Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Predictors of acute urinary retention in men


Bandolier often hears the complaint that too much attention is given to whether treatments work, and how good they are compared to other treatments, and too little attention is given to help to decide which patient to treat. A possible solution to this problem can come from combined analysis of patients entered into trials, and especially those treated with placebo. If we could determine which of those patients were likely to have an event, then we could treat them to try and avoid it.

Examples are only too rare because pharmaceutical companies, the custodians of the data, often cannot see the need. Praise then when evidence from such analysis looks likely to help decisions about men with clinical BPH [1].


Several large clinical trials have examined the effect of finasteride on men with clinical BPH over two to four years, during which time they were randomised and double-blind. The studies typically excluded men with PSA values above 10 ng/mL, and prostate cancer had to be excluded by biopsy in men with PSA over 4 ng/mL. Men included had moderate symptoms of BPH and a urine flow rate of less than 15 mL/sec.

In all, more than 3,000 men were given placebo, and the rate of spontaneous acute urinary retention (AUR) was known. Data from the men given placebo was used in the analysis. Two thirds of men were randomly selected to form a development set, and the remaining third formed a validation set.

Firstly, 110 baseline clinical variables were identified as potential predictors on the basis of clinical and epidemiological judgement and availability. They included demography, symptoms, bother, urinary flow parameters, comorbidity and concomitant therapy. Prostate volume was available only in a subset of men in one trial.

Logistic regression analysis was used to assess variables in a singly and together in predicting AUR. A potential scoring algorithm was derived using weights from logistic regression coefficients.


On average, men were 64 years old at baseline, had a peak urine flow of 11 mL/sec, and an average symptom score of 15 (moderate), and a mean PSA of 2.9 ng/mL. The spontaneous AUR rate was 2.5% over two years and 3.7% over four years.

Eventually three systems were chosen for testing. One was a five variable model that included urinating more than every two hours, symptom problem index, maximum urinary flow rate, hesitancy when urinating and PSA. Another was an algorithm using symptom problem index, PSA, urinating more often than every two hours and peak urine flow (Figure 1). The third was PSA alone.

Figure 1: Algorithm for predicting acute urinary retention

Start with serum PSA value
PSA ≥ 3 ng/mL AUR likely if peak flow <11.8 mL/sec
PSA<3 ng/mL AUR likely if:
1 Symptom problem index >8.5, AND
2 Frequent urination few times or less in last month, AND
3 Peak flow less than 8.1 mL/sec

Each of these performed about equally well in development and validation sets, and with the combined data (Table 1). If a test performed perfectly in prediction, the area under the ROC curve would be 1, and if it was no better than chance it would be 0.5. Figures of about 0.7 and higher are typical of many tests we use today. PSA alone, for instance, had a sensitivity of 75% and a specificity of 64%, giving a positive likelihood ratio of 2.1.

Table 1: Results for model, algorithm and PSA alone for prediction of AUR

Area under ROC curve
Data set Number of men Number of AURs Five element model Algorithm PSA alone
Model development 2146 67 0.71 0.76 0.68
Validation 1016 30 0.74 0.73 0.72
All data 3162 97 0.71 0.75 0.71

With PSA alone, the incidence of spontaneous AUR over four and two years was much higher in men in the upper tertile of 3.3 to 12 ng/mL (Figure 2). Over two years at least one man in 20 will have spontaneous AUR if they have symptoms of BPH and a moderately reduced maximum urinary flow rate (less than 15 mL/sec).

Figure 2: Acute urinary retention in RCTs over two and four years according to initial PSA


This is useful stuff, though readers should note a couple of things from the paper. Firstly, the decision point for PSA alone is implied as being 3.0 ng/mL, though that is not explicit. Secondly, when the algorithm uses frequent urination, the choice of this being a few times or less in the past month rather than more than a few times looks as if it may be the wrong way round.

The implications, though, are interesting. Since the mean PSA value at baseline in the trials was below 3 ng/mL, it would suggest that perhaps only about half the men who could form a potential treatment group needed to be treated. It would imply that better results of treatment with finasteride might come from men with higher PSA values, though the evidence for this is not available, so far as we know. Perhaps it will form the basis of further analysis of data at the single patient level.


  1. CG Roehrborn et al. Clinical predictors of spontaneous acute urinary retention in men with LUTS and clinical BPH: a comprehensive analysis of the pooled placebo groups of several large clinical trials. Urology 2001 58: 210-216.
previous or next story in this issue