Should Modest Elevations in Prostate-Specific Antigen, International Prostate Symptom Score, or Their Rates of Increase Over Time be Used as Surrogate Measures of Incident Benign Prostatic Hyperplasia?

Abstract

Although surrogate measures of benign prostatic hyperplasia (BPH) are often used in epidemiologic studies, their performance characteristics are unknown. Using data from the Prostate Cancer Prevention Trial (n = 5,986), we evaluated prostate-specific antigen (PSA), International Prostate Symptom Score (IPSS), and their rates of change as predictors of incident BPH. BPH (n = 842 cases) was defined as medical or surgical treatment or at least 2 IPSS of 15 or higher. Proportional hazards models were used to measure the associations of baseline PSA, IPSS, and their velocities over 2 years with BPH risk, and time-dependent receiver-operating characteristic curves were used to measure their discriminatory performance. Unit increases in PSA, IPSS, and IPSS velocity were associated with 34%, 35%, and 29% (all P < 0.001) increases in BPH risk, respectively. The areas under the receiver-operating characteristic curves were significantly greater than 0.5 for PSA (0.58, 95% confidence interval (CI): 0.56, 0.60), IPSS (0.77, 95% CI: 0.75, 0.78), and IPSS velocity (0.63, 95% CI: 0.61, 0.65); however there were no cut points at which sensitivity and specificity were both above 75%. We concluded that moderate elevations in PSA, IPSS, or their rates of change should not be used as surrogate measures of incident BPH.

Benign prostatic hyperplasia (BPH) is a common medical condition that affects 40%–50% of men by the time they reach 50 years of age and nearly 80% of men by the time they reach 70 years of age (1–4). BPH is characterized by hyperplasia of the stromal and, to a lesser extent, epithelial regions of the prostate, which can result in a constellation of lower urinary tract symptoms (LUTS) that affect quality of life and prompt many men to seek treatment (5–7). A principal challenge in epidemiologic studies of BPH is the definition of disease. Histologically, BPH is characterized by the presence of hyperplastic nodules in the prostate. Clinically, BPH is typically diagnosed based on a combination of criteria, including reduced urinary flow, increased prostate size, LUTS, and response to medical treatment; however, there is no consensus definition for the absence or presence of disease. Epidemiologic studies, for which neither histologic nor comprehensive clinical diagnoses are feasible, usually define BPH based on either receipt of medical or surgical treatment or LUTS that are sufficiently severe to motivate a patient to seek medical or surgical treatment (8).

Because of the challenges involved in diagnosing BPH and the large sample sizes needed to study incident BPH in longitudinal studies, investigators in some studies have used surrogate markers of BPH under the assumption that these have a high predictive value for clinically relevant disease. Two surrogate markers of BPH are based on prostate-specific antigen (PSA), which in healthy men is an indirect measure of prostate size (9), and the International Prostate Symptom Score (IPSS), which measures the severity of LUTS on a scale ranging from 0 to 35; either a threshold value that is still within the “normal” range or a high rate of increase in these markers could identify men who will develop clinically significant BPH. However, no studies have formally examined the predictive ability of these markers as surrogate measures of BPH.

In the present study, we evaluated the performance characteristics of PSA and IPSS as surrogate markers of BPH, based on single measures or on multiple measures over time. Specifically, we examined whether any PSA values less than 3 ng/mL, an IPSS less than 14, or their rates of increase over 2 years had sufficiently high sensitivity and specificity to allow their use as surrogate measures of BPH for epidemiologic research. The present study is based on data from the Prostate Cancer Prevention Trial (PCPT), which over the course of the 7-year trial collected detailed information on BPH-related surgical procedures, medication use, and LUTS. If valid as surrogate measures of BPH, PSA-based or IPSS-based surrogate endpoints could be further evaluated as candidate surrogate endpoints for use in intervention trials for the primary prevention of BPH.

