- Email: [email protected]
total, and complexed prostate-specific antigen for prostate cancer screening
COST-BENEFIT ANALYSIS OF TOTAL, FREE/TOTAL, AND COMPLEXED PROSTATE-SPECIFIC ANTIGEN FOR PROSTATE CANCER SCREENING LARS ELLISON, CAROL D. CHELI, STEVEN BRIGHT, ROBERT W. VELTRI,
AND
ALAN W. PARTIN
ABSTRACT Refinements in prostate-specific antigen (PSA) through the use of its derivatives have augmented early detection rates of prostate cancer. However, these improvements are coupled with relatively large increases in unit cost per detected cancer. We used decision-analytic modeling to determine the most appropriate PSA derivative for population-based screening. We constructed a decision-analytic model to determine the PSA derivative with the highest cost-benefit ratio for prostate cancer screening. We defined 5 screening strategies: total PSA (tPSA) 4.0 ng/mL; free PSA/tPSA (f/tPSA) in conjunction with tPSA; and complexed PSA (cPSA) 3.8, 3.4, and 3.0 ng/mL. Prostate cancer prevalence, false-positive rates, and false-negative rates for each test strategy were calculated from a database of 2138 men. The direct costs were obtained from literature review and our department of clinical chemistry. The derivative cPSA with a positive threshold of 3.8 ng/mL was the dominant strategy. The average cost of screening was $138.93. The strategy of tPSA became dominant when the cost of cPSA was ⬎$35.00 or the cost of a prostate biopsy was ⬍$67.30. To match the false-negative rate of tPSA 4.0 ng/mL, a cPSA threshold of 3.0 ng/mL is necessary (sensitivity 92.5%). At this level, the marginal cost increase over tPSA is $9.40. The dominant strategy for populationbased prostate cancer screening is use of cPSA with a positive threshold of 3.8 ng/mL. The use of cPSA with a threshold of 3.0 ng/mL identifies a similar number of cancers with fewer biopsies than tPSA at 4.0 ng/mL. UROLOGY 60 (Suppl 4A): 42–46, 2002. © 2002, Elsevier Science Inc.
P
rostate-specific antigen (PSA) has changed our definitions and management of prostate cancer. However, as a screening biomarker, total serum PSA (tPSA) has a poor positive predictive value because of a high false-positive rate.1 Recognition of different molecular forms of PSA (free PSA [fPSA] and complexed PSA [cPSA]) has led to incremental improvements in these test characteristics.2 Identification of appropriate positive thresholds for these PSA isoforms may allow clinicians to more accurately identify patients for biopsy and greatly diminish the numbers of unnecessary biopsies. The unmeasured corollary of the introduction of From the Brady Urological Institute, the Johns Hopkins Medical Institution, Baltimore Maryland, USA (LE, RWV, AWP); Robert Wood Johnson Clinical Scholars Program, the Johns Hopkins Medical Institution, Baltimore Maryland, USA (LE); Bayer Corporation, Tarrytown, New York, USA (CDC); and UroCor, Inc., Oklahoma City, Oklahoma, USA (SB) Reprint requests: Lars Ellison, MD, Brady Urological Institute, Johns Hopkins-RWJ Program, 600 North Wolfe Street, Carnegie 291, Baltimore, MD 21287. E-mail: [email protected]
42
© 2002, ELSEVIER SCIENCE INC. ALL RIGHTS RESERVED
improved and advanced biomarkers is increased cost of screening. The cost relation is complicated by the relative contribution of downstream diagnostic tests (specifically the transrectal ultrasound [TRUS]-guided biopsy). In addition, this cost relation is amplified by the necessary broad definition of the population at risk (men ⬎50 years old). We used (1) a large multi-institutional patient sample of men, aged 40 to 75 years, who underwent standard screening and prostate biopsy; (2) cost and utility data from the literature; and (3) decision-analytic modeling. This information was used to determine, from a societal perspective, which PSA isoform and serum threshold of positivity is most cost-beneficial for prostate cancer screening. METHODS PATIENT SAMPLE We merged 2 contemporary datasets, the first from UroCor Labs (Oklahoma City, OK) and the second from Bayer Diagnostics (Tarrytown, NY). The UroCor data have been previously described.3 For the Bayer dataset, the study sites were: 0090-4295/02/$22.00 PII S0090-4295(02)01694-1
TABLE I. Demographics of the combined UroCor* and Bayer Diagnostics† datasets No Cancer (n ⴝ 1519), mean (SE) Age (yr) tPSA (ng/mL) cPSA (ng/mL) f/tPSA (ng/mL)
64.2 7.24 5.60 18.87
(0.19) (0.11) (0.09) (0.29)
Cancer (n ⴝ 619), mean (SE) 66.6 10.49 8.40 16.16
(0.29) (1.55) (1.38) (0.38)
P-value ⬍0.001 0.001 0.002 ⬍0.001
cPSA ⫽ complexed prostate-specific antigen; f/tPSA ⫽ free to total prostate-specific antigen; tPSA ⫽ total prostate-specific antigen. * Oklahoma, OK. † Tarrytown, NY.
