The Utility of Patient Age in Evaluating Prostate Cancer

Transcription

1 CLINICAL CHEMISTRY Original Article The Utility of Patient Age in Evaluating Prostate Cancer HARRY T. POTEAT, MD, ScD, 1 GEORGE T. HO, MD, 2 MEI-LING TING LEE, PhD, 3 WILLIAM R. WELCH, MD, 1 KEVIN R. LOUGHLIN, MD, 2 AND DAVID B. SACKS, MB, ChB 1 The utility of age in examining patients for prostate cancer was assessed. Of the 462 patients in the study, 138 had prostate cancer. The age distribution of the patients with cancer was similar to that found in patients with prostate cancer in the US population, and a correlation between age and the serum prostate-specific antigen (PSA) value was noted (r=.4, P<.002). Selection of reference intervals had a significant effect on test performance. Using an interval of 0 to 4.0 ng/ml, sensitivity of the PSA assay was 90% overall and varied from 78% (patients aged years) to 94% (patients aged years). In contrast, age-adjusted reference ranges yielded corresponding sensitivities of 84%, 78%, and 88%. With a single, fixed reference range, specificity decreased with advancing patient age (P<.001). This trend was eliminated by adjusting the cutoff in different age groups. In addition, age-adjusted reference ranges improved specificity by 10%, and by using the results of examination of a biopsy specimen as the "gold standard," the total number of patients classified correctly by PSA increased from 226 to 250 (49%-54%). For staging before treatment, patient age, clinical stage, and Gleason score were combined to yield a single probability estimate for organ-confined disease (P-c.001). The use of age-adjusted reference ranges is supported by this study, which demonstrates that assay efficiency and specificity improve and sensitivity, although decreased overall, becomes more uniform across age groups. In this patient population, age was useful in determining the probability of organ-confined prostate cancer. Use of this model in clinical decision making should await evaluation in a prospective trial. (Key words: Prostate-specific antigen; Age; Neoplasms; Staging; Reference ranges; Multivariate models) Am J Clin Pathol 1997;107: Carcinoma of the prostate is the second most common malignancy in men in the United States and accounts for approximately 38,000 deaths each year. 1 ' 2 Because the disease may progress to an advanced incurable stage without eliciting symptoms, early detection of prostate cancer is desirable. However, the high prevalence of clinically inconsequential prostate cancer discovered at autopsy argues against early treatment of nonaggressive adenocarcinoma. 3 ' 4 Given this dichotomy, a clear understanding of the variables that influence the sensitivity and specificity of the PSA test is desirable. Prostate-specific antigen (PSA) is a 30-kd serine protease produced by prostatic epithelial cells. 5-7 The concentration of the protein in the serum can be accurately determined Prostatitis, benign prostatic hypertrophy (BPH), and adenocarcinoma of the From the ^Departments of Pathology, 2 Urology, and 3 Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts. Manuscript received May 16, 1996; revision accepted September 4,1996. Address reprint requests to Dr Poteat: Department of Pathology, Brigham and Women's Hospital, 75 Francis St, Boston, MA prostate are all associated with increased concentration of PSA in the serum. 7 ' 12 Measurement of the serum PSA concentration has become an important adjunct to digital rectal examination in the diagnosis of carcinoma of the prostate 13,14 and can also be used to detect residual or recurrent disease after radical prostatectomy or radiation therapy. 15,16 The serum PSA value increases with age in populations of healthy men This observation has led several authors to propose age-adjusted reference ranges for the assay. However, the influence of these age-adjusted reference ranges on the sensitivity and specificity of the PSA assay requires evaluation of the assay against a "gold standard" such as the results of examination of a prostate biopsy specimen. Moreover, the relationship between serum PSA concentration and age in patients with cancer has not been clearly established. Therefore, in a population of patients with prostate cancer whose age distribution was similar to that for patients with prostate cancer across the United States, 2 we studied the relationship between PSA value and patient age, as well as the effect of using age-adjusted reference ranges on assay efficiency. In addition, the utility of patient age and several other parameters known before therapy to predict the presence of organ-confined disease were assessed. 337