MATERIALS AND METHODS

Data are from the PCPT, a randomized, placebo-controlled trial to test whether finasteride reduced the risk of prostate cancer. In this trial, 18,880 men who were 55 years of age or older and had a normal digital rectal exam, a PSA concentration less than 3 ng/mL, no history of prostate cancer, and an IPSS less than 20 were randomized to receive finasteride (5 mg/day) or placebo. Enrollment occurred over 3 years, and each participant was followed for up to 7 years from their date of recruitment (10).

Data collection

Extensive medical data, including participant-reported physician diagnosis of and medical and surgical treatment for BPH, were collected at baseline and at each subsequent contact. In addition, at recruitment, randomization, and each annual clinic visit, participants completed the IPSS, a 7-item self-administered questionnaire assessing the frequency of LUTS (incomplete bladder emptying, frequent urination, intermittency, urgency, weak urinary stream, hesitancy, and nocturia) over the past month. The IPSS is a validated and reliable quantitative instrument for measuring male urinary symptoms (11) and is the primary method by which BPH symptoms are assessed in clinical practice (12). PSA concentrations were measured at baseline and each annual clinic visit. At baseline, data on age, race/ethnicity, physical activity level, and history of smoking were collected using self-administered questionnaires, and height and weight were measured by clinic staff. Body mass index was calculated as weight in kilograms divided by height in meters squared.

Definition of incident symptomatic BPH

Incident symptomatic BPH was defined as the first event of either surgical or medical treatment or sustained, clinically significant BPH symptoms. Surgical treatment included transurethral prostatectomy, balloon dilation, and laser or open prostatectomy. Medical treatments included 5α-reductase inhibitors (finasteride) and uroselective α-blockers (tamsulosin), as well as the use of nonspecific α-blocker therapy (doxazosin, prazosin, or terazosin) if a physician diagnosis of BPH or elevated IPSS measures (a single IPSS >14 or any 2 reports of an IPSS ≥12) were reported preceding medication use. Men who reported use of nonspecific α-blocker therapy without concomitant evidence of BPH and men who reported a physician diagnosis of BPH alone in the absence of treatment or symptoms were not included as BPH events. We defined the onset of clinically significant BPH symptoms as the second report of an IPSS of 15 or greater.

Definition of surrogate and criterion measures of BPH

Baseline values for PSA and IPSS were used to evaluate whether a threshold value within the “normal” range could serve as a surrogate measure of BPH. In these analyses, the cumulative 7-year incidence of BPH, as described above, was used as the criterion for evaluation. The slopes of PSA and IPSS measured at baseline and years 1 and 2 of the trial were also used to evaluate whether high PSA or IPSS velocity could serve as a surrogate measure. In these analyses, the 5-year cumulative incidence of BPH starting at year 2 of the trial was used as the criterion for evaluation.

Study population

Analyses were restricted to the 9,455 placebo-arm participants who had a baseline PSA concentration of 3 ng/dL or lower. Men were excluded if there was evidence of BPH at baseline (history of medical or surgical treatment for BPH (n = 735), an IPSS ≥15 (n = 1,306), or a self-reported history of BPH (n = 1,429)), leaving 5,986 men eligible for this study. Analyses based on baseline PSA concentration and IPSS were conducted in all 5,986 eligible participants. Analyses of PSA velocity were restricted to the 4,717 men with a minimum of 3 consecutive PSA measurements (at baseline and years 1 and 2) and 2 or more additional years of follow-up. Analyses of IPSS velocity were restricted to the 4,729 men with a minimum of 3 consecutive IPSS measurements (at baseline and years 1 and 2) and 2 or more years of follow-up.