Johns Hopkins Hospital (Baltimore, MD), Northwest Prostate Institute (Seattle, WA), New York University (New York, NY), Stanford University (Stanford, CA), University of Innsbruck (Innsbruck, Austria), MD Anderson Cancer Center (Houston, TX), and Wyoming Research Foundation (Cheyenne, WY). Subjects were enrolled when recommended for a prostate biopsy by the physician according to established local practice and when the patient consented to TRUS-guided prostate biopsy. The biopsy procedure consisted of ⱖ10 cores of prostate tissue, including a minimum of 6 systematic sectors and 4 lateral cores (2 from each side). All patients underwent a digital rectal examination (DRE). Patients were excluded from the study if they had a personal history of prostate cancer, or a previous history of a transurethral prostatic resection. Patients taking any medication that could decrease levels of serum PSA, such as estrogen, finasteride, or a course of ⱖ7 days of quinolone antibiotic therapy within 30 days of biopsy, were excluded from the study. In addition, patients taking any medication or food supplement that could increase serum PSA level, such as but not limited to dehydroepiandrosterone or testosterone, were excluded. The patient’s age, race, and family history of prostate cancer were obtained. Patients had blood drawn ⱕ1 month before biopsy. Blood was drawn either before DRE or ⱖ1 week and ⱕ1 month after DRE. If biopsy results were positive for prostate cancer, then Gleason score and clinical stage were provided, and if prostatectomy was performed, pathologic stage of disease was determined. Patients who underwent a previous biopsy or multiple biopsies and met the above inclusion criteria were enrolled into the study. This study was performed in compliance with the requirements of the respective institutional review boards. Serum samples were processed immediately upon collection of blood and kept at 4°C ⱕ8 hours before being frozen at ⫺70°C. Serum samples were subsequently shipped on dry ice to Johns Hopkins Hospital Clinical Chemistry Laboratory for testing of Bayer Immuno 1 cPSA and tPSA methods (Bayer Diagnostics) and Beckman Access fPSA and tPSA methods (Beckman Inc, San Diego, CA). Serum samples were tested immediately upon sample thaw. Individual determinations were obtained on each specimen tested. The same serum sample was used to determine Bayer tPSA, cPSA, and Beckman fPSA and tPSA concentrations. In aggregate, complete records of patient age, TRUS biopsy, tPSA, cPSA, fPSA values, and DRE results were available for 3757 men. Individuals were excluded if they were ⬎75 years of age or were coded with a suspicious or abnormal DRE. The final dataset included 2138 men (Table I). Prostate cancer prevalence as well as false-positive rates and false-negative rates for each test strategy were calculated from the database (Table II). The distributions of Gleason scores for detected cases by test strategy are shown in Table III. UROLOGY 60 (Supplement 4A), October 2002
UTILITIES AND COSTS A “utility” is a numerical value between 0 and 1, which represents the relative value of a health state (eg, living with the diagnosis of localized prostate cancer) as compared with perfect health. Utility and cost estimates (Table IV) were derived from literature review.4,5 Specifically, we identified utilities associated with biomarker-defined false-positive and false-negative status. Included among the screening costs were: (1) office and staffing costs, (2) the cost of running the various assays, (3) TRUS biopsy costs, (4) pathologic review costs, and (5) indirect costs. Values were indexed to US dollars (fiscal year 2001).
MODELING We constructed a decision-analytic model to determine the PSA derivative with the highest cost-benefit ratio for prostate cancer screening. We defined 5 treatment strategies: tPSA with a positive threshold of 4.0 ng/mL; fPSA in conjunction with tPSA (f/tPSA positive threshold ⬍25) when tPSA is 4 to 10 ng/mL; and cPSA with a positive thresholds of 3.8, 3.4, and 3.0 ng/mL. Sensitivity analyses were performed on all variables to determine the impact of reasonable variation in the baseline assumptions on the cost-benefit ratio of the different strategies. Statistical analysis was done with the STATA version 7.0 software package (Stata Corporation, College Station, TX), and modeling was completed with DATA version 3.5 (TreeAge Software, Inc., Williamstown, MA)
RESULTS Table I summarizes baseline demographic data for the 2138 men in the sample. Altogether, 29% of men were found to have evidence of cancer at biopsy. The mean age of men with biopsy evidence of cancer was slightly older than those without evidence of cancer. As expected, men with a diagnosis of cancer had higher mean tPSA and cPSA values and lower f/tPSA values. Receiver operating curves (ROC) were generated for the 3 isoforms of PSA. The area under the curve for each is reported in Figure 1. Analysis of the ROC curves shows them to be significantly different (P ⫽ 0.03). Isoforms cPSA and tPSA appear to have better performance in minimizing false-negatives, whereas f/tPSA appears to have better performance in minimizing false-positives. From the model, cPSA with a positive threshold 43
TABLE II. Test characteristics for assays at different threshold levels Biopsy Assay (ng/mL) tPSA ⬎4.0 f/tPSA* ⬍25 cPSA ⬎3.8 ⬎3.6 ⬎3.4 ⬎3.2 ⬎3.0
n
Rate (%)
Specificity (%)
Sensitivity (%)
1844 1577
86.3 73.7
16.3 29.0
92.4 81.2
1526 1613 1721 1784 1836
71.4 75.4 80.5 83.4 85.7
32.6 28.2 22.8 19.6 16.9
81.1 84.3 88.5 91.3 92.6
For abbreviations, see Table I. * Biopsy if tPSA is 4 –10 ng/mL and f/tPSA is ⬍25 ng/mL or tPSA ⬎10 ng/mL.