2 338 CLINICAL CHEMISTRY Article Patients PATIENTS AND METHODS All patients undergoing needle biopsy of the prostate at Brigham and Women's Hospital (Boston, Mass), from July 1, 1990 to March 31, 1993, were selected by review of pathology reports. Patients who had undergone biopsies of the right and left lobes and who had had a PSA determination at the same hospital within 6 weeks preceding the biopsy were included in the study. A review of 1,158 consecutive biopsy reports yielded 462 cases that met the inclusion criteria. The study protocol was approved by the human subjects committee. PSA Determination All study patients had a PSA determination performed in the clinical laboratory at the Brigham and Women's Hospital using the Hybritech Tandem E method (Hybritech Inc, San Diego, Calif). The assay has a stated reference range of 0 to 4.0 ng/ml and was performed as described by the manufacturer. 8 Biopsy Specimens Prostate-needle biopsy specimens were approximately 0.2 cm in diameter and varied in length from 1.0 to 2.5 cm. The number of cores removed from each lobe varied from two to five, with most specimens containing three cores per lobe. Biopsy specimens designated right and left were required for a patient to be included in the study. Histologic Evaluation Histologic evaluation was performed by the surgical pathology staff at Brigham and Women's Hospital. The diagnosis of adenocarcinoma was dependent on the presence of crowded, infiltrating glands of variable size composed of cells with abundant, clear to pale, eosinophilic cytoplasm and, especially in higher grade tumors, enlarged nuclei with large prominent nucleoli. The tumor in each biopsy specimen was graded according to the method of Gleason. 20 Cases for which determination of the Gleason score was controversial were reviewed by three to five senior pathologists, and a consensus score was recorded. Biopsy specimens that were initially graded only as moderately, well, or poorly differentiated, were reexamined by one of the authors (H.P. or W.W.) and Gleason graded. In two cases, only a single Gleason grade 2 or 3 had been given. These biopsy specimens were inspected by an author (H.P.) and classified as and 3 + 3, respectively. Information about the patient's PSA value and medical history, including age, appeared on some biopsy requisitions, which were seen by the surgical pathologist. Patient Classification and Staging Patients with biopsy-proved adenocarcinoma were segregated on the basis of their PSA values: if the PSA value was elevated, the patient was classified in the true-positive group, while if the PSA value was in the reference range, the patient was classed in the falsenegative group. Similarly, the true-negative group included patients without evidence of prostatic disease on needle biopsy whose serum PSA value was within the reference range. The false-positive group showed an elevation of the PSA above reference range without evidence of biopsy-proved adenocarcinoma of the prostate. The fixed reference range was 0 to 4.0 ng/ml, and age-adjusted ranges were as follows: age 40 to 49 years, 0 to 2.5 ng/ml; age 50 to 59 years, 0 to 3.5 ng/ml; age 60 to 69 years, 0 to 4.5 ng/ml; and age 70 years or older, 0 to 6.5 ng/ml, according to Oesterling et al. 17 The American-Whitmore system and the TNM system were used for staging, and the stage of B0, Tic was assigned to patients found to have prostate cancer on the basis of elevated serum PSA without palpable evidence of disease by digital rectal exam of the prostate. Clinical stage DO, T3NXMX was assigned to patients found to have persistently elevated (more than one measurement) serum prostatic acid phosphatase on metastatic workup for otherwise localized prostate cancer; the initial bone scan in each of these patients had failed to detect evidence of metastatic disease. Pathologic and clinical stage were determined by chart review by two of the authors who are urologists (G.H. and K.R.L.). Statistical Analysis \ The Kruskal-Wallis test and % 2 were used to test for differences between the ages of subjects included in this study and the general US population. Using multiple linear regression, relationships between patient age and the serum PSA value were analyzed after controlling for clinical stage and Gleason grade on the needlebiopsy specimen. The Armitage test 21 was used to test for linear trend in age for sensitivity and specificity. AJCF* irch 1997