Statistical analyses

Cox proportional hazards models were used to measure the associations of baseline values and rates of increase in continuous and categorical PSA and IPSS with the risk of incident BPH. Linearity for all measures was confirmed by examining the trend in hazard ratios across deciles of each marker. We used time-dependent receiver-operating characteristic (ROC) curves, which take into account the time-dependent nature of incident BPH, to measure the performance characteristics (sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV)) of the PSA- and IPSS-based measures as surrogate endpoints of BPH. Because the length of observation for incident BPH may be censored or subject to the competing risk of a prostate cancer diagnosis (we had no information from participants after cancer was diagnosed), we used methods for estimating time-dependent ROC curves that accounted for competing risks. In this setting, the sensitivity of a marker is a function of the time t from the measurement of the marker to the time of diagnosis; it is defined as the probability that the marker will be positive if that diagnosis occurs within time t of the marker measurement. Conversely, specificity at time t is defined as the probability that the marker will be negative if neither the event of interest nor the competing event has occurred by time t. Similarly, the PPV is defined as the probability that the event of interest will occur by time t if the marker is positive, and the NPV is the probability that neither the event of interest nor the competing event will occur by time t if the marker is negative. Estimation of the time-dependent ROC curves and PPVs/NPVs was conducted using the algorithm of Zheng et al. (13) based on a cause-specific (14) proportional hazards model for the risks of BPH and prostate cancer. A nonparametric method was used for models (baseline IPSS and IPSS velocity in the total sample and IPSS velocity among men with baseline IPSS <8) that did not meet the proportional hazards assumptions (13). Confidence intervals for the areas under the ROC curve (AUC) were derived based on 1,000 bootstrap replications. For analyses based on baseline PSA and IPSS, follow-up was started at baseline, and for rates of increase in PSA and IPSS, follow-up was started at year 2. Analyses were conducted in the full study sample and additionally stratified by age (<65 years vs. ≥65 years) and by baseline IPSS (0–7 vs. 8–13) to consider whether the performance of surrogate endpoints differed by whether the men had already developed moderate LUTS. Analyses of PSA velocity were also stratified by baseline PSA (<1.0, 1.0–1.9, ≥2.0). There are no set cut points for the sensitivity and specificity of proposed markers that would make it a useful surrogate measure for BPH; however, our judgment would be a conservative minimum of 80% sensitivity and 90% specificity.

RESULTS

Table 1 shows baseline demographic and health-related characteristics of the study sample. Participants were mostly white, overweight or obese, inactive, and nonsmokers. The mean age was 62.5 years. Mean PSA concentration (1.2 ng/mL; standard deviation, 0.7) was low, and only 14% of participants had values greater than 2.0 ng/mL. The mean PSA velocity was 0.13 ng/mL/year, and only 16% had increases greater than 0.25 ng/mL/year. Mean IPSS was moderately low (4.8; standard deviation, 2.9), and only 17% had values consistent with moderate symptoms (IPSS ≥8); mean IPSS velocity was 0.77 units/year, and 26.0% had an IPSS velocity greater than 1.5 units/year.

Table 1.

Distribution of Demographic and Lifestyle Characteristics in the Prostate Cancer Prevention Trial Placebo Arm, 1993–2003

Characteristic	Mean (SD)	No.	%
Age, years	62.5 (5.5)
54–59		2,116	35.3
60–64		1,857	31.0
≥65		2,013	33.6
Race/ethnicity
White		5,506	92.0
Black		231	3.9
Hispanic		148	2.5
Other		98	1.6
Body mass indexa	27.8 (4.1)
<25		1,467	24.8
25–29		3,057	51.6
≥30		1,398	23.6
Physical activity level
Sedentary		996	16.7
Light		2,561	43.0
Moderate		1,829	30.7
High		568	9.6
Diabetes
Yes		327	5.5
No		5,659	94.5
Current smoker
Yes		507	8.5
No		5,476	91.5
Incident BPH
Yes		842	14.1
No		5,144	85.9
Interim prostate cancerb
Yes		291	4.9
No		5,695	95.1
Baseline PSA, ng/mL	1.2 (0.7)
<1.0		2,785	46.5
1.0–1.9		2,242	37.5
>2.0		959	16.0
PSA velocity, ng/mL/yearc	0.13 (1.81)
<0		1,735	34.0
0–0.24		2,573	50.5
≥0.25		792	15.5
Baseline IPSS	4.8 (2.9)
<4.0		2,517	42.0
4.0–7.5		2,443	40.8
≥8.0		1,026	17.1
IPSS velocity, units/yeard	0.59 (1.77)
<0		2,155	37.5
0–1.49		2,095	36.5
≥1.50		1,492	26.0

Abbreviations: BPH, benign prostatic hyperplasia; IPSS, International Prostate Symptom Score; PSA, prostate-specific antigen; SD, standard deviation.

