayo Clinic series was discovered on the basis of a manual differential, which at the time was more likely to be ordered on patients who were known to be seriously ill. In an earlier study, we postulated that mild to moderate eosinophilia associated with an otherwise normal hematology profile discovered by automated WBC counting would probably differ in etiology from eosinophilia reported in earlier studies detected by manual differential.4 This hypothesis is further substantiated by the results of our larger investigation, inasmuch as 36% of study patients had no obvious cause for eosinophilia, while fewer than 9% manifested a serious systemic illness or parasitemia. The largest patient group with an established etiology in the study consisted of patients with seasonal allergies or allergic rhinitis. Not surprisingly, an allergic etiology was also predominantly implicated in the Arch Pathol Lab Med-Vol 121. Seotember 1qq7
rn
ophilia was- anticipated. a· much smaller investigatio of outpatients in Singapore, an allergic process was als found to be the most common identifiable cause of eosii ophilia.14 Some authors have postulated that apart from the a: sociated presence of underlying allergic disease or par< sitemia, the finding of a moderate or high eosinoph count is unusual.6•13 This observation is not consistent wit our data, since over 90% of patients fell into the moderat to high eosinophil category, but a significant number sti had no obvious etiology for the eosinophilia. It is interesting to speculate on the possible physiologi m echanism underlying eosinophilia in the group of p< tients for whom referring physicians could not identify a obvious cause. Studies of an urban North American pof ulation have shown that there is a significant tendency fc the absolute eosinophil count to show a seasonal variatior with highest counts occurring during the months corr monly associated with allergen release.15 Thus, a mild ur suspected response to seasonal allergens may have bee: responsible for the eosinophilia in the m ajority of this pz tient group. Smoking has also been postulated as a po~ sible cause of eosinophilia, but there appears to have bee: no follow-up studies of the initial clinical observation. 16 Considerable debate exists in the literature regardin: the optimum workup for patients with eosinophilia. Var ious authors have advised an organized approach to dat gathering, including a complete history and physical ex amination followed by an extensive laboratory evaluatio1 to thoroughly investigate possible causes. Various labora tory tests that have been recommended include a complet hematology profile; renal and liver functions studie~ erythrocyte sediinentation rate; urinalysis; stool exami nation for ova and parasites; chest radiograph; antinuclea antibodies; rheumatoid factor; serum immunoglobulin de terminations, including immunoglobulin E, radioallergo sorbent, or allergic skin testing; and a possible bone mar row examination. The recommendations for such a sys tematic and extensive workup have been based, in part on earlier reports in the literature that suggested up t< 40% of patients with eosinophilia will be found to have • serious underlying condition, such as malignancy, lym phoproliferative disorder, collagen disease, or parasitic in fection.J.7.12.t3 Although it is true that a variety of serious disease state: may be associated with eosinophilia, the majority of pa tients with these illnesses will have other systemic sign: and symptoms, physical findings, or abnormalities in thei history that would prompt further investigation. Radiatior has been linked to eosinophilia, but this exposure shoulc be elicited by history. 17 The various dermatologic abnor malities, including scabies, that may be associated witl eosinophilia should be detected during the history an< physical examination.18·19 Medication reactions have beer noted to be a frequent cause of mild to moderate eosino philia20; however, as in the current study, most adversE drug reactions associated with eosinophilia are accompa· nied by systemic m anifestations, such as fever, skin rash adenopathy, or signs of hepatic, renal, or cardiac toxicity which would prompt further investigation. An exceptior might occur in cases in which the absolute eosinophi count is specifically monitored as a possible indicator 0 1 gold therapy toxicity. When drug-induced eosinophilia i~ suspected on clinical grounds, cessation of the offendin~