TABLE III. Distribution of Gleason score for assays at different thresholds Assay (ng/mL) tPSA ⬎4.0 f/tPSA* ⬍25 cPSA ⬎3.8 ⬎3.6 ⬎3.4 ⬎3.2 ⬎3.0
Gleason Score (n) 5
6
7
8
9
Total Cases (n)
5 4
226 183
111 89
25 25
1 1
368 302
5 5 5 5 5
194 201 216 225 228
101 104 109 112 112
25 25 25 25 25
1 1 1 1 1
326 336 356 368 371
For abbreviations, see Table I. * Biopsy if tPSA is 4 –10 ng/mL and f/tPSA is ⬍25 ng/mL or tPSA ⬎10 ng/mL.
TABLE IV. Base-case costs and disutilities Variable Cost ($)* Assay Total PSA Free/total PSA Complexed PSA TRUS biopsy Pathology Disutility False positive False negative
Base Case
10 40 20 100 68 0.99 0.93
Sensitivity (%), range
5–20 20–60 10–40 50–300 34–140 0.95–1.0 0.85–1.0
* Costs in US$ indexed to fiscal year 2001.
of 3.8 ng/mL was the dominant strategy, based on the underlying assumptions. A dominant strategy is an approach that is both more effective and less costly. We defined the cost of screening as the aggregate cost of phlebotomy, running the assay, and, for individuals whose PSA isoform levels were above the predetermined positive threshold, we included the downstream costs associated with an office-based TRUS biopsy and specimen review. 44
The average cost of screening for the dominant strategy was $138.93. The marginal cost (ie, the cost difference between 2 strategies) to the next most appropriate strategy, tPSA, was $15.00 (Table V). Using sensitivity analysis, the strategy of tPSA becomes dominant when the cost of cPSA is ⬎$35.0 (75% increase over current levels) or the cost of a prostate biopsy is ⬍$67.3 (60% reduction of current levels). UROLOGY 60 (Supplement 4A), October 2002
FIGURE 1. Receiver operating characteristic curves (ROCs) for total prostatespecific antigen (tPSA), complexed prostate-specific antigen (cPSA), and free/ total prostate-specific antigen (f/tPSA).
TABLE V. Cost-benefit analysis: nondominated strategies Strategy
Cost ($)
Benefit (utility)
cPSA 3.8 tPSA 4.0 cPSA 3.0
139.9 154.9 164.3
0.9908 0.9922 0.9923
Cost/Benefit ($/utility) 141 156 166
cPSA ⫽ complexed prostate-specific antigen; tPSA ⫽ total prostate-specific antigen. The strategies using cPSA 3.2, 3.4 and 3.6 ng/mL, as well as f/tPSA, were dominated and therefore the data are not reported.