3 POTEAT ET AL 339 Patient Age in Prostate Cancer Stepwise multivariate logistic regression was used to construct a model for predicting the probability of organ-confined disease. Possible variable choices in the stepwise regression included serum PSA value, an interaction term for the serum PSA value and the patient age, the patient age, the Gleason score on the needle biopsy specimen, and the clinical stage. Statistical computations were performed using the statistical packages SAS and STATXACT. RESULTS We reviewed 1,158 consecutive biopsy reports. To meet inclusion criteria, patients were required to have had a PSA determination at the Brigham and Women's clinical laboratory within 6 weeks preceding the biopsy, have had biopsy specimens designated for the right and left lobes, and have no history of treatment for prostate cancer. Cases were excluded from the study if the pathologist could not exclude metastatic spread to the prostate from an alternative primary site or if a malignancy was not epithelial in origin. Among 462 patients who met the inclusion criteria, 324 had benign prostatic hypertrophy (BPH), prostatitis, or normal prostatic tissue in their biopsy specimens, and 138 had adenocarcinoma of the prostate. Selected characteristics of these patients are shown in Tables 1 and 2. Because the selection criteria for this study were distinct from those used for patients undergoing radical prostatectomy or patients randomly selected from the general population for screening, further characterization was performed. The age distribution of the patients reported in this study was not significantly different from that reported for patients with prostate cancer in the United States as a whole 2 (Table 3). Of the 138 patients, 82 (59%) had medium-grade disease (Gleason score 5-7) on examination of the biopsy specimen. This observation, considered with the data in Table 2, suggests that the study population was not skewed toward low- or high-grade disease. A significant positive correlation between serum PSA concentration and age has been reported for men with no evidence of prostate cancer. 17 We determined the relationship between the serum PSA and patient age in men with prostate cancer. Univariate linear TABLE 2. CHARACTERISTICS OF PATIENTS WITH ADENOCARCINOMA DIAGNOSED FROM THE BIOPSY SPECIMEN* No. of patients Mean ± SEM age, y (range) Gleason score (n = 138) Mean ± SEM Clinical stage (n = 138) + A,T(la,lb,lc)N0M0 B, T2N0M0 D, T3NXMX* Pathologic stage (n = 56) No. of patients B, T2N0M0 C, T3N0M0 D, TXN1MX <4.0 nglml ±2 (56-81) 4.8 ± (8) 36(5) 0(0) 7(1) 0(0) 86 (12) 14(2) 6 67(4) 33(2) 0(0) Patient Serum PSA nglml ±1 (50-86) 5.4 ± (15) 42(19) 9(4) 16(7) 4(2) 79 (35) 17(8) 22 86(19) 9(2) 5(1) >10 nglml ±1 (54-92) 6.6 ±0.2 8(6) 32 (25) 37 (29) 24(19) 1(1) 76 (60) 23(18) (12) 39(11) 18(5) *Data are presented as % of total (No. of cases) unless otherwise indicated. tthere were no stage C cases. JSee "Patients and Methods" section for the discussion of patient classification and staging. There were no stage A cases. TABLE 1. SUMMARY STATISTICS FOR AGE AND SERUM PSA VALUES FOR ALL PATIENTS IN THE STUDY Benign conditions/ n = 324 Adenocarcinoma, n = 138 H^y Mean, 67.6 ± 0.4+ Median, 68 Range, Mean, 72.3 ± 0.7 Median, 72 Range, PSA, nglml 9.0 ± ± PSA - prostate-specific antigen. 'Diagnoses of benign prostatic hypertrophy (glandular and/or stromal hyperplasia/hypertrophy), no significant pathologic change, and chronic inflammation consistent with prostatitis. + Standard error of the mean. TABLE 3. AGE DISTRIBUTION OF PROSTATE CANCER CASES IN THIS STUDY AND IN THE GENERAL us ruruuviiuw Age,y 25^ Present Study 0.0 (0) 6.5 (9) 29.0 (40) 44.9 (62) 19.6(27) United States (246) 6.4 (2,870) 31.7(14,251) 42.5 (19,078) 18.9(8,474) "Data are presented as % of total (No. of cases) unless otherwise indicated. P=.83 (Kruskal- Wallis) and P=.86 (x 2 )- The null hypothesis, that the distribution of cases in this study differs from that in the general US population only by a degree attributable to random chance, is accepted. This study includes only cases of primary adenocarcinoma of the prostate. The data from the general US population include all cases diagnosed as cancer of the prostate. Vol. 107 No. 3