A wide variety of solid tumors, predominately carcinomas, arising in the lung, head, neck, uterus, breast, pancreas, colon and rectum, thyroid gland, adrenal glands, and biliary tract have been associated with eosinophilia.21 However, as was the case in the current study, tumor-associated eosinophilia is usually an indication of late-stage disease and is frequently associated with widespread metastasis. Acute eosinophilic leukemia is characterized by increased numbers of immature eosinophils in the blood and bone marrow, as well as anemia and thrombocytopenia.22.23 Eosinophilia has been noted in association with a variety of other primary hematologic diseases, such as acute lymphoblastic and myelomonocytic leukemia, myeloma, adult T-cellleukemia, Hodgkin's and non-Hodgkins lymphoma, systemic mast cell disease, Kimura's disease, and the myelodysplastic syndromes. These patients have other abnormalities on physical examination or in the hematology or laboratory profile that should prompt further investigation. 2"-34 The pulmonary eosinophilia syndromes or vasculitis associated with eosinophilia should provide clues in the history and physical examination to stimulate further investigation.35·36 Similarly, collagen diseases associated with eosinophilia usually present as systemic diseases.37-39 For instance, the recently described eosinophilia-myalgia syndrome secondary to contaminated tryptophan was associated with fatigue, shortness of breath, cough, rash, headache, pulmonary vasculitis, alveolitis, and fasciitis.40 Lastly, the hypereosinophilic syndrome has arbitrarily been defined as an absolute eosinophil count of more than 1.5 X 109 I L for longer than 6 months, occurring in the absence of other obvious etiologic causes. 41 •42 In one series, 12% of patients with the hypereosinophilic syndrome had their eosinophilia detected incidentally. However, the majority of these patients had systemic symptoms, including cough, breathlessness, muscle pains, angiodema, rash, or fever.42 Many patients with the hypereosinophilic syndrome also have other hematologic findings, such as absolute neutrophilia, anemia, or thrombocytopenia, which would prompt further investigation. The ultimate value of any laboratory test should be measured not only by technical and clinical performance, but also by its impact on patient management. In our study, the discovery of eosinophilia in those cases where it was not anticipated did not result in significant changes in either patient management or clinical outcome. Thus, our results do not justify the extended workup recommended by some authors for patients in this particular clinical setting. These findings represent important information for those working in the hematology laboratory, since they are frequently consulted regarding the extent of investigation considered appropriate for individual cases of eosinophilia. With regard to the degree of follow-up testing, certainly, if there is an appropriate travel, dietary, or other exposure history, stool examination for ova and parasites would be worthwhile.43·44 If there is suspicion of a possible allergic etiology, tests for evidence of atopy may be performed concomitantly with the parasitic investigation. For routine testing for atopy in cases of eosinophilia, skin-prick tests, which are usually cheaper and more effective than radioallergosorbent testing, may be used.45 Finally, repeat WBC differential counts to ascertain that the eosinophilia has ultimately disappeared is also reasonable. A repeat 966
Arch Pathol Lab M ed-Vol 121 , September 1997