Use of cPSA with the threshold of 3.8 ng/mL had a 17% lower biopsy rate and an increased positive predictive value (32.8% vs 31.0%) compared with tPSA 4.0 ng/mL. At the cPSA threshold of 3.8 ng/ mL, the test sensitivity is 81.0%. In comparison, tPSA test sensitivity is 92.4%. Given current consensus on the value of 92% sensitivity, a cPSA threshold of 3.0 ng/mL is necessary (sensitivity, 92.5%) to achieve similar test performance. At this level, there is a $9.40 marginal cost increase over tPSA, with ⬍1% fewer biopsies. The sited utilities for false-positive and false-negative results associated with each screening strategy are conservative estimates for the relative impact of an “unnecessary” or conversely, a “missed” biopsy. For false-positives, we assume minimal impact (utility of 0.99). For false-negatives, men were assigned an impact equivalent to that previously reported for asymptomatic stable prostate cancer (utility of 0.93). Wide variation of these base-case utilities using sensitivity analysis did not significantly change the model output. UROLOGY 60 (Supplement 4A), October 2002
DISCUSSION The introduction of new PSA isoforms has refined the clinician’s ability to identify men at a higher than expected risk of harboring prostate cancer. The relative improvement obtained by these new assays comes with the potential for marked changes in the cost structure of population-based screening. Using decision-analytic modeling and a large international dataset, we have determined that the least costly strategy for prostate cancer screening is use of cPSA with a positive threshold of 3.8 ng/mL. The primary determinant of cost for prostate cancer screening is the biopsy rate. Given the relatively small unit contribution of the cost of the serum assays, it is not surprising that the strategy that most effectively balances the positive predictive value and test specificity would be expected to dominate all other strategies. To our knowledge, there are no prior studies that have applied cost-benefit analytic techniques to direct comparisons of different PSA assays. Prior published studies have compared test characteris45
tics of tPSA, cPSA, and f/tPSA. Brawer et al.6 found in aggregate that cPSA performance was equivalent to f/tPSA and improved over tPSA. They concluded that cPSA could serve as a single-assay replacement of f/tPSA. Miller et al.3 found similar relations among the 3 isoforms. They too advocated for adoption of cPSA, especially for patients with a tPSA in the 2 to 10 ng/mL range. Finne et al.7 further corroborated the observations. In their Finnish-based study, cPSA demonstrated better specificity than either tPSA or f/tPSA within sensitivity ranges of 85% to 90%. In contrast to direct comparisons of different assays, Ross et al.8 used Markov decision modeling to compare alternative screening intervals and PSA thresholds. Their elegant analysis demonstrated that a strategy of standard use of PSA beginning at age 40 to 45 followed by biennial testing prevented more prostate cancer deaths with less PSA testing and fewer biopsies. In addition, they found that standard intervals of testing with lower thresholds of PSA (2.5 ng/mL vs 4.0 ng/mL) came at the price of 48% more PSA tests and 63% more biopsies. A conservative interpretation of these data is that lowering the threshold for biopsy or maintaining the current recommendations of annual screening intervals is illogical from a cost-effectiveness perspective. The investigators did not discuss the impact of increasing the PSA threshold. As such, an alternative conclusion may be that the optimal level of sensitivity for PSA has not been determined. From this perspective, it may be the case that increasing the accepted serum threshold for a screening PSA test may, in fact, prevent similar numbers of prostate cancer deaths with fewer biopsies. The results of the current study must be interpreted with several caveats. First, decision-analytic modeling is only as strong and generalizable as the probabilities on which it is based. We are confident the population-based dataset is an accurate and representative sample of the general population. The rates of disease and associated test characteristics approximate those previously published. Second, the cost estimates have been extrapolated from an analysis performed nearly 10 years ago. In substantiating these costs, we found little difference from those observed at our institution. We are therefore confident that although the estimates are
46
not perfect, the relative contribution of each component is accurate. Third, the modeling used was based on simple decision analysis rather than the more complex Markov technique. In this regard, the estimates of cost are for a single screening year, and do not account for the total lifetime costs an individual may accrue. Our choice of this technique is based on limited data on changes in cPSA and f/tPSA over time. CONCLUSION Using cPSA as a screening tool for prostate cancer is more cost-effective than f/t PSA (with a threshold of 25%). The most appropriate strategy for population-based screening may be the use of cPSA with a threshold of 3.8 ng/mL. The most appropriate sensitivity for PSA screening has not been determined. If, however, we must choose a sensitivity of 92% as a critical value, cPSA at 3.0 ng/mL could be used interchangeably with tPSA at 4.0 ng/mL. Further analysis with Markov modeling is warranted, but will be dependent on better understanding changes in expression of this marker over time. REFERENCES 1. Brawer MK: Prostate-specific antigen. Semin Surg Oncol 18: 3–9, 2000. 2. Stephan C, Jung K, Brux B, et al: ACT-PSA and complexed PSA elimination kinetics in serum after radical retropubic prostatectomy: proof of new complex forming of PSA after release into circulation. Urology 55: 560 –563, 2000. 3. Miller MC, O’Dowd GJ, Partin AW, et al: Contemporary use of complexed PSA and calculated percent free PSA for early detection of prostate cancer: impact of changing disease demographics. Urology 57: 1105–1111, 2001. 4. Earle CC, Chapman RH, Baker CS, et al: Systematic overview of cost-utility assessments in oncology. J Clin Oncol 18: 3302–3317, 2000. 5. Gustafsson O, Carlsson P, Norming U, et al: Cost-effectiveness analysis in early detection of prostate cancer: an evaluation of six screening strategies in a randomly selected population of 2,400 men. Prostate 26: 299 –309, 1995. 6. Brawer MK, Cheli CD, Neaman IE, et al: Complexed prostate specific antigen provides significant enhancement of specificity compared with total prostate specific antigen for detecting prostate cancer. J Urol 163: 1476 –1480, 2000. 7. Finne P, Zhang WM, Auvinen A, et al: Use of the complex between prostate specific antigen and alpha 1-protease inhibitor for screening prostate cancer. J Urol 164: 1956 – 1960, 2000. 8. Ross KS, Carter HB, Pearson JD, et al: Comparative efficiency of prostate-specific antigen screening strategies for prostate cancer detection. JAMA 284: 1399 –1405, 2000.