4 340 CLINICAL CHEMISTRY Article regression indicated that patient age correlated with the serum PSA value (r=.4; P<.002) (Table 4). Controlling for clinical stage and Gleason score on the needle biopsy specimen, we found a significant positive correlation between patient age and the serum PSA value (P<.038). Thus, in the older patients with cancer, PSA tended to be higher even when the grade and stage of disease were considered. When we controlled for pathologic stage and Gleason score, a similar positive correlation held in the 56 patients for whom pathologic staging information was available. Next, we investigated what effect the use of age-specific reference ranges, proposed by Oesterling et al, 17 would have on how patients were classified for the presence of cancer. Assay sensitivity was examined using fixed and age-dependent reference ranges (Table 5). Overall sensitivity of the assay was 90% when the fixed reference range was used and 84% when age-dependent ranges were used. Compared with the fixed reference range, the use of age-adjusted reference ranges yielded eight new false-negative results but no new true-positive TABLE 4. CLINICAL VARIABLES CORRELATED WITH SERUM PSA VALUE IN MEN WITH ADENOCARCINOMA OF THE PROSTATE, UNIVARIATE REGRESSION Variable* r + r 2 * P Age.4.2 <.002 Gleason score.4.2 <.001 *The variable used in the regression. ^Correlation coefficient ' ^Square of the correlation coefficient, which indicates the degree of variability in the serum PSA value that can be accounted for by a given variable. A value of 1.0 would indicate 100%. TABLE 5. SENSITIVITY OF PSA IN PATIENTS WITH ADENO CARCINOMA OF THE PROSTATE* Fixed Reference Age-Dependent Age, y Range, (0-4.0 ng/ml) Reference Range 1 All ages 90 (124/138) 84 (116/138)' (7/9) 78 (7/9) (36/40) 88 (35/40) (58/62) 82(51/62) >80 85 (23/27) 85 (23/27) 'Data are reported as % of patients, or the sensitivity (No. of true-positive cases/no. of true-positive cases plus false-negative cases). + From Oesterling et al, 17 by patient age in years: 40 to 49, 0 to 2.5 ng/ml; 50 to 59, 0 to 3.5 ng/ml; 60 to 69, 0 to 4.5 ng/ml; and >70, ng/ml. results. Seven of the eight new false-negative results involved the classification of men 70 to 79 years of age. Clinical stages of the eight cases reclassified as false negative included four cases of clinical stage B0, Tic disease, one case of A2, Tib, and three cases of DO, T3NXMX disease. Pathologic staging information was available for two of the DO cases that were both staged as Cl, T3N0M0. The mean Gleason score of the new false-negative cases was 5.3 ± 0.45 (mean ± standard error of mean) compared with 4.8 ± 0.4 observed in cancer cases with PSA values from 0 to 4.0 ng/ml. The fixed reference range of 0 to 4.0 ng/ml yielded specificity that declined significantly with age (P<.001) (Table 6). When patients were reclassified using the age-adjusted reference ranges, 17 the trend for assay specificity to deteriorate with increasing patient age was no longer evident (P=.98) (Table 6). In addition, such patient reclassification resulted in 29 new true-negative cases. The majority (74%) of reclassified patients were in the 70- to 79-year-old age range. Only in the 50- to 59-year-old age category did specificity decrease with 2 cases (6%) newly classified as false positive. Overall assay specificity increased by 10%. Test specificity may be underestimated because of the presence of a verification bias, the necessity of performing needle biopsy only in patients deemed likely to have prostate cancer. 22 Assay efficiency increased from 49% using the fixed reference ranges to 54% using the age-adjusted ranges, with the total number of correctly classified patients increasing from 226 to 250 (Table 7). Next, we addressed the question of whether patient age was useful in predicting the probability of organconfined disease before treatment. To answer this question/ we set up a forward selection logistic regression model with the output variable being the probability of organ-confined disease on pathologic staging. Given the noted correlation between the PSA value and patient age, an interaction term of PSA and age was included. Thus, input variables that could be selected in the stepwise regression included clinical stage, serum PSA value, Gleason score on the needle biopsy specimen, patient age and an interaction term made up of patient age and serum PSA value. Surprisingly the stepwise regression identified only clinical stage, Gleason score, and patient age as the variables yielding the best fit for the data (P<.001) (Table 8). Patient age was of borderline significance (P=.091), and serum PSA value was not represented in the model. Only if Gleason score was eliminated as a possible variable choice could the serum PSA value be selected into the model, implying strong covariation of these two parameters. AJCP- irch 1997