In the· prese~t study; the vast majo~ity of eosinophilia in ambulatory outpatients were associated with allergic processes or idiopathic in nature, and an extensive investigation does not appear to be warranted unless specifically indicated by physical examination or other features in the clinical history or hematology profile. Other authors have recently reached similar conclusions regarding the economic consequences of excessive diagnostic testing in cases of eosinophilia.12.46 Perhaps the results of this investigation will stimulate further research into the etiology and clinical significance of eosinophilia in other patient groups and settings. References 1. Zacharski LR, Elveback LR, Unman )W. Leukocyte counts i n healthy adults. J Clin Path ol. 1971;56:148-150. 2. O rfanakis NG, Ostlund RE, Bishop CR, Athens JW. Normal blood leukocyte concentration values. J Clin Patho l. 1970;53:641-647. 3. Hildebrand FL, Cristensen NA, Hanlon DG. Eosinophi lia of unknown cause. Arch Intern Med. 1964;113:129-134. 4. Brigden M L, Horak MG. Incidence and clinical significance of unsuspected eosinophilia discovered by automated WBC differential counting. Lab Med. 1993;24:173-176. 5. Wykoff RF. Eosinophilia. South Med }. 1986;79:608-612. 6. Spry CJF. Eosinophilia. Practitioner. 1982;2 6:79-88. 7. Zucker-Franklin D. Eosinophilia of unknown etiology: a diagnostic dilemma. Hasp Pracr. 1971;7: 11 9-127. 8. Gleich G). Current understandi ng of eosi nophil function. Hasp Pract. 1988;23:137-160. 9. Weller PF. The immunobiology of eosinophi ls. N Eng/ J Med. 1991;324: 11 10-1118. 10. Wardlaw A). Eosinophils in the 1990s: new perspecti ves on their role in health and disease. Postgrad Med}. 1994;70:536-552. 11 . Lokich ). Clinical problem-solving: hypereosi nophilic syndrome. N Eng / J M ed. 1996;334:538. 12. Putterman C, Ben-Chetrit E. Clinical problem sol ving: testi ng, testing, testing... . N Eng/ J M ed. 1995;333: 1208-1211. 13. Lukens )N. Eosinophilia in children. Pecl iatr Clin North Am. 1972;1 9:969-981. 14. Teo CG, Singh M , ling WC, 1-lo LC, Ong VW, Seet LC. Evaluation of the common conditions associated w ith eosinophilia. I C/in Pathol. 1985;38:305-308. 15. Burrows B, Hansan FM, Barbee RA, H alonen M, Lebowitz MO. Epidemiologic observations on eosi nophilia and its relation to respiratory disorders. Am Rev Respi r Dis. 1980;122:709-719. 16. Schoen ), Pizer M . Eosinophilia apparently related to cigarette smoking. N Eng/ j Med. 1964;270:1344-1347. 17. Ghossein NA, Bosworth JL, Stacey P, Muggi a FM, Krishnaswamy V. Radiation related eosinophilia. Radiology. 1975; 11 7:413-417. 18. Aberer W, Konrad K, Wolff K. Wells' syndrome is a distinctive disease entity and not a histologic diagnosis. f Am A cad Dermato /1 988; 18:105- 11 4. 19. Deshpande AD. Eosi nophilia associated with scabies. Practitio ner. 1987;231 :455. 20. Spy CJF. Eosinophilia and allergic reactions to drugs. C/in Haematol. 1980;9:52 1-534. 21. Lowe 0 , Jori zzo ). Tumour-associated eosinophilia: a review. / C/in Pathol. 1981;34:1343-1348. 22. Benvenisti OS, Ullmann IE. Eosi nophilic leukemia: report of five cases and review of literature. Ann /nl M ed. 1969;7 1 :731-745. 23 . Ackerman GA. Eosi nophil ic leukemia: a morphologic and histochemical study. Blood. 1964;2 4:372. 24 . Xavier AM . Eosinophil ia as precursor of acute myelomonocytic leukemia. Arch Path ol Lab M ed. 1988; 11 2:577. Letter. 25. Fishel RS, Farnen )P, Hanson CA, Silver SM, Emerson SG. Acute lymphoblastic leukemia w ith eosinophilia. Medicine. t 990;69:232-243. 26. Glantz L, Rintels P, Samoszuk M, Medeiros). Plasma cell myeloma associated w ith eosi nophilia . Am ) C/i n Pathol. 1995;1 03:583-587. 27. M urata K, Yamada Y, Kami hira S, et al. Frequency of eosinophilia in adult T-cell leukemia/lymphoma. Cancer. 1992;69:966-971. 28. O 'Shea J), Jaffe ES, Lane HC, MacDermott RP, Fauci AS. Peri pheral T-cell lymphoma presenting as hypereosinophilia w ith vasculitis. Am f Med. 1987;82: 539-545. 29. Desenne JJ, Acquatella G, Stern R, M uller A, Sanchez M , Somoza R. Blood eosinophilia in Hodgkin's disease-a follow-up of 25 cases in Venezuela. Cancer. 1992;69: 1248-1253. 30. Watanabe K, Shinbo T, Kojima M , et al. B-cell lymphoma associated with eosinophilia. Cancer. 1989;64: 1682-1685. 3 1. Miranda RN, Esparaza AR, Sambandam S, Medeiros ). Systemic mast cell disease presenting with peripheral blood eosinophilia. Hum ~1thol. 1994;25:727- 730. 32. Liao KT, Rosai ), Doneshbod K. Malignant histiocytosis with cutaneous involvement and eosinophilia. Am j C/in Pathol. 1970;57:438-448.
