Cases and controls
Two hundred and forty-eight cases of prostate cancer were identified from the cancer register (n = 247) and practice searches (n = 1). Thirty-one were ineligible: nine had previous prostate cancer; five had other or unconfirmed cancers; six resided outside Exeter at diagnosis; and in 11 the records were unobtainable (10 had left Exeter, one had died). For the 217 eligible cases, 1080 controls were studied (in five elderly cases only four controls were available within the maximum 5-year age band). Originally generated were 1272 controls but 192 were ineligible: 18 had previous prostate cancer; 83 (6.5%) had not consulted in the 2 years; 20 resided outside Exeter; and in 71 the records were unobtainable (58 had left Exeter, 13 had died). These totals include 44 (20%) patients and 106 (9.8%) controls who had died but whose notes were retrievable.
Two hundred and five cases (94%) had histological proof of cancer. The remaining 12 had strong clinical evidence for the diagnosis, with at least two of the following: an extremely high PSA, a prostate deemed malignant by a consultant urologist, or radiology results suggestive of bony metastases. Biopsy had not been performed because of concurrent ill-health: all had been treated with depot anti-androgens. Eighteen (8%) cases had been identified by PSA screening in asymptomatic men and 39 (18%) were identified solely by histology after prostatectomy: 20 after an elective prostatectomy for presumed benign hyperplasia, and 19 semi-urgently after admission for retention. These cancers had been clinically unsuspected.11
Patient details are shown in .
Characteristics of patients with prostate cancer and matched controls.
Quality of coding
Inter-observer variation in coding was tested by repeat coding of 188 randomly selected codes by all four coders. The reliability coefficient was 0.83 (95% CI = 0.75 to 0.90).12
Identification of independent associations with cancer
In 2.5% or more of either cases or controls, 172 variables occurred. Abnormal rectal examination findings were classified as benign or malignant depending on the doctor's description. Hard, craggy or nodular glands were classified as malignant. From univariable conditional logistic regressions, 60 variables were considered for multivariable analyses. Selected univariable analyses are shown in .
Univariable analyses of selected variables.
The first and second records of loss of weight were both associated with prostate cancer in the univariable analyses. When they were both added to the same multivariable model, an independent association with prostate cancer was only identified for the second record. Thus the second record was used for all multivariable modelling. In the final model (), no interactions were found.
Multivariable conditional logistic regression analysis of pre-diagnostic features of prostate cancer.
A separate multivariable analysis added the results of PSA testing to the variables in . This model included only the 208 subjects who had been PSA tested (137 patients and 71 controls). In this model, using unmatched regression, the only variable associated with prostate cancer was a PSA >4 ng/ml, with an odds ratio of 29 (95% CI = 3.9 to 220): P = 0.001.
Timing of variable occurrence
Multivariable analysis using data excluding the last 180 days is shown in . The timings of the four variables from , in relation to the date of diagnosis, are shown in . These graphs show the monthly moving average number of presentations to primary care for each variable.
Multivariable conditional logistics regression analysis of pre-diagnositc features of prostate cancer axcluding the final 180 days.
Timing of symptom presentation to primary care in cases and controls.
PPVs for patients consulting a doctor in primary care
The PPVs for particular features of prostate cancer are shown in , both individually and together with a second variable. The variables in were selected from the multivariable analysis, after the exclusion of retention (PPV = 3.1%, 95% CI = 1.5 to 6.0) and impotence (PPV = 3.0%, 95% CI = 1.7 to 4.9). PPVs for retention with a second symptom were not calculated, as retention generally requires hospital admission, as a result of which a prostate cancer is likely to be identified. Impotence was generally an isolated symptom: the maximum number of cases reporting both impotence and any second symptom was two. Lower urinary tract symptoms had the highest univariable PPV, which rose considerably when accompanied by a prostate examination, which the GP considered to be malignant.
Positive predictive values (%) for prostate cancer for individual features, repeat presentations and for pairs of features (against a background risk of 0.35%).
The following PPVs were higher in older patients: impotence 1.1% aged 40–69 years (against a background risk of 0.12%), 8.4% aged ≥70 years (against a background risk of 1.1%); frequency 0.61% and 7.4%; nocturia 1.1% and 5.9%, rectal examination, deemed benign 0.85% and 8.7%, respectively. There were too few younger patients with the other variables for reliable analysis.
In the sub-analysis excluding the 39 patients who had previously unsuspected cancer identified at prostatectomy, the PPVs of symptoms were little changed (available from authors), other than for retention, which fell to 1.6%.