^a Weight (kg)/height (m)².

^b Cancers detected during the Prostate Cancer Prevention Trial as a result of abnormal digital rectal examination or PSA concentration.

^c Restricted to men with a minimum of 3 consecutive PSA measurements between baseline and year 2 and a minimum of 2 years follow-up.

^d Restricted to men with a minimum of 2 IPSS measurements between baseline and year 2 and a minimum of 2 years follow-up.

Table 2 shows the raw and covariate-adjusted associations of baseline PSA, baseline IPSS, and their velocities with BPH risk. Both baseline PSA concentration and IPSS were strongly and significantly associated with increased risk of BPH; each 1-ng/mL increase in baseline PSA was associated with a 34% (P < 0.0001) increase in risk, and each 1-unit increase in IPSS was associated with a 35% (P < 0.0001) increase in risk. The rate of increase in IPSS between baseline and year 2 (IPSS velocity) was also associated with an increase in BPH risk; each 1-unit increase in slope was associated with a 29% increase in risk (95% confidence interval (CI): 26, 32; P < 0.0001). Only increases of 0.25 ng/mL/year or more in PSA velocity were associated with an increase in risk, and the association of continuous PSA velocity with BPH risk did not reach statistical significance (P = 0.06).

Table 2.

Associations of Prostate-specific Antigen and International Prostate Symptom Score With Benign Prostatic Hyperplasia Risk in the Prostate Cancer Prevention Trial Placebo Arm, 1993–2003

Surrogate Measure of BPH	No. of BPH Cases	Person-years	Hazard Ratio	95% CI	Hazard Ratioa	95% CI
Baseline PSA, ng/mL
<1.0	331	14,220	1.00		1.00
1.0–1.9	323	11,079	1.27	1.09, 1.47	1.21	1.04, 1.41
≥2.0	188	4,600	1.80	1.51, 2.15	1.67	1.39, 2.00
P for trend			<0.0001		<0.0001
Continuous	842	29,898	1.39	1.27, 1.52	1.34	1.22, 1.47
PSA velocity, ng/mL/yearb
<0	203	6,255	1.00		1.00
0–0.24	297	9,408	0.97	0.81, 1.16	0.97	0.81, 1.16
≥0.25	113	2,572	1.36	1.08, 1.72	1.32	1.05, 1.67
P for trend			0.04		0.06
Continuous	613	18,519	1.06	0.97, 1.09	1.03	0.97, 1.10
Baseline IPSS, units
0–3	121	13,343	1.00		1.00
4–7	354	12,317	3.22	2.62, 3.95	3.10	2.51, 3.81
8–13	367	4,238	10.15	8.26, 12.47	9.62	7.82, 11.85
P for trend			<0.0001		<0.0001
Continuous	842	29,898	1.36	1.33, 1.39	1.35	1.32, 1.38
IPSS velocity, unit/yearc
<0	142	7,470	1.00		1.00
0–1.4	187	7,340	1.34	1.08, 1.67	1.32	1.06, 1.64
≥1.5	287	3,526	4.21	3.45, 5.16	4.13	3.37, 5.05
P for trend			<0.0001		<0.0001
Continuous	616	18,513	1.45	1.40, 1.50	1.45	1.40, 1.51

Abbreviations: BPH, benign prostatic hyperplasia; CI, confidence interval; IPSS, International Prostate Symptom Score; PSA, prostate-specific antigen.

^a Adjusted for age (linear), race (white, black, and other), and body mass index (linear).

^b Restricted to men with a minimum of 3 consecutive PSA measurements between baseline and year 2 and a minimum of 2 years of follow-up.

^c Restricted to men with a minimum of 2 IPSS measurements between baseline and year 2 and a minimum of 2 years of follow-up.