Eosinophilia Detected by Automated Cell Counting-Brigden & Graydon
~VJ t.Jdtltlll;) diiU lt::)U H!> Ul Ht!dlmem L years aner onset An Intern Med. 1995; 122:851-855 . 41. Jameson MD, Segraves SD. Idiopathic hypereosinophil ic syndrome. Pos. grad Med. 1988;84:93-101. 42 . Weller PF, Bubley G). The idiopathic hypereosinophilic syndrome. 8/oo( vovlln... .:J\U\U3 Vf
1992;27:954-958. 34. Matsushima T, Murakami 1-1, Sawamura M, et al. Myelodysplastic syndrome with eosinophilia in bone marrow. Br J Haemato/. 1993;84:636-638. 35. Enright T, Chua S, Lira DT. Pulmonary eosinophilic syndromes. Ann Allergy. 1989;62:277-283. 36. Lanham )G, Elkon KB, Pusey CD, Hughes GR. Systemic vasculitis with asthma and eosinophil ia: a clinical approach to the Churg-Strauss syndrome. Medicine. 1984;63:65-81. 37. Michel C), Doyle )A, Ginsburg WW. Eosinophi lic fasciitis. Mayo Clin Proc.
1981 ;56:27-34. 38. Lakhanpal S, Duffy ), Engel AG. Eosinophilia associated with perimyositis and pneumonitis. Mayo C/in Proc. 1988;63:37-41. 39. Falanga V, Medsger TA. Frequency, levels, and significance of blood eo-
1994;83:2759-2779. 43. Weller PF. Eosinophilia in travelers. Travel Medicine. 1992;76:1413-1431 44. Nutman TB, Ottesen EA, leng S, et al. Eosinophilia in Southeast Asia1 refugees: evaluation at a referra l center. J Infect Disease. 1987; 155:309-313. 45. Ten RM, Klein )5, Fri gas E. Laboratory medicine and pathology-al lerg skin testing. Mayo C/in Proc. 1995;70:783-784. 46. Lokich ). Clinical problem solving: hypereosinophilic syndrome. N Eng/ Med. 1996;334:538.
sinophi lia in systemic sclerosis, localized scleroderma, and eosinophilic fasciitis. JAm Acad Dermatol. 1987;17:648-656.
Arch Pathol Lab Med- Vol 12 1, September 1997
Eosinoph ilia Detec ted bv AutomatP.rl C:P./1 rnun tin--Rrio,jpn
flv
r. r~"rtnn
Oh7
Receiver Operating Characteristic Curves for Analysis of the Results of Cervicovaginal Smears A Useful Quality Improvement ToolAndrew A Renshaw, MD; Barbara R. Dean, CT(ASCP); Edmund 5. Cibas, MD
• Objectives.-To assess the value of receiver operating characteristic curves (ROCs) as a tool for improvement in the evaluation of cervicovaginal smear results. Design.-We reviewed the results of cervicovaginal smear interpretation for 1 year by 1 0 different cytopathologists working in the same laboratory (total number of cases, 45 356) and compared them with results of histologic evaluation of corresponding biopsies (n = 2090, 4.6%), the latter taken to be the gold standard. Receiver operating characteristic curves were generated for the laboratory as a whole and for individual cytopathologists.