5 POTEAT ET AL 341 Patient Age in Cancer TABLE 6. SPECIFICITY OF PSA IN PATIENTS WITH ADENOCARCINOMA OF THE PROSTATE* Fixed Reference Age-Dependent Age, y Range, ng/ml Reference Range* All ages 31 (102/324) 41 (134/324) (5/7)* 71 (5/7) (20/34) 53 (18/34) (47/153) 35 (53/153) (26/114) 43 (49/114) >80 25 (4/16) 56 (9/16) 'Data are reported as % of patients, or the specificity (No. of true-negative cases/no. of true-negative cases plus false-positive cases). + From Oesterling et al, 17 by patient age in years: 40 to 49, 0 to 2.5 ng/ml; 50 to 59, 0 to 3.5 ng/ml; 60 to 69, 0 to 4.5 ng/ml; and >70, ng/ml. ^Significant P value for trend. To determine the probability level at which the maximum number of patients would be correctly classified as having or not having organ-confined disease, a receiver-operator curve was constructed for the equation shown in Table 8. Patient classification was optimal if those with a 50% or higher probability of having organ-confined disease were classified as such and those with less than a 50% probability were classified as having extensive disease. Thus if 50% probability was used as a cutoff, then the equation was 86% sensitive and 71% specific for the presence of organ-confined disease. The correct patient classification was made 80% of the time. A graph of probability level vs accuracy of patient classification is shown in the Figure. DISCUSSION Four important results of this study are presented. First, the age distribution of the patients with cancer in this study is similar to that for patients with prostate cancer throughout the United States. Second, the serum PSA values in patients with cancer correlate with patient age. Third, test efficiency and specificity are improved by age-adjusted reference ranges. Fourth, age may need to be considered as a variable when calculating the probability of organ-confined cancer. Before instituting the age-specific reference ranges, an understanding of the relationship between serum PSA concentration in men with prostate cancer and age is important. This study showed a correlation between the serum PSA value and the patient's age (r=.4; P<.002), and the correlation remained significant after controlling for stage and grade of disease. This finding suggests that age-adjusted reference ranges, similar to those advocated for improving Age,y All ages >80 TABLE 7. EFFICIENCY OF PSA* Fixed Reference Range, ng/ml 49* (226/462) 63 (27/43) 43(83/193) 48(84/176) 63 (27/43) Age-Dependent Reference Range* 54 (250/462) 58 (25/43) 46 (88/193) 57 (100/176) 74 (32/43) + Data are reported as % of patients, or the efficiency (No. of true-positive cases plus true negative cases/total No. of patients in the study). tfrom Oesterling et al, 17 by patient age in years: 40 to 49, 0 to 2.5 ng/ml; 50 to 59, 0 to 3.5 ng/ml; 60 to 69, 0 to 4.5 ng/ml; and 270, ng/ml. ^Significant P value for trend. specificity, could be selected to make PSA assay sensitivity uniform across age groups. In fact, the range of sensitivities across age groups narrowed from 74% to 94% when a fixed reference range was used, to 78% to 88% when age-adjusted ranges were substituted. Labrie and colleagues 23 showed a significant correlation of age with serum PSA concentration in 57 patients with cancer (r-.04); however, grade and stage of disease were not controlled. In 66 patients with prostate cancer, Bangma et al 24 showed a trend for PSA to increase with age. While they differentiated between patients with organ-confined disease and those with extracapsular spread, no statistical analysis was indicated in their work. Moreover, although both of these studies had a broad age range of patients, neither established the age distribution of patients with cancer relative to that of the endemic population. In the publication of series derived from surgical patients alone, such analysis may be of less utility because these patients tend to be uniformly younger and have early stage disease. 25 The increased prevalence of BPH in older men possibly accounts for the increases of the serum PSA with age present in this population of patients with cancer; however, this is not the only possible explanation. Meshref et al 26 have noted that even after correcting for prostate volume, serum PSA increases with age. Oesterling et al 27 suggested that aging changes the structure of the prostate, allowing PSA to be drawn into the serum more easily in older men. Thus, it seems plausible that an older man with the same stage of prostate cancer as a younger man would, on average, have a higher serum PSA value. A limitation of this study is its retrospective nature. Data were not available from which to estimate gland size. Several methods exist for establishing reference ranges for laboratory tests. For PSA, investigators Vol. 107 No. 3