The ROC curves for PSA and PSA velocity in the full population are given in Figure 1. The AUCs that resulted from using baseline PSA to predict the 7-year incidence of BPH and from using PSA velocity to predict the 5-year incidence of BPH were 0.58 (95% CI: 0.56, 0.60) and 0.51 (95% CI: 0.50, 0.52), respectively (Figure 1). The sensitivities and specificities at selected cut points are shown in Table 3. There were no cut point values for either baseline PSA or PSA velocity for which both sensitivity and specificity were sufficiently high to consider these measures as surrogates for incident BPH. For example, at 80% sensitivity for baseline PSA (cut point of 0.60 ng/mL), specificity was 22%, and at 80% sensitivity for PSA velocity (cut point of −0.046 ng/mL/year), specificity was 20%. AUCs for PSA and PSA velocity stratified by baseline IPSS and age did not differ by more than 0.02, and AUCs for PSA velocity stratified by PSA at baseline did not differ by more than 0.01. The PPVs for baseline PSA and PSA velocity were generally low (<25% for all cut points), whereas the NPVs for both baseline PSA and PSA velocity were all above 75% (Table 3).

Figure 1.

Receiver-operating characteristic curves created when using prostate-specific antigen (PSA) concentration (ng/mL) and International Prostate Symptom Score (IPSS) to predict incident symptomatic benign prostatic hyperplasia (BPH) in the Prostate Cancer Prevention Trial placebo arm, 1993–2003. Men with a PSA level of 4 ng/mL or higher were excluded from participation in the Prostate Cancer Prevention Trial. A) Baseline PSA used to predict the 7-year incidence of BPH (area under the receiver-operating characteristic curve (AUC) = 0.58, 95% confidence interval (CI): 0.56, 0.60). B) PSA velocity used to predict the 5-year incidence of BPH (AUC = 0.51, 95% CI: 0.50, 0.52). Data were limited to men for whom we had complete data (measurements at the baseline, year 1, and year 2 visits). C) Baseline IPSS used to predict the 7-year incidence of BPH (AUC = 0.77, 95% CI: 0.76, 0.79). D) IPSS velocity used to predict the 5-year incidence of BPH (AUC = 0.68, 95% CI: 0.65, 0.70). Data were limited to men for whom we had complete data (measurements at the baseline, year 1, and year 2 visits).

Table 3.

Performance Characteristics of Baseline Prostate-specific Antigen, International Prostate Symptom Score, and Their Rates of Change Over 2 Years, in the Prostate Cancer Prevention Trial Placebo Arm, 1993–2003

Surrogate Measure of BPH	No. of BPH Cases	Incidence Ratea	Sensitivity	Specificity	PPV	NPV
Baseline PSA, ng/mL
Total	842	28.2
≥0.5	758	25.3	0.896	0.147	0.188	0.836
≥1.0	511	17.1	0.607	0.505	0.207	0.819
≥1.5	323	10.8	0.371	0.746	0.227	0.802
≥2.0	188	6.2	0.215	0.876	0.244	0.787
≥2.5	90	3.0	0.103	0.950	0.259	0.773
PSA velocity, ng/mL/year
Total	613	33.1
≥0	410	22.1	0.662	0.341	0.151	0.796
≥0.025	360	19.4	0.570	0.433	0.151	0.796
≥0.05	300	16.2	0.456	0.548	0.151	0.796
≥0.10	231	12.5	0.333	0.671	0.151	0.795
≥0.20	140	7.6	0.200	0.803	0.152	0.795
≥0.30	83	4.5	0.124	0.878	0.152	0.795
≥0.40	58	3.1	0.085	0.917	0.153	0.795
Baseline IPSS
Total	842	28.2
≥1	835	27.9	0.991	0.054	0.190	0.842
≥2	805	26.9	0.956	0.181	0.206	0.870
≥3	771	25.8	0.915	0.674	0.232	0.873
≥4	721	24.1	0.853	0.483	0.268	0.866
≥5	649	21.7	0.764	0.619	0.306	0.855
≥6	563	18.8	0.660	0.730	0.349	0.840
≥7	477	16.0	0.556	0.819	0.402	0.826
≥8	367	12.3	0.420	0.887	0.446	0.807
≥9	269	9.0	0.304	0.936	0.506	0.792
≥10	178	6.0	0.203	0.968	0.583	0.779
≥11	96	3.2	0.108	0.984	0.598	0.767
≥12	41	1.4	0.045	0.994	0.604	0.760
IPSS velocity, units/year
Total	616	33.3
≥0	474	25.6	0.774	0.406	0.182	0.864
≥0.50	407	22.0	0.662	0.595	0.218	0.857
≥1.00	345	18.6	0.564	0.722	0.259	0.848
≥1.50	287	15.5	0.454	0.812	0.293	0.840
≥2.00	232	12.5	0.366	0.877	0.339	0.832
≥2.50	190	10.3	0.295	0.925	0.402	0.828