UROLOGY 60 (Supplement 4A), October 2002
AND
ALAN W. PARTIN
ABSTRACT Refinements in prostate-specific antigen (PSA) through the use of its derivatives have augmented early detection rates of prostate cancer. However, these improvements are coupled with relatively large increases in unit cost per detected cancer. We used decision-analytic modeling to determine the most appropriate PSA derivative for population-based screening. We constructed a decision-analytic model to determine the PSA derivative with the highest cost-benefit ratio for prostate cancer screening. We defined 5 screening strategies: total PSA (tPSA) 4.0 ng/mL; free PSA/tPSA (f/tPSA) in conjunction with tPSA; and complexed PSA (cPSA) 3.8, 3.4, and 3.0 ng/mL. Prostate cancer prevalence, false-positive rates, and false-negative rates for each test strategy were calculated from a database of 2138 men. The direct costs were obtained from literature review and our department of clinical chemistry. The derivative cPSA with a positive threshold of 3.8 ng/mL was the dominant strategy. The average cost of screening was $138.93. The strategy of tPSA became dominant when the cost of cPSA was ⬎$35.00 or the cost of a prostate biopsy was ⬍$67.30. To match the false-negative rate of tPSA 4.0 ng/mL, a cPSA threshold of 3.0 ng/mL is necessary (sensitivity 92.5%). At this level, the marginal cost increase over tPSA is $9.40. The dominant strategy for populationbased prostate cancer screening is use of cPSA with a positive threshold of 3.8 ng/mL. The use of cPSA with a threshold of 3.0 ng/mL identifies a similar number of cancers with fewer biopsies than tPSA at 4.0 ng/mL. UROLOGY 60 (Suppl 4A): 42–46, 2002. © 2002, Elsevier Science Inc.
P
rostate-specific antigen (PSA) has changed our definitions and management of prostate cancer. However, as a screening biomarker, total serum PSA (tPSA) has a poor positive predictive value because of a high false-positive rate.1 Recognition of different molecular forms of PSA (free PSA [fPSA] and complexed PSA [cPSA]) has led to incremental improvements in these test characteristics.2 Identification of appropriate positive thresholds for these PSA isoforms may allow clinicians to more accurately identify patients for biopsy and greatly diminish the numbers of unnecessary biopsies. The unmeasured corollary of the introduction of From the Brady Urological Institute, the Johns Hopkins Medical Institution, Baltimore Maryland, USA (LE, RWV, AWP); Robert Wood Johnson Clinical Scholars Program, the Johns Hopkins Medical Institution, Baltimore Maryland, USA (LE); Bayer Corporation, Tarrytown, New York, USA (CDC); and UroCor, Inc., Oklahoma City, Oklahoma, USA (SB) Reprint requests: Lars Ellison, MD, Brady Urological Institute, Johns Hopkins-RWJ Program, 600 North Wolfe Street, Carnegie 291, Baltimore, MD 21287. E-mail: [email protected]
42
© 2002, ELSEVIER SCIENCE INC. ALL RIGHTS RESERVED
improved and advanced biomarkers is increased cost of screening. The cost relation is complicated by the relative contribution of downstream diagnostic tests (specifically the transrectal ultrasound [TRUS]-guided biopsy). In addition, this cost relation is amplified by the necessary broad definition of the population at risk (men ⬎50 years old). We used (1) a large multi-institutional patient sample of men, aged 40 to 75 years, who underwent standard screening and prostate biopsy; (2) cost and utility data from the literature; and (3) decision-analytic modeling. This information was used to determine, from a societal perspective, which PSA isoform and serum threshold of positivity is most cost-beneficial for prostate cancer screening. METHODS PATIENT SAMPLE We merged 2 contemporary datasets, the first from UroCor Labs (Oklahoma City, OK) and the second from Bayer Diagnostics (Tarrytown, NY). The UroCor data have been previously described.3 For the Bayer dataset, the study sites were: 0090-4295/02/$22.00 PII S0090-4295(02)01694-1
TABLE I. Demographics of the combined UroCor* and Bayer Diagnostics† datasets No Cancer (n ⴝ 1519), mean (SE) Age (yr) tPSA (ng/mL) cPSA (ng/mL) f/tPSA (ng/mL)
64.2 7.24 5.60 18.87
(0.19) (0.11) (0.09) (0.29)
Cancer (n ⴝ 619), mean (SE) 66.6 10.49 8.40 16.16
(0.29) (1.55) (1.38) (0.38)
P-value ⬍0.001 0.001 0.002 ⬍0.001
cPSA ⫽ complexed prostate-specific antigen; f/tPSA ⫽ free to total prostate-specific antigen; tPSA ⫽ total prostate-specific antigen. * Oklahoma, OK. † Tarrytown, NY.