6 342 CLINICAL CHEMISTRY Original Article TABLE 8. CALCULATING THE PROBABILITY OF ORGAN CONFINED DISEASE USING MULTIVARIATE STEPWISE LOGISTIC REGRESSION Variable" Clinical stage Gleason score Age Coefficient* ± ± ±.0646 c.0007 <.002 <.09 Equation to calculate probability of organ-confined disease* p _ e-t c+32g+p3a^(1+e p1c+p2g+p3a) 'Possible variable choices in the stepwise regression included serum prostate-specific antigen (PSA) value, an interaction term for the serum PSA value and the patient age, the patient age, the Gleason score on the needle biopsy specimen, and the clinical stage. The best fit to the data is achieved when the three listed variables are used. ^he coefficient by which each variable is to be multiplied is shown ± standard error. *pi is the coefficient for clinical stage; C is the clinical stage with clinical stages of A, T1N0M0 and B and, T2N0M0 receiving the value 1 and C, T3N0M0 and D, TnNXMX (where X * 0) receiving the value 0; P2 is the coefficient for Gleason score; G is the Gleason score P, is the coefficient for age and A is the patient age in years. Organ-confined disease is defined as pathologic stage of A, T1N0M0 or B, T2N0M0. have studied populations of apparently healthy men, stratified them by age, and established an upper reference limit A second method is to optimize the sensitivity and specificity of the test through use of a receiver-operator core. Using this method in a study group randomly selected from the general population, Labrie et al 23 identified 3.0 ng/ml as an appropriate cutoff for men of any age. El-Galley et al 28 selected 3.5 ng/ml in men aged 50 to 59 years, 5 ng/ml in men aged 60 to 69 years, and 7 ng/ml in men aged 70 to 79 years. There is general agreement that for men younger than 60 years, the reference range is below 4.0 ng/ml and for men 60 years and older the reference range is higher than 4.0 ng/ml ' 23,27 Several studies, including ours, have applied the age-adjusted reference ranges to a patient series and found decreased sensitivity and increased specificity relative to a fixed reference interval. 26 ' 27,29 The age distribution of the patients with cancer in our study makes our calculated 5% improvement in assay efficiency particularly salient. Thus, one could anticipate an overall improvement in patient classification by use of age-adjusted reference ranges (see Table 7). It is also significant that the majority of patient reclassification occurs in men older than 70 years. In this regard, Dugan et al 30 have observed that prostate cancer has age-adjusted clinical relevance, suggesting that an early stage lesion in a young man is potentially more serious than a similar stage lesion in an elderly man. Using these criteria, it is possible that some of the five Tic and Tib tumors that were reclassified from true i r Probability of Organ-Confined Disease FIG. Probability level vs accuracy of patient classification. The X axis indicates the probability of finding organ-confined disease on pathological staging as indicated by the equation in Table 8. The Y axis shows the accuracy of the equation when applied to the 56 patients for whom pathologic staging information was available. negative to false positive when age-adjusted ranges were used did not represent clinically relevant disease. However, in our study age-adjusted ranges did result in the reclassification of three T3 lesions, two of which were shown by pathologic stage to be clinically significant. We acknowledge that this failure of detecting prostate cancer may be unacceptable to some urologists. Even so, the use of age-adjusted reference ranges makes the sensitivity of the assay more uniform across age groups. Thus, one expects an advantage of ageadjusted reference ranges will be increased sensitivity in younger men. Allocation of resources by the medical community is under intense scrutiny. For prostate cancer, there is heated debate about which staging procedures are necessary. Gee et al, 31 analyzing trends in the management of prostate cancer, have observed that extensive numbers of staging studies are used in patients with newly diagnosed prostate cancer. Other authors have advocated reducing the number of such procedures because the serum PSA level frequently gives sufficient staging information. 32,33 Regardless of how this debate on resource allocation is resolved, studies such as those by Murphy et al, 34 in which only 57.5% of radical prostatectomy patients had stage B2 or ear- AJCP March 1997