Abbreviations: BPH, benign prostatic hyperplasia; IPSS, International Prostate Symptom Score; NPV, negative predictive value; PPV, positive predictive value; PSA, prostate-specific antigen.

^a Per 1,000 person-years.

The time-dependent ROC curves created when using baseline IPSS and IPSS velocity to predict incident BPH in the full population (baseline IPSS <14) are shown in Figure 1. The AUCs created by using baseline IPSS to predict the 7-year incidence of BPH and IPSS velocity to predict the 5-year incidence of BPH were 0.77 (95% CI: 0.76, 0.79) and 0.68 (95% CI: 0.65, 0.70), respectively. Table 3 shows the sensitivities and specificities at selected cut points. Although the AUCs are higher than those for PSA, there were no cut point values for either baseline IPSS or IPSS velocity that yielded sufficiently high sensitivities and specificities to consider these measures as surrogates for BPH. For example, at 80% sensitivity for baseline IPSS (cut point of 4.5), specificity was 55%, and at 80% sensitivity for IPSS velocity (cut point of −0.2/year), specificity was 33%. Results for IPSS and IPSS velocity did not differ by age. The PPVs for baseline IPSS varied by cut point, with cut points of 9 or higher yielding PPVs over 50%. The PPVs for IPSS velocity were generally low (<50% for all cut points). At all cut points, the NPVs for both baseline IPSS and IPSS velocity were greater than 75% (Table 3).

There was some evidence that the AUCs for both baseline IPSS and IPSS velocity were higher among men who had minimal LUTS at baseline than among men with moderate LUTS (Figure 2). For baseline IPSS, men with minimal and moderate baseline symptoms had AUCs of 0.70 (95% CI: 0.67, 0.72) and 0.62 (95% CI: 0.59, 0.66), respectively, and for IPSS velocity, the AUCs were 0.71 (95% CI: 0.68, 0.74) and 0.68 (95% CI: 0.64, 0.72), respectively. Table 4 gives the sensitivities and specificities for selected cut points of baseline IPSS and IPSS velocity in men stratified by baseline symptoms. Even though the AUCs for men with a baseline IPSS of 7 or lower were higher than those for men with a baseline IPSS greater than 7, there were still no cut points with sufficiently high sensitivities and specificities to make these measures useful as surrogates for incident BPH. For example, in the subgroup of men with few or no symptoms at baseline, at 80% sensitivity for baseline IPSS (cut point of 3.5), specificity was 46%, and at 80% sensitivity for IPSS velocity (cut point of 0.1/year), specificity was 44%.

Figure 2.

Receiver-operating characteristic curves created using International Prostate Symptom Score (IPSS) and IPSS velocity to predict incident symptomatic benign prostatic hyperplasia (BPH), stratified by baseline IPSS, in the Prostate Cancer Prevention Trial placebo arm, 1993–2003. A) Using data from men with a baseline IPSS of 0–7, we used baseline IPSS predict the 7-year incidence of BPH (area under the receiver-operating characteristic curve (AUC) = 0.70, 95% confidence interval (CI): 0.67, 0.72). B) Using data from men with a baseline IPSS of 0–7, we used IPSS velocity to predict the 5-year incidence of BPH (AUC = 0.71, 95% CI: 0.68, 0.74). Data were limited to men for whom we had complete data (measurements at the baseline, year 1, and year 2 visits). C) Using data from men with a baseline IPSS of 8–13, we used baseline IPSS to predict the 7-year incidence of BPH (AUC = 0.62, 95% CI: 0.59, 0.66). D) Using data from men with a baseline IPSS of 8–13, we used IPSS velocity to predict the 5-year incidence of BPH (AUC = 0.68, 95% CI: 0.64, 0.72). Data were limited to men for whom we had complete data (measurements at the baseline, year 1, and year 2 visits).