Johns Hopkins Hospital (Baltimore, MD), Northwest Prostate Institute (Seattle, WA), New York University (New York, NY), Stanford University (Stanford, CA), University of Innsbruck (Innsbruck, Austria), MD Anderson Cancer Center (Houston, TX), and Wyoming Research Foundation (Cheyenne, WY). Subjects were enrolled when recommended for a prostate biopsy by the physician according to established local practice and when the patient consented to TRUS-guided prostate biopsy. The biopsy procedure consisted of ⱖ10 cores of prostate tissue, including a minimum of 6 systematic sectors and 4 lateral cores (2 from each side). All patients underwent a digital rectal examination (DRE). Patients were excluded from the study if they had a personal history of prostate cancer, or a previous history of a transurethral prostatic resection. Patients taking any medication that could decrease levels of serum PSA, such as estrogen, finasteride, or a course of ⱖ7 days of quinolone antibiotic therapy within 30 days of biopsy, were excluded from the study. In addition, patients taking any medication or food supplement that could increase serum PSA level, such as but not limited to dehydroepiandrosterone or testosterone, were excluded. The patient’s age, race, and family history of prostate cancer were obtained. Patients had blood drawn ⱕ1 month before biopsy. Blood was drawn either before DRE or ⱖ1 week and ⱕ1 month after DRE. If biopsy results were positive for prostate cancer, then Gleason score and clinical stage were provided, and if prostatectomy was performed, pathologic stage of disease was determined. Patients who underwent a previous biopsy or multiple biopsies and met the above inclusion criteria were enrolled into the study. This study was performed in compliance with the requirements of the respective institutional review boards. Serum samples were processed immediately upon collection of blood and kept at 4°C ⱕ8 hours before being frozen at ⫺70°C. Serum samples were subsequently shipped on dry ice to Johns Hopkins Hospital Clinical Chemistry Laboratory for testing of Bayer Immuno 1 cPSA and tPSA methods (Bayer Diagnostics) and Beckman Access fPSA and tPSA methods (Beckman Inc, San Diego, CA). Serum samples were tested immediately upon sample thaw. Individual determinations were obtained on each specimen tested. The same serum sample was used to determine Bayer tPSA, cPSA, and Beckman fPSA and tPSA concentrations. In aggregate, complete records of patient age, TRUS biopsy, tPSA, cPSA, fPSA values, and DRE results were available for 3757 men. Individuals were excluded if they were ⬎75 years of age or were coded with a suspicious or abnormal DRE. The final dataset included 2138 men (Table I). Prostate cancer prevalence as well as false-positive rates and false-negative rates for each test strategy were calculated from the database (Table II). The distributions of Gleason scores for detected cases by test strategy are shown in Table III. UROLOGY 60 (Supplement 4A), October 2002
UTILITIES AND COSTS A “utility” is a numerical value between 0 and 1, which represents the relative value of a health state (eg, living with the diagnosis of localized prostate cancer) as compared with perfect health. Utility and cost estimates (Table IV) were derived from literature review.4,5 Specifically, we identified utilities associated with biomarker-defined false-positive and false-negative status. Included among the screening costs were: (1) office and staffing costs, (2) the cost of running the various assays, (3) TRUS biopsy costs, (4) pathologic review costs, and (5) indirect costs. Values were indexed to US dollars (fiscal year 2001).
MODELING We constructed a decision-analytic model to determine the PSA derivative with the highest cost-benefit ratio for prostate cancer screening. We defined 5 treatment strategies: tPSA with a positive threshold of 4.0 ng/mL; fPSA in conjunction with tPSA (f/tPSA positive threshold ⬍25) when tPSA is 4 to 10 ng/mL; and cPSA with a positive thresholds of 3.8, 3.4, and 3.0 ng/mL. Sensitivity analyses were performed on all variables to determine the impact of reasonable variation in the baseline assumptions on the cost-benefit ratio of the different strategies. Statistical analysis was done with the STATA version 7.0 software package (Stata Corporation, College Station, TX), and modeling was completed with DATA version 3.5 (TreeAge Software, Inc., Williamstown, MA)
RESULTS Table I summarizes baseline demographic data for the 2138 men in the sample. Altogether, 29% of men were found to have evidence of cancer at biopsy. The mean age of men with biopsy evidence of cancer was slightly older than those without evidence of cancer. As expected, men with a diagnosis of cancer had higher mean tPSA and cPSA values and lower f/tPSA values. Receiver operating curves (ROC) were generated for the 3 isoforms of PSA. The area under the curve for each is reported in Figure 1. Analysis of the ROC curves shows them to be significantly different (P ⫽ 0.03). Isoforms cPSA and tPSA appear to have better performance in minimizing false-negatives, whereas f/tPSA appears to have better performance in minimizing false-positives. From the model, cPSA with a positive threshold 43
TABLE II. Test characteristics for assays at different threshold levels Biopsy Assay (ng/mL) tPSA ⬎4.0 f/tPSA* ⬍25 cPSA ⬎3.8 ⬎3.6 ⬎3.4 ⬎3.2 ⬎3.0
n
Rate (%)
Specificity (%)
Sensitivity (%)
1844 1577
86.3 73.7
16.3 29.0
92.4 81.2
1526 1613 1721 1784 1836
71.4 75.4 80.5 83.4 85.7
32.6 28.2 22.8 19.6 16.9
81.1 84.3 88.5 91.3 92.6
For abbreviations, see Table I. * Biopsy if tPSA is 4 –10 ng/mL and f/tPSA is ⬍25 ng/mL or tPSA ⬎10 ng/mL.