7 POTEAT ET AL 343 Patient Age in Prostate Cancer lier disease, underscore the need for improved patient selection. For reasons such as these, we set out to derive a clinical staging algorithm based on variables known before treatment that would calculate the probability of organ-confined disease. The Gleason score and clinical stage recorded in the patient record were used to derive the algorithm. To limit interlaboratory variability, only patients with a PSA determination at our institution's laboratory were included. We found that the algorithm correctly classified 80% of patients. While this finding awaits prospective validation, this level of accuracy is comparable with that reported for endorectal coil magnetic resonance imaging (88%) and superior to that for digital rectal exam alone (56%). 35 In deriving a staging algorithm, we used stepwise logistic regression to select combinations of variables that together best described the probability of organconfined disease. Surprisingly, a combination of Gleason score, clinical disease stage, and patient age gave the best model (P<.001). Based on the work of others, 25,36 ' 37 we had anticipated that serum PSA or an interaction term between PSA and age would be included in the model that gave the best fit to the data. Possible reasons for the differences between our model and those derived by two other groups include sample size, variation in clinical and laboratory practices between institutions, and differences in the statistical methods used to build the algorithm. We view the potential significance of the algorithm as giving the clinicians at our hospital a method to integrate several clinical variables into a single probability to be used in decision making and outcomes analysis. Models derived at other tertiary care centers differ from ours because they include the serum PSA but exclude patient age. 25 ' 37 These differences bring into question the transferability or portability of such models among institutions. It may be advisable for each large treatment center to derive such an algorithm based on local practice parameters of staging and of grading the biopsy specimens. The serum PSA value is highly correlated with disease stage and will remain an important tool for the clinical staging of prostate cancer. 25 The data indicate that the use of age-adjusted reference ranges for PSA may be appropriate provided they allow for uniform sensitivity and specificity across age groups and, thus, do not discriminate against older persons. Age-adjusted reference ranges should be viewed as an adjunct to and not a replacement for appropriate patient selection and rational laboratory utilization. Staging algorithms should continue to be evaluated for their accuracy and cost benefit relative to other staging modalities, such as magnetic resonance imaging. It may be appropriate for each institution to derive its own staging algorithm. Acknowledgments: We thank Paul Oppedisano of Medical Scientists Incorporated (Boston, Mass) and Mike Elashoff of the Harvard School of Public Health (Boston) for advice on epidemiologic and statistical issues and the staff of the Brigham and Women's Hospital Tumor Registry: Linda Baba, Nancy Hinchliffe, Agnes McGlaughlin, and Andrea Brandfon. REFERENCES 1. Parker SL, Tong T, Bolden S, Wingo PA. Cancer statistics. CA Cancer ] Clin. 1996;46: Ries LAG, Miller BA, Hankey BF, Kosary CL, Harras A, Edwards BK, eds. SEER Cancer Statistics Review, : Tables and Graphs. Bethesda, Md: National Cancer Institute; 1994;11. NIH publication Rich AR. On the frequency of occurrence of occult carcinoma of the prostate. / Urol. 1935;33: Breslow N, Chan CW, Dhom G, et al. Latent carcinoma of the prostate at autopsy in seven areas, lnt ] Cancer. 1977;20: Wang MC, Valenzuela LA, Murphy GP, Chu TM. The purification of a human prostate specific antigen. Invest Urol. 1979;17: Lilja H: A Kallikrein-like serine protease in prostatic fluid cleaves the predominant seminal vesicle protein. / Clin Invest. 1985;76: Wang MC, Papsidero LD, Kuriyama M, Valenzuela LA, Murphy GP, Chu TM. Prostate antigen: a new potential marker for prostate cancer. Prostate. 1981;2: Kuriyama M, Wang MC, Papsidero LD, et al. Quantitation of prostate-specific antigen in serum by a sensitive enzyme immunoassay. Cancer Res. 1980;40: Chan DW, Bruzek DJ, Oesterling JE, Rock RC, Walsh PC. Prostate specific antigen as a marker for prostatic cancer: a monoclonal and a polyclonal immunoassay compared. Clin Chem. 1987;33: Hortin GL, Bahnson RR, Daft M, Chan KM, Catalona WJ, Ladenson JH. Differences in values obtained with 2 assays of prostate specific antigen. / Urol. 1988;139: Graves HC, Wehner N, Stamey TA. Ultrasensitive radioimmunoassay of prostate specific antigen. Clin Chem. 1992;38: Seamonds B, Yang N, Whitaker B. Evaluation of prostate specific antigen and prostatic acid phosphatase as prostatic cancer markers. Urology. 