Table 4.

Performance Characteristics of Baseline International Prostate Symptom Score and Its Rate of Change Over 2 Years as Surrogate Endpoints of Benign Prostatic Hyperplasia, Stratified by Baseline International Prostate Symptom Score, in the Prostate Cancer Prevention Trial Placebo Arm, 1993–2003

Surrogate Measure of BPH	No. of BPH Cases	Incidence Ratea	Sensitivity	Specificity	PPV	NPV
Baseline IPSS 0–7
Baseline IPSS
Total	475	18.5
≥1	468	18.2	0.984	0.065	0.134	0.889
≥2	438	17.1	0.937	0.207	0.148	0.882
≥3	404	15.7	0.860	0.369	0.167	0.874
≥4	354	13.8	0.743	0.543	0.194	0.864
≥5	282	11.0	0.598	0.696	0.225	0.852
≥6	196	7.6	0.424	0.821	0.258	0.840
≥7	110	4.3	0.228	0.922	0.301	0.825
IPSS velocity
Total	391	24.3
≥0	321	20.0	0.823	0.378	0.139	0.897
≥0.5	283	17.6	0.722	0.579	0.173	0.889
≥1.0	240	14.9	0.612	0.715	0.210	0.878
≥1.5	200	12.4	0.501	0.806	0.241	0.871
≥2.0	164	10.2	0.406	0.872	0.281	0.865
≥2.5	134	8.3	0.323	0.920	0.332	0.860
Baseline IPSS 8–13
Baseline IPSS
Total	367	86.6
≥9	269	63.5	0.723	0.433	0.514	0.584
≥10	178	42.0	0.471	0.716	0.578	0.554
≥11	96	22.7	0.270	0.876	0.641	0.528
≥12	41	9.7	0.122	0.958	0.701	0.507
IPSS velocity
Total	225	92.8
≥0	153	63.1	0.644	0.627	0.484	0.710
≥0.5	124	51.2	0.548	0.727	0.519	0.691
≥1.0	105	43.3	0.460	0.803	0.552	0.675
≥1.5	87	35.9	0.361	0.874	0.596	0.657
≥2.0	68	28.1	0.278	0.924	0.644	0.642
≥2.5	56	23.1	0.210	0.955	0.691	0.630

Abbreviations: BPH, benign prostatic hyperplasia; IPSS, International Prostate Symptom Score; NPV, negative predictive value; PPV, positive predictive value.

^a Per 1,000 person-years.

DISCUSSION

In the present large study of participants in the placebo-arm of the PCPT, there was little evidence to support the use of PSA, IPSS, or their rates of increase over 2 years as surrogate measures of incident BPH. Although each of these measures was associated with a significantly increased risk of BPH, the AUCs created when they were used to predict incident symptomatic BPH were at best modest. There were no cut points for any marker for which sensitivity or specificity was sufficiently high to make these measures useful as surrogates for either 7-year (for PSA and IPSS) or 5-year (for PSA and IPSS velocities) incidence of BPH. With the exception of baseline IPSS, which was modest only at high cut points, PPVs for all other measures were low at all cut points.