Results.-Analysis of these receiver operating characteristic curves was surprisingly informative. They showed that cytopathologists with different diagnostic thresholds could be distinguished from each other. A difference in diagnostic threshold could be distinguished from a difference in diagnostic accuracy. Conclusions.-We conclude that while receiver operating characteristic curves for cervicovaginal smear interpretation have limitations, the results can be used for quality improvement. (Arch Patholl.ab Med. 1997;121 :968-975)
he interpretation of cervicovaginal smears has come under intense scrutiny in recent years. A new reportT ing terminology, the Bethesda system, has been widely
SIL diagnosis, as well as altered diagnostic accuracy.21 The ASCUS-SIL ratio alone cannot distinguish between any of these factors. By analogy, using the ASCUS-SIL ratio alone to evaluate the quality of cervical vaginal smear interpretation is like trying to navigate without a fixed reference point; one may know that two points are 20 miles apart, but one cannot know the exact location of either one. Thus, to effectively change one's ASCUS-SIL ratio, additional information may be useful. Receiver operating characteristic curves (ROC) are a widely used tool for the evaluation of laboratory tests, particularly clinical laboratory tests. Receiver operating characteristic curves graphically display the true-positive rate (sensitivity) versus the false-positive rate (1 - specificity) for a particular test. By examining an ROC, one can distinguish a difference in diagnostic threshold, which is represented by points at different places along the same ROC curve, from a difference in diagnostic accuracy, which is represented by entirely separate ROC curves. A test with greater accuracy has an ROC curve that is shifted upward and to the left, resulting in a larger area under the curve than one with less accuracy. In effect, the greater the diagnostic accuracy, the greater the area under the curve. While evaluation of ROCs has been advocated as a better way of assessing the results of cytologic tests in general,22-24 the actual experience with ROCs in cytology is very limited. In what appears to be one of the few such studies,24 the accuracy of several different observers evaluating bronchial brush specimens was compared. To the best of our knowledge, ROCs have not been constructed for assessing the accuracy of observers evaluating cervicovaginal smears. Whether reliable ROCs can be constructed in a real life laboratory situation is not clear. First, a standard to which the results are compared must be defined. The standard can be adjudicated review of the
adopted, 1•2 and federal guidelines included in the Clinical Laboratories Improvement Amendments of 1988 have mandated several tools for quality assessment. 3 Review of smears originally diagnosed as within normal limits has shown that conventional screening is not 100% sensitive, and that false-negative results, due to both sampling and laboratory error (either screening or interpretation), are inevitable.4-11 Correlation of the results of cervicovaginal smear evaluation with those of corresponding biopsies have demonstrated that most discrepancies are due to sampling rather than laboratory errors.4•12-16 More recently, investigators have proposed monitoring the ratio of cases reported as atypical squamous cells of undetermined significance (ASCUS) to those interpreted as squamous intraepitheliallesion (SIL) as a quality control measure. 17 In one survey, while the absolute rate of ASCUS diagnoses varied considerably between laboratories, the ASCUS-SIL ratio varied much less. 17 Interim guidelines propose that the ASCUS-SIL ratio of a laboratory, and by implication, of an individual cytopathologist, should be kept at less than 3, or even less than 1.18 While ASCUS-SIL ratios provide useful quality assessment information, it is important to point out that an altered ASCUS-SIL ratio may be the result of several factors. These include an altered diagnostic threshold for a negative diagnosis, 19•20 an altered diagnostic threshold for an Accepted for publication April 15, 1997. From the Division of Cytology, Department of Pathology, Brigham & Women's Hospital and H arvard Medical School, Boston, Mass. Reprint requests to Department of Pathology, Brigham & Women's H ospital, 75 Francis St, Boston, MA 02115 (Dr Renshaw). 968
Arch Pathol Lab Med-Vol121 , September 1997
ROC Analysis in Quality lmprovement'-Renshaw et al