1986;28: Cooner WH, Mosley BR, Rutherford CL Jr, et al. Prostate cancer detection in a clinical urological practice by ultrasonography, digital rectal examination, and prostate specific antigen. / Urol. 1990;143: Mettlin C, Lee F, Drago J, Murphy GP. The American Cancer Society national prostate cancer detection project. Findings on the detection of early prostate cancer in 2425 men. Cancer. 1991;67: Oesterling JE, Chan DW, Epstein Jl, et al. Prostate specific antigen in the preparation and post operative evaluation of localized prostatic cancer treated with radical prostatectomy. / Urol. 1988;139: Stamey TA, Kabalin JN, McNealy JE, et al. Prostate specific antigen in the diagnosis and treatment of adenocarcinoma of the prostate. II: radical prostatectomy treated patients. / Urol. 1989;141: Vol. 107 No. 3

8 344 CLINICAL CHEMISTRY Original Article 17. Oesterling JE, Jacobson SJ, Chute CJ, et al. Serum prostate specific antigen in a community based population of healthy men: establishment of age specific reference ranges. JAMA. 1993;270: Dalkin B, Ahmann FR, Kopp JB. Prostate specific antigen levels in men older than 50 years without clinical evidence of prostate carcinoma. / Urol. 1993;150: Anderson R, Strickland D, Corbin D, Byvres VA, Zweiback E. Age specific reference ranges for serum prostate specific antigen. Urology. 1995;46: Gleason DF, The Veterans Administration Cooperative Urological Research Group. Histological grading and clinical staging of prostatic carcinoma. In: Tannenbaum M, ed. Urologic Pathology: The Prostate. Philadelphia, PA: Lea and Febiger; 1977: Armitage P. Tests for linear trends in proportions and frequencies. Biometrics. 1955;V.ll, Begg C, Greenes R. Assessment of diagnostic tests when disease verification is subject to bias. Biometrics. 1983;39: Labrie F, Dupont A, Suburu R, et al. Serum prostate specific antigen as a prescreening test for prostate cancer. / Urol. 1992;147: Bangma CRL, Kranse R, Bligenberg BG, Schroder FH. The value of screening tests in the detection of prostate cancer, part II: Retrospective analysis of free/total prostate specific antigen ratio, age specific reference ranges and PSA density. Urology. 1995;46: Partin AW, Carter HB, Chan DW, et al. Prostate specific antigen in the staging of localized prostate cancer: influence of tumor differentiation, tumor volume and benign hyperplasia. / Urol. 1990;43: Meshref AW, Nachabe M, Bazinet M, et al. Role of prostate specific antigen density after applying age specific prostate specific antigen reference ranges. Urology. 1995;45: Oesterling JE, Jacobsen SJ, Cooner WH. The use of age specific reference ranges for serum prostate specific antigen in men 60 years or older. Urology. 1995;53: El Galley RS, Petros JA, Sanders WH. Normal range prostatespecific antigen versus age specific prostate specific antigen in screening prostate adenocarcinoma. Urology. 1995;46: Speights VO, Spiekman AM, Brawn PN, et al. Evaluation of agespecific normal ranges for prostate specific antigen. Urology. 1995;45:454-^ Dugan JA, Bostwick DG, Myers RP, et al. The definition and preoperative prediction of clinically insignificant prostate cancer. JAMA. 1996;275: Gee WF, Holtgrewe HL, Albertsen PC, et al. Practice trends in the diagnosis and management of prostate cancer in the United States. / Urol. 1995;154: Chybowski FM, Keller JJL, Bergstralh EJ, Oesterling JE. Predicting radionuclide bone scan findings in patients with newly diagnosed untreated prostate cancer. / Urol. 1991;145: Levran Z, Gonzalez JA, Diokono AC, Jafri SZ, Steinart BW. Are pelvic computed tomography, bone scan and pelvic lymphadenectomy necessary in the staging of prostate cancer? Br J Urol. 1995;75: Murphy GP, Mettlin C, Menck H, Winchester DP, Davidson AM. National patterns of prostate cancer treatment by radical prostatectomy: results of a survey by the American College of Surgeons Commission on Cancer. / Urol. 1994;152: Hoch Boni RA, Boner JA, Debatin JF, et al. Optimization of prostate carcinoma staging: comparison of imaging and clinical methods. Clin Radiol. 1995;50: Kleer E, Larson-Keller JJ, Kineke H, Oesterling JE. Ability of preoperative serum prostate specific antigen value to predict pathologic stage and DNA ploidy: influence of clinical stage and tumor grade. Urology. 1993;41: Partin AW, Yoo J, Carter HB, et al. The use of prostate specific antigen, clinical stage and Gleason score to predict pathologic stage in men with localized prostate cancer. ] Urol. 1993;150: AJCP March 1997