Several previously published epidemiologic studies have used PSA and IPSS measures as surrogate BPH measures. For PSA, one set of studies used a PSA level greater than 1.4 ng/mL as a surrogate BPH measure (15–17); however, based on our data, this cut point yields a sensitivity of only 41% and a specificity of 71%. For PSA and IPSS velocities, cut points at the 80th percentile have been used as surrogate BPH measures (18), which in our sample corresponded to slopes of 0.19 ng/mL/year and 1.95 units/year, respectively. The sensitivities and specificities for these values were 21% and 80% for PSA velocity and 37% and 87% for IPSS velocity, respectively. Another measure that has been used in the literature is an IPSS greater than 7 (15, 16, 19–24). Using our data, this cut point yielded a sensitivity of 56% and a specificity of 82%. One possible outcome of the poor performance of these measures is inconsistent findings, in which some but not other measures of the surrogate were associated with BPH risk (15, 16).

The definition of incident symptomatic BPH used in the present study deserves comment. We used a definition that was developed by consensus among experts, including research urologists, epidemiologists, and statisticians, and has been used in a series of previously published epidemiologic articles (25–33). This definition uses multiple criteria to define BPH that capture both medical and surgical treatments, as well as clinically significant LUTS. It is relatively conservative, as it requires men to have repeated IPSS of 15 or above to be considered as having BPH, and it specifically excludes men who reported a physician diagnosis of BPH in the absence of symptoms or treatment. Furthermore, diagnoses based on medication use were specific to BPH, as men who reported nonspecific α-blockers without concomitant evidence of BPH by either self-report of physician diagnosis or symptoms (one report of an IPSS ≥15 or any 2 reports of an IPSS ≥12 at any time before the report of medication use) were not considered to have BPH. It is unknown whether quantitative measures of BPH, such as prostate size or urinary flow rate, would affect this definition; however, these clinical measures are rarely available in large epidemiologic studies, and their relevance to a clinically defined endpoint based on symptom severity is uncertain.

There are several strengths to the present study. Data are from a large 7-year clinical trial; therefore, annual assessments of LUTS and serum PSA concentrations are available from a large number of men. Furthermore, because the PCPT was focused on prostate health, there were careful assessments of treatment and medication use related to prostate disease. The PCPT also enrolled men with a wide range of symptomatology (baseline IPSS <20); therefore, we were able to look at different populations (based on baseline IPSS scores) relevant to a variety of potential hypotheses related to BPH risk and prevention.

We also recognize several limitations to this study that should be considered when interpreting its results. First, our definition of BPH by LUTS is not specific and cannot distinguish between LUTS due to BPH and LUTS due to other urological, neurological, or bladder conditions (34). However, the presence of other conditions associated with LUTS does not substantially affect these results, as the prevalence of these conditions in the PCPT study population is estimated to be only 6.5% (based on the age-specific prevalence of these conditions (13.4%) within the Olmstead County cohort, which also includes men with a history of prostate surgery (34)). Second, men with baseline PSA concentrations greater than 3.0 ng/mL were excluded from participation in the PCPT; inclusion of men with a wider range of PSA values at baseline may yield improved performance of PSA or PSA velocity as surrogate measures of incident BPH. Third, there is no “gold-standard” definition of symptomatic BPH and no universally accepted standard definition of incident symptomatic BPH; therefore, use of a different definition of BPH may modify these results. Fourth, the performances of baseline IPSS and IPSS velocity were likely improved by the fact that IPSS was used as part of the endpoint definition of incident BPH; thus, the interpretation of these results is nuanced. Fifth, analyses of PSA and IPSS velocities were based on changes in these measures over 2 years, and any BPH cases that occurred before year 2 were excluded. This would not affect specificity but could reduce sensitivity; however, it is unlikely to sufficiently increase the AUC to modify this study's conclusions. Lastly, the PCPT was limited to 7 years of follow-up for baseline measures and 5 years of follow-up for velocities, and longer follow-up may modify these results.

In conclusion, we found that modest elevations in PSA, IPSS, and their rates of change over 2 years were not useful surrogates for incident BPH. With the exception of PSA velocity, all measures strongly predicted BPH risk and had AUCs significantly greater than 0.5. Nevertheless, the AUCs were not high enough to make any cut points sufficiently sensitive and specific to consider their use as surrogate measures. Observational epidemiologic studies and primary prevention trials should use standardized measures of clinically significant BPH as study endpoints.