TABLE III. Distribution of Gleason score for assays at different thresholds Assay (ng/mL) tPSA ⬎4.0 f/tPSA* ⬍25 cPSA ⬎3.8 ⬎3.6 ⬎3.4 ⬎3.2 ⬎3.0
Gleason Score (n) 5
6
7
8
9
Total Cases (n)
5 4
226 183
111 89
25 25
1 1
368 302
5 5 5 5 5
194 201 216 225 228
101 104 109 112 112
25 25 25 25 25
1 1 1 1 1
326 336 356 368 371
For abbreviations, see Table I. * Biopsy if tPSA is 4 –10 ng/mL and f/tPSA is ⬍25 ng/mL or tPSA ⬎10 ng/mL.
TABLE IV. Base-case costs and disutilities Variable Cost ($)* Assay Total PSA Free/total PSA Complexed PSA TRUS biopsy Pathology Disutility False positive False negative
Base Case
10 40 20 100 68 0.99 0.93
Sensitivity (%), range
5–20 20–60 10–40 50–300 34–140 0.95–1.0 0.85–1.0
* Costs in US$ indexed to fiscal year 2001.
of 3.8 ng/mL was the dominant strategy, based on the underlying assumptions. A dominant strategy is an approach that is both more effective and less costly. We defined the cost of screening as the aggregate cost of phlebotomy, running the assay, and, for individuals whose PSA isoform levels were above the predetermined positive threshold, we included the downstream costs associated with an office-based TRUS biopsy and specimen review. 44
The average cost of screening for the dominant strategy was $138.93. The marginal cost (ie, the cost difference between 2 strategies) to the next most appropriate strategy, tPSA, was $15.00 (Table V). Using sensitivity analysis, the strategy of tPSA becomes dominant when the cost of cPSA is ⬎$35.0 (75% increase over current levels) or the cost of a prostate biopsy is ⬍$67.3 (60% reduction of current levels). UROLOGY 60 (Supplement 4A), October 2002
FIGURE 1. Receiver operating characteristic curves (ROCs) for total prostatespecific antigen (tPSA), complexed prostate-specific antigen (cPSA), and free/ total prostate-specific antigen (f/tPSA).
TABLE V. Cost-benefit analysis: nondominated strategies Strategy
Cost ($)
Benefit (utility)
cPSA 3.8 tPSA 4.0 cPSA 3.0
139.9 154.9 164.3
0.9908 0.9922 0.9923
Cost/Benefit ($/utility) 141 156 166
cPSA ⫽ complexed prostate-specific antigen; tPSA ⫽ total prostate-specific antigen. The strategies using cPSA 3.2, 3.4 and 3.6 ng/mL, as well as f/tPSA, were dominated and therefore the data are not reported.
Use of cPSA with the threshold of 3.8 ng/mL had a 17% lower biopsy rate and an increased positive predictive value (32.8% vs 31.0%) compared with tPSA 4.0 ng/mL. At the cPSA threshold of 3.8 ng/ mL, the test sensitivity is 81.0%. In comparison, tPSA test sensitivity is 92.4%. Given current consensus on the value of 92% sensitivity, a cPSA threshold of 3.0 ng/mL is necessary (sensitivity, 92.5%) to achieve similar test performance. At this level, there is a $9.40 marginal cost increase over tPSA, with ⬍1% fewer biopsies. The sited utilities for false-positive and false-negative results associated with each screening strategy are conservative estimates for the relative impact of an “unnecessary” or conversely, a “missed” biopsy. For false-positives, we assume minimal impact (utility of 0.99). For false-negatives, men were assigned an impact equivalent to that previously reported for asymptomatic stable prostate cancer (utility of 0.93). Wide variation of these base-case utilities using sensitivity analysis did not significantly change the model output. UROLOGY 60 (Supplement 4A), October 2002
DISCUSSION The introduction of new PSA isoforms has refined the clinician’s ability to identify men at a higher than expected risk of harboring prostate cancer. The relative improvement obtained by these new assays comes with the potential for marked changes in the cost structure of population-based screening. Using decision-analytic modeling and a large international dataset, we have determined that the least costly strategy for prostate cancer screening is use of cPSA with a positive threshold of 3.8 ng/mL. The primary determinant of cost for prostate cancer screening is the biopsy rate. Given the relatively small unit contribution of the cost of the serum assays, it is not surprising that the strategy that most effectively balances the positive predictive value and test specificity would be expected to dominate all other strategies. To our knowledge, there are no prior studies that have applied cost-benefit analytic techniques to direct comparisons of different PSA assays. Prior published studies have compared test characteris45
tics of tPSA, cPSA, and f/tPSA. Brawer et al.6 found in aggregate that cPSA performance was equivalent to f/tPSA and improved over tPSA. They concluded that cPSA could serve as a single-assay replacement of f/tPSA. Miller et al.3 found similar relations among the 3 isoforms. They too advocated for adoption of cPSA, especially for patients with a tPSA in the 2 to 10 ng/mL range. Finne et al.7 further corroborated the observations. In their Finnish-based study, cPSA demonstrated better specificity than either tPSA or f/tPSA within sensitivity ranges of 85% to 90%. In contrast to direct comparisons of different assays, Ross et al.8 used Markov decision modeling to compare alternative screening intervals and PSA thresholds. Their elegant analysis demonstrated that a strategy of standard use of PSA beginning at age 40 to 45 followed by biennial testing prevented more prostate cancer deaths with less PSA testing and fewer biopsies. In addition, they found that standard intervals of testing with lower thresholds of PSA (2.5 ng/mL vs 4.0 ng/mL) came at the price of 48% more PSA tests and 63% more biopsies. A conservative interpretation of these data is that lowering the threshold for biopsy or maintaining the current recommendations of annual screening intervals is illogical from a cost-effectiveness perspective. The investigators did not discuss the impact of increasing the PSA threshold. As such, an alternative conclusion may be that the optimal level of sensitivity for PSA has not been determined. From this perspective, it may be the case that increasing the accepted serum threshold for a screening PSA test may, in fact, prevent similar numbers of prostate cancer deaths with fewer biopsies. The results of the current study must be interpreted with several caveats. First, decision-analytic modeling is only as strong and generalizable as the probabilities on which it is based. We are confident the population-based dataset is an accurate and representative sample of the general population. The rates of disease and associated test characteristics approximate those previously published. Second, the cost estimates have been extrapolated from an analysis performed nearly 10 years ago. In substantiating these costs, we found little difference from those observed at our institution. We are therefore confident that although the estimates are
46
not perfect, the relative contribution of each component is accurate. Third, the modeling used was based on simple decision analysis rather than the more complex Markov technique. In this regard, the estimates of cost are for a single screening year, and do not account for the total lifetime costs an individual may accrue. Our choice of this technique is based on limited data on changes in cPSA and f/tPSA over time. CONCLUSION Using cPSA as a screening tool for prostate cancer is more cost-effective than f/t PSA (with a threshold of 25%). The most appropriate strategy for population-based screening may be the use of cPSA with a threshold of 3.8 ng/mL. The most appropriate sensitivity for PSA screening has not been determined. If, however, we must choose a sensitivity of 92% as a critical value, cPSA at 3.0 ng/mL could be used interchangeably with tPSA at 4.0 ng/mL. Further analysis with Markov modeling is warranted, but will be dependent on better understanding changes in expression of this marker over time. REFERENCES 1. Brawer MK: Prostate-specific antigen. Semin Surg Oncol 18: 3–9, 2000. 2. Stephan C, Jung K, Brux B, et al: ACT-PSA and complexed PSA elimination kinetics in serum after radical retropubic prostatectomy: proof of new complex forming of PSA after release into circulation. Urology 55: 560 –563, 2000. 3. Miller MC, O’Dowd GJ, Partin AW, et al: Contemporary use of complexed PSA and calculated percent free PSA for early detection of prostate cancer: impact of changing disease demographics. Urology 57: 1105–1111, 2001. 4. Earle CC, Chapman RH, Baker CS, et al: Systematic overview of cost-utility assessments in oncology. J Clin Oncol 18: 3302–3317, 2000. 5. Gustafsson O, Carlsson P, Norming U, et al: Cost-effectiveness analysis in early detection of prostate cancer: an evaluation of six screening strategies in a randomly selected population of 2,400 men. Prostate 26: 299 –309, 1995. 6. Brawer MK, Cheli CD, Neaman IE, et al: Complexed prostate specific antigen provides significant enhancement of specificity compared with total prostate specific antigen for detecting prostate cancer. J Urol 163: 1476 –1480, 2000. 7. Finne P, Zhang WM, Auvinen A, et al: Use of the complex between prostate specific antigen and alpha 1-protease inhibitor for screening prostate cancer. J Urol 164: 1956 – 1960, 2000. 8. Ross KS, Carter HB, Pearson JD, et al: Comparative efficiency of prostate-specific antigen screening strategies for prostate cancer detection. JAMA 284: 1399 –1405, 2000.
UROLOGY 60 (Supplement 4A), October 2002